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  2008 Sun Microsystems, Inc. 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/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 CFS_LIST_HEAD(ll_super_blocks);
  63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) cfs_ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC(sbi, sizeof(*sbi));
  82         if (!sbi)
  83                 RETURN(NULL);
  84
  85         cfs_spin_lock_init(&sbi->ll_lock);
  86         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  87         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  88         cfs_spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         cfs_spin_lock(&ll_sb_lock);
 116         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         cfs_spin_unlock(&ll_sb_lock);
 118
 119 #ifdef ENABLE_LLITE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127 #ifdef HAVE_EXPORT___IGET
 128         CFS_INIT_LIST_HEAD(&sbi->ll_deathrow);
 129         cfs_spin_lock_init(&sbi->ll_deathrow_lock);
 130 #endif
 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
 141         RETURN(sbi);
 142 }
 143
 144 void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 cfs_spin_lock(&ll_sb_lock);
 151                 cfs_list_del(&sbi->ll_list);
 152                 cfs_spin_unlock(&ll_sb_lock);
 153                 OBD_FREE(sbi, sizeof(*sbi));
 154         }
 155         EXIT;
 156 }
 157
 158 static struct dentry_operations ll_d_root_ops = {
 159         .d_compare = ll_dcompare,
 160         .d_revalidate = ll_revalidate_nd,
 161 };
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 164 {
 165         struct inode *root = 0;
 166         struct ll_sb_info *sbi = ll_s2sbi(sb);
 167         struct obd_device *obd;
 168         struct obd_capa *oc = NULL;
 169         struct obd_statfs osfs;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         obd_valid valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         if (proc_lustre_fs_root) {
 190                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 191                                                   dt, md);
 192                 if (err < 0)
 193                         CERROR("could not register mount in /proc/fs/lustre\n");
 194         }
 195
 196         /* indicate the features supported by this client */
 197         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 198                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 199                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 200                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 201                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 202                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 203                                   OBD_CONNECT_VBR      | OBD_CONNECT_SOM;
 204
 205 #ifdef HAVE_LRU_RESIZE_SUPPORT
 206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208 #endif
 209 #ifdef CONFIG_FS_POSIX_ACL
 210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 211 #endif
 212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 213         data->ocd_version = LUSTRE_VERSION_CODE;
 214
 215         if (sb->s_flags & MS_RDONLY)
 216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 219
 220 #ifdef HAVE_MS_FLOCK_LOCK
 221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 222         sb->s_flags |= MS_FLOCK_LOCK;
 223 #endif
 224
 225         if (sbi->ll_flags & LL_SBI_FLOCK)
 226                 sbi->ll_fop = &ll_file_operations_flock;
 227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 228                 sbi->ll_fop = &ll_file_operations;
 229         else
 230                 sbi->ll_fop = &ll_file_operations_noflock;
 231
 232         /* real client */
 233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 235                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 236
 237         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 238         if (err == -EBUSY) {
 239                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 240                                    "recovery, of which this client is not a "
 241                                    "part. Please wait for recovery to complete,"
 242                                    " abort, or time out.\n", md);
 243                 GOTO(out, err);
 244         } else if (err) {
 245                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 246                 GOTO(out, err);
 247         }
 248
 249         err = obd_fid_init(sbi->ll_md_exp);
 250         if (err) {
 251                 CERROR("Can't init metadata layer FID infrastructure, "
 252                        "rc %d\n", err);
 253                 GOTO(out_md, err);
 254         }
 255
 256         err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
 257         if (err)
 258                 GOTO(out_md_fid, err);
 259
 260         size = sizeof(*data);
 261         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 262                            KEY_CONN_DATA,  &size, data, NULL);
 263         if (err) {
 264                 CERROR("Get connect data failed: %d \n", err);
 265                 GOTO(out_md, err);
 266         }
 267
 268         LASSERT(osfs.os_bsize);
 269         sb->s_blocksize = osfs.os_bsize;
 270         sb->s_blocksize_bits = log2(osfs.os_bsize);
 271         sb->s_magic = LL_SUPER_MAGIC;
 272
 273         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 274          *         retval = in_file->f_op->sendfile(...);
 275          *         if (*ppos > max)
 276          *                 retval = -EOVERFLOW;
 277          *
 278          * it will check if *ppos is greater than max. However, max equals to
 279          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 280          * has been defined as a signed long long integer in linux kernel. */
 281 #if BITS_PER_LONG == 64
 282         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 283 #else
 284         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 285 #endif
 286         sbi->ll_namelen = osfs.os_namelen;
 287         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 288
 289         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 290             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 291                 LCONSOLE_INFO("Disabling user_xattr feature because "
 292                               "it is not supported on the server\n");
 293                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 294         }
 295
 296         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 297 #ifdef MS_POSIXACL
 298                 sb->s_flags |= MS_POSIXACL;
 299 #endif
 300                 sbi->ll_flags |= LL_SBI_ACL;
 301         } else {
 302                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 303 #ifdef MS_POSIXACL
 304                 sb->s_flags &= ~MS_POSIXACL;
 305 #endif
 306                 sbi->ll_flags &= ~LL_SBI_ACL;
 307         }
 308
 309         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 310                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 311                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 312                         LCONSOLE_INFO("client is set as remote by default.\n");
 313                 }
 314         } else {
 315                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 316                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 317                         LCONSOLE_INFO("client claims to be remote, but server "
 318                                       "rejected, forced to be local.\n");
 319                 }
 320         }
 321
 322         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 323                 LCONSOLE_INFO("client enabled MDS capability!\n");
 324                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 325         }
 326
 327         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 328                 LCONSOLE_INFO("client enabled OSS capability!\n");
 329                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 330         }
 331
 332         obd = class_name2obd(dt);
 333         if (!obd) {
 334                 CERROR("DT %s: not setup or attached\n", dt);
 335                 GOTO(out_md_fid, err = -ENODEV);
 336         }
 337
 338         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 339                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 340                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 341                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 342                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 343                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 344                                   OBD_CONNECT_SOM;
 345
 346         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 347                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 348                  * disabled by default, because it can still be enabled on the
 349                  * fly via /proc. As a consequence, we still need to come to an
 350                  * agreement on the supported algorithms at connect time */
 351                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 352
 353                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 354                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 355                 else
