lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) cfs_ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         cfs_spin_lock_init(&sbi->ll_lock);
  89         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  90         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  91         cfs_spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  97 #ifdef HAVE_BGL_SUPPORT
  98                 lru_page_max = pages / 4;
  99 #else
 100                 lru_page_max = pages / 2;
 101 #endif
 102         } else {
 103                 lru_page_max = (pages / 4) * 3;
 104         }
 105
 106         /* initialize lru data */
 107         cfs_atomic_set(&sbi->ll_lru.ccl_users, 0);
 108         sbi->ll_lru.ccl_page_max = lru_page_max;
 109         cfs_atomic_set(&sbi->ll_lru.ccl_page_left, lru_page_max);
 110         cfs_spin_lock_init(&sbi->ll_lru.ccl_lock);
 111         CFS_INIT_LIST_HEAD(&sbi->ll_lru.ccl_list);
 112
 113         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 114                                            SBI_DEFAULT_READAHEAD_MAX);
 115         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 117                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 118         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 119         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 120
 121         ll_generate_random_uuid(uuid);
 122         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 123         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 124
 125         cfs_spin_lock(&ll_sb_lock);
 126         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 127         cfs_spin_unlock(&ll_sb_lock);
 128
 129         sbi->ll_flags |= LL_SBI_VERBOSE;
 130 #ifdef ENABLE_CHECKSUM
 131         sbi->ll_flags |= LL_SBI_CHECKSUM;
 132 #endif
 133
 134 #ifdef HAVE_LRU_RESIZE_SUPPORT
 135         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 136 #endif
 137
 138         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 139                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 140                                    pp_r_hist.oh_lock);
 141                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 142                                    pp_w_hist.oh_lock);
 143         }
 144
 145         /* metadata statahead is enabled by default */
 146         sbi->ll_sa_max = LL_SA_RPC_DEF;
 147         cfs_atomic_set(&sbi->ll_sa_total, 0);
 148         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 149         cfs_atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151
 152         RETURN(sbi);
 153 }
 154
 155 void ll_free_sbi(struct super_block *sb)
 156 {
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         ENTRY;
 159
 160         if (sbi != NULL) {
 161                 cfs_spin_lock(&ll_sb_lock);
 162                 cfs_list_del(&sbi->ll_list);
 163                 cfs_spin_unlock(&ll_sb_lock);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static struct dentry_operations ll_d_root_ops = {
 170         .d_compare = ll_dcompare,
 171         .d_revalidate = ll_revalidate_nd,
 172 };
 173
 174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 175                                     struct vfsmount *mnt)
 176 {
 177         struct inode *root = 0;
 178         struct ll_sb_info *sbi = ll_s2sbi(sb);
 179         struct obd_device *obd;
 180         struct obd_capa *oc = NULL;
 181         struct obd_statfs *osfs = NULL;
 182         struct ptlrpc_request *request = NULL;
 183         struct obd_connect_data *data = NULL;
 184         struct obd_uuid *uuid;
 185         struct md_op_data *op_data;
 186         struct lustre_md lmd;
 187         obd_valid valid;
 188         int size, err, checksum;
 189         ENTRY;
 190
 191         obd = class_name2obd(md);
 192         if (!obd) {
 193                 CERROR("MD %s: not setup or attached\n", md);
 194                 RETURN(-EINVAL);
 195         }
 196
 197         OBD_ALLOC_PTR(data);
 198         if (data == NULL)
 199                 RETURN(-ENOMEM);
 200
 201         OBD_ALLOC_PTR(osfs);
 202         if (osfs == NULL) {
 203                 OBD_FREE_PTR(data);
 204                 RETURN(-ENOMEM);
 205         }
 206
 207         if (proc_lustre_fs_root) {
 208                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 209                                                   dt, md);
 210                 if (err < 0)
 211                         CERROR("could not register mount in /proc/fs/lustre\n");
 212         }
 213
 214         /* indicate the features supported by this client */
 215         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 216                                   OBD_CONNECT_ATTRFID  |
 217                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 218                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 219                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 220                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 221                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 222                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 223                                   OBD_CONNECT_EINPROGRESS |
 224                                   OBD_CONNECT_JOBSTATS;
 225
 226         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 227                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 228
 229 #ifdef HAVE_LRU_RESIZE_SUPPORT
 230         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 231                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 232 #endif
 233 #ifdef CONFIG_FS_POSIX_ACL
 234         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 235 #endif
 236
 237         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 238                 /* flag mdc connection as lightweight, only used for test
 239                  * purpose, use with care */
 240                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 241
 242         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 243         data->ocd_version = LUSTRE_VERSION_CODE;
 244
 245         if (sb->s_flags & MS_RDONLY)
 246                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 247         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 248                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 249
 250 #ifdef HAVE_MS_FLOCK_LOCK
 251         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 252         sb->s_flags |= MS_FLOCK_LOCK;
 253 #endif
 254 #ifdef MS_HAS_NEW_AOPS
 255         sb->s_flags |= MS_HAS_NEW_AOPS;
 256 #endif
 257
 258         if (sbi->ll_flags & LL_SBI_FLOCK)
 259                 sbi->ll_fop = &ll_file_operations_flock;
 260         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 261                 sbi->ll_fop = &ll_file_operations;
 262         else
 263                 sbi->ll_fop = &ll_file_operations_noflock;
 264
 265         /* real client */
 266         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 267         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 268                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 269
 270         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 271
 272         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 273         if (err == -EBUSY) {
 274                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 275                                    "recovery, of which this client is not a "
 276                                    "part. Please wait for recovery to complete,"
 277                                    " abort, or time out.\n", md);
 278                 GOTO(out, err);
 279         } else if (err) {
 280                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 281                 GOTO(out, err);
 282         }
 283
 284         err = obd_fid_init(sbi->ll_md_exp);
 285         if (err) {
 286                 CERROR("Can't init metadata layer FID infrastructure, "
 287                        "rc %d\n", err);
 288                 GOTO(out_md, err);
 289         }
 290
 291         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 292                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 293         if (err)
 294                 GOTO(out_md_fid, err);
 295
 296         /* This needs to be after statfs to ensure connect has finished.
 297          * Note that "data" does NOT contain the valid connect reply.
 298          * If connecting to a 1.8 server there will be no LMV device, so
 299          * we can access the MDC export directly and exp_connect_flags will
 300          * be non-zero, but if accessing an upgraded 2.1 server it will
 301          * have the correct flags filled in.
