lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = 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) 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         spin_lock_init(&sbi->ll_lock);
  86         spin_lock_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         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 = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         spin_lock(&ll_sb_lock);
 115         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_LLITE_CHECKSUM
 119         sbi->ll_flags |= LL_SBI_CHECKSUM;
 120 #endif
 121
 122 #ifdef HAVE_LRU_RESIZE_SUPPORT
 123         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 124 #endif
 125
 126 #ifdef HAVE_EXPORT___IGET
 127         INIT_LIST_HEAD(&sbi->ll_deathrow);
 128         spin_lock_init(&sbi->ll_deathrow_lock);
 129 #endif
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137
 138         RETURN(sbi);
 139 }
 140
 141 void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144         ENTRY;
 145
 146         if (sbi != NULL) {
 147                 spin_lock(&ll_sb_lock);
 148                 list_del(&sbi->ll_list);
 149                 spin_unlock(&ll_sb_lock);
 150                 OBD_FREE(sbi, sizeof(*sbi));
 151         }
 152         EXIT;
 153 }
 154
 155 static struct dentry_operations ll_d_root_ops = {
 156 #ifdef DCACHE_LUSTRE_INVALID
 157         .d_compare = ll_dcompare,
 158 #endif
 159 };
 160
 161 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 162  * us to make MDS RPCs with large enough reply buffers to hold the
 163  * maximum-sized (= maximum striped) EA and cookie without having to
 164  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 165 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 166 {
 167         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
 168         __u32 valsize = sizeof(struct lov_desc);
 169         int rc, easize, def_easize, cookiesize;
 170         struct lov_desc desc;
 171         __u32 stripes;
 172         ENTRY;
 173
 174         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 175                           &valsize, &desc, NULL);
 176         if (rc)
 177                 RETURN(rc);
 178
 179         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 180         lsm.lsm_stripe_count = stripes;
 181         easize = obd_size_diskmd(dt_exp, &lsm);
 182
 183         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 184         def_easize = obd_size_diskmd(dt_exp, &lsm);
 185
 186         cookiesize = stripes * sizeof(struct llog_cookie);
 187
 188         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 189                easize, cookiesize);
 190
 191         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 192         RETURN(rc);
 193 }
 194
 195 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 196 {
 197         struct inode *root = 0;
 198         struct ll_sb_info *sbi = ll_s2sbi(sb);
 199         struct obd_device *obd;
 200         struct obd_capa *oc = NULL;
 201         struct obd_statfs osfs;
 202         struct ptlrpc_request *request = NULL;
 203         struct lustre_handle dt_conn = {0, };
 204         struct lustre_handle md_conn = {0, };
 205         struct obd_connect_data *data = NULL;
 206         struct obd_uuid *uuid;
 207         struct lustre_md lmd;
 208         obd_valid valid;
 209         int size, err, checksum;
 210         ENTRY;
 211
 212         obd = class_name2obd(md);
 213         if (!obd) {
 214                 CERROR("MD %s: not setup or attached\n", md);
 215                 RETURN(-EINVAL);
 216         }
 217
 218         OBD_ALLOC_PTR(data);
 219         if (data == NULL)
 220                 RETURN(-ENOMEM);
 221
 222         if (proc_lustre_fs_root) {
 223                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 224                                                   dt, md);
 225                 if (err < 0)
 226                         CERROR("could not register mount in /proc/fs/lustre\n");
 227         }
 228
 229         /* indicate the features supported by this client */
 230         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 231                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 232                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 233                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 234                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 235                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT;
 236
 237 #ifdef HAVE_LRU_RESIZE_SUPPORT
 238         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 239                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 240 #endif
 241 #ifdef CONFIG_FS_POSIX_ACL
 242         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 243 #endif
 244         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 245         data->ocd_version = LUSTRE_VERSION_CODE;
 246
 247         if (sb->s_flags & MS_RDONLY)
 248                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 249         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 250                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 251
 252 #ifdef HAVE_MS_FLOCK_LOCK
 253         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 254         sb->s_flags |= MS_FLOCK_LOCK;
 255 #endif
 256
 257         if (sbi->ll_flags & LL_SBI_FLOCK)
 258                 sbi->ll_fop = &ll_file_operations_flock;
 259         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 260                 sbi->ll_fop = &ll_file_operations;
 261         else
 262                 sbi->ll_fop = &ll_file_operations_noflock;
 263
 264         /* real client */
 265         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 266         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 267                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 268
 269         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 270         if (err == -EBUSY) {
 271                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 272                                    "recovery, of which this client is not a "
 273                                    "part. Please wait for recovery to complete,"
 274                                    " abort, or time out.\n", md);
 275                 GOTO(out, err);
 276         } else if (err) {
 277                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 278                 GOTO(out, err);
 279         }
 280         sbi->ll_md_exp = class_conn2export(&md_conn);
 281
 282         err = obd_fid_init(sbi->ll_md_exp);
 283         if (err) {
 284                 CERROR("Can't init metadata layer FID infrastructure, "
 285                        "rc %d\n", err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 290         if (err)
 291                 GOTO(out_md_fid, err);
 292
 293         size = sizeof(*data);
 294         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 295                            KEY_CONN_DATA,  &size, data, NULL);
 296         if (err) {
 297                 CERROR("Get connect data failed: %d \n", err);
 298                 GOTO(out_md, err);
 299         }
 300
 301         LASSERT(osfs.os_bsize);
 302         sb->s_blocksize = osfs.os_bsize;
 303         sb->s_blocksize_bits = log2(osfs.os_bsize);
 304         sb->s_magic = LL_SUPER_MAGIC;
 305
 306         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 307          *         retval = in_file->f_op->sendfile(...);
 308          *         if (*ppos > max)
 309          *                 retval = -EOVERFLOW;
 310          *
 311          * it will check if *ppos is greater than max. However, max equals to
 312          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 313          * has been defined as a signed long long ineger in linux kernel. */
 314 #if BITS_PER_LONG == 64
 315         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 316 #else
 317         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 318 #endif
 319         sbi->ll_namelen = osfs.os_namelen;
 320         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 321
 322         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 323             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 324                 LCONSOLE_INFO("Disabling user_xattr feature because "
 325                               "it is not supported on the server\n");
 326                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 327         }
 328
 329         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 330 #ifdef MS_POSIXACL
 331                 sb->s_flags |= MS_POSIXACL;
 332 #endif
 333                 sbi->ll_flags |= LL_SBI_ACL;
 334         } else {
 335                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 336 #ifdef MS_POSIXACL
 337                 sb->s_flags &= ~MS_POSIXACL;
 338 #endif
 339                 sbi->ll_flags &= ~LL_SBI_ACL;
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 343                 sbi->ll_flags |= LL_SBI_JOIN;
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 346                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 347                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 348                         LCONSOLE_INFO("client is set as remote by default.\n");
 349                 }
 350         } else {
 351                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 352                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 353                         LCONSOLE_INFO("client claims to be remote, but server "
 354                                       "rejected, forced to be local.\n");
 355                 }
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 359                 LCONSOLE_INFO("client enabled MDS capability!\n");
 360                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 364                 LCONSOLE_INFO("client enabled OSS capability!\n");
 365                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 366         }
