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) 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         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         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_cache.ccc_users, 0);
 108         sbi->ll_cache.ccc_lru_max = lru_page_max;
 109         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 110         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 111         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 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         spin_lock(&ll_sb_lock);
 126         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 127         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                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 140                                pp_r_hist.oh_lock);
 141                 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                 spin_lock(&ll_sb_lock);
 162                 cfs_list_del(&sbi->ll_list);
 163                 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 | OBD_CONNECT_LVB_TYPE;
 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 | OBD_CONNECT_UMASK;
 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 | OBD_CONNECT_LVB_TYPE;
 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         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         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_CACHE_SET),
 555                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 556                                  &sbi->ll_cache, NULL);
 557
 558         sb->s_root = d_make_root(root);
 559         if (sb->s_root == NULL) {
 560                 CERROR("%s: can't make root dentry\n",
 561                         ll_get_fsname(sb, NULL, 0));
 562                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 563         }
 564
 565 #ifdef HAVE_DCACHE_LOCK
 566         sb->s_root->d_op = &ll_d_root_ops;
 567 #else
 568         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 569         d_set_d_op(sb->s_root, &ll_d_root_ops);
 570         sb->s_d_op = &ll_d_ops;
 571 #endif
 572
 573         sbi->ll_sdev_orig = sb->s_dev;
 574
 575         /* We set sb->s_dev equal on all lustre clients in order to support
 576          * NFS export clustering.  NFSD requires that the FSID be the same
 577          * on all clients. */
 578         /* s_dev is also used in lt_compare() to compare two fs, but that is
 579          * only a node-local comparison. */
 580         uuid = obd_get_uuid(sbi->ll_md_exp);
 581         if (uuid != NULL)
 582                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 583
 584         if (data != NULL)
 585                 OBD_FREE_PTR(data);
 586         if (osfs != NULL)
 587                 OBD_FREE_PTR(osfs);
 588
 589         RETURN(err);
 590 out_root:
 591         if (root)
 592                 iput(root);
 593 out_lock_cn_cb:
 594         obd_fid_fini(sbi->ll_dt_exp);
 595 out_dt:
 596         obd_disconnect(sbi->ll_dt_exp);
 597         sbi->ll_dt_exp = NULL;
 598 out_md_fid:
 599         obd_fid_fini(sbi->ll_md_exp);
 600 out_md:
 601         obd_disconnect(sbi->ll_md_exp);
 602         sbi->ll_md_exp = NULL;
 603 out:
 604         if (data != NULL)
 605                 OBD_FREE_PTR(data);
 606         if (osfs != NULL)
 607                 OBD_FREE_PTR(osfs);
 608         lprocfs_unregister_mountpoint(sbi);
 609         return err;
 610 }
 611
 612 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 613 {
 614         int size, rc;
 615
 616         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 617         size = sizeof(int);
 618         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 619                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 620         if (rc)
 621                 CERROR("Get max mdsize error rc %d \n", rc);
 622
 623         RETURN(rc);
 624 }
 625
 626 void ll_dump_inode(struct inode *inode)
 627 {
 628         struct ll_d_hlist_node *tmp;
 629         int dentry_count = 0;
 630
 631         LASSERT(inode != NULL);
 632
 633         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 634                 dentry_count++;
 635
 636         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 637                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 638                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 639 }
 640
 641 void lustre_dump_dentry(struct dentry *dentry, int recur)
 642 {
 643         struct list_head *tmp;
 644         int subdirs = 0;
 645
 646         LASSERT(dentry != NULL);
 647
 648         list_for_each(tmp, &dentry->d_subdirs)
 649                 subdirs++;
 650
 651         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 652                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 653                dentry->d_name.len, dentry->d_name.name,
 654                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 655                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 656                dentry->d_flags, dentry->d_fsdata, subdirs);
 657         if (dentry->d_inode != NULL)
 658                 ll_dump_inode(dentry->d_inode);
 659
 660         if (recur == 0)
 661                 return;
 662
 663         list_for_each(tmp, &dentry->d_subdirs) {
 664                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 665                 lustre_dump_dentry(d, recur - 1);
 666         }
 667 }
 668
 669 void client_common_put_super(struct super_block *sb)
 670 {
 671         struct ll_sb_info *sbi = ll_s2sbi(sb);
 672         ENTRY;
 673
 674 #ifdef CONFIG_FS_POSIX_ACL
 675         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 676                 et_fini(&sbi->ll_et);
 677                 rct_fini(&sbi->ll_rct);
 678         }
 679 #endif
 680
 681         ll_close_thread_shutdown(sbi->ll_lcq);
 682
 683         cl_sb_fini(sb);
 684
 685         cfs_list_del(&sbi->ll_conn_chain);
 686
 687         obd_fid_fini(sbi->ll_dt_exp);
 688         obd_disconnect(sbi->ll_dt_exp);
 689         sbi->ll_dt_exp = NULL;
 690
 691         lprocfs_unregister_mountpoint(sbi);
 692
 693         obd_fid_fini(sbi->ll_md_exp);
 694         obd_disconnect(sbi->ll_md_exp);
 695         sbi->ll_md_exp = NULL;
 696
 697         EXIT;
 698 }
 699
 700 void ll_kill_super(struct super_block *sb)
 701 {
 702         struct ll_sb_info *sbi;
 703
 704         ENTRY;
 705
 706         /* not init sb ?*/
 707         if (!(sb->s_flags & MS_ACTIVE))
 708                 return;
 709
 710         sbi = ll_s2sbi(sb);
 711         /* we need restore s_dev from changed for clustred NFS before put_super
 712          * because new kernels have cached s_dev and change sb->s_dev in
 713          * put_super not affected real removing devices */
 714         if (sbi)
 715                 sb->s_dev = sbi->ll_sdev_orig;
 716         EXIT;
 717 }
 718
 719 char *ll_read_opt(const char *opt, char *data)
 720 {
 721         char *value;
 722         char *retval;
 723         ENTRY;
 724
 725         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 726         if (strncmp(opt, data, strlen(opt)))
 727                 RETURN(NULL);
 728         if ((value = strchr(data, '=')) == NULL)
 729                 RETURN(NULL);
 730
 731         value++;
 732         OBD_ALLOC(retval, strlen(value) + 1);
 733         if (!retval) {
 734                 CERROR("out of memory!\n");
 735                 RETURN(NULL);
 736         }
 737
 738         memcpy(retval, value, strlen(value)+1);
 739         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 740         RETURN(retval);
 741 }
 742
 743 static inline int ll_set_opt(const char *opt, char *data, int fl)
 744 {
 745         if (strncmp(opt, data, strlen(opt)) != 0)
 746                 return(0);
 747         else
 748                 return(fl);
 749 }
 750
 751 /* non-client-specific mount options are parsed in lmd_parse */
 752 static int ll_options(char *options, int *flags)
 753 {
 754         int tmp;
 755         char *s1 = options, *s2;
 756         ENTRY;
 757
 758         if (!options)
 759                 RETURN(0);
 760