 302          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 303         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 304         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 305             valid != CLIENT_CONNECT_MDT_REQD) {
 306                 char *buf;
 307
 308                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 309                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 310                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 311                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 312                                    "feature(s) needed for correct operation "
 313                                    "of this client (%s). Please upgrade "
 314                                    "server or downgrade client.\n",
 315                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 316                 OBD_FREE(buf, CFS_PAGE_SIZE);
 317                 GOTO(out_md, err = -EPROTO);
 318         }
 319
 320         size = sizeof(*data);
 321         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 322                            KEY_CONN_DATA,  &size, data, NULL);
 323         if (err) {
 324                 CERROR("Get connect data failed: %d \n", err);
 325                 GOTO(out_md, err);
 326         }
 327
 328         LASSERT(osfs->os_bsize);
 329         sb->s_blocksize = osfs->os_bsize;
 330         sb->s_blocksize_bits = log2(osfs->os_bsize);
 331         sb->s_magic = LL_SUPER_MAGIC;
 332         sb->s_maxbytes = MAX_LFS_FILESIZE;
 333         sbi->ll_namelen = osfs->os_namelen;
 334         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 335
 336         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 337             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 338                 LCONSOLE_INFO("Disabling user_xattr feature because "
 339                               "it is not supported on the server\n");
 340                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 344 #ifdef MS_POSIXACL
 345                 sb->s_flags |= MS_POSIXACL;
 346 #endif
 347                 sbi->ll_flags |= LL_SBI_ACL;
 348         } else {
 349                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 350 #ifdef MS_POSIXACL
 351                 sb->s_flags &= ~MS_POSIXACL;
 352 #endif
 353                 sbi->ll_flags &= ~LL_SBI_ACL;
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 357                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 358                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client is set as remote by default.\n");
 360                 }
 361         } else {
 362                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 363                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client claims to be remote, but server "
 365                                       "rejected, forced to be local.\n");
 366                 }
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 370                 LCONSOLE_INFO("client enabled MDS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 375                 LCONSOLE_INFO("client enabled OSS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 380                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 383                 sbi->ll_md_brw_size = data->ocd_brw_size;
 384         else
 385                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 388                 LCONSOLE_INFO("Layout lock feature supported.\n");
 389                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 390         }
 391
 392         obd = class_name2obd(dt);
 393         if (!obd) {
 394                 CERROR("DT %s: not setup or attached\n", dt);
 395                 GOTO(out_md_fid, err = -ENODEV);
 396         }
 397
 398         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 399                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 400                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 401                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 402                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 403                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 404                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 405                                   OBD_CONNECT_MAXBYTES |
 406                                   OBD_CONNECT_EINPROGRESS |
 407                                   OBD_CONNECT_JOBSTATS;
 408
 409         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 410                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 411
 412         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 413                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 414                  * disabled by default, because it can still be enabled on the
 415                  * fly via /proc. As a consequence, we still need to come to an
 416                  * agreement on the supported algorithms at connect time */
 417                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 418
 419                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 420                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 421                 else
 422                         data->ocd_cksum_types = cksum_types_supported_client();
 423         }
 424
 425 #ifdef HAVE_LRU_RESIZE_SUPPORT
 426         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 427 #endif
 428         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 429                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 430
 431         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 432                "ocd_grant: %d\n", data->ocd_connect_flags,
 433                data->ocd_version, data->ocd_grant);
 434
 435         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 436         obd->obd_upcall.onu_upcall = cl_ocd_update;
 437
 438         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 439
 440         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 441         if (err == -EBUSY) {
 442                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 443                                    "recovery, of which this client is not a "
 444                                    "part.  Please wait for recovery to "
 445                                    "complete, abort, or time out.\n", dt);
 446                 GOTO(out_md_fid, err);
 447         } else if (err) {
 448                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 449                 GOTO(out_md_fid, err);
 450         }
 451
 452         err = obd_fid_init(sbi->ll_dt_exp);
 453         if (err) {
 454                 CERROR("Can't init data layer FID infrastructure, "
 455                        "rc %d\n", err);
 456                 GOTO(out_dt, err);
 457         }
 458
 459         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 460         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 461         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 462         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 463         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 464
 465         fid_zero(&sbi->ll_root_fid);
 466         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 467         if (err) {
 468                 CERROR("cannot mds_connect: rc = %d\n", err);
 469                 GOTO(out_lock_cn_cb, err);
 470         }
 471         if (!fid_is_sane(&sbi->ll_root_fid)) {
 472                 CERROR("Invalid root fid during mount\n");
 473                 GOTO(out_lock_cn_cb, err = -EINVAL);
 474         }
 475         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 476
 477         sb->s_op = &lustre_super_operations;
 478 #if THREAD_SIZE >= 8192 /*b=17630*/
 479         sb->s_export_op = &lustre_export_operations;
 480 #endif
 481
 482         /* make root inode
 483          * XXX: move this to after cbd setup? */
 484         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 485         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 486                 valid |= OBD_MD_FLRMTPERM;
 487         else if (sbi->ll_flags & LL_SBI_ACL)
 488                 valid |= OBD_MD_FLACL;
 489
 490         OBD_ALLOC_PTR(op_data);
 491         if (op_data == NULL)
 492                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 493
 494         op_data->op_fid1 = sbi->ll_root_fid;
 495         op_data->op_mode = 0;
 496         op_data->op_capa1 = oc;
 497         op_data->op_valid = valid;
 498
 499         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 500         if (oc)
 501                 capa_put(oc);
 502         OBD_FREE_PTR(op_data);
 503         if (err) {
 504                 CERROR("md_getattr failed for root: rc = %d\n", err);
 505                 GOTO(out_lock_cn_cb, err);
 506         }
 507         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 508                                sbi->ll_md_exp, &lmd);
 509         if (err) {
 510                 CERROR("failed to understand root inode md: rc = %d\n", err);
 511                 ptlrpc_req_finished (request);
 512                 GOTO(out_lock_cn_cb, err);
 513         }
 514
 515         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 516         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 517         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 518         ptlrpc_req_finished(request);
 519
 520         if (root == NULL || IS_ERR(root)) {
 521                 if (lmd.lsm)
 522                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 523 #ifdef CONFIG_FS_POSIX_ACL
 524                 if (lmd.posix_acl) {
 525                         posix_acl_release(lmd.posix_acl);
 526                         lmd.posix_acl = NULL;
 527                 }
 528 #endif
 529                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 530                 root = NULL;
 531                 CERROR("lustre_lite: bad iget4 for root\n");
 532                 GOTO(out_root, err);
 533         }
 534
 535         err = ll_close_thread_start(&sbi->ll_lcq);
 536         if (err) {
 537                 CERROR("cannot start close thread: rc %d\n", err);
 538                 GOTO(out_root, err);
 539         }
 540
 541 #ifdef CONFIG_FS_POSIX_ACL
 542         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 543                 rct_init(&sbi->ll_rct);
 544                 et_init(&sbi->ll_et);
 545         }
 546 #endif
 547
 548         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 549         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 550                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 551                                  NULL);
 552         cl_sb_init(sb);
 553
 554         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_LRU_SET),
 555                                  KEY_LRU_SET, sizeof(sbi->ll_lru),
 556                                  &sbi->ll_lru, NULL);
 557
 558         sb->s_root = d_alloc_root(root);
 559 #ifdef HAVE_DCACHE_LOCK
 560         sb->s_root->d_op = &ll_d_root_ops;
 561 #else
 562         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 563         d_set_d_op(sb->s_root, &ll_d_root_ops);
 564         sb->s_d_op = &ll_d_ops;
 565 #endif
 566
 567         sbi->ll_sdev_orig = sb->s_dev;
 568
 569         /* We set sb->s_dev equal on all lustre clients in order to support
 570          * NFS export clustering.  NFSD requires that the FSID be the same
 571          * on all clients. */
 572         /* s_dev is also used in lt_compare() to compare two fs, but that is
 573          * only a node-local comparison. */
 574         uuid = obd_get_uuid(sbi->ll_md_exp);
 575         if (uuid != NULL)
 576                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 577
 578         if (data != NULL)
 579                 OBD_FREE_PTR(data);
 580         if (osfs != NULL)
 581                 OBD_FREE_PTR(osfs);