 367
 368         obd = class_name2obd(dt);
 369         if (!obd) {
 370                 CERROR("DT %s: not setup or attached\n", dt);
 371                 GOTO(out_md_fid, err = -ENODEV);
 372         }
 373
 374         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 375                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 376                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 377                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 378                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 379                                   OBD_CONNECT_OSS_CAPA;
 380
 381         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 382                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 383                  * disabled by default, because it can still be enabled on the
 384                  * fly via /proc. As a consequence, we still need to come to an
 385                  * agreement on the supported algorithms at connect time */
 386                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 387
 388                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 389                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 390                 else
 391                         /* send the list of supported checksum types */
 392                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 393         }
 394
 395 #ifdef HAVE_LRU_RESIZE_SUPPORT
 396         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 397 #endif
 398         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 399                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 400
 401         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 402                "ocd_grant: %d\n", data->ocd_connect_flags,
 403                data->ocd_version, data->ocd_grant);
 404
 405         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 406         obd->obd_upcall.onu_upcall = ll_ocd_update;
 407         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 408
 409         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 410         if (err == -EBUSY) {
 411                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 412                                    "recovery, of which this client is not a "
 413                                    "part.  Please wait for recovery to "
 414                                    "complete, abort, or time out.\n", dt);
 415                 GOTO(out_md_fid, err);
 416         } else if (err) {
 417                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 418                 GOTO(out_md_fid, err);
 419         }
 420
 421         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 422
 423         err = obd_fid_init(sbi->ll_dt_exp);
 424         if (err) {
 425                 CERROR("Can't init data layer FID infrastructure, "
 426                        "rc %d\n", err);
 427                 GOTO(out_dt, err);
 428         }
 429
 430         spin_lock(&sbi->ll_lco.lco_lock);
 431         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 432         spin_unlock(&sbi->ll_lco.lco_lock);
 433
 434         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 435         if (err) {
 436                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 437                 GOTO(out_lock_cn_cb, err);
 438         }
 439
 440         fid_zero(&sbi->ll_root_fid);
 441         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 442         if (err) {
 443                 CERROR("cannot mds_connect: rc = %d\n", err);
 444                 GOTO(out_lock_cn_cb, err);
 445         }
 446         if (!fid_is_sane(&sbi->ll_root_fid)) {
 447                 CERROR("Invalid root fid during mount\n");
 448                 GOTO(out_lock_cn_cb, err = -EINVAL);
 449         }
 450         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 451
 452         sb->s_op = &lustre_super_operations;
 453         sb->s_export_op = &lustre_export_operations;
 454
 455         /* make root inode
 456          * XXX: move this to after cbd setup? */
 457         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 458         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 459                 valid |= OBD_MD_FLRMTPERM;
 460         else if (sbi->ll_flags & LL_SBI_ACL)
 461                 valid |= OBD_MD_FLACL;
 462
 463         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 464                          &request);
 465         if (oc)
 466                 capa_put(oc);
 467         if (err) {
 468                 CERROR("md_getattr failed for root: rc = %d\n", err);
 469                 GOTO(out_lock_cn_cb, err);
 470         }
 471         memset(&lmd, 0, sizeof(lmd));
 472         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 473                                sbi->ll_md_exp, &lmd);
 474         if (err) {
 475                 CERROR("failed to understand root inode md: rc = %d\n", err);
 476                 ptlrpc_req_finished (request);
 477                 GOTO(out_lock_cn_cb, err);
 478         }
 479
 480         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 481         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 482         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 483         ptlrpc_req_finished(request);
 484
 485         if (root == NULL || IS_ERR(root)) {
 486                 if (lmd.lsm)
 487                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 488 #ifdef CONFIG_FS_POSIX_ACL
 489                 if (lmd.posix_acl) {
 490                         posix_acl_release(lmd.posix_acl);
 491                         lmd.posix_acl = NULL;
 492                 }
 493 #endif
 494                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 495                 root = NULL;
 496                 CERROR("lustre_lite: bad iget4 for root\n");
 497                 GOTO(out_root, err);
 498         }
 499
 500         err = ll_close_thread_start(&sbi->ll_lcq);
 501         if (err) {
 502                 CERROR("cannot start close thread: rc %d\n", err);
 503                 GOTO(out_root, err);
 504         }
 505
 506 #ifdef CONFIG_FS_POSIX_ACL
 507         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 508                 rct_init(&sbi->ll_rct);
 509                 et_init(&sbi->ll_et);
 510         }
 511 #endif
 512
 513         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 514         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 515                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 516                                  NULL);
 517         cl_sb_init(sb);
 518
 519         sb->s_root = d_alloc_root(root);
 520         if (data != NULL)
 521                 OBD_FREE(data, sizeof(*data));
 522
 523         sb->s_root->d_op = &ll_d_root_ops;
 524
 525         sbi->ll_sdev_orig = sb->s_dev;
 526
 527         /* We set sb->s_dev equal on all lustre clients in order to support
 528          * NFS export clustering.  NFSD requires that the FSID be the same
 529          * on all clients. */
 530         /* s_dev is also used in lt_compare() to compare two fs, but that is
 531          * only a node-local comparison. */
 532         uuid = obd_get_uuid(sbi->ll_md_exp);
 533         if (uuid != NULL)
 534                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 535
 536         RETURN(err);
 537 out_root:
 538         if (root)
 539                 iput(root);
 540 out_lock_cn_cb:
 541         obd_fid_fini(sbi->ll_dt_exp);
 542 out_dt:
 543         obd_disconnect(sbi->ll_dt_exp);
 544         sbi->ll_dt_exp = NULL;
 545 out_md_fid:
 546         obd_fid_fini(sbi->ll_md_exp);
 547 out_md:
 548         obd_disconnect(sbi->ll_md_exp);
 549         sbi->ll_md_exp = NULL;
 550 out:
 551         if (data != NULL)
 552                 OBD_FREE_PTR(data);
 553         lprocfs_unregister_mountpoint(sbi);
 554         return err;
 555 }
 556
 557 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 558 {
 559         int size, rc;
 560
 561         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 562         size = sizeof(int);
 563         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 564                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 565         if (rc)
 566                 CERROR("Get max mdsize error rc %d \n", rc);
 567
 568         RETURN(rc);
 569 }
 570
 571 void ll_dump_inode(struct inode *inode)
 572 {
 573         struct list_head *tmp;
 574         int dentry_count = 0;
 575
 576         LASSERT(inode != NULL);
 577
 578         list_for_each(tmp, &inode->i_dentry)
 579                 dentry_count++;
 580
 581         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 582                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 583                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 584 }
 585
 586 void lustre_dump_dentry(struct dentry *dentry, int recur)
 587 {
 588         struct list_head *tmp;
 589         int subdirs = 0;
 590
 591         LASSERT(dentry != NULL);
 592
 593         list_for_each(tmp, &dentry->d_subdirs)
 594                 subdirs++;
 595
 596         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 597                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 598                dentry->d_name.len, dentry->d_name.name,
 599                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 600                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 601                dentry->d_flags, dentry->d_fsdata, subdirs);
 602         if (dentry->d_inode != NULL)
 603                 ll_dump_inode(dentry->d_inode);
 604
 605         if (recur == 0)
 606                 return;
 607
 608         list_for_each(tmp, &dentry->d_subdirs) {
 609                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 610                 lustre_dump_dentry(d, recur - 1);
 611         }
 612 }
 613
 614 #ifdef HAVE_EXPORT___IGET
 615 static void prune_dir_dentries(struct inode *inode)
 616 {
