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 (c) 2003, 2010, Oracle and/or its affiliates. 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/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  63
  64 extern struct address_space_operations ll_aops;
  65 extern struct address_space_operations ll_dir_aops;
  66
  67 #ifndef log2
  68 #define log2(n) cfs_ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC(sbi, sizeof(*sbi));
  81         if (!sbi)
  82                 RETURN(NULL);
  83
  84         cfs_spin_lock_init(&sbi->ll_lock);
  85         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  86         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  87         cfs_spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  93 #ifdef HAVE_BGL_SUPPORT
  94                 sbi->ll_async_page_max = pages / 4;
  95 #else
  96                 sbi->ll_async_page_max = pages / 2;
  97 #endif
  98         } else {
  99                 sbi->ll_async_page_max = (pages / 4) * 3;
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         CFS_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         cfs_spin_lock(&ll_sb_lock);
 115         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         cfs_spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_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         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 127                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 128                                    pp_r_hist.oh_lock);
 129                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 130                                    pp_w_hist.oh_lock);
 131         }
 132
 133         /* metadata statahead is enabled by default */
 134         sbi->ll_sa_max = LL_SA_RPC_DEF;
 135         atomic_set(&sbi->ll_sa_total, 0);
 136         atomic_set(&sbi->ll_sa_wrong, 0);
 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                 cfs_spin_lock(&ll_sb_lock);
 148                 cfs_list_del(&sbi->ll_list);
 149                 cfs_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         .d_compare = ll_dcompare,
 157         .d_revalidate = ll_revalidate_nd,
 158 };
 159
 160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 161 {
 162         struct inode *root = 0;
 163         struct ll_sb_info *sbi = ll_s2sbi(sb);
 164         struct obd_device *obd;
 165         struct obd_capa *oc = NULL;
 166         struct obd_statfs osfs;
 167         struct ptlrpc_request *request = NULL;
 168         struct obd_connect_data *data = NULL;
 169         struct obd_uuid *uuid;
 170         struct md_op_data *op_data;
 171         struct lustre_md lmd;
 172         obd_valid valid;
 173         int size, err, checksum;
 174         ENTRY;
 175
 176         obd = class_name2obd(md);
 177         if (!obd) {
 178                 CERROR("MD %s: not setup or attached\n", md);
 179                 RETURN(-EINVAL);
 180         }
 181
 182         OBD_ALLOC_PTR(data);
 183         if (data == NULL)
 184                 RETURN(-ENOMEM);
 185
 186         if (proc_lustre_fs_root) {
 187                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 188                                                   dt, md);
 189                 if (err < 0)
 190                         CERROR("could not register mount in /proc/fs/lustre\n");
 191         }
 192
 193         /* indicate the features supported by this client */
 194         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 195                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 196                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 197                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 198                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 199                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 200                                   OBD_CONNECT_VBR      | OBD_CONNECT_FULL20;
 201
 202         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 203                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 204
 205 #ifdef HAVE_LRU_RESIZE_SUPPORT
 206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208 #endif
 209 #ifdef CONFIG_FS_POSIX_ACL
 210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 211 #endif
 212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 213         data->ocd_version = LUSTRE_VERSION_CODE;
 214
 215         if (sb->s_flags & MS_RDONLY)
 216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 219
 220 #ifdef HAVE_MS_FLOCK_LOCK
 221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 222         sb->s_flags |= MS_FLOCK_LOCK;
 223 #endif
 224
 225         if (sbi->ll_flags & LL_SBI_FLOCK)
 226                 sbi->ll_fop = &ll_file_operations_flock;
 227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 228                 sbi->ll_fop = &ll_file_operations;
 229         else
 230                 sbi->ll_fop = &ll_file_operations_noflock;
 231
 232         /* real client */
 233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 235                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 236
 237         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 238         if (err == -EBUSY) {
 239                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 240                                    "recovery, of which this client is not a "
 241                                    "part. Please wait for recovery to complete,"
 242                                    " abort, or time out.\n", md);
 243                 GOTO(out, err);
 244         } else if (err) {
 245                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 246                 GOTO(out, err);
 247         }
 248
 249         err = obd_fid_init(sbi->ll_md_exp);
 250         if (err) {
 251                 CERROR("Can't init metadata layer FID infrastructure, "
 252                        "rc %d\n", err);
 253                 GOTO(out_md, err);
 254         }
 255
 256         err = obd_statfs(obd, &osfs,
 257                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 258         if (err)
 259                 GOTO(out_md_fid, err);
 260
 261         size = sizeof(*data);
 262         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 263                            KEY_CONN_DATA,  &size, data, NULL);
 264         if (err) {
 265                 CERROR("Get connect data failed: %d \n", err);
 266                 GOTO(out_md, err);
 267         }
 268
 269         LASSERT(osfs.os_bsize);
 270         sb->s_blocksize = osfs.os_bsize;
 271         sb->s_blocksize_bits = log2(osfs.os_bsize);
 272         sb->s_magic = LL_SUPER_MAGIC;
 273
 274         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 275          *         retval = in_file->f_op->sendfile(...);
 276          *         if (*ppos > max)
 277          *                 retval = -EOVERFLOW;
 278          *
 279          * it will check if *ppos is greater than max. However, max equals to
 280          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 281          * has been defined as a signed long long integer in linux kernel. */
 282 #if BITS_PER_LONG == 64
 283         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 284 #else
 285         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 286 #endif
 287         sbi->ll_namelen = osfs.os_namelen;
 288         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 289
 290         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 291             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 292                 LCONSOLE_INFO("Disabling user_xattr feature because "
 293                               "it is not supported on the server\n");
 294                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 295         }
 296
 297         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 298 #ifdef MS_POSIXACL
 299                 sb->s_flags |= MS_POSIXACL;
 300 #endif
 301                 sbi->ll_flags |= LL_SBI_ACL;
 302         } else {
 303                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 304 #ifdef MS_POSIXACL
 305                 sb->s_flags &= ~MS_POSIXACL;
 306 #endif
 307                 sbi->ll_flags &= ~LL_SBI_ACL;
 308         }
 309
 310         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 311                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 312                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 313                         LCONSOLE_INFO("client is set as remote by default.\n");
 314                 }
 315         } else {
 316                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 317                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 318                         LCONSOLE_INFO("client claims to be remote, but server "
 319                                       "rejected, forced to be local.\n");
 320                 }
 321         }
 322
 323         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 324                 LCONSOLE_INFO("client enabled MDS capability!\n");
 325                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 326         }
 327
 328         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 329                 LCONSOLE_INFO("client enabled OSS capability!\n");
 330                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 331         }
 332
 333         obd = class_name2obd(dt);
 334         if (!obd) {
 335                 CERROR("DT %s: not setup or attached\n", dt);
 336                 GOTO(out_md_fid, err = -ENODEV);
 337         }
 338
 339         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 340                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 341                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 342                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 343                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 344                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 345                                   OBD_CONNECT_FULL20;
 346
 347         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 348                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 349
 350         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 351                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 352                  * disabled by default, because it can still be enabled on the
