lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67 #else
  68 extern struct address_space_operations_ext ll_aops;
  69 extern struct address_space_operations_ext ll_dir_aops;
  70 #endif
  71
  72 #ifndef log2
  73 #define log2(n) cfs_ffz(~(n))
  74 #endif
  75
  76 static struct ll_sb_info *ll_init_sbi(void)
  77 {
  78         struct ll_sb_info *sbi = NULL;
  79         unsigned long pages;
  80         struct sysinfo si;
  81         class_uuid_t uuid;
  82         int i;
  83         ENTRY;
  84
  85         OBD_ALLOC(sbi, sizeof(*sbi));
  86         if (!sbi)
  87                 RETURN(NULL);
  88
  89         cfs_spin_lock_init(&sbi->ll_lock);
  90         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  91         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  92         cfs_spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  98 #ifdef HAVE_BGL_SUPPORT
  99                 sbi->ll_async_page_max = pages / 4;
 100 #else
 101                 sbi->ll_async_page_max = pages / 2;
 102 #endif
 103         } else {
 104                 sbi->ll_async_page_max = (pages / 4) * 3;
 105         }
 106
 107         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 108                                            SBI_DEFAULT_READAHEAD_MAX);
 109         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 110         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 111                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 112         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 113         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 114
 115         ll_generate_random_uuid(uuid);
 116         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 117         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 118
 119         cfs_spin_lock(&ll_sb_lock);
 120         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 121         cfs_spin_unlock(&ll_sb_lock);
 122
 123 #ifdef ENABLE_CHECKSUM
 124         sbi->ll_flags |= LL_SBI_CHECKSUM;
 125 #endif
 126
 127 #ifdef HAVE_LRU_RESIZE_SUPPORT
 128         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 129 #endif
 130
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 133                                    pp_r_hist.oh_lock);
 134                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 135                                    pp_w_hist.oh_lock);
 136         }
 137
 138         /* metadata statahead is enabled by default */
 139         sbi->ll_sa_max = LL_SA_RPC_DEF;
 140         atomic_set(&sbi->ll_sa_total, 0);
 141         atomic_set(&sbi->ll_sa_wrong, 0);
 142
 143         RETURN(sbi);
 144 }
 145
 146 void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 cfs_spin_lock(&ll_sb_lock);
 153                 cfs_list_del(&sbi->ll_list);
 154                 cfs_spin_unlock(&ll_sb_lock);
 155                 OBD_FREE(sbi, sizeof(*sbi));
 156         }
 157         EXIT;
 158 }
 159
 160 static struct dentry_operations ll_d_root_ops = {
 161         .d_compare = ll_dcompare,
 162         .d_revalidate = ll_revalidate_nd,
 163 };
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_capa *oc = NULL;
 171         struct obd_statfs osfs;
 172         struct ptlrpc_request *request = NULL;
 173         struct obd_connect_data *data = NULL;
 174         struct obd_uuid *uuid;
 175         struct md_op_data *op_data;
 176         struct lustre_md lmd;
 177         obd_valid valid;
 178         int size, err, checksum;
 179         ENTRY;
 180
 181         obd = class_name2obd(md);
 182         if (!obd) {
 183                 CERROR("MD %s: not setup or attached\n", md);
 184                 RETURN(-EINVAL);
 185         }
 186
 187         OBD_ALLOC_PTR(data);
 188         if (data == NULL)
 189                 RETURN(-ENOMEM);
 190
 191         if (proc_lustre_fs_root) {
 192                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 193                                                   dt, md);
 194                 if (err < 0)
 195                         CERROR("could not register mount in /proc/fs/lustre\n");
 196         }
 197
 198         /* indicate the features supported by this client */
 199         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 200                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 201                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 202                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 203                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 204                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 205                                   OBD_CONNECT_VBR      | OBD_CONNECT_FULL20 |
 206                                   OBD_CONNECT_64BITHASH;
 207
 208         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 209                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 210
 211 #ifdef HAVE_LRU_RESIZE_SUPPORT
 212         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 213                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 214 #endif
 215 #ifdef CONFIG_FS_POSIX_ACL
 216         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 217 #endif
 218         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 219         data->ocd_version = LUSTRE_VERSION_CODE;
 220
 221         if (sb->s_flags & MS_RDONLY)
 222                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 223         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 224                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 225
 226 #ifdef HAVE_MS_FLOCK_LOCK
 227         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 228         sb->s_flags |= MS_FLOCK_LOCK;
 229 #endif
 230 #ifdef MS_HAS_NEW_AOPS
 231         sb->s_flags |= MS_HAS_NEW_AOPS;
 232 #endif
 233
 234         if (sbi->ll_flags & LL_SBI_FLOCK)
 235                 sbi->ll_fop = &ll_file_operations_flock;
 236         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 237                 sbi->ll_fop = &ll_file_operations;
 238         else
 239                 sbi->ll_fop = &ll_file_operations_noflock;
 240
 241         /* real client */
 242         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 243         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 244                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 245
 246         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 247         if (err == -EBUSY) {
 248                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 249                                    "recovery, of which this client is not a "
 250                                    "part. Please wait for recovery to complete,"
 251                                    " abort, or time out.\n", md);
 252                 GOTO(out, err);
 253         } else if (err) {
 254                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 255                 GOTO(out, err);
 256         }
 257
 258         err = obd_fid_init(sbi->ll_md_exp);
 259         if (err) {
 260                 CERROR("Can't init metadata layer FID infrastructure, "
 261                        "rc %d\n", err);
 262                 GOTO(out_md, err);
 263         }
 264
 265         err = obd_statfs(obd, &osfs,
 266                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 267         if (err)
 268                 GOTO(out_md_fid, err);
 269
 270         size = sizeof(*data);
 271         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 272                            KEY_CONN_DATA,  &size, data, NULL);
 273         if (err) {
 274                 CERROR("Get connect data failed: %d \n", err);
 275                 GOTO(out_md, err);
 276         }
 277
 278         LASSERT(osfs.os_bsize);
 279         sb->s_blocksize = osfs.os_bsize;
 280         sb->s_blocksize_bits = log2(osfs.os_bsize);
 281         sb->s_magic = LL_SUPER_MAGIC;
 282
 283         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 284          *         retval = in_file->f_op->sendfile(...);
 285          *         if (*ppos > max)
 286          *                 retval = -EOVERFLOW;
 287          *
 288          * it will check if *ppos is greater than max. However, max equals to
 289          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 290          * has been defined as a signed long long integer in linux kernel. */
 291 #if BITS_PER_LONG == 64
 292         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 293 #else
 294         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 295 #endif
 296         sbi->ll_namelen = osfs.os_namelen;
 297         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 298
 299         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 300             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 301                 LCONSOLE_INFO("Disabling user_xattr feature because "
 302                               "it is not supported on the server\n");
 303                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 304         }
 305
 306         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 307 #ifdef MS_POSIXACL
 308                 sb->s_flags |= MS_POSIXACL;
 309 #endif
 310                 sbi->ll_flags |= LL_SBI_ACL;
 311         } else {
 312                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 313 #ifdef MS_POSIXACL
 314                 sb->s_flags &= ~MS_POSIXACL;
 315 #endif
 316                 sbi->ll_flags &= ~LL_SBI_ACL;
 317         }
 318
 319         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 320                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 321                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 322                         LCONSOLE_INFO("client is set as remote by default.\n");
 323                 }
 324         } else {
 325                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 326                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 327                         LCONSOLE_INFO("client claims to be remote, but server "
 328                                       "rejected, forced to be local.\n");
 329                 }
 330         }
 331
 332         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 333                 LCONSOLE_INFO("client enabled MDS capability!\n");
 334                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 335         }
 336
 337         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 338                 LCONSOLE_INFO("client enabled OSS capability!\n");