 617         struct dentry *dentry, *prev = NULL;
 618
 619         /* due to lustre specific logic, a directory
 620          * can have few dentries - a bug from VFS POV */
 621 restart:
 622         spin_lock(&dcache_lock);
 623         if (!list_empty(&inode->i_dentry)) {
 624                 dentry = list_entry(inode->i_dentry.prev,
 625                                     struct dentry, d_alias);
 626                 /* in order to prevent infinite loops we
 627                  * break if previous dentry is busy */
 628                 if (dentry != prev) {
 629                         prev = dentry;
 630                         dget_locked(dentry);
 631                         spin_unlock(&dcache_lock);
 632
 633                         /* try to kill all child dentries */
 634                         lock_dentry(dentry);
 635                         shrink_dcache_parent(dentry);
 636                         unlock_dentry(dentry);
 637                         dput(dentry);
 638
 639                         /* now try to get rid of current dentry */
 640                         d_prune_aliases(inode);
 641                         goto restart;
 642                 }
 643         }
 644         spin_unlock(&dcache_lock);
 645 }
 646
 647 static void prune_deathrow_one(struct ll_inode_info *lli)
 648 {
 649         struct inode *inode = ll_info2i(lli);
 650
 651         /* first, try to drop any dentries - they hold a ref on the inode */
 652         if (S_ISDIR(inode->i_mode))
 653                 prune_dir_dentries(inode);
 654         else
 655                 d_prune_aliases(inode);
 656
 657
 658         /* if somebody still uses it, leave it */
 659         LASSERT(atomic_read(&inode->i_count) > 0);
 660         if (atomic_read(&inode->i_count) > 1)
 661                 goto out;
 662
 663         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 664                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 665
 666         /* seems nobody uses it anymore */
 667         inode->i_nlink = 0;
 668
 669 out:
 670         iput(inode);
 671         return;
 672 }
 673
 674 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 675 {
 676         struct ll_inode_info *lli;
 677         int empty;
 678
 679         do {
 680                 if (need_resched() && try)
 681                         break;
 682
 683                 if (try) {
 684                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 685                                 break;
 686                 } else {
 687                         spin_lock(&sbi->ll_deathrow_lock);
 688                 }
 689
 690                 empty = 1;
 691                 lli = NULL;
 692                 if (!list_empty(&sbi->ll_deathrow)) {
 693                         lli = list_entry(sbi->ll_deathrow.next,
 694                                          struct ll_inode_info,
 695                                          lli_dead_list);
 696                         list_del_init(&lli->lli_dead_list);
 697                         if (!list_empty(&sbi->ll_deathrow))
 698                                 empty = 0;
 699                 }
 700                 spin_unlock(&sbi->ll_deathrow_lock);
 701
 702                 if (lli)
 703                         prune_deathrow_one(lli);
 704
 705         } while (empty == 0);
 706 }
 707 #else /* !HAVE_EXPORT___IGET */
 708 #define prune_deathrow(sbi, try) do {} while (0)
 709 #endif /* HAVE_EXPORT___IGET */
 710
 711 void client_common_put_super(struct super_block *sb)
 712 {
 713         struct ll_sb_info *sbi = ll_s2sbi(sb);
 714         ENTRY;
 715
 716 #ifdef CONFIG_FS_POSIX_ACL
 717         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 718                 et_fini(&sbi->ll_et);
 719                 rct_fini(&sbi->ll_rct);
 720         }
 721 #endif
 722
 723         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 724
 725         ll_close_thread_shutdown(sbi->ll_lcq);
 726
 727         /* destroy inodes in deathrow */
 728         prune_deathrow(sbi, 0);
 729
 730         list_del(&sbi->ll_conn_chain);
 731
 732         obd_fid_fini(sbi->ll_dt_exp);
 733         obd_disconnect(sbi->ll_dt_exp);
 734         sbi->ll_dt_exp = NULL;
 735
 736         lprocfs_unregister_mountpoint(sbi);
 737
 738         obd_fid_fini(sbi->ll_md_exp);
 739         obd_disconnect(sbi->ll_md_exp);
 740         sbi->ll_md_exp = NULL;
 741
 742         EXIT;
 743 }
 744
 745 void ll_kill_super(struct super_block *sb)
 746 {
 747         struct ll_sb_info *sbi;
 748
 749         ENTRY;
 750
 751         /* not init sb ?*/
 752         if (!(sb->s_flags & MS_ACTIVE))
 753                 return;
 754
 755         sbi = ll_s2sbi(sb);
 756         /* we need restore s_dev from changed for clustred NFS before put_super
 757          * because new kernels have cached s_dev and change sb->s_dev in
 758          * put_super not affected real removing devices */
 759         if (sbi)
 760                 sb->s_dev = sbi->ll_sdev_orig;
 761         EXIT;
 762 }
 763
 764 char *ll_read_opt(const char *opt, char *data)
 765 {
 766         char *value;
 767         char *retval;
 768         ENTRY;
 769
 770         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 771         if (strncmp(opt, data, strlen(opt)))
 772                 RETURN(NULL);
 773         if ((value = strchr(data, '=')) == NULL)
 774                 RETURN(NULL);
 775
 776         value++;
 777         OBD_ALLOC(retval, strlen(value) + 1);
 778         if (!retval) {
 779                 CERROR("out of memory!\n");
 780                 RETURN(NULL);
 781         }
 782
 783         memcpy(retval, value, strlen(value)+1);
 784         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 785         RETURN(retval);
 786 }
 787
 788 static inline int ll_set_opt(const char *opt, char *data, int fl)
 789 {
 790         if (strncmp(opt, data, strlen(opt)) != 0)
 791                 return(0);
 792         else
 793                 return(fl);
 794 }
 795
 796 /* non-client-specific mount options are parsed in lmd_parse */
 797 static int ll_options(char *options, int *flags)
 798 {
 799         int tmp;
 800         char *s1 = options, *s2;
 801         ENTRY;
 802
 803         if (!options)
 804                 RETURN(0);
 805
 806         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 807
 808         while (*s1) {
 809                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 810                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 821                 if (tmp) {
 822                         *flags |= tmp;
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 826                 if (tmp) {
 827                         *flags &= ~tmp;
 828                         goto next;
 829                 }
 830                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 836                 if (tmp) {
 837                         *flags &= ~tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 841                 if (tmp) {
 842                         /* Ignore deprecated mount option.  The client will
 843                          * always try to mount with ACL support, whether this
 844                          * is used depends on whether server supports it. */
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 848                 if (tmp) {
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 852                 if (tmp) {
 853                         *flags |= tmp;
 854                         goto next;
 855                 }
 856
 857                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877
 878                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 879                                    s1);
 880                 RETURN(-EINVAL);
 881
 882 next:
 883                 /* Find next opt */
 884                 s2 = strchr(s1, ',');
 885                 if (s2 == NULL)
 886                         break;
 887                 s1 = s2 + 1;
 888         }
 889         RETURN(0);
 890 }
 891
 892 void ll_lli_init(struct ll_inode_info *lli)
 893 {
 894         lli->lli_inode_magic = LLI_INODE_MAGIC;
 895         sema_init(&lli->lli_size_sem, 1);
 896         sema_init(&lli->lli_write_sem, 1);
 897         lli->lli_flags = 0;
 898         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 899         spin_lock_init(&lli->lli_lock);
 900         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 901         INIT_LIST_HEAD(&lli->lli_close_list);
 902         lli->lli_inode_magic = LLI_INODE_MAGIC;
 903         sema_init(&lli->lli_och_sem, 1);
 904         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 905         lli->lli_mds_exec_och = NULL;
 906         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 907         lli->lli_open_fd_exec_count = 0;
 908         INIT_LIST_HEAD(&lli->lli_dead_list);
 909         lli->lli_remote_perms = NULL;
 910         lli->lli_rmtperm_utime = 0;
 911         sema_init(&lli->lli_rmtperm_sem, 1);
 912         INIT_LIST_HEAD(&lli->lli_oss_capas);
 913 }
 914
 915 int ll_fill_super(struct super_block *sb)
 916 {
 917         struct lustre_profile *lprof;
 918         struct lustre_sb_info *lsi = s2lsi(sb);
 919         struct ll_sb_info *sbi;
 920         char  *dt = NULL, *md = NULL;
 921         char  *profilenm = get_profile_name(sb);
 922         struct config_llog_instance cfg = {0, };
 923         char   ll_instance[sizeof(sb) * 2 + 1];
 924         int    err;
 925         ENTRY;
 926
 927         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 928
 929         cfs_module_get();
 930
 931         /* client additional sb info */
 932         lsi->lsi_llsbi = sbi = ll_init_sbi();
 933         if (!sbi) {
 934                 cfs_module_put();
 935                 RETURN(-ENOMEM);
 936         }
 937
 938         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 939         if (err)
 940                 GOTO(out_free, err);
 941
 942         /* Generate a string unique to this super, in case some joker tries
 943            to mount the same fs at two mount points.