 353                  * fly via /proc. As a consequence, we still need to come to an
 354                  * agreement on the supported algorithms at connect time */
 355                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 356
 357                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 358                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 359                 else
 360                         /* send the list of supported checksum types */
 361                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 362         }
 363
 364 #ifdef HAVE_LRU_RESIZE_SUPPORT
 365         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 366 #endif
 367         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 368                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 369
 370         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 371                "ocd_grant: %d\n", data->ocd_connect_flags,
 372                data->ocd_version, data->ocd_grant);
 373
 374         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 375         obd->obd_upcall.onu_upcall = cl_ocd_update;
 376         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 377
 378         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 379         if (err == -EBUSY) {
 380                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 381                                    "recovery, of which this client is not a "
 382                                    "part.  Please wait for recovery to "
 383                                    "complete, abort, or time out.\n", dt);
 384                 GOTO(out_md_fid, err);
 385         } else if (err) {
 386                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 387                 GOTO(out_md_fid, err);
 388         }
 389
 390         err = obd_fid_init(sbi->ll_dt_exp);
 391         if (err) {
 392                 CERROR("Can't init data layer FID infrastructure, "
 393                        "rc %d\n", err);
 394                 GOTO(out_dt, err);
 395         }
 396
 397         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 398         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 399         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 400         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 401         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 402
 403         fid_zero(&sbi->ll_root_fid);
 404         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 405         if (err) {
 406                 CERROR("cannot mds_connect: rc = %d\n", err);
 407                 GOTO(out_lock_cn_cb, err);
 408         }
 409         if (!fid_is_sane(&sbi->ll_root_fid)) {
 410                 CERROR("Invalid root fid during mount\n");
 411                 GOTO(out_lock_cn_cb, err = -EINVAL);
 412         }
 413         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 414
 415         sb->s_op = &lustre_super_operations;
 416 #if THREAD_SIZE >= 8192
 417         sb->s_export_op = &lustre_export_operations;
 418 #endif
 419
 420         /* make root inode
 421          * XXX: move this to after cbd setup? */
 422         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 423         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 424                 valid |= OBD_MD_FLRMTPERM;
 425         else if (sbi->ll_flags & LL_SBI_ACL)
 426                 valid |= OBD_MD_FLACL;
 427
 428         OBD_ALLOC_PTR(op_data);
 429         if (op_data == NULL)
 430                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 431
 432         op_data->op_fid1 = sbi->ll_root_fid;
 433         op_data->op_mode = 0;
 434         op_data->op_capa1 = oc;
 435         op_data->op_valid = valid;
 436
 437         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 438         if (oc)
 439                 capa_put(oc);
 440         OBD_FREE_PTR(op_data);
 441         if (err) {
 442                 CERROR("md_getattr failed for root: rc = %d\n", err);
 443                 GOTO(out_lock_cn_cb, err);
 444         }
 445         memset(&lmd, 0, sizeof(lmd));
 446         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 447                                sbi->ll_md_exp, &lmd);
 448         if (err) {
 449                 CERROR("failed to understand root inode md: rc = %d\n", err);
 450                 ptlrpc_req_finished (request);
 451                 GOTO(out_lock_cn_cb, err);
 452         }
 453
 454         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 455         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
 456         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 457         ptlrpc_req_finished(request);
 458
 459         if (root == NULL || IS_ERR(root)) {
 460                 if (lmd.lsm)
 461                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 462 #ifdef CONFIG_FS_POSIX_ACL
 463                 if (lmd.posix_acl) {
 464                         posix_acl_release(lmd.posix_acl);
 465                         lmd.posix_acl = NULL;
 466                 }
 467 #endif
 468                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 469                 root = NULL;
 470                 CERROR("lustre_lite: bad iget4 for root\n");
 471                 GOTO(out_root, err);
 472         }
 473
 474         err = ll_close_thread_start(&sbi->ll_lcq);
 475         if (err) {
 476                 CERROR("cannot start close thread: rc %d\n", err);
 477                 GOTO(out_root, err);
 478         }
 479
 480 #ifdef CONFIG_FS_POSIX_ACL
 481         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 482                 rct_init(&sbi->ll_rct);
 483                 et_init(&sbi->ll_et);
 484         }
 485 #endif
 486
 487         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 488         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 489                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 490                                  NULL);
 491         cl_sb_init(sb);
 492
 493         sb->s_root = d_alloc_root(root);
 494         if (data != NULL)
 495                 OBD_FREE(data, sizeof(*data));
 496
 497         sb->s_root->d_op = &ll_d_root_ops;
 498
 499         sbi->ll_sdev_orig = sb->s_dev;
 500
 501         /* We set sb->s_dev equal on all lustre clients in order to support
 502          * NFS export clustering.  NFSD requires that the FSID be the same
 503          * on all clients. */
 504         /* s_dev is also used in lt_compare() to compare two fs, but that is
 505          * only a node-local comparison. */
 506         uuid = obd_get_uuid(sbi->ll_md_exp);
 507         if (uuid != NULL)
 508                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 509
 510         RETURN(err);
 511 out_root:
 512         if (root)
 513                 iput(root);
 514 out_lock_cn_cb:
 515         obd_fid_fini(sbi->ll_dt_exp);
 516 out_dt:
 517         obd_disconnect(sbi->ll_dt_exp);
 518         sbi->ll_dt_exp = NULL;
 519 out_md_fid:
 520         obd_fid_fini(sbi->ll_md_exp);
 521 out_md:
 522         obd_disconnect(sbi->ll_md_exp);
 523         sbi->ll_md_exp = NULL;
 524 out:
 525         if (data != NULL)
 526                 OBD_FREE_PTR(data);
 527         lprocfs_unregister_mountpoint(sbi);
 528         return err;
 529 }
 530
 531 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 532 {
 533         int size, rc;
 534
 535         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 536         size = sizeof(int);
 537         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 538                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 539         if (rc)
 540                 CERROR("Get max mdsize error rc %d \n", rc);
 541
 542         RETURN(rc);
 543 }
 544
 545 void ll_dump_inode(struct inode *inode)
 546 {
 547         struct list_head *tmp;
 548         int dentry_count = 0;
 549
 550         LASSERT(inode != NULL);
 551
 552         list_for_each(tmp, &inode->i_dentry)
 553                 dentry_count++;
 554
 555         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 556                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 557                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 558 }
 559
 560 void lustre_dump_dentry(struct dentry *dentry, int recur)
 561 {
 562         struct list_head *tmp;
 563         int subdirs = 0;
 564
 565         LASSERT(dentry != NULL);
 566
 567         list_for_each(tmp, &dentry->d_subdirs)
 568                 subdirs++;
 569
 570         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 571                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 572                dentry->d_name.len, dentry->d_name.name,
 573                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 574                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 575                dentry->d_flags, dentry->d_fsdata, subdirs);
 576         if (dentry->d_inode != NULL)
 577                 ll_dump_inode(dentry->d_inode);
 578
 579         if (recur == 0)
 580                 return;
 581
 582        list_for_each(tmp, &dentry->d_subdirs) {
 583                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 584                 lustre_dump_dentry(d, recur - 1);
 585         }
 586 }
 587
 588 void client_common_put_super(struct super_block *sb)
 589 {
 590         struct ll_sb_info *sbi = ll_s2sbi(sb);
 591         ENTRY;
 592
 593 #ifdef CONFIG_FS_POSIX_ACL
 594         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 595                 et_fini(&sbi->ll_et);
 596                 rct_fini(&sbi->ll_rct);
 597         }
 598 #endif
 599
 600         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 601
 602         ll_close_thread_shutdown(sbi->ll_lcq);
 603
 604         cl_sb_fini(sb);
 605
 606         cfs_list_del(&sbi->ll_conn_chain);
 607
 608         obd_fid_fini(sbi->ll_dt_exp);
 609         obd_disconnect(sbi->ll_dt_exp);
 610         sbi->ll_dt_exp = NULL;
 611
 612         lprocfs_unregister_mountpoint(sbi);
 613
 614         obd_fid_fini(sbi->ll_md_exp);
 615         obd_disconnect(sbi->ll_md_exp);
 616         sbi->ll_md_exp = NULL;
 617
 618         EXIT;
 619 }
 620
 621 void ll_kill_super(struct super_block *sb)
 622 {
 623         struct ll_sb_info *sbi;
 624
 625         ENTRY;
 626
 627         /* not init sb ?*/
 628         if (!(sb->s_flags & MS_ACTIVE))
 629                 return;
 630
 631         sbi = ll_s2sbi(sb);
