lustre/llite/llite_lib.c

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