 761         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 762
 763         while (*s1) {
 764                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 765                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 766                 if (tmp) {
 767                         *flags |= tmp;
 768                         goto next;
 769                 }
 770                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 771                 if (tmp) {
 772                         *flags |= tmp;
 773                         goto next;
 774                 }
 775                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 776                 if (tmp) {
 777                         *flags |= tmp;
 778                         goto next;
 779                 }
 780                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 781                 if (tmp) {
 782                         *flags &= ~tmp;
 783                         goto next;
 784                 }
 785                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 786                 if (tmp) {
 787                         *flags |= tmp;
 788                         goto next;
 789                 }
 790                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 791                 if (tmp) {
 792                         *flags &= ~tmp;
 793                         goto next;
 794                 }
 795 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 796                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 797                 if (tmp) {
 798                         /* Ignore deprecated mount option.  The client will
 799                          * always try to mount with ACL support, whether this
 800                          * is used depends on whether server supports it. */
 801                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 802                                                   "mount option 'acl'.\n");
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 806                 if (tmp) {
 807                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 808                                                   "mount option 'noacl'.\n");
 809                         goto next;
 810                 }
 811 #else
 812 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 813 #endif
 814                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 825                 if (tmp) {
 826                         *flags &= ~tmp;
 827                         goto next;
 828                 }
 829
 830                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags &= ~tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 846                 if (tmp) {
 847                         *flags &= ~tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 856                 if (tmp) {
 857                         *flags &= ~tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 861                 if (tmp) {
 862                         *flags |= tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 871                 if (tmp) {
 872                         *flags |= tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 876                 if (tmp) {
 877                         *flags &= ~tmp;
 878                         goto next;
 879                 }
 880                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 881                                    s1);
 882                 RETURN(-EINVAL);
 883
 884 next:
 885                 /* Find next opt */
 886                 s2 = strchr(s1, ',');
 887                 if (s2 == NULL)
 888                         break;
 889                 s1 = s2 + 1;
 890         }
 891         RETURN(0);
 892 }
 893
 894 void ll_lli_init(struct ll_inode_info *lli)
 895 {
 896         lli->lli_inode_magic = LLI_INODE_MAGIC;
 897         lli->lli_flags = 0;
 898         lli->lli_ioepoch = 0;
 899         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 900         spin_lock_init(&lli->lli_lock);
 901         lli->lli_posix_acl = NULL;
 902         lli->lli_remote_perms = NULL;
 903         mutex_init(&lli->lli_rmtperm_mutex);
 904         /* Do not set lli_fid, it has been initialized already. */
 905         fid_zero(&lli->lli_pfid);
 906         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 907         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 908         cfs_atomic_set(&lli->lli_open_count, 0);
 909         lli->lli_mds_capa = NULL;
 910         lli->lli_rmtperm_time = 0;
 911         lli->lli_pending_och = NULL;
 912         lli->lli_mds_read_och = NULL;
 913         lli->lli_mds_write_och = NULL;
 914         lli->lli_mds_exec_och = NULL;
 915         lli->lli_open_fd_read_count = 0;
 916         lli->lli_open_fd_write_count = 0;
 917         lli->lli_open_fd_exec_count = 0;
 918         mutex_init(&lli->lli_och_mutex);
 919         spin_lock_init(&lli->lli_agl_lock);
 920         lli->lli_has_smd = false;
 921         lli->lli_clob = NULL;
 922
 923         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 924         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 925                 mutex_init(&lli->lli_readdir_mutex);
 926                 lli->lli_opendir_key = NULL;
 927                 lli->lli_sai = NULL;
 928                 lli->lli_def_acl = NULL;
 929                 spin_lock_init(&lli->lli_sa_lock);
 930                 lli->lli_opendir_pid = 0;
 931         } else {
 932                 sema_init(&lli->lli_size_sem, 1);
 933                 lli->lli_size_sem_owner = NULL;
 934                 lli->lli_symlink_name = NULL;
 935                 init_rwsem(&lli->lli_trunc_sem);
 936                 mutex_init(&lli->lli_write_mutex);
 937                 init_rwsem(&lli->lli_glimpse_sem);
 938                 lli->lli_glimpse_time = 0;
 939                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 940                 lli->lli_agl_index = 0;
 941                 lli->lli_async_rc = 0;
 942         }
 943         mutex_init(&lli->lli_layout_mutex);
 944 }
 945
 946 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 947 {
 948 #ifdef HAVE_BDI_REGISTER
 949         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 950
 951 #ifdef HAVE_BDI_NAME
 952         bdi->name = "lustre";
 953 #endif
 954         return bdi_register(bdi, NULL, "lustre-%d",
 955                             atomic_inc_return(&ll_bdi_num));
 956 #else
 957         return 0;
 958 #endif
 959 }
 960
 961 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 962 {
 963         struct lustre_profile *lprof = NULL;
 964         struct lustre_sb_info *lsi = s2lsi(sb);
 965         struct ll_sb_info *sbi;
 966         char  *dt = NULL, *md = NULL;
 967         char  *profilenm = get_profile_name(sb);
 968         struct config_llog_instance *cfg;
 969         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 970         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 971         int    err;
 972         ENTRY;
 973
 974         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 975
 976         OBD_ALLOC_PTR(cfg);
 977         if (cfg == NULL)
 978                 RETURN(-ENOMEM);
 979
 980         cfs_module_get();
 981
 982         /* client additional sb info */
 983         lsi->lsi_llsbi = sbi = ll_init_sbi();
 984         if (!sbi) {
 985                 cfs_module_put(THIS_MODULE);
 986                 OBD_FREE_PTR(cfg);
 987                 RETURN(-ENOMEM);
 988         }
 989
 990         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 991         if (err)
 992                 GOTO(out_free, err);
 993
 994         err = ll_bdi_init(&lsi->lsi_bdi);
 995         if (err)
 996                 GOTO(out_free, err);
 997         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 998         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 999         err = ll_bdi_register(&lsi->lsi_bdi);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003 #ifdef HAVE_SB_BDI
1004         sb->s_bdi = &lsi->lsi_bdi;
1005 #endif
1006
1007         /* Generate a string unique to this super, in case some joker tries
1008            to mount the same fs at two mount points.
