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 CFS_LIST_HEAD(ll_super_blocks);
  63 cfs_spinlock_t ll_sb_lock = CFS_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) cfs_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         cfs_spin_lock_init(&sbi->ll_lock);
  86         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  87         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  88         cfs_spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         cfs_spin_lock(&ll_sb_lock);
 116         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         cfs_spin_unlock(&ll_sb_lock);
 118
 119 #ifdef ENABLE_LLITE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127 #ifdef HAVE_EXPORT___IGET
 128         CFS_INIT_LIST_HEAD(&sbi->ll_deathrow);
 129         cfs_spin_lock_init(&sbi->ll_deathrow_lock);
 130 #endif
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 133                                    pp_r_hist.oh_lock);
 134                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 135                                    pp_w_hist.oh_lock);
 136         }
 137
 138         /* metadata statahead is enabled by default */
 139         sbi->ll_sa_max = LL_SA_RPC_DEF;
 140         atomic_set(&sbi->ll_sa_total, 0);
 141         atomic_set(&sbi->ll_sa_wrong, 0);
 142
 143         RETURN(sbi);
 144 }
 145
 146 void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 cfs_spin_lock(&ll_sb_lock);
 153                 cfs_list_del(&sbi->ll_list);
 154                 cfs_spin_unlock(&ll_sb_lock);
 155                 OBD_FREE(sbi, sizeof(*sbi));
 156         }
 157         EXIT;
 158 }
 159
 160 static struct dentry_operations ll_d_root_ops = {
 161         .d_compare = ll_dcompare,
 162         .d_revalidate = ll_revalidate_nd,
 163 };
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_capa *oc = NULL;
 171         struct obd_statfs osfs;
 172         struct ptlrpc_request *request = NULL;
 173         struct obd_connect_data *data = NULL;
 174         struct obd_uuid *uuid;
 175         struct md_op_data *op_data;
 176         struct lustre_md lmd;
 177         obd_valid valid;
 178         int size, err, checksum;
 179         ENTRY;
 180
 181         obd = class_name2obd(md);
 182         if (!obd) {
 183                 CERROR("MD %s: not setup or attached\n", md);
 184                 RETURN(-EINVAL);
 185         }
 186
 187         OBD_ALLOC_PTR(data);
 188         if (data == NULL)
 189                 RETURN(-ENOMEM);
 190
 191         if (proc_lustre_fs_root) {
 192                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 193                                                   dt, md);
 194                 if (err < 0)
 195                         CERROR("could not register mount in /proc/fs/lustre\n");
 196         }
 197
 198         /* indicate the features supported by this client */
 199         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 200                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 201                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 202                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 203                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 204                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 205                                   OBD_CONNECT_VBR      | OBD_CONNECT_SOM;
 206
 207 #ifdef HAVE_LRU_RESIZE_SUPPORT
 208         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 209                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 210 #endif
 211 #ifdef CONFIG_FS_POSIX_ACL
 212         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 213 #endif
 214         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 215         data->ocd_version = LUSTRE_VERSION_CODE;
 216
 217         if (sb->s_flags & MS_RDONLY)
 218                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 219         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 220                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 221
 222 #ifdef HAVE_MS_FLOCK_LOCK
 223         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 224         sb->s_flags |= MS_FLOCK_LOCK;
 225 #endif
 226
 227         if (sbi->ll_flags & LL_SBI_FLOCK)
 228                 sbi->ll_fop = &ll_file_operations_flock;
 229         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 230                 sbi->ll_fop = &ll_file_operations;
 231         else
 232                 sbi->ll_fop = &ll_file_operations_noflock;
 233
 234         /* real client */
 235         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 236         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 237                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 238
 239         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 240         if (err == -EBUSY) {
 241                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 242                                    "recovery, of which this client is not a "
 243                                    "part. Please wait for recovery to complete,"
 244                                    " abort, or time out.\n", md);
 245                 GOTO(out, err);
 246         } else if (err) {
 247                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 248                 GOTO(out, err);
 249         }
 250
 251         err = obd_fid_init(sbi->ll_md_exp);
 252         if (err) {
 253                 CERROR("Can't init metadata layer FID infrastructure, "
 254                        "rc %d\n", err);
 255                 GOTO(out_md, err);
 256         }
 257
 258         err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
 259         if (err)
 260                 GOTO(out_md_fid, err);
 261
 262         size = sizeof(*data);
 263         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 264                            KEY_CONN_DATA,  &size, data, NULL);
 265         if (err) {
 266                 CERROR("Get connect data failed: %d \n", err);
 267                 GOTO(out_md, err);
 268         }
 269
 270         LASSERT(osfs.os_bsize);
 271         sb->s_blocksize = osfs.os_bsize;
 272         sb->s_blocksize_bits = log2(osfs.os_bsize);
 273         sb->s_magic = LL_SUPER_MAGIC;
 274
 275         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 276          *         retval = in_file->f_op->sendfile(...);
 277          *         if (*ppos > max)
 278          *                 retval = -EOVERFLOW;
 279          *
 280          * it will check if *ppos is greater than max. However, max equals to
 281          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 282          * has been defined as a signed long long integer in linux kernel. */
 283 #if BITS_PER_LONG == 64
 284         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 285 #else
 286         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 287 #endif
 288         sbi->ll_namelen = osfs.os_namelen;
 289         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 290
 291         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 292             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 293                 LCONSOLE_INFO("Disabling user_xattr feature because "
 294                               "it is not supported on the server\n");
 295                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 296         }
 297
 298         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 299 #ifdef MS_POSIXACL
 300                 sb->s_flags |= MS_POSIXACL;
 301 #endif
 302                 sbi->ll_flags |= LL_SBI_ACL;
 303         } else {
 304                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 305 #ifdef MS_POSIXACL
 306                 sb->s_flags &= ~MS_POSIXACL;
 307 #endif
 308                 sbi->ll_flags &= ~LL_SBI_ACL;
 309         }
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 312                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 313                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 314                         LCONSOLE_INFO("client is set as remote by default.\n");
 315                 }
 316         } else {
 317                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 318                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 319                         LCONSOLE_INFO("client claims to be remote, but server "
 320                                       "rejected, forced to be local.\n");
 321                 }
 322         }
 323
 324         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 325                 LCONSOLE_INFO("client enabled MDS capability!\n");
 326                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 327         }
 328
 329         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 330                 LCONSOLE_INFO("client enabled OSS capability!\n");
 331                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 332         }
 333
 334         obd = class_name2obd(dt);
 335         if (!obd) {
 336                 CERROR("DT %s: not setup or attached\n", dt);
 337                 GOTO(out_md_fid, err = -ENODEV);
 338         }
 339
 340         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 341                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 342                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 343                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 344                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 345                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 346                                   OBD_CONNECT_SOM;
 347
 348         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 349                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 350                  * disabled by default, because it can still be enabled on the
 351                  * fly via /proc. As a consequence, we still need to come to an
 352                  * agreement on the supported algorithms at connect time */
