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 "llite_internal.h"
  37
  38 kmem_cache_t *ll_file_data_slab;
  39
  40 LIST_HEAD(ll_super_blocks);
  41 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  42
  43 extern struct address_space_operations ll_aops;
  44 extern struct address_space_operations ll_dir_aops;
  45
  46 #ifndef log2
  47 #define log2(n) ffz(~(n))
  48 #endif
  49
  50
  51 struct ll_sb_info *ll_init_sbi(void)
  52 {
  53         struct ll_sb_info *sbi = NULL;
  54         class_uuid_t uuid;
  55         ENTRY;
  56
  57         OBD_ALLOC(sbi, sizeof(*sbi));
  58         if (!sbi)
  59                 RETURN(NULL);
  60
  61         spin_lock_init(&sbi->ll_lock);
  62         spin_lock_init(&sbi->ll_lco.lco_lock);
  63         INIT_LIST_HEAD(&sbi->ll_pglist);
  64         sbi->ll_pglist_gen = 0;
  65         if (num_physpages >> (20 - PAGE_SHIFT) < 512)
  66                 sbi->ll_async_page_max = num_physpages / 2;
  67         else
  68                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  69         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  70                                            SBI_DEFAULT_READAHEAD_MAX);
  71         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  72                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  73
  74         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  75         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  76
  77         class_generate_random_uuid(uuid);
  78         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  79         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  80
  81         spin_lock(&ll_sb_lock);
  82         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  83         spin_unlock(&ll_sb_lock);
  84
  85         INIT_LIST_HEAD(&sbi->ll_deathrow);
  86         spin_lock_init(&sbi->ll_deathrow_lock);
  87         RETURN(sbi);
  88 }
  89
  90 void ll_free_sbi(struct super_block *sb)
  91 {
  92         struct ll_sb_info *sbi = ll_s2sbi(sb);
  93         ENTRY;
  94
  95         if (sbi != NULL) {
  96                 spin_lock(&ll_sb_lock);
  97                 list_del(&sbi->ll_list);
  98                 spin_unlock(&ll_sb_lock);
  99                 OBD_FREE(sbi, sizeof(*sbi));
 100         }
 101         EXIT;
 102 }
 103
 104 static struct dentry_operations ll_d_root_ops = {
 105         .d_compare = ll_dcompare,
 106 };
 107
 108 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 109  * us to make MDS RPCs with large enough reply buffers to hold the
 110  * maximum-sized (= maximum striped) EA and cookie without having to
 111  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 112 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 113 {
 114         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 115         __u32 valsize = sizeof(struct lov_desc);
 116         int rc, easize, def_easize, cookiesize;
 117         struct lov_desc desc;
 118         __u32 stripes;
 119         ENTRY;
 120
 121         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 122                           &valsize, &desc);
 123         if (rc)
 124                 RETURN(rc);
 125
 126         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 127         lsm.lsm_stripe_count = stripes;
 128         easize = obd_size_diskmd(dt_exp, &lsm);
 129
 130         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 131         def_easize = obd_size_diskmd(dt_exp, &lsm);
 132
 133         cookiesize = stripes * sizeof(struct llog_cookie);
 134
 135         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 136                easize, cookiesize);
 137
 138         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 139         RETURN(rc);
 140 }
 141
 142 int client_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 143 {
 144         struct inode *root = 0;
 145         struct ll_sb_info *sbi = ll_s2sbi(sb);
 146         struct obd_device *obd;
 147         struct lu_fid rootfid;
 148         struct obd_statfs osfs;
 149         struct ptlrpc_request *request = NULL;
 150         struct lustre_handle osc_conn = {0, };
 151         struct lustre_handle md_conn = {0, };
 152         struct obd_connect_data *data = NULL;
 153         struct lustre_md md;
 154         int err;
 155         ENTRY;
 156
 157         obd = class_name2obd(mdc);
 158         if (!obd) {
 159                 CERROR("MDC %s: not setup or attached\n", mdc);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         OBD_ALLOC_PTR(data);
 164         if (data == NULL)
 165                 RETURN(-ENOMEM);
 166
 167         if (proc_lustre_fs_root) {