 356                         /* send the list of supported checksum types */
 357                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 358         }
 359
 360 #ifdef HAVE_LRU_RESIZE_SUPPORT
 361         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 362 #endif
 363         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 364                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 365
 366         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 367                "ocd_grant: %d\n", data->ocd_connect_flags,
 368                data->ocd_version, data->ocd_grant);
 369
 370         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 371         obd->obd_upcall.onu_upcall = cl_ocd_update;
 372         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 373
 374         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 375         if (err == -EBUSY) {
 376                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 377                                    "recovery, of which this client is not a "
 378                                    "part.  Please wait for recovery to "
 379                                    "complete, abort, or time out.\n", dt);
 380                 GOTO(out_md_fid, err);
 381         } else if (err) {
 382                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 383                 GOTO(out_md_fid, err);
 384         }
 385
 386         err = obd_fid_init(sbi->ll_dt_exp);
 387         if (err) {
 388                 CERROR("Can't init data layer FID infrastructure, "
 389                        "rc %d\n", err);
 390                 GOTO(out_dt, err);
 391         }
 392
 393         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 394         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 395         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 396         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 397         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 398
 399         fid_zero(&sbi->ll_root_fid);
 400         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 401         if (err) {
 402                 CERROR("cannot mds_connect: rc = %d\n", err);
 403                 GOTO(out_lock_cn_cb, err);
 404         }
 405         if (!fid_is_sane(&sbi->ll_root_fid)) {
 406                 CERROR("Invalid root fid during mount\n");
 407                 GOTO(out_lock_cn_cb, err = -EINVAL);
 408         }
 409         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 410
 411         sb->s_op = &lustre_super_operations;
 412         sb->s_export_op = &lustre_export_operations;
 413
 414         /* make root inode
 415          * XXX: move this to after cbd setup? */
 416         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 417         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 418                 valid |= OBD_MD_FLRMTPERM;
 419         else if (sbi->ll_flags & LL_SBI_ACL)
 420                 valid |= OBD_MD_FLACL;
 421
 422         OBD_ALLOC_PTR(op_data);
 423         if (op_data == NULL)
 424                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 425
 426         op_data->op_fid1 = sbi->ll_root_fid;
 427         op_data->op_mode = 0;
 428         op_data->op_capa1 = oc;
 429         op_data->op_valid = valid;
 430
 431         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 432         if (oc)
 433                 capa_put(oc);
 434         OBD_FREE_PTR(op_data);
 435         if (err) {
 436                 CERROR("md_getattr failed for root: rc = %d\n", err);
 437                 GOTO(out_lock_cn_cb, err);
 438         }
 439         memset(&lmd, 0, sizeof(lmd));
 440         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 441                                sbi->ll_md_exp, &lmd);
 442         if (err) {
 443                 CERROR("failed to understand root inode md: rc = %d\n", err);
 444                 ptlrpc_req_finished (request);
 445                 GOTO(out_lock_cn_cb, err);
 446         }
 447
 448         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 449         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
 450         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 451         ptlrpc_req_finished(request);
 452
 453         if (root == NULL || IS_ERR(root)) {
 454                 if (lmd.lsm)
 455                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 456 #ifdef CONFIG_FS_POSIX_ACL
 457                 if (lmd.posix_acl) {
 458                         posix_acl_release(lmd.posix_acl);
 459                         lmd.posix_acl = NULL;
 460                 }
 461 #endif
 462                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 463                 root = NULL;
 464                 CERROR("lustre_lite: bad iget4 for root\n");
 465                 GOTO(out_root, err);
 466         }
 467
 468         err = ll_close_thread_start(&sbi->ll_lcq);
 469         if (err) {
 470                 CERROR("cannot start close thread: rc %d\n", err);
 471                 GOTO(out_root, err);
 472         }
 473
 474 #ifdef CONFIG_FS_POSIX_ACL
 475         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 476                 rct_init(&sbi->ll_rct);
 477                 et_init(&sbi->ll_et);
 478         }
 479 #endif
 480
 481         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 482         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 483                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 484                                  NULL);
 485         cl_sb_init(sb);
 486
 487         sb->s_root = d_alloc_root(root);
 488         if (data != NULL)
 489                 OBD_FREE(data, sizeof(*data));
 490
 491         sb->s_root->d_op = &ll_d_root_ops;
 492
 493         sbi->ll_sdev_orig = sb->s_dev;
 494
 495         /* We set sb->s_dev equal on all lustre clients in order to support
 496          * NFS export clustering.  NFSD requires that the FSID be the same
 497          * on all clients. */
 498         /* s_dev is also used in lt_compare() to compare two fs, but that is
 499          * only a node-local comparison. */
 500         uuid = obd_get_uuid(sbi->ll_md_exp);
 501         if (uuid != NULL)
 502                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 503
 504         RETURN(err);
 505 out_root:
 506         if (root)
 507                 iput(root);
 508 out_lock_cn_cb:
 509         obd_fid_fini(sbi->ll_dt_exp);
 510 out_dt:
 511         obd_disconnect(sbi->ll_dt_exp);
 512         sbi->ll_dt_exp = NULL;
 513 out_md_fid:
 514         obd_fid_fini(sbi->ll_md_exp);
 515 out_md:
 516         obd_disconnect(sbi->ll_md_exp);
 517         sbi->ll_md_exp = NULL;
 518 out:
 519         if (data != NULL)
 520                 OBD_FREE_PTR(data);
 521         lprocfs_unregister_mountpoint(sbi);
 522         return err;
 523 }
 524
 525 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 526 {
 527         int size, rc;
 528
 529         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 530         size = sizeof(int);
 531         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 532                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 533         if (rc)
 534                 CERROR("Get max mdsize error rc %d \n", rc);
 535
 536         RETURN(rc);
 537 }
 538
 539 void ll_dump_inode(struct inode *inode)
 540 {
 541         struct list_head *tmp;
 542         int dentry_count = 0;
 543
 544         LASSERT(inode != NULL);
 545
 546         list_for_each(tmp, &inode->i_dentry)
 547                 dentry_count++;
 548
 549         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 550                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 551                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 552 }
 553
 554 void lustre_dump_dentry(struct dentry *dentry, int recur)
 555 {
 556         struct list_head *tmp;
 557         int subdirs = 0;
 558
 559         LASSERT(dentry != NULL);
 560
 561         list_for_each(tmp, &dentry->d_subdirs)
 562                 subdirs++;
 563
 564         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 565                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 566                dentry->d_name.len, dentry->d_name.name,
 567                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 568                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 569                dentry->d_flags, dentry->d_fsdata, subdirs);
 570         if (dentry->d_inode != NULL)
 571                 ll_dump_inode(dentry->d_inode);
 572
 573         if (recur == 0)
 574                 return;
 575
 576        list_for_each(tmp, &dentry->d_subdirs) {
 577                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 578                 lustre_dump_dentry(d, recur - 1);
 579         }
 580 }
 581
 582 #ifdef HAVE_EXPORT___IGET
 583 static void prune_dir_dentries(struct inode *inode)
 584 {
 585         struct dentry *dentry, *prev = NULL;
 586
 587         /* due to lustre specific logic, a directory
 588          * can have few dentries - a bug from VFS POV */
 589 restart:
 590         spin_lock(&dcache_lock);
 591         if (!list_empty(&inode->i_dentry)) {
 592                 dentry = list_entry(inode->i_dentry.prev,
 593                                     struct dentry, d_alias);
 594                 /* in order to prevent infinite loops we