 339                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 343                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 344
 345         obd = class_name2obd(dt);
 346         if (!obd) {
 347                 CERROR("DT %s: not setup or attached\n", dt);
 348                 GOTO(out_md_fid, err = -ENODEV);
 349         }
 350
 351         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 352                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 353                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 354                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 355                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 356                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 357                                   OBD_CONNECT_FULL20;
 358
 359         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 360                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 361
 362         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 363                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 364                  * disabled by default, because it can still be enabled on the
 365                  * fly via /proc. As a consequence, we still need to come to an
 366                  * agreement on the supported algorithms at connect time */
 367                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 368
 369                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 370                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 371                 else
 372                         /* send the list of supported checksum types */
 373                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 374         }
 375
 376 #ifdef HAVE_LRU_RESIZE_SUPPORT
 377         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 378 #endif
 379         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 380                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 381
 382         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 383                "ocd_grant: %d\n", data->ocd_connect_flags,
 384                data->ocd_version, data->ocd_grant);
 385
 386         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 387         obd->obd_upcall.onu_upcall = cl_ocd_update;
 388         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 389
 390         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 391         if (err == -EBUSY) {
 392                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 393                                    "recovery, of which this client is not a "
 394                                    "part.  Please wait for recovery to "
 395                                    "complete, abort, or time out.\n", dt);
 396                 GOTO(out_md_fid, err);
 397         } else if (err) {
 398                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 399                 GOTO(out_md_fid, err);
 400         }
 401
 402         err = obd_fid_init(sbi->ll_dt_exp);
 403         if (err) {
 404                 CERROR("Can't init data layer FID infrastructure, "
 405                        "rc %d\n", err);
 406                 GOTO(out_dt, err);
 407         }
 408
 409         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 410         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 411         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 412         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 413         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 414
 415         fid_zero(&sbi->ll_root_fid);
 416         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 417         if (err) {
 418                 CERROR("cannot mds_connect: rc = %d\n", err);
 419                 GOTO(out_lock_cn_cb, err);
 420         }
 421         if (!fid_is_sane(&sbi->ll_root_fid)) {
 422                 CERROR("Invalid root fid during mount\n");
 423                 GOTO(out_lock_cn_cb, err = -EINVAL);
 424         }
 425         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 426
 427         sb->s_op = &lustre_super_operations;
 428 #if THREAD_SIZE >= 8192
 429         sb->s_export_op = &lustre_export_operations;
 430 #endif
 431
 432         /* make root inode
 433          * XXX: move this to after cbd setup? */
 434         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 435         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 436                 valid |= OBD_MD_FLRMTPERM;
 437         else if (sbi->ll_flags & LL_SBI_ACL)
 438                 valid |= OBD_MD_FLACL;
 439
 440         OBD_ALLOC_PTR(op_data);
 441         if (op_data == NULL)
 442                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 443
 444         op_data->op_fid1 = sbi->ll_root_fid;
 445         op_data->op_mode = 0;
 446         op_data->op_capa1 = oc;
 447         op_data->op_valid = valid;
 448
 449         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 450         if (oc)
 451                 capa_put(oc);
 452         OBD_FREE_PTR(op_data);
 453         if (err) {
 454                 CERROR("md_getattr failed for root: rc = %d\n", err);
 455                 GOTO(out_lock_cn_cb, err);
 456         }
 457         memset(&lmd, 0, sizeof(lmd));
 458         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 459                                sbi->ll_md_exp, &lmd);
 460         if (err) {
 461                 CERROR("failed to understand root inode md: rc = %d\n", err);
 462                 ptlrpc_req_finished (request);
 463                 GOTO(out_lock_cn_cb, err);
 464         }
 465
 466         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 467         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 468         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 469         ptlrpc_req_finished(request);
 470
 471         if (root == NULL || IS_ERR(root)) {
 472                 if (lmd.lsm)
 473                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 474 #ifdef CONFIG_FS_POSIX_ACL
 475                 if (lmd.posix_acl) {
 476                         posix_acl_release(lmd.posix_acl);
 477                         lmd.posix_acl = NULL;
 478                 }
 479 #endif
 480                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 481                 root = NULL;
 482                 CERROR("lustre_lite: bad iget4 for root\n");
 483                 GOTO(out_root, err);
 484         }
 485
 486         err = ll_close_thread_start(&sbi->ll_lcq);
 487         if (err) {
 488                 CERROR("cannot start close thread: rc %d\n", err);
 489                 GOTO(out_root, err);
 490         }
 491
 492 #ifdef CONFIG_FS_POSIX_ACL
 493         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 494                 rct_init(&sbi->ll_rct);
 495                 et_init(&sbi->ll_et);
 496         }
 497 #endif
 498
 499         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 500         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 501                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 502                                  NULL);
 503         cl_sb_init(sb);
 504
 505         sb->s_root = d_alloc_root(root);
 506         if (data != NULL)
 507                 OBD_FREE(data, sizeof(*data));
 508
 509         sb->s_root->d_op = &ll_d_root_ops;
 510
 511         sbi->ll_sdev_orig = sb->s_dev;
 512
 513         /* We set sb->s_dev equal on all lustre clients in order to support
 514          * NFS export clustering.  NFSD requires that the FSID be the same
 515          * on all clients. */
 516         /* s_dev is also used in lt_compare() to compare two fs, but that is
 517          * only a node-local comparison. */
 518         uuid = obd_get_uuid(sbi->ll_md_exp);
 519         if (uuid != NULL)
 520                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 521
 522         RETURN(err);
 523 out_root:
 524         if (root)
 525                 iput(root);
 526 out_lock_cn_cb:
 527         obd_fid_fini(sbi->ll_dt_exp);
 528 out_dt:
 529         obd_disconnect(sbi->ll_dt_exp);
 530         sbi->ll_dt_exp = NULL;
 531 out_md_fid:
 532         obd_fid_fini(sbi->ll_md_exp);
 533 out_md:
 534         obd_disconnect(sbi->ll_md_exp);
 535         sbi->ll_md_exp = NULL;
 536 out:
 537         if (data != NULL)
 538                 OBD_FREE_PTR(data);
 539         lprocfs_unregister_mountpoint(sbi);
 540         return err;
 541 }
 542
 543 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 544 {
 545         int size, rc;
 546
 547         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 548         size = sizeof(int);
 549         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 550                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 551         if (rc)
 552                 CERROR("Get max mdsize error rc %d \n", rc);
 553
 554         RETURN(rc);
 555 }
 556
 557 void ll_dump_inode(struct inode *inode)
 558 {
 559         struct list_head *tmp;
 560         int dentry_count = 0;
 561
 562         LASSERT(inode != NULL);
 563
 564         list_for_each(tmp, &inode->i_dentry)
 565                 dentry_count++;
 566
 567         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 568                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 569                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 570 }
 571
 572 void lustre_dump_dentry(struct dentry *dentry, int recur)
 573 {
 574         struct list_head *tmp;
 575         int subdirs = 0;
 576
 577         LASSERT(dentry != NULL);
 578
 579         list_for_each(tmp, &dentry->d_subdirs)
 580                 subdirs++;
 581
 582         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 583                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 584                dentry->d_name.len, dentry->d_name.name,
 585                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 586                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 587                dentry->d_flags, dentry->d_fsdata, subdirs);
 588         if (dentry->d_inode != NULL)
 589                 ll_dump_inode(dentry->d_inode);
 590
 591         if (recur == 0)
 592                 return;
 593
 594        list_for_each(tmp, &dentry->d_subdirs) {
 595                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 596                 lustre_dump_dentry(d, recur - 1);
 597         }
 598 }
 599
