lustre/llite/llite_lib.c

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