 168                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 169                                                   osc, mdc);
 170                 if (err < 0)
 171                         CERROR("could not register mount in /proc/lustre");
 172         }
 173
 174         /* indicate that inodebits locking is supported by this client */
 175         data->ocd_connect_flags |= OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH;
 176         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 177
 178         if (sb->s_flags & MS_RDONLY)
 179                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 180         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 181                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 182         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_JOIN;
 183
 184         if (sbi->ll_flags & LL_SBI_FLOCK) {
 185                 sbi->ll_fop = &ll_file_operations_flock;
 186         } else {
 187                 sbi->ll_fop = &ll_file_operations;
 188         }
 189
 190         data->ocd_connect_flags |= OBD_CONNECT_VERSION;
 191         data->ocd_version = LUSTRE_VERSION_CODE;
 192
 193         /* real client */
 194         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 195
 196         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data);
 197         if (err == -EBUSY) {
 198                 CERROR("An MDT (mdc %s) is performing recovery, of which this"
 199                        " client is not a part.  Please wait for recovery to "
 200                        "complete, abort, or time out.\n", mdc);
 201                 GOTO(out, err);
 202         } else if (err) {
 203                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 204                 GOTO(out, err);
 205         }
 206         sbi->ll_md_exp = class_conn2export(&md_conn);
 207
 208         err = obd_statfs(obd, &osfs, jiffies - HZ);
 209         if (err)
 210                 GOTO(out_mdc, err);
 211
 212         LASSERT(osfs.os_bsize);
 213         sb->s_blocksize = osfs.os_bsize;
 214         sb->s_blocksize_bits = log2(osfs.os_bsize);
 215         sb->s_magic = LL_SUPER_MAGIC;
 216         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 217         sbi->ll_namelen = osfs.os_namelen;
 218
 219         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 220             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 221                 LCONSOLE_INFO("Disabling user_xattr feature because "
 222                               "it is not supported on the server\n");
 223                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 224         }
 225
 226         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 227 #ifdef MS_POSIXACL
 228                 sb->s_flags |= MS_POSIXACL;
 229 #endif
 230                 sbi->ll_flags |= LL_SBI_ACL;
 231         } else
 232                 sbi->ll_flags &= ~LL_SBI_ACL;
 233
 234         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 235                 sbi->ll_flags |= LL_SBI_JOIN;
 236
 237 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 238         /* We set sb->s_dev equal on all lustre clients in order to support
 239          * NFS export clustering.  NFSD requires that the FSID be the same
 240          * on all clients. */
 241         /* s_dev is also used in lt_compare() to compare two fs, but that is
 242          * only a node-local comparison. */
 243         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 244                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 245 #endif
 246
 247         obd = class_name2obd(osc);
 248         if (!obd) {
 249                 CERROR("OSC %s: not setup or attached\n", osc);
 250                 GOTO(out_mdc, err = -ENODEV);
 251         }
 252
 253         data->ocd_connect_flags =
 254                 OBD_CONNECT_GRANT | OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL;
 255
 256         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 257                "ocd_grant: %d\n", data->ocd_connect_flags,
 258                data->ocd_version, data->ocd_grant);
 259
 260         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 261         obd->obd_upcall.onu_upcall = ll_ocd_update;
 262
 263         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data);
 264         if (err == -EBUSY) {
 265                 CERROR("An OST (osc %s) is performing recovery, of which this"
 266                        " client is not a part.  Please wait for recovery to "
 267                        "complete, abort, or time out.\n", osc);
 268                 GOTO(out, err);