 600 void client_common_put_super(struct super_block *sb)
 601 {
 602         struct ll_sb_info *sbi = ll_s2sbi(sb);
 603         ENTRY;
 604
 605 #ifdef CONFIG_FS_POSIX_ACL
 606         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 607                 et_fini(&sbi->ll_et);
 608                 rct_fini(&sbi->ll_rct);
 609         }
 610 #endif
 611
 612         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 613
 614         ll_close_thread_shutdown(sbi->ll_lcq);
 615
 616         cl_sb_fini(sb);
 617
 618         cfs_list_del(&sbi->ll_conn_chain);
 619
 620         obd_fid_fini(sbi->ll_dt_exp);
 621         obd_disconnect(sbi->ll_dt_exp);
 622         sbi->ll_dt_exp = NULL;
 623
 624         lprocfs_unregister_mountpoint(sbi);
 625
 626         obd_fid_fini(sbi->ll_md_exp);
 627         obd_disconnect(sbi->ll_md_exp);
 628         sbi->ll_md_exp = NULL;
 629
 630         EXIT;
 631 }
 632
 633 void ll_kill_super(struct super_block *sb)
 634 {
 635         struct ll_sb_info *sbi;
 636
 637         ENTRY;
 638
 639         /* not init sb ?*/
 640         if (!(sb->s_flags & MS_ACTIVE))
 641                 return;
 642
 643         sbi = ll_s2sbi(sb);
 644         /* we need restore s_dev from changed for clustred NFS before put_super
 645          * because new kernels have cached s_dev and change sb->s_dev in
 646          * put_super not affected real removing devices */
 647         if (sbi)
 648                 sb->s_dev = sbi->ll_sdev_orig;
 649         EXIT;
 650 }
 651
 652 char *ll_read_opt(const char *opt, char *data)
 653 {
 654         char *value;
 655         char *retval;
 656         ENTRY;
 657
 658         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 659         if (strncmp(opt, data, strlen(opt)))
 660                 RETURN(NULL);
 661         if ((value = strchr(data, '=')) == NULL)
 662                 RETURN(NULL);
 663
 664         value++;
 665         OBD_ALLOC(retval, strlen(value) + 1);
 666         if (!retval) {
 667                 CERROR("out of memory!\n");
 668                 RETURN(NULL);
 669         }
 670
 671         memcpy(retval, value, strlen(value)+1);
 672         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 673         RETURN(retval);
 674 }
 675
 676 static inline int ll_set_opt(const char *opt, char *data, int fl)
 677 {
 678         if (strncmp(opt, data, strlen(opt)) != 0)
 679                 return(0);
 680         else
 681                 return(fl);
 682 }
 683
 684 /* non-client-specific mount options are parsed in lmd_parse */
 685 static int ll_options(char *options, int *flags)
 686 {
 687         int tmp;
 688         char *s1 = options, *s2;
 689         ENTRY;
 690
 691         if (!options)
 692                 RETURN(0);
 693
 694         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 695
 696         while (*s1) {
 697                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 698                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 699                 if (tmp) {
 700                         *flags |= tmp;
 701                         goto next;
 702                 }
 703                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 704                 if (tmp) {
 705                         *flags |= tmp;
 706                         goto next;
 707                 }
 708                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 709                 if (tmp) {
 710                         *flags |= tmp;
 711                         goto next;
 712                 }
 713                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 714                 if (tmp) {
 715                         *flags &= ~tmp;
 716                         goto next;
 717                 }
 718                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 719                 if (tmp) {
 720                         *flags |= tmp;
 721                         goto next;
 722                 }
 723                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 724                 if (tmp) {
 725                         *flags &= ~tmp;
 726                         goto next;
 727                 }
 728 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 729                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 730                 if (tmp) {
 731                         /* Ignore deprecated mount option.  The client will
 732                          * always try to mount with ACL support, whether this
 733                          * is used depends on whether server supports it. */
 734                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 735                                                   "mount option 'acl'.\n");
 736                         goto next;
 737                 }
 738                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 739                 if (tmp) {
 740                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 741                                                   "mount option 'noacl'.\n");
 742                         goto next;
 743                 }
 744 #else
 745 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 746 #endif
 747                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 748                 if (tmp) {
 749                         *flags |= tmp;
 750                         goto next;
 751                 }
 752
 753                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 754                 if (tmp) {
 755                         *flags |= tmp;
 756                         goto next;
 757                 }
 758                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 759                 if (tmp) {
 760                         *flags &= ~tmp;
 761                         goto next;
 762                 }
 763                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 764                 if (tmp) {
 765                         *flags |= tmp;
 766                         goto next;
 767                 }
 768                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 769                 if (tmp) {
 770                         *flags &= ~tmp;
 771                         goto next;
 772                 }
 773                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 774                 if (tmp) {
 775                         *flags |= tmp;
 776                         goto next;
 777                 }
 778                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 779                 if (tmp) {
 780                         *flags &= ~tmp;
 781                         goto next;
 782                 }
 783                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 784                 if (tmp) {
 785                         *flags |= tmp;
 786                         goto next;
 787                 }
 788                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 789                 if (tmp) {
 790                         *flags |= tmp;
 791                         goto next;
 792                 }
 793
 794                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 795                                    s1);
 796                 RETURN(-EINVAL);
 797
 798 next:
 799                 /* Find next opt */
 800                 s2 = strchr(s1, ',');
 801                 if (s2 == NULL)
 802                         break;
 803                 s1 = s2 + 1;
 804         }
 805         RETURN(0);
 806 }
 807
 808 void ll_lli_init(struct ll_inode_info *lli)
 809 {
 810         lli->lli_inode_magic = LLI_INODE_MAGIC;
 811         cfs_sema_init(&lli->lli_size_sem, 1);
 812         cfs_sema_init(&lli->lli_write_sem, 1);
 813         cfs_init_rwsem(&lli->lli_trunc_sem);
 814         lli->lli_flags = 0;
 815         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 816         cfs_spin_lock_init(&lli->lli_lock);
 817         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 818         lli->lli_inode_magic = LLI_INODE_MAGIC;
 819         cfs_sema_init(&lli->lli_och_sem, 1);
 820         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 821         lli->lli_mds_exec_och = NULL;
 822         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 823         lli->lli_open_fd_exec_count = 0;
 824         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 825         lli->lli_remote_perms = NULL;
 826         lli->lli_rmtperm_utime = 0;
 827         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 828         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 829         cfs_spin_lock_init(&lli->lli_sa_lock);
 830         cfs_sema_init(&lli->lli_readdir_sem, 1);
 831 }
 832
 833 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 834 {
 835 #ifdef HAVE_BDI_REGISTER
 836         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 837
 838         bdi->name = "lustre";
 839         return bdi_register(bdi, NULL, "lustre-%d",
 840                             atomic_inc_return(&ll_bdi_num));
 841 #else
 842         return 0;
 843 #endif
 844 }
 845
 846 int ll_fill_super(struct super_block *sb)
 847 {
 848         struct lustre_profile *lprof;
 849         struct lustre_sb_info *lsi = s2lsi(sb);
 850         struct ll_sb_info *sbi;
 851         char  *dt = NULL, *md = NULL;
 852         char  *profilenm = get_profile_name(sb);
 853         struct config_llog_instance cfg = {0, };
 854         char   ll_instance[sizeof(sb) * 2 + 1];
 855         int    err;
 856         ENTRY;
 857
 858         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 859
 860         cfs_module_get();
 861
 862         /* client additional sb info */
 863         lsi->lsi_llsbi = sbi = ll_init_sbi();
 864         if (!sbi) {
 865                 cfs_module_put(THIS_MODULE);
 866                 RETURN(-ENOMEM);
 867         }
 868
 869         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 870         if (err)
 871                 GOTO(out_free, err);
 872
 873         err = ll_bdi_init(&lsi->lsi_bdi);
 874         if (err)
 875                 GOTO(out_free, err);
 876         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 877         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 878         err = ll_bdi_register(&lsi->lsi_bdi);
 879         if (err)
 880                 GOTO(out_free, err);
 881
 882 #ifdef HAVE_SB_BDI
 883         sb->s_bdi = &lsi->lsi_bdi;
 884 #endif
 885
 886         /* Generate a string unique to this super, in case some joker tries
 887            to mount the same fs at two mount points.