 582
 583         RETURN(err);
 584 out_root:
 585         if (root)
 586                 iput(root);
 587 out_lock_cn_cb:
 588         obd_fid_fini(sbi->ll_dt_exp);
 589 out_dt:
 590         obd_disconnect(sbi->ll_dt_exp);
 591         sbi->ll_dt_exp = NULL;
 592 out_md_fid:
 593         obd_fid_fini(sbi->ll_md_exp);
 594 out_md:
 595         obd_disconnect(sbi->ll_md_exp);
 596         sbi->ll_md_exp = NULL;
 597 out:
 598         if (data != NULL)
 599                 OBD_FREE_PTR(data);
 600         if (osfs != NULL)
 601                 OBD_FREE_PTR(osfs);
 602         lprocfs_unregister_mountpoint(sbi);
 603         return err;
 604 }
 605
 606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 607 {
 608         int size, rc;
 609
 610         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 611         size = sizeof(int);
 612         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 613                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 614         if (rc)
 615                 CERROR("Get max mdsize error rc %d \n", rc);
 616
 617         RETURN(rc);
 618 }
 619
 620 void ll_dump_inode(struct inode *inode)
 621 {
 622         struct list_head *tmp;
 623         int dentry_count = 0;
 624
 625         LASSERT(inode != NULL);
 626
 627         list_for_each(tmp, &inode->i_dentry)
 628                 dentry_count++;
 629
 630         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 631                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 632                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 633 }
 634
 635 void lustre_dump_dentry(struct dentry *dentry, int recur)
 636 {
 637         struct list_head *tmp;
 638         int subdirs = 0;
 639
 640         LASSERT(dentry != NULL);
 641
 642         list_for_each(tmp, &dentry->d_subdirs)
 643                 subdirs++;
 644
 645         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 646                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 647                dentry->d_name.len, dentry->d_name.name,
 648                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 649                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 650                dentry->d_flags, dentry->d_fsdata, subdirs);
 651         if (dentry->d_inode != NULL)
 652                 ll_dump_inode(dentry->d_inode);
 653
 654         if (recur == 0)
 655                 return;
 656
 657         list_for_each(tmp, &dentry->d_subdirs) {
 658                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 659                 lustre_dump_dentry(d, recur - 1);
 660         }
 661 }
 662
 663 void client_common_put_super(struct super_block *sb)
 664 {
 665         struct ll_sb_info *sbi = ll_s2sbi(sb);
 666         ENTRY;
 667
 668 #ifdef CONFIG_FS_POSIX_ACL
 669         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 670                 et_fini(&sbi->ll_et);
 671                 rct_fini(&sbi->ll_rct);
 672         }
 673 #endif
 674
 675         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 676
 677         ll_close_thread_shutdown(sbi->ll_lcq);
 678
 679         cl_sb_fini(sb);
 680
 681         cfs_list_del(&sbi->ll_conn_chain);
 682
 683         obd_fid_fini(sbi->ll_dt_exp);
 684         obd_disconnect(sbi->ll_dt_exp);
 685         sbi->ll_dt_exp = NULL;
 686
 687         lprocfs_unregister_mountpoint(sbi);
 688
 689         obd_fid_fini(sbi->ll_md_exp);
 690         obd_disconnect(sbi->ll_md_exp);
 691         sbi->ll_md_exp = NULL;
 692
 693         EXIT;
 694 }
 695
 696 void ll_kill_super(struct super_block *sb)
 697 {
 698         struct ll_sb_info *sbi;
 699
 700         ENTRY;
 701
 702         /* not init sb ?*/
 703         if (!(sb->s_flags & MS_ACTIVE))
 704                 return;
 705
 706         sbi = ll_s2sbi(sb);
 707         /* we need restore s_dev from changed for clustred NFS before put_super
 708          * because new kernels have cached s_dev and change sb->s_dev in
 709          * put_super not affected real removing devices */
 710         if (sbi)
 711                 sb->s_dev = sbi->ll_sdev_orig;
 712         EXIT;
 713 }
 714
 715 char *ll_read_opt(const char *opt, char *data)
 716 {
 717         char *value;
 718         char *retval;
 719         ENTRY;
 720
 721         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 722         if (strncmp(opt, data, strlen(opt)))
 723                 RETURN(NULL);
 724         if ((value = strchr(data, '=')) == NULL)
 725                 RETURN(NULL);
 726
 727         value++;
 728         OBD_ALLOC(retval, strlen(value) + 1);
 729         if (!retval) {
 730                 CERROR("out of memory!\n");
 731                 RETURN(NULL);
 732         }
 733
 734         memcpy(retval, value, strlen(value)+1);
 735         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 736         RETURN(retval);
 737 }
 738
 739 static inline int ll_set_opt(const char *opt, char *data, int fl)
 740 {
 741         if (strncmp(opt, data, strlen(opt)) != 0)
 742                 return(0);
 743         else
 744                 return(fl);
 745 }
 746
 747 /* non-client-specific mount options are parsed in lmd_parse */
 748 static int ll_options(char *options, int *flags)
 749 {
 750         int tmp;
 751         char *s1 = options, *s2;
 752         ENTRY;
 753
 754         if (!options)
 755                 RETURN(0);
 756
 757         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 758
 759         while (*s1) {
 760                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 761                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 767                 if (tmp) {
 768                         *flags |= tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 777                 if (tmp) {
 778                         *flags &= ~tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 787                 if (tmp) {
 788                         *flags &= ~tmp;
 789                         goto next;
 790                 }
 791 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 792                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 793                 if (tmp) {
 794                         /* Ignore deprecated mount option.  The client will
 795                          * always try to mount with ACL support, whether this
 796                          * is used depends on whether server supports it. */
 797                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 798                                                   "mount option 'acl'.\n");
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 804                                                   "mount option 'noacl'.\n");
 805                         goto next;
 806                 }
 807 #else
 808 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 809 #endif
 810                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 821                 if (tmp) {
 822                         *flags &= ~tmp;
 823                         goto next;
 824                 }
 825
 826                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 877                                    s1);
 878                 RETURN(-EINVAL);
 879
 880 next:
 881                 /* Find next opt */
 882                 s2 = strchr(s1, ',');
 883                 if (s2 == NULL)
 884                         break;
 885                 s1 = s2 + 1;
 886         }
 887         RETURN(0);
 888 }
 889
 890 void ll_lli_init(struct ll_inode_info *lli)
 891 {
 892         lli->lli_inode_magic = LLI_INODE_MAGIC;
 893         lli->lli_flags = 0;
 894         lli->lli_ioepoch = 0;
 895         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 896         cfs_spin_lock_init(&lli->lli_lock);
 897         lli->lli_posix_acl = NULL;
 898         lli->lli_remote_perms = NULL;
 899         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 900         /* Do not set lli_fid, it has been initialized already. */
 901         fid_zero(&lli->lli_pfid);
 902         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 903         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 904         cfs_atomic_set(&lli->lli_open_count, 0);
 905         lli->lli_mds_capa = NULL;
 906         lli->lli_rmtperm_time = 0;
 907         lli->lli_pending_och = NULL;
 908         lli->lli_mds_read_och = NULL;
 909         lli->lli_mds_write_och = NULL;
 910         lli->lli_mds_exec_och = NULL;
 911         lli->lli_open_fd_read_count = 0;
 912         lli->lli_open_fd_write_count = 0;
 913         lli->lli_open_fd_exec_count = 0;
 914         cfs_mutex_init(&lli->lli_och_mutex);
 915         cfs_spin_lock_init(&lli->lli_agl_lock);
 916         lli->lli_has_smd = false;
 917         lli->lli_clob = NULL;
 918
 919         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 920         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 921                 cfs_mutex_init(&lli->lli_readdir_mutex);
 922                 lli->lli_opendir_key = NULL;
 923                 lli->lli_sai = NULL;
 924                 lli->lli_def_acl = NULL;
 925                 cfs_spin_lock_init(&lli->lli_sa_lock);
 926                 lli->lli_opendir_pid = 0;
 927         } else {
 928                 cfs_sema_init(&lli->lli_size_sem, 1);
 929                 lli->lli_size_sem_owner = NULL;
 930                 lli->lli_symlink_name = NULL;
 931                 cfs_init_rwsem(&lli->lli_trunc_sem);
 932                 cfs_mutex_init(&lli->lli_write_mutex);
 933                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 934                 lli->lli_glimpse_time = 0;
 935                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 936                 lli->lli_agl_index = 0;
 937                 lli->lli_async_rc = 0;
 938         }
 939         cfs_mutex_init(&lli->lli_layout_mutex);
 940 }
 941
 942 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 943 {
 944 #ifdef HAVE_BDI_REGISTER
 945         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 946
 947 #ifdef HAVE_BDI_NAME
 948         bdi->name = "lustre";
 949 #endif
 950         return bdi_register(bdi, NULL, "lustre-%d",
 951                             atomic_inc_return(&ll_bdi_num));
 952 #else
 953         return 0;
 954 #endif
 955 }
 956
 957 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 958 {
 959         struct lustre_profile *lprof = NULL;
 960         struct lustre_sb_info *lsi = s2lsi(sb);
 961         struct ll_sb_info *sbi;
 962         char  *dt = NULL, *md = NULL;
 963         char  *profilenm = get_profile_name(sb);
 964         struct config_llog_instance *cfg;
 965         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 966         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 967         int    err;
 968         ENTRY;
 969
 970         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 971
 972         OBD_ALLOC_PTR(cfg);
 973         if (cfg == NULL)
 974                 RETURN(-ENOMEM);
 975
 976         cfs_module_get();
 977
 978         /* client additional sb info */
 979         lsi->lsi_llsbi = sbi = ll_init_sbi();
 980         if (!sbi) {
 981                 cfs_module_put(THIS_MODULE);
 982                 OBD_FREE_PTR(cfg);
 983                 RETURN(-ENOMEM);
 984         }
 985
 986         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 987         if (err)
 988                 GOTO(out_free, err);
 989
 990         err = ll_bdi_init(&lsi->lsi_bdi);
 991         if (err)
 992                 GOTO(out_free, err);
 993         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 994         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 995         err = ll_bdi_register(&lsi->lsi_bdi);
 996         if (err)
 997                 GOTO(out_free, err);
 998
 999 #ifdef HAVE_SB_BDI
1000         sb->s_bdi = &lsi->lsi_bdi;
1001 #endif
1002
1003         /* Generate a string unique to this super, in case some joker tries
1004            to mount the same fs at two mount points.
