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  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002-2005 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <lustre_lite.h>
  32 #include <lustre_ha.h>
  33 #include <lustre_dlm.h>
  34 #include <lprocfs_status.h>
  35 #include <lustre_disk.h>
  36 #include <lustre_param.h>
  37 #include <lustre_log.h>
  38 #include <obd_cksum.h>
  39 #include "llite_internal.h"
  40
  41 cfs_mem_cache_t *ll_file_data_slab;
  42
  43 LIST_HEAD(ll_super_blocks);
  44 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  45
  46 extern struct address_space_operations ll_aops;
  47 extern struct address_space_operations ll_dir_aops;
  48
  49 #ifndef log2
  50 #define log2(n) ffz(~(n))
  51 #endif
  52
  53
  54 static struct ll_sb_info *ll_init_sbi(void)
  55 {
  56         struct ll_sb_info *sbi = NULL;
  57         class_uuid_t uuid;
  58         int i;
  59         ENTRY;
  60
  61         OBD_ALLOC(sbi, sizeof(*sbi));
  62         if (!sbi)
  63                 RETURN(NULL);
  64
  65         spin_lock_init(&sbi->ll_lock);
  66         spin_lock_init(&sbi->ll_lco.lco_lock);
  67         spin_lock_init(&sbi->ll_pp_extent_lock);
  68         spin_lock_init(&sbi->ll_process_lock);
  69         sbi->ll_rw_stats_on = 0;
  70         INIT_LIST_HEAD(&sbi->ll_pglist);
  71         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
  72                 sbi->ll_async_page_max = num_physpages / 2;
  73         else
  74                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  75         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  76                                            SBI_DEFAULT_READAHEAD_MAX);
  77         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  78                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  79
  80         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  81         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
  82
  83         ll_generate_random_uuid(uuid);
  84         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  85         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  86
  87         spin_lock(&ll_sb_lock);
  88         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  89         spin_unlock(&ll_sb_lock);
  90
  91 #ifdef ENABLE_LLITE_CHECKSUM
  92         sbi->ll_flags |= LL_SBI_CHECKSUM;
  93 #endif
  94
  95 #ifdef HAVE_LRU_RESIZE_SUPPORT
  96         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
  97 #endif
  98
  99 #ifdef HAVE_EXPORT___IGET
 100         INIT_LIST_HEAD(&sbi->ll_deathrow);
 101         spin_lock_init(&sbi->ll_deathrow_lock);
 102 #endif
 103         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 104                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 105                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 106         }
 107
 108         RETURN(sbi);
 109 }
 110
 111 void ll_free_sbi(struct super_block *sb)
 112 {
 113         struct ll_sb_info *sbi = ll_s2sbi(sb);
 114         ENTRY;
 115
 116         if (sbi != NULL) {
 117                 spin_lock(&ll_sb_lock);
 118                 list_del(&sbi->ll_list);
 119                 spin_unlock(&ll_sb_lock);
 120                 OBD_FREE(sbi, sizeof(*sbi));
 121         }
 122         EXIT;
 123 }
 124
 125 static struct dentry_operations ll_d_root_ops = {
 126 #ifdef DCACHE_LUSTRE_INVALID
 127         .d_compare = ll_dcompare,
 128 #endif
 129 };
 130
 131 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 132  * us to make MDS RPCs with large enough reply buffers to hold the
 133  * maximum-sized (= maximum striped) EA and cookie without having to
 134  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 135 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 136 {
 137         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 138         __u32 valsize = sizeof(struct lov_desc);
 139         int rc, easize, def_easize, cookiesize;
 140         struct lov_desc desc;
 141         __u32 stripes;
 142         ENTRY;
 143
 144         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 145                           &valsize, &desc);
 146         if (rc)
 147                 RETURN(rc);
 148
 149         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 150         lsm.lsm_stripe_count = stripes;
 151         easize = obd_size_diskmd(dt_exp, &lsm);
 152
 153         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 154         def_easize = obd_size_diskmd(dt_exp, &lsm);
 155
 156         cookiesize = stripes * sizeof(struct llog_cookie);
 157
 158         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 159                easize, cookiesize);
 160
 161         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 162         RETURN(rc);
 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 lu_fid rootfid;
 171         struct obd_capa *oc = NULL;
 172         struct obd_statfs osfs;
 173         struct ptlrpc_request *request = NULL;
 174         struct lustre_handle dt_conn = {0, };
 175         struct lustre_handle md_conn = {0, };
 176         struct obd_connect_data *data = NULL;
 177         struct lustre_md lmd;
 178         obd_valid valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         if (proc_lustre_fs_root) {
 193                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 194                                                   dt, md);
 195                 if (err < 0)
 196                         CERROR("could not register mount in /proc/lustre");
 197         }
 198
 199         /* indicate the features supported by this client */
 200         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 201                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 202                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 203                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 204                                   OBD_CONNECT_FID;
 205
 206 #ifdef HAVE_LRU_RESIZE_SUPPORT
 207         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 208                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 209 #endif
 210 #ifdef CONFIG_FS_POSIX_ACL
 211         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 212 #endif
 213         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 214         data->ocd_version = LUSTRE_VERSION_CODE;
 215
 216         if (sb->s_flags & MS_RDONLY)
 217                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 218         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 219                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 220
 221 #ifdef HAVE_MS_FLOCK_LOCK
 222         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 223         sb->s_flags |= MS_FLOCK_LOCK;
 224 #endif
 225
 226         if (sbi->ll_flags & LL_SBI_FLOCK)
 227                 sbi->ll_fop = &ll_file_operations_flock;
 228         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 229                 sbi->ll_fop = &ll_file_operations;
 230         else
 231                 sbi->ll_fop = &ll_file_operations_noflock;
 232
 233         /* real client */
 234         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 235         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 236                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 237                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 238         } else {
 239                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 240                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 241         }
 242
 243         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 244         if (err == -EBUSY) {
 245                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 246                                    "recovery, of which this client is not a "
 247                                    "part. Please wait for recovery to complete,"
 248                                    " abort, or time out.\n", md);
 249                 GOTO(out, err);
 250         } else if (err) {
 251                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 252                 GOTO(out, err);
 253         }
 254         sbi->ll_md_exp = class_conn2export(&md_conn);
 255
 256         err = obd_fid_init(sbi->ll_md_exp);
 257         if (err) {
 258                 CERROR("Can't init metadata layer FID infrastructure, "
 259                        "rc %d\n", err);
 260                 GOTO(out_md, err);
 261         }
 262
 263         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 264         if (err)
 265                 GOTO(out_md_fid, err);
 266
 267         size = sizeof(*data);
 268         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 269                            KEY_CONN_DATA,  &size, data);
 270         if (err) {
 271                 CERROR("Get connect data failed: %d \n", err);
 272                 GOTO(out_md, err);
 273         }
 274
 275         LASSERT(osfs.os_bsize);
 276         sb->s_blocksize = osfs.os_bsize;
 277         sb->s_blocksize_bits = log2(osfs.os_bsize);
 278         sb->s_magic = LL_SUPER_MAGIC;
 279
 280         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 281          *         retval = in_file->f_op->sendfile(...);
 282          *         if (*ppos > max)
 283          *                 retval = -EOVERFLOW;
 284          *
 285          * it will check if *ppos is greater than max. However, max equals to
 286          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 287          * has been defined as a signed long long ineger in linux kernel. */
 288 #if BITS_PER_LONG == 64
 289         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 290 #else
 291         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 292 #endif
 293         sbi->ll_namelen = osfs.os_namelen;
 294         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 295
 296         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 297             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 298                 LCONSOLE_INFO("Disabling user_xattr feature because "
 299                               "it is not supported on the server\n");
 300                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 301         }
 302
 303         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 304 #ifdef MS_POSIXACL
 305                 sb->s_flags |= MS_POSIXACL;
 306 #endif
 307                 sbi->ll_flags |= LL_SBI_ACL;
 308         } else {
 309                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 310 #ifdef MS_POSIXACL
 311                 sb->s_flags &= ~MS_POSIXACL;
 312 #endif
 313                 sbi->ll_flags &= ~LL_SBI_ACL;