 595                  * break if previous dentry is busy */
 596                 if (dentry != prev) {
 597                         prev = dentry;
 598                         dget_locked(dentry);
 599                         spin_unlock(&dcache_lock);
 600
 601                         /* try to kill all child dentries */
 602                         lock_dentry(dentry);
 603                         shrink_dcache_parent(dentry);
 604                         unlock_dentry(dentry);
 605                         dput(dentry);
 606
 607                         /* now try to get rid of current dentry */
 608                         d_prune_aliases(inode);
 609                         goto restart;
 610                 }
 611         }
 612         spin_unlock(&dcache_lock);
 613 }
 614
 615 static void prune_deathrow_one(struct ll_inode_info *lli)
 616 {
 617         struct inode *inode = ll_info2i(lli);
 618
 619         /* first, try to drop any dentries - they hold a ref on the inode */
 620         if (S_ISDIR(inode->i_mode))
 621                 prune_dir_dentries(inode);
 622         else
 623                 d_prune_aliases(inode);
 624
 625
 626         /* if somebody still uses it, leave it */
 627         LASSERT(atomic_read(&inode->i_count) > 0);
 628         if (atomic_read(&inode->i_count) > 1)
 629                 goto out;
 630
 631         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 632                inode->i_ino,inode->i_generation,
 633                atomic_read(&inode->i_count));
 634
 635         /* seems nobody uses it anymore */
 636         inode->i_nlink = 0;
 637
 638 out:
 639         iput(inode);
 640         return;
 641 }
 642
 643 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 644 {
 645         struct ll_inode_info *lli;
 646         int empty;
 647
 648         do {
 649                 if (need_resched() && try)
 650                         break;
 651
 652                 if (try) {
 653                         if (!cfs_spin_trylock(&sbi->ll_deathrow_lock))
 654                                 break;
 655                 } else {
 656                         cfs_spin_lock(&sbi->ll_deathrow_lock);
 657                 }
 658
 659                 empty = 1;
 660                 lli = NULL;
 661                 if (!cfs_list_empty(&sbi->ll_deathrow)) {
 662                         lli = cfs_list_entry(sbi->ll_deathrow.next,
 663                                              struct ll_inode_info,
 664                                              lli_dead_list);
 665                         cfs_list_del_init(&lli->lli_dead_list);
 666                         if (!cfs_list_empty(&sbi->ll_deathrow))
 667                                 empty = 0;
 668                 }
 669                 cfs_spin_unlock(&sbi->ll_deathrow_lock);
 670
 671                 if (lli)
 672                         prune_deathrow_one(lli);
 673
 674         } while (empty == 0);
 675 }
 676 #else /* !HAVE_EXPORT___IGET */
 677 #define prune_deathrow(sbi, try) do {} while (0)
 678 #endif /* HAVE_EXPORT___IGET */
 679
 680 void client_common_put_super(struct super_block *sb)
 681 {
 682         struct ll_sb_info *sbi = ll_s2sbi(sb);
 683         ENTRY;
 684
 685 #ifdef CONFIG_FS_POSIX_ACL
 686         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 687                 et_fini(&sbi->ll_et);
 688                 rct_fini(&sbi->ll_rct);
 689         }
 690 #endif
 691
 692         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 693
 694         ll_close_thread_shutdown(sbi->ll_lcq);
 695
 696         cl_sb_fini(sb);
 697
 698         /* destroy inodes in deathrow */
 699         prune_deathrow(sbi, 0);
 700
 701         cfs_list_del(&sbi->ll_conn_chain);
 702
 703         obd_fid_fini(sbi->ll_dt_exp);
 704         obd_disconnect(sbi->ll_dt_exp);
 705         sbi->ll_dt_exp = NULL;
 706
 707         lprocfs_unregister_mountpoint(sbi);
 708
 709         obd_fid_fini(sbi->ll_md_exp);
 710         obd_disconnect(sbi->ll_md_exp);
 711         sbi->ll_md_exp = NULL;
 712
 713         EXIT;
 714 }
 715
 716 void ll_kill_super(struct super_block *sb)
 717 {
 718         struct ll_sb_info *sbi;
 719
 720         ENTRY;
 721
 722         /* not init sb ?*/
 723         if (!(sb->s_flags & MS_ACTIVE))
 724                 return;
 725
 726         sbi = ll_s2sbi(sb);
 727         /* we need restore s_dev from changed for clustred NFS before put_super
 728          * because new kernels have cached s_dev and change sb->s_dev in
 729          * put_super not affected real removing devices */
 730         if (sbi)
 731                 sb->s_dev = sbi->ll_sdev_orig;
 732         EXIT;
 733 }
 734
 735 char *ll_read_opt(const char *opt, char *data)
 736 {
 737         char *value;
 738         char *retval;
 739         ENTRY;
 740
 741         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 742         if (strncmp(opt, data, strlen(opt)))
 743                 RETURN(NULL);
 744         if ((value = strchr(data, '=')) == NULL)
 745                 RETURN(NULL);
 746
 747         value++;
 748         OBD_ALLOC(retval, strlen(value) + 1);
 749         if (!retval) {
 750                 CERROR("out of memory!\n");
 751                 RETURN(NULL);
 752         }
 753
 754         memcpy(retval, value, strlen(value)+1);
 755         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 756         RETURN(retval);
 757 }
 758
 759 static inline int ll_set_opt(const char *opt, char *data, int fl)
 760 {
 761         if (strncmp(opt, data, strlen(opt)) != 0)
 762                 return(0);
 763         else
 764                 return(fl);
 765 }
 766
 767 /* non-client-specific mount options are parsed in lmd_parse */
 768 static int ll_options(char *options, int *flags)
 769 {
 770         int tmp;
 771         char *s1 = options, *s2;
 772         ENTRY;
 773
 774         if (!options)
 775                 RETURN(0);
 776
 777         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 778
 779         while (*s1) {
 780                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 781                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 787                 if (tmp) {
 788                         *flags |= tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 797                 if (tmp) {
 798                         *flags &= ~tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 812                 if (tmp) {
 813                         /* Ignore deprecated mount option.  The client will
 814                          * always try to mount with ACL support, whether this
 815                          * is used depends on whether server supports it. */
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 819                 if (tmp) {
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827
 828                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858
 859                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 860                                    s1);
 861                 RETURN(-EINVAL);
 862
 863 next:
 864                 /* Find next opt */
 865                 s2 = strchr(s1, ',');
 866                 if (s2 == NULL)
 867                         break;
 868                 s1 = s2 + 1;
 869         }
 870         RETURN(0);
 871 }
 872
 873 void ll_lli_init(struct ll_inode_info *lli)
 874 {
 875         lli->lli_inode_magic = LLI_INODE_MAGIC;
 876         cfs_sema_init(&lli->lli_size_sem, 1);
 877         cfs_sema_init(&lli->lli_write_sem, 1);
 878         cfs_sema_init(&lli->lli_trunc_sem, 1);
 879         lli->lli_flags = 0;
 880         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 881         cfs_spin_lock_init(&lli->lli_lock);
 882         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 883         lli->lli_inode_magic = LLI_INODE_MAGIC;
 884         cfs_sema_init(&lli->lli_och_sem, 1);
 885         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 886         lli->lli_mds_exec_och = NULL;
 887         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 888         lli->lli_open_fd_exec_count = 0;
 889         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 890         lli->lli_remote_perms = NULL;
 891         lli->lli_rmtperm_utime = 0;
 892         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 893         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 894 }
 895
 896 int ll_fill_super(struct super_block *sb)
 897 {
 898         struct lustre_profile *lprof;
 899         struct lustre_sb_info *lsi = s2lsi(sb);
 900         struct ll_sb_info *sbi;
 901         char  *dt = NULL, *md = NULL;
 902         char  *profilenm = get_profile_name(sb);
 903         struct config_llog_instance cfg = {0, };
 904         char   ll_instance[sizeof(sb) * 2 + 1];
 905         int    err;
 906         ENTRY;
 907
 908         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 909
 910         cfs_module_get();
 911
 912         /* client additional sb info */
 913         lsi->lsi_llsbi = sbi = ll_init_sbi();
 914         if (!sbi) {
 915                 cfs_module_put(THIS_MODULE);
 916                 RETURN(-ENOMEM);
 917         }
 918
 919         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 920         if (err)
 921                 GOTO(out_free, err);
 922
 923         /* Generate a string unique to this super, in case some joker tries
 924            to mount the same fs at two mount points.