1005            Use the address of the super itself.*/
1006         cfg->cfg_instance = sb;
1007         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1008         cfg->cfg_callback = class_config_llog_handler;
1009         /* set up client obds */
1010         err = lustre_process_log(sb, profilenm, cfg);
1011         if (err < 0) {
1012                 CERROR("Unable to process log: %d\n", err);
1013                 GOTO(out_free, err);
1014         }
1015
1016         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1017         lprof = class_get_profile(profilenm);
1018         if (lprof == NULL) {
1019                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1020                                    " read from the MGS.  Does that filesystem "
1021                                    "exist?\n", profilenm);
1022                 GOTO(out_free, err = -EINVAL);
1023         }
1024         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1025                lprof->lp_md, lprof->lp_dt);
1026
1027         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1028         if (!dt)
1029                 GOTO(out_free, err = -ENOMEM);
1030         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1031
1032         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1033         if (!md)
1034                 GOTO(out_free, err = -ENOMEM);
1035         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1036
1037         /* connections, registrations, sb setup */
1038         err = client_common_fill_super(sb, md, dt, mnt);
1039
1040 out_free:
1041         if (md)
1042                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1043         if (dt)
1044                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1045         if (err)
1046                 ll_put_super(sb);
1047         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1048                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1049
1050         OBD_FREE_PTR(cfg);
1051         RETURN(err);
1052 } /* ll_fill_super */
1053
1054
1055 void lu_context_keys_dump(void);
1056
1057 void ll_put_super(struct super_block *sb)
1058 {
1059         struct config_llog_instance cfg;
1060         struct obd_device *obd;
1061         struct lustre_sb_info *lsi = s2lsi(sb);
1062         struct ll_sb_info *sbi = ll_s2sbi(sb);
1063         char *profilenm = get_profile_name(sb);
1064         int force = 1, next;
1065         ENTRY;
1066
1067         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1068
1069         ll_print_capa_stat(sbi);
1070
1071         cfg.cfg_instance = sb;
1072         lustre_end_log(sb, profilenm, &cfg);
1073
1074         if (sbi->ll_md_exp) {
1075                 obd = class_exp2obd(sbi->ll_md_exp);
1076                 if (obd)
1077                         force = obd->obd_force;
1078         }
1079
1080         /* We need to set force before the lov_disconnect in
1081            lustre_common_put_super, since l_d cleans up osc's as well. */
1082         if (force) {
1083                 next = 0;
1084                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1085                                                      &next)) != NULL) {
1086                         obd->obd_force = force;
1087                 }
1088         }
1089
1090         if (sbi->ll_lcq) {
1091                 /* Only if client_common_fill_super succeeded */
1092                 client_common_put_super(sb);
1093         }
1094
1095         next = 0;
1096         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1097                 class_manual_cleanup(obd);
1098         }
1099
1100         if (sbi->ll_flags & LL_SBI_VERBOSE)
1101                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1102
1103         if (profilenm)
1104                 class_del_profile(profilenm);
1105
1106         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1107                 ll_bdi_destroy(&lsi->lsi_bdi);
1108                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1109         }
1110
1111         ll_free_sbi(sb);
1112         lsi->lsi_llsbi = NULL;
1113
1114         lustre_common_put_super(sb);
1115
1116         cl_env_cache_purge(~0);
1117
1118         cfs_module_put(THIS_MODULE);
1119
1120         EXIT;
1121 } /* client_put_super */
1122
1123 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1124 {
1125         struct inode *inode = NULL;
1126         /* NOTE: we depend on atomic igrab() -bzzz */
1127         lock_res_and_lock(lock);
1128         if (lock->l_ast_data) {
1129                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1130                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1131                         inode = igrab(lock->l_ast_data);
1132                 } else {
1133                         inode = lock->l_ast_data;
1134                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1135                                          D_WARNING, lock, "l_ast_data %p is "
1136                                          "bogus: magic %08x", lock->l_ast_data,
1137                                          lli->lli_inode_magic);
1138                         inode = NULL;
1139                 }
1140         }
1141         unlock_res_and_lock(lock);
1142         return inode;
1143 }
1144
1145 static int null_if_equal(struct ldlm_lock *lock, void *data)
1146 {
1147         if (data == lock->l_ast_data) {
1148                 lock->l_ast_data = NULL;
1149
1150                 if (lock->l_req_mode != lock->l_granted_mode)
1151                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1152         }
1153
1154         return LDLM_ITER_CONTINUE;
1155 }
1156
1157 void ll_clear_inode(struct inode *inode)
1158 {
1159         struct ll_inode_info *lli = ll_i2info(inode);
1160         struct ll_sb_info *sbi = ll_i2sbi(inode);
1161         ENTRY;
1162
1163         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1164                inode->i_generation, inode);
1165
1166         if (S_ISDIR(inode->i_mode)) {
1167                 /* these should have been cleared in ll_file_release */
1168                 LASSERT(lli->lli_opendir_key == NULL);
1169                 LASSERT(lli->lli_sai == NULL);
1170                 LASSERT(lli->lli_opendir_pid == 0);
1171         }
1172
1173         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1174         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1175                          null_if_equal, inode);
1176
1177         LASSERT(!lli->lli_open_fd_write_count);
1178         LASSERT(!lli->lli_open_fd_read_count);
1179         LASSERT(!lli->lli_open_fd_exec_count);
1180
1181         if (lli->lli_mds_write_och)
1182                 ll_md_real_close(inode, FMODE_WRITE);
1183         if (lli->lli_mds_exec_och)
1184                 ll_md_real_close(inode, FMODE_EXEC);
1185         if (lli->lli_mds_read_och)
1186                 ll_md_real_close(inode, FMODE_READ);
1187
1188         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1189                 OBD_FREE(lli->lli_symlink_name,
1190                          strlen(lli->lli_symlink_name) + 1);
1191                 lli->lli_symlink_name = NULL;
1192         }
1193
1194         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1195                 LASSERT(lli->lli_posix_acl == NULL);
1196                 if (lli->lli_remote_perms) {
1197                         free_rmtperm_hash(lli->lli_remote_perms);
1198                         lli->lli_remote_perms = NULL;
1199                 }
1200         }
1201 #ifdef CONFIG_FS_POSIX_ACL
1202         else if (lli->lli_posix_acl) {
1203                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1204                 LASSERT(lli->lli_remote_perms == NULL);
1205                 posix_acl_release(lli->lli_posix_acl);
1206                 lli->lli_posix_acl = NULL;
1207         }
1208 #endif
1209         lli->lli_inode_magic = LLI_INODE_DEAD;
1210
1211         ll_clear_inode_capas(inode);
1212         if (!S_ISDIR(inode->i_mode))
1213                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1214
1215         /*
1216          * XXX This has to be done before lsm is freed below, because
1217          * cl_object still uses inode lsm.
1218          */
1219         cl_inode_fini(inode);
1220         lli->lli_has_smd = false;
1221
1222         EXIT;
1223 }
1224
1225 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1226                   struct md_open_data **mod)
1227 {
1228         struct lustre_md md;
1229         struct inode *inode = dentry->d_inode;
1230         struct ll_sb_info *sbi = ll_i2sbi(inode);
1231         struct ptlrpc_request *request = NULL;
1232         int rc, ia_valid;
1233         ENTRY;
1234
1235         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1236                                      LUSTRE_OPC_ANY, NULL);
1237         if (IS_ERR(op_data))
1238                 RETURN(PTR_ERR(op_data));
1239
1240         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1241                         &request, mod);
1242         if (rc) {
1243                 ptlrpc_req_finished(request);
1244                 if (rc == -ENOENT) {
1245                         inode->i_nlink = 0;
1246                         /* Unlinked special device node? Or just a race?
1247                          * Pretend we done everything. */
1248                         if (!S_ISREG(inode->i_mode) &&
1249                             !S_ISDIR(inode->i_mode)) {
1250                                 ia_valid = op_data->op_attr.ia_valid;
1251                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252                                 rc = simple_setattr(dentry, &op_data->op_attr);
1253                                 op_data->op_attr.ia_valid = ia_valid;
1254                         }
1255                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1256                         CERROR("md_setattr fails: rc = %d\n", rc);
1257                 }
1258                 RETURN(rc);
1259         }
1260
1261         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1262                               sbi->ll_md_exp, &md);
1263         if (rc) {
1264                 ptlrpc_req_finished(request);
1265                 RETURN(rc);
1266         }
1267
1268         /* We want to adjust timestamps.
1269          * If there is at least some data in file, we cleared ATTR_SIZE
1270          * to avoid update size, otherwise it is important to do.(SOM case)
1271          * (bug 6196) */
1272         ia_valid = op_data->op_attr.ia_valid;
1273         /* Since we set ATTR_*_SET flags above, and already done permission
1274          * check, So don't let inode_change_ok() check it again.  */
1275         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1276         rc = simple_setattr(dentry, &op_data->op_attr);
1277         op_data->op_attr.ia_valid = ia_valid;
1278
1279         /* Extract epoch data if obtained. */
1280         op_data->op_handle = md.body->handle;
1281         op_data->op_ioepoch = md.body->ioepoch;
1282
1283         ll_update_inode(inode, &md);
1284         ptlrpc_req_finished(request);
1285
1286         RETURN(rc);
1287 }
1288
1289 /* Close IO epoch and send Size-on-MDS attribute update. */
1290 static int ll_setattr_done_writing(struct inode *inode,
1291                                    struct md_op_data *op_data,
1292                                    struct md_open_data *mod)
1293 {
1294         struct ll_inode_info *lli = ll_i2info(inode);
1295         int rc = 0;
1296         ENTRY;
1297
1298         LASSERT(op_data != NULL);
1299         if (!S_ISREG(inode->i_mode))
1300                 RETURN(0);
1301
1302         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1303                op_data->op_ioepoch, PFID(&lli->lli_fid));
1304
1305         op_data->op_flags = MF_EPOCH_CLOSE;
1306         ll_done_writing_attr(inode, op_data);
1307         ll_pack_inode2opdata(inode, op_data, NULL);
1308
1309         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1310         if (rc == -EAGAIN) {
1311                 /* MDS has instructed us to obtain Size-on-MDS attribute
1312                  * from OSTs and send setattr to back to MDS. */
1313                 rc = ll_som_update(inode, op_data);
1314         } else if (rc) {
1315                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1316                        inode->i_ino, rc);
1317         }
1318         RETURN(rc);
1319 }
1320
1321 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1322 {
1323         struct obd_capa *capa;
1324         int rc;
1325
1326         if (attr->ia_valid & ATTR_SIZE)
1327                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1328         else
1329                 capa = ll_mdscapa_get(inode);
1330
1331         rc = cl_setattr_ost(inode, attr, capa);
1332
1333         if (attr->ia_valid & ATTR_SIZE)
1334                 ll_truncate_free_capa(capa);
1335         else
1336                 capa_put(capa);
1337
1338         return rc;
1339 }
1340
1341 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1342 /**
1343  * inode_newsize_ok - may this inode be truncated to a given size
1344  * @inode:      the inode to be truncated
1345  * @offset:     the new size to assign to the inode
1346  * @Returns:    0 on success, -ve errno on failure
1347  *
1348  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1349  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1350  * when necessary. Caller must not proceed with inode size change if failure is
1351  * returned. @inode must be a file (not directory), with appropriate
1352  * permissions to allow truncate (inode_newsize_ok does NOT check these
1353  * conditions).
