lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC(sbi, sizeof(*sbi));
  82         if (!sbi)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         spin_lock_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         spin_lock(&ll_sb_lock);
 115         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_LLITE_CHECKSUM
 119         sbi->ll_flags |= LL_SBI_CHECKSUM;
 120 #endif
 121
 122 #ifdef HAVE_LRU_RESIZE_SUPPORT
 123         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 124 #endif
 125
 126 #ifdef HAVE_EXPORT___IGET
 127         INIT_LIST_HEAD(&sbi->ll_deathrow);
 128         spin_lock_init(&sbi->ll_deathrow_lock);
 129 #endif
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137
 138         RETURN(sbi);
 139 }
 140
 141 void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144         ENTRY;
 145
 146         if (sbi != NULL) {
 147                 spin_lock(&ll_sb_lock);
 148                 list_del(&sbi->ll_list);
 149                 spin_unlock(&ll_sb_lock);
 150                 OBD_FREE(sbi, sizeof(*sbi));
 151         }
 152         EXIT;
 153 }
 154
 155 static struct dentry_operations ll_d_root_ops = {
 156 #ifdef DCACHE_LUSTRE_INVALID
 157         .d_compare = ll_dcompare,
 158 #endif
 159 };
 160
 161 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 162  * us to make MDS RPCs with large enough reply buffers to hold the
 163  * maximum-sized (= maximum striped) EA and cookie without having to
 164  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 165 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 166 {
 167         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
 168         __u32 valsize = sizeof(struct lov_desc);
 169         int rc, easize, def_easize, cookiesize;
 170         struct lov_desc desc;
 171         __u32 stripes;
 172         ENTRY;
 173
 174         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 175                           &valsize, &desc, NULL);
 176         if (rc)
 177                 RETURN(rc);
 178
 179         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 180         lsm.lsm_stripe_count = stripes;
 181         easize = obd_size_diskmd(dt_exp, &lsm);
 182
 183         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 184         def_easize = obd_size_diskmd(dt_exp, &lsm);
 185
 186         cookiesize = stripes * sizeof(struct llog_cookie);
 187
 188         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 189                easize, cookiesize);
 190
 191         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 192         RETURN(rc);
 193 }
 194
 195 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 196 {
 197         struct inode *root = 0;
 198         struct ll_sb_info *sbi = ll_s2sbi(sb);
 199         struct obd_device *obd;
 200         struct obd_capa *oc = NULL;
 201         struct obd_statfs osfs;
 202         struct ptlrpc_request *request = NULL;
 203         struct lustre_handle dt_conn = {0, };
 204         struct lustre_handle md_conn = {0, };
 205         struct obd_connect_data *data = NULL;
 206         struct obd_uuid *uuid;
 207         struct lustre_md lmd;
 208         obd_valid valid;
 209         int size, err, checksum;
 210         ENTRY;
 211
 212         obd = class_name2obd(md);
 213         if (!obd) {
 214                 CERROR("MD %s: not setup or attached\n", md);
 215                 RETURN(-EINVAL);
 216         }
 217
 218         OBD_ALLOC_PTR(data);
 219         if (data == NULL)
 220                 RETURN(-ENOMEM);
 221
 222         if (proc_lustre_fs_root) {
 223                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 224                                                   dt, md);
 225                 if (err < 0)
 226                         CERROR("could not register mount in /proc/fs/lustre\n");
 227         }
 228
 229         /* indicate the features supported by this client */
 230         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 231                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 232                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 233                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 234                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 235                                   OBD_CONNECT_LOV_V3;
 236
 237 #ifdef HAVE_LRU_RESIZE_SUPPORT
 238         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 239                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 240 #endif
 241 #ifdef CONFIG_FS_POSIX_ACL
 242         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 243 #endif
 244         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 245         data->ocd_version = LUSTRE_VERSION_CODE;
 246
 247         if (sb->s_flags & MS_RDONLY)
 248                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 249         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 250                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 251
 252 #ifdef HAVE_MS_FLOCK_LOCK
 253         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 254         sb->s_flags |= MS_FLOCK_LOCK;
 255 #endif
 256
 257         if (sbi->ll_flags & LL_SBI_FLOCK)
 258                 sbi->ll_fop = &ll_file_operations_flock;
 259         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 260                 sbi->ll_fop = &ll_file_operations;
 261         else
 262                 sbi->ll_fop = &ll_file_operations_noflock;
 263
 264         /* real client */
 265         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 266         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 267                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 268                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 269         } else {
 270                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 271                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 272         }
 273
 274         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 275         if (err == -EBUSY) {
 276                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 277                                    "recovery, of which this client is not a "
 278                                    "part. Please wait for recovery to complete,"
 279                                    " abort, or time out.\n", md);
 280                 GOTO(out, err);
 281         } else if (err) {
 282                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 283                 GOTO(out, err);
 284         }
 285         sbi->ll_md_exp = class_conn2export(&md_conn);
 286
 287         err = obd_fid_init(sbi->ll_md_exp);
 288         if (err) {
 289                 CERROR("Can't init metadata layer FID infrastructure, "
 290                        "rc %d\n", err);
 291                 GOTO(out_md, err);
 292         }
 293
 294         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         size = sizeof(*data);
 299         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 300                            KEY_CONN_DATA,  &size, data, NULL);
 301         if (err) {
 302                 CERROR("Get connect data failed: %d \n", err);
 303                 GOTO(out_md, err);
 304         }
 305
 306         LASSERT(osfs.os_bsize);
 307         sb->s_blocksize = osfs.os_bsize;
 308         sb->s_blocksize_bits = log2(osfs.os_bsize);
 309         sb->s_magic = LL_SUPER_MAGIC;
 310
 311         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 312          *         retval = in_file->f_op->sendfile(...);
 313          *         if (*ppos > max)
 314          *                 retval = -EOVERFLOW;
 315          *
 316          * it will check if *ppos is greater than max. However, max equals to
 317          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 318          * has been defined as a signed long long ineger in linux kernel. */
 319 #if BITS_PER_LONG == 64
 320         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 321 #else
 322         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 323 #endif
 324         sbi->ll_namelen = osfs.os_namelen;
 325         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 326
 327         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 328             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 329                 LCONSOLE_INFO("Disabling user_xattr feature because "
 330                               "it is not supported on the server\n");
 331                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 332         }
 333
 334         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 335 #ifdef MS_POSIXACL
 336                 sb->s_flags |= MS_POSIXACL;
 337 #endif
 338                 sbi->ll_flags |= LL_SBI_ACL;
 339         } else {
 340                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 341 #ifdef MS_POSIXACL
 342                 sb->s_flags &= ~MS_POSIXACL;
 343 #endif
 344                 sbi->ll_flags &= ~LL_SBI_ACL;
 345         }
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 348                 sbi->ll_flags |= LL_SBI_JOIN;
 349
 350         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 351                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 352                         /* sometimes local client claims to be remote, but mdt
 353                          * will disagree when client gss not applied. */
 354                         LCONSOLE_INFO("client claims to be remote, but server "
 355                                       "rejected, forced to be local.\n");
 356                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 357                 }
 358         } else {
 359                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 360                         /* with gss applied, remote client can not claim to be
 361                          * local, so mdt maybe force client to be remote. */
 362                         LCONSOLE_INFO("client claims to be local, but server "
 363                                       "rejected, forced to be remote.\n");
 364                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 365                 }
