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  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/llite_lib.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/module.h>
  47 #include <linux/types.h>
  48 #include <linux/version.h>
  49 #include <linux/mm.h>
  50
  51 #include <lustre_lite.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 cfs_mem_cache_t *ll_file_data_slab;
  63
  64 CFS_LIST_HEAD(ll_super_blocks);
  65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  66
  67 #ifndef MS_HAS_NEW_AOPS
  68 extern struct address_space_operations ll_aops;
  69 extern struct address_space_operations ll_dir_aops;
  70 #else
  71 extern struct address_space_operations_ext ll_aops;
  72 extern struct address_space_operations_ext ll_dir_aops;
  73 #endif
  74
  75 #ifndef log2
  76 #define log2(n) cfs_ffz(~(n))
  77 #endif
  78
  79 static struct ll_sb_info *ll_init_sbi(void)
  80 {
  81         struct ll_sb_info *sbi = NULL;
  82         unsigned long pages;
  83         struct sysinfo si;
  84         class_uuid_t uuid;
  85         int i;
  86         ENTRY;
  87
  88         OBD_ALLOC(sbi, sizeof(*sbi));
  89         if (!sbi)
  90                 RETURN(NULL);
  91
  92         cfs_spin_lock_init(&sbi->ll_lock);
  93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  95         cfs_spin_lock_init(&sbi->ll_process_lock);
  96         sbi->ll_rw_stats_on = 0;
  97
  98         si_meminfo(&si);
  99         pages = si.totalram - si.totalhigh;
 100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 101 #ifdef HAVE_BGL_SUPPORT
 102                 sbi->ll_async_page_max = pages / 4;
 103 #else
 104                 sbi->ll_async_page_max = pages / 2;
 105 #endif
 106         } else {
 107                 sbi->ll_async_page_max = (pages / 4) * 3;
 108         }
 109
 110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 111                                            SBI_DEFAULT_READAHEAD_MAX);
 112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 117
 118         ll_generate_random_uuid(uuid);
 119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 121
 122         cfs_spin_lock(&ll_sb_lock);
 123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 124         cfs_spin_unlock(&ll_sb_lock);
 125
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_r_hist.oh_lock);
 137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 138                                    pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         atomic_set(&sbi->ll_sa_total, 0);
 144         atomic_set(&sbi->ll_sa_wrong, 0);
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         if (proc_lustre_fs_root) {
 202                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 203                                                   dt, md);
 204                 if (err < 0)
 205                         CERROR("could not register mount in /proc/fs/lustre\n");
 206         }
 207
 208         /* indicate the features supported by this client */
 209         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 210                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 211                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 212                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 213                                   OBD_CONNECT_CANCELSET| OBD_CONNECT_FID      |
 214                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 215                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 216                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH;
 217
 218         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 219                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 220
 221 #ifdef HAVE_LRU_RESIZE_SUPPORT
 222         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 223                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 224 #endif
 225 #ifdef CONFIG_FS_POSIX_ACL
 226         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 227 #endif
 228         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 229         data->ocd_version = LUSTRE_VERSION_CODE;
 230
 231         if (sb->s_flags & MS_RDONLY)
 232                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 233         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 234                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 235
 236 #ifdef HAVE_MS_FLOCK_LOCK
 237         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 238         sb->s_flags |= MS_FLOCK_LOCK;
 239 #endif
 240 #ifdef MS_HAS_NEW_AOPS
 241         sb->s_flags |= MS_HAS_NEW_AOPS;
 242 #endif
 243
 244         if (sbi->ll_flags & LL_SBI_FLOCK)
 245                 sbi->ll_fop = &ll_file_operations_flock;
 246         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 247                 sbi->ll_fop = &ll_file_operations;
 248         else
 249                 sbi->ll_fop = &ll_file_operations_noflock;
 250
 251         /* real client */
 252         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 253         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 254                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 255
 256         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 257
 258         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 259         if (err == -EBUSY) {
 260                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 261                                    "recovery, of which this client is not a "
 262                                    "part. Please wait for recovery to complete,"
 263                                    " abort, or time out.\n", md);
 264                 GOTO(out, err);
 265         } else if (err) {
 266                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 267                 GOTO(out, err);
 268         }
 269
 270         err = obd_fid_init(sbi->ll_md_exp);
 271         if (err) {
 272                 CERROR("Can't init metadata layer FID infrastructure, "
 273                        "rc %d\n", err);
 274                 GOTO(out_md, err);
 275         }
 276
 277         err = obd_statfs(obd, osfs,
 278                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 279         if (err)
 280                 GOTO(out_md_fid, err);
 281
 282         size = sizeof(*data);
 283         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 284                            KEY_CONN_DATA,  &size, data, NULL);
 285         if (err) {
 286                 CERROR("Get connect data failed: %d \n", err);
 287                 GOTO(out_md, err);
 288         }
 289
 290         LASSERT(osfs->os_bsize);
 291         sb->s_blocksize = osfs->os_bsize;
 292         sb->s_blocksize_bits = log2(osfs->os_bsize);
 293         sb->s_magic = LL_SUPER_MAGIC;
 294
 295         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 296          *         retval = in_file->f_op->sendfile(...);
 297          *         if (*ppos > max)
 298          *                 retval = -EOVERFLOW;
 299          *
 300          * it will check if *ppos is greater than max. However, max equals to
 301          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 302          * has been defined as a signed long long integer in linux kernel. */
 303 #if BITS_PER_LONG == 64
 304         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 305 #else
 306         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 307 #endif
 308         sbi->ll_namelen = osfs->os_namelen;
 309         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 310
 311         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 312             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 313                 LCONSOLE_INFO("Disabling user_xattr feature because "
 314                               "it is not supported on the server\n");
 315                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 316         }
 317
 318         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 319 #ifdef MS_POSIXACL
 320                 sb->s_flags |= MS_POSIXACL;
 321 #endif
 322                 sbi->ll_flags |= LL_SBI_ACL;
 323         } else {
 324                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 325 #ifdef MS_POSIXACL
 326                 sb->s_flags &= ~MS_POSIXACL;
 327 #endif
 328                 sbi->ll_flags &= ~LL_SBI_ACL;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 332                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 333                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 334                         LCONSOLE_INFO("client is set as remote by default.\n");
 335                 }
 336         } else {
 337                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 338                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 339                         LCONSOLE_INFO("client claims to be remote, but server "
 340                                       "rejected, forced to be local.\n");
 341                 }
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 345                 LCONSOLE_INFO("client enabled MDS capability!\n");
 346                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 347         }
 348
 349         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 350                 LCONSOLE_INFO("client enabled OSS capability!\n");
 351                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 352         }
 353
 354         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 355                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 358                 sbi->ll_md_brw_size = data->ocd_brw_size;
 359         else
 360                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 361
 362         obd = class_name2obd(dt);
 363         if (!obd) {
 364                 CERROR("DT %s: not setup or attached\n", dt);
 365                 GOTO(out_md_fid, err = -ENODEV);
 366         }
 367