 632         /* we need restore s_dev from changed for clustred NFS before put_super
 633          * because new kernels have cached s_dev and change sb->s_dev in
 634          * put_super not affected real removing devices */
 635         if (sbi)
 636                 sb->s_dev = sbi->ll_sdev_orig;
 637         EXIT;
 638 }
 639
 640 char *ll_read_opt(const char *opt, char *data)
 641 {
 642         char *value;
 643         char *retval;
 644         ENTRY;
 645
 646         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 647         if (strncmp(opt, data, strlen(opt)))
 648                 RETURN(NULL);
 649         if ((value = strchr(data, '=')) == NULL)
 650                 RETURN(NULL);
 651
 652         value++;
 653         OBD_ALLOC(retval, strlen(value) + 1);
 654         if (!retval) {
 655                 CERROR("out of memory!\n");
 656                 RETURN(NULL);
 657         }
 658
 659         memcpy(retval, value, strlen(value)+1);
 660         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 661         RETURN(retval);
 662 }
 663
 664 static inline int ll_set_opt(const char *opt, char *data, int fl)
 665 {
 666         if (strncmp(opt, data, strlen(opt)) != 0)
 667                 return(0);
 668         else
 669                 return(fl);
 670 }
 671
 672 /* non-client-specific mount options are parsed in lmd_parse */
 673 static int ll_options(char *options, int *flags)
 674 {
 675         int tmp;
 676         char *s1 = options, *s2;
 677         ENTRY;
 678
 679         if (!options)
 680                 RETURN(0);
 681
 682         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 683
 684         while (*s1) {
 685                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 686                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 687                 if (tmp) {
 688                         *flags |= tmp;
 689                         goto next;
 690                 }
 691                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 692                 if (tmp) {
 693                         *flags |= tmp;
 694                         goto next;
 695                 }
 696                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 697                 if (tmp) {
 698                         *flags |= tmp;
 699                         goto next;
 700                 }
 701                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 702                 if (tmp) {
 703                         *flags &= ~tmp;
 704                         goto next;
 705                 }
 706                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 707                 if (tmp) {
 708                         *flags |= tmp;
 709                         goto next;
 710                 }
 711                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 712                 if (tmp) {
 713                         *flags &= ~tmp;
 714                         goto next;
 715                 }
 716 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 717                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 718                 if (tmp) {
 719                         /* Ignore deprecated mount option.  The client will
 720                          * always try to mount with ACL support, whether this
 721                          * is used depends on whether server supports it. */
 722                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 723                                                   "mount option 'acl'.\n");
 724                         goto next;
 725                 }
 726                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 727                 if (tmp) {
 728                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 729                                                   "mount option 'noacl'.\n");
 730                         goto next;
 731                 }
 732 #else
 733 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 734 #endif
 735                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 736                 if (tmp) {
 737                         *flags |= tmp;
 738                         goto next;
 739                 }
 740
 741                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 742                 if (tmp) {
 743                         *flags |= tmp;
 744                         goto next;
 745                 }
 746                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 747                 if (tmp) {
 748                         *flags &= ~tmp;
 749                         goto next;
 750                 }
 751                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 752                 if (tmp) {
 753                         *flags |= tmp;
 754                         goto next;
 755                 }
 756                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 757                 if (tmp) {
 758                         *flags &= ~tmp;
 759                         goto next;
 760                 }
 761                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 767                 if (tmp) {
 768                         *flags &= ~tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781
 782                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 783                                    s1);
 784                 RETURN(-EINVAL);
 785
 786 next:
 787                 /* Find next opt */
 788                 s2 = strchr(s1, ',');
 789                 if (s2 == NULL)
 790                         break;
 791                 s1 = s2 + 1;
 792         }
 793         RETURN(0);
 794 }
 795
 796 void ll_lli_init(struct ll_inode_info *lli)
 797 {
 798         lli->lli_inode_magic = LLI_INODE_MAGIC;
 799         cfs_sema_init(&lli->lli_size_sem, 1);
 800         cfs_sema_init(&lli->lli_write_sem, 1);
 801         cfs_init_rwsem(&lli->lli_trunc_sem);
 802         lli->lli_flags = 0;
 803         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 804         cfs_spin_lock_init(&lli->lli_lock);
 805         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 806         lli->lli_inode_magic = LLI_INODE_MAGIC;
 807         cfs_sema_init(&lli->lli_och_sem, 1);
 808         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 809         lli->lli_mds_exec_och = NULL;
 810         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 811         lli->lli_open_fd_exec_count = 0;
 812         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 813         lli->lli_remote_perms = NULL;
 814         lli->lli_rmtperm_utime = 0;
 815         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 816         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 817         cfs_spin_lock_init(&lli->lli_sa_lock);
 818         cfs_sema_init(&lli->lli_readdir_sem, 1);
 819 }
 820
 821 #ifdef HAVE_NEW_BACKING_DEV_INFO
 822 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 823 #endif
 824
 825 int ll_fill_super(struct super_block *sb)
 826 {
 827         struct lustre_profile *lprof;
 828         struct lustre_sb_info *lsi = s2lsi(sb);
 829         struct ll_sb_info *sbi;
 830         char  *dt = NULL, *md = NULL;
 831         char  *profilenm = get_profile_name(sb);
 832         struct config_llog_instance cfg = {0, };
 833         char   ll_instance[sizeof(sb) * 2 + 1];
 834         int    err;
 835         ENTRY;
 836
 837         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 838
 839         cfs_module_get();
 840
 841         /* client additional sb info */
 842         lsi->lsi_llsbi = sbi = ll_init_sbi();
 843         if (!sbi) {
 844                 cfs_module_put(THIS_MODULE);
 845                 RETURN(-ENOMEM);
 846         }
 847
 848         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 849         if (err)
 850                 GOTO(out_free, err);
 851
 852         err = ll_bdi_init(&lsi->bdi);
 853         if (err)
 854                 GOTO(out_free, err);
 855
 856 #ifdef HAVE_NEW_BACKING_DEV_INFO
 857         lsi->bdi.name = "lustre";
 858         lsi->bdi.capabilities = BDI_CAP_MAP_COPY;
 859         err = bdi_register(&lsi->bdi, NULL, "lustre-%d",
 860                            atomic_inc_return(&ll_bdi_num));
 861         sb->s_bdi = &lsi->bdi;
 862 #endif
 863
 864         /* Generate a string unique to this super, in case some joker tries
 865            to mount the same fs at two mount points.
 866            Use the address of the super itself.*/
 867         sprintf(ll_instance, "%p", sb);
 868         cfg.cfg_instance = ll_instance;
 869         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 870
 871         /* set up client obds */
 872         err = lustre_process_log(sb, profilenm, &cfg);
 873         if (err < 0) {
 874                 CERROR("Unable to process log: %d\n", err);
 875                 GOTO(out_free, err);
 876         }
 877
 878         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 879         lprof = class_get_profile(profilenm);
 880         if (lprof == NULL) {
 881                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 882                                    " read from the MGS.  Does that filesystem "
 883                                    "exist?\n", profilenm);
 884                 GOTO(out_free, err = -EINVAL);
 885         }
 886         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 887                lprof->lp_md, lprof->lp_dt);
 888
 889         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 890                   strlen(ll_instance) + 2);
 891         if (!dt)
 892                 GOTO(out_free, err = -ENOMEM);
 893         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 894
 895         OBD_ALLOC(md, strlen(lprof->lp_md) +
 896                   strlen(ll_instance) + 2);
 897         if (!md)
 898                 GOTO(out_free, err = -ENOMEM);
 899         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 900
 901         /* connections, registrations, sb setup */
 902         err = client_common_fill_super(sb, md, dt);
 903
 904 out_free:
 905         if (md)
 906                 OBD_FREE(md, strlen(md) + 1);
 907         if (dt)
 908                 OBD_FREE(dt, strlen(dt) + 1);
 909         if (err)
 910                 ll_put_super(sb);
 911         else
 912                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 913
 914         RETURN(err);
 915 } /* ll_fill_super */
 916
 917
 918 void lu_context_keys_dump(void);