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 369                 LCONSOLE_INFO("client enabled MDS capability!\n");
 370                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 374                 LCONSOLE_INFO("client enabled OSS capability!\n");
 375                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 376         }
 377
 378         obd = class_name2obd(dt);
 379         if (!obd) {
 380                 CERROR("DT %s: not setup or attached\n", dt);
 381                 GOTO(out_md_fid, err = -ENODEV);
 382         }
 383
 384         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 385                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 386                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 387                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 388                                   OBD_CONNECT_AT;
 389         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 390                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 391
 392         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 393                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 394                  * disabled by default, because it can still be enabled on the
 395                  * fly via /proc. As a consequence, we still need to come to an
 396                  * agreement on the supported algorithms at connect time */
 397                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 398
 399                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 400                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 401                 else
 402                         /* send the list of supported checksum types */
 403                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 404         }
 405
 406 #ifdef HAVE_LRU_RESIZE_SUPPORT
 407         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 408 #endif
 409         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 410                "ocd_grant: %d\n", data->ocd_connect_flags,
 411                data->ocd_version, data->ocd_grant);
 412
 413         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 414         obd->obd_upcall.onu_upcall = ll_ocd_update;
 415         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 416
 417         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 418         if (err == -EBUSY) {
 419                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 420                                    "recovery, of which this client is not a "
 421                                    "part.  Please wait for recovery to "
 422                                    "complete, abort, or time out.\n", dt);
 423                 GOTO(out_md_fid, err);
 424         } else if (err) {
 425                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 426                 GOTO(out_md_fid, err);
 427         }
 428
 429         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 430
 431         err = obd_fid_init(sbi->ll_dt_exp);
 432         if (err) {
 433                 CERROR("Can't init data layer FID infrastructure, "
 434                        "rc %d\n", err);
 435                 GOTO(out_dt, err);
 436         }
 437
 438         spin_lock(&sbi->ll_lco.lco_lock);
 439         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 440         spin_unlock(&sbi->ll_lco.lco_lock);
 441
 442         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 443         if (err) {
 444                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 445                 GOTO(out_lock_cn_cb, err);
 446         }
 447
 448         fid_zero(&sbi->ll_root_fid);
 449         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 450         if (err) {
 451                 CERROR("cannot mds_connect: rc = %d\n", err);
 452                 GOTO(out_lock_cn_cb, err);
 453         }
 454         if (!fid_is_sane(&sbi->ll_root_fid)) {
 455                 CERROR("Invalid root fid during mount\n");
 456                 GOTO(out_lock_cn_cb, err = -EINVAL);
 457         }
 458         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 459
 460         sb->s_op = &lustre_super_operations;
 461         sb->s_export_op = &lustre_export_operations;
 462
 463         /* make root inode
 464          * XXX: move this to after cbd setup? */
 465         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 466         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 467                 valid |= OBD_MD_FLRMTPERM;
 468         else if (sbi->ll_flags & LL_SBI_ACL)
 469                 valid |= OBD_MD_FLACL;
 470
 471         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 472                          &request);
 473         if (oc)
 474                 free_capa(oc);
 475         if (err) {
 476                 CERROR("md_getattr failed for root: rc = %d\n", err);
 477                 GOTO(out_lock_cn_cb, err);
 478         }
 479         memset(&lmd, 0, sizeof(lmd));
 480         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 481                                sbi->ll_md_exp, &lmd);
 482         if (err) {
 483                 CERROR("failed to understand root inode md: rc = %d\n", err);
 484                 ptlrpc_req_finished (request);
 485                 GOTO(out_lock_cn_cb, err);
 486         }
 487
 488         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 489         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 490         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 491         ptlrpc_req_finished(request);
 492
 493         if (root == NULL || IS_ERR(root)) {
 494                 if (lmd.lsm)
 495                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 496 #ifdef CONFIG_FS_POSIX_ACL
 497                 if (lmd.posix_acl) {
 498                         posix_acl_release(lmd.posix_acl);
 499                         lmd.posix_acl = NULL;
 500                 }
 501 #endif
 502                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 503                 root = NULL;
 504                 CERROR("lustre_lite: bad iget4 for root\n");
 505                 GOTO(out_root, err);
 506         }
 507
 508         err = ll_close_thread_start(&sbi->ll_lcq);
 509         if (err) {
 510                 CERROR("cannot start close thread: rc %d\n", err);
 511                 GOTO(out_root, err);
 512         }
 513
 514 #ifdef CONFIG_FS_POSIX_ACL
 515         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 516                 rct_init(&sbi->ll_rct);
 517                 et_init(&sbi->ll_et);
 518         }
 519 #endif
 520
 521         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 522         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 523                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 524                                  NULL);
 525         cl_sb_init(sb);
 526
 527         sb->s_root = d_alloc_root(root);
 528         if (data != NULL)
 529                 OBD_FREE(data, sizeof(*data));
 530
 531         sb->s_root->d_op = &ll_d_root_ops;
 532
 533         sbi->ll_sdev_orig = sb->s_dev;
 534
 535         /* We set sb->s_dev equal on all lustre clients in order to support
 536          * NFS export clustering.  NFSD requires that the FSID be the same
 537          * on all clients. */
 538         /* s_dev is also used in lt_compare() to compare two fs, but that is
 539          * only a node-local comparison. */
 540         uuid = obd_get_uuid(sbi->ll_md_exp);
 541         if (uuid != NULL)
 542                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 543
 544         RETURN(err);
 545 out_root:
 546         if (root)
 547                 iput(root);
 548 out_lock_cn_cb:
 549         obd_fid_fini(sbi->ll_dt_exp);
 550 out_dt:
 551         obd_disconnect(sbi->ll_dt_exp);
 552         sbi->ll_dt_exp = NULL;
 553 out_md_fid:
 554         obd_fid_fini(sbi->ll_md_exp);
 555 out_md:
 556         obd_disconnect(sbi->ll_md_exp);
 557         sbi->ll_md_exp = NULL;
 558 out:
 559         if (data != NULL)
 560                 OBD_FREE_PTR(data);
 561         lprocfs_unregister_mountpoint(sbi);
 562         return err;
 563 }
 564
 565 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 566 {
 567         int size, rc;
 568
 569         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 570         size = sizeof(int);
 571         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 572                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 573         if (rc)
 574                 CERROR("Get max mdsize error rc %d \n", rc);
 575
 576         RETURN(rc);
 577 }
 578
 579 void ll_dump_inode(struct inode *inode)
 580 {
 581         struct list_head *tmp;
 582         int dentry_count = 0;
 583
 584         LASSERT(inode != NULL);
 585
 586         list_for_each(tmp, &inode->i_dentry)
 587                 dentry_count++;
 588
 589         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 590                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 591                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 592 }
 593
 594 void lustre_dump_dentry(struct dentry *dentry, int recur)
 595 {
 596         struct list_head *tmp;
 597         int subdirs = 0;
 598
 599         LASSERT(dentry != NULL);
 600
 601         list_for_each(tmp, &dentry->d_subdirs)
 602                 subdirs++;
 603
 604         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 605                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 606                dentry->d_name.len, dentry->d_name.name,
 607                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 608                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 609                dentry->d_flags, dentry->d_fsdata, subdirs);
 610         if (dentry->d_inode != NULL)
 611                 ll_dump_inode(dentry->d_inode);
 612
 613         if (recur == 0)
 614                 return;
 615
 616         list_for_each(tmp, &dentry->d_subdirs) {