 269         } else if (err) {
 270                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 271                 GOTO(out_mdc, err);
 272         }
 273
 274         sbi->ll_dt_exp = class_conn2export(&osc_conn);
 275
 276         spin_lock(&sbi->ll_lco.lco_lock);
 277         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 278         spin_unlock(&sbi->ll_lco.lco_lock);
 279
 280         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 281
 282         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 283                                   0, NULL, NULL, NULL);
 284         if (err < 0) {
 285                 LCONSOLE_ERROR("There are no OST's in this filesystem. "
 286                                "There must be at least one active OST for "
 287                                "a client to start.\n");
 288                 GOTO(out_osc, err);
 289         }
 290
 291         if (!ll_async_page_slab) {
 292                 ll_async_page_slab_size =
 293                         size_round(sizeof(struct ll_async_page)) + err;
 294                 ll_async_page_slab = kmem_cache_create("ll_async_page",
 295                                                        ll_async_page_slab_size,
 296                                                        0, 0, NULL, NULL);
 297                 if (!ll_async_page_slab)
 298                         GOTO(out_osc, -ENOMEM);
 299         }
 300
 301         err = md_getstatus(sbi->ll_md_exp, &rootfid);
 302         if (err) {
 303                 CERROR("cannot mds_connect: rc = %d\n", err);
 304                 GOTO(out_osc, err);
 305         }
 306         CDEBUG(D_SUPER, "rootfid "DFID3"\n", PFID3(&rootfid));
 307         sbi->ll_root_fid = rootfid;
 308
 309         sb->s_op = &lustre_super_operations;
 310 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 311         sb->s_export_op = &lustre_export_operations;
 312 #endif
 313
 314         /* make root inode
 315          * XXX: move this to after cbd setup? */
 316         err = md_getattr(sbi->ll_md_exp, &rootfid,
 317                          OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
 318                          (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
 319                          0, &request);
 320         if (err) {
 321                 CERROR("md_getattr failed for root: rc = %d\n", err);
 322                 GOTO(out_osc, err);
 323         }
 324
 325         err = md_get_lustre_md(sbi->ll_md_exp, request, 0, sbi->ll_dt_exp, &md);
 326         if (err) {
 327                 CERROR("failed to understand root inode md: rc = %d\n", err);
 328                 ptlrpc_req_finished (request);
 329                 GOTO(out_osc, err);
 330         }
 331
 332         LASSERT(fid_oid(&sbi->ll_root_fid) != 0);
 333         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &md);
 334         ptlrpc_req_finished(request);
 335
 336         if (root == NULL || is_bad_inode(root)) {
 337                 md_free_lustre_md(sbi->ll_dt_exp, &md);
 338                 CERROR("lustre_lite: bad iget4 for root\n");
 339                 GOTO(out_root, err = -EBADF);
 340         }
 341
 342         err = ll_close_thread_start(&sbi->ll_lcq);
 343         if (err) {
 344                 CERROR("cannot start close thread: rc %d\n", err);
 345                 GOTO(out_root, err);
 346         }
 347
 348         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 349            backing dev info assigned to inode mapping is used for
 350            determining maximal readahead. */
 351 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 352     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 353         /* bug 2805 - set VM readahead to zero */
 354         vm_max_readahead = vm_min_readahead = 0;
 355 #endif
 356
 357         sb->s_root = d_alloc_root(root);
 358         if (data != NULL)
 359                 OBD_FREE(data, sizeof(*data));
 360         sb->s_root->d_op = &ll_d_root_ops;
 361         RETURN(err);
 362
 363 out_root:
 364         if (root)
 365                 iput(root);
 366 out_osc:
 367         obd_disconnect(sbi->ll_dt_exp);
 368 out_mdc:
 369         obd_disconnect(sbi->ll_md_exp);
 370 out:
 371         if (data != NULL)
 372                 OBD_FREE_PTR(data);
 373         lprocfs_unregister_mountpoint(sbi);
 374         RETURN(err);
 375 }
 376
 377 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 378 {
 379         int size, rc;
 380
 381         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 382         size = sizeof(int);
 383         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 384                           &size, lmmsize);