1354  *
1355  * inode_newsize_ok must be called with i_mutex held.
1356  */
1357 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1358 {
1359         if (inode->i_size < offset) {
1360                 unsigned long limit;
1361
1362                 limit = rlimit(RLIMIT_FSIZE);
1363                 if (limit != RLIM_INFINITY && offset > limit)
1364                         goto out_sig;
1365                 if (offset > inode->i_sb->s_maxbytes)
1366                         goto out_big;
1367         } else {
1368                 /*
1369                  * truncation of in-use swapfiles is disallowed - it would
1370                  * cause subsequent swapout to scribble on the now-freed
1371                  * blocks.
1372                  */
1373                 if (IS_SWAPFILE(inode))
1374                         return -ETXTBSY;
1375         }
1376
1377         return 0;
1378 out_sig:
1379         send_sig(SIGXFSZ, current, 0);
1380 out_big:
1381         return -EFBIG;
1382 }
1383 #endif
1384
1385 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1386  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1387  * keep these values until such a time that objects are allocated for it.
1388  * We do the MDS operations first, as it is checking permissions for us.
1389  * We don't to the MDS RPC if there is nothing that we want to store there,
1390  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1391  * going to do an RPC anyways.
1392  *
1393  * If we are doing a truncate, we will send the mtime and ctime updates
1394  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1395  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1396  * at the same time.
1397  */
1398 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1399 {
1400         struct lov_stripe_md *lsm;
1401         struct inode *inode = dentry->d_inode;
1402         struct ll_inode_info *lli = ll_i2info(inode);
1403         struct md_op_data *op_data = NULL;
1404         struct md_open_data *mod = NULL;
1405         int ia_valid = attr->ia_valid;
1406         int rc = 0, rc1 = 0;
1407         ENTRY;
1408
1409         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1410                attr->ia_valid);
1411
1412         if (ia_valid & ATTR_SIZE) {
1413                 /* Check new size against VFS/VM file size limit and rlimit */
1414                 rc = inode_newsize_ok(inode, attr->ia_size);
1415                 if (rc)
1416                         RETURN(rc);
1417
1418                 /* The maximum Lustre file size is variable, based on the
1419                  * OST maximum object size and number of stripes.  This
1420                  * needs another check in addition to the VFS check above. */
1421                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1422                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1423                                PFID(&lli->lli_fid), attr->ia_size,
1424                                ll_file_maxbytes(inode));
1425                         RETURN(-EFBIG);
1426                 }
1427
1428                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1429         }
1430
1431         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1432         if (ia_valid & TIMES_SET_FLAGS) {
1433                 if (cfs_curproc_fsuid() != inode->i_uid &&
1434                     !cfs_capable(CFS_CAP_FOWNER))
1435                         RETURN(-EPERM);
1436         }
1437
1438         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1439         if (attr->ia_valid & ATTR_CTIME) {
1440                 attr->ia_ctime = CFS_CURRENT_TIME;
1441                 attr->ia_valid |= ATTR_CTIME_SET;
1442         }
1443         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1444                 attr->ia_atime = CFS_CURRENT_TIME;
1445                 attr->ia_valid |= ATTR_ATIME_SET;
1446         }
1447         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1448                 attr->ia_mtime = CFS_CURRENT_TIME;
1449                 attr->ia_valid |= ATTR_MTIME_SET;
1450         }
1451
1452         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1453                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1454                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1455                        cfs_time_current_sec());
1456
1457         /* We always do an MDS RPC, even if we're only changing the size;
1458          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1459
1460         OBD_ALLOC_PTR(op_data);
1461         if (op_data == NULL)
1462                 RETURN(-ENOMEM);
1463
1464         if (!S_ISDIR(inode->i_mode)) {
1465                 if (ia_valid & ATTR_SIZE)
1466                         UP_WRITE_I_ALLOC_SEM(inode);
1467                 mutex_unlock(&inode->i_mutex);
1468                 cfs_down_write(&lli->lli_trunc_sem);
1469                 mutex_lock(&inode->i_mutex);
1470                 if (ia_valid & ATTR_SIZE)
1471                         DOWN_WRITE_I_ALLOC_SEM(inode);
1472         }
1473
1474         /* We need a steady stripe configuration for setattr to avoid
1475          * confusion. */
1476         lsm = ccc_inode_lsm_get(inode);
1477
1478         /* NB: ATTR_SIZE will only be set after this point if the size
1479          * resides on the MDS, ie, this file has no objects. */
1480         if (lsm != NULL)
1481                 attr->ia_valid &= ~ATTR_SIZE;
1482
1483         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1484
1485         /* Open epoch for truncate. */
1486         if (exp_connect_som(ll_i2mdexp(inode)) &&
1487             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1488                 op_data->op_flags = MF_EPOCH_OPEN;
1489
1490         rc = ll_md_setattr(dentry, op_data, &mod);
1491         if (rc)
1492                 GOTO(out, rc);
1493
1494         ll_ioepoch_open(lli, op_data->op_ioepoch);
1495         if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1496                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1497                 GOTO(out, rc = 0);
1498         }
1499
1500         if (ia_valid & ATTR_SIZE)
1501                 attr->ia_valid |= ATTR_SIZE;
1502         if (ia_valid & (ATTR_SIZE |
1503                         ATTR_ATIME | ATTR_ATIME_SET |
1504                         ATTR_MTIME | ATTR_MTIME_SET))
1505                 /* For truncate and utimes sending attributes to OSTs, setting
1506                  * mtime/atime to the past will be performed under PW [0:EOF]
1507                  * extent lock (new_size:EOF for truncate).  It may seem
1508                  * excessive to send mtime/atime updates to OSTs when not
1509                  * setting times to past, but it is necessary due to possible
1510                  * time de-synchronization between MDT inode and OST objects */
1511                 rc = ll_setattr_ost(inode, attr);
1512         EXIT;
1513 out:
1514         ccc_inode_lsm_put(inode, lsm);
1515         if (op_data) {
1516                 if (op_data->op_ioepoch) {
1517                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1518                         if (!rc)
1519                                 rc = rc1;
1520                 }
1521                 ll_finish_md_op_data(op_data);
1522         }
1523         if (!S_ISDIR(inode->i_mode))
1524                 cfs_up_write(&lli->lli_trunc_sem);
1525
1526         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1527                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1528
1529         return rc;
1530 }
1531
1532 int ll_setattr(struct dentry *de, struct iattr *attr)
1533 {
1534         int mode = de->d_inode->i_mode;
1535
1536         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1537                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1538                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1539
1540         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1541                                (ATTR_SIZE|ATTR_MODE)) &&
1542             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1543              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1544               !(attr->ia_mode & S_ISGID))))
1545                 attr->ia_valid |= ATTR_FORCE;
1546
1547         if ((mode & S_ISUID) &&
1548             !(attr->ia_mode & S_ISUID) &&
1549             !(attr->ia_valid & ATTR_KILL_SUID))
1550                 attr->ia_valid |= ATTR_KILL_SUID;
1551
1552         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1553             !(attr->ia_mode & S_ISGID) &&
1554             !(attr->ia_valid & ATTR_KILL_SGID))
1555                 attr->ia_valid |= ATTR_KILL_SGID;
1556
1557         return ll_setattr_raw(de, attr);
1558 }
1559
1560 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1561                        __u64 max_age, __u32 flags)
1562 {
1563         struct ll_sb_info *sbi = ll_s2sbi(sb);
1564         struct obd_statfs obd_osfs;
1565         int rc;
1566         ENTRY;
1567
1568         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1569         if (rc) {
1570                 CERROR("md_statfs fails: rc = %d\n", rc);
1571                 RETURN(rc);
1572         }
1573
1574         osfs->os_type = sb->s_magic;
1575
1576         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1577                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1578
1579         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1580                 flags |= OBD_STATFS_NODELAY;
1581
1582         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1583         if (rc) {
1584                 CERROR("obd_statfs fails: rc = %d\n", rc);
1585                 RETURN(rc);
1586         }
1587
1588         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1589                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1590                obd_osfs.os_files);
1591
1592         osfs->os_bsize = obd_osfs.os_bsize;
1593         osfs->os_blocks = obd_osfs.os_blocks;
1594         osfs->os_bfree = obd_osfs.os_bfree;
1595         osfs->os_bavail = obd_osfs.os_bavail;
1596
1597         /* If we don't have as many objects free on the OST as inodes
1598          * on the MDS, we reduce the total number of inodes to
1599          * compensate, so that the "inodes in use" number is correct.