 919
 920 void ll_put_super(struct super_block *sb)
 921 {
 922         struct config_llog_instance cfg;
 923         char   ll_instance[sizeof(sb) * 2 + 1];
 924         struct obd_device *obd;
 925         struct lustre_sb_info *lsi = s2lsi(sb);
 926         struct ll_sb_info *sbi = ll_s2sbi(sb);
 927         char *profilenm = get_profile_name(sb);
 928         int force = 1, next;
 929         ENTRY;
 930
 931         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 932
 933         ll_print_capa_stat(sbi);
 934
 935         sprintf(ll_instance, "%p", sb);
 936         cfg.cfg_instance = ll_instance;
 937         lustre_end_log(sb, NULL, &cfg);
 938
 939         if (sbi->ll_md_exp) {
 940                 obd = class_exp2obd(sbi->ll_md_exp);
 941                 if (obd)
 942                         force = obd->obd_force;
 943         }
 944
 945         /* We need to set force before the lov_disconnect in
 946            lustre_common_put_super, since l_d cleans up osc's as well. */
 947         if (force) {
 948                 next = 0;
 949                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 950                                                      &next)) != NULL) {
 951                         obd->obd_force = force;
 952                 }
 953         }
 954
 955         if (sbi->ll_lcq) {
 956                 /* Only if client_common_fill_super succeeded */
 957                 client_common_put_super(sb);
 958         }
 959
 960         next = 0;
 961         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 962                 class_manual_cleanup(obd);
 963         }
 964
 965         if (profilenm)
 966                 class_del_profile(profilenm);
 967
 968         if (ll_bdi_wb_cnt(lsi->bdi) > 0)
 969                 ll_bdi_destroy(&lsi->bdi);
 970
 971         ll_free_sbi(sb);
 972         lsi->lsi_llsbi = NULL;
 973
 974         lustre_common_put_super(sb);
 975
 976         cl_env_cache_purge(~0);
 977
 978         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
 979
 980         cfs_module_put(THIS_MODULE);
 981
 982         EXIT;
 983 } /* client_put_super */
 984
 985 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 986 {
 987         struct inode *inode = NULL;
 988         /* NOTE: we depend on atomic igrab() -bzzz */
 989         lock_res_and_lock(lock);
 990         if (lock->l_ast_data) {
 991                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 992                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 993                         inode = igrab(lock->l_ast_data);
 994                 } else {
 995                         inode = lock->l_ast_data;
 996                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
 997                                          D_WARNING, lock, "l_ast_data %p is "
 998                                          "bogus: magic %08x", lock->l_ast_data,
 999                                          lli->lli_inode_magic);
1000                         inode = NULL;
1001                 }
1002         }
1003         unlock_res_and_lock(lock);
1004         return inode;
1005 }
1006
1007 static int null_if_equal(struct ldlm_lock *lock, void *data)
1008 {
1009         if (data == lock->l_ast_data) {
1010                 lock->l_ast_data = NULL;
1011
1012                 if (lock->l_req_mode != lock->l_granted_mode)
1013                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1014         }
1015
1016         return LDLM_ITER_CONTINUE;
1017 }
1018
1019 void ll_clear_inode(struct inode *inode)
1020 {
1021         struct ll_inode_info *lli = ll_i2info(inode);
1022         struct ll_sb_info *sbi = ll_i2sbi(inode);
1023         ENTRY;
1024
1025         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1026                inode->i_generation, inode);
1027
1028         if (S_ISDIR(inode->i_mode)) {
1029                 /* these should have been cleared in ll_file_release */
1030                 LASSERT(lli->lli_sai == NULL);
1031                 LASSERT(lli->lli_opendir_key == NULL);
1032                 LASSERT(lli->lli_opendir_pid == 0);
1033         }
1034
1035         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1036         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1037                          null_if_equal, inode);
1038
1039         LASSERT(!lli->lli_open_fd_write_count);
1040         LASSERT(!lli->lli_open_fd_read_count);
1041         LASSERT(!lli->lli_open_fd_exec_count);
1042
1043         if (lli->lli_mds_write_och)
1044                 ll_md_real_close(inode, FMODE_WRITE);
1045         if (lli->lli_mds_exec_och)
1046                 ll_md_real_close(inode, FMODE_EXEC);
1047         if (lli->lli_mds_read_och)
1048                 ll_md_real_close(inode, FMODE_READ);
1049
1050         if (lli->lli_symlink_name) {
1051                 OBD_FREE(lli->lli_symlink_name,
1052                          strlen(lli->lli_symlink_name) + 1);
1053                 lli->lli_symlink_name = NULL;
1054         }
1055
1056         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1057                 LASSERT(lli->lli_posix_acl == NULL);
1058                 if (lli->lli_remote_perms) {
1059                         free_rmtperm_hash(lli->lli_remote_perms);
1060                         lli->lli_remote_perms = NULL;
1061                 }
1062         }
1063 #ifdef CONFIG_FS_POSIX_ACL
1064         else if (lli->lli_posix_acl) {
1065                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1066                 LASSERT(lli->lli_remote_perms == NULL);
1067                 posix_acl_release(lli->lli_posix_acl);
1068                 lli->lli_posix_acl = NULL;
1069         }
1070 #endif
1071         lli->lli_inode_magic = LLI_INODE_DEAD;
1072
1073         ll_clear_inode_capas(inode);
1074         /*
1075          * XXX This has to be done before lsm is freed below, because
1076          * cl_object still uses inode lsm.
1077          */
1078         cl_inode_fini(inode);
1079
1080         if (lli->lli_smd) {
1081                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1082                 lli->lli_smd = NULL;
1083         }
1084
1085
1086         EXIT;
1087 }
1088
1089 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1090                   struct md_open_data **mod)
1091 {
1092         struct lustre_md md;
1093         struct ll_sb_info *sbi = ll_i2sbi(inode);
1094         struct ptlrpc_request *request = NULL;
1095         int rc;
1096         ENTRY;
1097
1098         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1099                                      LUSTRE_OPC_ANY, NULL);
1100         if (IS_ERR(op_data))
1101                 RETURN(PTR_ERR(op_data));
1102
1103         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1104                         &request, mod);
1105         if (rc) {
1106                 ptlrpc_req_finished(request);
1107                 if (rc == -ENOENT) {
1108                         inode->i_nlink = 0;
1109                         /* Unlinked special device node? Or just a race?
1110                          * Pretend we done everything. */
1111                         if (!S_ISREG(inode->i_mode) &&
1112                             !S_ISDIR(inode->i_mode))
1113                                 rc = inode_setattr(inode, &op_data->op_attr);
1114                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1115                         CERROR("md_setattr fails: rc = %d\n", rc);
1116                 }
1117                 RETURN(rc);
1118         }
1119
1120         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1121                               sbi->ll_md_exp, &md);
1122         if (rc) {
1123                 ptlrpc_req_finished(request);
1124                 RETURN(rc);
1125         }
1126
1127         /* We call inode_setattr to adjust timestamps.
1128          * If there is at least some data in file, we cleared ATTR_SIZE
1129          * above to avoid invoking vmtruncate, otherwise it is important
1130          * to call vmtruncate in inode_setattr to update inode->i_size
1131          * (bug 6196) */
1132         rc = inode_setattr(inode, &op_data->op_attr);
1133
1134         /* Extract epoch data if obtained. */
1135         op_data->op_handle = md.body->handle;
1136         op_data->op_ioepoch = md.body->ioepoch;
1137
1138         ll_update_inode(inode, &md);
1139         ptlrpc_req_finished(request);
1140
1141         RETURN(rc);
1142 }
1143
1144 /* Close IO epoch and send Size-on-MDS attribute update. */
1145 static int ll_setattr_done_writing(struct inode *inode,
1146                                    struct md_op_data *op_data,
1147                                    struct md_open_data *mod)
1148 {
1149         struct ll_inode_info *lli = ll_i2info(inode);
1150         int rc = 0;
1151         ENTRY;
1152
1153         LASSERT(op_data != NULL);
1154         if (!S_ISREG(inode->i_mode))
1155                 RETURN(0);
1156
1157         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1158                op_data->op_ioepoch, PFID(&lli->lli_fid));
1159
1160         op_data->op_flags = MF_EPOCH_CLOSE;
1161         ll_done_writing_attr(inode, op_data);
1162         ll_pack_inode2opdata(inode, op_data, NULL);
1163
1164         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1165         if (rc == -EAGAIN) {
1166                 /* MDS has instructed us to obtain Size-on-MDS attribute
1167                  * from OSTs and send setattr to back to MDS. */
1168                 rc = ll_som_update(inode, op_data);
1169         } else if (rc) {
1170                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1171                        inode->i_ino, rc);
1172         }
1173         RETURN(rc);
1174 }
1175
1176 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1177 {
1178         struct obd_capa *capa;
1179         int rc;
1180
1181         if (attr->ia_valid & ATTR_SIZE)
1182                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1183         else
1184                 capa = ll_mdscapa_get(inode);
1185
1186         rc = cl_setattr_ost(inode, attr, capa);
1187
1188         if (attr->ia_valid & ATTR_SIZE)
1189                 ll_truncate_free_capa(capa);
1190         else
1191                 capa_put(capa);
1192
1193         return rc;
1194 }
1195
1196 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1197  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1198  * keep these values until such a time that objects are allocated for it.