 368         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 369                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 370                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 371                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 372                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 373                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 374                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 375                                   OBD_CONNECT_MAXBYTES;
 376
 377         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 378                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 379
 380         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 381                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 382                  * disabled by default, because it can still be enabled on the
 383                  * fly via /proc. As a consequence, we still need to come to an
 384                  * agreement on the supported algorithms at connect time */
 385                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 386
 387                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 388                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 389                 else
 390                         /* send the list of supported checksum types */
 391                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 392         }
 393
 394 #ifdef HAVE_LRU_RESIZE_SUPPORT
 395         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 396 #endif
 397         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 398                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 399
 400         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 401                "ocd_grant: %d\n", data->ocd_connect_flags,
 402                data->ocd_version, data->ocd_grant);
 403
 404         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 405         obd->obd_upcall.onu_upcall = cl_ocd_update;
 406
 407         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 408
 409         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 410         if (err == -EBUSY) {
 411                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 412                                    "recovery, of which this client is not a "
 413                                    "part.  Please wait for recovery to "
 414                                    "complete, abort, or time out.\n", dt);
 415                 GOTO(out_md_fid, err);
 416         } else if (err) {
 417                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 418                 GOTO(out_md_fid, err);
 419         }
 420
 421         err = obd_fid_init(sbi->ll_dt_exp);
 422         if (err) {
 423                 CERROR("Can't init data layer FID infrastructure, "
 424                        "rc %d\n", err);
 425                 GOTO(out_dt, err);
 426         }
 427
 428         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 429         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 430         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 431         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 432         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 433
 434         fid_zero(&sbi->ll_root_fid);
 435         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 436         if (err) {
 437                 CERROR("cannot mds_connect: rc = %d\n", err);
 438                 GOTO(out_lock_cn_cb, err);
 439         }
 440         if (!fid_is_sane(&sbi->ll_root_fid)) {
 441                 CERROR("Invalid root fid during mount\n");
 442                 GOTO(out_lock_cn_cb, err = -EINVAL);
 443         }
 444         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 445
 446         sb->s_op = &lustre_super_operations;
 447 #if THREAD_SIZE >= 8192
 448         sb->s_export_op = &lustre_export_operations;
 449 #endif
 450
 451         /* make root inode
 452          * XXX: move this to after cbd setup? */
 453         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 454         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 455                 valid |= OBD_MD_FLRMTPERM;
 456         else if (sbi->ll_flags & LL_SBI_ACL)
 457                 valid |= OBD_MD_FLACL;
 458
 459         OBD_ALLOC_PTR(op_data);
 460         if (op_data == NULL)
 461                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 462
 463         op_data->op_fid1 = sbi->ll_root_fid;
 464         op_data->op_mode = 0;
 465         op_data->op_capa1 = oc;
 466         op_data->op_valid = valid;
 467
 468         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 469         if (oc)
 470                 capa_put(oc);
 471         OBD_FREE_PTR(op_data);
 472         if (err) {
 473                 CERROR("md_getattr failed for root: rc = %d\n", err);
 474                 GOTO(out_lock_cn_cb, err);
 475         }
 476         memset(&lmd, 0, sizeof(lmd));
 477         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 478                                sbi->ll_md_exp, &lmd);
 479         if (err) {
 480                 CERROR("failed to understand root inode md: rc = %d\n", err);
 481                 ptlrpc_req_finished (request);
 482                 GOTO(out_lock_cn_cb, err);
 483         }
 484
 485         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 486         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 487         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 488         ptlrpc_req_finished(request);
 489
 490         if (root == NULL || IS_ERR(root)) {
 491                 if (lmd.lsm)
 492                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 493 #ifdef CONFIG_FS_POSIX_ACL
 494                 if (lmd.posix_acl) {
 495                         posix_acl_release(lmd.posix_acl);
 496                         lmd.posix_acl = NULL;
 497                 }
 498 #endif
 499                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 500                 root = NULL;
 501                 CERROR("lustre_lite: bad iget4 for root\n");
 502                 GOTO(out_root, err);
 503         }
 504
 505         err = ll_close_thread_start(&sbi->ll_lcq);
 506         if (err) {
 507                 CERROR("cannot start close thread: rc %d\n", err);
 508                 GOTO(out_root, err);
 509         }
 510
 511 #ifdef CONFIG_FS_POSIX_ACL
 512         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 513                 rct_init(&sbi->ll_rct);
 514                 et_init(&sbi->ll_et);
 515         }
 516 #endif
 517
 518         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 519         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 520                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 521                                  NULL);
 522         cl_sb_init(sb);
 523
 524         sb->s_root = d_alloc_root(root);
 525         sb->s_root->d_op = &ll_d_root_ops;
 526
 527         sbi->ll_sdev_orig = sb->s_dev;
 528
 529         /* We set sb->s_dev equal on all lustre clients in order to support
 530          * NFS export clustering.  NFSD requires that the FSID be the same
 531          * on all clients. */
 532         /* s_dev is also used in lt_compare() to compare two fs, but that is
 533          * only a node-local comparison. */
 534         uuid = obd_get_uuid(sbi->ll_md_exp);
 535         if (uuid != NULL)
 536                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 537         sbi->ll_mnt = mnt;
 538
 539         if (data != NULL)
 540                 OBD_FREE_PTR(data);
 541         if (osfs != NULL)
 542                 OBD_FREE_PTR(osfs);
 543
 544         RETURN(err);
 545 out_root:
 546         if (root)
 547                 iput(root);
 548 out_lock_cn_cb:
 549         obd_fid_fini(sbi->ll_dt_exp);
 550 out_dt:
 551         obd_disconnect(sbi->ll_dt_exp);
 552         sbi->ll_dt_exp = NULL;
 553 out_md_fid:
 554         obd_fid_fini(sbi->ll_md_exp);
 555 out_md:
 556         obd_disconnect(sbi->ll_md_exp);
 557         sbi->ll_md_exp = NULL;
 558 out:
 559         if (data != NULL)
 560                 OBD_FREE_PTR(data);
 561         if (osfs != NULL)
 562                 OBD_FREE_PTR(osfs);
 563         lprocfs_unregister_mountpoint(sbi);
 564         return err;
 565 }
 566
 567 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 568 {
 569         int size, rc;
 570
 571         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 572         size = sizeof(int);
 573         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 574                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 575         if (rc)
 576                 CERROR("Get max mdsize error rc %d \n", rc);
 577
 578         RETURN(rc);
 579 }
 580
 581 void ll_dump_inode(struct inode *inode)
 582 {
 583         struct list_head *tmp;
 584         int dentry_count = 0;
 585
 586         LASSERT(inode != NULL);
 587
 588         list_for_each(tmp, &inode->i_dentry)
 589                 dentry_count++;
 590
 591         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 592                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 593                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 594 }
 595
 596 void lustre_dump_dentry(struct dentry *dentry, int recur)
 597 {
 598         struct list_head *tmp;
 599         int subdirs = 0;
 600
 601         LASSERT(dentry != NULL);
 602
 603         list_for_each(tmp, &dentry->d_subdirs)
 604                 subdirs++;
 605
 606         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 607                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 608                dentry->d_name.len, dentry->d_name.name,
 609                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 610                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 611                dentry->d_flags, dentry->d_fsdata, subdirs);
 612         if (dentry->d_inode != NULL)
 613                 ll_dump_inode(dentry->d_inode);
 614
 615         if (recur == 0)