1600          */
1601         if (obd_osfs.os_ffree < osfs->os_ffree) {
1602                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1603                         obd_osfs.os_ffree;
1604                 osfs->os_ffree = obd_osfs.os_ffree;
1605         }
1606
1607         RETURN(rc);
1608 }
1609 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1610 {
1611         struct super_block *sb = de->d_sb;
1612         struct obd_statfs osfs;
1613         int rc;
1614
1615         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1616         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1617
1618         /* Some amount of caching on the client is allowed */
1619         rc = ll_statfs_internal(sb, &osfs,
1620                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1621                                 0);
1622         if (rc)
1623                 return rc;
1624
1625         statfs_unpack(sfs, &osfs);
1626
1627         /* We need to downshift for all 32-bit kernels, because we can't
1628          * tell if the kernel is being called via sys_statfs64() or not.
1629          * Stop before overflowing f_bsize - in which case it is better
1630          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1631         if (sizeof(long) < 8) {
1632                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1633                         sfs->f_bsize <<= 1;
1634
1635                         osfs.os_blocks >>= 1;
1636                         osfs.os_bfree >>= 1;
1637                         osfs.os_bavail >>= 1;
1638                 }
1639         }
1640
1641         sfs->f_blocks = osfs.os_blocks;
1642         sfs->f_bfree = osfs.os_bfree;
1643         sfs->f_bavail = osfs.os_bavail;
1644
1645         return 0;
1646 }
1647
1648 void ll_inode_size_lock(struct inode *inode)
1649 {
1650         struct ll_inode_info *lli;
1651
1652         LASSERT(!S_ISDIR(inode->i_mode));
1653
1654         lli = ll_i2info(inode);
1655         LASSERT(lli->lli_size_sem_owner != current);
1656         cfs_down(&lli->lli_size_sem);
1657         LASSERT(lli->lli_size_sem_owner == NULL);
1658         lli->lli_size_sem_owner = current;
1659 }
1660
1661 void ll_inode_size_unlock(struct inode *inode)
1662 {
1663         struct ll_inode_info *lli;
1664
1665         lli = ll_i2info(inode);
1666         LASSERT(lli->lli_size_sem_owner == current);
1667         lli->lli_size_sem_owner = NULL;
1668         cfs_up(&lli->lli_size_sem);
1669 }
1670
1671 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1672 {
1673         struct ll_inode_info *lli = ll_i2info(inode);
1674         struct mdt_body *body = md->body;
1675         struct lov_stripe_md *lsm = md->lsm;
1676         struct ll_sb_info *sbi = ll_i2sbi(inode);
1677
1678         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1679         if (lsm != NULL) {
1680                 LASSERT(S_ISREG(inode->i_mode));
1681                 cfs_mutex_lock(&lli->lli_och_mutex);
1682                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1683                                 lsm, inode->i_ino, inode->i_generation, inode);
1684                 /* cl_file_inode_init must go before lli_has_smd or a race
1685                  * is possible where client thinks the file has stripes,
1686                  * but lov raid0 is not setup yet and parallel e.g.
1687                  * glimpse would try to use uninitialized lov */
1688                 if (cl_file_inode_init(inode, md) == 0)
1689                         lli->lli_has_smd = true;
1690                 cfs_mutex_unlock(&lli->lli_och_mutex);
1691
1692                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1693                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1694                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1695                 if (md->lsm != NULL)
1696                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1697         }
1698
1699         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1700                 if (body->valid & OBD_MD_FLRMTPERM)
1701                         ll_update_remote_perm(inode, md->remote_perm);
1702         }
1703 #ifdef CONFIG_FS_POSIX_ACL
1704         else if (body->valid & OBD_MD_FLACL) {
1705                 cfs_spin_lock(&lli->lli_lock);
1706                 if (lli->lli_posix_acl)
1707                         posix_acl_release(lli->lli_posix_acl);
1708                 lli->lli_posix_acl = md->posix_acl;
1709                 cfs_spin_unlock(&lli->lli_lock);
1710         }
1711 #endif
1712         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1713         inode->i_generation = cl_fid_build_gen(&body->fid1);
1714
1715         if (body->valid & OBD_MD_FLATIME) {
1716                 if (body->atime > LTIME_S(inode->i_atime))
1717                         LTIME_S(inode->i_atime) = body->atime;
1718                 lli->lli_lvb.lvb_atime = body->atime;
1719         }
1720         if (body->valid & OBD_MD_FLMTIME) {
1721                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1722                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1723                                "to "LPU64"\n", inode->i_ino,
1724                                LTIME_S(inode->i_mtime), body->mtime);
1725                         LTIME_S(inode->i_mtime) = body->mtime;
1726                 }
1727                 lli->lli_lvb.lvb_mtime = body->mtime;
1728         }
1729         if (body->valid & OBD_MD_FLCTIME) {
1730                 if (body->ctime > LTIME_S(inode->i_ctime))
1731                         LTIME_S(inode->i_ctime) = body->ctime;
1732                 lli->lli_lvb.lvb_ctime = body->ctime;
1733         }
1734         if (body->valid & OBD_MD_FLMODE)
1735                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1736         if (body->valid & OBD_MD_FLTYPE)
1737                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1738         LASSERT(inode->i_mode != 0);
1739         if (S_ISREG(inode->i_mode)) {
1740                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1741         } else {
1742                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1743         }
1744         if (body->valid & OBD_MD_FLUID)
1745                 inode->i_uid = body->uid;
1746         if (body->valid & OBD_MD_FLGID)
1747                 inode->i_gid = body->gid;
1748         if (body->valid & OBD_MD_FLFLAGS)
1749                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1750         if (body->valid & OBD_MD_FLNLINK)
1751                 inode->i_nlink = body->nlink;
1752         if (body->valid & OBD_MD_FLRDEV)
1753                 inode->i_rdev = old_decode_dev(body->rdev);
1754
1755         if (body->valid & OBD_MD_FLID) {
1756                 /* FID shouldn't be changed! */
1757                 if (fid_is_sane(&lli->lli_fid)) {
1758                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1759                                  "Trying to change FID "DFID
1760                                  " to the "DFID", inode %lu/%u(%p)\n",
1761                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1762                                  inode->i_ino, inode->i_generation, inode);
1763                 } else
1764                         lli->lli_fid = body->fid1;
1765         }
1766
1767         LASSERT(fid_seq(&lli->lli_fid) != 0);
1768
1769         if (body->valid & OBD_MD_FLSIZE) {
1770                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1771                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1772                         struct lustre_handle lockh;
1773                         ldlm_mode_t mode;
1774
1775                         /* As it is possible a blocking ast has been processed
1776                          * by this time, we need to check there is an UPDATE
1777                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1778                          * it. */
1779                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1780                                                &lockh);
1781                         if (mode) {
1782                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1783                                                       LLIF_EPOCH_PENDING |
1784                                                       LLIF_SOM_DIRTY)) {
1785                                         CERROR("ino %lu flags %u still has "
1786                                                "size authority! do not trust "
1787                                                "the size got from MDS\n",
1788                                                inode->i_ino, lli->lli_flags);
1789                                 } else {
1790                                         /* Use old size assignment to avoid
1791                                          * deadlock bz14138 & bz14326 */
1792                                         i_size_write(inode, body->size);
1793                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1794                                 }
1795                                 ldlm_lock_decref(&lockh, mode);
1796                         }
1797                 } else {
1798                         /* Use old size assignment to avoid
1799                          * deadlock bz14138 & bz14326 */
1800                         i_size_write(inode, body->size);
1801
1802                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1803                                inode->i_ino, (unsigned long long)body->size);
1804                 }
1805
1806                 if (body->valid & OBD_MD_FLBLOCKS)
1807                         inode->i_blocks = body->blocks;
1808         }
1809
1810         if (body->valid & OBD_MD_FLMDSCAPA) {
1811                 LASSERT(md->mds_capa);
1812                 ll_add_capa(inode, md->mds_capa);
1813         }
1814         if (body->valid & OBD_MD_FLOSSCAPA) {
1815                 LASSERT(md->oss_capa);
1816                 ll_add_capa(inode, md->oss_capa);
1817         }
1818 }
1819
1820 void ll_read_inode2(struct inode *inode, void *opaque)
1821 {
1822         struct lustre_md *md = opaque;
1823         struct ll_inode_info *lli = ll_i2info(inode);
1824         ENTRY;
1825
1826         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1827                PFID(&lli->lli_fid), inode);
1828
1829         LASSERT(!lli->lli_has_smd);
1830
1831         /* Core attributes from the MDS first.  This is a new inode, and
1832          * the VFS doesn't zero times in the core inode so we have to do
1833          * it ourselves.  They will be overwritten by either MDS or OST
1834          * attributes - we just need to make sure they aren't newer. */
1835         LTIME_S(inode->i_mtime) = 0;
1836         LTIME_S(inode->i_atime) = 0;
1837         LTIME_S(inode->i_ctime) = 0;
1838         inode->i_rdev = 0;
1839         ll_update_inode(inode, md);
1840
1841         /* OIDEBUG(inode); */
1842
1843         /* initializing backing dev info. */
1844         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1845
1846
1847         if (S_ISREG(inode->i_mode)) {
1848                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1849                 inode->i_op = &ll_file_inode_operations;
1850                 inode->i_fop = sbi->ll_fop;
1851                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1852                 EXIT;
1853         } else if (S_ISDIR(inode->i_mode)) {
1854                 inode->i_op = &ll_dir_inode_operations;