 888            Use the address of the super itself.*/
 889         sprintf(ll_instance, "%p", sb);
 890         cfg.cfg_instance = ll_instance;
 891         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 892
 893         /* set up client obds */
 894         err = lustre_process_log(sb, profilenm, &cfg);
 895         if (err < 0) {
 896                 CERROR("Unable to process log: %d\n", err);
 897                 GOTO(out_free, err);
 898         }
 899
 900         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 901         lprof = class_get_profile(profilenm);
 902         if (lprof == NULL) {
 903                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 904                                    " read from the MGS.  Does that filesystem "
 905                                    "exist?\n", profilenm);
 906                 GOTO(out_free, err = -EINVAL);
 907         }
 908         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 909                lprof->lp_md, lprof->lp_dt);
 910
 911         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 912                   strlen(ll_instance) + 2);
 913         if (!dt)
 914                 GOTO(out_free, err = -ENOMEM);
 915         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 916
 917         OBD_ALLOC(md, strlen(lprof->lp_md) +
 918                   strlen(ll_instance) + 2);
 919         if (!md)
 920                 GOTO(out_free, err = -ENOMEM);
 921         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 922
 923         /* connections, registrations, sb setup */
 924         err = client_common_fill_super(sb, md, dt);
 925
 926 out_free:
 927         if (md)
 928                 OBD_FREE(md, strlen(md) + 1);
 929         if (dt)
 930                 OBD_FREE(dt, strlen(dt) + 1);
 931         if (err)
 932                 ll_put_super(sb);
 933         else
 934                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 935
 936         RETURN(err);
 937 } /* ll_fill_super */
 938
 939
 940 void lu_context_keys_dump(void);
 941
 942 void ll_put_super(struct super_block *sb)
 943 {
 944         struct config_llog_instance cfg;
 945         char   ll_instance[sizeof(sb) * 2 + 1];
 946         struct obd_device *obd;
 947         struct lustre_sb_info *lsi = s2lsi(sb);
 948         struct ll_sb_info *sbi = ll_s2sbi(sb);
 949         char *profilenm = get_profile_name(sb);
 950         int force = 1, next;
 951         ENTRY;
 952
 953         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 954
 955         ll_print_capa_stat(sbi);
 956
 957         sprintf(ll_instance, "%p", sb);
 958         cfg.cfg_instance = ll_instance;
 959         lustre_end_log(sb, NULL, &cfg);
 960
 961         if (sbi->ll_md_exp) {
 962                 obd = class_exp2obd(sbi->ll_md_exp);
 963                 if (obd)
 964                         force = obd->obd_force;
 965         }
 966
 967         /* We need to set force before the lov_disconnect in
 968            lustre_common_put_super, since l_d cleans up osc's as well. */
 969         if (force) {
 970                 next = 0;
 971                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 972                                                      &next)) != NULL) {
 973                         obd->obd_force = force;
 974                 }
 975         }
 976
 977         if (sbi->ll_lcq) {
 978                 /* Only if client_common_fill_super succeeded */
 979                 client_common_put_super(sb);
 980         }
 981
 982         next = 0;
 983         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 984                 class_manual_cleanup(obd);
 985         }
 986
 987         if (profilenm)
 988                 class_del_profile(profilenm);
 989
 990         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
 991                 ll_bdi_destroy(&lsi->lsi_bdi);
 992                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
 993         }
 994
 995         ll_free_sbi(sb);
 996         lsi->lsi_llsbi = NULL;
 997
 998         lustre_common_put_super(sb);
 999
1000         cl_env_cache_purge(~0);
1001
1002         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1003
1004         cfs_module_put(THIS_MODULE);
1005
1006         EXIT;
1007 } /* client_put_super */
1008
1009 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1010 {
1011         struct inode *inode = NULL;
1012         /* NOTE: we depend on atomic igrab() -bzzz */
1013         lock_res_and_lock(lock);
1014         if (lock->l_ast_data) {
1015                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1016                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1017                         inode = igrab(lock->l_ast_data);
1018                 } else {
1019                         inode = lock->l_ast_data;
1020                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1021                                          D_WARNING, lock, "l_ast_data %p is "
1022                                          "bogus: magic %08x", lock->l_ast_data,
1023                                          lli->lli_inode_magic);
1024                         inode = NULL;
1025                 }
1026         }
1027         unlock_res_and_lock(lock);
1028         return inode;
1029 }
1030
1031 static int null_if_equal(struct ldlm_lock *lock, void *data)
1032 {
1033         if (data == lock->l_ast_data) {
1034                 lock->l_ast_data = NULL;
1035
1036                 if (lock->l_req_mode != lock->l_granted_mode)
1037                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1038         }
1039
1040         return LDLM_ITER_CONTINUE;
1041 }
1042
1043 void ll_clear_inode(struct inode *inode)
1044 {
1045         struct ll_inode_info *lli = ll_i2info(inode);
1046         struct ll_sb_info *sbi = ll_i2sbi(inode);
1047         ENTRY;
1048
1049         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1050                inode->i_generation, inode);
1051
1052         if (S_ISDIR(inode->i_mode)) {
1053                 /* these should have been cleared in ll_file_release */
1054                 LASSERT(lli->lli_sai == NULL);
1055                 LASSERT(lli->lli_opendir_key == NULL);
1056                 LASSERT(lli->lli_opendir_pid == 0);
1057         }
1058
1059         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1060         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1061                          null_if_equal, inode);
1062
1063         LASSERT(!lli->lli_open_fd_write_count);
1064         LASSERT(!lli->lli_open_fd_read_count);
1065         LASSERT(!lli->lli_open_fd_exec_count);
1066
1067         if (lli->lli_mds_write_och)
1068                 ll_md_real_close(inode, FMODE_WRITE);
1069         if (lli->lli_mds_exec_och)
1070                 ll_md_real_close(inode, FMODE_EXEC);
1071         if (lli->lli_mds_read_och)
1072                 ll_md_real_close(inode, FMODE_READ);
1073
1074         if (lli->lli_symlink_name) {
1075                 OBD_FREE(lli->lli_symlink_name,
1076                          strlen(lli->lli_symlink_name) + 1);
1077                 lli->lli_symlink_name = NULL;
1078         }
1079
1080         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1081                 LASSERT(lli->lli_posix_acl == NULL);
1082                 if (lli->lli_remote_perms) {
1083                         free_rmtperm_hash(lli->lli_remote_perms);
1084                         lli->lli_remote_perms = NULL;
1085                 }
1086         }
1087 #ifdef CONFIG_FS_POSIX_ACL
1088         else if (lli->lli_posix_acl) {
1089                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1090                 LASSERT(lli->lli_remote_perms == NULL);
1091                 posix_acl_release(lli->lli_posix_acl);
1092                 lli->lli_posix_acl = NULL;
1093         }
1094 #endif
1095         lli->lli_inode_magic = LLI_INODE_DEAD;
1096
1097         ll_clear_inode_capas(inode);
1098         /*
1099          * XXX This has to be done before lsm is freed below, because
1100          * cl_object still uses inode lsm.
1101          */
1102         cl_inode_fini(inode);
1103
1104         if (lli->lli_smd) {
1105                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1106                 lli->lli_smd = NULL;
1107         }
1108
1109
1110         EXIT;
1111 }
1112
1113 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1114                   struct md_open_data **mod)
1115 {
1116         struct lustre_md md;
1117         struct ll_sb_info *sbi = ll_i2sbi(inode);
1118         struct ptlrpc_request *request = NULL;
1119         int rc;
1120         ENTRY;
1121
1122         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1123                                      LUSTRE_OPC_ANY, NULL);
1124         if (IS_ERR(op_data))
1125                 RETURN(PTR_ERR(op_data));
1126
1127         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1128                         &request, mod);
1129         if (rc) {
1130                 ptlrpc_req_finished(request);
1131                 if (rc == -ENOENT) {
1132                         inode->i_nlink = 0;
1133                         /* Unlinked special device node? Or just a race?
1134                          * Pretend we done everything. */
1135                         if (!S_ISREG(inode->i_mode) &&
1136                             !S_ISDIR(inode->i_mode))
1137                                 rc = inode_setattr(inode, &op_data->op_attr);
1138                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1139                         CERROR("md_setattr fails: rc = %d\n", rc);
1140                 }
1141                 RETURN(rc);
1142         }
1143
1144         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1145                               sbi->ll_md_exp, &md);
1146         if (rc) {
1147                 ptlrpc_req_finished(request);
1148                 RETURN(rc);
1149         }
1150
1151         /* We call inode_setattr to adjust timestamps.
1152          * If there is at least some data in file, we cleared ATTR_SIZE
1153          * above to avoid invoking vmtruncate, otherwise it is important
1154          * to call vmtruncate in inode_setattr to update inode->i_size
1155          * (bug 6196) */
1156         rc = inode_setattr(inode, &op_data->op_attr);
1157
1158         /* Extract epoch data if obtained. */
1159         op_data->op_handle = md.body->handle;
1160         op_data->op_ioepoch = md.body->ioepoch;
1161
1162         ll_update_inode(inode, &md);
1163         ptlrpc_req_finished(request);
1164
1165         RETURN(rc);
1166 }
1167
1168 /* Close IO epoch and send Size-on-MDS attribute update. */
1169 static int ll_setattr_done_writing(struct inode *inode,
1170                                    struct md_op_data *op_data,
1171                                    struct md_open_data *mod)
1172 {
1173         struct ll_inode_info *lli = ll_i2info(inode);
1174         int rc = 0;
1175         ENTRY;
1176
1177         LASSERT(op_data != NULL);
1178         if (!S_ISREG(inode->i_mode))
1179                 RETURN(0);
1180
1181         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1182                op_data->op_ioepoch, PFID(&lli->lli_fid));
1183
1184         op_data->op_flags = MF_EPOCH_CLOSE;
1185         ll_done_writing_attr(inode, op_data);
1186         ll_pack_inode2opdata(inode, op_data, NULL);
1187
1188         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1189         if (rc == -EAGAIN) {
1190                 /* MDS has instructed us to obtain Size-on-MDS attribute
1191                  * from OSTs and send setattr to back to MDS. */
1192                 rc = ll_som_update(inode, op_data);
1193         } else if (rc) {
1194                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1195                        inode->i_ino, rc);
1196         }
1197         RETURN(rc);
1198 }
1199
1200 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1201 {
1202         struct obd_capa *capa;
1203         int rc;
1204
1205         if (attr->ia_valid & ATTR_SIZE)
1206                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1207         else
1208                 capa = ll_mdscapa_get(inode);
1209
1210         rc = cl_setattr_ost(inode, attr, capa);
1211
1212         if (attr->ia_valid & ATTR_SIZE)
1213                 ll_truncate_free_capa(capa);
1214         else
1215                 capa_put(capa);
1216
1217         return rc;
1218 }
1219
1220 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1221  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1222  * keep these values until such a time that objects are allocated for it.