 353                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 354
 355                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 356                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 357                 else
 358                         /* send the list of supported checksum types */
 359                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 360         }
 361
 362 #ifdef HAVE_LRU_RESIZE_SUPPORT
 363         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 364 #endif
 365         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 366                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 367
 368         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 369                "ocd_grant: %d\n", data->ocd_connect_flags,
 370                data->ocd_version, data->ocd_grant);
 371
 372         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 373         obd->obd_upcall.onu_upcall = cl_ocd_update;
 374         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 375
 376         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 377         if (err == -EBUSY) {
 378                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 379                                    "recovery, of which this client is not a "
 380                                    "part.  Please wait for recovery to "
 381                                    "complete, abort, or time out.\n", dt);
 382                 GOTO(out_md_fid, err);
 383         } else if (err) {
 384                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 385                 GOTO(out_md_fid, err);
 386         }
 387
 388         err = obd_fid_init(sbi->ll_dt_exp);
 389         if (err) {
 390                 CERROR("Can't init data layer FID infrastructure, "
 391                        "rc %d\n", err);
 392                 GOTO(out_dt, err);
 393         }
 394
 395         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 396         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 397         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 398         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 399         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 400
 401         fid_zero(&sbi->ll_root_fid);
 402         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 403         if (err) {
 404                 CERROR("cannot mds_connect: rc = %d\n", err);
 405                 GOTO(out_lock_cn_cb, err);
 406         }
 407         if (!fid_is_sane(&sbi->ll_root_fid)) {
 408                 CERROR("Invalid root fid during mount\n");
 409                 GOTO(out_lock_cn_cb, err = -EINVAL);
 410         }
 411         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 412
 413         sb->s_op = &lustre_super_operations;
 414         sb->s_export_op = &lustre_export_operations;
 415
 416         /* make root inode
 417          * XXX: move this to after cbd setup? */
 418         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 419         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 420                 valid |= OBD_MD_FLRMTPERM;
 421         else if (sbi->ll_flags & LL_SBI_ACL)
 422                 valid |= OBD_MD_FLACL;
 423
 424         OBD_ALLOC_PTR(op_data);
 425         if (op_data == NULL)
 426                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 427
 428         op_data->op_fid1 = sbi->ll_root_fid;
 429         op_data->op_mode = 0;
 430         op_data->op_capa1 = oc;
 431         op_data->op_valid = valid;
 432
 433         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 434         if (oc)
 435                 capa_put(oc);
 436         OBD_FREE_PTR(op_data);
 437         if (err) {
 438                 CERROR("md_getattr failed for root: rc = %d\n", err);
 439                 GOTO(out_lock_cn_cb, err);
 440         }
 441         memset(&lmd, 0, sizeof(lmd));
 442         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 443                                sbi->ll_md_exp, &lmd);
 444         if (err) {
 445                 CERROR("failed to understand root inode md: rc = %d\n", err);
 446                 ptlrpc_req_finished (request);
 447                 GOTO(out_lock_cn_cb, err);
 448         }
 449
 450         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 451         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
 452         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 453         ptlrpc_req_finished(request);
 454
 455         if (root == NULL || IS_ERR(root)) {
 456                 if (lmd.lsm)
 457                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 458 #ifdef CONFIG_FS_POSIX_ACL
 459                 if (lmd.posix_acl) {
 460                         posix_acl_release(lmd.posix_acl);
 461                         lmd.posix_acl = NULL;
 462                 }
 463 #endif
 464                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 465                 root = NULL;
 466                 CERROR("lustre_lite: bad iget4 for root\n");
 467                 GOTO(out_root, err);
 468         }
 469
 470         err = ll_close_thread_start(&sbi->ll_lcq);
 471         if (err) {
 472                 CERROR("cannot start close thread: rc %d\n", err);
 473                 GOTO(out_root, err);
 474         }
 475
 476 #ifdef CONFIG_FS_POSIX_ACL
 477         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 478                 rct_init(&sbi->ll_rct);
 479                 et_init(&sbi->ll_et);
 480         }
 481 #endif
 482
 483         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 484         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 485                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 486                                  NULL);
 487         cl_sb_init(sb);
 488
 489         sb->s_root = d_alloc_root(root);
 490         if (data != NULL)
 491                 OBD_FREE(data, sizeof(*data));
 492
 493         sb->s_root->d_op = &ll_d_root_ops;
 494
 495         sbi->ll_sdev_orig = sb->s_dev;
 496
 497         /* We set sb->s_dev equal on all lustre clients in order to support
 498          * NFS export clustering.  NFSD requires that the FSID be the same
 499          * on all clients. */
 500         /* s_dev is also used in lt_compare() to compare two fs, but that is
 501          * only a node-local comparison. */
 502         uuid = obd_get_uuid(sbi->ll_md_exp);
 503         if (uuid != NULL)
 504                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 505
 506         RETURN(err);
 507 out_root:
 508         if (root)
 509                 iput(root);
 510 out_lock_cn_cb:
 511         obd_fid_fini(sbi->ll_dt_exp);
 512 out_dt:
 513         obd_disconnect(sbi->ll_dt_exp);
 514         sbi->ll_dt_exp = NULL;
 515 out_md_fid:
 516         obd_fid_fini(sbi->ll_md_exp);
 517 out_md:
 518         obd_disconnect(sbi->ll_md_exp);
 519         sbi->ll_md_exp = NULL;
 520 out:
 521         if (data != NULL)
 522                 OBD_FREE_PTR(data);
 523         lprocfs_unregister_mountpoint(sbi);
 524         return err;
 525 }
 526
 527 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 528 {
 529         int size, rc;
 530
 531         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 532         size = sizeof(int);
 533         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 534                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 535         if (rc)
 536                 CERROR("Get max mdsize error rc %d \n", rc);
 537
 538         RETURN(rc);
 539 }
 540
 541 void ll_dump_inode(struct inode *inode)
 542 {
 543         struct list_head *tmp;
 544         int dentry_count = 0;
 545
 546         LASSERT(inode != NULL);
 547
 548         list_for_each(tmp, &inode->i_dentry)
 549                 dentry_count++;
 550
 551         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 552                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 553                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 554 }
 555
 556 void lustre_dump_dentry(struct dentry *dentry, int recur)
 557 {
 558         struct list_head *tmp;
 559         int subdirs = 0;
 560
 561         LASSERT(dentry != NULL);
 562
 563         list_for_each(tmp, &dentry->d_subdirs)
 564                 subdirs++;
 565
 566         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 567                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 568                dentry->d_name.len, dentry->d_name.name,
 569                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 570                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 571                dentry->d_flags, dentry->d_fsdata, subdirs);
 572         if (dentry->d_inode != NULL)
 573                 ll_dump_inode(dentry->d_inode);
 574
 575         if (recur == 0)
 576                 return;
 577
 578        list_for_each(tmp, &dentry->d_subdirs) {
 579                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 580                 lustre_dump_dentry(d, recur - 1);
 581         }
 582 }
 583
 584 #ifdef HAVE_EXPORT___IGET
 585 static void prune_dir_dentries(struct inode *inode)
 586 {
 587         struct dentry *dentry, *prev = NULL;
 588
 589         /* due to lustre specific logic, a directory
 590          * can have few dentries - a bug from VFS POV */
 591 restart:
 592         spin_lock(&dcache_lock);
 593         if (!list_empty(&inode->i_dentry)) {
 594                 dentry = list_entry(inode->i_dentry.prev,
 595                                     struct dentry, d_alias);
 596                 /* in order to prevent infinite loops we
 597                  * break if previous dentry is busy */
 598                 if (dentry != prev) {
 599                         prev = dentry;
 600                         dget_locked(dentry);
 601                         spin_unlock(&dcache_lock);
 602
 603                         /* try to kill all child dentries */
 604                         lock_dentry(dentry);
 605                         shrink_dcache_parent(dentry);
 606                         unlock_dentry(dentry);
