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