1223  * We do the MDS operations first, as it is checking permissions for us.
1224  * We don't to the MDS RPC if there is nothing that we want to store there,
1225  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1226  * going to do an RPC anyways.
1227  *
1228  * If we are doing a truncate, we will send the mtime and ctime updates
1229  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1230  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1231  * at the same time.
1232  */
1233 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1234 {
1235         struct ll_inode_info *lli = ll_i2info(inode);
1236         struct lov_stripe_md *lsm = lli->lli_smd;
1237         struct md_op_data *op_data = NULL;
1238         struct md_open_data *mod = NULL;
1239         int ia_valid = attr->ia_valid;
1240         int rc = 0, rc1 = 0;
1241         ENTRY;
1242
1243         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1244                attr->ia_valid);
1245         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1246
1247         if (ia_valid & ATTR_SIZE) {
1248                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1249                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1250                                attr->ia_size, ll_file_maxbytes(inode));
1251                         RETURN(-EFBIG);
1252                 }
1253
1254                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1255         }
1256
1257         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1258         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1259                 if (cfs_curproc_fsuid() != inode->i_uid &&
1260                     !cfs_capable(CFS_CAP_FOWNER))
1261                         RETURN(-EPERM);
1262         }
1263
1264         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1265         if (attr->ia_valid & ATTR_CTIME) {
1266                 attr->ia_ctime = CFS_CURRENT_TIME;
1267                 attr->ia_valid |= ATTR_CTIME_SET;
1268         }
1269         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1270                 attr->ia_atime = CFS_CURRENT_TIME;
1271                 attr->ia_valid |= ATTR_ATIME_SET;
1272         }
1273         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1274                 attr->ia_mtime = CFS_CURRENT_TIME;
1275                 attr->ia_valid |= ATTR_MTIME_SET;
1276         }
1277
1278         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1279                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1280                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1281                        cfs_time_current_sec());
1282
1283         /* NB: ATTR_SIZE will only be set after this point if the size
1284          * resides on the MDS, ie, this file has no objects. */
1285         if (lsm)
1286                 attr->ia_valid &= ~ATTR_SIZE;
1287
1288         /* We always do an MDS RPC, even if we're only changing the size;
1289          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1290
1291         OBD_ALLOC_PTR(op_data);
1292         if (op_data == NULL)
1293                 RETURN(-ENOMEM);
1294
1295         UNLOCK_INODE_MUTEX(inode);
1296         if (ia_valid & ATTR_SIZE)
1297                 UP_WRITE_I_ALLOC_SEM(inode);
1298         cfs_down_write(&lli->lli_trunc_sem);
1299         LOCK_INODE_MUTEX(inode);
1300         if (ia_valid & ATTR_SIZE)
1301                 DOWN_WRITE_I_ALLOC_SEM(inode);
1302
1303         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1304
1305         /* Open epoch for truncate. */
1306         if (exp_connect_som(ll_i2mdexp(inode)) &&
1307             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1308                 op_data->op_flags = MF_EPOCH_OPEN;
1309
1310         rc = ll_md_setattr(inode, op_data, &mod);
1311         if (rc)
1312                 GOTO(out, rc);
1313
1314         ll_ioepoch_open(lli, op_data->op_ioepoch);
1315         if (!lsm || !S_ISREG(inode->i_mode)) {
1316                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1317                 GOTO(out, rc = 0);
1318         }
1319
1320         if (ia_valid & ATTR_SIZE)
1321                 attr->ia_valid |= ATTR_SIZE;
1322         if (ia_valid & (ATTR_SIZE |
1323                         ATTR_ATIME | ATTR_ATIME_SET |
1324                         ATTR_MTIME | ATTR_MTIME_SET))
1325                 /* on truncate and utimes send attributes to osts, setting
1326                  * mtime/atime to past will be performed under PW 0:EOF extent
1327                  * lock (new_size:EOF for truncate)
1328                  * it may seem excessive to send mtime/atime updates to osts
1329                  * when not setting times to past, but it is necessary due to
1330                  * possible time de-synchronization */
1331                 rc = ll_setattr_ost(inode, attr);
1332         EXIT;
1333 out:
1334         if (op_data) {
1335                 if (op_data->op_ioepoch)
1336                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1337                 ll_finish_md_op_data(op_data);
1338         }
1339         cfs_up_write(&lli->lli_trunc_sem);
1340         return rc ? rc : rc1;
1341 }
1342
1343 int ll_setattr(struct dentry *de, struct iattr *attr)
1344 {
1345         int mode = de->d_inode->i_mode;
1346
1347         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1348                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1349                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1350
1351         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1352                                (ATTR_SIZE|ATTR_MODE)) &&
1353             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1354              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1355               !(attr->ia_mode & S_ISGID))))
1356                 attr->ia_valid |= ATTR_FORCE;
1357
1358         if ((mode & S_ISUID) &&
1359             !(attr->ia_mode & S_ISUID) &&
1360             !(attr->ia_valid & ATTR_KILL_SUID))
1361                 attr->ia_valid |= ATTR_KILL_SUID;
1362
1363         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1364             !(attr->ia_mode & S_ISGID) &&
1365             !(attr->ia_valid & ATTR_KILL_SGID))
1366                 attr->ia_valid |= ATTR_KILL_SGID;
1367
1368         return ll_setattr_raw(de->d_inode, attr);
1369 }
1370
1371 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1372                        __u64 max_age, __u32 flags)
1373 {
1374         struct ll_sb_info *sbi = ll_s2sbi(sb);
1375         struct obd_statfs obd_osfs;
1376         int rc;
1377         ENTRY;
1378
1379         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1380         if (rc) {
1381                 CERROR("md_statfs fails: rc = %d\n", rc);
1382                 RETURN(rc);
1383         }
1384
1385         osfs->os_type = sb->s_magic;
1386
1387         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1388                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1389
1390         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1391                 flags |= OBD_STATFS_NODELAY;
1392
1393         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1394                               &obd_osfs, max_age, flags);
1395         if (rc) {
1396                 CERROR("obd_statfs fails: rc = %d\n", rc);
1397                 RETURN(rc);
1398         }
1399
1400         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1401                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1402                obd_osfs.os_files);
1403
1404         osfs->os_bsize = obd_osfs.os_bsize;
1405         osfs->os_blocks = obd_osfs.os_blocks;
1406         osfs->os_bfree = obd_osfs.os_bfree;
1407         osfs->os_bavail = obd_osfs.os_bavail;
1408
1409         /* If we don't have as many objects free on the OST as inodes
1410          * on the MDS, we reduce the total number of inodes to
1411          * compensate, so that the "inodes in use" number is correct.
1412          */
1413         if (obd_osfs.os_ffree < osfs->os_ffree) {
1414                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1415                         obd_osfs.os_ffree;
1416                 osfs->os_ffree = obd_osfs.os_ffree;
1417         }
1418
1419         RETURN(rc);
1420 }
1421 #ifndef HAVE_STATFS_DENTRY_PARAM
1422 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1423 {
1424 #else
1425 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1426 {
1427         struct super_block *sb = de->d_sb;
1428 #endif
1429         struct obd_statfs osfs;
1430         int rc;
1431
1432         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1433         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1434
1435         /* Some amount of caching on the client is allowed */
1436         rc = ll_statfs_internal(sb, &osfs,
1437                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1438                                 0);
1439         if (rc)
1440                 return rc;
1441
1442         statfs_unpack(sfs, &osfs);
1443
1444         /* We need to downshift for all 32-bit kernels, because we can't
1445          * tell if the kernel is being called via sys_statfs64() or not.