 607                         dput(dentry);
 608
 609                         /* now try to get rid of current dentry */
 610                         d_prune_aliases(inode);
 611                         goto restart;
 612                 }
 613         }
 614         spin_unlock(&dcache_lock);
 615 }
 616
 617 static void prune_deathrow_one(struct ll_inode_info *lli)
 618 {
 619         struct inode *inode = ll_info2i(lli);
 620
 621         /* first, try to drop any dentries - they hold a ref on the inode */
 622         if (S_ISDIR(inode->i_mode))
 623                 prune_dir_dentries(inode);
 624         else
 625                 d_prune_aliases(inode);
 626
 627
 628         /* if somebody still uses it, leave it */
 629         LASSERT(atomic_read(&inode->i_count) > 0);
 630         if (atomic_read(&inode->i_count) > 1)
 631                 goto out;
 632
 633         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 634                inode->i_ino,inode->i_generation,
 635                atomic_read(&inode->i_count));
 636
 637         /* seems nobody uses it anymore */
 638         inode->i_nlink = 0;
 639
 640 out:
 641         iput(inode);
 642         return;
 643 }
 644
 645 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 646 {
 647         struct ll_inode_info *lli;
 648         int empty;
 649
 650         do {
 651                 if (need_resched() && try)
 652                         break;
 653
 654                 if (try) {
 655                         if (!cfs_spin_trylock(&sbi->ll_deathrow_lock))
 656                                 break;
 657                 } else {
 658                         cfs_spin_lock(&sbi->ll_deathrow_lock);
 659                 }
 660
 661                 empty = 1;
 662                 lli = NULL;
 663                 if (!cfs_list_empty(&sbi->ll_deathrow)) {
 664                         lli = cfs_list_entry(sbi->ll_deathrow.next,
 665                                              struct ll_inode_info,
 666                                              lli_dead_list);
 667                         cfs_list_del_init(&lli->lli_dead_list);
 668                         if (!cfs_list_empty(&sbi->ll_deathrow))
 669                                 empty = 0;
 670                 }
 671                 cfs_spin_unlock(&sbi->ll_deathrow_lock);
 672
 673                 if (lli)
 674                         prune_deathrow_one(lli);
 675
 676         } while (empty == 0);
 677 }
 678 #else /* !HAVE_EXPORT___IGET */
 679 #define prune_deathrow(sbi, try) do {} while (0)
 680 #endif /* HAVE_EXPORT___IGET */
 681
 682 void client_common_put_super(struct super_block *sb)
 683 {
 684         struct ll_sb_info *sbi = ll_s2sbi(sb);
 685         ENTRY;
 686
 687 #ifdef CONFIG_FS_POSIX_ACL
 688         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 689                 et_fini(&sbi->ll_et);
 690                 rct_fini(&sbi->ll_rct);
 691         }
 692 #endif
 693
 694         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 695
 696         ll_close_thread_shutdown(sbi->ll_lcq);
 697
 698         cl_sb_fini(sb);
 699
 700         /* destroy inodes in deathrow */
 701         prune_deathrow(sbi, 0);
 702
 703         cfs_list_del(&sbi->ll_conn_chain);
 704
 705         obd_fid_fini(sbi->ll_dt_exp);
 706         obd_disconnect(sbi->ll_dt_exp);
 707         sbi->ll_dt_exp = NULL;
 708
 709         lprocfs_unregister_mountpoint(sbi);
 710
 711         obd_fid_fini(sbi->ll_md_exp);
 712         obd_disconnect(sbi->ll_md_exp);
 713         sbi->ll_md_exp = NULL;
 714
 715         EXIT;
 716 }
 717
 718 void ll_kill_super(struct super_block *sb)
 719 {
 720         struct ll_sb_info *sbi;
 721
 722         ENTRY;
 723
 724         /* not init sb ?*/
 725         if (!(sb->s_flags & MS_ACTIVE))
 726                 return;
 727
 728         sbi = ll_s2sbi(sb);
 729         /* we need restore s_dev from changed for clustred NFS before put_super
 730          * because new kernels have cached s_dev and change sb->s_dev in
 731          * put_super not affected real removing devices */
 732         if (sbi)
 733                 sb->s_dev = sbi->ll_sdev_orig;
 734         EXIT;
 735 }
 736
 737 char *ll_read_opt(const char *opt, char *data)
 738 {
 739         char *value;
 740         char *retval;
 741         ENTRY;
 742
 743         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 744         if (strncmp(opt, data, strlen(opt)))
 745                 RETURN(NULL);
 746         if ((value = strchr(data, '=')) == NULL)
 747                 RETURN(NULL);
 748
 749         value++;
 750         OBD_ALLOC(retval, strlen(value) + 1);
 751         if (!retval) {
 752                 CERROR("out of memory!\n");
 753                 RETURN(NULL);
 754         }
 755
 756         memcpy(retval, value, strlen(value)+1);
 757         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 758         RETURN(retval);
 759 }
 760
 761 static inline int ll_set_opt(const char *opt, char *data, int fl)
 762 {
 763         if (strncmp(opt, data, strlen(opt)) != 0)
 764                 return(0);
 765         else
 766                 return(fl);
 767 }
 768
 769 /* non-client-specific mount options are parsed in lmd_parse */
 770 static int ll_options(char *options, int *flags)
 771 {
 772         int tmp;
 773         char *s1 = options, *s2;
 774         ENTRY;
 775
 776         if (!options)
 777                 RETURN(0);
 778
 779         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 780
 781         while (*s1) {
 782                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 783                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 784                 if (tmp) {
 785                         *flags |= tmp;
 786                         goto next;
 787                 }
 788                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 789                 if (tmp) {
 790                         *flags |= tmp;
 791                         goto next;
 792                 }
 793                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 794                 if (tmp) {
 795                         *flags |= tmp;
 796                         goto next;
 797                 }
 798                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 799                 if (tmp) {
 800                         *flags &= ~tmp;
 801                         goto next;
 802                 }
 803                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 804                 if (tmp) {
 805                         *flags |= tmp;
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 809                 if (tmp) {
 810                         *flags &= ~tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 814                 if (tmp) {
 815                         /* Ignore deprecated mount option.  The client will
 816                          * always try to mount with ACL support, whether this
 817                          * is used depends on whether server supports it. */
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 821                 if (tmp) {
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829
 830                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags &= ~tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 846                 if (tmp) {
 847                         *flags &= ~tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 856                 if (tmp) {
 857                         *flags &= ~tmp;
 858                         goto next;
 859                 }
 860
 861                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 862                                    s1);
 863                 RETURN(-EINVAL);
 864
 865 next:
 866                 /* Find next opt */
 867                 s2 = strchr(s1, ',');
 868                 if (s2 == NULL)
 869                         break;
 870                 s1 = s2 + 1;
 871         }
 872         RETURN(0);
 873 }
 874
 875 void ll_lli_init(struct ll_inode_info *lli)
 876 {
 877         lli->lli_inode_magic = LLI_INODE_MAGIC;
 878         cfs_sema_init(&lli->lli_size_sem, 1);
 879         cfs_sema_init(&lli->lli_write_sem, 1);
 880         cfs_sema_init(&lli->lli_trunc_sem, 1);
 881         lli->lli_flags = 0;
 882         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 883         cfs_spin_lock_init(&lli->lli_lock);
 884         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 885         lli->lli_inode_magic = LLI_INODE_MAGIC;
 886         cfs_sema_init(&lli->lli_och_sem, 1);
 887         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 888         lli->lli_mds_exec_och = NULL;
 889         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 890         lli->lli_open_fd_exec_count = 0;
 891         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 892         lli->lli_remote_perms = NULL;
 893         lli->lli_rmtperm_utime = 0;
 894         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 895         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 896 }
 897
 898 int ll_fill_super(struct super_block *sb)
 899 {
 900         struct lustre_profile *lprof;
 901         struct lustre_sb_info *lsi = s2lsi(sb);
 902         struct ll_sb_info *sbi;
 903         char  *dt = NULL, *md = NULL;
 904         char  *profilenm = get_profile_name(sb);
 905         struct config_llog_instance cfg = {0, };
 906         char   ll_instance[sizeof(sb) * 2 + 1];
 907         int    err;
 908         ENTRY;
 909
 910         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 911
 912         cfs_module_get();
 913
 914         /* client additional sb info */
 915         lsi->lsi_llsbi = sbi = ll_init_sbi();
 916         if (!sbi) {
 917                 cfs_module_put(THIS_MODULE);
 918                 RETURN(-ENOMEM);
 919         }
 920
 921         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 922         if (err)
 923                 GOTO(out_free, err);
 924
 925         /* Generate a string unique to this super, in case some joker tries
 926            to mount the same fs at two mount points.