1855                 inode->i_fop = &ll_dir_operations;
1856                 EXIT;
1857         } else if (S_ISLNK(inode->i_mode)) {
1858                 inode->i_op = &ll_fast_symlink_inode_operations;
1859                 EXIT;
1860         } else {
1861                 inode->i_op = &ll_special_inode_operations;
1862
1863                 init_special_inode(inode, inode->i_mode,
1864                                    inode->i_rdev);
1865
1866                 EXIT;
1867         }
1868 }
1869
1870 void ll_delete_inode(struct inode *inode)
1871 {
1872         struct cl_inode_info *lli = cl_i2info(inode);
1873         struct ll_sb_info    *sbi = ll_i2sbi(inode);
1874         int rc;
1875         ENTRY;
1876
1877         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1878         if (rc)
1879                 CERROR("fid_delete() failed, rc %d\n", rc);
1880
1881         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1882                 /* discard all dirty pages before truncating them, required by
1883                  * osc_extent implementation at LU-1030. */
1884                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1885
1886         truncate_inode_pages(&inode->i_data, 0);
1887
1888         /* Workaround for LU-118 */
1889         if (inode->i_data.nrpages) {
1890                 TREE_READ_LOCK_IRQ(&inode->i_data);
1891                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1892                 LASSERTF(inode->i_data.nrpages == 0,
1893                          "inode=%lu/%u(%p) nrpages=%lu, see "
1894                          "http://jira.whamcloud.com/browse/LU-118\n",
1895                          inode->i_ino, inode->i_generation, inode,
1896                          inode->i_data.nrpages);
1897         }
1898         /* Workaround end */
1899
1900 #ifdef HAVE_SBOPS_EVICT_INODE
1901         ll_clear_inode(inode);
1902         end_writeback(inode);
1903 #else
1904         clear_inode(inode);
1905 #endif
1906
1907         EXIT;
1908 }
1909
1910 int ll_iocontrol(struct inode *inode, struct file *file,
1911                  unsigned int cmd, unsigned long arg)
1912 {
1913         struct ll_sb_info *sbi = ll_i2sbi(inode);
1914         struct ptlrpc_request *req = NULL;
1915         int rc, flags = 0;
1916         ENTRY;
1917
1918         switch(cmd) {
1919         case FSFILT_IOC_GETFLAGS: {
1920                 struct mdt_body *body;
1921                 struct md_op_data *op_data;
1922
1923                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1924                                              0, 0, LUSTRE_OPC_ANY,
1925                                              NULL);
1926                 if (IS_ERR(op_data))
1927                         RETURN(PTR_ERR(op_data));
1928
1929                 op_data->op_valid = OBD_MD_FLFLAGS;
1930                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1931                 ll_finish_md_op_data(op_data);
1932                 if (rc) {
1933                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1934                         RETURN(-abs(rc));
1935                 }
1936
1937                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1938
1939                 flags = body->flags;
1940
1941                 ptlrpc_req_finished(req);
1942
1943                 RETURN(put_user(flags, (int *)arg));
1944         }
1945         case FSFILT_IOC_SETFLAGS: {
1946                 struct lov_stripe_md *lsm;
1947                 struct obd_info oinfo = { { { 0 } } };
1948                 struct md_op_data *op_data;
1949
1950                 if (get_user(flags, (int *)arg))
1951                         RETURN(-EFAULT);
1952
1953                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1954                                              LUSTRE_OPC_ANY, NULL);
1955                 if (IS_ERR(op_data))
1956                         RETURN(PTR_ERR(op_data));
1957
1958                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1959                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1960                 rc = md_setattr(sbi->ll_md_exp, op_data,
1961                                 NULL, 0, NULL, 0, &req, NULL);
1962                 ll_finish_md_op_data(op_data);
1963                 ptlrpc_req_finished(req);
1964                 if (rc)
1965                         RETURN(rc);
1966
1967                 inode->i_flags = ll_ext_to_inode_flags(flags);
1968
1969                 lsm = ccc_inode_lsm_get(inode);
1970                 if (lsm == NULL)
1971                         RETURN(0);
1972
1973                 OBDO_ALLOC(oinfo.oi_oa);
1974                 if (!oinfo.oi_oa) {
1975                         ccc_inode_lsm_put(inode, lsm);
1976                         RETURN(-ENOMEM);
1977                 }
1978                 oinfo.oi_md = lsm;
1979                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1980                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1981                 oinfo.oi_oa->o_flags = flags;
1982                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1983                                        OBD_MD_FLGROUP;
1984                 oinfo.oi_capa = ll_mdscapa_get(inode);
1985                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1986                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1987                 capa_put(oinfo.oi_capa);
1988                 OBDO_FREE(oinfo.oi_oa);
1989                 ccc_inode_lsm_put(inode, lsm);
1990
1991                 if (rc && rc != -EPERM && rc != -EACCES)
1992                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1993
1994                 RETURN(rc);
1995         }
1996         default:
1997                 RETURN(-ENOSYS);
1998         }
1999
2000         RETURN(0);
2001 }
2002
2003 int ll_flush_ctx(struct inode *inode)
2004 {
2005         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2006
2007         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2008
2009         obd_set_info_async(NULL, sbi->ll_md_exp,
2010                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2011                            0, NULL, NULL);
2012         obd_set_info_async(NULL, sbi->ll_dt_exp,
2013                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2014                            0, NULL, NULL);
2015         return 0;
2016 }
2017
2018 /* umount -f client means force down, don't save state */
2019 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2020 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2021 {
2022         struct super_block *sb = vfsmnt->mnt_sb;
2023 #else
2024 void ll_umount_begin(struct super_block *sb)
2025 {
2026 #endif
2027         struct ll_sb_info *sbi = ll_s2sbi(sb);
2028         struct obd_device *obd;
2029         struct obd_ioctl_data *ioc_data;
2030         ENTRY;
2031
2032 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2033         if (!(flags & MNT_FORCE)) {
2034                 EXIT;
2035                 return;
2036         }
2037 #endif
2038
2039         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2040                sb->s_count, atomic_read(&sb->s_active));
2041
2042         obd = class_exp2obd(sbi->ll_md_exp);
2043         if (obd == NULL) {
2044                 CERROR("Invalid MDC connection handle "LPX64"\n",
2045                        sbi->ll_md_exp->exp_handle.h_cookie);
2046                 EXIT;
2047                 return;
2048         }
2049         obd->obd_force = 1;
2050
2051         obd = class_exp2obd(sbi->ll_dt_exp);
2052         if (obd == NULL) {
2053                 CERROR("Invalid LOV connection handle "LPX64"\n",
2054                        sbi->ll_dt_exp->exp_handle.h_cookie);
2055                 EXIT;
2056                 return;
2057         }
2058         obd->obd_force = 1;
2059
2060         OBD_ALLOC_PTR(ioc_data);
2061         if (ioc_data) {
2062                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2063                               sizeof ioc_data, ioc_data, NULL);
2064
2065                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2066                               sizeof ioc_data, ioc_data, NULL);
2067
2068                 OBD_FREE_PTR(ioc_data);
2069         }
2070
2071
2072         /* Really, we'd like to wait until there are no requests outstanding,
2073          * and then continue.  For now, we just invalidate the requests,
2074          * schedule() and sleep one second if needed, and hope.
2075          */
2076         cfs_schedule();
2077 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2078         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2079                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2080                                                    cfs_time_seconds(1));
2081                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2082                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2083                                       "may try to umount it a bit later\n",
2084                                       atomic_read(&vfsmnt->mnt_count));
2085         }
2086 #endif
2087
2088         EXIT;
2089 }
2090
2091 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2092 {
2093         struct ll_sb_info *sbi = ll_s2sbi(sb);
2094         char *profilenm = get_profile_name(sb);
2095         int err;
2096         __u32 read_only;
2097
2098         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2099                 read_only = *flags & MS_RDONLY;
2100                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2101                                          sizeof(KEY_READ_ONLY),
2102                                          KEY_READ_ONLY, sizeof(read_only),
2103                                          &read_only, NULL);
2104                 if (err) {
2105                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2106                                       profilenm, read_only ?
2107                                       "read-only" : "read-write", err);
2108                         return err;
2109                 }
2110
2111                 if (read_only)
2112                         sb->s_flags |= MS_RDONLY;
2113                 else
2114                         sb->s_flags &= ~MS_RDONLY;
2115
2116                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2117                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2118                                       read_only ?  "read-only" : "read-write");
2119         }
2120         return 0;
2121 }
2122
2123 int ll_prep_inode(struct inode **inode,
2124                   struct ptlrpc_request *req,
2125                   struct super_block *sb)
2126 {
2127         struct ll_sb_info *sbi = NULL;
2128         struct lustre_md md;
2129         __u64 ibits;
2130         int rc;
2131         ENTRY;
2132
2133         LASSERT(*inode || sb);
2134         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2135         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2136                               sbi->ll_md_exp, &md);
2137         if (rc)
2138                 RETURN(rc);
2139
2140         if (*inode) {
2141                 ll_update_inode(*inode, &md);
2142         } else {
2143                 LASSERT(sb != NULL);
2144
2145                 /*
2146                  * At this point server returns to client's same fid as client
2147                  * generated for creating. So using ->fid1 is okay here.