 925            Use the address of the super itself.*/
 926         sprintf(ll_instance, "%p", sb);
 927         cfg.cfg_instance = ll_instance;
 928         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 929
 930         /* set up client obds */
 931         err = lustre_process_log(sb, profilenm, &cfg);
 932         if (err < 0) {
 933                 CERROR("Unable to process log: %d\n", err);
 934                 GOTO(out_free, err);
 935         }
 936
 937         lprof = class_get_profile(profilenm);
 938         if (lprof == NULL) {
 939                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 940                                    " read from the MGS.  Does that filesystem "
 941                                    "exist?\n", profilenm);
 942                 GOTO(out_free, err = -EINVAL);
 943         }
 944         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 945                lprof->lp_md, lprof->lp_dt);
 946
 947         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 948                   strlen(ll_instance) + 2);
 949         if (!dt)
 950                 GOTO(out_free, err = -ENOMEM);
 951         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 952
 953         OBD_ALLOC(md, strlen(lprof->lp_md) +
 954                   strlen(ll_instance) + 2);
 955         if (!md)
 956                 GOTO(out_free, err = -ENOMEM);
 957         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 958
 959         /* connections, registrations, sb setup */
 960         err = client_common_fill_super(sb, md, dt);
 961
 962 out_free:
 963         if (md)
 964                 OBD_FREE(md, strlen(md) + 1);
 965         if (dt)
 966                 OBD_FREE(dt, strlen(dt) + 1);
 967         if (err)
 968                 ll_put_super(sb);
 969         else
 970                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 971
 972         RETURN(err);
 973 } /* ll_fill_super */
 974
 975
 976 void lu_context_keys_dump(void);
 977
 978 void ll_put_super(struct super_block *sb)
 979 {
 980         struct config_llog_instance cfg;
 981         char   ll_instance[sizeof(sb) * 2 + 1];
 982         struct obd_device *obd;
 983         struct lustre_sb_info *lsi = s2lsi(sb);
 984         struct ll_sb_info *sbi = ll_s2sbi(sb);
 985         char *profilenm = get_profile_name(sb);
 986         int force = 1, next;
 987         ENTRY;
 988
 989         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 990
 991         ll_print_capa_stat(sbi);
 992
 993         sprintf(ll_instance, "%p", sb);
 994         cfg.cfg_instance = ll_instance;
 995         lustre_end_log(sb, NULL, &cfg);
 996
 997         if (sbi->ll_md_exp) {
 998                 obd = class_exp2obd(sbi->ll_md_exp);
 999                 if (obd)
1000                         force = obd->obd_force;
1001         }
1002
1003         /* We need to set force before the lov_disconnect in
1004            lustre_common_put_super, since l_d cleans up osc's as well. */
1005         if (force) {
1006                 next = 0;
1007                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1008                                                      &next)) != NULL) {
1009                         obd->obd_force = force;
1010                 }
1011         }
1012
1013         if (sbi->ll_lcq) {
1014                 /* Only if client_common_fill_super succeeded */
1015                 client_common_put_super(sb);
1016         }
1017
1018         next = 0;
1019         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1020                 class_manual_cleanup(obd);
1021         }
1022
1023         if (profilenm)
1024                 class_del_profile(profilenm);
1025
1026         ll_free_sbi(sb);
1027         lsi->lsi_llsbi = NULL;
1028
1029         lustre_common_put_super(sb);
1030
1031         cl_env_cache_purge(~0);
1032
1033         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1034
1035         cfs_module_put(THIS_MODULE);
1036
1037         EXIT;
1038 } /* client_put_super */
1039
1040 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1041 {
1042         struct inode *inode = NULL;
1043         /* NOTE: we depend on atomic igrab() -bzzz */
1044         lock_res_and_lock(lock);
1045         if (lock->l_ast_data) {
1046                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1047                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1048                         inode = igrab(lock->l_ast_data);
1049                 } else {
1050                         inode = lock->l_ast_data;
1051                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1052                                                 D_INFO : D_WARNING,
1053                                         lock, __FILE__, __func__, __LINE__,
1054                                         "l_ast_data %p is bogus: magic %08x",
1055                                         lock->l_ast_data, lli->lli_inode_magic);
1056                         inode = NULL;
1057                 }
1058         }
1059         unlock_res_and_lock(lock);
1060         return inode;
1061 }
1062
1063 static int null_if_equal(struct ldlm_lock *lock, void *data)
1064 {
1065         if (data == lock->l_ast_data) {
1066                 lock->l_ast_data = NULL;
1067
1068                 if (lock->l_req_mode != lock->l_granted_mode)
1069                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1070         }
1071
1072         return LDLM_ITER_CONTINUE;
1073 }
1074
1075 void ll_clear_inode(struct inode *inode)
1076 {
1077         struct ll_inode_info *lli = ll_i2info(inode);
1078         struct ll_sb_info *sbi = ll_i2sbi(inode);
1079         ENTRY;
1080
1081         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1082                inode->i_generation, inode);
1083
1084         if (S_ISDIR(inode->i_mode)) {
1085                 /* these should have been cleared in ll_file_release */
1086                 LASSERT(lli->lli_sai == NULL);
1087                 LASSERT(lli->lli_opendir_key == NULL);
1088                 LASSERT(lli->lli_opendir_pid == 0);
1089         }
1090
1091         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1092         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1093                          null_if_equal, inode);
1094
1095         LASSERT(!lli->lli_open_fd_write_count);
1096         LASSERT(!lli->lli_open_fd_read_count);
1097         LASSERT(!lli->lli_open_fd_exec_count);
1098
1099         if (lli->lli_mds_write_och)
1100                 ll_md_real_close(inode, FMODE_WRITE);
1101         if (lli->lli_mds_exec_och)
1102                 ll_md_real_close(inode, FMODE_EXEC);
1103         if (lli->lli_mds_read_och)
1104                 ll_md_real_close(inode, FMODE_READ);
1105
1106         if (lli->lli_symlink_name) {
1107                 OBD_FREE(lli->lli_symlink_name,
1108                          strlen(lli->lli_symlink_name) + 1);
1109                 lli->lli_symlink_name = NULL;
1110         }
1111
1112         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1113                 LASSERT(lli->lli_posix_acl == NULL);
1114                 if (lli->lli_remote_perms) {
1115                         free_rmtperm_hash(lli->lli_remote_perms);
1116                         lli->lli_remote_perms = NULL;
1117                 }
1118         }
1119 #ifdef CONFIG_FS_POSIX_ACL
1120         else if (lli->lli_posix_acl) {
1121                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1122                 LASSERT(lli->lli_remote_perms == NULL);
1123                 posix_acl_release(lli->lli_posix_acl);
1124                 lli->lli_posix_acl = NULL;
1125         }
1126 #endif
1127         lli->lli_inode_magic = LLI_INODE_DEAD;
1128
1129 #ifdef HAVE_EXPORT___IGET
1130         cfs_spin_lock(&sbi->ll_deathrow_lock);
1131         cfs_list_del_init(&lli->lli_dead_list);
1132         cfs_spin_unlock(&sbi->ll_deathrow_lock);
1133 #endif
1134         ll_clear_inode_capas(inode);
1135         /*
1136          * XXX This has to be done before lsm is freed below, because
1137          * cl_object still uses inode lsm.
1138          */
1139         cl_inode_fini(inode);
1140
1141         if (lli->lli_smd) {
1142                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1143                 lli->lli_smd = NULL;
1144         }
1145
1146
1147         EXIT;
1148 }
1149
1150 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1151                   struct md_open_data **mod)
1152 {
1153         struct lustre_md md;
1154         struct ll_sb_info *sbi = ll_i2sbi(inode);
1155         struct ptlrpc_request *request = NULL;
1156         int rc;
1157         ENTRY;
1158
1159         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1160                                      LUSTRE_OPC_ANY, NULL);
1161         if (IS_ERR(op_data))
1162                 RETURN(PTR_ERR(op_data));
1163
1164         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1165                         &request, mod);
1166         if (rc) {
1167                 ptlrpc_req_finished(request);
1168                 if (rc == -ENOENT) {
1169                         inode->i_nlink = 0;
1170                         /* Unlinked special device node? Or just a race?
1171                          * Pretend we done everything. */
1172                         if (!S_ISREG(inode->i_mode) &&
1173                             !S_ISDIR(inode->i_mode))
1174                                 rc = inode_setattr(inode, &op_data->op_attr);
1175                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1176                         CERROR("md_setattr fails: rc = %d\n", rc);
1177                 }
1178                 RETURN(rc);
1179         }
1180
1181         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1182                               sbi->ll_md_exp, &md);
1183         if (rc) {
1184                 ptlrpc_req_finished(request);
1185                 RETURN(rc);
1186         }
1187
1188         /* We call inode_setattr to adjust timestamps.
1189          * If there is at least some data in file, we cleared ATTR_SIZE
1190          * above to avoid invoking vmtruncate, otherwise it is important
1191          * to call vmtruncate in inode_setattr to update inode->i_size
1192          * (bug 6196) */
1193         rc = inode_setattr(inode, &op_data->op_attr);
1194
1195         /* Extract epoch data if obtained. */
1196         op_data->op_handle = md.body->handle;
1197         op_data->op_ioepoch = md.body->ioepoch;
1198
1199         ll_update_inode(inode, &md);
1200         ptlrpc_req_finished(request);
1201
1202         RETURN(rc);
1203 }
1204
1205 /* Close IO epoch and send Size-on-MDS attribute update. */
1206 static int ll_setattr_done_writing(struct inode *inode,
1207                                    struct md_op_data *op_data,
1208                                    struct md_open_data *mod)
1209 {
1210         struct ll_inode_info *lli = ll_i2info(inode);
1211         int rc = 0;
1212         ENTRY;
1213
1214         LASSERT(op_data != NULL);
1215         if (!S_ISREG(inode->i_mode))
1216                 RETURN(0);
1217
1218         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1219                op_data->op_ioepoch, PFID(&lli->lli_fid));
1220
1221         op_data->op_flags = MF_EPOCH_CLOSE;
1222         ll_done_writing_attr(inode, op_data);
1223         ll_pack_inode2opdata(inode, op_data, NULL);
1224
1225         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1226         if (rc == -EAGAIN) {
1227                 /* MDS has instructed us to obtain Size-on-MDS attribute
1228                  * from OSTs and send setattr to back to MDS. */
1229                 rc = ll_som_update(inode, op_data);
1230         } else if (rc) {
1231                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1232                        inode->i_ino, rc);
1233         }
1234         RETURN(rc);
1235 }
1236
1237 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1238 {
1239         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1240         int rc;
1241
1242         rc = cl_setattr_do_truncate(inode, size, capa);
1243         ll_truncate_free_capa(capa);
1244         return rc;
1245 }
1246
1247 static int ll_setattr_ost(struct inode *inode)
1248 {
1249         struct obd_capa *capa = ll_mdscapa_get(inode);
1250         int rc;
1251
1252         rc = cl_setattr_ost(inode, capa);
1253         capa_put(capa);
1254
1255         return rc;
1256 }
1257
1258 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1259  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1260  * keep these values until such a time that objects are allocated for it.