 927            Use the address of the super itself.*/
 928         sprintf(ll_instance, "%p", sb);
 929         cfg.cfg_instance = ll_instance;
 930         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 931
 932         /* set up client obds */
 933         err = lustre_process_log(sb, profilenm, &cfg);
 934         if (err < 0) {
 935                 CERROR("Unable to process log: %d\n", err);
 936                 GOTO(out_free, err);
 937         }
 938
 939         lprof = class_get_profile(profilenm);
 940         if (lprof == NULL) {
 941                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 942                                    " read from the MGS.  Does that filesystem "
 943                                    "exist?\n", profilenm);
 944                 GOTO(out_free, err = -EINVAL);
 945         }
 946         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 947                lprof->lp_md, lprof->lp_dt);
 948
 949         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 950                   strlen(ll_instance) + 2);
 951         if (!dt)
 952                 GOTO(out_free, err = -ENOMEM);
 953         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 954
 955         OBD_ALLOC(md, strlen(lprof->lp_md) +
 956                   strlen(ll_instance) + 2);
 957         if (!md)
 958                 GOTO(out_free, err = -ENOMEM);
 959         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 960
 961         /* connections, registrations, sb setup */
 962         err = client_common_fill_super(sb, md, dt);
 963
 964 out_free:
 965         if (md)
 966                 OBD_FREE(md, strlen(md) + 1);
 967         if (dt)
 968                 OBD_FREE(dt, strlen(dt) + 1);
 969         if (err)
 970                 ll_put_super(sb);
 971         else
 972                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 973
 974         RETURN(err);
 975 } /* ll_fill_super */
 976
 977
 978 void lu_context_keys_dump(void);
 979
 980 void ll_put_super(struct super_block *sb)
 981 {
 982         struct config_llog_instance cfg;
 983         char   ll_instance[sizeof(sb) * 2 + 1];
 984         struct obd_device *obd;
 985         struct lustre_sb_info *lsi = s2lsi(sb);
 986         struct ll_sb_info *sbi = ll_s2sbi(sb);
 987         char *profilenm = get_profile_name(sb);
 988         int force = 1, next;
 989         ENTRY;
 990
 991         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 992
 993         ll_print_capa_stat(sbi);
 994
 995         sprintf(ll_instance, "%p", sb);
 996         cfg.cfg_instance = ll_instance;
 997         lustre_end_log(sb, NULL, &cfg);
 998
 999         if (sbi->ll_md_exp) {
1000                 obd = class_exp2obd(sbi->ll_md_exp);
1001                 if (obd)
1002                         force = obd->obd_force;
1003         }
1004
1005         /* We need to set force before the lov_disconnect in
1006            lustre_common_put_super, since l_d cleans up osc's as well. */
1007         if (force) {
1008                 next = 0;
1009                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1010                                                      &next)) != NULL) {
1011                         obd->obd_force = force;
1012                 }
1013         }
1014
1015         if (sbi->ll_lcq) {
1016                 /* Only if client_common_fill_super succeeded */
1017                 client_common_put_super(sb);
1018         }
1019
1020         next = 0;
1021         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1022                 class_manual_cleanup(obd);
1023         }
1024
1025         if (profilenm)
1026                 class_del_profile(profilenm);
1027
1028         ll_free_sbi(sb);
1029         lsi->lsi_llsbi = NULL;
1030
1031         lustre_common_put_super(sb);
1032
1033         cl_env_cache_purge(~0);
1034
1035         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1036
1037         cfs_module_put(THIS_MODULE);
1038
1039         EXIT;
1040 } /* client_put_super */
1041
1042 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1043 {
1044         struct inode *inode = NULL;
1045         /* NOTE: we depend on atomic igrab() -bzzz */
1046         lock_res_and_lock(lock);
1047         if (lock->l_ast_data) {
1048                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1049                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1050                         inode = igrab(lock->l_ast_data);
1051                 } else {
1052                         inode = lock->l_ast_data;
1053                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1054                                                 D_INFO : D_WARNING,
1055                                         lock, __FILE__, __func__, __LINE__,
1056                                         "l_ast_data %p is bogus: magic %08x",
1057                                         lock->l_ast_data, lli->lli_inode_magic);
1058                         inode = NULL;
1059                 }
1060         }
1061         unlock_res_and_lock(lock);
1062         return inode;
1063 }
1064
1065 static int null_if_equal(struct ldlm_lock *lock, void *data)
1066 {
1067         if (data == lock->l_ast_data) {
1068                 lock->l_ast_data = NULL;
1069
1070                 if (lock->l_req_mode != lock->l_granted_mode)
1071                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1072         }
1073
1074         return LDLM_ITER_CONTINUE;
1075 }
1076
1077 void ll_clear_inode(struct inode *inode)
1078 {
1079         struct ll_inode_info *lli = ll_i2info(inode);
1080         struct ll_sb_info *sbi = ll_i2sbi(inode);
1081         ENTRY;
1082
1083         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1084                inode->i_generation, inode);
1085
1086         if (S_ISDIR(inode->i_mode)) {
1087                 /* these should have been cleared in ll_file_release */
1088                 LASSERT(lli->lli_sai == NULL);
1089                 LASSERT(lli->lli_opendir_key == NULL);
1090                 LASSERT(lli->lli_opendir_pid == 0);
1091         }
1092
1093         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1094         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1095                          null_if_equal, inode);
1096
1097         LASSERT(!lli->lli_open_fd_write_count);
1098         LASSERT(!lli->lli_open_fd_read_count);
1099         LASSERT(!lli->lli_open_fd_exec_count);
1100
1101         if (lli->lli_mds_write_och)
1102                 ll_md_real_close(inode, FMODE_WRITE);
1103         if (lli->lli_mds_exec_och)
1104                 ll_md_real_close(inode, FMODE_EXEC);
1105         if (lli->lli_mds_read_och)
1106                 ll_md_real_close(inode, FMODE_READ);
1107
1108         if (lli->lli_symlink_name) {
1109                 OBD_FREE(lli->lli_symlink_name,
1110                          strlen(lli->lli_symlink_name) + 1);
1111                 lli->lli_symlink_name = NULL;
1112         }
1113
1114         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1115                 LASSERT(lli->lli_posix_acl == NULL);
1116                 if (lli->lli_remote_perms) {
1117                         free_rmtperm_hash(lli->lli_remote_perms);
1118                         lli->lli_remote_perms = NULL;
1119                 }
1120         }
1121 #ifdef CONFIG_FS_POSIX_ACL
1122         else if (lli->lli_posix_acl) {
1123                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1124                 LASSERT(lli->lli_remote_perms == NULL);
1125                 posix_acl_release(lli->lli_posix_acl);
1126                 lli->lli_posix_acl = NULL;
1127         }
1128 #endif
1129         lli->lli_inode_magic = LLI_INODE_DEAD;
1130
1131 #ifdef HAVE_EXPORT___IGET
1132         cfs_spin_lock(&sbi->ll_deathrow_lock);
1133         cfs_list_del_init(&lli->lli_dead_list);
1134         cfs_spin_unlock(&sbi->ll_deathrow_lock);
1135 #endif
1136         ll_clear_inode_capas(inode);
1137         /*
1138          * XXX This has to be done before lsm is freed below, because
1139          * cl_object still uses inode lsm.