 314         }
 315
 316         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 317                 sbi->ll_flags |= LL_SBI_JOIN;
 318
 319         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 320                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 321                         /* sometimes local client claims to be remote, but mdt
 322                          * will disagree when client gss not applied. */
 323                         LCONSOLE_INFO("client claims to be remote, but server "
 324                                       "rejected, forced to be local.\n");
 325                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 326                 }
 327         } else {
 328                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 329                         /* with gss applied, remote client can not claim to be
 330                          * local, so mdt maybe force client to be remote. */
 331                         LCONSOLE_INFO("client claims to be local, but server "
 332                                       "rejected, forced to be remote.\n");
 333                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 334                 }
 335         }
 336
 337         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 338                 LCONSOLE_INFO("client enabled MDS capability!\n");
 339                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 343                 LCONSOLE_INFO("client enabled OSS capability!\n");
 344                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 345         }
 346
 347         sbi->ll_sdev_orig = sb->s_dev;
 348 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 349         /* We set sb->s_dev equal on all lustre clients in order to support
 350          * NFS export clustering.  NFSD requires that the FSID be the same
 351          * on all clients. */
 352         /* s_dev is also used in lt_compare() to compare two fs, but that is
 353          * only a node-local comparison. */
 354
 355         /* XXX: this will not work with LMV */
 356         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 357                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 358 #endif
 359
 360         obd = class_name2obd(dt);
 361         if (!obd) {
 362                 CERROR("DT %s: not setup or attached\n", dt);
 363                 GOTO(out_md_fid, err = -ENODEV);
 364         }
 365
 366         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 367                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 368                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID;
 369         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 370                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 371
 372         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 373                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 374                  * disabled by default, because it can still be enabled on the
 375                  * fly via /proc. As a consequence, we still need to come to an
 376                  * agreement on the supported algorithms at connect time */
 377                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 378
 379                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 380                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 381                 else
 382                         /* send the list of supported checksum types */
 383                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 384         }
 385
 386 #ifdef HAVE_LRU_RESIZE_SUPPORT
 387         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 388 #endif
 389         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 390                "ocd_grant: %d\n", data->ocd_connect_flags,
 391                data->ocd_version, data->ocd_grant);
 392
 393         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 394         obd->obd_upcall.onu_upcall = ll_ocd_update;
 395         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 396
 397         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 398         if (err == -EBUSY) {
 399                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 400                                    "recovery, of which this client is not a "
 401                                    "part.  Please wait for recovery to "
 402                                    "complete, abort, or time out.\n", dt);
 403                 GOTO(out_md_fid, err);
 404         } else if (err) {
 405                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 406                 GOTO(out_md_fid, err);
 407         }
 408
 409         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 410
 411         err = obd_fid_init(sbi->ll_dt_exp);
 412         if (err) {
 413                 CERROR("Can't init data layer FID infrastructure, "
 414                        "rc %d\n", err);
 415                 GOTO(out_dt, err);
 416         }
 417
 418         spin_lock(&sbi->ll_lco.lco_lock);
 419         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 420         spin_unlock(&sbi->ll_lco.lco_lock);
 421
 422         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 423
 424         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 425                                   0, NULL, NULL, NULL);
 426         if (err < 0) {
 427                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 428                                    "filesystem. There must be at least one "
 429                                    "active OST for a client to start.\n");
 430                 GOTO(out_dt_fid, err);
 431         }
 432
 433         if (!ll_async_page_slab) {
 434                 ll_async_page_slab_size =
 435                         size_round(sizeof(struct ll_async_page)) + err;
 436                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 437                                                           ll_async_page_slab_size,
 438                                                           0, 0);
 439                 if (!ll_async_page_slab)
 440                         GOTO(out_dt_fid, err = -ENOMEM);
 441         }
 442
 443         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 444         if (err) {
 445                 CERROR("cannot mds_connect: rc = %d\n", err);
 446                 GOTO(out_dt_fid, err);
 447         }
 448         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 449         sbi->ll_root_fid = rootfid;
 450
 451         sb->s_op = &lustre_super_operations;
 452         sb->s_export_op = &lustre_export_operations;
 453
 454         /* make root inode
 455          * XXX: move this to after cbd setup? */
 456         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 457         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 458                 valid |= OBD_MD_FLRMTPERM;
 459         else if (sbi->ll_flags & LL_SBI_ACL)
 460                 valid |= OBD_MD_FLACL;
 461
 462         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 463         if (oc)
 464                 free_capa(oc);
 465         if (err) {
 466                 CERROR("md_getattr failed for root: rc = %d\n", err);
 467                 GOTO(out_dt_fid, err);
 468         }
 469         memset(&lmd, 0, sizeof(lmd));
 470         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 471                                sbi->ll_md_exp, &lmd);
 472         if (err) {
 473                 CERROR("failed to understand root inode md: rc = %d\n", err);
 474                 ptlrpc_req_finished (request);
 475                 GOTO(out_dt_fid, err);
 476         }
 477
 478         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 479         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 480         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 481         ptlrpc_req_finished(request);
 482
 483         if (root == NULL || is_bad_inode(root)) {
 484                 if (lmd.lsm)
 485                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 486 #ifdef CONFIG_FS_POSIX_ACL
 487                 if (lmd.posix_acl) {
 488                         posix_acl_release(lmd.posix_acl);
 489                         lmd.posix_acl = NULL;
 490                 }
 491 #endif
 492                 CERROR("lustre_lite: bad iget4 for root\n");
 493                 GOTO(out_root, err = -EBADF);
 494         }
 495
 496         err = ll_close_thread_start(&sbi->ll_lcq);
 497         if (err) {
 498                 CERROR("cannot start close thread: rc %d\n", err);
 499                 GOTO(out_root, err);
 500         }
 501
 502 #ifdef CONFIG_FS_POSIX_ACL
 503         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 504                 rct_init(&sbi->ll_rct);
 505                 et_init(&sbi->ll_et);
 506         }
 507 #endif
 508
 509         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 510         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 511                                  sizeof(checksum), &checksum, NULL);
 512
 513         sb->s_root = d_alloc_root(root);
 514         if (data != NULL)
 515                 OBD_FREE(data, sizeof(*data));
 516         sb->s_root->d_op = &ll_d_root_ops;
 517         RETURN(err);
 518 out_root:
 519         if (root)
 520                 iput(root);
 521 out_dt_fid:
 522         obd_fid_fini(sbi->ll_dt_exp);
 523 out_dt:
 524         obd_disconnect(sbi->ll_dt_exp);
 525         sbi->ll_dt_exp = NULL;
 526 out_md_fid:
 527         obd_fid_fini(sbi->ll_md_exp);
 528 out_md:
 529         obd_disconnect(sbi->ll_md_exp);
 530         sbi->ll_md_exp = NULL;
 531 out:
 532         if (data != NULL)
 533                 OBD_FREE_PTR(data);
 534         lprocfs_unregister_mountpoint(sbi);
 535         return err;
 536 }
 537
 538 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 539 {
 540         int size, rc;
 541
 542         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 543         size = sizeof(int);
 544         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 545                           &size, lmmsize);
 546         if (rc)
 547                 CERROR("Get max mdsize error rc %d \n", rc);
 548
 549         RETURN(rc);
 550 }
 551
 552 void ll_dump_inode(struct inode *inode)
 553 {
 554         struct list_head *tmp;
 555         int dentry_count = 0;
 556
 557         LASSERT(inode != NULL);
 558
 559         list_for_each(tmp, &inode->i_dentry)
 560                 dentry_count++;
 561
 562         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 563                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 564                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 565 }