1009            Use the address of the super itself.*/
1010         cfg->cfg_instance = sb;
1011         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1012         cfg->cfg_callback = class_config_llog_handler;
1013         /* set up client obds */
1014         err = lustre_process_log(sb, profilenm, cfg);
1015         if (err < 0) {
1016                 CERROR("Unable to process log: %d\n", err);
1017                 GOTO(out_free, err);
1018         }
1019
1020         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1021         lprof = class_get_profile(profilenm);
1022         if (lprof == NULL) {
1023                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1024                                    " read from the MGS.  Does that filesystem "
1025                                    "exist?\n", profilenm);
1026                 GOTO(out_free, err = -EINVAL);
1027         }
1028         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1029                lprof->lp_md, lprof->lp_dt);
1030
1031         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1032         if (!dt)
1033                 GOTO(out_free, err = -ENOMEM);
1034         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1035
1036         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1037         if (!md)
1038                 GOTO(out_free, err = -ENOMEM);
1039         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1040
1041         /* connections, registrations, sb setup */
1042         err = client_common_fill_super(sb, md, dt, mnt);
1043
1044 out_free:
1045         if (md)
1046                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1047         if (dt)
1048                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1049         if (err)
1050                 ll_put_super(sb);
1051         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1052                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1053
1054         OBD_FREE_PTR(cfg);
1055         RETURN(err);
1056 } /* ll_fill_super */
1057
1058
1059 void lu_context_keys_dump(void);
1060
1061 void ll_put_super(struct super_block *sb)
1062 {
1063         struct config_llog_instance cfg;
1064         struct obd_device *obd;
1065         struct lustre_sb_info *lsi = s2lsi(sb);
1066         struct ll_sb_info *sbi = ll_s2sbi(sb);
1067         char *profilenm = get_profile_name(sb);
1068         int force = 1, next;
1069         ENTRY;
1070
1071         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072
1073         ll_print_capa_stat(sbi);
1074
1075         cfg.cfg_instance = sb;
1076         lustre_end_log(sb, profilenm, &cfg);
1077
1078         if (sbi->ll_md_exp) {
1079                 obd = class_exp2obd(sbi->ll_md_exp);
1080                 if (obd)
1081                         force = obd->obd_force;
1082         }
1083
1084         /* We need to set force before the lov_disconnect in
1085            lustre_common_put_super, since l_d cleans up osc's as well. */
1086         if (force) {
1087                 next = 0;
1088                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1089                                                      &next)) != NULL) {
1090                         obd->obd_force = force;
1091                 }
1092         }
1093
1094         if (sbi->ll_lcq) {
1095                 /* Only if client_common_fill_super succeeded */
1096                 client_common_put_super(sb);
1097         }
1098
1099         next = 0;
1100         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1101                 class_manual_cleanup(obd);
1102         }
1103
1104         if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1106
1107         if (profilenm)
1108                 class_del_profile(profilenm);
1109
1110         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1111                 ll_bdi_destroy(&lsi->lsi_bdi);
1112                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1113         }
1114
1115         ll_free_sbi(sb);
1116         lsi->lsi_llsbi = NULL;
1117
1118         lustre_common_put_super(sb);
1119
1120         cl_env_cache_purge(~0);
1121
1122         cfs_module_put(THIS_MODULE);
1123
1124         EXIT;
1125 } /* client_put_super */
1126
1127 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1128 {
1129         struct inode *inode = NULL;
1130         /* NOTE: we depend on atomic igrab() -bzzz */
1131         lock_res_and_lock(lock);
1132         if (lock->l_ast_data) {
1133                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1134                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1135                         inode = igrab(lock->l_ast_data);
1136                 } else {
1137                         inode = lock->l_ast_data;
1138                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1139                                          D_WARNING, lock, "l_ast_data %p is "
1140                                          "bogus: magic %08x", lock->l_ast_data,
1141                                          lli->lli_inode_magic);
1142                         inode = NULL;
1143                 }
1144         }
1145         unlock_res_and_lock(lock);
1146         return inode;
1147 }
1148
1149 static int null_if_equal(struct ldlm_lock *lock, void *data)
1150 {
1151         if (data == lock->l_ast_data) {
1152                 lock->l_ast_data = NULL;
1153
1154                 if (lock->l_req_mode != lock->l_granted_mode)
1155                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1156         }
1157
1158         return LDLM_ITER_CONTINUE;
1159 }
1160
1161 void ll_clear_inode(struct inode *inode)
1162 {
1163         struct ll_inode_info *lli = ll_i2info(inode);
1164         struct ll_sb_info *sbi = ll_i2sbi(inode);
1165         ENTRY;
1166
1167         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1168                inode->i_generation, inode);
1169
1170         if (S_ISDIR(inode->i_mode)) {
1171                 /* these should have been cleared in ll_file_release */
1172                 LASSERT(lli->lli_opendir_key == NULL);
1173                 LASSERT(lli->lli_sai == NULL);
1174                 LASSERT(lli->lli_opendir_pid == 0);
1175         }
1176
1177         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1178         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1179                          null_if_equal, inode);
1180
1181         LASSERT(!lli->lli_open_fd_write_count);
1182         LASSERT(!lli->lli_open_fd_read_count);
1183         LASSERT(!lli->lli_open_fd_exec_count);
1184
1185         if (lli->lli_mds_write_och)
1186                 ll_md_real_close(inode, FMODE_WRITE);
1187         if (lli->lli_mds_exec_och)
1188                 ll_md_real_close(inode, FMODE_EXEC);
1189         if (lli->lli_mds_read_och)
1190                 ll_md_real_close(inode, FMODE_READ);
1191
1192         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1193                 OBD_FREE(lli->lli_symlink_name,
1194                          strlen(lli->lli_symlink_name) + 1);
1195                 lli->lli_symlink_name = NULL;
1196         }
1197
1198         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1199                 LASSERT(lli->lli_posix_acl == NULL);
1200                 if (lli->lli_remote_perms) {
1201                         free_rmtperm_hash(lli->lli_remote_perms);
1202                         lli->lli_remote_perms = NULL;
1203                 }
1204         }
1205 #ifdef CONFIG_FS_POSIX_ACL
1206         else if (lli->lli_posix_acl) {
1207                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1208                 LASSERT(lli->lli_remote_perms == NULL);
1209                 posix_acl_release(lli->lli_posix_acl);
1210                 lli->lli_posix_acl = NULL;
1211         }
1212 #endif
1213         lli->lli_inode_magic = LLI_INODE_DEAD;
1214
1215         ll_clear_inode_capas(inode);
1216         if (!S_ISDIR(inode->i_mode))
1217                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1218
1219         /*
1220          * XXX This has to be done before lsm is freed below, because
1221          * cl_object still uses inode lsm.
1222          */
1223         cl_inode_fini(inode);
1224         lli->lli_has_smd = false;
1225
1226         EXIT;
1227 }
1228
1229 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1230                   struct md_open_data **mod)
1231 {
1232         struct lustre_md md;
1233         struct inode *inode = dentry->d_inode;
1234         struct ll_sb_info *sbi = ll_i2sbi(inode);
1235         struct ptlrpc_request *request = NULL;
1236         int rc, ia_valid;
1237         ENTRY;
1238
1239         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1240                                      LUSTRE_OPC_ANY, NULL);
1241         if (IS_ERR(op_data))
1242                 RETURN(PTR_ERR(op_data));
1243
1244         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1245                         &request, mod);
1246         if (rc) {
1247                 ptlrpc_req_finished(request);
1248                 if (rc == -ENOENT) {
1249                         clear_nlink(inode);
1250                         /* Unlinked special device node? Or just a race?
1251                          * Pretend we done everything. */
1252                         if (!S_ISREG(inode->i_mode) &&
1253                             !S_ISDIR(inode->i_mode)) {
1254                                 ia_valid = op_data->op_attr.ia_valid;
1255                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1256                                 rc = simple_setattr(dentry, &op_data->op_attr);
1257                                 op_data->op_attr.ia_valid = ia_valid;
1258                         }
1259                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1260                         CERROR("md_setattr fails: rc = %d\n", rc);
1261                 }
1262                 RETURN(rc);
1263         }
1264
1265         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1266                               sbi->ll_md_exp, &md);
1267         if (rc) {
1268                 ptlrpc_req_finished(request);
1269                 RETURN(rc);
1270         }
1271
1272         /* We want to adjust timestamps.
1273          * If there is at least some data in file, we cleared ATTR_SIZE
1274          * to avoid update size, otherwise it is important to do.(SOM case)
1275          * (bug 6196) */
1276         ia_valid = op_data->op_attr.ia_valid;
1277         /* Since we set ATTR_*_SET flags above, and already done permission
1278          * check, So don't let inode_change_ok() check it again.  */
1279         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1280         rc = simple_setattr(dentry, &op_data->op_attr);
1281         op_data->op_attr.ia_valid = ia_valid;
1282
1283         /* Extract epoch data if obtained. */
1284         op_data->op_handle = md.body->handle;
1285         op_data->op_ioepoch = md.body->ioepoch;
1286
1287         ll_update_inode(inode, &md);
1288         ptlrpc_req_finished(request);
1289
1290         RETURN(rc);
1291 }
1292
1293 /* Close IO epoch and send Size-on-MDS attribute update. */
1294 static int ll_setattr_done_writing(struct inode *inode,
1295                                    struct md_op_data *op_data,
1296                                    struct md_open_data *mod)
1297 {
1298         struct ll_inode_info *lli = ll_i2info(inode);
1299         int rc = 0;
1300         ENTRY;
1301
1302         LASSERT(op_data != NULL);
1303         if (!S_ISREG(inode->i_mode))
1304                 RETURN(0);
1305
1306         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1307                op_data->op_ioepoch, PFID(&lli->lli_fid));
1308
1309         op_data->op_flags = MF_EPOCH_CLOSE;
1310         ll_done_writing_attr(inode, op_data);
1311         ll_pack_inode2opdata(inode, op_data, NULL);
1312
1313         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1314         if (rc == -EAGAIN) {
1315                 /* MDS has instructed us to obtain Size-on-MDS attribute
1316                  * from OSTs and send setattr to back to MDS. */
1317                 rc = ll_som_update(inode, op_data);
1318         } else if (rc) {
1319                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1320                        inode->i_ino, rc);
1321         }
1322         RETURN(rc);
1323 }
1324
1325 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1326 {
1327         struct obd_capa *capa;
1328         int rc;
1329
1330         if (attr->ia_valid & ATTR_SIZE)
1331                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1332         else
1333                 capa = ll_mdscapa_get(inode);
1334
1335         rc = cl_setattr_ost(inode, attr, capa);
1336
1337         if (attr->ia_valid & ATTR_SIZE)
1338                 ll_truncate_free_capa(capa);
1339         else
1340                 capa_put(capa);
1341
1342         return rc;
1343 }
1344
1345 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1346 /**
1347  * inode_newsize_ok - may this inode be truncated to a given size
1348  * @inode:      the inode to be truncated
1349  * @offset:     the new size to assign to the inode
1350  * @Returns:    0 on success, -ve errno on failure
1351  *
1352  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1353  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1354  * when necessary. Caller must not proceed with inode size change if failure is
1355  * returned. @inode must be a file (not directory), with appropriate
1356  * permissions to allow truncate (inode_newsize_ok does NOT check these
1357  * conditions).