 616                 return;
 617
 618        list_for_each(tmp, &dentry->d_subdirs) {
 619                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 620                 lustre_dump_dentry(d, recur - 1);
 621         }
 622 }
 623
 624 void client_common_put_super(struct super_block *sb)
 625 {
 626         struct ll_sb_info *sbi = ll_s2sbi(sb);
 627         ENTRY;
 628
 629 #ifdef CONFIG_FS_POSIX_ACL
 630         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 631                 et_fini(&sbi->ll_et);
 632                 rct_fini(&sbi->ll_rct);
 633         }
 634 #endif
 635
 636         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 637
 638         ll_close_thread_shutdown(sbi->ll_lcq);
 639
 640         cl_sb_fini(sb);
 641
 642         cfs_list_del(&sbi->ll_conn_chain);
 643
 644         obd_fid_fini(sbi->ll_dt_exp);
 645         obd_disconnect(sbi->ll_dt_exp);
 646         sbi->ll_dt_exp = NULL;
 647
 648         lprocfs_unregister_mountpoint(sbi);
 649
 650         obd_fid_fini(sbi->ll_md_exp);
 651         obd_disconnect(sbi->ll_md_exp);
 652         sbi->ll_md_exp = NULL;
 653
 654         EXIT;
 655 }
 656
 657 void ll_kill_super(struct super_block *sb)
 658 {
 659         struct ll_sb_info *sbi;
 660
 661         ENTRY;
 662
 663         /* not init sb ?*/
 664         if (!(sb->s_flags & MS_ACTIVE))
 665                 return;
 666
 667         sbi = ll_s2sbi(sb);
 668         /* we need restore s_dev from changed for clustred NFS before put_super
 669          * because new kernels have cached s_dev and change sb->s_dev in
 670          * put_super not affected real removing devices */
 671         if (sbi)
 672                 sb->s_dev = sbi->ll_sdev_orig;
 673         EXIT;
 674 }
 675
 676 char *ll_read_opt(const char *opt, char *data)
 677 {
 678         char *value;
 679         char *retval;
 680         ENTRY;
 681
 682         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 683         if (strncmp(opt, data, strlen(opt)))
 684                 RETURN(NULL);
 685         if ((value = strchr(data, '=')) == NULL)
 686                 RETURN(NULL);
 687
 688         value++;
 689         OBD_ALLOC(retval, strlen(value) + 1);
 690         if (!retval) {
 691                 CERROR("out of memory!\n");
 692                 RETURN(NULL);
 693         }
 694
 695         memcpy(retval, value, strlen(value)+1);
 696         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 697         RETURN(retval);
 698 }
 699
 700 static inline int ll_set_opt(const char *opt, char *data, int fl)
 701 {
 702         if (strncmp(opt, data, strlen(opt)) != 0)
 703                 return(0);
 704         else
 705                 return(fl);
 706 }
 707
 708 /* non-client-specific mount options are parsed in lmd_parse */
 709 static int ll_options(char *options, int *flags)
 710 {
 711         int tmp;
 712         char *s1 = options, *s2;
 713         ENTRY;
 714
 715         if (!options)
 716                 RETURN(0);
 717
 718         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 719
 720         while (*s1) {
 721                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 722                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 723                 if (tmp) {
 724                         *flags |= tmp;
 725                         goto next;
 726                 }
 727                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 728                 if (tmp) {
 729                         *flags |= tmp;
 730                         goto next;
 731                 }
 732                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 733                 if (tmp) {
 734                         *flags |= tmp;
 735                         goto next;
 736                 }
 737                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 738                 if (tmp) {
 739                         *flags &= ~tmp;
 740                         goto next;
 741                 }
 742                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 743                 if (tmp) {
 744                         *flags |= tmp;
 745                         goto next;
 746                 }
 747                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 748                 if (tmp) {
 749                         *flags &= ~tmp;
 750                         goto next;
 751                 }
 752 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 753                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 754                 if (tmp) {
 755                         /* Ignore deprecated mount option.  The client will
 756                          * always try to mount with ACL support, whether this
 757                          * is used depends on whether server supports it. */
 758                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 759                                                   "mount option 'acl'.\n");
 760                         goto next;
 761                 }
 762                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 763                 if (tmp) {
 764                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 765                                                   "mount option 'noacl'.\n");
 766                         goto next;
 767                 }
 768 #else
 769 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 770 #endif
 771                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776
 777                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 778                 if (tmp) {
 779                         *flags |= tmp;
 780                         goto next;
 781                 }
 782                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 783                 if (tmp) {
 784                         *flags &= ~tmp;
 785                         goto next;
 786                 }
 787                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 788                 if (tmp) {
 789                         *flags |= tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 793                 if (tmp) {
 794                         *flags &= ~tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 803                 if (tmp) {
 804                         *flags &= ~tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817
 818                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 819                                    s1);
 820                 RETURN(-EINVAL);
 821
 822 next:
 823                 /* Find next opt */
 824                 s2 = strchr(s1, ',');
 825                 if (s2 == NULL)
 826                         break;
 827                 s1 = s2 + 1;
 828         }
 829         RETURN(0);
 830 }
 831
 832 void ll_lli_init(struct ll_inode_info *lli)
 833 {
 834         lli->lli_inode_magic = LLI_INODE_MAGIC;
 835         cfs_sema_init(&lli->lli_size_sem, 1);
 836         cfs_sema_init(&lli->lli_write_sem, 1);
 837         cfs_init_rwsem(&lli->lli_trunc_sem);
 838         lli->lli_flags = 0;
 839         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 840         cfs_spin_lock_init(&lli->lli_lock);
 841         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 842         lli->lli_inode_magic = LLI_INODE_MAGIC;
 843         cfs_sema_init(&lli->lli_och_sem, 1);
 844         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 845         lli->lli_mds_exec_och = NULL;
 846         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 847         lli->lli_open_fd_exec_count = 0;
 848         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 849         lli->lli_remote_perms = NULL;
 850         lli->lli_rmtperm_utime = 0;
 851         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 852         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 853         cfs_spin_lock_init(&lli->lli_sa_lock);
 854         cfs_sema_init(&lli->lli_readdir_sem, 1);
 855         fid_zero(&lli->lli_pfid);
 856 }
 857
 858 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 859 {
 860 #ifdef HAVE_BDI_REGISTER
 861         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 862
 863         bdi->name = "lustre";
 864         return bdi_register(bdi, NULL, "lustre-%d",
 865                             atomic_inc_return(&ll_bdi_num));
 866 #else
 867         return 0;
 868 #endif
 869 }
 870
 871 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 872 {
 873         struct lustre_profile *lprof = NULL;
 874         struct lustre_sb_info *lsi = s2lsi(sb);
 875         struct ll_sb_info *sbi;
 876         char  *dt = NULL, *md = NULL;
 877         char  *profilenm = get_profile_name(sb);
 878         struct config_llog_instance *cfg;
 879         const int instlen = sizeof(cfg->cfg_instance) * 2 + 1;
 880         int    err;
 881         ENTRY;
 882
 883         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 884
 885         OBD_ALLOC_PTR(cfg);
 886         if (cfg == NULL)
 887                 RETURN(-ENOMEM);
 888
 889         cfs_module_get();
 890
 891         /* client additional sb info */
 892         lsi->lsi_llsbi = sbi = ll_init_sbi();
 893         if (!sbi) {
 894                 cfs_module_put(THIS_MODULE);
 895                 OBD_FREE_PTR(cfg);
 896                 RETURN(-ENOMEM);
 897         }
 898
 899         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 900         if (err)
 901                 GOTO(out_free, err);
 902
 903         err = ll_bdi_init(&lsi->lsi_bdi);
 904         if (err)
 905                 GOTO(out_free, err);
 906         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 907         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 908         err = ll_bdi_register(&lsi->lsi_bdi);
 909         if (err)
 910                 GOTO(out_free, err);
 911
 912 #ifdef HAVE_SB_BDI
 913         sb->s_bdi = &lsi->lsi_bdi;
 914 #endif
 915
 916         /* Generate a string unique to this super, in case some joker tries
 917            to mount the same fs at two mount points.