 385         if (rc)
 386                 CERROR("Get max mdsize error rc %d \n", rc);
 387
 388         RETURN(rc);
 389 }
 390
 391 void ll_dump_inode(struct inode *inode)
 392 {
 393         struct list_head *tmp;
 394         int dentry_count = 0;
 395
 396         LASSERT(inode != NULL);
 397
 398         list_for_each(tmp, &inode->i_dentry)
 399                 dentry_count++;
 400
 401         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 402                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 403                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 404 }
 405
 406 void lustre_dump_dentry(struct dentry *dentry, int recur)
 407 {
 408         struct list_head *tmp;
 409         int subdirs = 0;
 410
 411         LASSERT(dentry != NULL);
 412
 413         list_for_each(tmp, &dentry->d_subdirs)
 414                 subdirs++;
 415
 416         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 417                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 418                dentry->d_name.len, dentry->d_name.name,
 419                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 420                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 421                dentry->d_flags, dentry->d_fsdata, subdirs);
 422         if (dentry->d_inode != NULL)
 423                 ll_dump_inode(dentry->d_inode);
 424
 425         if (recur == 0)
 426                 return;
 427
 428         list_for_each(tmp, &dentry->d_subdirs) {
 429                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 430                 lustre_dump_dentry(d, recur - 1);
 431         }
 432 }
 433
 434 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 435 void lustre_throw_orphan_dentries(struct super_block *sb)
 436 {
 437         struct hlist_node *tmp, *next;
 438         struct ll_sb_info *sbi = ll_s2sbi(sb);
 439
 440         /* Do this to get rid of orphaned dentries. That is not really trw. */
 441         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 442                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 443                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 444                       "before and after shrink_dcache_parent\n",
 445                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 446                 lustre_dump_dentry(dentry, 1);
 447                 shrink_dcache_parent(dentry);
 448                 lustre_dump_dentry(dentry, 1);
 449         }
 450 }
 451 #else
 452 #define lustre_throw_orphan_dentries(sb)
 453 #endif
 454
 455 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 456 {
 457         LIST_HEAD(throw_away);
 458         int locked = 0;
 459         ENTRY;
 460
 461         if (try) {
 462                 locked = spin_trylock(&sbi->ll_deathrow_lock);
 463         } else {
 464                 spin_lock(&sbi->ll_deathrow_lock);
 465                 locked = 1;
 466         }
 467
 468         if (!locked) {
 469                 EXIT;
 470                 return;
 471         }
 472
 473         list_splice_init(&sbi->ll_deathrow, &throw_away);
 474         spin_unlock(&sbi->ll_deathrow_lock);
 475
 476         while (!list_empty(&throw_away)) {
 477                 struct ll_inode_info *lli;
 478                 struct inode *inode;
 479
 480                 lli = list_entry(throw_away.next, struct ll_inode_info,
 481                                  lli_dead_list);
 482                 list_del_init(&lli->lli_dead_list);
 483
 484                 inode = ll_info2i(lli);
 485                 d_prune_aliases(inode);
 486
 487                 CDEBUG(D_INODE, "prune duplicate inode %p inum %lu count %u\n",
 488                        inode, inode->i_ino, atomic_read(&inode->i_count));
 489                 iput(inode);
 490         }
 491         EXIT;
 492 }
 493
 494 void client_common_put_super(struct super_block *sb)
 495 {
 496         struct ll_sb_info *sbi = ll_s2sbi(sb);
 497         ENTRY;
 498
 499         ll_close_thread_shutdown(sbi->ll_lcq);
 500
 501         /* destroy inodes in deathrow */
 502         prune_deathrow(sbi, 0);
 503
 504         list_del(&sbi->ll_conn_chain);
 505         obd_disconnect(sbi->ll_dt_exp);
 506
 507         lprocfs_unregister_mountpoint(sbi);
 508         if (sbi->ll_proc_root) {
 509                 lprocfs_remove(sbi->ll_proc_root);
 510                 sbi->ll_proc_root = NULL;
 511         }
 512
 513         obd_disconnect(sbi->ll_md_exp);
 514
 515         lustre_throw_orphan_dentries(sb);
 516         EXIT;
 517 }
 518
 519 char *ll_read_opt(const char *opt, char *data)
 520 {