 566
 567 void lustre_dump_dentry(struct dentry *dentry, int recur)
 568 {
 569         struct list_head *tmp;
 570         int subdirs = 0;
 571
 572         LASSERT(dentry != NULL);
 573
 574         list_for_each(tmp, &dentry->d_subdirs)
 575                 subdirs++;
 576
 577         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 578                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 579                dentry->d_name.len, dentry->d_name.name,
 580                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 581                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 582                dentry->d_flags, dentry->d_fsdata, subdirs);
 583         if (dentry->d_inode != NULL)
 584                 ll_dump_inode(dentry->d_inode);
 585
 586         if (recur == 0)
 587                 return;
 588
 589         list_for_each(tmp, &dentry->d_subdirs) {
 590                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 591                 lustre_dump_dentry(d, recur - 1);
 592         }
 593 }
 594
 595 #ifdef HAVE_EXPORT___IGET
 596 static void prune_dir_dentries(struct inode *inode)
 597 {
 598         struct dentry *dentry, *prev = NULL;
 599
 600         /* due to lustre specific logic, a directory
 601          * can have few dentries - a bug from VFS POV */
 602 restart:
 603         spin_lock(&dcache_lock);
 604         if (!list_empty(&inode->i_dentry)) {
 605                 dentry = list_entry(inode->i_dentry.prev,
 606                                     struct dentry, d_alias);
 607                 /* in order to prevent infinite loops we
 608                  * break if previous dentry is busy */
 609                 if (dentry != prev) {
 610                         prev = dentry;
 611                         dget_locked(dentry);
 612                         spin_unlock(&dcache_lock);
 613
 614                         /* try to kill all child dentries */
 615                         lock_dentry(dentry);
 616                         shrink_dcache_parent(dentry);
 617                         unlock_dentry(dentry);
 618                         dput(dentry);
 619
 620                         /* now try to get rid of current dentry */
 621                         d_prune_aliases(inode);
 622                         goto restart;
 623                 }
 624         }
 625         spin_unlock(&dcache_lock);
 626 }
 627
 628 static void prune_deathrow_one(struct ll_inode_info *lli)
 629 {
 630         struct inode *inode = ll_info2i(lli);
 631
 632         /* first, try to drop any dentries - they hold a ref on the inode */
 633         if (S_ISDIR(inode->i_mode))
 634                 prune_dir_dentries(inode);
 635         else
 636                 d_prune_aliases(inode);
 637
 638
 639         /* if somebody still uses it, leave it */
 640         LASSERT(atomic_read(&inode->i_count) > 0);
 641         if (atomic_read(&inode->i_count) > 1)
 642                 goto out;
 643
 644         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 645                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 646
 647         /* seems nobody uses it anymore */
 648         inode->i_nlink = 0;
 649
 650 out:
 651         iput(inode);
 652         return;
 653 }
 654
 655 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 656 {
 657         struct ll_inode_info *lli;
 658         int empty;
 659
 660         do {
 661                 if (need_resched() && try)
 662                         break;
 663
 664                 if (try) {
 665                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 666                                 break;
 667                 } else {
 668                         spin_lock(&sbi->ll_deathrow_lock);
 669                 }
 670
 671                 empty = 1;
 672                 lli = NULL;
 673                 if (!list_empty(&sbi->ll_deathrow)) {
 674                         lli = list_entry(sbi->ll_deathrow.next,
 675                                          struct ll_inode_info,
 676                                          lli_dead_list);
 677                         list_del_init(&lli->lli_dead_list);
 678                         if (!list_empty(&sbi->ll_deathrow))
 679                                 empty = 0;
 680                 }
 681                 spin_unlock(&sbi->ll_deathrow_lock);
 682
 683                 if (lli)
 684                         prune_deathrow_one(lli);
 685
 686         } while (empty == 0);
 687 }
 688 #else /* !HAVE_EXPORT___IGET */
 689 #define prune_deathrow(sbi, try) do {} while (0)
 690 #endif /* HAVE_EXPORT___IGET */
 691
 692 void client_common_put_super(struct super_block *sb)
 693 {
 694         struct ll_sb_info *sbi = ll_s2sbi(sb);
 695         ENTRY;
 696
 697 #ifdef CONFIG_FS_POSIX_ACL
 698         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 699                 et_fini(&sbi->ll_et);
 700                 rct_fini(&sbi->ll_rct);
 701         }
 702 #endif
 703
 704         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 705
 706         ll_close_thread_shutdown(sbi->ll_lcq);
 707
 708         /* destroy inodes in deathrow */
 709         prune_deathrow(sbi, 0);
 710
 711         list_del(&sbi->ll_conn_chain);
 712
 713         obd_fid_fini(sbi->ll_dt_exp);
 714         obd_disconnect(sbi->ll_dt_exp);
 715         sbi->ll_dt_exp = NULL;
 716
 717         lprocfs_unregister_mountpoint(sbi);
 718
 719         obd_fid_fini(sbi->ll_md_exp);
 720         obd_disconnect(sbi->ll_md_exp);
 721         sbi->ll_md_exp = NULL;
 722
 723         EXIT;
 724 }
 725
 726 void ll_kill_super(struct super_block *sb)
 727 {
 728         struct ll_sb_info *sbi;
 729
 730         ENTRY;
 731
 732         /* not init sb ?*/
 733         if (!(sb->s_flags & MS_ACTIVE))
 734                 return;
 735
 736         sbi = ll_s2sbi(sb);
 737         /* we need restore s_dev from changed for clustred NFS before put_super
 738          * because new kernels have cached s_dev and change sb->s_dev in
 739          * put_super not affected real removing devices */
 740         if (sbi)
 741                 sb->s_dev = sbi->ll_sdev_orig;
 742         EXIT;
 743 }
 744
 745 char *ll_read_opt(const char *opt, char *data)
 746 {
 747         char *value;
 748         char *retval;
 749         ENTRY;
 750
 751         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 752         if (strncmp(opt, data, strlen(opt)))
 753                 RETURN(NULL);
 754         if ((value = strchr(data, '=')) == NULL)
 755                 RETURN(NULL);
 756
 757         value++;
 758         OBD_ALLOC(retval, strlen(value) + 1);
 759         if (!retval) {
 760                 CERROR("out of memory!\n");
 761                 RETURN(NULL);
 762         }
 763
 764         memcpy(retval, value, strlen(value)+1);
 765         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 766         RETURN(retval);
 767 }
 768
 769 static inline int ll_set_opt(const char *opt, char *data, int fl)
 770 {
 771         if (strncmp(opt, data, strlen(opt)) != 0)
 772                 return(0);
 773         else
 774                 return(fl);
 775 }
 776
 777 /* non-client-specific mount options are parsed in lmd_parse */
 778 static int ll_options(char *options, int *flags)
 779 {
 780         int tmp;
 781         char *s1 = options, *s2;
 782         ENTRY;
 783
 784         if (!options)
 785                 RETURN(0);
 786
 787         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 788
 789         while (*s1) {
 790                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 791                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 822                 if (tmp) {
 823                         /* Ignore deprecated mount option.  The client will
 824                          * always try to mount with ACL support, whether this
 825                          * is used depends on whether server supports it. */
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 829                 if (tmp) {
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837
 838                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858
 859                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 860                                    s1);
 861                 RETURN(-EINVAL);
 862
 863 next:
 864                 /* Find next opt */
 865                 s2 = strchr(s1, ',');
 866                 if (s2 == NULL)
 867                         break;
 868                 s1 = s2 + 1;
 869         }
 870         RETURN(0);
 871 }
 872
 873 void ll_lli_init(struct ll_inode_info *lli)
 874 {
 875         lli->lli_inode_magic = LLI_INODE_MAGIC;
 876         sema_init(&lli->lli_size_sem, 1);
 877         sema_init(&lli->lli_write_sem, 1);
 878         lli->lli_flags = 0;
 879         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 880         spin_lock_init(&lli->lli_lock);
 881         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 882         INIT_LIST_HEAD(&lli->lli_close_list);
 883         lli->lli_inode_magic = LLI_INODE_MAGIC;
 884         sema_init(&lli->lli_och_sem, 1);
 885         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 886         lli->lli_mds_exec_och = NULL;
 887         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 888         lli->lli_open_fd_exec_count = 0;
 889         INIT_LIST_HEAD(&lli->lli_dead_list);
 890         lli->lli_remote_perms = NULL;
 891         lli->lli_rmtperm_utime = 0;
 892         sema_init(&lli->lli_rmtperm_sem, 1);
 893         INIT_LIST_HEAD(&lli->lli_oss_capas);
 894 }
 895
 896 int ll_fill_super(struct super_block *sb)
 897 {
 898         struct lustre_profile *lprof;
 899         struct lustre_sb_info *lsi = s2lsi(sb);
 900         struct ll_sb_info *sbi;
 901         char  *dt = NULL, *md = NULL;
 902         char  *profilenm = get_profile_name(sb);
 903         struct config_llog_instance cfg = {0, };
 904         char   ll_instance[sizeof(sb) * 2 + 1];
 905         int    err;
 906         ENTRY;
 907
 908         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 909
 910         cfs_module_get();
 911
 912         sb->s_type->fs_flags |= FS_RENAME_DOES_D_MOVE;
 913         /* client additional sb info */
 914         lsi->lsi_llsbi = sbi = ll_init_sbi();
 915         if (!sbi) {
 916                 cfs_module_put();
 917                 RETURN(-ENOMEM);
 918         }
 919
 920         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 921         if (err)
 922                 GOTO(out_free, err);
 923
 924         /* Generate a string unique to this super, in case some joker tries
 925            to mount the same fs at two mount points.