1358  *
1359  * inode_newsize_ok must be called with i_mutex held.
1360  */
1361 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1362 {
1363         if (inode->i_size < offset) {
1364                 unsigned long limit;
1365
1366                 limit = rlimit(RLIMIT_FSIZE);
1367                 if (limit != RLIM_INFINITY && offset > limit)
1368                         goto out_sig;
1369                 if (offset > inode->i_sb->s_maxbytes)
1370                         goto out_big;
1371         } else {
1372                 /*
1373                  * truncation of in-use swapfiles is disallowed - it would
1374                  * cause subsequent swapout to scribble on the now-freed
1375                  * blocks.
1376                  */
1377                 if (IS_SWAPFILE(inode))
1378                         return -ETXTBSY;
1379         }
1380
1381         return 0;
1382 out_sig:
1383         send_sig(SIGXFSZ, current, 0);
1384 out_big:
1385         return -EFBIG;
1386 }
1387 #endif
1388
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1391  * keep these values until such a time that objects are allocated for it.
1392  * We do the MDS operations first, as it is checking permissions for us.
1393  * We don't to the MDS RPC if there is nothing that we want to store there,
1394  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395  * going to do an RPC anyways.
1396  *
1397  * If we are doing a truncate, we will send the mtime and ctime updates
1398  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1400  * at the same time.
1401  */
1402 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1403 {
1404         struct lov_stripe_md *lsm;
1405         struct inode *inode = dentry->d_inode;
1406         struct ll_inode_info *lli = ll_i2info(inode);
1407         struct md_op_data *op_data = NULL;
1408         struct md_open_data *mod = NULL;
1409         int ia_valid = attr->ia_valid;
1410         int rc = 0, rc1 = 0;
1411         ENTRY;
1412
1413         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1414                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1415                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1416                 attr->ia_valid);
1417
1418         if (ia_valid & ATTR_SIZE) {
1419                 /* Check new size against VFS/VM file size limit and rlimit */
1420                 rc = inode_newsize_ok(inode, attr->ia_size);
1421                 if (rc)
1422                         RETURN(rc);
1423
1424                 /* The maximum Lustre file size is variable, based on the
1425                  * OST maximum object size and number of stripes.  This
1426                  * needs another check in addition to the VFS check above. */
1427                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1428                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1429                                PFID(&lli->lli_fid), attr->ia_size,
1430                                ll_file_maxbytes(inode));
1431                         RETURN(-EFBIG);
1432                 }
1433
1434                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1435         }
1436
1437         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1438         if (ia_valid & TIMES_SET_FLAGS) {
1439                 if (cfs_curproc_fsuid() != inode->i_uid &&
1440                     !cfs_capable(CFS_CAP_FOWNER))
1441                         RETURN(-EPERM);
1442         }
1443
1444         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1445         if (attr->ia_valid & ATTR_CTIME) {
1446                 attr->ia_ctime = CFS_CURRENT_TIME;
1447                 attr->ia_valid |= ATTR_CTIME_SET;
1448         }
1449         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1450                 attr->ia_atime = CFS_CURRENT_TIME;
1451                 attr->ia_valid |= ATTR_ATIME_SET;
1452         }
1453         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1454                 attr->ia_mtime = CFS_CURRENT_TIME;
1455                 attr->ia_valid |= ATTR_MTIME_SET;
1456         }
1457
1458         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1459                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1460                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1461                        cfs_time_current_sec());
1462
1463         /* We always do an MDS RPC, even if we're only changing the size;
1464          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1465
1466         OBD_ALLOC_PTR(op_data);
1467         if (op_data == NULL)
1468                 RETURN(-ENOMEM);
1469
1470         if (!S_ISDIR(inode->i_mode)) {
1471                 if (ia_valid & ATTR_SIZE)
1472                         inode_dio_write_done(inode);
1473                 mutex_unlock(&inode->i_mutex);
1474                 down_write(&lli->lli_trunc_sem);
1475                 mutex_lock(&inode->i_mutex);
1476                 if (ia_valid & ATTR_SIZE)
1477                         inode_dio_wait(inode);
1478         }
1479
1480         /* We need a steady stripe configuration for setattr to avoid
1481          * confusion. */
1482         lsm = ccc_inode_lsm_get(inode);
1483
1484         /* NB: ATTR_SIZE will only be set after this point if the size
1485          * resides on the MDS, ie, this file has no objects. */
1486         if (lsm != NULL)
1487                 attr->ia_valid &= ~ATTR_SIZE;
1488         /* can't call ll_setattr_ost() while holding a refcount of lsm */
1489         ccc_inode_lsm_put(inode, lsm);
1490
1491         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1492
1493         /* Open epoch for truncate. */
1494         if (exp_connect_som(ll_i2mdexp(inode)) &&
1495             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1496                 op_data->op_flags = MF_EPOCH_OPEN;
1497
1498         rc = ll_md_setattr(dentry, op_data, &mod);
1499         if (rc)
1500                 GOTO(out, rc);
1501
1502         ll_ioepoch_open(lli, op_data->op_ioepoch);
1503         if (!S_ISREG(inode->i_mode))
1504                 GOTO(out, rc = 0);
1505
1506         if (ia_valid & ATTR_SIZE)
1507                 attr->ia_valid |= ATTR_SIZE;
1508         if (ia_valid & (ATTR_SIZE |
1509                         ATTR_ATIME | ATTR_ATIME_SET |
1510                         ATTR_MTIME | ATTR_MTIME_SET))
1511                 /* For truncate and utimes sending attributes to OSTs, setting
1512                  * mtime/atime to the past will be performed under PW [0:EOF]
1513                  * extent lock (new_size:EOF for truncate).  It may seem
1514                  * excessive to send mtime/atime updates to OSTs when not
1515                  * setting times to past, but it is necessary due to possible
1516                  * time de-synchronization between MDT inode and OST objects */
1517                 rc = ll_setattr_ost(inode, attr);
1518         EXIT;
1519 out:
1520         if (op_data) {
1521                 if (op_data->op_ioepoch) {
1522                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1523                         if (!rc)
1524                                 rc = rc1;
1525                 }
1526                 ll_finish_md_op_data(op_data);
1527         }
1528         if (!S_ISDIR(inode->i_mode))
1529                 up_write(&lli->lli_trunc_sem);
1530
1531         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1532                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1533
1534         return rc;
1535 }
1536
1537 int ll_setattr(struct dentry *de, struct iattr *attr)
1538 {
1539         int mode = de->d_inode->i_mode;
1540
1541         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1542                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1543                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1544
1545         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1546                                (ATTR_SIZE|ATTR_MODE)) &&
1547             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1548              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1549               !(attr->ia_mode & S_ISGID))))
1550                 attr->ia_valid |= ATTR_FORCE;
1551
1552         if ((mode & S_ISUID) &&
1553             !(attr->ia_mode & S_ISUID) &&
1554             !(attr->ia_valid & ATTR_KILL_SUID))
1555                 attr->ia_valid |= ATTR_KILL_SUID;
1556
1557         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1558             !(attr->ia_mode & S_ISGID) &&
1559             !(attr->ia_valid & ATTR_KILL_SGID))
1560                 attr->ia_valid |= ATTR_KILL_SGID;
1561
1562         return ll_setattr_raw(de, attr);
1563 }
1564
1565 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1566                        __u64 max_age, __u32 flags)
1567 {
1568         struct ll_sb_info *sbi = ll_s2sbi(sb);
1569         struct obd_statfs obd_osfs;
1570         int rc;
1571         ENTRY;
1572
1573         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1574         if (rc) {
1575                 CERROR("md_statfs fails: rc = %d\n", rc);
1576                 RETURN(rc);
1577         }
1578
1579         osfs->os_type = sb->s_magic;
1580
1581         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1582                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1583
1584         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1585                 flags |= OBD_STATFS_NODELAY;
1586
1587         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1588         if (rc) {
1589                 CERROR("obd_statfs fails: rc = %d\n", rc);
1590                 RETURN(rc);
1591         }
1592
1593         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1594                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1595                obd_osfs.os_files);
1596
1597         osfs->os_bsize = obd_osfs.os_bsize;
1598         osfs->os_blocks = obd_osfs.os_blocks;
1599         osfs->os_bfree = obd_osfs.os_bfree;
1600         osfs->os_bavail = obd_osfs.os_bavail;
1601
1602         /* If we don't have as many objects free on the OST as inodes
1603          * on the MDS, we reduce the total number of inodes to
1604          * compensate, so that the "inodes in use" number is correct.