 918            Use the address of the super itself.*/
 919         cfg->cfg_instance = sb;
 920         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 921
 922         /* set up client obds */
 923         err = lustre_process_log(sb, profilenm, cfg);
 924         if (err < 0) {
 925                 CERROR("Unable to process log: %d\n", err);
 926                 GOTO(out_free, err);
 927         }
 928
 929         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 930         lprof = class_get_profile(profilenm);
 931         if (lprof == NULL) {
 932                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 933                                    " read from the MGS.  Does that filesystem "
 934                                    "exist?\n", profilenm);
 935                 GOTO(out_free, err = -EINVAL);
 936         }
 937         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 938                lprof->lp_md, lprof->lp_dt);
 939
 940         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen);
 941         if (!dt)
 942                 GOTO(out_free, err = -ENOMEM);
 943         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 944
 945         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen);
 946         if (!md)
 947                 GOTO(out_free, err = -ENOMEM);
 948         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
 949
 950         /* connections, registrations, sb setup */
 951         err = client_common_fill_super(sb, md, dt, mnt);
 952
 953 out_free:
 954         if (md)
 955                 OBD_FREE(md, strlen(lprof->lp_md) + instlen);
 956         if (dt)
 957                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen);
 958         if (err)
 959                 ll_put_super(sb);
 960         else
 961                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 962
 963         OBD_FREE_PTR(cfg);
 964         RETURN(err);
 965 } /* ll_fill_super */
 966
 967
 968 void lu_context_keys_dump(void);
 969
 970 void ll_put_super(struct super_block *sb)
 971 {
 972         struct config_llog_instance cfg;
 973         struct obd_device *obd;
 974         struct lustre_sb_info *lsi = s2lsi(sb);
 975         struct ll_sb_info *sbi = ll_s2sbi(sb);
 976         char *profilenm = get_profile_name(sb);
 977         int force = 1, next;
 978         ENTRY;
 979
 980         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 981
 982         ll_print_capa_stat(sbi);
 983
 984         cfg.cfg_instance = sb;
 985         lustre_end_log(sb, profilenm, &cfg);
 986
 987         if (sbi->ll_md_exp) {
 988                 obd = class_exp2obd(sbi->ll_md_exp);
 989                 if (obd)
 990                         force = obd->obd_force;
 991         }
 992
 993         /* We need to set force before the lov_disconnect in
 994            lustre_common_put_super, since l_d cleans up osc's as well. */
 995         if (force) {
 996                 next = 0;
 997                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 998                                                      &next)) != NULL) {
 999                         obd->obd_force = force;
1000                 }
1001         }
1002
1003         if (sbi->ll_lcq) {
1004                 /* Only if client_common_fill_super succeeded */
1005                 client_common_put_super(sb);
1006         }
1007
1008         next = 0;
1009         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1010                 class_manual_cleanup(obd);
1011         }
1012
1013         if (profilenm)
1014                 class_del_profile(profilenm);
1015
1016         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1017                 ll_bdi_destroy(&lsi->lsi_bdi);
1018                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1019         }
1020
1021         ll_free_sbi(sb);
1022         lsi->lsi_llsbi = NULL;
1023
1024         lustre_common_put_super(sb);
1025
1026         cl_env_cache_purge(~0);
1027
1028         LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1029
1030         cfs_module_put(THIS_MODULE);
1031
1032         EXIT;
1033 } /* client_put_super */
1034
1035 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1036 {
1037         struct inode *inode = NULL;
1038         /* NOTE: we depend on atomic igrab() -bzzz */
1039         lock_res_and_lock(lock);
1040         if (lock->l_ast_data) {
1041                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1042                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1043                         inode = igrab(lock->l_ast_data);
1044                 } else {
1045                         inode = lock->l_ast_data;
1046                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1047                                          D_WARNING, lock, "l_ast_data %p is "
1048                                          "bogus: magic %08x", lock->l_ast_data,
1049                                          lli->lli_inode_magic);
1050                         inode = NULL;
1051                 }
1052         }
1053         unlock_res_and_lock(lock);
1054         return inode;
1055 }
1056
1057 static int null_if_equal(struct ldlm_lock *lock, void *data)
1058 {
1059         if (data == lock->l_ast_data) {
1060                 lock->l_ast_data = NULL;
1061
1062                 if (lock->l_req_mode != lock->l_granted_mode)
1063                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1064         }
1065
1066         return LDLM_ITER_CONTINUE;
1067 }
1068
1069 void ll_clear_inode(struct inode *inode)
1070 {
1071         struct ll_inode_info *lli = ll_i2info(inode);
1072         struct ll_sb_info *sbi = ll_i2sbi(inode);
1073         ENTRY;
1074
1075         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1076                inode->i_generation, inode);
1077
1078         if (S_ISDIR(inode->i_mode)) {
1079                 /* these should have been cleared in ll_file_release */
1080                 LASSERT(lli->lli_sai == NULL);
1081                 LASSERT(lli->lli_opendir_key == NULL);
1082                 LASSERT(lli->lli_opendir_pid == 0);
1083         }
1084
1085         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1086         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1087                          null_if_equal, inode);
1088
1089         LASSERT(!lli->lli_open_fd_write_count);
1090         LASSERT(!lli->lli_open_fd_read_count);
1091         LASSERT(!lli->lli_open_fd_exec_count);
1092
1093         if (lli->lli_mds_write_och)
1094                 ll_md_real_close(inode, FMODE_WRITE);
1095         if (lli->lli_mds_exec_och)
1096                 ll_md_real_close(inode, FMODE_EXEC);
1097         if (lli->lli_mds_read_och)
1098                 ll_md_real_close(inode, FMODE_READ);
1099
1100         if (lli->lli_symlink_name) {
1101                 OBD_FREE(lli->lli_symlink_name,
1102                          strlen(lli->lli_symlink_name) + 1);
1103                 lli->lli_symlink_name = NULL;
1104         }
1105
1106         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1107                 LASSERT(lli->lli_posix_acl == NULL);
1108                 if (lli->lli_remote_perms) {
1109                         free_rmtperm_hash(lli->lli_remote_perms);
1110                         lli->lli_remote_perms = NULL;
1111                 }
1112         }
1113 #ifdef CONFIG_FS_POSIX_ACL
1114         else if (lli->lli_posix_acl) {
1115                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1116                 LASSERT(lli->lli_remote_perms == NULL);
1117                 posix_acl_release(lli->lli_posix_acl);
1118                 lli->lli_posix_acl = NULL;
1119         }
1120 #endif
1121         lli->lli_inode_magic = LLI_INODE_DEAD;
1122
1123         ll_clear_inode_capas(inode);
1124         /*
1125          * XXX This has to be done before lsm is freed below, because
1126          * cl_object still uses inode lsm.
1127          */
1128         cl_inode_fini(inode);
1129
1130         if (lli->lli_smd) {
1131                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1132                 lli->lli_smd = NULL;
1133         }
1134
1135
1136         EXIT;
1137 }
1138
1139 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1140                   struct md_open_data **mod)
1141 {
1142         struct lustre_md md;
1143         struct ll_sb_info *sbi = ll_i2sbi(inode);
1144         struct ptlrpc_request *request = NULL;
1145         int rc;
1146         ENTRY;
1147
1148         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1149                                      LUSTRE_OPC_ANY, NULL);
1150         if (IS_ERR(op_data))
1151                 RETURN(PTR_ERR(op_data));
1152
1153         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1154                         &request, mod);
1155         if (rc) {
1156                 ptlrpc_req_finished(request);
1157                 if (rc == -ENOENT) {
1158                         inode->i_nlink = 0;
1159                         /* Unlinked special device node? Or just a race?
1160                          * Pretend we done everything. */
1161                         if (!S_ISREG(inode->i_mode) &&
1162                             !S_ISDIR(inode->i_mode))
1163                                 rc = inode_setattr(inode, &op_data->op_attr);
1164                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1165                         CERROR("md_setattr fails: rc = %d\n", rc);
1166                 }
1167                 RETURN(rc);
1168         }
1169
1170         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1171                               sbi->ll_md_exp, &md);
1172         if (rc) {
1173                 ptlrpc_req_finished(request);
1174                 RETURN(rc);
1175         }
1176
1177         /* We call inode_setattr to adjust timestamps.
1178          * If there is at least some data in file, we cleared ATTR_SIZE
1179          * above to avoid invoking vmtruncate, otherwise it is important
1180          * to call vmtruncate in inode_setattr to update inode->i_size
1181          * (bug 6196) */
1182         rc = inode_setattr(inode, &op_data->op_attr);
1183
1184         /* Extract epoch data if obtained. */
1185         op_data->op_handle = md.body->handle;
1186         op_data->op_ioepoch = md.body->ioepoch;
1187
1188         ll_update_inode(inode, &md);
1189         ptlrpc_req_finished(request);
1190
1191         RETURN(rc);
1192 }
1193
1194 /* Close IO epoch and send Size-on-MDS attribute update. */
1195 static int ll_setattr_done_writing(struct inode *inode,
1196                                    struct md_op_data *op_data,
1197                                    struct md_open_data *mod)
1198 {
1199         struct ll_inode_info *lli = ll_i2info(inode);
1200         int rc = 0;
1201         ENTRY;
1202
1203         LASSERT(op_data != NULL);
1204         if (!S_ISREG(inode->i_mode))
1205                 RETURN(0);
1206
1207         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1208                op_data->op_ioepoch, PFID(&lli->lli_fid));
1209
1210         op_data->op_flags = MF_EPOCH_CLOSE;
1211         ll_done_writing_attr(inode, op_data);
1212         ll_pack_inode2opdata(inode, op_data, NULL);
1213
1214         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1215         if (rc == -EAGAIN) {
1216                 /* MDS has instructed us to obtain Size-on-MDS attribute
1217                  * from OSTs and send setattr to back to MDS. */
1218                 rc = ll_som_update(inode, op_data);
1219         } else if (rc) {
1220                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1221                        inode->i_ino, rc);
1222         }
1223         RETURN(rc);
1224 }
1225
1226 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1227 {
1228         struct obd_capa *capa;
1229         int rc;
1230
1231         if (attr->ia_valid & ATTR_SIZE)
1232                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1233         else
1234                 capa = ll_mdscapa_get(inode);
1235
1236         rc = cl_setattr_ost(inode, attr, capa);
1237
1238         if (attr->ia_valid & ATTR_SIZE)
1239                 ll_truncate_free_capa(capa);
1240         else
1241                 capa_put(capa);
1242
1243         return rc;
1244 }
1245
1246 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1247  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1248  * keep these values until such a time that objects are allocated for it.
1249  * We do the MDS operations first, as it is checking permissions for us.