1446          * Stop before overflowing f_bsize - in which case it is better
1447          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1448         if (sizeof(long) < 8) {
1449                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1450                         sfs->f_bsize <<= 1;
1451
1452                         osfs.os_blocks >>= 1;
1453                         osfs.os_bfree >>= 1;
1454                         osfs.os_bavail >>= 1;
1455                 }
1456         }
1457
1458         sfs->f_blocks = osfs.os_blocks;
1459         sfs->f_bfree = osfs.os_bfree;
1460         sfs->f_bavail = osfs.os_bavail;
1461
1462         return 0;
1463 }
1464
1465 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1466 {
1467         struct ll_inode_info *lli;
1468         struct lov_stripe_md *lsm;
1469
1470         lli = ll_i2info(inode);
1471         LASSERT(lli->lli_size_sem_owner != current);
1472         cfs_down(&lli->lli_size_sem);
1473         LASSERT(lli->lli_size_sem_owner == NULL);
1474         lli->lli_size_sem_owner = current;
1475         lsm = lli->lli_smd;
1476         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1477                  lsm, lock_lsm);
1478         if (lock_lsm)
1479                 lov_stripe_lock(lsm);
1480 }
1481
1482 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1483 {
1484         struct ll_inode_info *lli;
1485         struct lov_stripe_md *lsm;
1486
1487         lli = ll_i2info(inode);
1488         lsm = lli->lli_smd;
1489         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1490                  lsm, unlock_lsm);
1491         if (unlock_lsm)
1492                 lov_stripe_unlock(lsm);
1493         LASSERT(lli->lli_size_sem_owner == current);
1494         lli->lli_size_sem_owner = NULL;
1495         cfs_up(&lli->lli_size_sem);
1496 }
1497
1498 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1499 {
1500         struct ll_inode_info *lli = ll_i2info(inode);
1501         struct mdt_body *body = md->body;
1502         struct lov_stripe_md *lsm = md->lsm;
1503         struct ll_sb_info *sbi = ll_i2sbi(inode);
1504
1505         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1506         if (lsm != NULL) {
1507                 cfs_down(&lli->lli_och_sem);
1508                 if (lli->lli_smd == NULL) {
1509                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1510                             lsm->lsm_magic != LOV_MAGIC_V3) {
1511                                 dump_lsm(D_ERROR, lsm);
1512                                 LBUG();
1513                         }
1514                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1515                                lsm, inode->i_ino, inode->i_generation, inode);
1516                         /* cl_inode_init must go before lli_smd or a race is
1517                          * possible where client thinks the file has stripes,
1518                          * but lov raid0 is not setup yet and parallel e.g.
1519                          * glimpse would try to use uninitialized lov */
1520                         cl_inode_init(inode, md);
1521                         lli->lli_smd = lsm;
1522                         cfs_up(&lli->lli_och_sem);
1523                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1524                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1525                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1526                 } else {
1527                         cfs_up(&lli->lli_och_sem);
1528                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1529                                 lli->lli_smd->lsm_stripe_count ==
1530                                 lsm->lsm_stripe_count);
1531                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1532                                 CERROR("lsm mismatch for inode %ld\n",
1533                                        inode->i_ino);
1534                                 CERROR("lli_smd:\n");
1535                                 dump_lsm(D_ERROR, lli->lli_smd);
1536                                 CERROR("lsm:\n");
1537                                 dump_lsm(D_ERROR, lsm);
1538                                 LBUG();
1539                         }
1540                 }
1541                 if (lli->lli_smd != lsm)
1542                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1543         }
1544
1545         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1546                 if (body->valid & OBD_MD_FLRMTPERM)
1547                         ll_update_remote_perm(inode, md->remote_perm);
1548         }
1549 #ifdef CONFIG_FS_POSIX_ACL
1550         else if (body->valid & OBD_MD_FLACL) {
1551                 cfs_spin_lock(&lli->lli_lock);
1552                 if (lli->lli_posix_acl)
1553                         posix_acl_release(lli->lli_posix_acl);
1554                 lli->lli_posix_acl = md->posix_acl;
1555                 cfs_spin_unlock(&lli->lli_lock);
1556         }
1557 #endif
1558         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1559         inode->i_generation = cl_fid_build_gen(&body->fid1);
1560
1561         if (body->valid & OBD_MD_FLATIME) {
1562                 if (body->atime > LTIME_S(inode->i_atime))
1563                         LTIME_S(inode->i_atime) = body->atime;
1564                 lli->lli_lvb.lvb_atime = body->atime;
1565         }
1566         if (body->valid & OBD_MD_FLMTIME) {
1567                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1568                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1569                                "to "LPU64"\n", inode->i_ino,
1570                                LTIME_S(inode->i_mtime), body->mtime);
1571                         LTIME_S(inode->i_mtime) = body->mtime;
1572                 }
1573                 lli->lli_lvb.lvb_mtime = body->mtime;
1574         }
1575         if (body->valid & OBD_MD_FLCTIME) {
1576                 if (body->ctime > LTIME_S(inode->i_ctime))
1577                         LTIME_S(inode->i_ctime) = body->ctime;
1578                 lli->lli_lvb.lvb_ctime = body->ctime;
1579         }
1580         if (body->valid & OBD_MD_FLMODE)
1581                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1582         if (body->valid & OBD_MD_FLTYPE)
1583                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1584         LASSERT(inode->i_mode != 0);
1585         if (S_ISREG(inode->i_mode)) {
1586                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1587         } else {
1588                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1589         }
1590 #ifdef HAVE_INODE_BLKSIZE
1591         inode->i_blksize = 1<<inode->i_blkbits;
1592 #endif
1593         if (body->valid & OBD_MD_FLUID)
1594                 inode->i_uid = body->uid;
1595         if (body->valid & OBD_MD_FLGID)
1596                 inode->i_gid = body->gid;
1597         if (body->valid & OBD_MD_FLFLAGS)
1598                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1599         if (body->valid & OBD_MD_FLNLINK)
1600                 inode->i_nlink = body->nlink;
1601         if (body->valid & OBD_MD_FLRDEV)
1602                 inode->i_rdev = old_decode_dev(body->rdev);
1603
1604         if (body->valid & OBD_MD_FLID) {
1605                 /* FID shouldn't be changed! */
1606                 if (fid_is_sane(&lli->lli_fid)) {
1607                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1608                                  "Trying to change FID "DFID
1609                                  " to the "DFID", inode %lu/%u(%p)\n",
1610                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1611                                  inode->i_ino, inode->i_generation, inode);
1612                 } else
1613                         lli->lli_fid = body->fid1;
1614         }
1615
1616         LASSERT(fid_seq(&lli->lli_fid) != 0);
1617
1618         if (body->valid & OBD_MD_FLSIZE) {
1619                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1620                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1621                         struct lustre_handle lockh;
1622                         ldlm_mode_t mode;
1623
1624                         /* As it is possible a blocking ast has been processed
1625                          * by this time, we need to check there is an UPDATE
1626                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1627                          * it. */
1628                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1629                                                &lockh);
1630                         if (mode) {
1631                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1632                                                       LLIF_EPOCH_PENDING |
1633                                                       LLIF_SOM_DIRTY)) {
1634                                         CERROR("ino %lu flags %lu still has "
1635                                                "size authority! do not trust "
1636                                                "the size got from MDS\n",
1637                                                inode->i_ino, lli->lli_flags);
1638                                 } else {
1639                                         /* Use old size assignment to avoid
1640                                          * deadlock bz14138 & bz14326 */
1641                                         inode->i_size = body->size;
1642                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1643                                 }
1644                                 ldlm_lock_decref(&lockh, mode);
1645                         }
1646                 } else {
1647                         /* Use old size assignment to avoid
1648                          * deadlock bz14138 & bz14326 */
1649                         inode->i_size = body->size;
1650
1651                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1652                                inode->i_ino, (unsigned long long)body->size);
1653                 }
1654
1655                 if (body->valid & OBD_MD_FLBLOCKS)
1656                         inode->i_blocks = body->blocks;
1657         }
1658