 926            Use the address of the super itself.*/
 927         sprintf(ll_instance, "%p", sb);
 928         cfg.cfg_instance = ll_instance;
 929         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 930
 931         /* set up client obds */
 932         err = lustre_process_log(sb, profilenm, &cfg);
 933         if (err < 0) {
 934                 CERROR("Unable to process log: %d\n", err);
 935                 GOTO(out_free, err);
 936         }
 937
 938         lprof = class_get_profile(profilenm);
 939         if (lprof == NULL) {
 940                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 941                                    " read from the MGS.  Does that filesystem "
 942                                    "exist?\n", profilenm);
 943                 GOTO(out_free, err = -EINVAL);
 944         }
 945         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 946                lprof->lp_md, lprof->lp_dt);
 947
 948         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 949                   strlen(ll_instance) + 2);
 950         if (!dt)
 951                 GOTO(out_free, err = -ENOMEM);
 952         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 953
 954         OBD_ALLOC(md, strlen(lprof->lp_md) +
 955                   strlen(ll_instance) + 2);
 956         if (!md)
 957                 GOTO(out_free, err = -ENOMEM);
 958         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 959
 960         /* connections, registrations, sb setup */
 961         err = client_common_fill_super(sb, md, dt);
 962
 963 out_free:
 964         if (md)
 965                 OBD_FREE(md, strlen(md) + 1);
 966         if (dt)
 967                 OBD_FREE(dt, strlen(dt) + 1);
 968         if (err)
 969                 ll_put_super(sb);
 970         else
 971                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 972
 973         RETURN(err);
 974 } /* ll_fill_super */
 975
 976
 977 void ll_put_super(struct super_block *sb)
 978 {
 979         struct config_llog_instance cfg;
 980         char   ll_instance[sizeof(sb) * 2 + 1];
 981         struct obd_device *obd;
 982         struct lustre_sb_info *lsi = s2lsi(sb);
 983         struct ll_sb_info *sbi = ll_s2sbi(sb);
 984         char *profilenm = get_profile_name(sb);
 985         int force = 1, next;
 986         ENTRY;
 987
 988         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 989
 990         ll_print_capa_stat(sbi);
 991
 992         sprintf(ll_instance, "%p", sb);
 993         cfg.cfg_instance = ll_instance;
 994         lustre_end_log(sb, NULL, &cfg);
 995
 996         if (sbi->ll_md_exp) {
 997                 obd = class_exp2obd(sbi->ll_md_exp);
 998                 if (obd)
 999                         force = obd->obd_force;
1000         }
1001
1002         /* We need to set force before the lov_disconnect in
1003            lustre_common_put_super, since l_d cleans up osc's as well. */
1004         if (force) {
1005                 next = 0;
1006                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1007                                                      &next)) != NULL) {
1008                         obd->obd_force = force;
1009                 }
1010         }
1011
1012         if (sbi->ll_lcq) {
1013                 /* Only if client_common_fill_super succeeded */
1014                 client_common_put_super(sb);
1015         }
1016         next = 0;
1017         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1018                 class_manual_cleanup(obd);
1019         }
1020
1021         if (profilenm)
1022                 class_del_profile(profilenm);
1023
1024         ll_free_sbi(sb);
1025         lsi->lsi_llsbi = NULL;
1026
1027         lustre_common_put_super(sb);
1028
1029         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1030
1031         cfs_module_put();
1032
1033         EXIT;
1034 } /* client_put_super */
1035
1036 #ifdef HAVE_REGISTER_CACHE
1037 #include <linux/cache_def.h>
1038 #ifdef HAVE_CACHE_RETURN_INT
1039 static int
1040 #else
1041 static void
1042 #endif
1043 ll_shrink_cache(int priority, unsigned int gfp_mask)
1044 {
1045         struct ll_sb_info *sbi;
1046         int count = 0;
1047
1048         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1049                 count += llap_shrink_cache(sbi, priority);
1050
1051 #ifdef HAVE_CACHE_RETURN_INT
1052         return count;
1053 #endif
1054 }
1055
1056 struct cache_definition ll_cache_definition = {
1057         .name = "llap_cache",
1058         .shrink = ll_shrink_cache
1059 };
1060 #endif /* HAVE_REGISTER_CACHE */
1061
1062 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1063 {
1064         struct inode *inode = NULL;
1065         /* NOTE: we depend on atomic igrab() -bzzz */
1066         lock_res_and_lock(lock);
1067         if (lock->l_ast_data) {
1068                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1069                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1070                         inode = igrab(lock->l_ast_data);
1071                 } else {
1072                         inode = lock->l_ast_data;
1073                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1074                                                 D_INFO : D_WARNING,
1075                                         lock, __FILE__, __func__, __LINE__,
1076                                         "l_ast_data %p is bogus: magic %08x",
1077                                         lock->l_ast_data, lli->lli_inode_magic);
1078                         inode = NULL;
1079                 }
1080         }
1081         unlock_res_and_lock(lock);
1082         return inode;
1083 }
1084
1085 static int null_if_equal(struct ldlm_lock *lock, void *data)
1086 {
1087         if (data == lock->l_ast_data) {
1088                 lock->l_ast_data = NULL;
1089
1090                 if (lock->l_req_mode != lock->l_granted_mode)
1091                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1092         }
1093
1094         return LDLM_ITER_CONTINUE;
1095 }
1096
1097 void ll_clear_inode(struct inode *inode)
1098 {
1099         struct ll_inode_info *lli = ll_i2info(inode);
1100         struct ll_sb_info *sbi = ll_i2sbi(inode);
1101         ENTRY;
1102
1103         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1104                inode->i_generation, inode);
1105
1106         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1107         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1108                          null_if_equal, inode);
1109
1110         LASSERT(!lli->lli_open_fd_write_count);
1111         LASSERT(!lli->lli_open_fd_read_count);
1112         LASSERT(!lli->lli_open_fd_exec_count);
1113
1114         if (lli->lli_mds_write_och)
1115                 ll_md_real_close(inode, FMODE_WRITE);
1116         if (lli->lli_mds_exec_och)
1117                 ll_md_real_close(inode, FMODE_EXEC);
1118         if (lli->lli_mds_read_och)
1119                 ll_md_real_close(inode, FMODE_READ);
1120
1121         if (lli->lli_smd) {
1122                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1123                                   null_if_equal, inode);
1124
1125                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1126                 lli->lli_smd = NULL;
1127         }
1128
1129         if (lli->lli_symlink_name) {
1130                 OBD_FREE(lli->lli_symlink_name,
1131                          strlen(lli->lli_symlink_name) + 1);
1132                 lli->lli_symlink_name = NULL;
1133         }
1134
1135         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1136                 LASSERT(lli->lli_posix_acl == NULL);
1137                 if (lli->lli_remote_perms) {
1138                         free_rmtperm_hash(lli->lli_remote_perms);
1139                         lli->lli_remote_perms = NULL;
1140                 }
1141         }
1142 #ifdef CONFIG_FS_POSIX_ACL
1143         else if (lli->lli_posix_acl) {
1144                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1145                 LASSERT(lli->lli_remote_perms == NULL);
1146                 posix_acl_release(lli->lli_posix_acl);
1147                 lli->lli_posix_acl = NULL;
1148         }
1149 #endif
1150         lli->lli_inode_magic = LLI_INODE_DEAD;
1151
1152 #ifdef HAVE_EXPORT___IGET
1153         spin_lock(&sbi->ll_deathrow_lock);
1154         list_del_init(&lli->lli_dead_list);
1155         spin_unlock(&sbi->ll_deathrow_lock);
1156 #endif
1157         ll_clear_inode_capas(inode);
1158
1159         EXIT;
1160 }
1161
1162 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1163                   struct md_open_data **mod)
1164 {
1165         struct lustre_md md;
1166         struct ll_sb_info *sbi = ll_i2sbi(inode);
1167         struct ptlrpc_request *request = NULL;
1168         int rc;
1169         ENTRY;
1170
1171         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1172                                      LUSTRE_OPC_ANY, NULL);
1173         if (IS_ERR(op_data))
1174                 RETURN(PTR_ERR(op_data));
1175
1176         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1177                         &request, mod);
1178         if (rc) {
1179                 ptlrpc_req_finished(request);
1180                 if (rc == -ENOENT) {
1181                         inode->i_nlink = 0;
1182                         /* Unlinked special device node? Or just a race?
1183                          * Pretend we done everything. */
1184                         if (!S_ISREG(inode->i_mode) &&
1185                             !S_ISDIR(inode->i_mode))
1186                                 rc = inode_setattr(inode, &op_data->op_attr);
1187                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1188                         CERROR("md_setattr fails: rc = %d\n", rc);
1189                 }
1190                 RETURN(rc);
1191         }
1192
1193         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1194                               sbi->ll_md_exp, &md);
1195         if (rc) {
1196                 ptlrpc_req_finished(request);
1197                 RETURN(rc);
1198         }
1199
1200         /* We call inode_setattr to adjust timestamps.