 944            Use the address of the super itself.*/
 945         sprintf(ll_instance, "%p", sb);
 946         cfg.cfg_instance = ll_instance;
 947         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 948
 949         /* set up client obds */
 950         err = lustre_process_log(sb, profilenm, &cfg);
 951         if (err < 0) {
 952                 CERROR("Unable to process log: %d\n", err);
 953                 GOTO(out_free, err);
 954         }
 955
 956         lprof = class_get_profile(profilenm);
 957         if (lprof == NULL) {
 958                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 959                                    " read from the MGS.  Does that filesystem "
 960                                    "exist?\n", profilenm);
 961                 GOTO(out_free, err = -EINVAL);
 962         }
 963         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 964                lprof->lp_md, lprof->lp_dt);
 965
 966         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 967                   strlen(ll_instance) + 2);
 968         if (!dt)
 969                 GOTO(out_free, err = -ENOMEM);
 970         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 971
 972         OBD_ALLOC(md, strlen(lprof->lp_md) +
 973                   strlen(ll_instance) + 2);
 974         if (!md)
 975                 GOTO(out_free, err = -ENOMEM);
 976         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 977
 978         /* connections, registrations, sb setup */
 979         err = client_common_fill_super(sb, md, dt);
 980
 981 out_free:
 982         if (md)
 983                 OBD_FREE(md, strlen(md) + 1);
 984         if (dt)
 985                 OBD_FREE(dt, strlen(dt) + 1);
 986         if (err)
 987                 ll_put_super(sb);
 988         else
 989                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 990
 991         RETURN(err);
 992 } /* ll_fill_super */
 993
 994
 995 void lu_context_keys_dump(void);
 996
 997 void ll_put_super(struct super_block *sb)
 998 {
 999         struct config_llog_instance cfg;
1000         char   ll_instance[sizeof(sb) * 2 + 1];
1001         struct obd_device *obd;
1002         struct lustre_sb_info *lsi = s2lsi(sb);
1003         struct ll_sb_info *sbi = ll_s2sbi(sb);
1004         char *profilenm = get_profile_name(sb);
1005         int force = 1, next;
1006         ENTRY;
1007
1008         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1009
1010         ll_print_capa_stat(sbi);
1011
1012         sprintf(ll_instance, "%p", sb);
1013         cfg.cfg_instance = ll_instance;
1014         lustre_end_log(sb, NULL, &cfg);
1015
1016         if (sbi->ll_md_exp) {
1017                 obd = class_exp2obd(sbi->ll_md_exp);
1018                 if (obd)
1019                         force = obd->obd_force;
1020         }
1021
1022         /* We need to set force before the lov_disconnect in
1023            lustre_common_put_super, since l_d cleans up osc's as well. */
1024         if (force) {
1025                 next = 0;
1026                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1027                                                      &next)) != NULL) {
1028                         obd->obd_force = force;
1029                 }
1030         }
1031
1032         cl_sb_fini(sb);
1033
1034         if (sbi->ll_lcq) {
1035                 /* Only if client_common_fill_super succeeded */
1036                 client_common_put_super(sb);
1037         }
1038         next = 0;
1039         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1040                 class_manual_cleanup(obd);
1041         }
1042
1043         if (profilenm)
1044                 class_del_profile(profilenm);
1045
1046         ll_free_sbi(sb);
1047         lsi->lsi_llsbi = NULL;
1048
1049         lustre_common_put_super(sb);
1050
1051         cl_env_cache_purge(~0);
1052         lu_context_keys_dump();
1053
1054         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1055
1056         cfs_module_put();
1057
1058         EXIT;
1059 } /* client_put_super */
1060
1061 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1062 {
1063         struct inode *inode = NULL;
1064         /* NOTE: we depend on atomic igrab() -bzzz */
1065         lock_res_and_lock(lock);
1066         if (lock->l_ast_data) {
1067                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1068                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1069                         inode = igrab(lock->l_ast_data);
1070                 } else {
1071                         inode = lock->l_ast_data;
1072                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1073                                                 D_INFO : D_WARNING,
1074                                         lock, __FILE__, __func__, __LINE__,
1075                                         "l_ast_data %p is bogus: magic %08x",
1076                                         lock->l_ast_data, lli->lli_inode_magic);
1077                         inode = NULL;
1078                 }
1079         }
1080         unlock_res_and_lock(lock);
1081         return inode;
1082 }
1083
1084 static int null_if_equal(struct ldlm_lock *lock, void *data)
1085 {
1086         if (data == lock->l_ast_data) {
1087                 lock->l_ast_data = NULL;
1088
1089                 if (lock->l_req_mode != lock->l_granted_mode)
1090                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1091         }
1092
1093         return LDLM_ITER_CONTINUE;
1094 }
1095
1096 void ll_clear_inode(struct inode *inode)
1097 {
1098         struct ll_inode_info *lli = ll_i2info(inode);
1099         struct ll_sb_info *sbi = ll_i2sbi(inode);
1100         ENTRY;
1101
1102         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1103                inode->i_generation, inode);
1104
1105         if (S_ISDIR(inode->i_mode)) {
1106                 /* these should have been cleared in ll_file_release */
1107                 LASSERT(lli->lli_sai == NULL);
1108                 LASSERT(lli->lli_opendir_key == NULL);
1109                 LASSERT(lli->lli_opendir_pid == 0);
1110         }
1111
1112         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1113         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1114                          null_if_equal, inode);
1115
1116         LASSERT(!lli->lli_open_fd_write_count);
1117         LASSERT(!lli->lli_open_fd_read_count);
1118         LASSERT(!lli->lli_open_fd_exec_count);
1119
1120         if (lli->lli_mds_write_och)
1121                 ll_md_real_close(inode, FMODE_WRITE);
1122         if (lli->lli_mds_exec_och)
1123                 ll_md_real_close(inode, FMODE_EXEC);
1124         if (lli->lli_mds_read_och)
1125                 ll_md_real_close(inode, FMODE_READ);
1126
1127         if (lli->lli_symlink_name) {
1128                 OBD_FREE(lli->lli_symlink_name,
1129                          strlen(lli->lli_symlink_name) + 1);
1130                 lli->lli_symlink_name = NULL;
1131         }
1132
1133         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1134                 LASSERT(lli->lli_posix_acl == NULL);
1135                 if (lli->lli_remote_perms) {
1136                         free_rmtperm_hash(lli->lli_remote_perms);
1137                         lli->lli_remote_perms = NULL;
1138                 }
1139         }
1140 #ifdef CONFIG_FS_POSIX_ACL
1141         else if (lli->lli_posix_acl) {
1142                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1143                 LASSERT(lli->lli_remote_perms == NULL);
1144                 posix_acl_release(lli->lli_posix_acl);
1145                 lli->lli_posix_acl = NULL;
1146         }
1147 #endif
1148         lli->lli_inode_magic = LLI_INODE_DEAD;
1149
1150 #ifdef HAVE_EXPORT___IGET
1151         spin_lock(&sbi->ll_deathrow_lock);
1152         list_del_init(&lli->lli_dead_list);
1153         spin_unlock(&sbi->ll_deathrow_lock);
1154 #endif
1155         ll_clear_inode_capas(inode);
1156         /*
1157          * XXX This has to be done before lsm is freed below, because
1158          * cl_object still uses inode lsm.
1159          */
1160         cl_inode_fini(inode);
1161
1162         if (lli->lli_smd) {
1163                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1164                 lli->lli_smd = NULL;
1165         }
1166
1167
1168         EXIT;
1169 }
1170
1171 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1172                   struct md_open_data **mod)
1173 {
1174         struct lustre_md md;
1175         struct ll_sb_info *sbi = ll_i2sbi(inode);
1176         struct ptlrpc_request *request = NULL;
1177         int rc;
1178         ENTRY;
1179
1180         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1181                                      LUSTRE_OPC_ANY, NULL);
1182         if (IS_ERR(op_data))
1183                 RETURN(PTR_ERR(op_data));
1184
1185         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1186                         &request, mod);
1187         if (rc) {
1188                 ptlrpc_req_finished(request);
1189                 if (rc == -ENOENT) {
1190                         inode->i_nlink = 0;
1191                         /* Unlinked special device node? Or just a race?
1192                          * Pretend we done everything. */
1193                         if (!S_ISREG(inode->i_mode) &&
1194                             !S_ISDIR(inode->i_mode))
1195                                 rc = inode_setattr(inode, &op_data->op_attr);
1196                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1197                         CERROR("md_setattr fails: rc = %d\n", rc);
1198                 }
1199                 RETURN(rc);
1200         }
1201
1202         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1203                               sbi->ll_md_exp, &md);
1204         if (rc) {
1205                 ptlrpc_req_finished(request);
1206                 RETURN(rc);
1207         }
1208
1209         /* We call inode_setattr to adjust timestamps.