1261  * We do the MDS operations first, as it is checking permissions for us.
1262  * We don't to the MDS RPC if there is nothing that we want to store there,
1263  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1264  * going to do an RPC anyways.
1265  *
1266  * If we are doing a truncate, we will send the mtime and ctime updates
1267  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1268  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1269  * at the same time.
1270  */
1271 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1272 {
1273         struct ll_inode_info *lli = ll_i2info(inode);
1274         struct lov_stripe_md *lsm = lli->lli_smd;
1275         struct md_op_data *op_data = NULL;
1276         struct md_open_data *mod = NULL;
1277         int ia_valid = attr->ia_valid;
1278         int rc = 0, rc1 = 0;
1279         ENTRY;
1280
1281         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1282                attr->ia_valid);
1283         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1284
1285         if (ia_valid & ATTR_SIZE) {
1286                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1287                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1288                                attr->ia_size, ll_file_maxbytes(inode));
1289                         RETURN(-EFBIG);
1290                 }
1291
1292                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1293         }
1294
1295         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1296         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1297                 if (cfs_curproc_fsuid() != inode->i_uid &&
1298                     !cfs_capable(CFS_CAP_FOWNER))
1299                         RETURN(-EPERM);
1300         }
1301
1302         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1303         if (attr->ia_valid & ATTR_CTIME) {
1304                 attr->ia_ctime = CFS_CURRENT_TIME;
1305                 attr->ia_valid |= ATTR_CTIME_SET;
1306         }
1307         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1308                 attr->ia_atime = CFS_CURRENT_TIME;
1309                 attr->ia_valid |= ATTR_ATIME_SET;
1310         }
1311         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1312                 attr->ia_mtime = CFS_CURRENT_TIME;
1313                 attr->ia_valid |= ATTR_MTIME_SET;
1314         }
1315         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1316                 /* To avoid stale mtime on mds, obtain it from ost and send
1317                    to mds. */
1318                 rc = cl_glimpse_size(inode);
1319                 if (rc)
1320                         RETURN(rc);
1321
1322                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1323                 attr->ia_mtime = inode->i_mtime;
1324         }
1325
1326         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1327                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1328                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1329                        cfs_time_current_sec());
1330
1331         /* NB: ATTR_SIZE will only be set after this point if the size
1332          * resides on the MDS, ie, this file has no objects. */
1333         if (lsm)
1334                 attr->ia_valid &= ~ATTR_SIZE;
1335
1336         /* We always do an MDS RPC, even if we're only changing the size;
1337          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1338
1339         OBD_ALLOC_PTR(op_data);
1340         if (op_data == NULL)
1341                 RETURN(-ENOMEM);
1342
1343         UNLOCK_INODE_MUTEX(inode);
1344         if (ia_valid & ATTR_SIZE)
1345                 UP_WRITE_I_ALLOC_SEM(inode);
1346         cfs_down(&lli->lli_trunc_sem);
1347         LOCK_INODE_MUTEX(inode);
1348         if (ia_valid & ATTR_SIZE)
1349                 DOWN_WRITE_I_ALLOC_SEM(inode);
1350
1351         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1352
1353         /* Open epoch for truncate. */
1354         if (exp_connect_som(ll_i2mdexp(inode)) && (ia_valid & ATTR_SIZE))
1355                 op_data->op_flags = MF_EPOCH_OPEN;
1356
1357         rc = ll_md_setattr(inode, op_data, &mod);
1358         if (rc)
1359                 GOTO(out, rc);
1360
1361         ll_ioepoch_open(lli, op_data->op_ioepoch);
1362         if (!lsm || !S_ISREG(inode->i_mode)) {
1363                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1364                 GOTO(out, rc = 0);
1365         }
1366
1367         if (ia_valid & ATTR_SIZE)
1368                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1369         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1370                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1371                        inode->i_ino, LTIME_S(attr->ia_mtime));
1372                 rc = ll_setattr_ost(inode);
1373         }
1374         EXIT;
1375 out:
1376         if (op_data) {
1377                 if (op_data->op_ioepoch)
1378                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1379                 ll_finish_md_op_data(op_data);
1380         }
1381         cfs_up(&lli->lli_trunc_sem);
1382         return rc ? rc : rc1;
1383 }
1384
1385 int ll_setattr(struct dentry *de, struct iattr *attr)
1386 {
1387         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1388             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1389                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1390
1391         if ((de->d_inode->i_mode & S_ISUID) &&
1392             !(attr->ia_mode & S_ISUID) &&
1393             !(attr->ia_valid & ATTR_KILL_SUID))
1394                 attr->ia_valid |= ATTR_KILL_SUID;
1395
1396         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1397             !(attr->ia_mode & S_ISGID) &&
1398             !(attr->ia_valid & ATTR_KILL_SGID))
1399                 attr->ia_valid |= ATTR_KILL_SGID;
1400
1401         return ll_setattr_raw(de->d_inode, attr);
1402 }
1403
1404 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1405                        __u64 max_age, __u32 flags)
1406 {
1407         struct ll_sb_info *sbi = ll_s2sbi(sb);
1408         struct obd_statfs obd_osfs;
1409         int rc;
1410         ENTRY;
1411
1412         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1413         if (rc) {
1414                 CERROR("md_statfs fails: rc = %d\n", rc);
1415                 RETURN(rc);
1416         }
1417
1418         osfs->os_type = sb->s_magic;
1419
1420         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1421                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1422
1423         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1424                 flags |= OBD_STATFS_NODELAY;
1425
1426         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1427                               &obd_osfs, max_age, flags);
1428         if (rc) {
1429                 CERROR("obd_statfs fails: rc = %d\n", rc);
1430                 RETURN(rc);
1431         }
1432
1433         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1434                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1435                obd_osfs.os_files);
1436
1437         osfs->os_bsize = obd_osfs.os_bsize;
1438         osfs->os_blocks = obd_osfs.os_blocks;
1439         osfs->os_bfree = obd_osfs.os_bfree;
1440         osfs->os_bavail = obd_osfs.os_bavail;
1441
1442         /* If we don't have as many objects free on the OST as inodes
1443          * on the MDS, we reduce the total number of inodes to
1444          * compensate, so that the "inodes in use" number is correct.
1445          */
1446         if (obd_osfs.os_ffree < osfs->os_ffree) {
1447                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1448                         obd_osfs.os_ffree;
1449                 osfs->os_ffree = obd_osfs.os_ffree;
1450         }
1451
1452         RETURN(rc);
1453 }
1454 #ifndef HAVE_STATFS_DENTRY_PARAM
1455 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1456 {
1457 #else
1458 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1459 {
1460         struct super_block *sb = de->d_sb;
1461 #endif
1462         struct obd_statfs osfs;
1463         int rc;
1464
1465         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1466         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1467
1468         /* For now we will always get up-to-date statfs values, but in the
1469          * future we may allow some amount of caching on the client (e.g.
1470          * from QOS or lprocfs updates). */
1471         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1472         if (rc)
1473                 return rc;
1474
1475         statfs_unpack(sfs, &osfs);
1476
1477         /* We need to downshift for all 32-bit kernels, because we can't
1478          * tell if the kernel is being called via sys_statfs64() or not.