1140          */
1141         cl_inode_fini(inode);
1142
1143         if (lli->lli_smd) {
1144                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1145                 lli->lli_smd = NULL;
1146         }
1147
1148
1149         EXIT;
1150 }
1151
1152 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1153                   struct md_open_data **mod)
1154 {
1155         struct lustre_md md;
1156         struct ll_sb_info *sbi = ll_i2sbi(inode);
1157         struct ptlrpc_request *request = NULL;
1158         int rc;
1159         ENTRY;
1160
1161         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1162                                      LUSTRE_OPC_ANY, NULL);
1163         if (IS_ERR(op_data))
1164                 RETURN(PTR_ERR(op_data));
1165
1166         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1167                         &request, mod);
1168         if (rc) {
1169                 ptlrpc_req_finished(request);
1170                 if (rc == -ENOENT) {
1171                         inode->i_nlink = 0;
1172                         /* Unlinked special device node? Or just a race?
1173                          * Pretend we done everything. */
1174                         if (!S_ISREG(inode->i_mode) &&
1175                             !S_ISDIR(inode->i_mode))
1176                                 rc = inode_setattr(inode, &op_data->op_attr);
1177                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1178                         CERROR("md_setattr fails: rc = %d\n", rc);
1179                 }
1180                 RETURN(rc);
1181         }
1182
1183         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1184                               sbi->ll_md_exp, &md);
1185         if (rc) {
1186                 ptlrpc_req_finished(request);
1187                 RETURN(rc);
1188         }
1189
1190         /* We call inode_setattr to adjust timestamps.
1191          * If there is at least some data in file, we cleared ATTR_SIZE
1192          * above to avoid invoking vmtruncate, otherwise it is important
1193          * to call vmtruncate in inode_setattr to update inode->i_size
1194          * (bug 6196) */
1195         rc = inode_setattr(inode, &op_data->op_attr);
1196
1197         /* Extract epoch data if obtained. */
1198         op_data->op_handle = md.body->handle;
1199         op_data->op_ioepoch = md.body->ioepoch;
1200
1201         ll_update_inode(inode, &md);
1202         ptlrpc_req_finished(request);
1203
1204         RETURN(rc);
1205 }
1206
1207 /* Close IO epoch and send Size-on-MDS attribute update. */
1208 static int ll_setattr_done_writing(struct inode *inode,
1209                                    struct md_op_data *op_data,
1210                                    struct md_open_data *mod)
1211 {
1212         struct ll_inode_info *lli = ll_i2info(inode);
1213         int rc = 0;
1214         ENTRY;
1215
1216         LASSERT(op_data != NULL);
1217         if (!S_ISREG(inode->i_mode))
1218                 RETURN(0);
1219
1220         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1221                op_data->op_ioepoch, PFID(&lli->lli_fid));
1222
1223         op_data->op_flags = MF_EPOCH_CLOSE;
1224         ll_done_writing_attr(inode, op_data);
1225         ll_pack_inode2opdata(inode, op_data, NULL);
1226
1227         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1228         if (rc == -EAGAIN) {
1229                 /* MDS has instructed us to obtain Size-on-MDS attribute
1230                  * from OSTs and send setattr to back to MDS. */
1231                 rc = ll_som_update(inode, op_data);
1232         } else if (rc) {
1233                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1234                        inode->i_ino, rc);
1235         }
1236         RETURN(rc);
1237 }
1238
1239 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1240 {
1241         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1242         int rc;
1243
1244         rc = cl_setattr_do_truncate(inode, size, capa);
1245         ll_truncate_free_capa(capa);
1246         return rc;
1247 }
1248
1249 static int ll_setattr_ost(struct inode *inode)
1250 {
1251         struct obd_capa *capa = ll_mdscapa_get(inode);
1252         int rc;
1253
1254         rc = cl_setattr_ost(inode, capa);
1255         capa_put(capa);
1256
1257         return rc;
1258 }
1259
1260 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1261  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1262  * keep these values until such a time that objects are allocated for it.
1263  * We do the MDS operations first, as it is checking permissions for us.
1264  * We don't to the MDS RPC if there is nothing that we want to store there,
1265  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1266  * going to do an RPC anyways.
1267  *
1268  * If we are doing a truncate, we will send the mtime and ctime updates
1269  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1270  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1271  * at the same time.