 521         char *value;
 522         char *retval;
 523         ENTRY;
 524
 525         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 526         if (strncmp(opt, data, strlen(opt)))
 527                 RETURN(NULL);
 528         if ((value = strchr(data, '=')) == NULL)
 529                 RETURN(NULL);
 530
 531         value++;
 532         OBD_ALLOC(retval, strlen(value) + 1);
 533         if (!retval) {
 534                 CERROR("out of memory!\n");
 535                 RETURN(NULL);
 536         }
 537
 538         memcpy(retval, value, strlen(value)+1);
 539         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 540         RETURN(retval);
 541 }
 542
 543 static inline int ll_set_opt(const char *opt, char *data, int fl)
 544 {
 545         if (strncmp(opt, data, strlen(opt)) != 0)
 546                 return(0);
 547         else
 548                 return(fl);
 549 }
 550
 551 /* non-client-specific mount options are parsed in lmd_parse */
 552 void ll_options(char *options, int *flags)
 553 {
 554         int tmp;
 555         char *s1 = options, *s2;
 556         ENTRY;
 557
 558         if (!options) {
 559                 EXIT;
 560                 return;
 561         }
 562
 563         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 564
 565         while (*s1) {
 566                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 567                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 568                 if (tmp) {
 569                         *flags |= tmp;
 570                         goto next;
 571                 }
 572                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 573                 if (tmp) {
 574                         *flags |= tmp;
 575                         goto next;
 576                 }
 577                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK);
 578                 if (tmp) {
 579                         *flags &= ~tmp;
 580                         goto next;
 581                 }
 582                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 583                 if (tmp) {
 584                         *flags |= tmp;
 585                         goto next;
 586                 }
 587                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 588                 if (tmp) {
 589                         *flags &= ~tmp;
 590                         goto next;
 591                 }
 592                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 593                 if (tmp) {
 594                         /* Ignore deprecated mount option.  The client will
 595                          * always try to mount with ACL support, whether this
 596                          * is used depends on whether server supports it. */
 597                         goto next;
 598                 }
 599                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 600                 if (tmp) {
 601                         goto next;
 602                 }
 603
 604 next:
 605                 /* Find next opt */
 606                 s2 = strchr(s1, ',');
 607                 if (s2 == NULL)
 608                         break;
 609                 s1 = s2 + 1;
 610         }
 611         EXIT;
 612 }
 613
 614 void ll_lli_init(struct ll_inode_info *lli)
 615 {
 616         sema_init(&lli->lli_open_sem, 1);
 617         sema_init(&lli->lli_size_sem, 1);
 618         lli->lli_flags = 0;
 619         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 620         spin_lock_init(&lli->lli_lock);
 621         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 622         lli->lli_inode_magic = LLI_INODE_MAGIC;
 623         INIT_LIST_HEAD(&lli->lli_dead_list);
 624 }
 625
 626 int ll_fill_super(struct super_block *sb)
 627 {
 628         struct lustre_profile *lprof;
 629         struct lustre_sb_info *lsi = s2lsi(sb);
 630         struct ll_sb_info *sbi;
 631         char  *osc = NULL;
 632         char  *mdc = NULL;
 633         char  *profilenm = get_profile_name(sb);
 634         struct config_llog_instance cfg;
 635         char   ll_instance[sizeof(sb) * 2 + 1];
 636         int    err;
 637         ENTRY;
 638
 639         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 640
 641         /* client additional sb info */
 642         lsi->lsi_llsbi = sbi = ll_init_sbi();
 643         if (!sbi)
 644                 RETURN(-ENOMEM);
 645
 646         ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 647
 648         /* Generate a string unique to this super, in case some joker tries
 649            to mount the same fs at two mount points.