1201          * If there is at least some data in file, we cleared ATTR_SIZE
1202          * above to avoid invoking vmtruncate, otherwise it is important
1203          * to call vmtruncate in inode_setattr to update inode->i_size
1204          * (bug 6196) */
1205         rc = inode_setattr(inode, &op_data->op_attr);
1206
1207         /* Extract epoch data if obtained. */
1208         op_data->op_handle = md.body->handle;
1209         op_data->op_ioepoch = md.body->ioepoch;
1210
1211         ll_update_inode(inode, &md);
1212         ptlrpc_req_finished(request);
1213
1214         RETURN(rc);
1215 }
1216
1217 /* Close IO epoch and send Size-on-MDS attribute update. */
1218 static int ll_setattr_done_writing(struct inode *inode,
1219                                    struct md_op_data *op_data,
1220                                    struct md_open_data *mod)
1221 {
1222         struct ll_inode_info *lli = ll_i2info(inode);
1223         int rc = 0;
1224         ENTRY;
1225
1226         LASSERT(op_data != NULL);
1227         if (!S_ISREG(inode->i_mode))
1228                 RETURN(0);
1229
1230         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1231                op_data->op_ioepoch, PFID(&lli->lli_fid));
1232
1233         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1234         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1235         if (rc == -EAGAIN) {
1236                 /* MDS has instructed us to obtain Size-on-MDS attribute
1237                  * from OSTs and send setattr to back to MDS. */
1238                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1239                                          op_data->op_ioepoch);
1240         } else if (rc) {
1241                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1242                        inode->i_ino, rc);
1243         }
1244         RETURN(rc);
1245 }
1246
1247 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1248  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1249  * keep these values until such a time that objects are allocated for it.
1250  * We do the MDS operations first, as it is checking permissions for us.
1251  * We don't to the MDS RPC if there is nothing that we want to store there,
1252  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1253  * going to do an RPC anyways.
1254  *
1255  * If we are doing a truncate, we will send the mtime and ctime updates
1256  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1257  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1258  * at the same time.
1259  */
1260 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1261 {
1262         struct ll_inode_info *lli = ll_i2info(inode);
1263         struct lov_stripe_md *lsm = lli->lli_smd;
1264         struct ll_sb_info *sbi = ll_i2sbi(inode);
1265         struct md_op_data *op_data = NULL;
1266         struct md_open_data *mod = NULL;
1267         int ia_valid = attr->ia_valid;
1268         int rc = 0, rc1 = 0;
1269         ENTRY;
1270
1271         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1272                attr->ia_valid);
1273         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1274
1275         if (ia_valid & ATTR_SIZE) {
1276                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1277                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1278                                attr->ia_size, ll_file_maxbytes(inode));
1279                         RETURN(-EFBIG);
1280                 }
1281
1282                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1283         }
1284
1285         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1286         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1287                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1288                         RETURN(-EPERM);
1289         }
1290
1291         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1292         if (attr->ia_valid & ATTR_CTIME) {
1293                 attr->ia_ctime = CURRENT_TIME;
1294                 attr->ia_valid |= ATTR_CTIME_SET;
1295         }
1296         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1297                 attr->ia_atime = CURRENT_TIME;
1298                 attr->ia_valid |= ATTR_ATIME_SET;
1299         }
1300         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1301                 attr->ia_mtime = CURRENT_TIME;
1302                 attr->ia_valid |= ATTR_MTIME_SET;
1303         }
1304         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1305                 /* To avoid stale mtime on mds, obtain it from ost and send
1306                    to mds. */
1307                 rc = ll_glimpse_size(inode, 0);
1308                 if (rc)
1309                         RETURN(rc);
1310
1311                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1312                 attr->ia_mtime = inode->i_mtime;
1313         }
1314
1315         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1316                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1317                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1318                        CURRENT_SECONDS);
1319
1320         /* NB: ATTR_SIZE will only be set after this point if the size
1321          * resides on the MDS, ie, this file has no objects. */
1322         if (lsm)
1323                 attr->ia_valid &= ~ATTR_SIZE;
1324
1325         /* We always do an MDS RPC, even if we're only changing the size;
1326          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1327
1328         OBD_ALLOC_PTR(op_data);
1329         if (op_data == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1333
1334         /* Open epoch for truncate. */
1335         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1336             (ia_valid & ATTR_SIZE))
1337                 op_data->op_flags = MF_EPOCH_OPEN;
1338
1339         rc = ll_md_setattr(inode, op_data, &mod);
1340         if (rc)
1341                 GOTO(out, rc);
1342
1343         if (op_data->op_ioepoch)
1344                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1345                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1346
1347         if (!lsm || !S_ISREG(inode->i_mode)) {
1348                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1349                 GOTO(out, rc = 0);
1350         }
1351
1352         /* We really need to get our PW lock before we change inode->i_size.
1353          * If we don't we can race with other i_size updaters on our node, like
1354          * ll_file_read.  We can also race with i_size propogation to other
1355          * nodes through dirtying and writeback of final cached pages.  This
1356          * last one is especially bad for racing o_append users on other
1357          * nodes. */
1358         if (ia_valid & ATTR_SIZE) {
1359                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1360                                                            OBD_OBJECT_EOF } };
1361                 struct lustre_handle lockh = { 0 };
1362                 int err, ast_flags = 0;
1363                 /* XXX when we fix the AST intents to pass the discard-range
1364                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1365                  * XXX here. */
1366                 if (attr->ia_size == 0)
1367                         ast_flags = LDLM_AST_DISCARD_DATA;
1368
1369                 UNLOCK_INODE_MUTEX(inode);
1370                 UP_WRITE_I_ALLOC_SEM(inode);
1371                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1372                                     ast_flags);
1373                 LOCK_INODE_MUTEX(inode);
1374                 DOWN_WRITE_I_ALLOC_SEM(inode);
1375
1376                 if (rc != 0)
1377                         GOTO(out, rc);
1378
1379                 /* Only ll_inode_size_lock is taken at this level.
1380                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1381                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1382                  * ll_truncate dropped ll_inode_size_lock() */
1383                 ll_inode_size_lock(inode, 0);
1384                 rc = vmtruncate(inode, attr->ia_size);
1385                 if (rc != 0) {
1386                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1387                         ll_inode_size_unlock(inode, 0);
1388                 }
1389
1390                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1391                 if (err) {
1392                         CERROR("ll_extent_unlock failed: %d\n", err);
1393                         if (!rc)
1394                                 rc = err;
1395                 }
1396         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1397                 obd_flag flags;
1398                 struct obd_info oinfo = { { { 0 } } };
1399                 struct obdo *oa;
1400
1401                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1402                        inode->i_ino, LTIME_S(attr->ia_mtime));
1403
1404                 OBDO_ALLOC(oa);
1405                 if (oa) {
1406                         oa->o_id = lsm->lsm_object_id;
1407                         oa->o_gr = lsm->lsm_object_gr;
1408                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1409
1410                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1411                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1412                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1413                                 OBD_MD_FLGROUP;
1414
1415                         obdo_from_inode(oa, inode, flags);
1416
1417                         oinfo.oi_oa = oa;
1418                         oinfo.oi_md = lsm;
1419                         oinfo.oi_capa = ll_mdscapa_get(inode);
1420
1421                         /* XXX: this looks unnecessary now. */
1422                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1423                         capa_put(oinfo.oi_capa);
1424                         if (rc)
1425                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1426                         OBDO_FREE(oa);
1427                 } else {
1428                         rc = -ENOMEM;
1429                 }
1430         }
1431         EXIT;
1432 out:
1433         if (op_data) {
1434                 if (op_data->op_ioepoch)
1435                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1436                 ll_finish_md_op_data(op_data);
1437         }
1438         return rc ? rc : rc1;
1439 }
1440
1441 int ll_setattr(struct dentry *de, struct iattr *attr)
1442 {
1443         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1444             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1445                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1446
1447         return ll_setattr_raw(de->d_inode, attr);
1448 }
1449
1450 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1451                        __u64 max_age, __u32 flags)
1452 {
1453         struct ll_sb_info *sbi = ll_s2sbi(sb);
1454         struct obd_statfs obd_osfs;
1455         int rc;
1456         ENTRY;
1457
1458         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1459         if (rc) {
1460                 CERROR("md_statfs fails: rc = %d\n", rc);
1461                 RETURN(rc);
1462         }
1463
1464         osfs->os_type = sb->s_magic;
1465
1466         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1467                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1468
1469         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1470                               &obd_osfs, max_age, flags);
1471         if (rc) {
1472                 CERROR("obd_statfs fails: rc = %d\n", rc);
1473                 RETURN(rc);
1474         }
1475
1476         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1477                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1478                obd_osfs.os_files);
1479
1480         osfs->os_bsize = obd_osfs.os_bsize;
1481         osfs->os_blocks = obd_osfs.os_blocks;
1482         osfs->os_bfree = obd_osfs.os_bfree;
1483         osfs->os_bavail = obd_osfs.os_bavail;
1484
1485         /* If we don't have as many objects free on the OST as inodes
1486          * on the MDS, we reduce the total number of inodes to
1487          * compensate, so that the "inodes in use" number is correct.