1605          */
1606         if (obd_osfs.os_ffree < osfs->os_ffree) {
1607                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1608                         obd_osfs.os_ffree;
1609                 osfs->os_ffree = obd_osfs.os_ffree;
1610         }
1611
1612         RETURN(rc);
1613 }
1614 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1615 {
1616         struct super_block *sb = de->d_sb;
1617         struct obd_statfs osfs;
1618         int rc;
1619
1620         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1621         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1622
1623         /* Some amount of caching on the client is allowed */
1624         rc = ll_statfs_internal(sb, &osfs,
1625                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1626                                 0);
1627         if (rc)
1628                 return rc;
1629
1630         statfs_unpack(sfs, &osfs);
1631
1632         /* We need to downshift for all 32-bit kernels, because we can't
1633          * tell if the kernel is being called via sys_statfs64() or not.
1634          * Stop before overflowing f_bsize - in which case it is better
1635          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1636         if (sizeof(long) < 8) {
1637                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1638                         sfs->f_bsize <<= 1;
1639
1640                         osfs.os_blocks >>= 1;
1641                         osfs.os_bfree >>= 1;
1642                         osfs.os_bavail >>= 1;
1643                 }
1644         }
1645
1646         sfs->f_blocks = osfs.os_blocks;
1647         sfs->f_bfree = osfs.os_bfree;
1648         sfs->f_bavail = osfs.os_bavail;
1649
1650         return 0;
1651 }
1652
1653 void ll_inode_size_lock(struct inode *inode)
1654 {
1655         struct ll_inode_info *lli;
1656
1657         LASSERT(!S_ISDIR(inode->i_mode));
1658
1659         lli = ll_i2info(inode);
1660         LASSERT(lli->lli_size_sem_owner != current);
1661         down(&lli->lli_size_sem);
1662         LASSERT(lli->lli_size_sem_owner == NULL);
1663         lli->lli_size_sem_owner = current;
1664 }
1665
1666 void ll_inode_size_unlock(struct inode *inode)
1667 {
1668         struct ll_inode_info *lli;
1669
1670         lli = ll_i2info(inode);
1671         LASSERT(lli->lli_size_sem_owner == current);
1672         lli->lli_size_sem_owner = NULL;
1673         up(&lli->lli_size_sem);
1674 }
1675
1676 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1677 {
1678         struct ll_inode_info *lli = ll_i2info(inode);
1679         struct mdt_body *body = md->body;
1680         struct lov_stripe_md *lsm = md->lsm;
1681         struct ll_sb_info *sbi = ll_i2sbi(inode);
1682
1683         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1684         if (lsm != NULL) {
1685                 LASSERT(S_ISREG(inode->i_mode));
1686                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1687                                 lsm, inode->i_ino, inode->i_generation, inode);
1688                 /* cl_file_inode_init must go before lli_has_smd or a race
1689                  * is possible where client thinks the file has stripes,
1690                  * but lov raid0 is not setup yet and parallel e.g.
1691                  * glimpse would try to use uninitialized lov */
1692                 if (cl_file_inode_init(inode, md) == 0)
1693                         lli->lli_has_smd = true;
1694
1695                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1696                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1697                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1698                 if (md->lsm != NULL)
1699                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1700         }
1701
1702         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1703                 if (body->valid & OBD_MD_FLRMTPERM)
1704                         ll_update_remote_perm(inode, md->remote_perm);
1705         }
1706 #ifdef CONFIG_FS_POSIX_ACL
1707         else if (body->valid & OBD_MD_FLACL) {
1708                 spin_lock(&lli->lli_lock);
1709                 if (lli->lli_posix_acl)
1710                         posix_acl_release(lli->lli_posix_acl);
1711                 lli->lli_posix_acl = md->posix_acl;
1712                 spin_unlock(&lli->lli_lock);
1713         }
1714 #endif
1715         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1716         inode->i_generation = cl_fid_build_gen(&body->fid1);
1717
1718         if (body->valid & OBD_MD_FLATIME) {
1719                 if (body->atime > LTIME_S(inode->i_atime))
1720                         LTIME_S(inode->i_atime) = body->atime;
1721                 lli->lli_lvb.lvb_atime = body->atime;
1722         }
1723         if (body->valid & OBD_MD_FLMTIME) {
1724                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1725                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1726                                "to "LPU64"\n", inode->i_ino,
1727                                LTIME_S(inode->i_mtime), body->mtime);
1728                         LTIME_S(inode->i_mtime) = body->mtime;
1729                 }
1730                 lli->lli_lvb.lvb_mtime = body->mtime;
1731         }
1732         if (body->valid & OBD_MD_FLCTIME) {
1733                 if (body->ctime > LTIME_S(inode->i_ctime))
1734                         LTIME_S(inode->i_ctime) = body->ctime;
1735                 lli->lli_lvb.lvb_ctime = body->ctime;
1736         }
1737         if (body->valid & OBD_MD_FLMODE)
1738                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1739         if (body->valid & OBD_MD_FLTYPE)
1740                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1741         LASSERT(inode->i_mode != 0);
1742         if (S_ISREG(inode->i_mode)) {
1743                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1744         } else {
1745                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1746         }
1747         if (body->valid & OBD_MD_FLUID)
1748                 inode->i_uid = body->uid;
1749         if (body->valid & OBD_MD_FLGID)
1750                 inode->i_gid = body->gid;
1751         if (body->valid & OBD_MD_FLFLAGS)
1752                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1753         if (body->valid & OBD_MD_FLNLINK)
1754                 set_nlink(inode, body->nlink);
1755         if (body->valid & OBD_MD_FLRDEV)
1756                 inode->i_rdev = old_decode_dev(body->rdev);
1757
1758         if (body->valid & OBD_MD_FLID) {
1759                 /* FID shouldn't be changed! */
1760                 if (fid_is_sane(&lli->lli_fid)) {
1761                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1762                                  "Trying to change FID "DFID
1763                                  " to the "DFID", inode %lu/%u(%p)\n",
1764                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1765                                  inode->i_ino, inode->i_generation, inode);
1766                 } else
1767                         lli->lli_fid = body->fid1;
1768         }
1769
1770         LASSERT(fid_seq(&lli->lli_fid) != 0);
1771
1772         if (body->valid & OBD_MD_FLSIZE) {
1773                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1774                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1775                         struct lustre_handle lockh;
1776                         ldlm_mode_t mode;
1777
1778                         /* As it is possible a blocking ast has been processed
1779                          * by this time, we need to check there is an UPDATE
1780                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1781                          * it. */
1782                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1783                                                &lockh, LDLM_FL_CBPENDING);
1784                         if (mode) {
1785                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1786                                                       LLIF_EPOCH_PENDING |
1787                                                       LLIF_SOM_DIRTY)) {
1788                                         CERROR("ino %lu flags %u still has "
1789                                                "size authority! do not trust "
1790                                                "the size got from MDS\n",