 617                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 618                 lustre_dump_dentry(d, recur - 1);
 619         }
 620 }
 621
 622 #ifdef HAVE_EXPORT___IGET
 623 static void prune_dir_dentries(struct inode *inode)
 624 {
 625         struct dentry *dentry, *prev = NULL;
 626
 627         /* due to lustre specific logic, a directory
 628          * can have few dentries - a bug from VFS POV */
 629 restart:
 630         spin_lock(&dcache_lock);
 631         if (!list_empty(&inode->i_dentry)) {
 632                 dentry = list_entry(inode->i_dentry.prev,
 633                                     struct dentry, d_alias);
 634                 /* in order to prevent infinite loops we
 635                  * break if previous dentry is busy */
 636                 if (dentry != prev) {
 637                         prev = dentry;
 638                         dget_locked(dentry);
 639                         spin_unlock(&dcache_lock);
 640
 641                         /* try to kill all child dentries */
 642                         lock_dentry(dentry);
 643                         shrink_dcache_parent(dentry);
 644                         unlock_dentry(dentry);
 645                         dput(dentry);
 646
 647                         /* now try to get rid of current dentry */
 648                         d_prune_aliases(inode);
 649                         goto restart;
 650                 }
 651         }
 652         spin_unlock(&dcache_lock);
 653 }
 654
 655 static void prune_deathrow_one(struct ll_inode_info *lli)
 656 {
 657         struct inode *inode = ll_info2i(lli);
 658
 659         /* first, try to drop any dentries - they hold a ref on the inode */
 660         if (S_ISDIR(inode->i_mode))
 661                 prune_dir_dentries(inode);
 662         else
 663                 d_prune_aliases(inode);
 664
 665
 666         /* if somebody still uses it, leave it */
 667         LASSERT(atomic_read(&inode->i_count) > 0);
 668         if (atomic_read(&inode->i_count) > 1)
 669                 goto out;
 670
 671         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 672                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 673
 674         /* seems nobody uses it anymore */
 675         inode->i_nlink = 0;
 676
 677 out:
 678         iput(inode);
 679         return;
 680 }
 681
 682 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 683 {
 684         struct ll_inode_info *lli;
 685         int empty;
 686
 687         do {
 688                 if (need_resched() && try)
 689                         break;
 690
 691                 if (try) {
 692                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 693                                 break;
 694                 } else {
 695                         spin_lock(&sbi->ll_deathrow_lock);
 696                 }
 697
 698                 empty = 1;
 699                 lli = NULL;
 700                 if (!list_empty(&sbi->ll_deathrow)) {
 701                         lli = list_entry(sbi->ll_deathrow.next,
 702                                          struct ll_inode_info,
 703                                          lli_dead_list);
 704                         list_del_init(&lli->lli_dead_list);
 705                         if (!list_empty(&sbi->ll_deathrow))
 706                                 empty = 0;
 707                 }
 708                 spin_unlock(&sbi->ll_deathrow_lock);
 709
 710                 if (lli)
 711                         prune_deathrow_one(lli);
 712
 713         } while (empty == 0);
 714 }
 715 #else /* !HAVE_EXPORT___IGET */
 716 #define prune_deathrow(sbi, try) do {} while (0)
 717 #endif /* HAVE_EXPORT___IGET */
 718
 719 void client_common_put_super(struct super_block *sb)
 720 {
 721         struct ll_sb_info *sbi = ll_s2sbi(sb);
 722         ENTRY;
 723
 724 #ifdef CONFIG_FS_POSIX_ACL
 725         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 726                 et_fini(&sbi->ll_et);
 727                 rct_fini(&sbi->ll_rct);
 728         }
 729 #endif
 730
 731         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 732
 733         ll_close_thread_shutdown(sbi->ll_lcq);
 734
 735         /* destroy inodes in deathrow */
 736         prune_deathrow(sbi, 0);
 737
 738         list_del(&sbi->ll_conn_chain);
 739
 740         obd_fid_fini(sbi->ll_dt_exp);
 741         obd_disconnect(sbi->ll_dt_exp);
 742         sbi->ll_dt_exp = NULL;
 743
 744         lprocfs_unregister_mountpoint(sbi);
 745
 746         obd_fid_fini(sbi->ll_md_exp);
 747         obd_disconnect(sbi->ll_md_exp);
 748         sbi->ll_md_exp = NULL;
 749
 750         EXIT;
 751 }
 752
 753 void ll_kill_super(struct super_block *sb)
 754 {
 755         struct ll_sb_info *sbi;
 756
 757         ENTRY;
 758
 759         /* not init sb ?*/
 760         if (!(sb->s_flags & MS_ACTIVE))
 761                 return;
 762
 763         sbi = ll_s2sbi(sb);
 764         /* we need restore s_dev from changed for clustred NFS before put_super
 765          * because new kernels have cached s_dev and change sb->s_dev in
 766          * put_super not affected real removing devices */
 767         if (sbi)
 768                 sb->s_dev = sbi->ll_sdev_orig;
 769         EXIT;
 770 }
 771
 772 char *ll_read_opt(const char *opt, char *data)
 773 {
 774         char *value;
 775         char *retval;
 776         ENTRY;
 777
 778         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 779         if (strncmp(opt, data, strlen(opt)))
 780                 RETURN(NULL);
 781         if ((value = strchr(data, '=')) == NULL)
 782                 RETURN(NULL);
 783
 784         value++;
 785         OBD_ALLOC(retval, strlen(value) + 1);
 786         if (!retval) {
 787                 CERROR("out of memory!\n");
 788                 RETURN(NULL);
 789         }
 790
 791         memcpy(retval, value, strlen(value)+1);
 792         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 793         RETURN(retval);
 794 }
 795
 796 static inline int ll_set_opt(const char *opt, char *data, int fl)
 797 {
 798         if (strncmp(opt, data, strlen(opt)) != 0)
 799                 return(0);
 800         else
 801                 return(fl);
 802 }
 803
 804 /* non-client-specific mount options are parsed in lmd_parse */
 805 static int ll_options(char *options, int *flags)
 806 {
 807         int tmp;
 808         char *s1 = options, *s2;
 809         ENTRY;
 810
 811         if (!options)
 812                 RETURN(0);
 813
 814         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 815
 816         while (*s1) {
 817                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 818                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 849                 if (tmp) {
 850                         /* Ignore deprecated mount option.  The client will
 851                          * always try to mount with ACL support, whether this
 852                          * is used depends on whether server supports it. */
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 856                 if (tmp) {
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 860                 if (tmp) {
 861                         *flags |= tmp;
 862                         goto next;
 863                 }
 864
 865                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 871                 if (tmp) {
 872                         *flags &= ~tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 876                 if (tmp) {
 877                         *flags |= tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 881                 if (tmp) {
 882                         *flags &= ~tmp;
 883                         goto next;
 884                 }
 885
 886                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 887                                    s1);
 888                 RETURN(-EINVAL);
 889
 890 next:
 891                 /* Find next opt */
 892                 s2 = strchr(s1, ',');
 893                 if (s2 == NULL)
 894                         break;
 895                 s1 = s2 + 1;
 896         }
 897         RETURN(0);
 898 }
 899
 900 void ll_lli_init(struct ll_inode_info *lli)
 901 {
 902         lli->lli_inode_magic = LLI_INODE_MAGIC;
 903         sema_init(&lli->lli_size_sem, 1);
 904         sema_init(&lli->lli_write_sem, 1);
 905         lli->lli_flags = 0;
 906         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 907         spin_lock_init(&lli->lli_lock);
 908         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 909         INIT_LIST_HEAD(&lli->lli_close_list);
 910         lli->lli_inode_magic = LLI_INODE_MAGIC;
 911         sema_init(&lli->lli_och_sem, 1);
 912         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 913         lli->lli_mds_exec_och = NULL;
 914         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 915         lli->lli_open_fd_exec_count = 0;
 916         INIT_LIST_HEAD(&lli->lli_dead_list);
 917         lli->lli_remote_perms = NULL;
 918         lli->lli_rmtperm_utime = 0;
 919         sema_init(&lli->lli_rmtperm_sem, 1);
 920         INIT_LIST_HEAD(&lli->lli_oss_capas);
 921 }
 922
 923 int ll_fill_super(struct super_block *sb)
 924 {
 925         struct lustre_profile *lprof;
 926         struct lustre_sb_info *lsi = s2lsi(sb);
 927         struct ll_sb_info *sbi;
 928         char  *dt = NULL, *md = NULL;
 929         char  *profilenm = get_profile_name(sb);
 930         struct config_llog_instance cfg = {0, };
 931         char   ll_instance[sizeof(sb) * 2 + 1];
 932         int    err;
 933         ENTRY;
 934
 935         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 936
 937         cfs_module_get();
 938
 939         /* client additional sb info */
 940         lsi->lsi_llsbi = sbi = ll_init_sbi();
 941         if (!sbi) {
 942                 cfs_module_put();
 943                 RETURN(-ENOMEM);
 944         }
 945
 946         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 947         if (err)
 948                 GOTO(out_free, err);
 949
 950         /* Generate a string unique to this super, in case some joker tries
 951            to mount the same fs at two mount points.