1210          * If there is at least some data in file, we cleared ATTR_SIZE
1211          * above to avoid invoking vmtruncate, otherwise it is important
1212          * to call vmtruncate in inode_setattr to update inode->i_size
1213          * (bug 6196) */
1214         rc = inode_setattr(inode, &op_data->op_attr);
1215
1216         /* Extract epoch data if obtained. */
1217         op_data->op_handle = md.body->handle;
1218         op_data->op_ioepoch = md.body->ioepoch;
1219
1220         ll_update_inode(inode, &md);
1221         ptlrpc_req_finished(request);
1222
1223         RETURN(rc);
1224 }
1225
1226 /* Close IO epoch and send Size-on-MDS attribute update. */
1227 static int ll_setattr_done_writing(struct inode *inode,
1228                                    struct md_op_data *op_data,
1229                                    struct md_open_data *mod)
1230 {
1231         struct ll_inode_info *lli = ll_i2info(inode);
1232         int rc = 0;
1233         ENTRY;
1234
1235         LASSERT(op_data != NULL);
1236         if (!S_ISREG(inode->i_mode))
1237                 RETURN(0);
1238
1239         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1240                op_data->op_ioepoch, PFID(&lli->lli_fid));
1241
1242         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1243         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1244         if (rc == -EAGAIN) {
1245                 /* MDS has instructed us to obtain Size-on-MDS attribute
1246                  * from OSTs and send setattr to back to MDS. */
1247                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1248                                          op_data->op_ioepoch);
1249         } else if (rc) {
1250                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1251                        inode->i_ino, rc);
1252         }
1253         RETURN(rc);
1254 }
1255
1256 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1257 {
1258         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1259         int rc;
1260
1261         rc = cl_setattr_do_truncate(inode, size, capa);
1262         ll_truncate_free_capa(capa);
1263         return rc;
1264 }
1265
1266 static int ll_setattr_ost(struct inode *inode)
1267 {
1268         struct obd_capa *capa = ll_mdscapa_get(inode);
1269         int rc;
1270
1271         rc = cl_setattr_ost(inode, capa);
1272         capa_put(capa);
1273
1274         return rc;
1275 }
1276
1277 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1278  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1279  * keep these values until such a time that objects are allocated for it.
1280  * We do the MDS operations first, as it is checking permissions for us.
1281  * We don't to the MDS RPC if there is nothing that we want to store there,
1282  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1283  * going to do an RPC anyways.
1284  *
1285  * If we are doing a truncate, we will send the mtime and ctime updates
1286  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1287  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1288  * at the same time.
1289  */
1290 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1291 {
1292         struct ll_inode_info *lli = ll_i2info(inode);
1293         struct lov_stripe_md *lsm = lli->lli_smd;
1294         struct md_op_data *op_data = NULL;
1295         struct md_open_data *mod = NULL;
1296         int ia_valid = attr->ia_valid;
1297         int rc = 0, rc1 = 0;
1298         ENTRY;
1299
1300         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1301                attr->ia_valid);
1302         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1303
1304         if (ia_valid & ATTR_SIZE) {
1305                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1306                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1307                                attr->ia_size, ll_file_maxbytes(inode));
1308                         RETURN(-EFBIG);
1309                 }
1310
1311                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1312         }
1313
1314         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1315         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1316                 if (current->fsuid != inode->i_uid &&
1317                     !cfs_capable(CFS_CAP_FOWNER))
1318                         RETURN(-EPERM);
1319         }
1320
1321         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1322         if (attr->ia_valid & ATTR_CTIME) {
1323                 attr->ia_ctime = CURRENT_TIME;
1324                 attr->ia_valid |= ATTR_CTIME_SET;
1325         }
1326         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1327                 attr->ia_atime = CURRENT_TIME;
1328                 attr->ia_valid |= ATTR_ATIME_SET;
1329         }
1330         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1331                 attr->ia_mtime = CURRENT_TIME;
1332                 attr->ia_valid |= ATTR_MTIME_SET;
1333         }
1334         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1335                 /* To avoid stale mtime on mds, obtain it from ost and send
1336                    to mds. */
1337                 rc = cl_glimpse_size(inode);
1338                 if (rc)
1339                         RETURN(rc);
1340
1341                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1342                 attr->ia_mtime = inode->i_mtime;
1343         }
1344
1345         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1346                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1347                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1348                        cfs_time_current_sec());
1349
1350         /* NB: ATTR_SIZE will only be set after this point if the size
1351          * resides on the MDS, ie, this file has no objects. */
1352         if (lsm)
1353                 attr->ia_valid &= ~ATTR_SIZE;
1354
1355         /* We always do an MDS RPC, even if we're only changing the size;
1356          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1357
1358         OBD_ALLOC_PTR(op_data);
1359         if (op_data == NULL)
1360                 RETURN(-ENOMEM);
1361
1362         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1363
1364         /* Open epoch for truncate. */
1365         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1366             (ia_valid & ATTR_SIZE))
1367                 op_data->op_flags = MF_EPOCH_OPEN;
1368
1369         rc = ll_md_setattr(inode, op_data, &mod);
1370         if (rc)
1371                 GOTO(out, rc);
1372
1373         if (op_data->op_ioepoch)
1374                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1375                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1376
1377         if (!lsm || !S_ISREG(inode->i_mode)) {
1378                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1379                 GOTO(out, rc = 0);
1380         }
1381
1382         if (ia_valid & ATTR_SIZE)
1383                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1384         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1385                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1386                        inode->i_ino, LTIME_S(attr->ia_mtime));
1387                 rc = ll_setattr_ost(inode);
1388         }
1389         EXIT;
1390 out:
1391         if (op_data) {
1392                 if (op_data->op_ioepoch)
1393                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1394                 ll_finish_md_op_data(op_data);
1395         }
1396         return rc ? rc : rc1;
1397 }
1398
1399 int ll_setattr(struct dentry *de, struct iattr *attr)
1400 {
1401         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1402             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1403                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1404
1405         if ((de->d_inode->i_mode & S_ISUID) &&
1406             !(attr->ia_mode & S_ISUID) &&
1407             !(attr->ia_valid & ATTR_KILL_SUID))
1408                 attr->ia_valid |= ATTR_KILL_SUID;
1409
1410         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1411             !(attr->ia_mode & S_ISGID) &&
1412             !(attr->ia_valid & ATTR_KILL_SGID))
1413                 attr->ia_valid |= ATTR_KILL_SGID;
1414
1415         return ll_setattr_raw(de->d_inode, attr);
1416 }
1417
1418 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1419                        __u64 max_age, __u32 flags)
1420 {
1421         struct ll_sb_info *sbi = ll_s2sbi(sb);
1422         struct obd_statfs obd_osfs;
1423         int rc;
1424         ENTRY;
1425
1426         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1427         if (rc) {
1428                 CERROR("md_statfs fails: rc = %d\n", rc);
1429                 RETURN(rc);
1430         }
1431
1432         osfs->os_type = sb->s_magic;
1433
1434         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1435                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1436
1437         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1438                               &obd_osfs, max_age, flags);
1439         if (rc) {
1440                 CERROR("obd_statfs fails: rc = %d\n", rc);
1441                 RETURN(rc);
1442         }
1443
1444         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1445                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1446                obd_osfs.os_files);
1447
1448         osfs->os_bsize = obd_osfs.os_bsize;
1449         osfs->os_blocks = obd_osfs.os_blocks;
1450         osfs->os_bfree = obd_osfs.os_bfree;
1451         osfs->os_bavail = obd_osfs.os_bavail;
1452
1453         /* If we don't have as many objects free on the OST as inodes
1454          * on the MDS, we reduce the total number of inodes to
1455          * compensate, so that the "inodes in use" number is correct.
1456          */
1457         if (obd_osfs.os_ffree < osfs->os_ffree) {
1458                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1459                         obd_osfs.os_ffree;
1460                 osfs->os_ffree = obd_osfs.os_ffree;
1461         }
1462
1463         RETURN(rc);
1464 }
1465 #ifndef HAVE_STATFS_DENTRY_PARAM
1466 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1467 {
1468 #else
1469 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1470 {
1471         struct super_block *sb = de->d_sb;
1472 #endif
1473         struct obd_statfs osfs;
1474         int rc;
1475
1476         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1477         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1478
1479         /* For now we will always get up-to-date statfs values, but in the
1480          * future we may allow some amount of caching on the client (e.g.