1791                                                inode->i_ino, lli->lli_flags);
1792                                 } else {
1793                                         /* Use old size assignment to avoid
1794                                          * deadlock bz14138 & bz14326 */
1795                                         i_size_write(inode, body->size);
1796                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1797                                 }
1798                                 ldlm_lock_decref(&lockh, mode);
1799                         }
1800                 } else {
1801                         /* Use old size assignment to avoid
1802                          * deadlock bz14138 & bz14326 */
1803                         i_size_write(inode, body->size);
1804
1805                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1806                                inode->i_ino, (unsigned long long)body->size);
1807                 }
1808
1809                 if (body->valid & OBD_MD_FLBLOCKS)
1810                         inode->i_blocks = body->blocks;
1811         }
1812
1813         if (body->valid & OBD_MD_FLMDSCAPA) {
1814                 LASSERT(md->mds_capa);
1815                 ll_add_capa(inode, md->mds_capa);
1816         }
1817         if (body->valid & OBD_MD_FLOSSCAPA) {
1818                 LASSERT(md->oss_capa);
1819                 ll_add_capa(inode, md->oss_capa);
1820         }
1821 }
1822
1823 void ll_read_inode2(struct inode *inode, void *opaque)
1824 {
1825         struct lustre_md *md = opaque;
1826         struct ll_inode_info *lli = ll_i2info(inode);
1827         ENTRY;
1828
1829         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1830                PFID(&lli->lli_fid), inode);
1831
1832         LASSERT(!lli->lli_has_smd);
1833
1834         /* Core attributes from the MDS first.  This is a new inode, and
1835          * the VFS doesn't zero times in the core inode so we have to do
1836          * it ourselves.  They will be overwritten by either MDS or OST
1837          * attributes - we just need to make sure they aren't newer. */
1838         LTIME_S(inode->i_mtime) = 0;
1839         LTIME_S(inode->i_atime) = 0;
1840         LTIME_S(inode->i_ctime) = 0;
1841         inode->i_rdev = 0;
1842         ll_update_inode(inode, md);
1843
1844         /* OIDEBUG(inode); */
1845
1846         /* initializing backing dev info. */
1847         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1848
1849
1850         if (S_ISREG(inode->i_mode)) {
1851                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1852                 inode->i_op = &ll_file_inode_operations;
1853                 inode->i_fop = sbi->ll_fop;
1854                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1855                 EXIT;
1856         } else if (S_ISDIR(inode->i_mode)) {
1857                 inode->i_op = &ll_dir_inode_operations;
1858                 inode->i_fop = &ll_dir_operations;
1859                 EXIT;
1860         } else if (S_ISLNK(inode->i_mode)) {
1861                 inode->i_op = &ll_fast_symlink_inode_operations;
1862                 EXIT;
1863         } else {
1864                 inode->i_op = &ll_special_inode_operations;
1865
1866                 init_special_inode(inode, inode->i_mode,
1867                                    inode->i_rdev);
1868
1869                 EXIT;
1870         }
1871 }
1872
1873 void ll_delete_inode(struct inode *inode)
1874 {
1875         struct cl_inode_info *lli = cl_i2info(inode);
1876         ENTRY;
1877
1878         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1879                 /* discard all dirty pages before truncating them, required by
1880                  * osc_extent implementation at LU-1030. */
1881                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1882
1883         truncate_inode_pages(&inode->i_data, 0);
1884
1885         /* Workaround for LU-118 */
1886         if (inode->i_data.nrpages) {
1887                 TREE_READ_LOCK_IRQ(&inode->i_data);
1888                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1889                 LASSERTF(inode->i_data.nrpages == 0,
1890                          "inode=%lu/%u(%p) nrpages=%lu, see "
1891                          "http://jira.whamcloud.com/browse/LU-118\n",
1892                          inode->i_ino, inode->i_generation, inode,
1893                          inode->i_data.nrpages);
1894         }
1895         /* Workaround end */
1896
1897 #ifdef HAVE_SBOPS_EVICT_INODE
1898         ll_clear_inode(inode);
1899 #endif
1900         clear_inode(inode);
1901
1902         EXIT;
1903 }
1904
1905 int ll_iocontrol(struct inode *inode, struct file *file,
1906                  unsigned int cmd, unsigned long arg)
1907 {
1908         struct ll_sb_info *sbi = ll_i2sbi(inode);
1909         struct ptlrpc_request *req = NULL;
1910         int rc, flags = 0;
1911         ENTRY;
1912
1913         switch(cmd) {
1914         case FSFILT_IOC_GETFLAGS: {
1915                 struct mdt_body *body;
1916                 struct md_op_data *op_data;
1917
1918                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1919                                              0, 0, LUSTRE_OPC_ANY,
1920                                              NULL);
1921                 if (IS_ERR(op_data))
1922                         RETURN(PTR_ERR(op_data));
1923
1924                 op_data->op_valid = OBD_MD_FLFLAGS;
1925                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1926                 ll_finish_md_op_data(op_data);
1927                 if (rc) {
1928                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1929                         RETURN(-abs(rc));
1930                 }
1931
1932                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1933
1934                 flags = body->flags;
1935
1936                 ptlrpc_req_finished(req);
1937
1938                 RETURN(put_user(flags, (int *)arg));
1939         }
1940         case FSFILT_IOC_SETFLAGS: {
1941                 struct lov_stripe_md *lsm;
1942                 struct obd_info oinfo = { { { 0 } } };
1943                 struct md_op_data *op_data;
1944
1945                 if (get_user(flags, (int *)arg))
1946                         RETURN(-EFAULT);
1947
1948                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1949                                              LUSTRE_OPC_ANY, NULL);
1950                 if (IS_ERR(op_data))
1951                         RETURN(PTR_ERR(op_data));
1952
1953                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1954                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1955                 rc = md_setattr(sbi->ll_md_exp, op_data,
1956                                 NULL, 0, NULL, 0, &req, NULL);
1957                 ll_finish_md_op_data(op_data);
1958                 ptlrpc_req_finished(req);
1959                 if (rc)
1960                         RETURN(rc);
1961
1962                 inode->i_flags = ll_ext_to_inode_flags(flags);
1963
1964                 lsm = ccc_inode_lsm_get(inode);
1965                 if (lsm == NULL)
1966                         RETURN(0);
1967
1968                 OBDO_ALLOC(oinfo.oi_oa);
1969                 if (!oinfo.oi_oa) {
1970                         ccc_inode_lsm_put(inode, lsm);
1971                         RETURN(-ENOMEM);
1972                 }
1973                 oinfo.oi_md = lsm;
1974                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1975                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1976                 oinfo.oi_oa->o_flags = flags;
1977                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1978                                        OBD_MD_FLGROUP;
1979                 oinfo.oi_capa = ll_mdscapa_get(inode);
1980                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1981                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1982                 capa_put(oinfo.oi_capa);
1983                 OBDO_FREE(oinfo.oi_oa);
1984                 ccc_inode_lsm_put(inode, lsm);
1985
1986                 if (rc && rc != -EPERM && rc != -EACCES)
1987                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1988
1989                 RETURN(rc);
1990         }
1991         default:
1992                 RETURN(-ENOSYS);
1993         }
1994
1995         RETURN(0);
1996 }
1997
1998 int ll_flush_ctx(struct inode *inode)
1999 {