1199  * We do the MDS operations first, as it is checking permissions for us.
1200  * We don't to the MDS RPC if there is nothing that we want to store there,
1201  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1202  * going to do an RPC anyways.
1203  *
1204  * If we are doing a truncate, we will send the mtime and ctime updates
1205  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1206  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1207  * at the same time.
1208  */
1209 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1210 {
1211         struct ll_inode_info *lli = ll_i2info(inode);
1212         struct lov_stripe_md *lsm = lli->lli_smd;
1213         struct md_op_data *op_data = NULL;
1214         struct md_open_data *mod = NULL;
1215         int ia_valid = attr->ia_valid;
1216         int rc = 0, rc1 = 0;
1217         ENTRY;
1218
1219         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1220                attr->ia_valid);
1221         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1222
1223         if (ia_valid & ATTR_SIZE) {
1224                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1225                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1226                                attr->ia_size, ll_file_maxbytes(inode));
1227                         RETURN(-EFBIG);
1228                 }
1229
1230                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1231         }
1232
1233         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1234         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1235                 if (cfs_curproc_fsuid() != inode->i_uid &&
1236                     !cfs_capable(CFS_CAP_FOWNER))
1237                         RETURN(-EPERM);
1238         }
1239
1240         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1241         if (attr->ia_valid & ATTR_CTIME) {
1242                 attr->ia_ctime = CFS_CURRENT_TIME;
1243                 attr->ia_valid |= ATTR_CTIME_SET;
1244         }
1245         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1246                 attr->ia_atime = CFS_CURRENT_TIME;
1247                 attr->ia_valid |= ATTR_ATIME_SET;
1248         }
1249         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1250                 attr->ia_mtime = CFS_CURRENT_TIME;
1251                 attr->ia_valid |= ATTR_MTIME_SET;
1252         }
1253
1254         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1255                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1256                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1257                        cfs_time_current_sec());
1258
1259         /* NB: ATTR_SIZE will only be set after this point if the size
1260          * resides on the MDS, ie, this file has no objects. */
1261         if (lsm)
1262                 attr->ia_valid &= ~ATTR_SIZE;
1263
1264         /* We always do an MDS RPC, even if we're only changing the size;
1265          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1266
1267         OBD_ALLOC_PTR(op_data);
1268         if (op_data == NULL)
1269                 RETURN(-ENOMEM);
1270
1271         UNLOCK_INODE_MUTEX(inode);
1272         if (ia_valid & ATTR_SIZE)
1273                 UP_WRITE_I_ALLOC_SEM(inode);
1274         cfs_down_write(&lli->lli_trunc_sem);
1275         LOCK_INODE_MUTEX(inode);
1276         if (ia_valid & ATTR_SIZE)
1277                 DOWN_WRITE_I_ALLOC_SEM(inode);
1278
1279         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1280
1281         /* Open epoch for truncate. */
1282         if (exp_connect_som(ll_i2mdexp(inode)) &&
1283             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1284                 op_data->op_flags = MF_EPOCH_OPEN;
1285
1286         rc = ll_md_setattr(inode, op_data, &mod);
1287         if (rc)
1288                 GOTO(out, rc);
1289
1290         ll_ioepoch_open(lli, op_data->op_ioepoch);
1291         if (!lsm || !S_ISREG(inode->i_mode)) {
1292                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1293                 GOTO(out, rc = 0);
1294         }
1295
1296         if (ia_valid & ATTR_SIZE)
1297                 attr->ia_valid |= ATTR_SIZE;
1298         if (ia_valid & (ATTR_SIZE |
1299                         ATTR_ATIME | ATTR_ATIME_SET |
1300                         ATTR_MTIME | ATTR_MTIME_SET))
1301                 /* on truncate and utimes send attributes to osts, setting
1302                  * mtime/atime to past will be performed under PW 0:EOF extent
1303                  * lock (new_size:EOF for truncate)
1304                  * it may seem excessive to send mtime/atime updates to osts
1305                  * when not setting times to past, but it is necessary due to
1306                  * possible time de-synchronization */
1307                 rc = ll_setattr_ost(inode, attr);
1308         EXIT;
1309 out:
1310         if (op_data) {
1311                 if (op_data->op_ioepoch)
1312                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1313                 ll_finish_md_op_data(op_data);
1314         }
1315         cfs_up_write(&lli->lli_trunc_sem);
1316         return rc ? rc : rc1;
1317 }
1318
1319 int ll_setattr(struct dentry *de, struct iattr *attr)
1320 {
1321         int mode = de->d_inode->i_mode;
1322
1323         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1324                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1325                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1326
1327         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1328                                (ATTR_SIZE|ATTR_MODE)) &&
1329             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1330              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1331               !(attr->ia_mode & S_ISGID))))
1332                 attr->ia_valid |= ATTR_FORCE;
1333
1334         if ((mode & S_ISUID) &&
1335             !(attr->ia_mode & S_ISUID) &&
1336             !(attr->ia_valid & ATTR_KILL_SUID))
1337                 attr->ia_valid |= ATTR_KILL_SUID;
1338
1339         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1340             !(attr->ia_mode & S_ISGID) &&
1341             !(attr->ia_valid & ATTR_KILL_SGID))
1342                 attr->ia_valid |= ATTR_KILL_SGID;
1343
1344         return ll_setattr_raw(de->d_inode, attr);
1345 }
1346
1347 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1348                        __u64 max_age, __u32 flags)
1349 {
1350         struct ll_sb_info *sbi = ll_s2sbi(sb);
1351         struct obd_statfs obd_osfs;
1352         int rc;
1353         ENTRY;
1354
1355         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1356         if (rc) {
1357                 CERROR("md_statfs fails: rc = %d\n", rc);
1358                 RETURN(rc);
1359         }
1360
1361         osfs->os_type = sb->s_magic;
1362
1363         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1364                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1365
1366         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1367                 flags |= OBD_STATFS_NODELAY;
1368
1369         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1370                               &obd_osfs, max_age, flags);
1371         if (rc) {
1372                 CERROR("obd_statfs fails: rc = %d\n", rc);
1373                 RETURN(rc);
1374         }
1375
1376         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1377                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1378                obd_osfs.os_files);
1379
1380         osfs->os_bsize = obd_osfs.os_bsize;
1381         osfs->os_blocks = obd_osfs.os_blocks;
1382         osfs->os_bfree = obd_osfs.os_bfree;
1383         osfs->os_bavail = obd_osfs.os_bavail;
1384
1385         /* If we don't have as many objects free on the OST as inodes
1386          * on the MDS, we reduce the total number of inodes to
1387          * compensate, so that the "inodes in use" number is correct.
1388          */
1389         if (obd_osfs.os_ffree < osfs->os_ffree) {
1390                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1391                         obd_osfs.os_ffree;
1392                 osfs->os_ffree = obd_osfs.os_ffree;
1393         }
1394
1395         RETURN(rc);
1396 }
1397 #ifndef HAVE_STATFS_DENTRY_PARAM
1398 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1399 {
1400 #else
1401 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1402 {
1403         struct super_block *sb = de->d_sb;
1404 #endif
1405         struct obd_statfs osfs;
1406         int rc;
1407
1408         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1409         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1410
1411         /* Some amount of caching on the client is allowed */
1412         rc = ll_statfs_internal(sb, &osfs,
1413                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1414                                 0);
1415         if (rc)
1416                 return rc;
1417
1418         statfs_unpack(sfs, &osfs);
1419
1420         /* We need to downshift for all 32-bit kernels, because we can't
1421          * tell if the kernel is being called via sys_statfs64() or not.