1479          * Stop before overflowing f_bsize - in which case it is better
1480          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1481         if (sizeof(long) < 8) {
1482                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1483                         sfs->f_bsize <<= 1;
1484
1485                         osfs.os_blocks >>= 1;
1486                         osfs.os_bfree >>= 1;
1487                         osfs.os_bavail >>= 1;
1488                 }
1489         }
1490
1491         sfs->f_blocks = osfs.os_blocks;
1492         sfs->f_bfree = osfs.os_bfree;
1493         sfs->f_bavail = osfs.os_bavail;
1494
1495         return 0;
1496 }
1497
1498 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1499 {
1500         struct ll_inode_info *lli;
1501         struct lov_stripe_md *lsm;
1502
1503         lli = ll_i2info(inode);
1504         LASSERT(lli->lli_size_sem_owner != current);
1505         cfs_down(&lli->lli_size_sem);
1506         LASSERT(lli->lli_size_sem_owner == NULL);
1507         lli->lli_size_sem_owner = current;
1508         lsm = lli->lli_smd;
1509         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1510                  lsm, lock_lsm);
1511         if (lock_lsm)
1512                 lov_stripe_lock(lsm);
1513 }
1514
1515 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1516 {
1517         struct ll_inode_info *lli;
1518         struct lov_stripe_md *lsm;
1519
1520         lli = ll_i2info(inode);
1521         lsm = lli->lli_smd;
1522         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1523                  lsm, unlock_lsm);
1524         if (unlock_lsm)
1525                 lov_stripe_unlock(lsm);
1526         LASSERT(lli->lli_size_sem_owner == current);
1527         lli->lli_size_sem_owner = NULL;
1528         cfs_up(&lli->lli_size_sem);
1529 }
1530
1531 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1532 {
1533         struct ll_inode_info *lli = ll_i2info(inode);
1534         struct mdt_body *body = md->body;
1535         struct lov_stripe_md *lsm = md->lsm;
1536         struct ll_sb_info *sbi = ll_i2sbi(inode);
1537
1538         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1539         if (lsm != NULL) {
1540                 if (lli->lli_smd == NULL) {
1541                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1542                             lsm->lsm_magic != LOV_MAGIC_V3) {
1543                                 dump_lsm(D_ERROR, lsm);
1544                                 LBUG();
1545                         }
1546                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1547                                lsm, inode->i_ino, inode->i_generation, inode);
1548                         cl_inode_init(inode, md);
1549                         /* ll_inode_size_lock() requires it is only
1550                          * called with lli_smd != NULL or lock_lsm == 0
1551                          *  or we can race between lock/unlock.
1552                          *  bug 9547 */
1553                         lli->lli_smd = lsm;
1554                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1555                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1556                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1557                 } else {
1558                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1559                                 lli->lli_smd->lsm_stripe_count ==
1560                                 lsm->lsm_stripe_count);
1561                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1562                                 CERROR("lsm mismatch for inode %ld\n",
1563                                        inode->i_ino);
1564                                 CERROR("lli_smd:\n");
1565                                 dump_lsm(D_ERROR, lli->lli_smd);
1566                                 CERROR("lsm:\n");
1567                                 dump_lsm(D_ERROR, lsm);
1568                                 LBUG();
1569                         }
1570                 }
1571                 if (lli->lli_smd != lsm)
1572                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1573         }
1574
1575         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1576                 if (body->valid & OBD_MD_FLRMTPERM)
1577                         ll_update_remote_perm(inode, md->remote_perm);
1578         }
1579 #ifdef CONFIG_FS_POSIX_ACL
1580         else if (body->valid & OBD_MD_FLACL) {
1581                 cfs_spin_lock(&lli->lli_lock);
1582                 if (lli->lli_posix_acl)
1583                         posix_acl_release(lli->lli_posix_acl);
1584                 lli->lli_posix_acl = md->posix_acl;
1585                 cfs_spin_unlock(&lli->lli_lock);
1586         }
1587 #endif
1588         inode->i_ino = cl_fid_build_ino(&body->fid1);
1589         inode->i_generation = cl_fid_build_gen(&body->fid1);
1590
1591         if (body->valid & OBD_MD_FLATIME &&
1592             body->atime > LTIME_S(inode->i_atime))
1593                 LTIME_S(inode->i_atime) = body->atime;
1594
1595         /* mtime is always updated with ctime, but can be set in past.
1596            As write and utime(2) may happen within 1 second, and utime's
1597            mtime has a priority over write's one, so take mtime from mds
1598            for the same ctimes. */
1599         if (body->valid & OBD_MD_FLCTIME &&
1600             body->ctime >= LTIME_S(inode->i_ctime)) {
1601                 LTIME_S(inode->i_ctime) = body->ctime;
1602                 if (body->valid & OBD_MD_FLMTIME) {
1603                         CDEBUG(D_INODE, "setting ino %lu mtime "
1604                                "from %lu to "LPU64"\n", inode->i_ino,
1605                                LTIME_S(inode->i_mtime), body->mtime);
1606                         LTIME_S(inode->i_mtime) = body->mtime;
1607                 }
1608         }
1609         if (body->valid & OBD_MD_FLMODE)
1610                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1611         if (body->valid & OBD_MD_FLTYPE)
1612                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1613         LASSERT(inode->i_mode != 0);
1614         if (S_ISREG(inode->i_mode)) {
1615                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1616         } else {
1617                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1618         }
1619 #ifdef HAVE_INODE_BLKSIZE
1620         inode->i_blksize = 1<<inode->i_blkbits;
1621 #endif
1622         if (body->valid & OBD_MD_FLUID)
1623                 inode->i_uid = body->uid;
1624         if (body->valid & OBD_MD_FLGID)
1625                 inode->i_gid = body->gid;
1626         if (body->valid & OBD_MD_FLFLAGS)
1627                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1628         if (body->valid & OBD_MD_FLNLINK)
1629                 inode->i_nlink = body->nlink;
1630         if (body->valid & OBD_MD_FLRDEV)
1631                 inode->i_rdev = old_decode_dev(body->rdev);
1632
1633         if (body->valid & OBD_MD_FLID) {
1634                 /* FID shouldn't be changed! */
1635                 if (fid_is_sane(&lli->lli_fid)) {
1636                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1637                                  "Trying to change FID "DFID
1638                                  " to the "DFID", inode %lu/%u(%p)\n",
1639                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1640                                  inode->i_ino, inode->i_generation, inode);
1641                 } else
1642                         lli->lli_fid = body->fid1;
1643         }
1644
1645         LASSERT(fid_seq(&lli->lli_fid) != 0);
1646
1647         if (body->valid & OBD_MD_FLSIZE) {
1648                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1649                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1650                         struct lustre_handle lockh;
1651                         ldlm_mode_t mode;
1652
1653                         /* As it is possible a blocking ast has been processed
1654                          * by this time, we need to check there is an UPDATE
1655                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1656                          * it. */
1657                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1658                                                &lockh);
1659                         if (mode) {
1660                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1661                                                       LLIF_EPOCH_PENDING |
1662                                                       LLIF_SOM_DIRTY)) {
1663                                         CERROR("ino %lu flags %lu still has "
1664                                                "size authority! do not trust "
1665                                                "the size got from MDS\n",
1666                                                inode->i_ino, lli->lli_flags);
1667                                 } else {
1668                                         /* Use old size assignment to avoid
1669                                          * deadlock bz14138 & bz14326 */
1670                                         inode->i_size = body->size;
1671                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1672                                 }
1673                                 ldlm_lock_decref(&lockh, mode);
1674                         }
1675                 } else {
1676                         /* Use old size assignment to avoid
1677                          * deadlock bz14138 & bz14326 */
1678                         inode->i_size = body->size;
1679                 }
1680
1681                 if (body->valid & OBD_MD_FLBLOCKS)
1682                         inode->i_blocks = body->blocks;
1683         }
1684
1685         if (body->valid & OBD_MD_FLMDSCAPA) {
1686                 LASSERT(md->mds_capa);