2000         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2001
2002         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2003
2004         obd_set_info_async(NULL, sbi->ll_md_exp,
2005                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2006                            0, NULL, NULL);
2007         obd_set_info_async(NULL, sbi->ll_dt_exp,
2008                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2009                            0, NULL, NULL);
2010         return 0;
2011 }
2012
2013 /* umount -f client means force down, don't save state */
2014 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2015 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2016 {
2017         struct super_block *sb = vfsmnt->mnt_sb;
2018 #else
2019 void ll_umount_begin(struct super_block *sb)
2020 {
2021 #endif
2022         struct ll_sb_info *sbi = ll_s2sbi(sb);
2023         struct obd_device *obd;
2024         struct obd_ioctl_data *ioc_data;
2025         ENTRY;
2026
2027 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2028         if (!(flags & MNT_FORCE)) {
2029                 EXIT;
2030                 return;
2031         }
2032 #endif
2033
2034         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2035                sb->s_count, atomic_read(&sb->s_active));
2036
2037         obd = class_exp2obd(sbi->ll_md_exp);
2038         if (obd == NULL) {
2039                 CERROR("Invalid MDC connection handle "LPX64"\n",
2040                        sbi->ll_md_exp->exp_handle.h_cookie);
2041                 EXIT;
2042                 return;
2043         }
2044         obd->obd_force = 1;
2045
2046         obd = class_exp2obd(sbi->ll_dt_exp);
2047         if (obd == NULL) {
2048                 CERROR("Invalid LOV connection handle "LPX64"\n",
2049                        sbi->ll_dt_exp->exp_handle.h_cookie);
2050                 EXIT;
2051                 return;
2052         }
2053         obd->obd_force = 1;
2054
2055         OBD_ALLOC_PTR(ioc_data);
2056         if (ioc_data) {
2057                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2058                               sizeof ioc_data, ioc_data, NULL);
2059
2060                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2061                               sizeof ioc_data, ioc_data, NULL);
2062
2063                 OBD_FREE_PTR(ioc_data);
2064         }
2065
2066
2067         /* Really, we'd like to wait until there are no requests outstanding,
2068          * and then continue.  For now, we just invalidate the requests,
2069          * schedule() and sleep one second if needed, and hope.
2070          */
2071         cfs_schedule();
2072 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2073         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2074                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2075                                                    cfs_time_seconds(1));
2076                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2077                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2078                                       "may try to umount it a bit later\n",
2079                                       atomic_read(&vfsmnt->mnt_count));
2080         }
2081 #endif
2082
2083         EXIT;
2084 }
2085
2086 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2087 {
2088         struct ll_sb_info *sbi = ll_s2sbi(sb);
2089         char *profilenm = get_profile_name(sb);
2090         int err;
2091         __u32 read_only;
2092
2093         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2094                 read_only = *flags & MS_RDONLY;
2095                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2096                                          sizeof(KEY_READ_ONLY),
2097                                          KEY_READ_ONLY, sizeof(read_only),
2098                                          &read_only, NULL);
2099                 if (err) {
2100                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2101                                       profilenm, read_only ?
2102                                       "read-only" : "read-write", err);
2103                         return err;
2104                 }
2105
2106                 if (read_only)
2107                         sb->s_flags |= MS_RDONLY;
2108                 else
2109                         sb->s_flags &= ~MS_RDONLY;
2110
2111                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2112                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2113                                       read_only ?  "read-only" : "read-write");
2114         }
2115         return 0;
2116 }
2117
2118 int ll_prep_inode(struct inode **inode,
2119                   struct ptlrpc_request *req,
2120                   struct super_block *sb)
2121 {
2122         struct ll_sb_info *sbi = NULL;
2123         struct lustre_md md;
2124         __u64 ibits;
2125         int rc;
2126         ENTRY;
2127
2128         LASSERT(*inode || sb);
2129         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2130         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2131                               sbi->ll_md_exp, &md);
2132         if (rc)
2133                 RETURN(rc);
2134
2135         if (*inode) {
2136                 ll_update_inode(*inode, &md);
2137         } else {
2138                 LASSERT(sb != NULL);
2139
2140                 /*
2141                  * At this point server returns to client's same fid as client
2142                  * generated for creating. So using ->fid1 is okay here.
2143                  */
2144                 LASSERT(fid_is_sane(&md.body->fid1));
2145
2146                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2147                 if (*inode == NULL || IS_ERR(*inode)) {
2148                         if (md.lsm)
2149                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2150 #ifdef CONFIG_FS_POSIX_ACL
2151                         if (md.posix_acl) {
2152                                 posix_acl_release(md.posix_acl);
2153                                 md.posix_acl = NULL;
2154                         }
2155 #endif
2156                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2157                         *inode = NULL;
2158                         CERROR("new_inode -fatal: rc %d\n", rc);
2159                         GOTO(out, rc);
2160                 }
2161         }
2162
2163         /* sanity check for LAYOUT lock. */
2164         ibits = MDS_INODELOCK_LAYOUT;
2165         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2166             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2167                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2168                         ll_get_fsname(sb, NULL, 0),
2169                         PFID(ll_inode2fid(*inode)), *inode);
2170         }
2171
2172 out:
2173         md_free_lustre_md(sbi->ll_md_exp, &md);
2174         RETURN(rc);
2175 }
2176
2177 int ll_obd_statfs(struct inode *inode, void *arg)
2178 {
2179         struct ll_sb_info *sbi = NULL;
2180         struct obd_export *exp;
2181         char *buf = NULL;
2182         struct obd_ioctl_data *data = NULL;
2183         __u32 type;
2184         __u32 flags;
2185         int len = 0, rc;
2186
2187         if (!inode || !(sbi = ll_i2sbi(inode)))
2188                 GOTO(out_statfs, rc = -EINVAL);
2189
2190         rc = obd_ioctl_getdata(&buf, &len, arg);
2191         if (rc)
2192                 GOTO(out_statfs, rc);
2193
2194         data = (void*)buf;
2195         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2196             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2197                 GOTO(out_statfs, rc = -EINVAL);
2198
2199         if (data->ioc_inllen1 != sizeof(__u32) ||
2200             data->ioc_inllen2 != sizeof(__u32) ||
2201             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2202             data->ioc_plen2 != sizeof(struct obd_uuid))
2203                 GOTO(out_statfs, rc = -EINVAL);
2204
2205         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2206         if (type & LL_STATFS_LMV)
2207                 exp = sbi->ll_md_exp;
2208         else if (type & LL_STATFS_LOV)
2209                 exp = sbi->ll_dt_exp;
2210         else
2211                 GOTO(out_statfs, rc = -ENODEV);
2212
2213         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2214         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2215         if (rc)