1250  * We don't to the MDS RPC if there is nothing that we want to store there,
1251  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1252  * going to do an RPC anyways.
1253  *
1254  * If we are doing a truncate, we will send the mtime and ctime updates
1255  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1256  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1257  * at the same time.
1258  */
1259 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1260 {
1261         struct ll_inode_info *lli = ll_i2info(inode);
1262         struct lov_stripe_md *lsm = lli->lli_smd;
1263         struct md_op_data *op_data = NULL;
1264         struct md_open_data *mod = NULL;
1265         int ia_valid = attr->ia_valid;
1266         int rc = 0, rc1 = 0;
1267         ENTRY;
1268
1269         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1270                attr->ia_valid);
1271         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1272
1273         if (ia_valid & ATTR_SIZE) {
1274                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1275                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1276                                attr->ia_size, ll_file_maxbytes(inode));
1277                         RETURN(-EFBIG);
1278                 }
1279
1280                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1281         }
1282
1283         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1284         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1285                 if (cfs_curproc_fsuid() != inode->i_uid &&
1286                     !cfs_capable(CFS_CAP_FOWNER))
1287                         RETURN(-EPERM);
1288         }
1289
1290         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1291         if (attr->ia_valid & ATTR_CTIME) {
1292                 attr->ia_ctime = CFS_CURRENT_TIME;
1293                 attr->ia_valid |= ATTR_CTIME_SET;
1294         }
1295         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1296                 attr->ia_atime = CFS_CURRENT_TIME;
1297                 attr->ia_valid |= ATTR_ATIME_SET;
1298         }
1299         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1300                 attr->ia_mtime = CFS_CURRENT_TIME;
1301                 attr->ia_valid |= ATTR_MTIME_SET;
1302         }
1303
1304         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1305                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1306                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1307                        cfs_time_current_sec());
1308
1309         /* NB: ATTR_SIZE will only be set after this point if the size
1310          * resides on the MDS, ie, this file has no objects. */
1311         if (lsm)
1312                 attr->ia_valid &= ~ATTR_SIZE;
1313
1314         /* We always do an MDS RPC, even if we're only changing the size;
1315          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1316
1317         OBD_ALLOC_PTR(op_data);
1318         if (op_data == NULL)
1319                 RETURN(-ENOMEM);
1320
1321         UNLOCK_INODE_MUTEX(inode);
1322         if (ia_valid & ATTR_SIZE)
1323                 UP_WRITE_I_ALLOC_SEM(inode);
1324         cfs_down_write(&lli->lli_trunc_sem);
1325         LOCK_INODE_MUTEX(inode);
1326         if (ia_valid & ATTR_SIZE)
1327                 DOWN_WRITE_I_ALLOC_SEM(inode);
1328
1329         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1330
1331         /* Open epoch for truncate. */
1332         if (exp_connect_som(ll_i2mdexp(inode)) &&
1333             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1334                 op_data->op_flags = MF_EPOCH_OPEN;
1335
1336         rc = ll_md_setattr(inode, op_data, &mod);
1337         if (rc)
1338                 GOTO(out, rc);
1339
1340         ll_ioepoch_open(lli, op_data->op_ioepoch);
1341         if (!lsm || !S_ISREG(inode->i_mode)) {
1342                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1343                 GOTO(out, rc = 0);
1344         }
1345
1346         if (ia_valid & ATTR_SIZE)
1347                 attr->ia_valid |= ATTR_SIZE;
1348         if (ia_valid & (ATTR_SIZE |
1349                         ATTR_ATIME | ATTR_ATIME_SET |
1350                         ATTR_MTIME | ATTR_MTIME_SET))
1351                 /* on truncate and utimes send attributes to osts, setting
1352                  * mtime/atime to past will be performed under PW 0:EOF extent
1353                  * lock (new_size:EOF for truncate)
1354                  * it may seem excessive to send mtime/atime updates to osts
1355                  * when not setting times to past, but it is necessary due to
1356                  * possible time de-synchronization */
1357                 rc = ll_setattr_ost(inode, attr);
1358         EXIT;
1359 out:
1360         if (op_data) {
1361                 if (op_data->op_ioepoch)
1362                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1363                 ll_finish_md_op_data(op_data);
1364         }
1365         cfs_up_write(&lli->lli_trunc_sem);
1366         return rc ? rc : rc1;
1367 }
1368
1369 int ll_setattr(struct dentry *de, struct iattr *attr)
1370 {
1371         int mode = de->d_inode->i_mode;
1372
1373         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1374                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1375                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1376
1377         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1378                                (ATTR_SIZE|ATTR_MODE)) &&
1379             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1380              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1381               !(attr->ia_mode & S_ISGID))))
1382                 attr->ia_valid |= ATTR_FORCE;
1383
1384         if ((mode & S_ISUID) &&
1385             !(attr->ia_mode & S_ISUID) &&
1386             !(attr->ia_valid & ATTR_KILL_SUID))
1387                 attr->ia_valid |= ATTR_KILL_SUID;
1388
1389         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1390             !(attr->ia_mode & S_ISGID) &&
1391             !(attr->ia_valid & ATTR_KILL_SGID))
1392                 attr->ia_valid |= ATTR_KILL_SGID;
1393
1394         return ll_setattr_raw(de->d_inode, attr);
1395 }
1396
1397 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1398                        __u64 max_age, __u32 flags)
1399 {
1400         struct ll_sb_info *sbi = ll_s2sbi(sb);
1401         struct obd_statfs obd_osfs;
1402         int rc;
1403         ENTRY;
1404
1405         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1406         if (rc) {
1407                 CERROR("md_statfs fails: rc = %d\n", rc);
1408                 RETURN(rc);
1409         }
1410
1411         osfs->os_type = sb->s_magic;
1412
1413         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1414                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1415
1416         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1417                 flags |= OBD_STATFS_NODELAY;
1418
1419         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1420                               &obd_osfs, max_age, flags);
1421         if (rc) {
1422                 CERROR("obd_statfs fails: rc = %d\n", rc);
1423                 RETURN(rc);
1424         }
1425
1426         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1427                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1428                obd_osfs.os_files);
1429
1430         osfs->os_bsize = obd_osfs.os_bsize;
1431         osfs->os_blocks = obd_osfs.os_blocks;
1432         osfs->os_bfree = obd_osfs.os_bfree;
1433         osfs->os_bavail = obd_osfs.os_bavail;
1434
1435         /* If we don't have as many objects free on the OST as inodes
1436          * on the MDS, we reduce the total number of inodes to
1437          * compensate, so that the "inodes in use" number is correct.
1438          */
1439         if (obd_osfs.os_ffree < osfs->os_ffree) {
1440                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1441                         obd_osfs.os_ffree;
1442                 osfs->os_ffree = obd_osfs.os_ffree;
1443         }
1444
1445         RETURN(rc);
1446 }
1447 #ifndef HAVE_STATFS_DENTRY_PARAM
1448 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1449 {
1450 #else
1451 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1452 {
1453         struct super_block *sb = de->d_sb;
1454 #endif
1455         struct obd_statfs osfs;
1456         int rc;
1457
1458         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1459         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1460
1461         /* Some amount of caching on the client is allowed */
1462         rc = ll_statfs_internal(sb, &osfs,
1463                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1464                                 0);
1465         if (rc)
1466                 return rc;
1467
1468         statfs_unpack(sfs, &osfs);
1469
1470         /* We need to downshift for all 32-bit kernels, because we can't
1471          * tell if the kernel is being called via sys_statfs64() or not.