1687                 ll_add_capa(inode, md->mds_capa);
1688         }
1689         if (body->valid & OBD_MD_FLOSSCAPA) {
1690                 LASSERT(md->oss_capa);
1691                 ll_add_capa(inode, md->oss_capa);
1692         }
1693 }
1694
1695 static struct backing_dev_info ll_backing_dev_info = {
1696         .ra_pages       = 0,    /* No readahead */
1697 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1698         .capabilities   = 0,    /* Does contribute to dirty memory */
1699 #else
1700         .memory_backed  = 0,    /* Does contribute to dirty memory */
1701 #endif
1702 };
1703
1704 void ll_read_inode2(struct inode *inode, void *opaque)
1705 {
1706         struct lustre_md *md = opaque;
1707         struct ll_inode_info *lli = ll_i2info(inode);
1708         ENTRY;
1709
1710         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1711                inode->i_ino, inode->i_generation, inode);
1712
1713         ll_lli_init(lli);
1714
1715         LASSERT(!lli->lli_smd);
1716
1717         /* Core attributes from the MDS first.  This is a new inode, and
1718          * the VFS doesn't zero times in the core inode so we have to do
1719          * it ourselves.  They will be overwritten by either MDS or OST
1720          * attributes - we just need to make sure they aren't newer. */
1721         LTIME_S(inode->i_mtime) = 0;
1722         LTIME_S(inode->i_atime) = 0;
1723         LTIME_S(inode->i_ctime) = 0;
1724         inode->i_rdev = 0;
1725         ll_update_inode(inode, md);
1726
1727         /* OIDEBUG(inode); */
1728
1729         if (S_ISREG(inode->i_mode)) {
1730                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1731                 inode->i_op = &ll_file_inode_operations;
1732                 inode->i_fop = sbi->ll_fop;
1733                 inode->i_mapping->a_ops = &ll_aops;
1734                 EXIT;
1735         } else if (S_ISDIR(inode->i_mode)) {
1736                 inode->i_op = &ll_dir_inode_operations;
1737                 inode->i_fop = &ll_dir_operations;
1738                 inode->i_mapping->a_ops = &ll_dir_aops;
1739                 EXIT;
1740         } else if (S_ISLNK(inode->i_mode)) {
1741                 inode->i_op = &ll_fast_symlink_inode_operations;
1742                 EXIT;
1743         } else {
1744                 inode->i_op = &ll_special_inode_operations;
1745
1746                 init_special_inode(inode, inode->i_mode,
1747                                    kdev_t_to_nr(inode->i_rdev));
1748
1749                 /* initializing backing dev info. */
1750                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1751
1752                 EXIT;
1753         }
1754 }
1755
1756 void ll_delete_inode(struct inode *inode)
1757 {
1758         struct ll_sb_info *sbi = ll_i2sbi(inode);
1759         int rc;
1760         ENTRY;
1761
1762         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1763         if (rc) {
1764                 CERROR("fid_delete() failed, rc %d\n", rc);
1765         }
1766         truncate_inode_pages(&inode->i_data, 0);
1767         clear_inode(inode);
1768
1769         EXIT;
1770 }
1771
1772 int ll_iocontrol(struct inode *inode, struct file *file,
1773                  unsigned int cmd, unsigned long arg)
1774 {
1775         struct ll_sb_info *sbi = ll_i2sbi(inode);
1776         struct ptlrpc_request *req = NULL;
1777         int rc, flags = 0;
1778         ENTRY;
1779
1780         switch(cmd) {
1781         case FSFILT_IOC_GETFLAGS: {
1782                 struct mdt_body *body;
1783                 struct md_op_data *op_data;
1784
1785                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1786                                              0, 0, LUSTRE_OPC_ANY,
1787                                              NULL);
1788                 if (op_data == NULL)
1789                         RETURN(-ENOMEM);
1790
1791                 op_data->op_valid = OBD_MD_FLFLAGS;
1792                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1793                 ll_finish_md_op_data(op_data);
1794                 if (rc) {
1795                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1796                         RETURN(-abs(rc));
1797                 }
1798
1799                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1800
1801                 flags = body->flags;
1802
1803                 ptlrpc_req_finished(req);
1804
1805                 RETURN(put_user(flags, (int *)arg));
1806         }
1807         case FSFILT_IOC_SETFLAGS: {
1808                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1809                 struct obd_info oinfo = { { { 0 } } };
1810                 struct md_op_data *op_data;
1811
1812                 if (get_user(flags, (int *)arg))
1813                         RETURN(-EFAULT);
1814
1815                 oinfo.oi_md = lsm;
1816                 OBDO_ALLOC(oinfo.oi_oa);
1817                 if (!oinfo.oi_oa)
1818                         RETURN(-ENOMEM);
1819
1820                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1821                                              LUSTRE_OPC_ANY, NULL);
1822                 if (IS_ERR(op_data))
1823                         RETURN(PTR_ERR(op_data));
1824
1825                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1826                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1827                 rc = md_setattr(sbi->ll_md_exp, op_data,
1828                                 NULL, 0, NULL, 0, &req, NULL);
1829                 ll_finish_md_op_data(op_data);
1830                 ptlrpc_req_finished(req);
1831                 if (rc) {
1832                         OBDO_FREE(oinfo.oi_oa);
1833                         RETURN(rc);
1834                 }
1835
1836                 if (lsm == NULL) {
1837                         OBDO_FREE(oinfo.oi_oa);
1838                         GOTO(update_cache, rc);
1839                 }
1840
1841                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1842                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1843                 oinfo.oi_oa->o_flags = flags;
1844                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1845                                        OBD_MD_FLGROUP;
1846                 oinfo.oi_capa = ll_mdscapa_get(inode);
1847
1848                 obdo_from_inode(oinfo.oi_oa, inode,
1849                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1850                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1851                 capa_put(oinfo.oi_capa);
1852                 OBDO_FREE(oinfo.oi_oa);
1853                 if (rc) {
1854                         if (rc != -EPERM && rc != -EACCES)
1855                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1856                         RETURN(rc);
1857                 }
1858
1859                 EXIT;
1860 update_cache:
1861                 inode->i_flags = ll_ext_to_inode_flags(flags);
1862                 return 0;
1863         }
1864         default:
1865                 RETURN(-ENOSYS);
1866         }
1867
1868         RETURN(0);
1869 }
1870
1871 int ll_flush_ctx(struct inode *inode)
1872 {
1873         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1874
1875         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1876
1877         obd_set_info_async(sbi->ll_md_exp,
1878                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1879                            0, NULL, NULL);
1880         obd_set_info_async(sbi->ll_dt_exp,
1881                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1882                            0, NULL, NULL);
1883         return 0;
1884 }
1885
1886 /* umount -f client means force down, don't save state */
1887 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1888 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1889 {
1890         struct super_block *sb = vfsmnt->mnt_sb;
1891 #else
1892 void ll_umount_begin(struct super_block *sb)
1893 {
1894 #endif
1895         struct lustre_sb_info *lsi = s2lsi(sb);
1896         struct ll_sb_info *sbi = ll_s2sbi(sb);
1897         struct obd_device *obd;
1898         struct obd_ioctl_data *ioc_data;
1899         ENTRY;
1900
1901 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1902         if (!(flags & MNT_FORCE)) {
1903                 EXIT;
1904                 return;
1905         }
1906 #endif
1907
1908         /* Tell the MGC we got umount -f */
1909         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1910
1911         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1912                sb->s_count, atomic_read(&sb->s_active));
1913
1914         obd = class_exp2obd(sbi->ll_md_exp);
1915         if (obd == NULL) {
1916                 CERROR("Invalid MDC connection handle "LPX64"\n",
1917                        sbi->ll_md_exp->exp_handle.h_cookie);
1918                 EXIT;
1919                 return;
1920         }
1921         obd->obd_force = 1;
1922
1923         obd = class_exp2obd(sbi->ll_dt_exp);
1924         if (obd == NULL) {
1925                 CERROR("Invalid LOV connection handle "LPX64"\n",
1926                        sbi->ll_dt_exp->exp_handle.h_cookie);
1927                 EXIT;
1928                 return;
1929         }
1930         obd->obd_force = 1;
1931
1932         OBD_ALLOC_PTR(ioc_data);
1933         if (ioc_data) {
1934                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1935                               sizeof ioc_data, ioc_data, NULL);
1936
1937                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1938                               sizeof ioc_data, ioc_data, NULL);
1939
1940                 OBD_FREE_PTR(ioc_data);
1941         }
1942
1943
1944         /* Really, we'd like to wait until there are no requests outstanding,
1945          * and then continue.  For now, we just invalidate the requests,
1946          * schedule() and sleep one second if needed, and hope.