1272  */
1273 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1274 {
1275         struct ll_inode_info *lli = ll_i2info(inode);
1276         struct lov_stripe_md *lsm = lli->lli_smd;
1277         struct md_op_data *op_data = NULL;
1278         struct md_open_data *mod = NULL;
1279         int ia_valid = attr->ia_valid;
1280         int rc = 0, rc1 = 0;
1281         ENTRY;
1282
1283         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1284                attr->ia_valid);
1285         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1286
1287         if (ia_valid & ATTR_SIZE) {
1288                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1289                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1290                                attr->ia_size, ll_file_maxbytes(inode));
1291                         RETURN(-EFBIG);
1292                 }
1293
1294                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1295         }
1296
1297         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1298         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1299                 if (cfs_curproc_fsuid() != inode->i_uid &&
1300                     !cfs_capable(CFS_CAP_FOWNER))
1301                         RETURN(-EPERM);
1302         }
1303
1304         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1305         if (attr->ia_valid & ATTR_CTIME) {
1306                 attr->ia_ctime = CFS_CURRENT_TIME;
1307                 attr->ia_valid |= ATTR_CTIME_SET;
1308         }
1309         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1310                 attr->ia_atime = CFS_CURRENT_TIME;
1311                 attr->ia_valid |= ATTR_ATIME_SET;
1312         }
1313         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1314                 attr->ia_mtime = CFS_CURRENT_TIME;
1315                 attr->ia_valid |= ATTR_MTIME_SET;
1316         }
1317         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1318                 /* To avoid stale mtime on mds, obtain it from ost and send
1319                    to mds. */
1320                 rc = cl_glimpse_size(inode);
1321                 if (rc)
1322                         RETURN(rc);
1323
1324                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1325                 attr->ia_mtime = inode->i_mtime;
1326         }
1327
1328         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1329                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1330                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1331                        cfs_time_current_sec());
1332
1333         /* NB: ATTR_SIZE will only be set after this point if the size
1334          * resides on the MDS, ie, this file has no objects. */
1335         if (lsm)
1336                 attr->ia_valid &= ~ATTR_SIZE;
1337
1338         /* We always do an MDS RPC, even if we're only changing the size;
1339          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1340
1341         OBD_ALLOC_PTR(op_data);
1342         if (op_data == NULL)
1343                 RETURN(-ENOMEM);
1344
1345         UNLOCK_INODE_MUTEX(inode);
1346         if (ia_valid & ATTR_SIZE)
1347                 UP_WRITE_I_ALLOC_SEM(inode);
1348         cfs_down(&lli->lli_trunc_sem);
1349         LOCK_INODE_MUTEX(inode);
1350         if (ia_valid & ATTR_SIZE)
1351                 DOWN_WRITE_I_ALLOC_SEM(inode);
1352
1353         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1354
1355         /* Open epoch for truncate. */
1356         if (exp_connect_som(ll_i2mdexp(inode)) && (ia_valid & ATTR_SIZE))
1357                 op_data->op_flags = MF_EPOCH_OPEN;
1358
1359         rc = ll_md_setattr(inode, op_data, &mod);
1360         if (rc)
1361                 GOTO(out, rc);
1362
1363         ll_ioepoch_open(lli, op_data->op_ioepoch);
1364         if (!lsm || !S_ISREG(inode->i_mode)) {
1365                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1366                 GOTO(out, rc = 0);
1367         }
1368
1369         if (ia_valid & ATTR_SIZE)
1370                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1371         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1372                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1373                        inode->i_ino, LTIME_S(attr->ia_mtime));
1374                 rc = ll_setattr_ost(inode);
1375         }
1376         EXIT;
1377 out:
1378         if (op_data) {
1379                 if (op_data->op_ioepoch)
1380                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1381                 ll_finish_md_op_data(op_data);
1382         }
1383         cfs_up(&lli->lli_trunc_sem);
1384         return rc ? rc : rc1;
1385 }
1386
1387 int ll_setattr(struct dentry *de, struct iattr *attr)
1388 {
1389         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1390             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1391                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1392
1393         if ((de->d_inode->i_mode & S_ISUID) &&
1394             !(attr->ia_mode & S_ISUID) &&
1395             !(attr->ia_valid & ATTR_KILL_SUID))
1396                 attr->ia_valid |= ATTR_KILL_SUID;
1397
1398         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1399             !(attr->ia_mode & S_ISGID) &&
1400             !(attr->ia_valid & ATTR_KILL_SGID))
1401                 attr->ia_valid |= ATTR_KILL_SGID;
1402
1403         return ll_setattr_raw(de->d_inode, attr);
1404 }
1405
1406 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1407                        __u64 max_age, __u32 flags)
1408 {
1409         struct ll_sb_info *sbi = ll_s2sbi(sb);
1410         struct obd_statfs obd_osfs;
1411         int rc;
1412         ENTRY;
1413
1414         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1415         if (rc) {
1416                 CERROR("md_statfs fails: rc = %d\n", rc);
1417                 RETURN(rc);
1418         }
1419
1420         osfs->os_type = sb->s_magic;
1421
1422         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1423                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1424
1425         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1426                 flags |= OBD_STATFS_NODELAY;
1427
1428         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1429                               &obd_osfs, max_age, flags);
1430         if (rc) {
1431                 CERROR("obd_statfs fails: rc = %d\n", rc);
1432                 RETURN(rc);
1433         }
1434
1435         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1436                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1437                obd_osfs.os_files);
1438
1439         osfs->os_bsize = obd_osfs.os_bsize;
1440         osfs->os_blocks = obd_osfs.os_blocks;
1441         osfs->os_bfree = obd_osfs.os_bfree;
1442         osfs->os_bavail = obd_osfs.os_bavail;
1443
1444         /* If we don't have as many objects free on the OST as inodes
1445          * on the MDS, we reduce the total number of inodes to
1446          * compensate, so that the "inodes in use" number is correct.
1447          */
1448         if (obd_osfs.os_ffree < osfs->os_ffree) {
1449                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1450                         obd_osfs.os_ffree;
1451                 osfs->os_ffree = obd_osfs.os_ffree;
1452         }
1453
1454         RETURN(rc);
1455 }
1456 #ifndef HAVE_STATFS_DENTRY_PARAM
1457 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1458 {
1459 #else
1460 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1461 {
1462         struct super_block *sb = de->d_sb;
1463 #endif
1464         struct obd_statfs osfs;
1465         int rc;
1466
1467         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1468         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1469
1470         /* For now we will always get up-to-date statfs values, but in the
1471          * future we may allow some amount of caching on the client (e.g.
1472          * from QOS or lprocfs updates). */
1473         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1474         if (rc)
1475                 return rc;
1476
1477         statfs_unpack(sfs, &osfs);
1478
1479         /* We need to downshift for all 32-bit kernels, because we can't
1480          * tell if the kernel is being called via sys_statfs64() or not.
1481          * Stop before overflowing f_bsize - in which case it is better
1482          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1483         if (sizeof(long) < 8) {
1484                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1485                         sfs->f_bsize <<= 1;
1486
1487                         osfs.os_blocks >>= 1;
1488                         osfs.os_bfree >>= 1;
1489                         osfs.os_bavail >>= 1;
1490                 }
1491         }
1492
1493         sfs->f_blocks = osfs.os_blocks;
1494         sfs->f_bfree = osfs.os_bfree;
1495         sfs->f_bavail = osfs.os_bavail;
1496
1497         return 0;
1498 }
1499
1500 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1501 {
1502         struct ll_inode_info *lli;
1503         struct lov_stripe_md *lsm;
1504
1505         lli = ll_i2info(inode);
1506         LASSERT(lli->lli_size_sem_owner != current);
1507         cfs_down(&lli->lli_size_sem);
1508         LASSERT(lli->lli_size_sem_owner == NULL);
1509         lli->lli_size_sem_owner = current;
1510         lsm = lli->lli_smd;
1511         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1512                  lsm, lock_lsm);
1513         if (lock_lsm)
1514                 lov_stripe_lock(lsm);
1515 }
1516
1517 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1518 {
1519         struct ll_inode_info *lli;
1520         struct lov_stripe_md *lsm;
1521
1522         lli = ll_i2info(inode);
1523         lsm = lli->lli_smd;
1524         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1525                  lsm, unlock_lsm);
1526         if (unlock_lsm)
1527                 lov_stripe_unlock(lsm);
1528         LASSERT(lli->lli_size_sem_owner == current);
1529         lli->lli_size_sem_owner = NULL;
1530         cfs_up(&lli->lli_size_sem);
1531 }
1532
1533 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1534 {
1535         struct ll_inode_info *lli = ll_i2info(inode);
1536         struct mdt_body *body = md->body;
1537         struct lov_stripe_md *lsm = md->lsm;
1538         struct ll_sb_info *sbi = ll_i2sbi(inode);
1539
1540         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1541         if (lsm != NULL) {
1542                 cfs_down(&lli->lli_och_sem);
1543                 if (lli->lli_smd == NULL) {
1544                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1545                             lsm->lsm_magic != LOV_MAGIC_V3) {
1546                                 dump_lsm(D_ERROR, lsm);
1547                                 LBUG();
1548                         }
1549                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1550                                lsm, inode->i_ino, inode->i_generation, inode);
1551                         /* cl_inode_init must go before lli_smd or a race is
1552                          * possible where client thinks the file has stripes,
1553                          * but lov raid0 is not setup yet and parallel e.g.