1472          * Stop before overflowing f_bsize - in which case it is better
1473          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1474         if (sizeof(long) < 8) {
1475                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1476                         sfs->f_bsize <<= 1;
1477
1478                         osfs.os_blocks >>= 1;
1479                         osfs.os_bfree >>= 1;
1480                         osfs.os_bavail >>= 1;
1481                 }
1482         }
1483
1484         sfs->f_blocks = osfs.os_blocks;
1485         sfs->f_bfree = osfs.os_bfree;
1486         sfs->f_bavail = osfs.os_bavail;
1487
1488         return 0;
1489 }
1490
1491 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1492 {
1493         struct ll_inode_info *lli;
1494         struct lov_stripe_md *lsm;
1495
1496         lli = ll_i2info(inode);
1497         LASSERT(lli->lli_size_sem_owner != current);
1498         cfs_down(&lli->lli_size_sem);
1499         LASSERT(lli->lli_size_sem_owner == NULL);
1500         lli->lli_size_sem_owner = current;
1501         lsm = lli->lli_smd;
1502         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1503                  lsm, lock_lsm);
1504         if (lock_lsm)
1505                 lov_stripe_lock(lsm);
1506 }
1507
1508 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1509 {
1510         struct ll_inode_info *lli;
1511         struct lov_stripe_md *lsm;
1512
1513         lli = ll_i2info(inode);
1514         lsm = lli->lli_smd;
1515         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1516                  lsm, unlock_lsm);
1517         if (unlock_lsm)
1518                 lov_stripe_unlock(lsm);
1519         LASSERT(lli->lli_size_sem_owner == current);
1520         lli->lli_size_sem_owner = NULL;
1521         cfs_up(&lli->lli_size_sem);
1522 }
1523
1524 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1525 {
1526         struct ll_inode_info *lli = ll_i2info(inode);
1527         struct mdt_body *body = md->body;
1528         struct lov_stripe_md *lsm = md->lsm;
1529         struct ll_sb_info *sbi = ll_i2sbi(inode);
1530
1531         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1532         if (lsm != NULL) {
1533                 cfs_down(&lli->lli_och_sem);
1534                 if (lli->lli_smd == NULL) {
1535                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1536                             lsm->lsm_magic != LOV_MAGIC_V3) {
1537                                 dump_lsm(D_ERROR, lsm);
1538                                 LBUG();
1539                         }
1540                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1541                                lsm, inode->i_ino, inode->i_generation, inode);
1542                         /* cl_inode_init must go before lli_smd or a race is
1543                          * possible where client thinks the file has stripes,
1544                          * but lov raid0 is not setup yet and parallel e.g.
1545                          * glimpse would try to use uninitialized lov */
1546                         cl_inode_init(inode, md);
1547                         cfs_spin_lock(&lli->lli_lock);
1548                         lli->lli_smd = lsm;
1549                         cfs_spin_unlock(&lli->lli_lock);
1550                         cfs_up(&lli->lli_och_sem);
1551                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1552                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1553                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1554                 } else {
1555                         cfs_up(&lli->lli_och_sem);
1556                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1557                                 lli->lli_smd->lsm_stripe_count ==
1558                                 lsm->lsm_stripe_count);
1559                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1560                                 CERROR("lsm mismatch for inode %ld\n",
1561                                        inode->i_ino);
1562                                 CERROR("lli_smd:\n");
1563                                 dump_lsm(D_ERROR, lli->lli_smd);
1564                                 CERROR("lsm:\n");
1565                                 dump_lsm(D_ERROR, lsm);
1566                                 LBUG();
1567                         }
1568                 }
1569                 if (lli->lli_smd != lsm)
1570                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1571         }
1572
1573         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1574                 if (body->valid & OBD_MD_FLRMTPERM)
1575                         ll_update_remote_perm(inode, md->remote_perm);
1576         }
1577 #ifdef CONFIG_FS_POSIX_ACL
1578         else if (body->valid & OBD_MD_FLACL) {
1579                 cfs_spin_lock(&lli->lli_lock);
1580                 if (lli->lli_posix_acl)
1581                         posix_acl_release(lli->lli_posix_acl);
1582                 lli->lli_posix_acl = md->posix_acl;
1583                 cfs_spin_unlock(&lli->lli_lock);
1584         }
1585 #endif
1586         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1587         inode->i_generation = cl_fid_build_gen(&body->fid1);
1588
1589         if (body->valid & OBD_MD_FLATIME) {
1590                 if (body->atime > LTIME_S(inode->i_atime))
1591                         LTIME_S(inode->i_atime) = body->atime;
1592                 lli->lli_lvb.lvb_atime = body->atime;
1593         }
1594         if (body->valid & OBD_MD_FLMTIME) {
1595                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1596                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1597                                "to "LPU64"\n", inode->i_ino,
1598                                LTIME_S(inode->i_mtime), body->mtime);
1599                         LTIME_S(inode->i_mtime) = body->mtime;
1600                 }
1601                 lli->lli_lvb.lvb_mtime = body->mtime;
1602         }
1603         if (body->valid & OBD_MD_FLCTIME) {
1604                 if (body->ctime > LTIME_S(inode->i_ctime))
1605                         LTIME_S(inode->i_ctime) = body->ctime;
1606                 lli->lli_lvb.lvb_ctime = body->ctime;
1607         }
1608         if (body->valid & OBD_MD_FLMODE)
1609                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1610         if (body->valid & OBD_MD_FLTYPE)
1611                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1612         LASSERT(inode->i_mode != 0);
1613         if (S_ISREG(inode->i_mode)) {
1614                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1615         } else {
1616                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1617         }
1618 #ifdef HAVE_INODE_BLKSIZE
1619         inode->i_blksize = 1<<inode->i_blkbits;
1620 #endif
1621         if (body->valid & OBD_MD_FLUID)
1622                 inode->i_uid = body->uid;
1623         if (body->valid & OBD_MD_FLGID)
1624                 inode->i_gid = body->gid;
1625         if (body->valid & OBD_MD_FLFLAGS)
1626                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1627         if (body->valid & OBD_MD_FLNLINK)
1628                 inode->i_nlink = body->nlink;
1629         if (body->valid & OBD_MD_FLRDEV)
1630                 inode->i_rdev = old_decode_dev(body->rdev);
1631
1632         if (body->valid & OBD_MD_FLID) {
1633                 /* FID shouldn't be changed! */
1634                 if (fid_is_sane(&lli->lli_fid)) {
1635                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1636                                  "Trying to change FID "DFID
1637                                  " to the "DFID", inode %lu/%u(%p)\n",
1638                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1639                                  inode->i_ino, inode->i_generation, inode);
1640                 } else
1641                         lli->lli_fid = body->fid1;
1642         }
1643
1644         LASSERT(fid_seq(&lli->lli_fid) != 0);
1645
1646         if (body->valid & OBD_MD_FLSIZE) {
1647                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1648                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1649                         struct lustre_handle lockh;
1650                         ldlm_mode_t mode;
1651
1652                         /* As it is possible a blocking ast has been processed
1653                          * by this time, we need to check there is an UPDATE
1654                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1655                          * it. */
1656                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1657                                                &lockh);
1658                         if (mode) {
1659                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1660                                                       LLIF_EPOCH_PENDING |
1661                                                       LLIF_SOM_DIRTY)) {
1662                                         CERROR("ino %lu flags %lu still has "
1663                                                "size authority! do not trust "
1664                                                "the size got from MDS\n",
1665                                                inode->i_ino, lli->lli_flags);
1666                                 } else {
1667                                         /* Use old size assignment to avoid
1668                                          * deadlock bz14138 & bz14326 */
1669                                         inode->i_size = body->size;
1670                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1671                                 }
1672                                 ldlm_lock_decref(&lockh, mode);
1673                         }
1674                 } else {
1675                         /* Use old size assignment to avoid
1676                          * deadlock bz14138 & bz14326 */
1677                         inode->i_size = body->size;
1678
1679                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1680                                inode->i_ino, (unsigned long long)body->size);
1681                 }
1682
1683                 if (body->valid & OBD_MD_FLBLOCKS)