1659         if (body->valid & OBD_MD_FLMDSCAPA) {
1660                 LASSERT(md->mds_capa);
1661                 ll_add_capa(inode, md->mds_capa);
1662         }
1663         if (body->valid & OBD_MD_FLOSSCAPA) {
1664                 LASSERT(md->oss_capa);
1665                 ll_add_capa(inode, md->oss_capa);
1666         }
1667 }
1668
1669 void ll_read_inode2(struct inode *inode, void *opaque)
1670 {
1671         struct lustre_md *md = opaque;
1672         struct ll_inode_info *lli = ll_i2info(inode);
1673         ENTRY;
1674
1675         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1676                inode->i_ino, inode->i_generation, inode);
1677
1678         ll_lli_init(lli);
1679
1680         LASSERT(!lli->lli_smd);
1681
1682         /* Core attributes from the MDS first.  This is a new inode, and
1683          * the VFS doesn't zero times in the core inode so we have to do
1684          * it ourselves.  They will be overwritten by either MDS or OST
1685          * attributes - we just need to make sure they aren't newer. */
1686         LTIME_S(inode->i_mtime) = 0;
1687         LTIME_S(inode->i_atime) = 0;
1688         LTIME_S(inode->i_ctime) = 0;
1689         inode->i_rdev = 0;
1690         ll_update_inode(inode, md);
1691
1692         /* OIDEBUG(inode); */
1693
1694         /* initializing backing dev info. */
1695         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1696
1697
1698         if (S_ISREG(inode->i_mode)) {
1699                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1700                 inode->i_op = &ll_file_inode_operations;
1701                 inode->i_fop = sbi->ll_fop;
1702                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1703                 EXIT;
1704         } else if (S_ISDIR(inode->i_mode)) {
1705                 inode->i_op = &ll_dir_inode_operations;
1706                 inode->i_fop = &ll_dir_operations;
1707                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1708                 EXIT;
1709         } else if (S_ISLNK(inode->i_mode)) {
1710                 inode->i_op = &ll_fast_symlink_inode_operations;
1711                 EXIT;
1712         } else {
1713                 inode->i_op = &ll_special_inode_operations;
1714
1715                 init_special_inode(inode, inode->i_mode,
1716                                    kdev_t_to_nr(inode->i_rdev));
1717
1718                 EXIT;
1719         }
1720 }
1721
1722 void ll_delete_inode(struct inode *inode)
1723 {
1724         struct ll_sb_info *sbi = ll_i2sbi(inode);
1725         int rc;
1726         ENTRY;
1727
1728         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1729         if (rc)
1730                 CERROR("fid_delete() failed, rc %d\n", rc);
1731
1732         truncate_inode_pages(&inode->i_data, 0);
1733
1734         /* Workaround for LU-118 */
1735         if (inode->i_data.nrpages) {
1736                 TREE_READ_LOCK_IRQ(&inode->i_data);
1737                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1738                 LASSERTF(inode->i_data.nrpages == 0,
1739                          "inode=%lu/%u(%p) nrpages=%lu, see "
1740                          "http://jira.whamcloud.com/browse/LU-118\n",
1741                          inode->i_ino, inode->i_generation, inode,
1742                          inode->i_data.nrpages);
1743         }
1744         /* Workaround end */
1745
1746         clear_inode(inode);
1747
1748         EXIT;
1749 }
1750
1751 int ll_iocontrol(struct inode *inode, struct file *file,
1752                  unsigned int cmd, unsigned long arg)
1753 {
1754         struct ll_sb_info *sbi = ll_i2sbi(inode);
1755         struct ptlrpc_request *req = NULL;
1756         int rc, flags = 0;
1757         ENTRY;
1758
1759         switch(cmd) {
1760         case FSFILT_IOC_GETFLAGS: {
1761                 struct mdt_body *body;
1762                 struct md_op_data *op_data;
1763
1764                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1765                                              0, 0, LUSTRE_OPC_ANY,
1766                                              NULL);
1767                 if (op_data == NULL)
1768                         RETURN(-ENOMEM);
1769
1770                 op_data->op_valid = OBD_MD_FLFLAGS;
1771                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1772                 ll_finish_md_op_data(op_data);
1773                 if (rc) {
1774                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1775                         RETURN(-abs(rc));
1776                 }
1777
1778                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1779
1780                 flags = body->flags;
1781
1782                 ptlrpc_req_finished(req);
1783
1784                 RETURN(put_user(flags, (int *)arg));
1785         }
1786         case FSFILT_IOC_SETFLAGS: {
1787                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1788                 struct obd_info oinfo = { { { 0 } } };
1789                 struct md_op_data *op_data;
1790
1791                 if (get_user(flags, (int *)arg))
1792                         RETURN(-EFAULT);
1793
1794                 oinfo.oi_md = lsm;
1795                 OBDO_ALLOC(oinfo.oi_oa);
1796                 if (!oinfo.oi_oa)
1797                         RETURN(-ENOMEM);
1798
1799                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1800                                              LUSTRE_OPC_ANY, NULL);
1801                 if (IS_ERR(op_data))
1802                         RETURN(PTR_ERR(op_data));
1803
1804                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1805                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1806                 rc = md_setattr(sbi->ll_md_exp, op_data,
1807                                 NULL, 0, NULL, 0, &req, NULL);
1808                 ll_finish_md_op_data(op_data);
1809                 ptlrpc_req_finished(req);
1810                 if (rc) {
1811                         OBDO_FREE(oinfo.oi_oa);
1812                         RETURN(rc);
1813                 }
1814
1815                 if (lsm == NULL) {
1816                         OBDO_FREE(oinfo.oi_oa);
1817                         GOTO(update_cache, rc);
1818                 }
1819
1820                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1821                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1822                 oinfo.oi_oa->o_flags = flags;
1823                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1824                                        OBD_MD_FLGROUP;
1825                 oinfo.oi_capa = ll_mdscapa_get(inode);
1826                 obdo_from_inode(oinfo.oi_oa, inode,
1827                                 &ll_i2info(inode)->lli_fid, 0);
1828                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1829                 capa_put(oinfo.oi_capa);
1830                 OBDO_FREE(oinfo.oi_oa);
1831                 if (rc) {
1832                         if (rc != -EPERM && rc != -EACCES)
1833                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1834                         RETURN(rc);
1835                 }
1836
1837                 EXIT;
1838 update_cache:
1839                 inode->i_flags = ll_ext_to_inode_flags(flags);
1840                 return 0;
1841         }
1842         default:
1843                 RETURN(-ENOSYS);
1844         }
1845
1846         RETURN(0);
1847 }
1848
1849 int ll_flush_ctx(struct inode *inode)
1850 {
1851         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1852
1853         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1854
1855         obd_set_info_async(sbi->ll_md_exp,
1856                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1857                            0, NULL, NULL);
1858         obd_set_info_async(sbi->ll_dt_exp,
1859                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1860                            0, NULL, NULL);
1861         return 0;
1862 }
1863
1864 /* umount -f client means force down, don't save state */
1865 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1866 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1867 {
1868         struct super_block *sb = vfsmnt->mnt_sb;
1869 #else
1870 void ll_umount_begin(struct super_block *sb)
1871 {
1872 #endif
1873         struct lustre_sb_info *lsi = s2lsi(sb);
1874         struct ll_sb_info *sbi = ll_s2sbi(sb);
1875         struct obd_device *obd;
1876         struct obd_ioctl_data *ioc_data;
1877         ENTRY;
1878
1879 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1880         if (!(flags & MNT_FORCE)) {
1881                 EXIT;
1882                 return;
1883         }
1884 #endif
1885
1886         /* Tell the MGC we got umount -f */
1887         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1888
1889         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1890                sb->s_count, atomic_read(&sb->s_active));
1891
1892         obd = class_exp2obd(sbi->ll_md_exp);
1893         if (obd == NULL) {
1894                 CERROR("Invalid MDC connection handle "LPX64"\n",
1895                        sbi->ll_md_exp->exp_handle.h_cookie);
1896                 EXIT;
1897                 return;
1898         }
1899         obd->obd_force = 1;
1900
1901         obd = class_exp2obd(sbi->ll_dt_exp);
1902         if (obd == NULL) {
1903                 CERROR("Invalid LOV connection handle "LPX64"\n",
1904                        sbi->ll_dt_exp->exp_handle.h_cookie);
1905                 EXIT;
1906                 return;
1907         }
1908         obd->obd_force = 1;
1909
1910         OBD_ALLOC_PTR(ioc_data);
1911         if (ioc_data) {
1912                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1913                               sizeof ioc_data, ioc_data, NULL);
1914
1915                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1916                               sizeof ioc_data, ioc_data, NULL);
1917
1918                 OBD_FREE_PTR(ioc_data);
1919         }
1920
1921
1922         /* Really, we'd like to wait until there are no requests outstanding,
1923          * and then continue.  For now, we just invalidate the requests,
1924          * schedule() and sleep one second if needed, and hope.