1554                          * glimpse would try to use uninitialized lov */
1555                         cl_inode_init(inode, md);
1556                         lli->lli_smd = lsm;
1557                         cfs_up(&lli->lli_och_sem);
1558                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1559                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1560                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1561                 } else {
1562                         cfs_up(&lli->lli_och_sem);
1563                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1564                                 lli->lli_smd->lsm_stripe_count ==
1565                                 lsm->lsm_stripe_count);
1566                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1567                                 CERROR("lsm mismatch for inode %ld\n",
1568                                        inode->i_ino);
1569                                 CERROR("lli_smd:\n");
1570                                 dump_lsm(D_ERROR, lli->lli_smd);
1571                                 CERROR("lsm:\n");
1572                                 dump_lsm(D_ERROR, lsm);
1573                                 LBUG();
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                 cfs_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                 cfs_spin_unlock(&lli->lli_lock);
1591         }
1592 #endif
1593         inode->i_ino = cl_fid_build_ino(&body->fid1);
1594         inode->i_generation = cl_fid_build_gen(&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 (exp_connect_som(ll_i2mdexp(inode)) &&
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 FSFILT_IOC_GETFLAGS: {
1787                 struct mdt_body *body;
1788                 struct md_op_data *op_data;
1789
1790                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1791                                              0, 0, LUSTRE_OPC_ANY,
1792                                              NULL);
1793                 if (op_data == NULL)
1794                         RETURN(-ENOMEM);
1795
1796                 op_data->op_valid = OBD_MD_FLFLAGS;
1797                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1798                 ll_finish_md_op_data(op_data);
1799                 if (rc) {
1800                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1801                         RETURN(-abs(rc));
1802                 }
1803
1804                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1805
1806                 flags = body->flags;
1807
1808                 ptlrpc_req_finished(req);
1809
1810                 RETURN(put_user(flags, (int *)arg));
1811         }
1812         case FSFILT_IOC_SETFLAGS: {
1813                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1814                 struct obd_info oinfo = { { { 0 } } };
1815                 struct md_op_data *op_data;
1816
1817                 if (get_user(flags, (int *)arg))
1818                         RETURN(-EFAULT);
1819
1820                 oinfo.oi_md = lsm;
1821                 OBDO_ALLOC(oinfo.oi_oa);
1822                 if (!oinfo.oi_oa)
1823                         RETURN(-ENOMEM);
1824
1825                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1826                                              LUSTRE_OPC_ANY, NULL);
1827                 if (IS_ERR(op_data))
1828                         RETURN(PTR_ERR(op_data));
1829
1830                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1831                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1832                 rc = md_setattr(sbi->ll_md_exp, op_data,
1833                                 NULL, 0, NULL, 0, &req, NULL);
1834                 ll_finish_md_op_data(op_data);
1835                 ptlrpc_req_finished(req);
1836                 if (rc) {
1837                         OBDO_FREE(oinfo.oi_oa);
1838                         RETURN(rc);
1839                 }
1840
1841                 if (lsm == NULL) {
1842                         OBDO_FREE(oinfo.oi_oa);
1843                         GOTO(update_cache, rc);
1844                 }
1845
1846                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1847                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1848                 oinfo.oi_oa->o_flags = flags;
1849                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1850                                        OBD_MD_FLGROUP;
1851                 oinfo.oi_capa = ll_mdscapa_get(inode);
1852
1853                 obdo_from_inode(oinfo.oi_oa, inode,
1854                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1855                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1856                 capa_put(oinfo.oi_capa);
1857                 OBDO_FREE(oinfo.oi_oa);
1858                 if (rc) {
1859                         if (rc != -EPERM && rc != -EACCES)
1860                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1861                         RETURN(rc);
1862                 }
1863
1864                 EXIT;
1865 update_cache:
1866                 inode->i_flags = ll_ext_to_inode_flags(flags);
1867                 return 0;
1868         }
1869         default:
1870                 RETURN(-ENOSYS);
1871         }
1872
1873         RETURN(0);
1874 }
1875
1876 int ll_flush_ctx(struct inode *inode)
1877 {
1878         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1879
1880         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1881
1882         obd_set_info_async(sbi->ll_md_exp,
1883                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1884                            0, NULL, NULL);
1885         obd_set_info_async(sbi->ll_dt_exp,
1886                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1887                            0, NULL, NULL);
1888         return 0;
1889 }
1890
1891 /* umount -f client means force down, don't save state */
1892 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1893 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1894 {
1895         struct super_block *sb = vfsmnt->mnt_sb;
1896 #else
1897 void ll_umount_begin(struct super_block *sb)
1898 {
1899 #endif
1900         struct lustre_sb_info *lsi = s2lsi(sb);
1901         struct ll_sb_info *sbi = ll_s2sbi(sb);
1902         struct obd_device *obd;
1903         struct obd_ioctl_data *ioc_data;
1904         ENTRY;
1905
1906 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1907         if (!(flags & MNT_FORCE)) {
1908                 EXIT;
1909                 return;
1910         }
1911 #endif
1912
1913         /* Tell the MGC we got umount -f */
1914         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1915
1916         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1917                sb->s_count, atomic_read(&sb->s_active));
1918
1919         obd = class_exp2obd(sbi->ll_md_exp);
1920         if (obd == NULL) {
1921                 CERROR("Invalid MDC connection handle "LPX64"\n",
1922                        sbi->ll_md_exp->exp_handle.h_cookie);
1923                 EXIT;
1924                 return;
1925         }
1926         obd->obd_force = 1;
1927
1928         obd = class_exp2obd(sbi->ll_dt_exp);
1929         if (obd == NULL) {
1930                 CERROR("Invalid LOV connection handle "LPX64"\n",
1931                        sbi->ll_dt_exp->exp_handle.h_cookie);
1932                 EXIT;
1933                 return;
1934         }
1935         obd->obd_force = 1;
1936
1937         OBD_ALLOC_PTR(ioc_data);
1938         if (ioc_data) {
1939                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1940                               sizeof ioc_data, ioc_data, NULL);
1941
1942                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1943                               sizeof ioc_data, ioc_data, NULL);
1944
1945                 OBD_FREE_PTR(ioc_data);
1946         }
1947
1948
1949         /* Really, we'd like to wait until there are no requests outstanding,
1950          * and then continue.  For now, we just invalidate the requests,
1951          * schedule() and sleep one second if needed, and hope.
