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