1684                         inode->i_blocks = body->blocks;
1685         }
1686
1687         if (body->valid & OBD_MD_FLMDSCAPA) {
1688                 LASSERT(md->mds_capa);
1689                 ll_add_capa(inode, md->mds_capa);
1690         }
1691         if (body->valid & OBD_MD_FLOSSCAPA) {
1692                 LASSERT(md->oss_capa);
1693                 ll_add_capa(inode, md->oss_capa);
1694         }
1695 }
1696
1697 void ll_read_inode2(struct inode *inode, void *opaque)
1698 {
1699         struct lustre_md *md = opaque;
1700         struct ll_inode_info *lli = ll_i2info(inode);
1701         ENTRY;
1702
1703         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1704                inode->i_ino, inode->i_generation, inode);
1705
1706         ll_lli_init(lli);
1707
1708         LASSERT(!lli->lli_smd);
1709
1710         /* Core attributes from the MDS first.  This is a new inode, and
1711          * the VFS doesn't zero times in the core inode so we have to do
1712          * it ourselves.  They will be overwritten by either MDS or OST
1713          * attributes - we just need to make sure they aren't newer. */
1714         LTIME_S(inode->i_mtime) = 0;
1715         LTIME_S(inode->i_atime) = 0;
1716         LTIME_S(inode->i_ctime) = 0;
1717         inode->i_rdev = 0;
1718         ll_update_inode(inode, md);
1719
1720         /* OIDEBUG(inode); */
1721
1722         /* initializing backing dev info. */
1723         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1724
1725
1726         if (S_ISREG(inode->i_mode)) {
1727                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1728                 inode->i_op = &ll_file_inode_operations;
1729                 inode->i_fop = sbi->ll_fop;
1730                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1731                 EXIT;
1732         } else if (S_ISDIR(inode->i_mode)) {
1733                 inode->i_op = &ll_dir_inode_operations;
1734                 inode->i_fop = &ll_dir_operations;
1735                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1736                 EXIT;
1737         } else if (S_ISLNK(inode->i_mode)) {
1738                 inode->i_op = &ll_fast_symlink_inode_operations;
1739                 EXIT;
1740         } else {
1741                 inode->i_op = &ll_special_inode_operations;
1742
1743                 init_special_inode(inode, inode->i_mode,
1744                                    kdev_t_to_nr(inode->i_rdev));
1745
1746                 EXIT;
1747         }
1748 }
1749
1750 void ll_delete_inode(struct inode *inode)
1751 {
1752         struct ll_sb_info *sbi = ll_i2sbi(inode);
1753         int rc;
1754         ENTRY;
1755
1756         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1757         if (rc)
1758                 CERROR("fid_delete() failed, rc %d\n", rc);
1759
1760         truncate_inode_pages(&inode->i_data, 0);
1761
1762         /* Workaround for LU-118 */
1763         if (inode->i_data.nrpages) {
1764                 TREE_READ_LOCK_IRQ(&inode->i_data);
1765                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1766                 LASSERTF(inode->i_data.nrpages == 0,
1767                          "inode=%lu/%u(%p) nrpages=%lu, see "
1768                          "http://jira.whamcloud.com/browse/LU-118\n",
1769                          inode->i_ino, inode->i_generation, inode,
1770                          inode->i_data.nrpages);
1771         }
1772         /* Workaround end */
1773
1774         clear_inode(inode);
1775
1776         EXIT;
1777 }
1778
1779 int ll_iocontrol(struct inode *inode, struct file *file,
1780                  unsigned int cmd, unsigned long arg)
1781 {
1782         struct ll_sb_info *sbi = ll_i2sbi(inode);
1783         struct ptlrpc_request *req = NULL;
1784         int rc, flags = 0;
1785         ENTRY;
1786
1787         switch(cmd) {
1788         case FSFILT_IOC_GETFLAGS: {
1789                 struct mdt_body *body;
1790                 struct md_op_data *op_data;
1791
1792                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1793                                              0, 0, LUSTRE_OPC_ANY,
1794                                              NULL);
1795                 if (op_data == NULL)
1796                         RETURN(-ENOMEM);
1797
1798                 op_data->op_valid = OBD_MD_FLFLAGS;
1799                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1800                 ll_finish_md_op_data(op_data);
1801                 if (rc) {
1802                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1803                         RETURN(-abs(rc));
1804                 }
1805
1806                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1807
1808                 flags = body->flags;
1809
1810                 ptlrpc_req_finished(req);
1811
1812                 RETURN(put_user(flags, (int *)arg));
1813         }
1814         case FSFILT_IOC_SETFLAGS: {
1815                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1816                 struct obd_info oinfo = { { { 0 } } };
1817                 struct md_op_data *op_data;
1818
1819                 if (get_user(flags, (int *)arg))
1820                         RETURN(-EFAULT);
1821
1822                 oinfo.oi_md = lsm;
1823                 OBDO_ALLOC(oinfo.oi_oa);
1824                 if (!oinfo.oi_oa)
1825                         RETURN(-ENOMEM);
1826
1827                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1828                                              LUSTRE_OPC_ANY, NULL);
1829                 if (IS_ERR(op_data))
1830                         RETURN(PTR_ERR(op_data));
1831
1832                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1833                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1834                 rc = md_setattr(sbi->ll_md_exp, op_data,
1835                                 NULL, 0, NULL, 0, &req, NULL);
1836                 ll_finish_md_op_data(op_data);
1837                 ptlrpc_req_finished(req);
1838                 if (rc) {
1839                         OBDO_FREE(oinfo.oi_oa);
1840                         RETURN(rc);
1841                 }
1842
1843                 if (lsm == NULL) {
1844                         OBDO_FREE(oinfo.oi_oa);
1845                         GOTO(update_cache, rc);
1846                 }
1847
1848                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1849                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1850                 oinfo.oi_oa->o_flags = flags;
1851                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1852                                        OBD_MD_FLGROUP;
1853                 oinfo.oi_capa = ll_mdscapa_get(inode);
1854                 obdo_from_inode(oinfo.oi_oa, inode,
1855                                 &ll_i2info(inode)->lli_fid, 0);
1856                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1857                 capa_put(oinfo.oi_capa);
1858                 OBDO_FREE(oinfo.oi_oa);
1859                 if (rc) {
1860                         if (rc != -EPERM && rc != -EACCES)
1861                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1862                         RETURN(rc);
1863                 }
1864
1865                 EXIT;
1866 update_cache:
1867                 inode->i_flags = ll_ext_to_inode_flags(flags);
1868                 return 0;
1869         }
1870         default:
1871                 RETURN(-ENOSYS);
1872         }
1873
1874         RETURN(0);
1875 }
1876
1877 int ll_flush_ctx(struct inode *inode)
1878 {
1879         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1880
1881         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1882
1883         obd_set_info_async(sbi->ll_md_exp,
1884                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1885                            0, NULL, NULL);
1886         obd_set_info_async(sbi->ll_dt_exp,
1887                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1888                            0, NULL, NULL);
1889         return 0;
1890 }
1891
1892 /* umount -f client means force down, don't save state */
1893 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1894 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1895 {
1896         struct super_block *sb = vfsmnt->mnt_sb;
1897 #else
1898 void ll_umount_begin(struct super_block *sb)
1899 {
1900 #endif
1901         struct lustre_sb_info *lsi = s2lsi(sb);
1902         struct ll_sb_info *sbi = ll_s2sbi(sb);
1903         struct obd_device *obd;
1904         struct obd_ioctl_data *ioc_data;
1905         ENTRY;
1906
1907 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1908         if (!(flags & MNT_FORCE)) {
1909                 EXIT;
1910                 return;
1911         }
1912 #endif
1913
1914         /* Tell the MGC we got umount -f */
1915         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1916
1917         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1918                sb->s_count, atomic_read(&sb->s_active));
1919
1920         obd = class_exp2obd(sbi->ll_md_exp);
1921         if (obd == NULL) {
1922                 CERROR("Invalid MDC connection handle "LPX64"\n",
1923                        sbi->ll_md_exp->exp_handle.h_cookie);
1924                 EXIT;
1925                 return;
1926         }
1927         obd->obd_force = 1;
1928
1929         obd = class_exp2obd(sbi->ll_dt_exp);
1930         if (obd == NULL) {
1931                 CERROR("Invalid LOV connection handle "LPX64"\n",
1932                        sbi->ll_dt_exp->exp_handle.h_cookie);
1933                 EXIT;
1934                 return;
1935         }
1936         obd->obd_force = 1;
1937
1938         OBD_ALLOC_PTR(ioc_data);
1939         if (ioc_data) {
1940                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1941                               sizeof ioc_data, ioc_data, NULL);
1942
1943                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1944                               sizeof ioc_data, ioc_data, NULL);
1945
1946                 OBD_FREE_PTR(ioc_data);
1947         }
1948
1949
1950         /* Really, we'd like to wait until there are no requests outstanding,
1951          * and then continue.  For now, we just invalidate the requests,
1952          * schedule() and sleep one second if needed, and hope.