 650            Use the address of the super itself.*/
 651         sprintf(ll_instance, "%p", sb);
 652         cfg.cfg_instance = ll_instance;
 653         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 654         cfg.cfg_last_idx = 0;
 655
 656         /* set up client obds */
 657         err = lustre_process_log(sb, profilenm, &cfg);
 658         if (err < 0) {
 659                 CERROR("Unable to process log: %d\n", err);
 660                 GOTO(out_free, err);
 661         }
 662
 663         lprof = class_get_profile(profilenm);
 664         if (lprof == NULL) {
 665                 CERROR("No profile found: %s\n", profilenm);
 666                 GOTO(out_free, err = -EINVAL);
 667         }
 668         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 669                lprof->lp_mdc, lprof->lp_osc);
 670
 671         OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 672                   strlen(ll_instance) + 2);
 673         if (!osc)
 674                 GOTO(out_free, err = -ENOMEM);
 675         sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
 676
 677         OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 678                   strlen(ll_instance) + 2);
 679         if (!mdc)
 680                 GOTO(out_free, err = -ENOMEM);
 681         sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
 682
 683         /* connections, registrations, sb setup */
 684         err = client_common_fill_super(sb, mdc, osc);
 685
 686 out_free:
 687         if (mdc)
 688                 OBD_FREE(mdc, strlen(mdc) + 1);
 689         if (osc)
 690                 OBD_FREE(osc, strlen(osc) + 1);
 691         if (err) {
 692                 struct obd_device *obd;
 693                 int next = 0;
 694                 /* like ll_put_super below */
 695                 lustre_end_log(sb, NULL, &cfg);
 696                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 697                        != NULL) {
 698                         class_manual_cleanup(obd);
 699                 }
 700                 class_del_profile(profilenm);
 701                 ll_free_sbi(sb);
 702                 lsi->lsi_llsbi = NULL;
 703                 lustre_common_put_super(sb);
 704         }
 705         RETURN(err);
 706 } /* ll_fill_super */
 707
 708
 709 void ll_put_super(struct super_block *sb)
 710 {
 711         struct config_llog_instance cfg;
 712         char   ll_instance[sizeof(sb) * 2 + 1];
 713         struct obd_device *obd;
 714         struct lustre_sb_info *lsi = s2lsi(sb);
 715         struct ll_sb_info *sbi = ll_s2sbi(sb);
 716         char *profilenm = get_profile_name(sb);
 717         int next;
 718         ENTRY;
 719
 720         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 721
 722         sprintf(ll_instance, "%p", sb);
 723         cfg.cfg_instance = ll_instance;
 724         lustre_end_log(sb, NULL, &cfg);
 725
 726         obd = class_exp2obd(sbi->ll_md_exp);
 727         if (obd) {
 728                 int force = obd->obd_no_recov;
 729                 /* We need to set force before the lov_disconnect in
 730                 lustre_common_put_super, since l_d cleans up osc's as well. */
 731                 next = 0;
 732                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 733                        != NULL) {
 734                         obd->obd_force = force;
 735                 }
 736         }
 737
 738         client_common_put_super(sb);
 739
 740         next = 0;
 741         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 742                 class_manual_cleanup(obd);
 743         }
 744
 745         if (profilenm)
 746                 class_del_profile(profilenm);
 747
 748         ll_free_sbi(sb);
 749         lsi->lsi_llsbi = NULL;
 750
 751         lustre_common_put_super(sb);
 752
 753         LCONSOLE_WARN("client umount complete\n");
 754         EXIT;
 755 } /* client_put_super */
 756
 757 #ifdef HAVE_REGISTER_CACHE
 758 #include <linux/cache_def.h>
 759 #ifdef HAVE_CACHE_RETURN_INT
 760 static int
 761 #else
 762 static void
 763 #endif
 764 ll_shrink_cache(int priority, unsigned int gfp_mask)
 765 {
 766         struct ll_sb_info *sbi;
 767         int count = 0;
 768
 769         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
 770                 count += llap_shrink_cache(sbi, priority);
 771
 772 #ifdef HAVE_CACHE_RETURN_INT
 773         return count;
 774 #endif
 775 }
 776
 777 struct cache_definition ll_cache_definition = {
 778         .name = "llap_cache",
 779         .shrink = ll_shrink_cache
 780 };
 781 #endif /* HAVE_REGISTER_CACHE */
 782
 783 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 784 {
 785         struct inode *inode = NULL;
 786         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 787         if (lock->l_ast_data) {
 788                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 789                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 790                         inode = igrab(lock->l_ast_data);
 791                 } else {
 792                         inode = lock->l_ast_data;
 793                         if (inode->i_state & I_FREEING)