1481          * from QOS or lprocfs updates). */
1482         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1483         if (rc)
1484                 return rc;
1485
1486         statfs_unpack(sfs, &osfs);
1487
1488         /* We need to downshift for all 32-bit kernels, because we can't
1489          * tell if the kernel is being called via sys_statfs64() or not.
1490          * Stop before overflowing f_bsize - in which case it is better
1491          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1492         if (sizeof(long) < 8) {
1493                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1494                         sfs->f_bsize <<= 1;
1495
1496                         osfs.os_blocks >>= 1;
1497                         osfs.os_bfree >>= 1;
1498                         osfs.os_bavail >>= 1;
1499                 }
1500         }
1501
1502         sfs->f_blocks = osfs.os_blocks;
1503         sfs->f_bfree = osfs.os_bfree;
1504         sfs->f_bavail = osfs.os_bavail;
1505
1506         return 0;
1507 }
1508
1509 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1510 {
1511         struct ll_inode_info *lli;
1512         struct lov_stripe_md *lsm;
1513
1514         lli = ll_i2info(inode);
1515         LASSERT(lli->lli_size_sem_owner != current);
1516         down(&lli->lli_size_sem);
1517         LASSERT(lli->lli_size_sem_owner == NULL);
1518         lli->lli_size_sem_owner = current;
1519         lsm = lli->lli_smd;
1520         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1521                  lsm, lock_lsm);
1522         if (lock_lsm)
1523                 lov_stripe_lock(lsm);
1524 }
1525
1526 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1527 {
1528         struct ll_inode_info *lli;
1529         struct lov_stripe_md *lsm;
1530
1531         lli = ll_i2info(inode);
1532         lsm = lli->lli_smd;
1533         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1534                  lsm, unlock_lsm);
1535         if (unlock_lsm)
1536                 lov_stripe_unlock(lsm);
1537         LASSERT(lli->lli_size_sem_owner == current);
1538         lli->lli_size_sem_owner = NULL;
1539         up(&lli->lli_size_sem);
1540 }
1541
1542 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1543 {
1544         struct ll_inode_info *lli = ll_i2info(inode);
1545
1546         dump_lsm(D_INODE, lsm);
1547         dump_lsm(D_INODE, lli->lli_smd);
1548         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1549                  "lsm must be joined lsm %p\n", lsm);
1550         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1551         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1552                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1553         lli->lli_smd = lsm;
1554         lli->lli_maxbytes = lsm->lsm_maxbytes;
1555         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1556                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1557 }
1558
1559 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1560 {
1561         struct ll_inode_info *lli = ll_i2info(inode);
1562         struct mdt_body *body = md->body;
1563         struct lov_stripe_md *lsm = md->lsm;
1564         struct ll_sb_info *sbi = ll_i2sbi(inode);
1565
1566         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1567         if (lsm != NULL) {
1568                 if (lli->lli_smd == NULL) {
1569                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1570                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1571                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1572                                 dump_lsm(D_ERROR, lsm);
1573                                 LBUG();
1574                         }
1575                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1576                                lsm, inode->i_ino, inode->i_generation, inode);
1577                         cl_inode_init(inode, md);
1578                         /* ll_inode_size_lock() requires it is only
1579                          * called with lli_smd != NULL or lock_lsm == 0
1580                          *  or we can race between lock/unlock.
1581                          *  bug 9547 */
1582                         lli->lli_smd = lsm;
1583                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1584                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1585                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1586                 } else {
1587                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1588                              lli->lli_smd->lsm_stripe_count ==
1589                                         lsm->lsm_stripe_count) {
1590                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1591                                         CERROR("lsm mismatch for inode %ld\n",
1592                                                 inode->i_ino);
1593                                         CERROR("lli_smd:\n");
1594                                         dump_lsm(D_ERROR, lli->lli_smd);
1595                                         CERROR("lsm:\n");
1596                                         dump_lsm(D_ERROR, lsm);
1597                                         LBUG();
1598                                 }
1599                         } else {
1600                                 cl_inode_init(inode, md);
1601                                 ll_replace_lsm(inode, lsm);
1602                         }
1603                 }
1604                 if (lli->lli_smd != lsm)
1605                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1606         }
1607
1608         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1609                 if (body->valid & OBD_MD_FLRMTPERM)
1610                         ll_update_remote_perm(inode, md->remote_perm);
1611         }
1612 #ifdef CONFIG_FS_POSIX_ACL
1613         else if (body->valid & OBD_MD_FLACL) {
1614                 spin_lock(&lli->lli_lock);
1615                 if (lli->lli_posix_acl)
1616                         posix_acl_release(lli->lli_posix_acl);
1617                 lli->lli_posix_acl = md->posix_acl;
1618                 spin_unlock(&lli->lli_lock);
1619         }
1620 #endif
1621         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1622         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1623
1624         if (body->valid & OBD_MD_FLATIME &&
1625             body->atime > LTIME_S(inode->i_atime))
1626                 LTIME_S(inode->i_atime) = body->atime;
1627
1628         /* mtime is always updated with ctime, but can be set in past.
1629            As write and utime(2) may happen within 1 second, and utime's
1630            mtime has a priority over write's one, so take mtime from mds
1631            for the same ctimes. */
1632         if (body->valid & OBD_MD_FLCTIME &&
1633             body->ctime >= LTIME_S(inode->i_ctime)) {
1634                 LTIME_S(inode->i_ctime) = body->ctime;
1635                 if (body->valid & OBD_MD_FLMTIME) {
1636                         CDEBUG(D_INODE, "setting ino %lu mtime "
1637                                "from %lu to "LPU64"\n", inode->i_ino,
1638                                LTIME_S(inode->i_mtime), body->mtime);
1639                         LTIME_S(inode->i_mtime) = body->mtime;
1640                 }
1641         }
1642         if (body->valid & OBD_MD_FLMODE)
1643                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1644         if (body->valid & OBD_MD_FLTYPE)
1645                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1646         LASSERT(inode->i_mode != 0);
1647         if (S_ISREG(inode->i_mode)) {
1648                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1649         } else {
1650                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1651         }
1652 #ifdef HAVE_INODE_BLKSIZE
1653         inode->i_blksize = 1<<inode->i_blkbits;
1654 #endif
1655         if (body->valid & OBD_MD_FLUID)
1656                 inode->i_uid = body->uid;
1657         if (body->valid & OBD_MD_FLGID)
1658                 inode->i_gid = body->gid;
1659         if (body->valid & OBD_MD_FLFLAGS)
1660                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1661         if (body->valid & OBD_MD_FLNLINK)
1662                 inode->i_nlink = body->nlink;
1663         if (body->valid & OBD_MD_FLRDEV)
1664                 inode->i_rdev = old_decode_dev(body->rdev);
1665
1666         if (body->valid & OBD_MD_FLID) {
1667                 /* FID shouldn't be changed! */
1668                 if (fid_is_sane(&lli->lli_fid)) {
1669                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1670                                  "Trying to change FID "DFID
1671                                  " to the "DFID", inode %lu/%u(%p)\n",
1672                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1673                                  inode->i_ino, inode->i_generation, inode);
1674                 } else
1675                         lli->lli_fid = body->fid1;
1676         }
1677
1678         LASSERT(fid_seq(&lli->lli_fid) != 0);
1679
1680         if (body->valid & OBD_MD_FLSIZE) {
1681                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1682                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1683                         struct lustre_handle lockh;
1684                         ldlm_mode_t mode;
1685
1686                         /* As it is possible a blocking ast has been processed
1687                          * by this time, we need to check there is an UPDATE
1688                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1689                          * it. */
1690                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1691                                                &lockh);
1692                         if (mode) {
1693                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1694                                                       LLIF_EPOCH_PENDING |
1695                                                       LLIF_SOM_DIRTY)) {
1696                                         CERROR("ino %lu flags %lu still has "
1697                                                "size authority! do not trust "
1698                                                "the size got from MDS\n",