1422          * Stop before overflowing f_bsize - in which case it is better
1423          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1424         if (sizeof(long) < 8) {
1425                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1426                         sfs->f_bsize <<= 1;
1427
1428                         osfs.os_blocks >>= 1;
1429                         osfs.os_bfree >>= 1;
1430                         osfs.os_bavail >>= 1;
1431                 }
1432         }
1433
1434         sfs->f_blocks = osfs.os_blocks;
1435         sfs->f_bfree = osfs.os_bfree;
1436         sfs->f_bavail = osfs.os_bavail;
1437
1438         return 0;
1439 }
1440
1441 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1442 {
1443         struct ll_inode_info *lli;
1444         struct lov_stripe_md *lsm;
1445
1446         lli = ll_i2info(inode);
1447         LASSERT(lli->lli_size_sem_owner != current);
1448         cfs_down(&lli->lli_size_sem);
1449         LASSERT(lli->lli_size_sem_owner == NULL);
1450         lli->lli_size_sem_owner = current;
1451         lsm = lli->lli_smd;
1452         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1453                  lsm, lock_lsm);
1454         if (lock_lsm)
1455                 lov_stripe_lock(lsm);
1456 }
1457
1458 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1459 {
1460         struct ll_inode_info *lli;
1461         struct lov_stripe_md *lsm;
1462
1463         lli = ll_i2info(inode);
1464         lsm = lli->lli_smd;
1465         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1466                  lsm, unlock_lsm);
1467         if (unlock_lsm)
1468                 lov_stripe_unlock(lsm);
1469         LASSERT(lli->lli_size_sem_owner == current);
1470         lli->lli_size_sem_owner = NULL;
1471         cfs_up(&lli->lli_size_sem);
1472 }
1473
1474 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1475 {
1476         struct ll_inode_info *lli = ll_i2info(inode);
1477         struct mdt_body *body = md->body;
1478         struct lov_stripe_md *lsm = md->lsm;
1479         struct ll_sb_info *sbi = ll_i2sbi(inode);
1480
1481         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1482         if (lsm != NULL) {
1483                 cfs_down(&lli->lli_och_sem);
1484                 if (lli->lli_smd == NULL) {
1485                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1486                             lsm->lsm_magic != LOV_MAGIC_V3) {
1487                                 dump_lsm(D_ERROR, lsm);
1488                                 LBUG();
1489                         }
1490                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1491                                lsm, inode->i_ino, inode->i_generation, inode);
1492                         /* cl_inode_init must go before lli_smd or a race is
1493                          * possible where client thinks the file has stripes,
1494                          * but lov raid0 is not setup yet and parallel e.g.
1495                          * glimpse would try to use uninitialized lov */
1496                         cl_inode_init(inode, md);
1497                         lli->lli_smd = lsm;
1498                         cfs_up(&lli->lli_och_sem);
1499                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1500                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1501                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1502                 } else {
1503                         cfs_up(&lli->lli_och_sem);
1504                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1505                                 lli->lli_smd->lsm_stripe_count ==
1506                                 lsm->lsm_stripe_count);
1507                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1508                                 CERROR("lsm mismatch for inode %ld\n",
1509                                        inode->i_ino);
1510                                 CERROR("lli_smd:\n");
1511                                 dump_lsm(D_ERROR, lli->lli_smd);
1512                                 CERROR("lsm:\n");
1513                                 dump_lsm(D_ERROR, lsm);
1514                                 LBUG();
1515                         }
1516                 }
1517                 if (lli->lli_smd != lsm)
1518                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1519         }
1520
1521         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1522                 if (body->valid & OBD_MD_FLRMTPERM)
1523                         ll_update_remote_perm(inode, md->remote_perm);
1524         }
1525 #ifdef CONFIG_FS_POSIX_ACL
1526         else if (body->valid & OBD_MD_FLACL) {
1527                 cfs_spin_lock(&lli->lli_lock);
1528                 if (lli->lli_posix_acl)
1529                         posix_acl_release(lli->lli_posix_acl);
1530                 lli->lli_posix_acl = md->posix_acl;
1531                 cfs_spin_unlock(&lli->lli_lock);
1532         }
1533 #endif
1534         inode->i_ino = cl_fid_build_ino(&body->fid1);
1535         inode->i_generation = cl_fid_build_gen(&body->fid1);
1536
1537         if (body->valid & OBD_MD_FLATIME) {
1538                 if (body->atime > LTIME_S(inode->i_atime))
1539                         LTIME_S(inode->i_atime) = body->atime;
1540                 lli->lli_lvb.lvb_atime = body->atime;
1541         }
1542         if (body->valid & OBD_MD_FLMTIME) {
1543                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1544                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1545                                "to "LPU64"\n", inode->i_ino,
1546                                LTIME_S(inode->i_mtime), body->mtime);
1547                         LTIME_S(inode->i_mtime) = body->mtime;
1548                 }
1549                 lli->lli_lvb.lvb_mtime = body->mtime;
1550         }
1551         if (body->valid & OBD_MD_FLCTIME) {
1552                 if (body->ctime > LTIME_S(inode->i_ctime))
1553                         LTIME_S(inode->i_ctime) = body->ctime;
1554                 lli->lli_lvb.lvb_ctime = body->ctime;
1555         }
1556         if (body->valid & OBD_MD_FLMODE)
1557                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1558         if (body->valid & OBD_MD_FLTYPE)
1559                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1560         LASSERT(inode->i_mode != 0);
1561         if (S_ISREG(inode->i_mode)) {
1562                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1563         } else {
1564                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1565         }
1566 #ifdef HAVE_INODE_BLKSIZE
1567         inode->i_blksize = 1<<inode->i_blkbits;
1568 #endif
1569         if (body->valid & OBD_MD_FLUID)
1570                 inode->i_uid = body->uid;
1571         if (body->valid & OBD_MD_FLGID)
1572                 inode->i_gid = body->gid;
1573         if (body->valid & OBD_MD_FLFLAGS)
1574                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1575         if (body->valid & OBD_MD_FLNLINK)
1576                 inode->i_nlink = body->nlink;
1577         if (body->valid & OBD_MD_FLRDEV)
1578                 inode->i_rdev = old_decode_dev(body->rdev);
1579
1580         if (body->valid & OBD_MD_FLID) {
1581                 /* FID shouldn't be changed! */
1582                 if (fid_is_sane(&lli->lli_fid)) {
1583                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1584                                  "Trying to change FID "DFID
1585                                  " to the "DFID", inode %lu/%u(%p)\n",
1586                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1587                                  inode->i_ino, inode->i_generation, inode);
1588                 } else
1589                         lli->lli_fid = body->fid1;
1590         }
1591
1592         LASSERT(fid_seq(&lli->lli_fid) != 0);
1593
1594         if (body->valid & OBD_MD_FLSIZE) {
1595                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1596                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1597                         struct lustre_handle lockh;
1598                         ldlm_mode_t mode;
1599
1600                         /* As it is possible a blocking ast has been processed
1601                          * by this time, we need to check there is an UPDATE
1602                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1603                          * it. */
1604                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1605                                                &lockh);
1606                         if (mode) {
1607                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1608                                                       LLIF_EPOCH_PENDING |
1609                                                       LLIF_SOM_DIRTY)) {
1610                                         CERROR("ino %lu flags %lu still has "
1611                                                "size authority! do not trust "
1612                                                "the size got from MDS\n",
1613                                                inode->i_ino, lli->lli_flags);
1614                                 } else {
1615                                         /* Use old size assignment to avoid
1616                                          * deadlock bz14138 & bz14326 */
1617                                         inode->i_size = body->size;
1618                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1619                                 }
1620                                 ldlm_lock_decref(&lockh, mode);
1621                         }
1622                 } else {
1623                         /* Use old size assignment to avoid
1624                          * deadlock bz14138 & bz14326 */
1625                         inode->i_size = body->size;
1626