1488          */
1489         if (obd_osfs.os_ffree < osfs->os_ffree) {
1490                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1491                         obd_osfs.os_ffree;
1492                 osfs->os_ffree = obd_osfs.os_ffree;
1493         }
1494
1495         RETURN(rc);
1496 }
1497 #ifndef HAVE_STATFS_DENTRY_PARAM
1498 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1499 {
1500 #else
1501 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1502 {
1503         struct super_block *sb = de->d_sb;
1504 #endif
1505         struct obd_statfs osfs;
1506         int rc;
1507
1508         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1509         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1510
1511         /* For now we will always get up-to-date statfs values, but in the
1512          * future we may allow some amount of caching on the client (e.g.
1513          * from QOS or lprocfs updates). */
1514         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1515         if (rc)
1516                 return rc;
1517
1518         statfs_unpack(sfs, &osfs);
1519
1520         /* We need to downshift for all 32-bit kernels, because we can't
1521          * tell if the kernel is being called via sys_statfs64() or not.
1522          * Stop before overflowing f_bsize - in which case it is better
1523          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1524         if (sizeof(long) < 8) {
1525                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1526                         sfs->f_bsize <<= 1;
1527
1528                         osfs.os_blocks >>= 1;
1529                         osfs.os_bfree >>= 1;
1530                         osfs.os_bavail >>= 1;
1531                 }
1532         }
1533
1534         sfs->f_blocks = osfs.os_blocks;
1535         sfs->f_bfree = osfs.os_bfree;
1536         sfs->f_bavail = osfs.os_bavail;
1537
1538         return 0;
1539 }
1540
1541 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1542 {
1543         struct ll_inode_info *lli;
1544         struct lov_stripe_md *lsm;
1545
1546         lli = ll_i2info(inode);
1547         LASSERT(lli->lli_size_sem_owner != current);
1548         down(&lli->lli_size_sem);
1549         LASSERT(lli->lli_size_sem_owner == NULL);
1550         lli->lli_size_sem_owner = current;
1551         lsm = lli->lli_smd;
1552         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1553                  lsm, lock_lsm);
1554         if (lock_lsm)
1555                 lov_stripe_lock(lsm);
1556 }
1557
1558 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1559 {
1560         struct ll_inode_info *lli;
1561         struct lov_stripe_md *lsm;
1562
1563         lli = ll_i2info(inode);
1564         lsm = lli->lli_smd;
1565         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1566                  lsm, unlock_lsm);
1567         if (unlock_lsm)
1568                 lov_stripe_unlock(lsm);
1569         LASSERT(lli->lli_size_sem_owner == current);
1570         lli->lli_size_sem_owner = NULL;
1571         up(&lli->lli_size_sem);
1572 }
1573
1574 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1575 {
1576         struct ll_inode_info *lli = ll_i2info(inode);
1577
1578         dump_lsm(D_INODE, lsm);
1579         dump_lsm(D_INODE, lli->lli_smd);
1580         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1581                  "lsm must be joined lsm %p\n", lsm);
1582         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1583         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1584                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1585         lli->lli_smd = lsm;
1586         lli->lli_maxbytes = lsm->lsm_maxbytes;
1587         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1588                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1589 }
1590
1591 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1592 {
1593         struct ll_inode_info *lli = ll_i2info(inode);
1594         struct mdt_body *body = md->body;
1595         struct lov_stripe_md *lsm = md->lsm;
1596         struct ll_sb_info *sbi = ll_i2sbi(inode);
1597
1598         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1599         if (lsm != NULL) {
1600                 if (lli->lli_smd == NULL) {
1601                         if (lsm->lsm_magic != LOV_MAGIC &&
1602                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1603                                 dump_lsm(D_ERROR, lsm);
1604                                 LBUG();
1605                         }
1606                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1607                                lsm, inode->i_ino, inode->i_generation, inode);
1608                         /* ll_inode_size_lock() requires it is only called
1609                          * with lli_smd != NULL or lock_lsm == 0 or we can
1610                          * race between lock/unlock.  bug 9547 */
1611                         lli->lli_smd = lsm;
1612                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1613                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1614                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1615                 } else {
1616                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1617                              lli->lli_smd->lsm_stripe_count ==
1618                                         lsm->lsm_stripe_count) {
1619                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1620                                         CERROR("lsm mismatch for inode %ld\n",
1621                                                 inode->i_ino);
1622                                         CERROR("lli_smd:\n");
1623                                         dump_lsm(D_ERROR, lli->lli_smd);
1624                                         CERROR("lsm:\n");
1625                                         dump_lsm(D_ERROR, lsm);
1626                                         LBUG();
1627                                 }
1628                         } else
1629                                 ll_replace_lsm(inode, lsm);
1630                 }
1631                 if (lli->lli_smd != lsm)
1632                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1633         }
1634
1635         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1636                 if (body->valid & OBD_MD_FLRMTPERM)
1637                         ll_update_remote_perm(inode, md->remote_perm);
1638         }
1639 #ifdef CONFIG_FS_POSIX_ACL
1640         else if (body->valid & OBD_MD_FLACL) {
1641                 spin_lock(&lli->lli_lock);
1642                 if (lli->lli_posix_acl)
1643                         posix_acl_release(lli->lli_posix_acl);
1644                 lli->lli_posix_acl = md->posix_acl;
1645                 spin_unlock(&lli->lli_lock);
1646         }
1647 #endif
1648         if (body->valid & OBD_MD_FLATIME &&
1649             body->atime > LTIME_S(inode->i_atime))
1650                 LTIME_S(inode->i_atime) = body->atime;
1651
1652         /* mtime is always updated with ctime, but can be set in past.