 952            Use the address of the super itself.*/
 953         sprintf(ll_instance, "%p", sb);
 954         cfg.cfg_instance = ll_instance;
 955         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 956
 957         /* set up client obds */
 958         err = lustre_process_log(sb, profilenm, &cfg);
 959         if (err < 0) {
 960                 CERROR("Unable to process log: %d\n", err);
 961                 GOTO(out_free, err);
 962         }
 963
 964         lprof = class_get_profile(profilenm);
 965         if (lprof == NULL) {
 966                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 967                                    " read from the MGS.  Does that filesystem "
 968                                    "exist?\n", profilenm);
 969                 GOTO(out_free, err = -EINVAL);
 970         }
 971         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 972                lprof->lp_md, lprof->lp_dt);
 973
 974         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 975                   strlen(ll_instance) + 2);
 976         if (!dt)
 977                 GOTO(out_free, err = -ENOMEM);
 978         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 979
 980         OBD_ALLOC(md, strlen(lprof->lp_md) +
 981                   strlen(ll_instance) + 2);
 982         if (!md)
 983                 GOTO(out_free, err = -ENOMEM);
 984         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 985
 986         /* connections, registrations, sb setup */
 987         err = client_common_fill_super(sb, md, dt);
 988
 989 out_free:
 990         if (md)
 991                 OBD_FREE(md, strlen(md) + 1);
 992         if (dt)
 993                 OBD_FREE(dt, strlen(dt) + 1);
 994         if (err)
 995                 ll_put_super(sb);
 996         else
 997                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 998
 999         RETURN(err);
1000 } /* ll_fill_super */
1001
1002
1003 void lu_context_keys_dump(void);
1004
1005 void ll_put_super(struct super_block *sb)
1006 {
1007         struct config_llog_instance cfg;
1008         char   ll_instance[sizeof(sb) * 2 + 1];
1009         struct obd_device *obd;
1010         struct lustre_sb_info *lsi = s2lsi(sb);
1011         struct ll_sb_info *sbi = ll_s2sbi(sb);
1012         char *profilenm = get_profile_name(sb);
1013         int force = 1, next;
1014         ENTRY;
1015
1016         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1017
1018         ll_print_capa_stat(sbi);
1019
1020         sprintf(ll_instance, "%p", sb);
1021         cfg.cfg_instance = ll_instance;
1022         lustre_end_log(sb, NULL, &cfg);
1023
1024         if (sbi->ll_md_exp) {
1025                 obd = class_exp2obd(sbi->ll_md_exp);
1026                 if (obd)
1027                         force = obd->obd_force;
1028         }
1029
1030         /* We need to set force before the lov_disconnect in
1031            lustre_common_put_super, since l_d cleans up osc's as well. */
1032         if (force) {
1033                 next = 0;
1034                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1035                                                      &next)) != NULL) {
1036                         obd->obd_force = force;
1037                 }
1038         }
1039
1040         cl_sb_fini(sb);
1041
1042         if (sbi->ll_lcq) {
1043                 /* Only if client_common_fill_super succeeded */
1044                 client_common_put_super(sb);
1045         }
1046         next = 0;
1047         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1048                 class_manual_cleanup(obd);
1049         }
1050
1051         if (profilenm)
1052                 class_del_profile(profilenm);
1053
1054         ll_free_sbi(sb);
1055         lsi->lsi_llsbi = NULL;
1056
1057         lustre_common_put_super(sb);
1058
1059         cl_env_cache_purge(~0);
1060         lu_context_keys_dump();
1061
1062         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1063
1064         cfs_module_put();
1065
1066         EXIT;
1067 } /* client_put_super */
1068
1069 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1070 {
1071         struct inode *inode = NULL;
1072         /* NOTE: we depend on atomic igrab() -bzzz */
1073         lock_res_and_lock(lock);
1074         if (lock->l_ast_data) {
1075                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1076                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1077                         inode = igrab(lock->l_ast_data);
1078                 } else {
1079                         inode = lock->l_ast_data;
1080                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1081                                                 D_INFO : D_WARNING,
1082                                         lock, __FILE__, __func__, __LINE__,
1083                                         "l_ast_data %p is bogus: magic %08x",
1084                                         lock->l_ast_data, lli->lli_inode_magic);
1085                         inode = NULL;
1086                 }
1087         }
1088         unlock_res_and_lock(lock);
1089         return inode;
1090 }
1091
1092 static int null_if_equal(struct ldlm_lock *lock, void *data)
1093 {
1094         if (data == lock->l_ast_data) {
1095                 lock->l_ast_data = NULL;
1096
1097                 if (lock->l_req_mode != lock->l_granted_mode)
1098                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1099         }
1100
1101         return LDLM_ITER_CONTINUE;
1102 }
1103
1104 void ll_clear_inode(struct inode *inode)
1105 {
1106         struct ll_inode_info *lli = ll_i2info(inode);
1107         struct ll_sb_info *sbi = ll_i2sbi(inode);
1108         ENTRY;
1109
1110         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1111                inode->i_generation, inode);
1112
1113         if (S_ISDIR(inode->i_mode)) {
1114                 /* these should have been cleared in ll_file_release */
1115                 LASSERT(lli->lli_sai == NULL);
1116                 LASSERT(lli->lli_opendir_key == NULL);
1117                 LASSERT(lli->lli_opendir_pid == 0);
1118         }
1119
1120         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1121         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1122                          null_if_equal, inode);
1123
1124         LASSERT(!lli->lli_open_fd_write_count);
1125         LASSERT(!lli->lli_open_fd_read_count);
1126         LASSERT(!lli->lli_open_fd_exec_count);
1127
1128         if (lli->lli_mds_write_och)
1129                 ll_md_real_close(inode, FMODE_WRITE);
1130         if (lli->lli_mds_exec_och)
1131                 ll_md_real_close(inode, FMODE_EXEC);
1132         if (lli->lli_mds_read_och)
1133                 ll_md_real_close(inode, FMODE_READ);
1134
1135         if (lli->lli_symlink_name) {
1136                 OBD_FREE(lli->lli_symlink_name,
1137                          strlen(lli->lli_symlink_name) + 1);
1138                 lli->lli_symlink_name = NULL;
1139         }
1140
1141         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1142                 LASSERT(lli->lli_posix_acl == NULL);
1143                 if (lli->lli_remote_perms) {
1144                         free_rmtperm_hash(lli->lli_remote_perms);
1145                         lli->lli_remote_perms = NULL;
1146                 }
1147         }
1148 #ifdef CONFIG_FS_POSIX_ACL
1149         else if (lli->lli_posix_acl) {
1150                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1151                 LASSERT(lli->lli_remote_perms == NULL);
1152                 posix_acl_release(lli->lli_posix_acl);
1153                 lli->lli_posix_acl = NULL;
1154         }
1155 #endif
1156         lli->lli_inode_magic = LLI_INODE_DEAD;
1157
1158 #ifdef HAVE_EXPORT___IGET
1159         spin_lock(&sbi->ll_deathrow_lock);
1160         list_del_init(&lli->lli_dead_list);
1161         spin_unlock(&sbi->ll_deathrow_lock);
1162 #endif
1163         ll_clear_inode_capas(inode);
1164         /*
1165          * XXX This has to be done before lsm is freed below, because
1166          * cl_object still uses inode lsm.
1167          */
1168         cl_inode_fini(inode);
1169
1170         if (lli->lli_smd) {
1171                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1172                 lli->lli_smd = NULL;
1173         }
1174
1175
1176         EXIT;
1177 }
1178
1179 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1180                   struct md_open_data **mod)
1181 {
1182         struct lustre_md md;
1183         struct ll_sb_info *sbi = ll_i2sbi(inode);
1184         struct ptlrpc_request *request = NULL;
1185         int rc;
1186         ENTRY;
1187
1188         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1189                                      LUSTRE_OPC_ANY, NULL);
1190         if (IS_ERR(op_data))
1191                 RETURN(PTR_ERR(op_data));
1192
1193         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1194                         &request, mod);
1195         if (rc) {
1196                 ptlrpc_req_finished(request);
1197                 if (rc == -ENOENT) {
1198                         inode->i_nlink = 0;
1199                         /* Unlinked special device node? Or just a race?
1200                          * Pretend we done everything. */
1201                         if (!S_ISREG(inode->i_mode) &&
1202                             !S_ISDIR(inode->i_mode))
1203                                 rc = inode_setattr(inode, &op_data->op_attr);
1204                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1205                         CERROR("md_setattr fails: rc = %d\n", rc);
1206                 }
1207                 RETURN(rc);
1208         }
1209
1210         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1211                               sbi->ll_md_exp, &md);
1212         if (rc) {
1213                 ptlrpc_req_finished(request);
1214                 RETURN(rc);
1215         }
1216
1217         /* We call inode_setattr to adjust timestamps.