1627                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1628                                inode->i_ino, (unsigned long long)body->size);
1629                 }
1630
1631                 if (body->valid & OBD_MD_FLBLOCKS)
1632                         inode->i_blocks = body->blocks;
1633         }
1634
1635         if (body->valid & OBD_MD_FLMDSCAPA) {
1636                 LASSERT(md->mds_capa);
1637                 ll_add_capa(inode, md->mds_capa);
1638         }
1639         if (body->valid & OBD_MD_FLOSSCAPA) {
1640                 LASSERT(md->oss_capa);
1641                 ll_add_capa(inode, md->oss_capa);
1642         }
1643 }
1644
1645 void ll_read_inode2(struct inode *inode, void *opaque)
1646 {
1647         struct lustre_md *md = opaque;
1648         struct ll_inode_info *lli = ll_i2info(inode);
1649         ENTRY;
1650
1651         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1652                inode->i_ino, inode->i_generation, inode);
1653
1654         ll_lli_init(lli);
1655
1656         LASSERT(!lli->lli_smd);
1657
1658         /* Core attributes from the MDS first.  This is a new inode, and
1659          * the VFS doesn't zero times in the core inode so we have to do
1660          * it ourselves.  They will be overwritten by either MDS or OST
1661          * attributes - we just need to make sure they aren't newer. */
1662         LTIME_S(inode->i_mtime) = 0;
1663         LTIME_S(inode->i_atime) = 0;
1664         LTIME_S(inode->i_ctime) = 0;
1665         inode->i_rdev = 0;
1666         ll_update_inode(inode, md);
1667
1668         /* OIDEBUG(inode); */
1669
1670         /* initializing backing dev info. */
1671         inode->i_mapping->backing_dev_info = &(s2lsi(inode->i_sb)->bdi);
1672
1673
1674         if (S_ISREG(inode->i_mode)) {
1675                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1676                 inode->i_op = &ll_file_inode_operations;
1677                 inode->i_fop = sbi->ll_fop;
1678                 inode->i_mapping->a_ops = &ll_aops;
1679                 EXIT;
1680         } else if (S_ISDIR(inode->i_mode)) {
1681                 inode->i_op = &ll_dir_inode_operations;
1682                 inode->i_fop = &ll_dir_operations;
1683                 inode->i_mapping->a_ops = &ll_dir_aops;
1684                 EXIT;
1685         } else if (S_ISLNK(inode->i_mode)) {
1686                 inode->i_op = &ll_fast_symlink_inode_operations;
1687                 EXIT;
1688         } else {
1689                 inode->i_op = &ll_special_inode_operations;
1690
1691                 init_special_inode(inode, inode->i_mode,
1692                                    kdev_t_to_nr(inode->i_rdev));
1693
1694                 EXIT;
1695         }
1696 }
1697
1698 void ll_delete_inode(struct inode *inode)
1699 {
1700         struct ll_sb_info *sbi = ll_i2sbi(inode);
1701         int rc;
1702         ENTRY;
1703
1704         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1705         if (rc) {
1706                 CERROR("fid_delete() failed, rc %d\n", rc);
1707         }
1708         truncate_inode_pages(&inode->i_data, 0);
1709         clear_inode(inode);
1710
1711         EXIT;
1712 }
1713
1714 int ll_iocontrol(struct inode *inode, struct file *file,
1715                  unsigned int cmd, unsigned long arg)
1716 {
1717         struct ll_sb_info *sbi = ll_i2sbi(inode);
1718         struct ptlrpc_request *req = NULL;
1719         int rc, flags = 0;
1720         ENTRY;
1721
1722         switch(cmd) {
1723         case FSFILT_IOC_GETFLAGS: {
1724                 struct mdt_body *body;
1725                 struct md_op_data *op_data;
1726
1727                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1728                                              0, 0, LUSTRE_OPC_ANY,
1729                                              NULL);
1730                 if (op_data == NULL)
1731                         RETURN(-ENOMEM);
1732
1733                 op_data->op_valid = OBD_MD_FLFLAGS;
1734                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1735                 ll_finish_md_op_data(op_data);
1736                 if (rc) {
1737                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1738                         RETURN(-abs(rc));
1739                 }
1740
1741                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1742
1743                 flags = body->flags;
1744
1745                 ptlrpc_req_finished(req);
1746
1747                 RETURN(put_user(flags, (int *)arg));
1748         }
1749         case FSFILT_IOC_SETFLAGS: {
1750                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1751                 struct obd_info oinfo = { { { 0 } } };
1752                 struct md_op_data *op_data;
1753
1754                 if (get_user(flags, (int *)arg))
1755                         RETURN(-EFAULT);
1756
1757                 oinfo.oi_md = lsm;
1758                 OBDO_ALLOC(oinfo.oi_oa);
1759                 if (!oinfo.oi_oa)
1760                         RETURN(-ENOMEM);
1761
1762                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1763                                              LUSTRE_OPC_ANY, NULL);
1764                 if (IS_ERR(op_data))
1765                         RETURN(PTR_ERR(op_data));
1766
1767                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1768                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1769                 rc = md_setattr(sbi->ll_md_exp, op_data,
1770                                 NULL, 0, NULL, 0, &req, NULL);
1771                 ll_finish_md_op_data(op_data);
1772                 ptlrpc_req_finished(req);
1773                 if (rc) {
1774                         OBDO_FREE(oinfo.oi_oa);
1775                         RETURN(rc);
1776                 }
1777
1778                 if (lsm == NULL) {
1779                         OBDO_FREE(oinfo.oi_oa);
1780                         GOTO(update_cache, rc);
1781                 }
1782
1783                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1784                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1785                 oinfo.oi_oa->o_flags = flags;
1786                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1787                                        OBD_MD_FLGROUP;
1788                 oinfo.oi_capa = ll_mdscapa_get(inode);
1789                 obdo_from_inode(oinfo.oi_oa, inode,
1790                                 &ll_i2info(inode)->lli_fid, 0);
1791                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1792                 capa_put(oinfo.oi_capa);
1793                 OBDO_FREE(oinfo.oi_oa);
1794                 if (rc) {
1795                         if (rc != -EPERM && rc != -EACCES)
1796                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1797                         RETURN(rc);
1798                 }
1799
1800                 EXIT;
1801 update_cache:
1802                 inode->i_flags = ll_ext_to_inode_flags(flags);
1803                 return 0;
1804         }
1805         default:
1806                 RETURN(-ENOSYS);
1807         }
1808
1809         RETURN(0);
1810 }
1811
1812 int ll_flush_ctx(struct inode *inode)
1813 {
1814         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1815
1816         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1817
1818         obd_set_info_async(sbi->ll_md_exp,
1819                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1820                            0, NULL, NULL);
1821         obd_set_info_async(sbi->ll_dt_exp,
1822                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1823                            0, NULL, NULL);
1824         return 0;
1825 }
1826
1827 /* umount -f client means force down, don't save state */
1828 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1829 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1830 {
1831         struct super_block *sb = vfsmnt->mnt_sb;
1832 #else
1833 void ll_umount_begin(struct super_block *sb)
1834 {
1835 #endif
1836         struct lustre_sb_info *lsi = s2lsi(sb);
1837         struct ll_sb_info *sbi = ll_s2sbi(sb);
1838         struct obd_device *obd;
1839         struct obd_ioctl_data *ioc_data;
1840         ENTRY;
1841
1842 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1843         if (!(flags & MNT_FORCE)) {
1844                 EXIT;
1845                 return;
1846         }
1847 #endif
1848
1849         /* Tell the MGC we got umount -f */
1850         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1851
1852         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1853                sb->s_count, atomic_read(&sb->s_active));
1854
1855         obd = class_exp2obd(sbi->ll_md_exp);
1856         if (obd == NULL) {
1857                 CERROR("Invalid MDC connection handle "LPX64"\n",
1858                        sbi->ll_md_exp->exp_handle.h_cookie);
1859                 EXIT;
1860                 return;
1861         }
1862         obd->obd_force = 1;
1863
1864         obd = class_exp2obd(sbi->ll_dt_exp);
1865         if (obd == NULL) {
1866                 CERROR("Invalid LOV connection handle "LPX64"\n",
1867                        sbi->ll_dt_exp->exp_handle.h_cookie);
1868                 EXIT;
1869                 return;
1870         }
1871         obd->obd_force = 1;
1872
1873         OBD_ALLOC_PTR(ioc_data);
1874         if (ioc_data) {
1875                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1876                               sizeof ioc_data, ioc_data, NULL);
1877
1878                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1879                               sizeof ioc_data, ioc_data, NULL);
1880
1881                 OBD_FREE_PTR(ioc_data);
1882         }
1883
1884
1885         /* Really, we'd like to wait until there are no requests outstanding,
1886          * and then continue.  For now, we just invalidate the requests,
1887          * schedule() and sleep one second if needed, and hope.