1953          */
1954         cfs_schedule();
1955 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1956         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1957                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1958                                                    cfs_time_seconds(1));
1959                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1960                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1961                                       "may try to umount it a bit later\n",
1962                                       atomic_read(&vfsmnt->mnt_count));
1963         }
1964 #endif
1965
1966         EXIT;
1967 }
1968
1969 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1970 {
1971         struct ll_sb_info *sbi = ll_s2sbi(sb);
1972         int err;
1973         __u32 read_only;
1974
1975         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1976                 read_only = *flags & MS_RDONLY;
1977                 err = obd_set_info_async(sbi->ll_md_exp,
1978                                          sizeof(KEY_READ_ONLY),
1979                                          KEY_READ_ONLY, sizeof(read_only),
1980                                          &read_only, NULL);
1981                 if (err) {
1982                         CERROR("Failed to change the read-only flag during "
1983                                "remount: %d\n", err);
1984                         return err;
1985                 }
1986
1987                 if (read_only)
1988                         sb->s_flags |= MS_RDONLY;
1989                 else
1990                         sb->s_flags &= ~MS_RDONLY;
1991         }
1992         return 0;
1993 }
1994
1995 int ll_prep_inode(struct inode **inode,
1996                   struct ptlrpc_request *req,
1997                   struct super_block *sb)
1998 {
1999         struct ll_sb_info *sbi = NULL;
2000         struct lustre_md md;
2001         int rc;
2002         ENTRY;
2003
2004         LASSERT(*inode || sb);
2005         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2006         memset(&md, 0, sizeof(struct lustre_md));
2007
2008         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2009                               sbi->ll_md_exp, &md);
2010         if (rc)
2011                 RETURN(rc);
2012
2013         if (*inode) {
2014                 ll_update_inode(*inode, &md);
2015         } else {
2016                 LASSERT(sb != NULL);
2017
2018                 /*
2019                  * At this point server returns to client's same fid as client
2020                  * generated for creating. So using ->fid1 is okay here.
2021                  */
2022                 LASSERT(fid_is_sane(&md.body->fid1));
2023
2024                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2025                 if (*inode == NULL || IS_ERR(*inode)) {
2026                         if (md.lsm)
2027                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2028 #ifdef CONFIG_FS_POSIX_ACL
2029                         if (md.posix_acl) {
2030                                 posix_acl_release(md.posix_acl);
2031                                 md.posix_acl = NULL;
2032                         }
2033 #endif
2034                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2035                         *inode = NULL;
2036                         CERROR("new_inode -fatal: rc %d\n", rc);
2037                         GOTO(out, rc);
2038                 }
2039         }
2040
2041 out:
2042         md_free_lustre_md(sbi->ll_md_exp, &md);
2043         RETURN(rc);
2044 }
2045
2046 int ll_obd_statfs(struct inode *inode, void *arg)
2047 {
2048         struct ll_sb_info *sbi = NULL;
2049         struct obd_export *exp;
2050         char *buf = NULL;
2051         struct obd_ioctl_data *data = NULL;
2052         __u32 type;
2053         int len = 0, rc;
2054
2055         if (!inode || !(sbi = ll_i2sbi(inode)))
2056                 GOTO(out_statfs, rc = -EINVAL);
2057
2058         rc = obd_ioctl_getdata(&buf, &len, arg);
2059         if (rc)
2060                 GOTO(out_statfs, rc);
2061
2062         data = (void*)buf;
2063         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2064             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2065                 GOTO(out_statfs, rc = -EINVAL);
2066
2067         if (data->ioc_inllen1 != sizeof(__u32) ||
2068             data->ioc_inllen2 != sizeof(__u32) ||
2069             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2070             data->ioc_plen2 != sizeof(struct obd_uuid))
2071                 GOTO(out_statfs, rc = -EINVAL);
2072
2073         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2074         if (type == LL_STATFS_MDC)
2075                 exp = sbi->ll_md_exp;
2076         else if (type == LL_STATFS_LOV)
2077                 exp = sbi->ll_dt_exp;
2078         else
2079                 GOTO(out_statfs, rc = -ENODEV);
2080
2081         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2082         if (rc)
2083                 GOTO(out_statfs, rc);
2084 out_statfs:
2085         if (buf)
2086                 obd_ioctl_freedata(buf, len);
2087         return rc;
2088 }
2089
2090 int ll_process_config(struct lustre_cfg *lcfg)
2091 {
2092         char *ptr;
2093         void *sb;
2094         struct lprocfs_static_vars lvars;
2095         unsigned long x;
2096         int rc = 0;
2097
2098         lprocfs_llite_init_vars(&lvars);
2099
2100         /* The instance name contains the sb: lustre-client-aacfe000 */
2101         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2102         if (!ptr || !*(++ptr))
2103                 return -EINVAL;
2104         if (sscanf(ptr, "%lx", &x) != 1)
2105                 return -EINVAL;
2106         sb = (void *)x;
2107         /* This better be a real Lustre superblock! */
2108         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2109
2110         /* Note we have not called client_common_fill_super yet, so
2111            proc fns must be able to handle that! */
2112         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2113                                       lcfg, sb);
2114         if (rc > 0)
2115                 rc = 0;
2116         return(rc);
2117 }
2118
2119 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2120 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2121                                        struct inode *i1, struct inode *i2,
2122                                        const char *name, int namelen,
2123                                        int mode, __u32 opc, void *data)
2124 {
2125         LASSERT(i1 != NULL);
2126
2127         if (namelen > ll_i2sbi(i1)->ll_namelen)
2128                 return ERR_PTR(-ENAMETOOLONG);
2129
2130         if (op_data == NULL)
2131                 OBD_ALLOC_PTR(op_data);
2132
2133         if (op_data == NULL)
2134                 return ERR_PTR(-ENOMEM);
2135
2136         ll_i2gids(op_data->op_suppgids, i1, i2);
2137         op_data->op_fid1 = *ll_inode2fid(i1);
2138         op_data->op_capa1 = ll_mdscapa_get(i1);
2139
2140         if (i2) {
2141                 op_data->op_fid2 = *ll_inode2fid(i2);
2142                 op_data->op_capa2 = ll_mdscapa_get(i2);
2143         } else {
2144                 fid_zero(&op_data->op_fid2);
2145                 op_data->op_capa2 = NULL;
2146         }
2147
2148         op_data->op_name = name;
2149         op_data->op_namelen = namelen;
2150         op_data->op_mode = mode;
2151         op_data->op_mod_time = cfs_time_current_sec();
2152         op_data->op_fsuid = cfs_curproc_fsuid();
2153         op_data->op_fsgid = cfs_curproc_fsgid();
2154         op_data->op_cap = cfs_curproc_cap_pack();
2155         op_data->op_bias = MDS_CHECK_SPLIT;
2156         op_data->op_opc = opc;
2157         op_data->op_mds = 0;
2158         op_data->op_data = data;
2159
2160         /* If the file is being opened after mknod() (normally due to NFS)
2161          * try to use the default stripe data from parent directory for
2162          * allocating OST objects.  Try to pass the parent FID to MDS. */
2163         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2164             ll_i2info(i2)->lli_smd == NULL) {
2165                 struct ll_inode_info *lli = ll_i2info(i2);
2166
2167                 cfs_spin_lock(&lli->lli_lock);
2168                 if (likely(lli->lli_smd == NULL &&
2169                            !fid_is_zero(&lli->lli_pfid)))
2170                         op_data->op_fid1 = lli->lli_pfid;
2171                 cfs_spin_unlock(&lli->lli_lock);
2172                 /** We ignore parent's capability temporary. */
2173         }
2174
2175         return op_data;
2176 }
2177
2178 void ll_finish_md_op_data(struct md_op_data *op_data)
2179 {
2180         capa_put(op_data->op_capa1);
2181         capa_put(op_data->op_capa2);
2182         OBD_FREE_PTR(op_data);
2183 }
2184
2185 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2186 {
2187         struct ll_sb_info *sbi;
2188
2189         LASSERT((seq != NULL) && (vfs != NULL));
2190         sbi = ll_s2sbi(vfs->mnt_sb);
2191
2192         if (sbi->ll_flags & LL_SBI_NOLCK)
2193                 seq_puts(seq, ",nolock");
2194
2195         if (sbi->ll_flags & LL_SBI_FLOCK)
2196                 seq_puts(seq, ",flock");
2197
2198         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2199                 seq_puts(seq, ",localflock");
2200
2201         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2202                 seq_puts(seq, ",user_xattr");
2203
2204         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2205                 seq_puts(seq, ",lazystatfs");
2206
2207         RETURN(0);
2208 }