1218          * If there is at least some data in file, we cleared ATTR_SIZE
1219          * above to avoid invoking vmtruncate, otherwise it is important
1220          * to call vmtruncate in inode_setattr to update inode->i_size
1221          * (bug 6196) */
1222         rc = inode_setattr(inode, &op_data->op_attr);
1223
1224         /* Extract epoch data if obtained. */
1225         op_data->op_handle = md.body->handle;
1226         op_data->op_ioepoch = md.body->ioepoch;
1227
1228         ll_update_inode(inode, &md);
1229         ptlrpc_req_finished(request);
1230
1231         RETURN(rc);
1232 }
1233
1234 /* Close IO epoch and send Size-on-MDS attribute update. */
1235 static int ll_setattr_done_writing(struct inode *inode,
1236                                    struct md_op_data *op_data,
1237                                    struct md_open_data *mod)
1238 {
1239         struct ll_inode_info *lli = ll_i2info(inode);
1240         int rc = 0;
1241         ENTRY;
1242
1243         LASSERT(op_data != NULL);
1244         if (!S_ISREG(inode->i_mode))
1245                 RETURN(0);
1246
1247         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1248                op_data->op_ioepoch, PFID(&lli->lli_fid));
1249
1250         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1251         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1252         if (rc == -EAGAIN) {
1253                 /* MDS has instructed us to obtain Size-on-MDS attribute
1254                  * from OSTs and send setattr to back to MDS. */
1255                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1256                                          op_data->op_ioepoch);
1257         } else if (rc) {
1258                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1259                        inode->i_ino, rc);
1260         }
1261         RETURN(rc);
1262 }
1263
1264 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1265 {
1266         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1267         int rc;
1268
1269         rc = cl_setattr_do_truncate(inode, size, capa);
1270         ll_truncate_free_capa(capa);
1271         return rc;
1272 }
1273
1274 static int ll_setattr_ost(struct inode *inode)
1275 {
1276         struct obd_capa *capa = ll_mdscapa_get(inode);
1277         int rc;
1278
1279         rc = cl_setattr_ost(inode, capa);
1280         capa_put(capa);
1281
1282         return rc;
1283 }
1284
1285 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1286  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1287  * keep these values until such a time that objects are allocated for it.
1288  * We do the MDS operations first, as it is checking permissions for us.
1289  * We don't to the MDS RPC if there is nothing that we want to store there,
1290  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1291  * going to do an RPC anyways.
1292  *
1293  * If we are doing a truncate, we will send the mtime and ctime updates
1294  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1295  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1296  * at the same time.
1297  */
1298 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1299 {
1300         struct ll_inode_info *lli = ll_i2info(inode);
1301         struct lov_stripe_md *lsm = lli->lli_smd;
1302         struct md_op_data *op_data = NULL;
1303         struct md_open_data *mod = NULL;
1304         int ia_valid = attr->ia_valid;
1305         int rc = 0, rc1 = 0;
1306         ENTRY;
1307
1308         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1309                attr->ia_valid);
1310         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1311
1312         if (ia_valid & ATTR_SIZE) {
1313                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1314                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1315                                attr->ia_size, ll_file_maxbytes(inode));
1316                         RETURN(-EFBIG);
1317                 }
1318
1319                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1320         }
1321
1322         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1323         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1324                 if (current->fsuid != inode->i_uid &&
1325                     !cfs_capable(CFS_CAP_FOWNER))
1326                         RETURN(-EPERM);
1327         }
1328
1329         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1330         if (attr->ia_valid & ATTR_CTIME) {
1331                 attr->ia_ctime = CURRENT_TIME;
1332                 attr->ia_valid |= ATTR_CTIME_SET;
1333         }
1334         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1335                 attr->ia_atime = CURRENT_TIME;
1336                 attr->ia_valid |= ATTR_ATIME_SET;
1337         }
1338         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1339                 attr->ia_mtime = CURRENT_TIME;
1340                 attr->ia_valid |= ATTR_MTIME_SET;
1341         }
1342         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1343                 /* To avoid stale mtime on mds, obtain it from ost and send
1344                    to mds. */
1345                 rc = cl_glimpse_size(inode);
1346                 if (rc)
1347                         RETURN(rc);
1348
1349                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1350                 attr->ia_mtime = inode->i_mtime;
1351         }
1352
1353         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1354                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1355                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1356                        cfs_time_current_sec());
1357
1358         /* NB: ATTR_SIZE will only be set after this point if the size
1359          * resides on the MDS, ie, this file has no objects. */
1360         if (lsm)
1361                 attr->ia_valid &= ~ATTR_SIZE;
1362
1363         /* We always do an MDS RPC, even if we're only changing the size;
1364          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1365
1366         OBD_ALLOC_PTR(op_data);
1367         if (op_data == NULL)
1368                 RETURN(-ENOMEM);
1369
1370         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1371
1372         /* Open epoch for truncate. */
1373         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1374             (ia_valid & ATTR_SIZE))
1375                 op_data->op_flags = MF_EPOCH_OPEN;
1376
1377         rc = ll_md_setattr(inode, op_data, &mod);
1378         if (rc)
1379                 GOTO(out, rc);
1380
1381         if (op_data->op_ioepoch)
1382                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1383                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1384
1385         if (!lsm || !S_ISREG(inode->i_mode)) {
1386                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1387                 GOTO(out, rc = 0);
1388         }
1389
1390         if (ia_valid & ATTR_SIZE)
1391                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1392         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1393                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1394                        inode->i_ino, LTIME_S(attr->ia_mtime));
1395                 rc = ll_setattr_ost(inode);
1396         }
1397         EXIT;
1398 out:
1399         if (op_data) {
1400                 if (op_data->op_ioepoch)
1401                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1402                 ll_finish_md_op_data(op_data);
1403         }
1404         return rc ? rc : rc1;
1405 }
1406
1407 int ll_setattr(struct dentry *de, struct iattr *attr)
1408 {
1409         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1410             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1411                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1412
1413         if ((de->d_inode->i_mode & S_ISUID) &&
1414             !(attr->ia_mode & S_ISUID) &&
1415             !(attr->ia_valid & ATTR_KILL_SUID))
1416                 attr->ia_valid |= ATTR_KILL_SUID;
1417
1418         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1419             !(attr->ia_mode & S_ISGID) &&
1420             !(attr->ia_valid & ATTR_KILL_SGID))
1421                 attr->ia_valid |= ATTR_KILL_SGID;
1422
1423         return ll_setattr_raw(de->d_inode, attr);
1424 }
1425
1426 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1427                        __u64 max_age, __u32 flags)
1428 {
1429         struct ll_sb_info *sbi = ll_s2sbi(sb);
1430         struct obd_statfs obd_osfs;
1431         int rc;
1432         ENTRY;
1433
1434         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1435         if (rc) {
1436                 CERROR("md_statfs fails: rc = %d\n", rc);
1437                 RETURN(rc);
1438         }
1439
1440         osfs->os_type = sb->s_magic;
1441
1442         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1443                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1444
1445         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1446                               &obd_osfs, max_age, flags);
1447         if (rc) {
1448                 CERROR("obd_statfs fails: rc = %d\n", rc);
1449                 RETURN(rc);
1450         }
1451
1452         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1453                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1454                obd_osfs.os_files);
1455
1456         osfs->os_bsize = obd_osfs.os_bsize;
1457         osfs->os_blocks = obd_osfs.os_blocks;
1458         osfs->os_bfree = obd_osfs.os_bfree;
1459         osfs->os_bavail = obd_osfs.os_bavail;
1460
1461         /* If we don't have as many objects free on the OST as inodes
1462          * on the MDS, we reduce the total number of inodes to
1463          * compensate, so that the "inodes in use" number is correct.
1464          */
1465         if (obd_osfs.os_ffree < osfs->os_ffree) {
1466                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1467                         obd_osfs.os_ffree;
1468                 osfs->os_ffree = obd_osfs.os_ffree;
1469         }
1470
1471         RETURN(rc);
1472 }
1473 #ifndef HAVE_STATFS_DENTRY_PARAM
1474 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1475 {
1476 #else
1477 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1478 {
1479         struct super_block *sb = de->d_sb;
1480 #endif
1481         struct obd_statfs osfs;
1482         int rc;
1483
1484         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1485         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1486
1487         /* For now we will always get up-to-date statfs values, but in the
1488          * future we may allow some amount of caching on the client (e.g.