1699                                                inode->i_ino, lli->lli_flags);
1700                                 } else {
1701                                         /* Use old size assignment to avoid
1702                                          * deadlock bz14138 & bz14326 */
1703                                         inode->i_size = body->size;
1704                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1705                                 }
1706                                 ldlm_lock_decref(&lockh, mode);
1707                         }
1708                 } else {
1709                         /* Use old size assignment to avoid
1710                          * deadlock bz14138 & bz14326 */
1711                         inode->i_size = body->size;
1712                 }
1713
1714                 if (body->valid & OBD_MD_FLBLOCKS)
1715                         inode->i_blocks = body->blocks;
1716         }
1717
1718         if (body->valid & OBD_MD_FLMDSCAPA) {
1719                 LASSERT(md->mds_capa);
1720                 ll_add_capa(inode, md->mds_capa);
1721         }
1722         if (body->valid & OBD_MD_FLOSSCAPA) {
1723                 LASSERT(md->oss_capa);
1724                 ll_add_capa(inode, md->oss_capa);
1725         }
1726 }
1727
1728 static struct backing_dev_info ll_backing_dev_info = {
1729         .ra_pages       = 0,    /* No readahead */
1730 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1731         .capabilities   = 0,    /* Does contribute to dirty memory */
1732 #else
1733         .memory_backed  = 0,    /* Does contribute to dirty memory */
1734 #endif
1735 };
1736
1737 void ll_read_inode2(struct inode *inode, void *opaque)
1738 {
1739         struct lustre_md *md = opaque;
1740         struct ll_inode_info *lli = ll_i2info(inode);
1741         ENTRY;
1742
1743         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1744                inode->i_ino, inode->i_generation, inode);
1745
1746         ll_lli_init(lli);
1747
1748         LASSERT(!lli->lli_smd);
1749
1750         /* Core attributes from the MDS first.  This is a new inode, and
1751          * the VFS doesn't zero times in the core inode so we have to do
1752          * it ourselves.  They will be overwritten by either MDS or OST
1753          * attributes - we just need to make sure they aren't newer. */
1754         LTIME_S(inode->i_mtime) = 0;
1755         LTIME_S(inode->i_atime) = 0;
1756         LTIME_S(inode->i_ctime) = 0;
1757         inode->i_rdev = 0;
1758         ll_update_inode(inode, md);
1759
1760         /* OIDEBUG(inode); */
1761
1762         if (S_ISREG(inode->i_mode)) {
1763                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1764                 inode->i_op = &ll_file_inode_operations;
1765                 inode->i_fop = sbi->ll_fop;
1766                 inode->i_mapping->a_ops = &ll_aops;
1767                 EXIT;
1768         } else if (S_ISDIR(inode->i_mode)) {
1769                 inode->i_op = &ll_dir_inode_operations;
1770                 inode->i_fop = &ll_dir_operations;
1771                 inode->i_mapping->a_ops = &ll_dir_aops;
1772                 EXIT;
1773         } else if (S_ISLNK(inode->i_mode)) {
1774                 inode->i_op = &ll_fast_symlink_inode_operations;
1775                 EXIT;
1776         } else {
1777                 inode->i_op = &ll_special_inode_operations;
1778
1779                 init_special_inode(inode, inode->i_mode,
1780                                    kdev_t_to_nr(inode->i_rdev));
1781
1782                 /* initializing backing dev info. */
1783                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1784
1785                 EXIT;
1786         }
1787 }
1788
1789 void ll_delete_inode(struct inode *inode)
1790 {
1791         struct ll_sb_info *sbi = ll_i2sbi(inode);
1792         int rc;
1793         ENTRY;
1794
1795         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1796         if (rc) {
1797                 CERROR("fid_delete() failed, rc %d\n", rc);
1798         }
1799         truncate_inode_pages(&inode->i_data, 0);
1800         clear_inode(inode);
1801
1802         EXIT;
1803 }
1804
1805 int ll_iocontrol(struct inode *inode, struct file *file,
1806                  unsigned int cmd, unsigned long arg)
1807 {
1808         struct ll_sb_info *sbi = ll_i2sbi(inode);
1809         struct ptlrpc_request *req = NULL;
1810         int rc, flags = 0;
1811         ENTRY;
1812
1813         switch(cmd) {
1814         case EXT3_IOC_GETFLAGS: {
1815                 struct mdt_body *body;
1816                 struct obd_capa *oc;
1817
1818                 oc = ll_mdscapa_get(inode);
1819                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1820                                 OBD_MD_FLFLAGS, 0, &req);
1821                 capa_put(oc);
1822                 if (rc) {
1823                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1824                         RETURN(-abs(rc));
1825                 }
1826
1827                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1828
1829                 flags = body->flags;
1830
1831                 ptlrpc_req_finished(req);
1832
1833                 RETURN(put_user(flags, (int *)arg));
1834         }
1835         case EXT3_IOC_SETFLAGS: {
1836                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1837                 struct obd_info oinfo = { { { 0 } } };
1838                 struct md_op_data *op_data;
1839
1840                 if (get_user(flags, (int *)arg))
1841                         RETURN(-EFAULT);
1842
1843                 oinfo.oi_md = lsm;
1844                 OBDO_ALLOC(oinfo.oi_oa);
1845                 if (!oinfo.oi_oa)
1846                         RETURN(-ENOMEM);
1847
1848                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1849                                              LUSTRE_OPC_ANY, NULL);
1850                 if (IS_ERR(op_data))
1851                         RETURN(PTR_ERR(op_data));
1852
1853                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1854                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1855                 rc = md_setattr(sbi->ll_md_exp, op_data,
1856                                 NULL, 0, NULL, 0, &req, NULL);
1857                 ll_finish_md_op_data(op_data);
1858                 ptlrpc_req_finished(req);
1859                 if (rc || lsm == NULL) {
1860                         OBDO_FREE(oinfo.oi_oa);
1861                         RETURN(rc);
1862                 }
1863
1864                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1865                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1866                 oinfo.oi_oa->o_flags = flags;
1867                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1868                                        OBD_MD_FLGROUP;
1869                 oinfo.oi_capa = ll_mdscapa_get(inode);
1870
1871                 obdo_from_inode(oinfo.oi_oa, inode,
1872                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1873                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1874                 capa_put(oinfo.oi_capa);
1875                 OBDO_FREE(oinfo.oi_oa);
1876                 if (rc) {
1877                         if (rc != -EPERM && rc != -EACCES)
1878                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1879                         RETURN(rc);
1880                 }
1881
1882                 inode->i_flags = ll_ext_to_inode_flags(flags |
1883                                                        MDS_BFLAG_EXT_FLAGS);
1884                 RETURN(0);
1885         }
1886         default:
1887                 RETURN(-ENOSYS);
1888         }
1889
1890         RETURN(0);
1891 }
1892
1893 int ll_flush_ctx(struct inode *inode)
1894 {
1895         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1896
1897         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1898
1899         obd_set_info_async(sbi->ll_md_exp,
1900                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1901                            0, NULL, NULL);
1902         obd_set_info_async(sbi->ll_dt_exp,
1903                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1904                            0, NULL, NULL);
1905         return 0;
1906 }
1907
1908 /* umount -f client means force down, don't save state */
1909 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1910 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1911 {
1912         struct super_block *sb = vfsmnt->mnt_sb;
1913 #else
1914 void ll_umount_begin(struct super_block *sb)
1915 {
1916 #endif
1917         struct lustre_sb_info *lsi = s2lsi(sb);
1918         struct ll_sb_info *sbi = ll_s2sbi(sb);
1919         struct obd_device *obd;
1920         struct obd_ioctl_data ioc_data = { 0 };
1921         ENTRY;
1922
1923 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1924         if (!(flags & MNT_FORCE)) {
1925                 EXIT;
1926                 return;
1927         }
1928 #endif
1929
1930         /* Tell the MGC we got umount -f */
1931         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1932
1933         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1934                sb->s_count, atomic_read(&sb->s_active));
1935
1936         obd = class_exp2obd(sbi->ll_md_exp);
1937         if (obd == NULL) {
1938                 CERROR("Invalid MDC connection handle "LPX64"\n",
1939                        sbi->ll_md_exp->exp_handle.h_cookie);
1940                 EXIT;
1941                 return;
1942         }
1943         obd->obd_force = 1;
1944         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1945                       &ioc_data, NULL);
1946
1947         obd = class_exp2obd(sbi->ll_dt_exp);
1948         if (obd == NULL) {
1949                 CERROR("Invalid LOV connection handle "LPX64"\n",
1950                        sbi->ll_dt_exp->exp_handle.h_cookie);
1951                 EXIT;
1952                 return;
1953         }
1954
1955         obd->obd_force = 1;
1956         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1957                       &ioc_data, NULL);
1958
1959         /* Really, we'd like to wait until there are no requests outstanding,
1960          * and then continue.  For now, we just invalidate the requests,
1961          * schedule, and hope.