1952          */
1953         cfs_schedule();
1954 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1955         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1956                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1957                                                    cfs_time_seconds(1));
1958                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1959                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1960                                       "may try to umount it a bit later\n",
1961                                       atomic_read(&vfsmnt->mnt_count));
1962         }
1963 #endif
1964
1965         EXIT;
1966 }
1967
1968 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1969 {
1970         struct ll_sb_info *sbi = ll_s2sbi(sb);
1971         int err;
1972         __u32 read_only;
1973
1974         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1975                 read_only = *flags & MS_RDONLY;
1976                 err = obd_set_info_async(sbi->ll_md_exp,
1977                                          sizeof(KEY_READ_ONLY),
1978                                          KEY_READ_ONLY, sizeof(read_only),
1979                                          &read_only, NULL);
1980                 if (err) {
1981                         CERROR("Failed to change the read-only flag during "
1982                                "remount: %d\n", err);
1983                         return err;
1984                 }
1985
1986                 if (read_only)
1987                         sb->s_flags |= MS_RDONLY;
1988                 else
1989                         sb->s_flags &= ~MS_RDONLY;
1990         }
1991         return 0;
1992 }
1993
1994 int ll_prep_inode(struct inode **inode,
1995                   struct ptlrpc_request *req,
1996                   struct super_block *sb)
1997 {
1998         struct ll_sb_info *sbi = NULL;
1999         struct lustre_md md;
2000         int rc;
2001         ENTRY;
2002
2003         LASSERT(*inode || sb);
2004         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2005         prune_deathrow(sbi, 1);
2006         memset(&md, 0, sizeof(struct lustre_md));
2007
2008         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2009                               sbi->ll_md_exp, &md);
2010         if (rc)
2011                 RETURN(rc);
2012
2013         if (*inode) {
2014                 ll_update_inode(*inode, &md);
2015         } else {
2016                 LASSERT(sb != NULL);
2017
2018                 /*
2019                  * At this point server returns to client's same fid as client
2020                  * generated for creating. So using ->fid1 is okay here.
2021                  */
2022                 LASSERT(fid_is_sane(&md.body->fid1));
2023
2024                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
2025                 if (*inode == NULL || IS_ERR(*inode)) {
2026                         if (md.lsm)
2027                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2028 #ifdef CONFIG_FS_POSIX_ACL
2029                         if (md.posix_acl) {
2030                                 posix_acl_release(md.posix_acl);
2031                                 md.posix_acl = NULL;
2032                         }
2033 #endif
2034                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2035                         *inode = NULL;
2036                         CERROR("new_inode -fatal: rc %d\n", rc);
2037                         GOTO(out, rc);
2038                 }
2039         }
2040
2041 out:
2042         md_free_lustre_md(sbi->ll_md_exp, &md);
2043         RETURN(rc);
2044 }
2045
2046 int ll_obd_statfs(struct inode *inode, void *arg)
2047 {
2048         struct ll_sb_info *sbi = NULL;
2049         struct obd_export *exp;
2050         char *buf = NULL;
2051         struct obd_ioctl_data *data = NULL;
2052         __u32 type;
2053         int len = 0, rc;
2054
2055         if (!inode || !(sbi = ll_i2sbi(inode)))
2056                 GOTO(out_statfs, rc = -EINVAL);
2057
2058         rc = obd_ioctl_getdata(&buf, &len, arg);
2059         if (rc)
2060                 GOTO(out_statfs, rc);
2061
2062         data = (void*)buf;
2063         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2064             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2065                 GOTO(out_statfs, rc = -EINVAL);
2066
2067         if (data->ioc_inllen1 != sizeof(__u32) ||
2068             data->ioc_inllen2 != sizeof(__u32) ||
2069             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2070             data->ioc_plen2 != sizeof(struct obd_uuid))
2071                 GOTO(out_statfs, rc = -EINVAL);
2072
2073         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2074         if (type == LL_STATFS_MDC)
2075                 exp = sbi->ll_md_exp;
2076         else if (type == LL_STATFS_LOV)
2077                 exp = sbi->ll_dt_exp;
2078         else
2079                 GOTO(out_statfs, rc = -ENODEV);
2080
2081         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2082         if (rc)
2083                 GOTO(out_statfs, rc);
2084 out_statfs:
2085         if (buf)
2086                 obd_ioctl_freedata(buf, len);
2087         return rc;
2088 }
2089
2090 int ll_process_config(struct lustre_cfg *lcfg)
2091 {
2092         char *ptr;
2093         void *sb;
2094         struct lprocfs_static_vars lvars;
2095         unsigned long x;
2096         int rc = 0;
2097
2098         lprocfs_llite_init_vars(&lvars);
2099
2100         /* The instance name contains the sb: lustre-client-aacfe000 */
2101         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2102         if (!ptr || !*(++ptr))
2103                 return -EINVAL;
2104         if (sscanf(ptr, "%lx", &x) != 1)
2105                 return -EINVAL;
2106         sb = (void *)x;
2107         /* This better be a real Lustre superblock! */
2108         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2109
2110         /* Note we have not called client_common_fill_super yet, so
2111            proc fns must be able to handle that! */
2112         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2113                                       lcfg, sb);
2114         if (rc > 0)
2115                 rc = 0;
2116         return(rc);
2117 }
2118
2119 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2120 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2121                                        struct inode *i1, struct inode *i2,
2122                                        const char *name, int namelen,
2123                                        int mode, __u32 opc, void *data)
2124 {
2125         LASSERT(i1 != NULL);
2126
2127         if (namelen > ll_i2sbi(i1)->ll_namelen)
2128                 return ERR_PTR(-ENAMETOOLONG);
2129
2130         if (op_data == NULL)
2131                 OBD_ALLOC_PTR(op_data);
2132
2133         if (op_data == NULL)
2134                 return ERR_PTR(-ENOMEM);
2135
2136         ll_i2gids(op_data->op_suppgids, i1, i2);
2137         op_data->op_fid1 = *ll_inode2fid(i1);
2138         op_data->op_capa1 = ll_mdscapa_get(i1);
2139
2140         if (i2) {
2141                 op_data->op_fid2 = *ll_inode2fid(i2);
2142                 op_data->op_capa2 = ll_mdscapa_get(i2);
2143         } else {
2144                 fid_zero(&op_data->op_fid2);
2145                 op_data->op_capa2 = NULL;
2146         }
2147
2148         op_data->op_name = name;
2149         op_data->op_namelen = namelen;
2150         op_data->op_mode = mode;
2151         op_data->op_mod_time = cfs_time_current_sec();
2152         op_data->op_fsuid = cfs_curproc_fsuid();
2153         op_data->op_fsgid = cfs_curproc_fsgid();
2154         op_data->op_cap = cfs_curproc_cap_pack();
2155         op_data->op_bias = MDS_CHECK_SPLIT;
2156         op_data->op_opc = opc;
2157         op_data->op_mds = 0;
2158         op_data->op_data = data;
2159
2160         return op_data;
2161 }
2162
2163 void ll_finish_md_op_data(struct md_op_data *op_data)
2164 {
2165         capa_put(op_data->op_capa1);
2166         capa_put(op_data->op_capa2);
2167         OBD_FREE_PTR(op_data);
2168 }
2169
2170 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2171 {
2172         struct ll_sb_info *sbi;
2173
2174         LASSERT((seq != NULL) && (vfs != NULL));
2175         sbi = ll_s2sbi(vfs->mnt_sb);
2176
2177         if (sbi->ll_flags & LL_SBI_NOLCK)
2178                 seq_puts(seq, ",nolock");
2179
2180         if (sbi->ll_flags & LL_SBI_FLOCK)
2181                 seq_puts(seq, ",flock");
2182
2183         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2184                 seq_puts(seq, ",localflock");
2185
2186         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2187                 seq_puts(seq, ",user_xattr");
2188
2189         if (sbi->ll_flags & LL_SBI_ACL)
2190                 seq_puts(seq, ",acl");
2191
2192         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2193                 seq_puts(seq, ",lazystatfs");
2194
2195         RETURN(0);
2196 }