1489          * from QOS or lprocfs updates). */
1490         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1491         if (rc)
1492                 return rc;
1493
1494         statfs_unpack(sfs, &osfs);
1495
1496         /* We need to downshift for all 32-bit kernels, because we can't
1497          * tell if the kernel is being called via sys_statfs64() or not.
1498          * Stop before overflowing f_bsize - in which case it is better
1499          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1500         if (sizeof(long) < 8) {
1501                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1502                         sfs->f_bsize <<= 1;
1503
1504                         osfs.os_blocks >>= 1;
1505                         osfs.os_bfree >>= 1;
1506                         osfs.os_bavail >>= 1;
1507                 }
1508         }
1509
1510         sfs->f_blocks = osfs.os_blocks;
1511         sfs->f_bfree = osfs.os_bfree;
1512         sfs->f_bavail = osfs.os_bavail;
1513
1514         return 0;
1515 }
1516
1517 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1518 {
1519         struct ll_inode_info *lli;
1520         struct lov_stripe_md *lsm;
1521
1522         lli = ll_i2info(inode);
1523         LASSERT(lli->lli_size_sem_owner != current);
1524         down(&lli->lli_size_sem);
1525         LASSERT(lli->lli_size_sem_owner == NULL);
1526         lli->lli_size_sem_owner = current;
1527         lsm = lli->lli_smd;
1528         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1529                  lsm, lock_lsm);
1530         if (lock_lsm)
1531                 lov_stripe_lock(lsm);
1532 }
1533
1534 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1535 {
1536         struct ll_inode_info *lli;
1537         struct lov_stripe_md *lsm;
1538
1539         lli = ll_i2info(inode);
1540         lsm = lli->lli_smd;
1541         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1542                  lsm, unlock_lsm);
1543         if (unlock_lsm)
1544                 lov_stripe_unlock(lsm);
1545         LASSERT(lli->lli_size_sem_owner == current);
1546         lli->lli_size_sem_owner = NULL;
1547         up(&lli->lli_size_sem);
1548 }
1549
1550 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1551 {
1552         struct ll_inode_info *lli = ll_i2info(inode);
1553
1554         dump_lsm(D_INODE, lsm);
1555         dump_lsm(D_INODE, lli->lli_smd);
1556         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1557                  "lsm must be joined lsm %p\n", lsm);
1558         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1559         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1560                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1561         lli->lli_smd = lsm;
1562         lli->lli_maxbytes = lsm->lsm_maxbytes;
1563         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1564                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1565 }
1566
1567 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1568 {
1569         struct ll_inode_info *lli = ll_i2info(inode);
1570         struct mdt_body *body = md->body;
1571         struct lov_stripe_md *lsm = md->lsm;
1572         struct ll_sb_info *sbi = ll_i2sbi(inode);
1573
1574         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1575         if (lsm != NULL) {
1576                 if (lli->lli_smd == NULL) {
1577                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1578                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1579                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1580                                 dump_lsm(D_ERROR, lsm);
1581                                 LBUG();
1582                         }
1583                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1584                                lsm, inode->i_ino, inode->i_generation, inode);
1585                         cl_inode_init(inode, md);
1586                         /* ll_inode_size_lock() requires it is only
1587                          * called with lli_smd != NULL or lock_lsm == 0
1588                          *  or we can race between lock/unlock.
1589                          *  bug 9547 */
1590                         lli->lli_smd = lsm;
1591                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1592                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1593                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1594                 } else {
1595                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1596                              lli->lli_smd->lsm_stripe_count ==
1597                                         lsm->lsm_stripe_count) {
1598                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1599                                         CERROR("lsm mismatch for inode %ld\n",
1600                                                 inode->i_ino);
1601                                         CERROR("lli_smd:\n");
1602                                         dump_lsm(D_ERROR, lli->lli_smd);
1603                                         CERROR("lsm:\n");
1604                                         dump_lsm(D_ERROR, lsm);
1605                                         LBUG();
1606                                 }
1607                         } else {
1608                                 cl_inode_init(inode, md);
1609                                 ll_replace_lsm(inode, lsm);
1610                         }
1611                 }
1612                 if (lli->lli_smd != lsm)
1613                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1614         }
1615
1616         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1617                 if (body->valid & OBD_MD_FLRMTPERM)
1618                         ll_update_remote_perm(inode, md->remote_perm);
1619         }
1620 #ifdef CONFIG_FS_POSIX_ACL
1621         else if (body->valid & OBD_MD_FLACL) {
1622                 spin_lock(&lli->lli_lock);
1623                 if (lli->lli_posix_acl)
1624                         posix_acl_release(lli->lli_posix_acl);
1625                 lli->lli_posix_acl = md->posix_acl;
1626                 spin_unlock(&lli->lli_lock);
1627         }
1628 #endif
1629         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1630
1631         if (body->valid & OBD_MD_FLATIME &&
1632             body->atime > LTIME_S(inode->i_atime))
1633                 LTIME_S(inode->i_atime) = body->atime;
1634
1635         /* mtime is always updated with ctime, but can be set in past.
1636            As write and utime(2) may happen within 1 second, and utime's
1637            mtime has a priority over write's one, so take mtime from mds
1638            for the same ctimes. */
1639         if (body->valid & OBD_MD_FLCTIME &&
1640             body->ctime >= LTIME_S(inode->i_ctime)) {
1641                 LTIME_S(inode->i_ctime) = body->ctime;
1642                 if (body->valid & OBD_MD_FLMTIME) {
1643                         CDEBUG(D_INODE, "setting ino %lu mtime "
1644                                "from %lu to "LPU64"\n", inode->i_ino,
1645                                LTIME_S(inode->i_mtime), body->mtime);
1646                         LTIME_S(inode->i_mtime) = body->mtime;
1647                 }
1648         }
1649         if (body->valid & OBD_MD_FLMODE)
1650                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1651         if (body->valid & OBD_MD_FLTYPE)
1652                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1653         LASSERT(inode->i_mode != 0);
1654         if (S_ISREG(inode->i_mode)) {
1655                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1656         } else {
1657                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1658         }
1659 #ifdef HAVE_INODE_BLKSIZE
1660         inode->i_blksize = 1<<inode->i_blkbits;
1661 #endif
1662         if (body->valid & OBD_MD_FLUID)
1663                 inode->i_uid = body->uid;
1664         if (body->valid & OBD_MD_FLGID)
1665                 inode->i_gid = body->gid;
1666         if (body->valid & OBD_MD_FLFLAGS)
1667                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1668         if (body->valid & OBD_MD_FLNLINK)
1669                 inode->i_nlink = body->nlink;
1670         if (body->valid & OBD_MD_FLRDEV)
1671                 inode->i_rdev = old_decode_dev(body->rdev);
1672
1673         if (body->valid & OBD_MD_FLID) {
1674                 /* FID shouldn't be changed! */
1675                 if (fid_is_sane(&lli->lli_fid)) {
1676                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1677                                  "Trying to change FID "DFID
1678                                  " to the "DFID", inode %lu/%u(%p)\n",
1679                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1680                                  inode->i_ino, inode->i_generation, inode);
1681                 } else
1682                         lli->lli_fid = body->fid1;
1683         }
1684
1685         LASSERT(fid_seq(&lli->lli_fid) != 0);
1686
1687         if (body->valid & OBD_MD_FLSIZE) {
1688                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1689                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1690                         struct lustre_handle lockh;
1691                         ldlm_mode_t mode;
1692
1693                         /* As it is possible a blocking ast has been processed
1694                          * by this time, we need to check there is an UPDATE
1695                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1696                          * it. */
1697                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1698                                                &lockh);
1699                         if (mode) {
1700                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1701                                                       LLIF_EPOCH_PENDING |
1702                                                       LLIF_SOM_DIRTY)) {
1703                                         CERROR("ino %lu flags %lu still has "
1704                                                "size authority! do not trust "
1705                                                "the size got from MDS\n",