2148                  */
2149                 LASSERT(fid_is_sane(&md.body->fid1));
2150
2151                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2152                 if (*inode == NULL || IS_ERR(*inode)) {
2153                         if (md.lsm)
2154                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2155 #ifdef CONFIG_FS_POSIX_ACL
2156                         if (md.posix_acl) {
2157                                 posix_acl_release(md.posix_acl);
2158                                 md.posix_acl = NULL;
2159                         }
2160 #endif
2161                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2162                         *inode = NULL;
2163                         CERROR("new_inode -fatal: rc %d\n", rc);
2164                         GOTO(out, rc);
2165                 }
2166         }
2167
2168         /* sanity check for LAYOUT lock. */
2169         ibits = MDS_INODELOCK_LAYOUT;
2170         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2171             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2172                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2173                         ll_get_fsname(sb, NULL, 0),
2174                         PFID(ll_inode2fid(*inode)), *inode);
2175         }
2176
2177 out:
2178         md_free_lustre_md(sbi->ll_md_exp, &md);
2179         RETURN(rc);
2180 }
2181
2182 int ll_obd_statfs(struct inode *inode, void *arg)
2183 {
2184         struct ll_sb_info *sbi = NULL;
2185         struct obd_export *exp;
2186         char *buf = NULL;
2187         struct obd_ioctl_data *data = NULL;
2188         __u32 type;
2189         __u32 flags;
2190         int len = 0, rc;
2191
2192         if (!inode || !(sbi = ll_i2sbi(inode)))
2193                 GOTO(out_statfs, rc = -EINVAL);
2194
2195         rc = obd_ioctl_getdata(&buf, &len, arg);
2196         if (rc)
2197                 GOTO(out_statfs, rc);
2198
2199         data = (void*)buf;
2200         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2201             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2202                 GOTO(out_statfs, rc = -EINVAL);
2203
2204         if (data->ioc_inllen1 != sizeof(__u32) ||
2205             data->ioc_inllen2 != sizeof(__u32) ||
2206             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2207             data->ioc_plen2 != sizeof(struct obd_uuid))
2208                 GOTO(out_statfs, rc = -EINVAL);
2209
2210         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2211         if (type & LL_STATFS_LMV)
2212                 exp = sbi->ll_md_exp;
2213         else if (type & LL_STATFS_LOV)
2214                 exp = sbi->ll_dt_exp;
2215         else
2216                 GOTO(out_statfs, rc = -ENODEV);
2217
2218         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2219         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2220         if (rc)
2221                 GOTO(out_statfs, rc);
2222 out_statfs:
2223         if (buf)
2224                 obd_ioctl_freedata(buf, len);
2225         return rc;
2226 }
2227
2228 int ll_process_config(struct lustre_cfg *lcfg)
2229 {
2230         char *ptr;
2231         void *sb;
2232         struct lprocfs_static_vars lvars;
2233         unsigned long x;
2234         int rc = 0;
2235
2236         lprocfs_llite_init_vars(&lvars);
2237
2238         /* The instance name contains the sb: lustre-client-aacfe000 */
2239         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2240         if (!ptr || !*(++ptr))
2241                 return -EINVAL;
2242         if (sscanf(ptr, "%lx", &x) != 1)
2243                 return -EINVAL;
2244         sb = (void *)x;
2245         /* This better be a real Lustre superblock! */
2246         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2247
2248         /* Note we have not called client_common_fill_super yet, so
2249            proc fns must be able to handle that! */
2250         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2251                                       lcfg, sb);
2252         if (rc > 0)
2253                 rc = 0;
2254         return(rc);
2255 }
2256
2257 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2258 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2259                                        struct inode *i1, struct inode *i2,
2260                                        const char *name, int namelen,
2261                                        int mode, __u32 opc, void *data)
2262 {
2263         LASSERT(i1 != NULL);
2264
2265         if (namelen > ll_i2sbi(i1)->ll_namelen)
2266                 return ERR_PTR(-ENAMETOOLONG);
2267
2268         if (op_data == NULL)
2269                 OBD_ALLOC_PTR(op_data);
2270
2271         if (op_data == NULL)
2272                 return ERR_PTR(-ENOMEM);
2273
2274         ll_i2gids(op_data->op_suppgids, i1, i2);
2275         op_data->op_fid1 = *ll_inode2fid(i1);
2276         op_data->op_capa1 = ll_mdscapa_get(i1);
2277
2278         if (i2) {
2279                 op_data->op_fid2 = *ll_inode2fid(i2);
2280                 op_data->op_capa2 = ll_mdscapa_get(i2);
2281         } else {
2282                 fid_zero(&op_data->op_fid2);
2283                 op_data->op_capa2 = NULL;
2284         }
2285
2286         op_data->op_name = name;
2287         op_data->op_namelen = namelen;
2288         op_data->op_mode = mode;
2289         op_data->op_mod_time = cfs_time_current_sec();
2290         op_data->op_fsuid = cfs_curproc_fsuid();
2291         op_data->op_fsgid = cfs_curproc_fsgid();
2292         op_data->op_cap = cfs_curproc_cap_pack();
2293         op_data->op_bias = MDS_CHECK_SPLIT;
2294         op_data->op_opc = opc;
2295         op_data->op_mds = 0;
2296         op_data->op_data = data;
2297
2298         /* If the file is being opened after mknod() (normally due to NFS)
2299          * try to use the default stripe data from parent directory for
2300          * allocating OST objects.  Try to pass the parent FID to MDS. */
2301         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2302             !ll_i2info(i2)->lli_has_smd) {
2303                 struct ll_inode_info *lli = ll_i2info(i2);
2304
2305                 cfs_spin_lock(&lli->lli_lock);
2306                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2307                         op_data->op_fid1 = lli->lli_pfid;
2308                 cfs_spin_unlock(&lli->lli_lock);
2309                 /** We ignore parent's capability temporary. */
2310         }
2311
2312         return op_data;
2313 }
2314
2315 void ll_finish_md_op_data(struct md_op_data *op_data)
2316 {
2317         capa_put(op_data->op_capa1);
2318         capa_put(op_data->op_capa2);
2319         OBD_FREE_PTR(op_data);
2320 }
2321
2322 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2323 {
2324         struct ll_sb_info *sbi;
2325
2326         LASSERT((seq != NULL) && (vfs != NULL));
2327         sbi = ll_s2sbi(vfs->mnt_sb);
2328
2329         if (sbi->ll_flags & LL_SBI_NOLCK)
2330                 seq_puts(seq, ",nolock");
2331
2332         if (sbi->ll_flags & LL_SBI_FLOCK)
2333                 seq_puts(seq, ",flock");
2334
2335         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2336                 seq_puts(seq, ",localflock");
2337
2338         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2339                 seq_puts(seq, ",user_xattr");
2340
2341         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2342                 seq_puts(seq, ",lazystatfs");
2343
2344         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2345                 seq_puts(seq, ",user_fid2path");
2346
2347         RETURN(0);
2348 }
2349
2350 /**
2351  * Get obd name by cmd, and copy out to user space
2352  */
2353 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2354 {
2355         struct ll_sb_info *sbi = ll_i2sbi(inode);
2356         struct obd_device *obd;
2357         ENTRY;
2358
2359         if (cmd == OBD_IOC_GETDTNAME)
2360                 obd = class_exp2obd(sbi->ll_dt_exp);
2361         else if (cmd == OBD_IOC_GETMDNAME)
2362                 obd = class_exp2obd(sbi->ll_md_exp);
2363         else
2364                 RETURN(-EINVAL);
2365
2366         if (!obd)
2367                 RETURN(-ENOENT);
2368
2369         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2370                              strlen(obd->obd_name) + 1))
2371                 RETURN(-EFAULT);
2372
2373         RETURN(0);
2374 }
2375
2376 /**
2377  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2378  * fsname will be returned in this buffer; otherwise, a static buffer will be
2379  * used to store the fsname and returned to caller.
2380  */
2381 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2382 {
2383         static char fsname_static[MTI_NAME_MAXLEN];
2384         struct lustre_sb_info *lsi = s2lsi(sb);
2385         char *ptr;
2386         int len;
2387
2388         if (buf == NULL) {
2389                 /* this means the caller wants to use static buffer
2390                  * and it doesn't care about race. Usually this is
2391                  * in error reporting path */
2392                 buf = fsname_static;
2393                 buflen = sizeof(fsname_static);
2394         }
2395
2396         len = strlen(lsi->lsi_lmd->lmd_profile);
2397         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2398         if (ptr && (strcmp(ptr, "-client") == 0))
2399                 len -= 7;
2400
2401         if (unlikely(len >= buflen))
2402                 len = buflen - 1;
2403         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2404         buf[len] = '\0';
2405
2406         return buf;
2407 }