1653            As write and utime(2) may happen within 1 second, and utime's
1654            mtime has a priority over write's one, so take mtime from mds
1655            for the same ctimes. */
1656         if (body->valid & OBD_MD_FLCTIME &&
1657             body->ctime >= LTIME_S(inode->i_ctime)) {
1658                 LTIME_S(inode->i_ctime) = body->ctime;
1659                 if (body->valid & OBD_MD_FLMTIME) {
1660                         CDEBUG(D_INODE, "setting ino %lu mtime "
1661                                "from %lu to "LPU64"\n", inode->i_ino,
1662                                LTIME_S(inode->i_mtime), body->mtime);
1663                         LTIME_S(inode->i_mtime) = body->mtime;
1664                 }
1665         }
1666         if (body->valid & OBD_MD_FLMODE)
1667                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1668         if (body->valid & OBD_MD_FLTYPE)
1669                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1670         if (S_ISREG(inode->i_mode)) {
1671                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1672         } else {
1673                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1674         }
1675 #ifdef HAVE_INODE_BLKSIZE
1676         inode->i_blksize = 1<<inode->i_blkbits;
1677 #endif
1678         if (body->valid & OBD_MD_FLUID)
1679                 inode->i_uid = body->uid;
1680         if (body->valid & OBD_MD_FLGID)
1681                 inode->i_gid = body->gid;
1682         if (body->valid & OBD_MD_FLFLAGS)
1683                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1684         if (body->valid & OBD_MD_FLNLINK)
1685                 inode->i_nlink = body->nlink;
1686         if (body->valid & OBD_MD_FLRDEV)
1687                 inode->i_rdev = old_decode_dev(body->rdev);
1688
1689         if (body->valid & OBD_MD_FLID) {
1690                 /* FID shouldn't be changed! */
1691                 if (fid_is_sane(&lli->lli_fid)) {
1692                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1693                                  "Trying to change FID "DFID
1694                                  " to the "DFID", inode %lu/%u(%p)\n",
1695                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1696                                  inode->i_ino, inode->i_generation, inode);
1697                 } else
1698                         lli->lli_fid = body->fid1;
1699         }
1700
1701         LASSERT(fid_seq(&lli->lli_fid) != 0);
1702
1703         if (body->valid & OBD_MD_FLSIZE) {
1704                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1705                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1706                         struct lustre_handle lockh;
1707                         ldlm_mode_t mode;
1708
1709                         /* As it is possible a blocking ast has been processed
1710                          * by this time, we need to check there is an UPDATE
1711                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1712                          * it. */
1713                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1714                                                &lockh);
1715                         if (mode) {
1716                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1717                                                       LLIF_EPOCH_PENDING |
1718                                                       LLIF_SOM_DIRTY)) {
1719                                         CERROR("ino %lu flags %lu still has "
1720                                                "size authority! do not trust "
1721                                                "the size got from MDS\n",
1722                                                inode->i_ino, lli->lli_flags);
1723                                 } else {
1724                                         /* Use old size assignment to avoid
1725                                          * deadlock bz14138 & bz14326 */
1726                                         inode->i_size = body->size;
1727                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1728                                 }
1729                                 ldlm_lock_decref(&lockh, mode);
1730                         }
1731                 } else {
1732                         /* Use old size assignment to avoid
1733                          * deadlock bz14138 & bz14326 */
1734                         inode->i_size = body->size;
1735                 }
1736
1737                 if (body->valid & OBD_MD_FLBLOCKS)
1738                         inode->i_blocks = body->blocks;
1739         }
1740
1741         if (body->valid & OBD_MD_FLMDSCAPA) {
1742                 LASSERT(md->mds_capa);
1743                 ll_add_capa(inode, md->mds_capa);
1744         }
1745         if (body->valid & OBD_MD_FLOSSCAPA) {
1746                 LASSERT(md->oss_capa);
1747                 ll_add_capa(inode, md->oss_capa);
1748         }
1749 }
1750
1751 static struct backing_dev_info ll_backing_dev_info = {
1752         .ra_pages       = 0,    /* No readahead */
1753 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1754         .capabilities   = 0,    /* Does contribute to dirty memory */
1755 #else
1756         .memory_backed  = 0,    /* Does contribute to dirty memory */
1757 #endif
1758 };
1759
1760 void ll_read_inode2(struct inode *inode, void *opaque)
1761 {
1762         struct lustre_md *md = opaque;
1763         struct ll_inode_info *lli = ll_i2info(inode);
1764         ENTRY;
1765
1766         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1767                inode->i_ino, inode->i_generation, inode);
1768
1769         ll_lli_init(lli);
1770
1771         LASSERT(!lli->lli_smd);
1772
1773         /* Core attributes from the MDS first.  This is a new inode, and
1774          * the VFS doesn't zero times in the core inode so we have to do
1775          * it ourselves.  They will be overwritten by either MDS or OST
1776          * attributes - we just need to make sure they aren't newer. */
1777         LTIME_S(inode->i_mtime) = 0;
1778         LTIME_S(inode->i_atime) = 0;
1779         LTIME_S(inode->i_ctime) = 0;
1780         inode->i_rdev = 0;
1781         ll_update_inode(inode, md);
1782
1783         /* OIDEBUG(inode); */
1784
1785         if (S_ISREG(inode->i_mode)) {
1786                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1787                 inode->i_op = &ll_file_inode_operations;
1788                 inode->i_fop = sbi->ll_fop;
1789                 inode->i_mapping->a_ops = &ll_aops;
1790                 EXIT;
1791         } else if (S_ISDIR(inode->i_mode)) {
1792                 inode->i_op = &ll_dir_inode_operations;
1793                 inode->i_fop = &ll_dir_operations;
1794                 inode->i_mapping->a_ops = &ll_dir_aops;
1795                 EXIT;
1796         } else if (S_ISLNK(inode->i_mode)) {
1797                 inode->i_op = &ll_fast_symlink_inode_operations;
1798                 EXIT;
1799         } else {
1800                 inode->i_op = &ll_special_inode_operations;
1801
1802                 init_special_inode(inode, inode->i_mode,
1803                                    kdev_t_to_nr(inode->i_rdev));
1804
1805                 /* initializing backing dev info. */
1806                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1807
1808                 EXIT;
1809         }
1810 }
1811
1812 void ll_delete_inode(struct inode *inode)
1813 {
1814         struct ll_sb_info *sbi = ll_i2sbi(inode);
1815         int rc;
1816         ENTRY;
1817
1818         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1819         if (rc) {
1820                 CERROR("fid_delete() failed, rc %d\n", rc);
1821         }
1822         truncate_inode_pages(&inode->i_data, 0);
1823         clear_inode(inode);
1824
1825         EXIT;
1826 }
1827
1828 int ll_iocontrol(struct inode *inode, struct file *file,
1829                  unsigned int cmd, unsigned long arg)
1830 {
1831         struct ll_sb_info *sbi = ll_i2sbi(inode);
1832         struct ptlrpc_request *req = NULL;
1833         int rc, flags = 0;
1834         ENTRY;
1835
1836         switch(cmd) {
1837         case EXT3_IOC_GETFLAGS: {
1838                 struct mdt_body *body;
1839                 struct obd_capa *oc;
1840
1841                 oc = ll_mdscapa_get(inode);
1842                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1843                                 OBD_MD_FLFLAGS, 0, &req);
1844                 capa_put(oc);
1845                 if (rc) {
1846                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1847                         RETURN(-abs(rc));
1848                 }
1849
1850                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1851
1852                 flags = body->flags;
1853
1854                 ptlrpc_req_finished(req);
1855
1856                 RETURN(put_user(flags, (int *)arg));
1857         }
1858         case EXT3_IOC_SETFLAGS: {
1859                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1860                 struct obd_info oinfo = { { { 0 } } };
1861                 struct md_op_data *op_data;
1862
1863                 if (get_user(flags, (int *)arg))
1864                         RETURN(-EFAULT);
1865
1866                 oinfo.oi_md = lsm;
1867                 OBDO_ALLOC(oinfo.oi_oa);
1868                 if (!oinfo.oi_oa)
1869                         RETURN(-ENOMEM);
1870
1871                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1872                                              LUSTRE_OPC_ANY, NULL);
1873                 if (IS_ERR(op_data))
1874                         RETURN(PTR_ERR(op_data));
1875
1876                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1877                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1878                 rc = md_setattr(sbi->ll_md_exp, op_data,
1879                                 NULL, 0, NULL, 0, &req, NULL);
1880                 ll_finish_md_op_data(op_data);
1881                 ptlrpc_req_finished(req);
1882                 if (rc || lsm == NULL) {
1883                         OBDO_FREE(oinfo.oi_oa);
1884                         RETURN(rc);
1885                 }
1886
1887                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1888                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1889                 oinfo.oi_oa->o_flags = flags;
1890                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1891                                        OBD_MD_FLGROUP;
1892                 oinfo.oi_capa = ll_mdscapa_get(inode);
1893
1894                 obdo_from_inode(oinfo.oi_oa, inode,
1895                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1896                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1897                 capa_put(oinfo.oi_capa);
1898                 OBDO_FREE(oinfo.oi_oa);
1899                 if (rc) {
1900                         if (rc != -EPERM && rc != -EACCES)
1901                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1902                         RETURN(rc);
1903                 }
1904
1905                 inode->i_flags = ll_ext_to_inode_flags(flags |
1906                                                        MDS_BFLAG_EXT_FLAGS);
1907                 RETURN(0);
1908         }
1909         default:
1910                 RETURN(-ENOSYS);
1911         }
1912
1913         RETURN(0);
1914 }
1915
1916 int ll_flush_ctx(struct inode *inode)
1917 {
1918         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1919
1920         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1921
1922         obd_set_info_async(sbi->ll_md_exp,
1923                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1924                            0, NULL, NULL);
1925         obd_set_info_async(sbi->ll_dt_exp,
1926                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1927                            0, NULL, NULL);
1928         return 0;
1929 }
1930
1931 /* umount -f client means force down, don't save state */
1932 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1933 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1934 {
1935         struct super_block *sb = vfsmnt->mnt_sb;
1936 #else
1937 void ll_umount_begin(struct super_block *sb)
1938 {
1939 #endif
1940         struct lustre_sb_info *lsi = s2lsi(sb);
1941         struct ll_sb_info *sbi = ll_s2sbi(sb);
1942         struct obd_device *obd;
1943         struct obd_ioctl_data ioc_data = { 0 };
1944         ENTRY;
1945
1946 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1947         if (!(flags & MNT_FORCE)) {
1948                 EXIT;
1949                 return;
1950         }
1951 #endif
1952
1953         /* Tell the MGC we got umount -f */
1954         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1955
1956         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1957                sb->s_count, atomic_read(&sb->s_active));
1958
1959         obd = class_exp2obd(sbi->ll_md_exp);
1960         if (obd == NULL) {
1961                 CERROR("Invalid MDC connection handle "LPX64"\n",
1962                        sbi->ll_md_exp->exp_handle.h_cookie);
1963                 EXIT;
1964                 return;
1965         }
1966         obd->obd_force = 1;
1967         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1968                       &ioc_data, NULL);
1969
1970         obd = class_exp2obd(sbi->ll_dt_exp);
1971         if (obd == NULL) {
1972                 CERROR("Invalid LOV connection handle "LPX64"\n",
1973                        sbi->ll_dt_exp->exp_handle.h_cookie);
1974                 EXIT;
1975                 return;
1976         }
1977
1978         obd->obd_force = 1;
1979         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1980                       &ioc_data, NULL);
1981
1982         /* Really, we'd like to wait until there are no requests outstanding,
1983          * and then continue.  For now, we just invalidate the requests,
1984          * schedule, and hope.