1925          */
1926         cfs_schedule();
1927 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1928         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1929                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1930                                                    cfs_time_seconds(1));
1931                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1932                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1933                                       "may try to umount it a bit later\n",
1934                                       atomic_read(&vfsmnt->mnt_count));
1935         }
1936 #endif
1937
1938         EXIT;
1939 }
1940
1941 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1942 {
1943         struct ll_sb_info *sbi = ll_s2sbi(sb);
1944         int err;
1945         __u32 read_only;
1946
1947         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1948                 read_only = *flags & MS_RDONLY;
1949                 err = obd_set_info_async(sbi->ll_md_exp,
1950                                          sizeof(KEY_READ_ONLY),
1951                                          KEY_READ_ONLY, sizeof(read_only),
1952                                          &read_only, NULL);
1953                 if (err) {
1954                         CERROR("Failed to change the read-only flag during "
1955                                "remount: %d\n", err);
1956                         return err;
1957                 }
1958
1959                 if (read_only)
1960                         sb->s_flags |= MS_RDONLY;
1961                 else
1962                         sb->s_flags &= ~MS_RDONLY;
1963         }
1964         return 0;
1965 }
1966
1967 int ll_prep_inode(struct inode **inode,
1968                   struct ptlrpc_request *req,
1969                   struct super_block *sb)
1970 {
1971         struct ll_sb_info *sbi = NULL;
1972         struct lustre_md md;
1973         int rc;
1974         ENTRY;
1975
1976         LASSERT(*inode || sb);
1977         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1978         memset(&md, 0, sizeof(struct lustre_md));
1979
1980         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1981                               sbi->ll_md_exp, &md);
1982         if (rc)
1983                 RETURN(rc);
1984
1985         if (*inode) {
1986                 ll_update_inode(*inode, &md);
1987         } else {
1988                 LASSERT(sb != NULL);
1989
1990                 /*
1991                  * At this point server returns to client's same fid as client
1992                  * generated for creating. So using ->fid1 is okay here.
1993                  */
1994                 LASSERT(fid_is_sane(&md.body->fid1));
1995
1996                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
1997                 if (*inode == NULL || IS_ERR(*inode)) {
1998                         if (md.lsm)
1999                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2000 #ifdef CONFIG_FS_POSIX_ACL
2001                         if (md.posix_acl) {
2002                                 posix_acl_release(md.posix_acl);
2003                                 md.posix_acl = NULL;
2004                         }
2005 #endif
2006                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2007                         *inode = NULL;
2008                         CERROR("new_inode -fatal: rc %d\n", rc);
2009                         GOTO(out, rc);
2010                 }
2011         }
2012
2013 out:
2014         md_free_lustre_md(sbi->ll_md_exp, &md);
2015         RETURN(rc);
2016 }
2017
2018 int ll_obd_statfs(struct inode *inode, void *arg)
2019 {
2020         struct ll_sb_info *sbi = NULL;
2021         struct obd_export *exp;
2022         char *buf = NULL;
2023         struct obd_ioctl_data *data = NULL;
2024         __u32 type;
2025         int len = 0, rc;
2026
2027         if (!inode || !(sbi = ll_i2sbi(inode)))
2028                 GOTO(out_statfs, rc = -EINVAL);
2029
2030         rc = obd_ioctl_getdata(&buf, &len, arg);
2031         if (rc)
2032                 GOTO(out_statfs, rc);
2033
2034         data = (void*)buf;
2035         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2036             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2037                 GOTO(out_statfs, rc = -EINVAL);
2038
2039         if (data->ioc_inllen1 != sizeof(__u32) ||
2040             data->ioc_inllen2 != sizeof(__u32) ||
2041             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2042             data->ioc_plen2 != sizeof(struct obd_uuid))
2043                 GOTO(out_statfs, rc = -EINVAL);
2044
2045         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2046         if (type == LL_STATFS_MDC)
2047                 exp = sbi->ll_md_exp;
2048         else if (type == LL_STATFS_LOV)
2049                 exp = sbi->ll_dt_exp;
2050         else
2051                 GOTO(out_statfs, rc = -ENODEV);
2052
2053         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2054         if (rc)
2055                 GOTO(out_statfs, rc);
2056 out_statfs:
2057         if (buf)
2058                 obd_ioctl_freedata(buf, len);
2059         return rc;
2060 }
2061
2062 int ll_process_config(struct lustre_cfg *lcfg)
2063 {
2064         char *ptr;
2065         void *sb;
2066         struct lprocfs_static_vars lvars;
2067         unsigned long x;
2068         int rc = 0;
2069
2070         lprocfs_llite_init_vars(&lvars);
2071
2072         /* The instance name contains the sb: lustre-client-aacfe000 */
2073         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2074         if (!ptr || !*(++ptr))
2075                 return -EINVAL;
2076         if (sscanf(ptr, "%lx", &x) != 1)
2077                 return -EINVAL;
2078         sb = (void *)x;
2079         /* This better be a real Lustre superblock! */
2080         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2081
2082         /* Note we have not called client_common_fill_super yet, so
2083            proc fns must be able to handle that! */
2084         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2085                                       lcfg, sb);
2086         if (rc > 0)
2087                 rc = 0;
2088         return(rc);
2089 }
2090
2091 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2092 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2093                                        struct inode *i1, struct inode *i2,
2094                                        const char *name, int namelen,
2095                                        int mode, __u32 opc, void *data)
2096 {
2097         LASSERT(i1 != NULL);
2098
2099         if (namelen > ll_i2sbi(i1)->ll_namelen)
2100                 return ERR_PTR(-ENAMETOOLONG);
2101
2102         if (op_data == NULL)
2103                 OBD_ALLOC_PTR(op_data);
2104
2105         if (op_data == NULL)
2106                 return ERR_PTR(-ENOMEM);
2107
2108         ll_i2gids(op_data->op_suppgids, i1, i2);
2109         op_data->op_fid1 = *ll_inode2fid(i1);
2110         op_data->op_capa1 = ll_mdscapa_get(i1);
2111
2112         if (i2) {
2113                 op_data->op_fid2 = *ll_inode2fid(i2);
2114                 op_data->op_capa2 = ll_mdscapa_get(i2);
2115         } else {
2116                 fid_zero(&op_data->op_fid2);
2117                 op_data->op_capa2 = NULL;
2118         }
2119
2120         op_data->op_name = name;
2121         op_data->op_namelen = namelen;
2122         op_data->op_mode = mode;
2123         op_data->op_mod_time = cfs_time_current_sec();
2124         op_data->op_fsuid = cfs_curproc_fsuid();
2125         op_data->op_fsgid = cfs_curproc_fsgid();
2126         op_data->op_cap = cfs_curproc_cap_pack();
2127         op_data->op_bias = MDS_CHECK_SPLIT;
2128         op_data->op_opc = opc;
2129         op_data->op_mds = 0;
2130         op_data->op_data = data;
2131
2132         return op_data;
2133 }
2134
2135 void ll_finish_md_op_data(struct md_op_data *op_data)
2136 {
2137         capa_put(op_data->op_capa1);
2138         capa_put(op_data->op_capa2);
2139         OBD_FREE_PTR(op_data);
2140 }
2141
2142 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2143 {
2144         struct ll_sb_info *sbi;
2145
2146         LASSERT((seq != NULL) && (vfs != NULL));
2147         sbi = ll_s2sbi(vfs->mnt_sb);
2148
2149         if (sbi->ll_flags & LL_SBI_NOLCK)
2150                 seq_puts(seq, ",nolock");
2151
2152         if (sbi->ll_flags & LL_SBI_FLOCK)
2153                 seq_puts(seq, ",flock");
2154
2155         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2156                 seq_puts(seq, ",localflock");
2157
2158         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2159                 seq_puts(seq, ",user_xattr");
2160
2161         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2162                 seq_puts(seq, ",lazystatfs");
2163
2164         RETURN(0);
2165 }