1947          */
1948         cfs_schedule();
1949 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1950         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1951                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1952                                                    cfs_time_seconds(1));
1953                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1954                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1955                                       "may try to umount it a bit later\n",
1956                                       atomic_read(&vfsmnt->mnt_count));
1957         }
1958 #endif
1959
1960         EXIT;
1961 }
1962
1963 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1964 {
1965         struct ll_sb_info *sbi = ll_s2sbi(sb);
1966         int err;
1967         __u32 read_only;
1968
1969         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1970                 read_only = *flags & MS_RDONLY;
1971                 err = obd_set_info_async(sbi->ll_md_exp,
1972                                          sizeof(KEY_READ_ONLY),
1973                                          KEY_READ_ONLY, sizeof(read_only),
1974                                          &read_only, NULL);
1975                 if (err) {
1976                         CERROR("Failed to change the read-only flag during "
1977                                "remount: %d\n", err);
1978                         return err;
1979                 }
1980
1981                 if (read_only)
1982                         sb->s_flags |= MS_RDONLY;
1983                 else
1984                         sb->s_flags &= ~MS_RDONLY;
1985         }
1986         return 0;
1987 }
1988
1989 int ll_prep_inode(struct inode **inode,
1990                   struct ptlrpc_request *req,
1991                   struct super_block *sb)
1992 {
1993         struct ll_sb_info *sbi = NULL;
1994         struct lustre_md md;
1995         int rc;
1996         ENTRY;
1997
1998         LASSERT(*inode || sb);
1999         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2000         prune_deathrow(sbi, 1);
2001         memset(&md, 0, sizeof(struct lustre_md));
2002
2003         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2004                               sbi->ll_md_exp, &md);
2005         if (rc)
2006                 RETURN(rc);
2007
2008         if (*inode) {
2009                 ll_update_inode(*inode, &md);
2010         } else {
2011                 LASSERT(sb != NULL);
2012
2013                 /*
2014                  * At this point server returns to client's same fid as client
2015                  * generated for creating. So using ->fid1 is okay here.
2016                  */
2017                 LASSERT(fid_is_sane(&md.body->fid1));
2018
2019                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
2020                 if (*inode == NULL || IS_ERR(*inode)) {
2021                         if (md.lsm)
2022                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2023 #ifdef CONFIG_FS_POSIX_ACL
2024                         if (md.posix_acl) {
2025                                 posix_acl_release(md.posix_acl);
2026                                 md.posix_acl = NULL;
2027                         }
2028 #endif
2029                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2030                         *inode = NULL;
2031                         CERROR("new_inode -fatal: rc %d\n", rc);
2032                         GOTO(out, rc);
2033                 }
2034         }
2035
2036 out:
2037         md_free_lustre_md(sbi->ll_md_exp, &md);
2038         RETURN(rc);
2039 }
2040
2041 int ll_obd_statfs(struct inode *inode, void *arg)
2042 {
2043         struct ll_sb_info *sbi = NULL;
2044         struct obd_export *exp;
2045         char *buf = NULL;
2046         struct obd_ioctl_data *data = NULL;
2047         __u32 type;
2048         int len = 0, rc;
2049
2050         if (!inode || !(sbi = ll_i2sbi(inode)))
2051                 GOTO(out_statfs, rc = -EINVAL);
2052
2053         rc = obd_ioctl_getdata(&buf, &len, arg);
2054         if (rc)
2055                 GOTO(out_statfs, rc);
2056
2057         data = (void*)buf;
2058         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2059             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2060                 GOTO(out_statfs, rc = -EINVAL);
2061
2062         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2063         if (type == LL_STATFS_MDC)
2064                 exp = sbi->ll_md_exp;
2065         else if (type == LL_STATFS_LOV)
2066                 exp = sbi->ll_dt_exp;
2067         else
2068                 GOTO(out_statfs, rc = -ENODEV);
2069
2070         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2071         if (rc)
2072                 GOTO(out_statfs, rc);
2073 out_statfs:
2074         if (buf)
2075                 obd_ioctl_freedata(buf, len);
2076         return rc;
2077 }
2078
2079 int ll_process_config(struct lustre_cfg *lcfg)
2080 {
2081         char *ptr;
2082         void *sb;
2083         struct lprocfs_static_vars lvars;
2084         unsigned long x;
2085         int rc = 0;
2086
2087         lprocfs_llite_init_vars(&lvars);
2088
2089         /* The instance name contains the sb: lustre-client-aacfe000 */
2090         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2091         if (!ptr || !*(++ptr))
2092                 return -EINVAL;
2093         if (sscanf(ptr, "%lx", &x) != 1)
2094                 return -EINVAL;
2095         sb = (void *)x;
2096         /* This better be a real Lustre superblock! */
2097         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2098
2099         /* Note we have not called client_common_fill_super yet, so
2100            proc fns must be able to handle that! */
2101         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2102                                       lcfg, sb);
2103         if (rc > 0)
2104                 rc = 0;
2105         return(rc);
2106 }
2107
2108 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2109 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2110                                        struct inode *i1, struct inode *i2,
2111                                        const char *name, int namelen,
2112                                        int mode, __u32 opc, void *data)
2113 {
2114         LASSERT(i1 != NULL);
2115
2116         if (namelen > ll_i2sbi(i1)->ll_namelen)
2117                 return ERR_PTR(-ENAMETOOLONG);
2118
2119         if (op_data == NULL)
2120                 OBD_ALLOC_PTR(op_data);
2121
2122         if (op_data == NULL)
2123                 return ERR_PTR(-ENOMEM);
2124
2125         ll_i2gids(op_data->op_suppgids, i1, i2);
2126         op_data->op_fid1 = *ll_inode2fid(i1);
2127         op_data->op_capa1 = ll_mdscapa_get(i1);
2128
2129         if (i2) {
2130                 op_data->op_fid2 = *ll_inode2fid(i2);
2131                 op_data->op_capa2 = ll_mdscapa_get(i2);
2132         } else {
2133                 fid_zero(&op_data->op_fid2);
2134                 op_data->op_capa2 = NULL;
2135         }
2136
2137         op_data->op_name = name;
2138         op_data->op_namelen = namelen;
2139         op_data->op_mode = mode;
2140         op_data->op_mod_time = cfs_time_current_sec();
2141         op_data->op_fsuid = cfs_curproc_fsuid();
2142         op_data->op_fsgid = cfs_curproc_fsgid();
2143         op_data->op_cap = cfs_curproc_cap_pack();
2144         op_data->op_bias = MDS_CHECK_SPLIT;
2145         op_data->op_opc = opc;
2146         op_data->op_mds = 0;
2147         op_data->op_data = data;
2148
2149         return op_data;
2150 }
2151
2152 void ll_finish_md_op_data(struct md_op_data *op_data)
2153 {
2154         capa_put(op_data->op_capa1);
2155         capa_put(op_data->op_capa2);
2156         OBD_FREE_PTR(op_data);
2157 }
2158
2159 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2160 {
2161         struct ll_sb_info *sbi;
2162
2163         LASSERT((seq != NULL) && (vfs != NULL));
2164         sbi = ll_s2sbi(vfs->mnt_sb);
2165
2166         if (sbi->ll_flags & LL_SBI_NOLCK)
2167                 seq_puts(seq, ",nolock");
2168
2169         if (sbi->ll_flags & LL_SBI_FLOCK)
2170                 seq_puts(seq, ",flock");
2171
2172         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2173                 seq_puts(seq, ",localflock");
2174
2175         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2176                 seq_puts(seq, ",user_xattr");
2177
2178         if (sbi->ll_flags & LL_SBI_ACL)
2179                 seq_puts(seq, ",acl");
2180
2181         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2182                 seq_puts(seq, ",lazystatfs");
2183
2184         RETURN(0);
2185 }