1962          */
1963         schedule();
1964
1965         EXIT;
1966 }
1967
1968 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1969 {
1970         struct ll_sb_info *sbi = ll_s2sbi(sb);
1971         int err;
1972         __u32 read_only;
1973
1974         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1975                 read_only = *flags & MS_RDONLY;
1976                 err = obd_set_info_async(sbi->ll_md_exp,
1977                                          sizeof(KEY_READ_ONLY),
1978                                          KEY_READ_ONLY, sizeof(read_only),
1979                                          &read_only, NULL);
1980                 if (err) {
1981                         CERROR("Failed to change the read-only flag during "
1982                                "remount: %d\n", err);
1983                         return err;
1984                 }
1985
1986                 if (read_only)
1987                         sb->s_flags |= MS_RDONLY;
1988                 else
1989                         sb->s_flags &= ~MS_RDONLY;
1990         }
1991         return 0;
1992 }
1993
1994 int ll_prep_inode(struct inode **inode,
1995                   struct ptlrpc_request *req,
1996                   struct super_block *sb)
1997 {
1998         struct ll_sb_info *sbi = NULL;
1999         struct lustre_md md;
2000         int rc;
2001         ENTRY;
2002
2003         LASSERT(*inode || sb);
2004         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2005         prune_deathrow(sbi, 1);
2006         memset(&md, 0, sizeof(struct lustre_md));
2007
2008         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2009                               sbi->ll_md_exp, &md);
2010         if (rc)
2011                 RETURN(rc);
2012
2013         if (*inode) {
2014                 ll_update_inode(*inode, &md);
2015         } else {
2016                 LASSERT(sb != NULL);
2017
2018                 /*
2019                  * At this point server returns to client's same fid as client
2020                  * generated for creating. So using ->fid1 is okay here.
2021                  */
2022                 LASSERT(fid_is_sane(&md.body->fid1));
2023
2024                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2025                 if (*inode == NULL || IS_ERR(*inode)) {
2026                         if (md.lsm)
2027                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2028 #ifdef CONFIG_FS_POSIX_ACL
2029                         if (md.posix_acl) {
2030                                 posix_acl_release(md.posix_acl);
2031                                 md.posix_acl = NULL;
2032                         }
2033 #endif
2034                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2035                         *inode = NULL;
2036                         CERROR("new_inode -fatal: rc %d\n", rc);
2037                         GOTO(out, rc);
2038                 }
2039         }
2040
2041         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2042                          ll_i2info(*inode)->lli_smd);
2043 out:
2044         md_free_lustre_md(sbi->ll_md_exp, &md);
2045         RETURN(rc);
2046 }
2047
2048 int ll_obd_statfs(struct inode *inode, void *arg)
2049 {
2050         struct ll_sb_info *sbi = NULL;
2051         struct obd_export *exp;
2052         char *buf = NULL;
2053         struct obd_ioctl_data *data = NULL;
2054         __u32 type;
2055         int len = 0, rc;
2056
2057         if (!inode || !(sbi = ll_i2sbi(inode)))
2058                 GOTO(out_statfs, rc = -EINVAL);
2059
2060         rc = obd_ioctl_getdata(&buf, &len, arg);
2061         if (rc)
2062                 GOTO(out_statfs, rc);
2063
2064         data = (void*)buf;
2065         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2066             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2067                 GOTO(out_statfs, rc = -EINVAL);
2068
2069         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2070         if (type == LL_STATFS_MDC)
2071                 exp = sbi->ll_md_exp;
2072         else if (type == LL_STATFS_LOV)
2073                 exp = sbi->ll_dt_exp;
2074         else
2075                 GOTO(out_statfs, rc = -ENODEV);
2076
2077         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2078         if (rc)
2079                 GOTO(out_statfs, rc);
2080 out_statfs:
2081         if (buf)
2082                 obd_ioctl_freedata(buf, len);
2083         return rc;
2084 }
2085
2086 int ll_process_config(struct lustre_cfg *lcfg)
2087 {
2088         char *ptr;
2089         void *sb;
2090         struct lprocfs_static_vars lvars;
2091         unsigned long x;
2092         int rc = 0;
2093
2094         lprocfs_llite_init_vars(&lvars);
2095
2096         /* The instance name contains the sb: lustre-client-aacfe000 */
2097         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2098         if (!ptr || !*(++ptr))
2099                 return -EINVAL;
2100         if (sscanf(ptr, "%lx", &x) != 1)
2101                 return -EINVAL;
2102         sb = (void *)x;
2103         /* This better be a real Lustre superblock! */
2104         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2105
2106         /* Note we have not called client_common_fill_super yet, so
2107            proc fns must be able to handle that! */
2108         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2109                                       lcfg, sb);
2110         if (rc > 0)
2111                 rc = 0;
2112         return(rc);
2113 }
2114
2115 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2116 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2117                                        struct inode *i1, struct inode *i2,
2118                                        const char *name, int namelen,
2119                                        int mode, __u32 opc, void *data)
2120 {
2121         LASSERT(i1 != NULL);
2122
2123         if (namelen > ll_i2sbi(i1)->ll_namelen)
2124                 return ERR_PTR(-ENAMETOOLONG);
2125
2126         if (op_data == NULL)
2127                 OBD_ALLOC_PTR(op_data);
2128
2129         if (op_data == NULL)
2130                 return ERR_PTR(-ENOMEM);
2131
2132         ll_i2gids(op_data->op_suppgids, i1, i2);
2133         op_data->op_fid1 = *ll_inode2fid(i1);
2134         op_data->op_capa1 = ll_mdscapa_get(i1);
2135
2136         if (i2) {
2137                 op_data->op_fid2 = *ll_inode2fid(i2);
2138                 op_data->op_capa2 = ll_mdscapa_get(i2);
2139         } else {
2140                 fid_zero(&op_data->op_fid2);
2141                 op_data->op_capa2 = NULL;
2142         }
2143
2144         op_data->op_name = name;
2145         op_data->op_namelen = namelen;
2146         op_data->op_mode = mode;
2147         op_data->op_mod_time = cfs_time_current_sec();
2148         op_data->op_fsuid = current->fsuid;
2149         op_data->op_fsgid = current->fsgid;
2150         op_data->op_cap = cfs_curproc_cap_pack();
2151         op_data->op_bias = MDS_CHECK_SPLIT;
2152         op_data->op_opc = opc;
2153         op_data->op_mds = 0;
2154         op_data->op_data = data;
2155
2156         return op_data;
2157 }
2158
2159 void ll_finish_md_op_data(struct md_op_data *op_data)
2160 {
2161         capa_put(op_data->op_capa1);
2162         capa_put(op_data->op_capa2);
2163         OBD_FREE_PTR(op_data);
2164 }
2165
2166 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2167 {
2168         struct ll_sb_info *sbi;
2169
2170         LASSERT((seq != NULL) && (vfs != NULL));
2171         sbi = ll_s2sbi(vfs->mnt_sb);
2172
2173         if (sbi->ll_flags & LL_SBI_NOLCK)
2174                 seq_puts(seq, ",nolock");
2175
2176         if (sbi->ll_flags & LL_SBI_FLOCK)
2177                 seq_puts(seq, ",flock");
2178
2179         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2180                 seq_puts(seq, ",localflock");
2181
2182         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2183                 seq_puts(seq, ",user_xattr");
2184
2185         if (sbi->ll_flags & LL_SBI_ACL)
2186                 seq_puts(seq, ",acl");
2187
2188         RETURN(0);
2189 }