1888          */
1889         cfs_schedule();
1890 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1891         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1892                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1893                                                    cfs_time_seconds(1));
1894                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1895                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1896                                       "may try to umount it a bit later\n",
1897                                       atomic_read(&vfsmnt->mnt_count));
1898         }
1899 #endif
1900
1901         EXIT;
1902 }
1903
1904 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1905 {
1906         struct ll_sb_info *sbi = ll_s2sbi(sb);
1907         int err;
1908         __u32 read_only;
1909
1910         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1911                 read_only = *flags & MS_RDONLY;
1912                 err = obd_set_info_async(sbi->ll_md_exp,
1913                                          sizeof(KEY_READ_ONLY),
1914                                          KEY_READ_ONLY, sizeof(read_only),
1915                                          &read_only, NULL);
1916                 if (err) {
1917                         CERROR("Failed to change the read-only flag during "
1918                                "remount: %d\n", err);
1919                         return err;
1920                 }
1921
1922                 if (read_only)
1923                         sb->s_flags |= MS_RDONLY;
1924                 else
1925                         sb->s_flags &= ~MS_RDONLY;
1926         }
1927         return 0;
1928 }
1929
1930 int ll_prep_inode(struct inode **inode,
1931                   struct ptlrpc_request *req,
1932                   struct super_block *sb)
1933 {
1934         struct ll_sb_info *sbi = NULL;
1935         struct lustre_md md;
1936         int rc;
1937         ENTRY;
1938
1939         LASSERT(*inode || sb);
1940         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1941         memset(&md, 0, sizeof(struct lustre_md));
1942
1943         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1944                               sbi->ll_md_exp, &md);
1945         if (rc)
1946                 RETURN(rc);
1947
1948         if (*inode) {
1949                 ll_update_inode(*inode, &md);
1950         } else {
1951                 LASSERT(sb != NULL);
1952
1953                 /*
1954                  * At this point server returns to client's same fid as client
1955                  * generated for creating. So using ->fid1 is okay here.
1956                  */
1957                 LASSERT(fid_is_sane(&md.body->fid1));
1958
1959                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1960                 if (*inode == NULL || IS_ERR(*inode)) {
1961                         if (md.lsm)
1962                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1963 #ifdef CONFIG_FS_POSIX_ACL
1964                         if (md.posix_acl) {
1965                                 posix_acl_release(md.posix_acl);
1966                                 md.posix_acl = NULL;
1967                         }
1968 #endif
1969                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1970                         *inode = NULL;
1971                         CERROR("new_inode -fatal: rc %d\n", rc);
1972                         GOTO(out, rc);
1973                 }
1974         }
1975
1976 out:
1977         md_free_lustre_md(sbi->ll_md_exp, &md);
1978         RETURN(rc);
1979 }
1980
1981 int ll_obd_statfs(struct inode *inode, void *arg)
1982 {
1983         struct ll_sb_info *sbi = NULL;
1984         struct obd_export *exp;
1985         char *buf = NULL;
1986         struct obd_ioctl_data *data = NULL;
1987         __u32 type;
1988         int len = 0, rc;
1989
1990         if (!inode || !(sbi = ll_i2sbi(inode)))
1991                 GOTO(out_statfs, rc = -EINVAL);
1992
1993         rc = obd_ioctl_getdata(&buf, &len, arg);
1994         if (rc)
1995                 GOTO(out_statfs, rc);
1996
1997         data = (void*)buf;
1998         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1999             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2000                 GOTO(out_statfs, rc = -EINVAL);
2001
2002         if (data->ioc_inllen1 != sizeof(__u32) ||
2003             data->ioc_inllen2 != sizeof(__u32) ||
2004             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2005             data->ioc_plen2 != sizeof(struct obd_uuid))
2006                 GOTO(out_statfs, rc = -EINVAL);
2007
2008         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2009         if (type == LL_STATFS_MDC)
2010                 exp = sbi->ll_md_exp;
2011         else if (type == LL_STATFS_LOV)
2012                 exp = sbi->ll_dt_exp;
2013         else
2014                 GOTO(out_statfs, rc = -ENODEV);
2015
2016         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2017         if (rc)
2018                 GOTO(out_statfs, rc);
2019 out_statfs:
2020         if (buf)
2021                 obd_ioctl_freedata(buf, len);
2022         return rc;
2023 }
2024
2025 int ll_process_config(struct lustre_cfg *lcfg)
2026 {
2027         char *ptr;
2028         void *sb;
2029         struct lprocfs_static_vars lvars;
2030         unsigned long x;
2031         int rc = 0;
2032
2033         lprocfs_llite_init_vars(&lvars);
2034
2035         /* The instance name contains the sb: lustre-client-aacfe000 */
2036         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2037         if (!ptr || !*(++ptr))
2038                 return -EINVAL;
2039         if (sscanf(ptr, "%lx", &x) != 1)
2040                 return -EINVAL;
2041         sb = (void *)x;
2042         /* This better be a real Lustre superblock! */
2043         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2044
2045         /* Note we have not called client_common_fill_super yet, so
2046            proc fns must be able to handle that! */
2047         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2048                                       lcfg, sb);
2049         if (rc > 0)
2050                 rc = 0;
2051         return(rc);
2052 }
2053
2054 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2055 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2056                                        struct inode *i1, struct inode *i2,
2057                                        const char *name, int namelen,
2058                                        int mode, __u32 opc, void *data)
2059 {
2060         LASSERT(i1 != NULL);
2061
2062         if (namelen > ll_i2sbi(i1)->ll_namelen)
2063                 return ERR_PTR(-ENAMETOOLONG);
2064
2065         if (op_data == NULL)
2066                 OBD_ALLOC_PTR(op_data);
2067
2068         if (op_data == NULL)
2069                 return ERR_PTR(-ENOMEM);
2070
2071         ll_i2gids(op_data->op_suppgids, i1, i2);
2072         op_data->op_fid1 = *ll_inode2fid(i1);
2073         op_data->op_capa1 = ll_mdscapa_get(i1);
2074
2075         if (i2) {
2076                 op_data->op_fid2 = *ll_inode2fid(i2);
2077                 op_data->op_capa2 = ll_mdscapa_get(i2);
2078         } else {
2079                 fid_zero(&op_data->op_fid2);
2080                 op_data->op_capa2 = NULL;
2081         }
2082
2083         op_data->op_name = name;
2084         op_data->op_namelen = namelen;
2085         op_data->op_mode = mode;
2086         op_data->op_mod_time = cfs_time_current_sec();
2087         op_data->op_fsuid = cfs_curproc_fsuid();
2088         op_data->op_fsgid = cfs_curproc_fsgid();
2089         op_data->op_cap = cfs_curproc_cap_pack();
2090         op_data->op_bias = MDS_CHECK_SPLIT;
2091         op_data->op_opc = opc;
2092         op_data->op_mds = 0;
2093         op_data->op_data = data;
2094
2095         return op_data;
2096 }
2097
2098 void ll_finish_md_op_data(struct md_op_data *op_data)
2099 {
2100         capa_put(op_data->op_capa1);
2101         capa_put(op_data->op_capa2);
2102         OBD_FREE_PTR(op_data);
2103 }
2104
2105 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2106 {
2107         struct ll_sb_info *sbi;
2108
2109         LASSERT((seq != NULL) && (vfs != NULL));
2110         sbi = ll_s2sbi(vfs->mnt_sb);
2111
2112         if (sbi->ll_flags & LL_SBI_NOLCK)
2113                 seq_puts(seq, ",nolock");
2114
2115         if (sbi->ll_flags & LL_SBI_FLOCK)
2116                 seq_puts(seq, ",flock");
2117
2118         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2119                 seq_puts(seq, ",localflock");
2120
2121         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2122                 seq_puts(seq, ",user_xattr");
2123
2124         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2125                 seq_puts(seq, ",lazystatfs");
2126
2127         RETURN(0);
2128 }