1706                                                inode->i_ino, lli->lli_flags);
1707                                 } else {
1708                                         /* Use old size assignment to avoid
1709                                          * deadlock bz14138 & bz14326 */
1710                                         inode->i_size = body->size;
1711                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1712                                 }
1713                                 ldlm_lock_decref(&lockh, mode);
1714                         }
1715                 } else {
1716                         /* Use old size assignment to avoid
1717                          * deadlock bz14138 & bz14326 */
1718                         inode->i_size = body->size;
1719                 }
1720
1721                 if (body->valid & OBD_MD_FLBLOCKS)
1722                         inode->i_blocks = body->blocks;
1723         }
1724
1725         if (body->valid & OBD_MD_FLMDSCAPA) {
1726                 LASSERT(md->mds_capa);
1727                 ll_add_capa(inode, md->mds_capa);
1728         }
1729         if (body->valid & OBD_MD_FLOSSCAPA) {
1730                 LASSERT(md->oss_capa);
1731                 ll_add_capa(inode, md->oss_capa);
1732         }
1733 }
1734
1735 static struct backing_dev_info ll_backing_dev_info = {
1736         .ra_pages       = 0,    /* No readahead */
1737 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1738         .capabilities   = 0,    /* Does contribute to dirty memory */
1739 #else
1740         .memory_backed  = 0,    /* Does contribute to dirty memory */
1741 #endif
1742 };
1743
1744 void ll_read_inode2(struct inode *inode, void *opaque)
1745 {
1746         struct lustre_md *md = opaque;
1747         struct ll_inode_info *lli = ll_i2info(inode);
1748         ENTRY;
1749
1750         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1751                inode->i_ino, inode->i_generation, inode);
1752
1753         ll_lli_init(lli);
1754
1755         LASSERT(!lli->lli_smd);
1756
1757         /* Core attributes from the MDS first.  This is a new inode, and
1758          * the VFS doesn't zero times in the core inode so we have to do
1759          * it ourselves.  They will be overwritten by either MDS or OST
1760          * attributes - we just need to make sure they aren't newer. */
1761         LTIME_S(inode->i_mtime) = 0;
1762         LTIME_S(inode->i_atime) = 0;
1763         LTIME_S(inode->i_ctime) = 0;
1764         inode->i_rdev = 0;
1765         ll_update_inode(inode, md);
1766
1767         /* OIDEBUG(inode); */
1768
1769         if (S_ISREG(inode->i_mode)) {
1770                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1771                 inode->i_op = &ll_file_inode_operations;
1772                 inode->i_fop = sbi->ll_fop;
1773                 inode->i_mapping->a_ops = &ll_aops;
1774                 EXIT;
1775         } else if (S_ISDIR(inode->i_mode)) {
1776                 inode->i_op = &ll_dir_inode_operations;
1777                 inode->i_fop = &ll_dir_operations;
1778                 inode->i_mapping->a_ops = &ll_dir_aops;
1779                 EXIT;
1780         } else if (S_ISLNK(inode->i_mode)) {
1781                 inode->i_op = &ll_fast_symlink_inode_operations;
1782                 EXIT;
1783         } else {
1784                 inode->i_op = &ll_special_inode_operations;
1785
1786                 init_special_inode(inode, inode->i_mode,
1787                                    kdev_t_to_nr(inode->i_rdev));
1788
1789                 /* initializing backing dev info. */
1790                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1791
1792                 EXIT;
1793         }
1794 }
1795
1796 void ll_delete_inode(struct inode *inode)
1797 {
1798         struct ll_sb_info *sbi = ll_i2sbi(inode);
1799         int rc;
1800         ENTRY;
1801
1802         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1803         if (rc) {
1804                 CERROR("fid_delete() failed, rc %d\n", rc);
1805         }
1806         truncate_inode_pages(&inode->i_data, 0);
1807         clear_inode(inode);
1808
1809         EXIT;
1810 }
1811
1812 int ll_iocontrol(struct inode *inode, struct file *file,
1813                  unsigned int cmd, unsigned long arg)
1814 {
1815         struct ll_sb_info *sbi = ll_i2sbi(inode);
1816         struct ptlrpc_request *req = NULL;
1817         int rc, flags = 0;
1818         ENTRY;
1819
1820         switch(cmd) {
1821         case EXT3_IOC_GETFLAGS: {
1822                 struct mdt_body *body;
1823                 struct obd_capa *oc;
1824
1825                 oc = ll_mdscapa_get(inode);
1826                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1827                                 OBD_MD_FLFLAGS, 0, &req);
1828                 capa_put(oc);
1829                 if (rc) {
1830                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1831                         RETURN(-abs(rc));
1832                 }
1833
1834                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1835
1836                 flags = body->flags;
1837
1838                 ptlrpc_req_finished(req);
1839
1840                 RETURN(put_user(flags, (int *)arg));
1841         }
1842         case EXT3_IOC_SETFLAGS: {
1843                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1844                 struct obd_info oinfo = { { { 0 } } };
1845                 struct md_op_data *op_data;
1846
1847                 if (get_user(flags, (int *)arg))
1848                         RETURN(-EFAULT);
1849
1850                 oinfo.oi_md = lsm;
1851                 OBDO_ALLOC(oinfo.oi_oa);
1852                 if (!oinfo.oi_oa)
1853                         RETURN(-ENOMEM);
1854
1855                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1856                                              LUSTRE_OPC_ANY, NULL);
1857                 if (IS_ERR(op_data))
1858                         RETURN(PTR_ERR(op_data));
1859
1860                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1861                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1862                 rc = md_setattr(sbi->ll_md_exp, op_data,
1863                                 NULL, 0, NULL, 0, &req, NULL);
1864                 ll_finish_md_op_data(op_data);
1865                 ptlrpc_req_finished(req);
1866                 if (rc || lsm == NULL) {
1867                         OBDO_FREE(oinfo.oi_oa);
1868                         RETURN(rc);
1869                 }
1870
1871                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1872                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1873                 oinfo.oi_oa->o_flags = flags;
1874                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1875                                        OBD_MD_FLGROUP;
1876                 oinfo.oi_capa = ll_mdscapa_get(inode);
1877
1878                 obdo_from_inode(oinfo.oi_oa, inode,
1879                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1880                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1881                 capa_put(oinfo.oi_capa);
1882                 OBDO_FREE(oinfo.oi_oa);
1883                 if (rc) {
1884                         if (rc != -EPERM && rc != -EACCES)
1885                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1886                         RETURN(rc);
1887                 }
1888
1889                 inode->i_flags = ll_ext_to_inode_flags(flags |
1890                                                        MDS_BFLAG_EXT_FLAGS);
1891                 RETURN(0);
1892         }
1893         default:
1894                 RETURN(-ENOSYS);
1895         }
1896
1897         RETURN(0);
1898 }
1899
1900 int ll_flush_ctx(struct inode *inode)
1901 {
1902         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1903
1904         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1905
1906         obd_set_info_async(sbi->ll_md_exp,
1907                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1908                            0, NULL, NULL);
1909         obd_set_info_async(sbi->ll_dt_exp,
1910                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1911                            0, NULL, NULL);
1912         return 0;
1913 }
1914
1915 /* umount -f client means force down, don't save state */
1916 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1917 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1918 {
1919         struct super_block *sb = vfsmnt->mnt_sb;
1920 #else
1921 void ll_umount_begin(struct super_block *sb)
1922 {
1923 #endif
1924         struct lustre_sb_info *lsi = s2lsi(sb);
1925         struct ll_sb_info *sbi = ll_s2sbi(sb);
1926         struct obd_device *obd;
1927         struct obd_ioctl_data ioc_data = { 0 };
1928         ENTRY;
1929
1930 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1931         if (!(flags & MNT_FORCE)) {
1932                 EXIT;
1933                 return;
1934         }
1935 #endif
1936
1937         /* Tell the MGC we got umount -f */
1938         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1939
1940         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1941                sb->s_count, atomic_read(&sb->s_active));
1942
1943         obd = class_exp2obd(sbi->ll_md_exp);
1944         if (obd == NULL) {
1945                 CERROR("Invalid MDC connection handle "LPX64"\n",
1946                        sbi->ll_md_exp->exp_handle.h_cookie);
1947                 EXIT;
1948                 return;
1949         }
1950         obd->obd_force = 1;
1951         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1952                       &ioc_data, NULL);
1953
1954         obd = class_exp2obd(sbi->ll_dt_exp);
1955         if (obd == NULL) {
1956                 CERROR("Invalid LOV connection handle "LPX64"\n",
1957                        sbi->ll_dt_exp->exp_handle.h_cookie);
1958                 EXIT;
1959                 return;
1960         }
1961
1962         obd->obd_force = 1;
1963         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1964                       &ioc_data, NULL);
1965
1966         /* Really, we'd like to wait until there are no requests outstanding,
1967          * and then continue.  For now, we just invalidate the requests,
1968          * schedule, and hope.