2216                 GOTO(out_statfs, rc);
2217 out_statfs:
2218         if (buf)
2219                 obd_ioctl_freedata(buf, len);
2220         return rc;
2221 }
2222
2223 int ll_process_config(struct lustre_cfg *lcfg)
2224 {
2225         char *ptr;
2226         void *sb;
2227         struct lprocfs_static_vars lvars;
2228         unsigned long x;
2229         int rc = 0;
2230
2231         lprocfs_llite_init_vars(&lvars);
2232
2233         /* The instance name contains the sb: lustre-client-aacfe000 */
2234         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2235         if (!ptr || !*(++ptr))
2236                 return -EINVAL;
2237         if (sscanf(ptr, "%lx", &x) != 1)
2238                 return -EINVAL;
2239         sb = (void *)x;
2240         /* This better be a real Lustre superblock! */
2241         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2242
2243         /* Note we have not called client_common_fill_super yet, so
2244            proc fns must be able to handle that! */
2245         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2246                                       lcfg, sb);
2247         if (rc > 0)
2248                 rc = 0;
2249         return(rc);
2250 }
2251
2252 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2253 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2254                                        struct inode *i1, struct inode *i2,
2255                                        const char *name, int namelen,
2256                                        int mode, __u32 opc, void *data)
2257 {
2258         LASSERT(i1 != NULL);
2259
2260         if (namelen > ll_i2sbi(i1)->ll_namelen)
2261                 return ERR_PTR(-ENAMETOOLONG);
2262
2263         if (op_data == NULL)
2264                 OBD_ALLOC_PTR(op_data);
2265
2266         if (op_data == NULL)
2267                 return ERR_PTR(-ENOMEM);
2268
2269         ll_i2gids(op_data->op_suppgids, i1, i2);
2270         op_data->op_fid1 = *ll_inode2fid(i1);
2271         op_data->op_capa1 = ll_mdscapa_get(i1);
2272
2273         if (i2) {
2274                 op_data->op_fid2 = *ll_inode2fid(i2);
2275                 op_data->op_capa2 = ll_mdscapa_get(i2);
2276         } else {
2277                 fid_zero(&op_data->op_fid2);
2278                 op_data->op_capa2 = NULL;
2279         }
2280
2281         op_data->op_name = name;
2282         op_data->op_namelen = namelen;
2283         op_data->op_mode = mode;
2284         op_data->op_mod_time = cfs_time_current_sec();
2285         op_data->op_fsuid = cfs_curproc_fsuid();
2286         op_data->op_fsgid = cfs_curproc_fsgid();
2287         op_data->op_cap = cfs_curproc_cap_pack();
2288         op_data->op_bias = MDS_CHECK_SPLIT;
2289         op_data->op_opc = opc;
2290         op_data->op_mds = 0;
2291         op_data->op_data = data;
2292
2293         /* If the file is being opened after mknod() (normally due to NFS)
2294          * try to use the default stripe data from parent directory for
2295          * allocating OST objects.  Try to pass the parent FID to MDS. */
2296         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2297             !ll_i2info(i2)->lli_has_smd) {
2298                 struct ll_inode_info *lli = ll_i2info(i2);
2299
2300                 spin_lock(&lli->lli_lock);
2301                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2302                         op_data->op_fid1 = lli->lli_pfid;
2303                 spin_unlock(&lli->lli_lock);
2304                 /** We ignore parent's capability temporary. */
2305         }
2306
2307         return op_data;
2308 }
2309
2310 void ll_finish_md_op_data(struct md_op_data *op_data)
2311 {
2312         capa_put(op_data->op_capa1);
2313         capa_put(op_data->op_capa2);
2314         OBD_FREE_PTR(op_data);
2315 }
2316
2317 #ifdef HAVE_SUPEROPS_USE_DENTRY
2318 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2319 #else
2320 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2321 #endif
2322 {
2323         struct ll_sb_info *sbi;
2324
2325 #ifdef HAVE_SUPEROPS_USE_DENTRY
2326         LASSERT((seq != NULL) && (dentry != NULL));
2327         sbi = ll_s2sbi(dentry->d_sb);
2328 #else
2329         LASSERT((seq != NULL) && (vfs != NULL));
2330         sbi = ll_s2sbi(vfs->mnt_sb);
2331 #endif
2332
2333         if (sbi->ll_flags & LL_SBI_NOLCK)
2334                 seq_puts(seq, ",nolock");
2335
2336         if (sbi->ll_flags & LL_SBI_FLOCK)
2337                 seq_puts(seq, ",flock");
2338
2339         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2340                 seq_puts(seq, ",localflock");
2341
2342         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2343                 seq_puts(seq, ",user_xattr");
2344
2345         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2346                 seq_puts(seq, ",lazystatfs");
2347
2348         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2349                 seq_puts(seq, ",user_fid2path");
2350
2351         RETURN(0);
2352 }
2353
2354 /**
2355  * Get obd name by cmd, and copy out to user space
2356  */
2357 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2358 {
2359         struct ll_sb_info *sbi = ll_i2sbi(inode);
2360         struct obd_device *obd;
2361         ENTRY;
2362
2363         if (cmd == OBD_IOC_GETDTNAME)
2364                 obd = class_exp2obd(sbi->ll_dt_exp);
2365         else if (cmd == OBD_IOC_GETMDNAME)
2366                 obd = class_exp2obd(sbi->ll_md_exp);
2367         else
2368                 RETURN(-EINVAL);
2369
2370         if (!obd)
2371                 RETURN(-ENOENT);
2372
2373         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2374                              strlen(obd->obd_name) + 1))
2375                 RETURN(-EFAULT);
2376
2377         RETURN(0);
2378 }
2379
2380 /**
2381  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2382  * fsname will be returned in this buffer; otherwise, a static buffer will be
2383  * used to store the fsname and returned to caller.
2384  */
2385 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2386 {
2387         static char fsname_static[MTI_NAME_MAXLEN];
2388         struct lustre_sb_info *lsi = s2lsi(sb);
2389         char *ptr;
2390         int len;
2391
2392         if (buf == NULL) {
2393                 /* this means the caller wants to use static buffer
2394                  * and it doesn't care about race. Usually this is
2395                  * in error reporting path */
2396                 buf = fsname_static;
2397                 buflen = sizeof(fsname_static);
2398         }
2399
2400         len = strlen(lsi->lsi_lmd->lmd_profile);
2401         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2402         if (ptr && (strcmp(ptr, "-client") == 0))
2403                 len -= 7;
2404
2405         if (unlikely(len >= buflen))
2406                 len = buflen - 1;
2407         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2408         buf[len] = '\0';
2409
2410         return buf;
2411 }
2412
2413 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2414 {
2415         char *path = NULL;
2416
2417 #ifdef HAVE_FS_STRUCT_USE_PATH
2418         struct path p;
2419
2420         p.dentry = dentry;
2421         p.mnt = current->fs->root.mnt;
2422         path_get(&p);
2423         path = d_path(&p, buf, bufsize);
2424         path_put(&p);
2425 #else
2426         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2427 #endif
2428
2429         return path;
2430 }
2431
2432 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2433 {
2434         char *buf, *path = NULL;
2435         struct dentry *dentry = NULL;
2436         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2437
2438         buf = (char *)__get_free_page(GFP_KERNEL);
2439         if (buf != NULL) {
2440                 dentry = d_find_alias(page->mapping->host);
2441                 if (dentry != NULL)
2442                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2443         }
2444
2445         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2446               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2447               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2448               PFID(&obj->cob_header.coh_lu.loh_fid),
2449               (path && !IS_ERR(path)) ? path : "", ioret);
2450
2451         if (dentry != NULL)
2452                 dput(dentry);
2453
2454         if (buf != NULL)
2455                 free_page((unsigned long)buf);
2456 }