1985          */
1986         schedule();
1987
1988         EXIT;
1989 }
1990
1991 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1992 {
1993         struct ll_sb_info *sbi = ll_s2sbi(sb);
1994         int err;
1995         __u32 read_only;
1996
1997         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1998                 read_only = *flags & MS_RDONLY;
1999                 err = obd_set_info_async(sbi->ll_md_exp,
2000                                          sizeof(KEY_READ_ONLY) - 1,
2001                                          KEY_READ_ONLY, sizeof(read_only),
2002                                          &read_only, NULL);
2003                 if (err) {
2004                         CERROR("Failed to change the read-only flag during "
2005                                "remount: %d\n", err);
2006                         return err;
2007                 }
2008
2009                 if (read_only)
2010                         sb->s_flags |= MS_RDONLY;
2011                 else
2012                         sb->s_flags &= ~MS_RDONLY;
2013         }
2014         return 0;
2015 }
2016
2017 int ll_prep_inode(struct inode **inode,
2018                   struct ptlrpc_request *req,
2019                   struct super_block *sb)
2020 {
2021         struct ll_sb_info *sbi = NULL;
2022         struct lustre_md md;
2023         int rc = 0;
2024         ENTRY;
2025
2026         LASSERT(*inode || sb);
2027         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2028         prune_deathrow(sbi, 1);
2029         memset(&md, 0, sizeof(struct lustre_md));
2030
2031         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2032                               sbi->ll_md_exp, &md);
2033         if (rc)
2034                 RETURN(rc);
2035
2036         if (*inode) {
2037                 ll_update_inode(*inode, &md);
2038         } else {
2039                 LASSERT(sb != NULL);
2040
2041                 /*
2042                  * At this point server returns to client's same fid as client
2043                  * generated for creating. So using ->fid1 is okay here.
2044                  */
2045                 LASSERT(fid_is_sane(&md.body->fid1));
2046
2047                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2048                 if (*inode == NULL || is_bad_inode(*inode)) {
2049                         if (md.lsm)
2050                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2051 #ifdef CONFIG_FS_POSIX_ACL
2052                         if (md.posix_acl) {
2053                                 posix_acl_release(md.posix_acl);
2054                                 md.posix_acl = NULL;
2055                         }
2056 #endif
2057                         rc = -ENOMEM;
2058                         CERROR("new_inode -fatal: rc %d\n", rc);
2059                         GOTO(out, rc);
2060                 }
2061         }
2062
2063         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2064                          ll_i2info(*inode)->lli_smd);
2065 out:
2066         md_free_lustre_md(sbi->ll_md_exp, &md);
2067         RETURN(rc);
2068 }
2069
2070 char *llap_origins[] = {
2071         [LLAP_ORIGIN_UNKNOWN] = "--",
2072         [LLAP_ORIGIN_READPAGE] = "rp",
2073         [LLAP_ORIGIN_READAHEAD] = "ra",
2074         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2075         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2076 };
2077
2078 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2079                                              struct list_head *list)
2080 {
2081         struct ll_async_page *llap;
2082         struct list_head *pos;
2083
2084         list_for_each(pos, list) {
2085                 if (pos == &sbi->ll_pglist)
2086                         return NULL;
2087                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2088                 if (llap->llap_page == NULL)
2089                         continue;
2090                 return llap;
2091         }
2092         LBUG();
2093         return NULL;
2094 }
2095
2096 int ll_obd_statfs(struct inode *inode, void *arg)
2097 {
2098         struct ll_sb_info *sbi = NULL;
2099         struct obd_export *exp;
2100         char *buf = NULL;
2101         struct obd_ioctl_data *data = NULL;
2102         __u32 type;
2103         int len = 0, rc;
2104
2105         if (!inode || !(sbi = ll_i2sbi(inode)))
2106                 GOTO(out_statfs, rc = -EINVAL);
2107
2108         rc = obd_ioctl_getdata(&buf, &len, arg);
2109         if (rc)
2110                 GOTO(out_statfs, rc);
2111
2112         data = (void*)buf;
2113         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2114             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2115                 GOTO(out_statfs, rc = -EINVAL);
2116
2117         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2118         if (type == LL_STATFS_MDC)
2119                 exp = sbi->ll_md_exp;
2120         else if (type == LL_STATFS_LOV)
2121                 exp = sbi->ll_dt_exp;
2122         else
2123                 GOTO(out_statfs, rc = -ENODEV);
2124
2125         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2126         if (rc)
2127                 GOTO(out_statfs, rc);
2128 out_statfs:
2129         if (buf)
2130                 obd_ioctl_freedata(buf, len);
2131         return rc;
2132 }
2133
2134 int ll_process_config(struct lustre_cfg *lcfg)
2135 {
2136         char *ptr;
2137         void *sb;
2138         struct lprocfs_static_vars lvars;
2139         unsigned long x;
2140         int rc = 0;
2141
2142         lprocfs_llite_init_vars(&lvars);
2143
2144         /* The instance name contains the sb: lustre-client-aacfe000 */
2145         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2146         if (!ptr || !*(++ptr))
2147                 return -EINVAL;
2148         if (sscanf(ptr, "%lx", &x) != 1)
2149                 return -EINVAL;
2150         sb = (void *)x;
2151         /* This better be a real Lustre superblock! */
2152         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2153
2154         /* Note we have not called client_common_fill_super yet, so
2155            proc fns must be able to handle that! */
2156         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2157                                       lcfg, sb);
2158         return(rc);
2159 }
2160
2161 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2162 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2163                                        struct inode *i1, struct inode *i2,
2164                                        const char *name, int namelen,
2165                                        int mode, __u32 opc, void *data)
2166 {
2167         LASSERT(i1 != NULL);
2168
2169         if (namelen > ll_i2sbi(i1)->ll_namelen)
2170                 return ERR_PTR(-ENAMETOOLONG);
2171
2172         if (op_data == NULL)
2173                 OBD_ALLOC_PTR(op_data);
2174
2175         if (op_data == NULL)
2176                 return ERR_PTR(-ENOMEM);
2177
2178         ll_i2gids(op_data->op_suppgids, i1, i2);
2179         op_data->op_fid1 = *ll_inode2fid(i1);
2180         op_data->op_capa1 = ll_mdscapa_get(i1);
2181
2182         if (i2) {
2183                 op_data->op_fid2 = *ll_inode2fid(i2);
2184                 op_data->op_capa2 = ll_mdscapa_get(i2);
2185         } else {
2186                 fid_zero(&op_data->op_fid2);
2187         }
2188
2189         op_data->op_name = name;
2190         op_data->op_namelen = namelen;
2191         op_data->op_mode = mode;
2192         op_data->op_mod_time = CURRENT_SECONDS;
2193         op_data->op_fsuid = current->fsuid;
2194         op_data->op_fsgid = current->fsgid;
2195         op_data->op_cap = current->cap_effective;
2196         op_data->op_bias = MDS_CHECK_SPLIT;
2197         op_data->op_opc = opc;
2198         op_data->op_mds = 0;
2199         op_data->op_data = data;
2200
2201         return op_data;
2202 }
2203
2204 void ll_finish_md_op_data(struct md_op_data *op_data)
2205 {
2206         capa_put(op_data->op_capa1);
2207         capa_put(op_data->op_capa2);
2208         OBD_FREE_PTR(op_data);
2209 }