1969          */
1970         schedule();
1971
1972         EXIT;
1973 }
1974
1975 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1976 {
1977         struct ll_sb_info *sbi = ll_s2sbi(sb);
1978         int err;
1979         __u32 read_only;
1980
1981         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1982                 read_only = *flags & MS_RDONLY;
1983                 err = obd_set_info_async(sbi->ll_md_exp,
1984                                          sizeof(KEY_READ_ONLY),
1985                                          KEY_READ_ONLY, sizeof(read_only),
1986                                          &read_only, NULL);
1987                 if (err) {
1988                         CERROR("Failed to change the read-only flag during "
1989                                "remount: %d\n", err);
1990                         return err;
1991                 }
1992
1993                 if (read_only)
1994                         sb->s_flags |= MS_RDONLY;
1995                 else
1996                         sb->s_flags &= ~MS_RDONLY;
1997         }
1998         return 0;
1999 }
2000
2001 int ll_prep_inode(struct inode **inode,
2002                   struct ptlrpc_request *req,
2003                   struct super_block *sb)
2004 {
2005         struct ll_sb_info *sbi = NULL;
2006         struct lustre_md md;
2007         int rc;
2008         ENTRY;
2009
2010         LASSERT(*inode || sb);
2011         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2012         prune_deathrow(sbi, 1);
2013         memset(&md, 0, sizeof(struct lustre_md));
2014
2015         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2016                               sbi->ll_md_exp, &md);
2017         if (rc)
2018                 RETURN(rc);
2019
2020         if (*inode) {
2021                 ll_update_inode(*inode, &md);
2022         } else {
2023                 LASSERT(sb != NULL);
2024
2025                 /*
2026                  * At this point server returns to client's same fid as client
2027                  * generated for creating. So using ->fid1 is okay here.
2028                  */
2029                 LASSERT(fid_is_sane(&md.body->fid1));
2030
2031                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2032                 if (*inode == NULL || IS_ERR(*inode)) {
2033                         if (md.lsm)
2034                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2035 #ifdef CONFIG_FS_POSIX_ACL
2036                         if (md.posix_acl) {
2037                                 posix_acl_release(md.posix_acl);
2038                                 md.posix_acl = NULL;
2039                         }
2040 #endif
2041                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2042                         *inode = NULL;
2043                         CERROR("new_inode -fatal: rc %d\n", rc);
2044                         GOTO(out, rc);
2045                 }
2046         }
2047
2048         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2049                          ll_i2info(*inode)->lli_smd);
2050 out:
2051         md_free_lustre_md(sbi->ll_md_exp, &md);
2052         RETURN(rc);
2053 }
2054
2055 int ll_obd_statfs(struct inode *inode, void *arg)
2056 {
2057         struct ll_sb_info *sbi = NULL;
2058         struct obd_export *exp;
2059         char *buf = NULL;
2060         struct obd_ioctl_data *data = NULL;
2061         __u32 type;
2062         int len = 0, rc;
2063
2064         if (!inode || !(sbi = ll_i2sbi(inode)))
2065                 GOTO(out_statfs, rc = -EINVAL);
2066
2067         rc = obd_ioctl_getdata(&buf, &len, arg);
2068         if (rc)
2069                 GOTO(out_statfs, rc);
2070
2071         data = (void*)buf;
2072         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2073             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2074                 GOTO(out_statfs, rc = -EINVAL);
2075
2076         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2077         if (type == LL_STATFS_MDC)
2078                 exp = sbi->ll_md_exp;
2079         else if (type == LL_STATFS_LOV)
2080                 exp = sbi->ll_dt_exp;
2081         else
2082                 GOTO(out_statfs, rc = -ENODEV);
2083
2084         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2085         if (rc)
2086                 GOTO(out_statfs, rc);
2087 out_statfs:
2088         if (buf)
2089                 obd_ioctl_freedata(buf, len);
2090         return rc;
2091 }
2092
2093 int ll_process_config(struct lustre_cfg *lcfg)
2094 {
2095         char *ptr;
2096         void *sb;
2097         struct lprocfs_static_vars lvars;
2098         unsigned long x;
2099         int rc = 0;
2100
2101         lprocfs_llite_init_vars(&lvars);
2102
2103         /* The instance name contains the sb: lustre-client-aacfe000 */
2104         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2105         if (!ptr || !*(++ptr))
2106                 return -EINVAL;
2107         if (sscanf(ptr, "%lx", &x) != 1)
2108                 return -EINVAL;
2109         sb = (void *)x;
2110         /* This better be a real Lustre superblock! */
2111         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2112
2113         /* Note we have not called client_common_fill_super yet, so
2114            proc fns must be able to handle that! */
2115         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2116                                       lcfg, sb);
2117         return(rc);
2118 }
2119
2120 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2121 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2122                                        struct inode *i1, struct inode *i2,
2123                                        const char *name, int namelen,
2124                                        int mode, __u32 opc, void *data)
2125 {
2126         LASSERT(i1 != NULL);
2127
2128         if (namelen > ll_i2sbi(i1)->ll_namelen)
2129                 return ERR_PTR(-ENAMETOOLONG);
2130
2131         if (op_data == NULL)
2132                 OBD_ALLOC_PTR(op_data);
2133
2134         if (op_data == NULL)
2135                 return ERR_PTR(-ENOMEM);
2136
2137         ll_i2gids(op_data->op_suppgids, i1, i2);
2138         op_data->op_fid1 = *ll_inode2fid(i1);
2139         op_data->op_capa1 = ll_mdscapa_get(i1);
2140
2141         if (i2) {
2142                 op_data->op_fid2 = *ll_inode2fid(i2);
2143                 op_data->op_capa2 = ll_mdscapa_get(i2);
2144         } else {
2145                 fid_zero(&op_data->op_fid2);
2146                 op_data->op_capa2 = NULL;
2147         }
2148
2149         op_data->op_name = name;
2150         op_data->op_namelen = namelen;
2151         op_data->op_mode = mode;
2152         op_data->op_mod_time = cfs_time_current_sec();
2153         op_data->op_fsuid = current->fsuid;
2154         op_data->op_fsgid = current->fsgid;
2155         op_data->op_cap = cfs_curproc_cap_pack();
2156         op_data->op_bias = MDS_CHECK_SPLIT;
2157         op_data->op_opc = opc;
2158         op_data->op_mds = 0;
2159         op_data->op_data = data;
2160
2161         return op_data;
2162 }
2163
2164 void ll_finish_md_op_data(struct md_op_data *op_data)
2165 {
2166         capa_put(op_data->op_capa1);
2167         capa_put(op_data->op_capa2);
2168         OBD_FREE_PTR(op_data);
2169 }
2170
2171 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2172 {
2173         struct ll_sb_info *sbi;
2174
2175         LASSERT((seq != NULL) && (vfs != NULL));
2176         sbi = ll_s2sbi(vfs->mnt_sb);
2177
2178         if (sbi->ll_flags & LL_SBI_NOLCK)
2179                 seq_puts(seq, ",nolock");
2180
2181         if (sbi->ll_flags & LL_SBI_FLOCK)
2182                 seq_puts(seq, ",flock");
2183
2184         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2185                 seq_puts(seq, ",localflock");
2186
2187         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2188                 seq_puts(seq, ",user_xattr");
2189
2190         if (sbi->ll_flags & LL_SBI_ACL)
2191                 seq_puts(seq, ",acl");
2192
2193         RETURN(0);
2194 }