 794                                 __LDLM_DEBUG(D_INFO, lock,
 795                                      "l_ast_data %p is bogus: magic %08x",
 796                                      lock->l_ast_data, lli->lli_inode_magic);
 797                         else
 798                                 __LDLM_DEBUG(D_WARNING, lock,
 799                                      "l_ast_data %p is bogus: magic %08x",
 800                                      lock->l_ast_data, lli->lli_inode_magic);
 801
 802                         inode = NULL;
 803                 }
 804         }
 805         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 806         return inode;
 807 }
 808
 809 static int null_if_equal(struct ldlm_lock *lock, void *data)
 810 {
 811         if (data == lock->l_ast_data) {
 812                 lock->l_ast_data = NULL;
 813
 814                 if (lock->l_req_mode != lock->l_granted_mode)
 815                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
 816         }
 817
 818         return LDLM_ITER_CONTINUE;
 819 }
 820
 821 void ll_clear_inode(struct inode *inode)
 822 {
 823         struct ll_inode_info *lli = ll_i2info(inode);
 824         struct ll_sb_info *sbi = ll_i2sbi(inode);
 825         ENTRY;
 826
 827         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 828                inode->i_generation, inode);
 829
 830         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 831         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
 832                          null_if_equal, inode);
 833
 834         if (lli->lli_smd) {
 835                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 836                                   null_if_equal, inode);
 837
 838                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 839                 lli->lli_smd = NULL;
 840         }
 841
 842         if (lli->lli_symlink_name) {
 843                 OBD_FREE(lli->lli_symlink_name,
 844                          strlen(lli->lli_symlink_name) + 1);
 845                 lli->lli_symlink_name = NULL;
 846         }
 847
 848 #ifdef CONFIG_FS_POSIX_ACL
 849         if (lli->lli_posix_acl) {
 850                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
 851                 posix_acl_release(lli->lli_posix_acl);
 852                 lli->lli_posix_acl = NULL;
 853         }
 854 #endif
 855
 856         lli->lli_inode_magic = LLI_INODE_DEAD;
 857
 858         spin_lock(&sbi->ll_deathrow_lock);
 859         list_del_init(&lli->lli_dead_list);
 860         spin_unlock(&sbi->ll_deathrow_lock);
 861
 862         EXIT;
 863 }
 864
 865 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 866  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 867  * keep these values until such a time that objects are allocated for it.
 868  * We do the MDS operations first, as it is checking permissions for us.
 869  * We don't to the MDS RPC if there is nothing that we want to store there,
 870  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 871  * going to do an RPC anyways.
 872  *
 873  * If we are doing a truncate, we will send the mtime and ctime updates
 874  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 875  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 876  * at the same time.
 877  */
 878 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 879 {
 880         struct ll_inode_info *lli = ll_i2info(inode);
 881         struct lov_stripe_md *lsm = lli->lli_smd;
 882         struct ll_sb_info *sbi = ll_i2sbi(inode);
 883         struct ptlrpc_request *request = NULL;
 884         struct md_op_data op_data = { { 0 } };
 885         int ia_valid = attr->ia_valid;
 886         int rc = 0;
 887         ENTRY;
 888
 889         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
 890                attr->ia_valid);
 891         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
 892
 893         if (ia_valid & ATTR_SIZE) {
 894                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 895                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 896                                attr->ia_size, ll_file_maxbytes(inode));
 897                         RETURN(-EFBIG);
 898                 }
 899
 900                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 901         }
 902
 903         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
 904         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
 905                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 906                         RETURN(-EPERM);
 907         }
 908
 909         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 910         if (attr->ia_valid & ATTR_CTIME) {
 911                 attr->ia_ctime = CURRENT_TIME;
 912                 attr->ia_valid |= ATTR_CTIME_SET;
 913         }
 914         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 915                 attr->ia_atime = CURRENT_TIME;
 916                 attr->ia_valid |= ATTR_ATIME_SET;
 917         }
 918         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 919                 attr->ia_mtime = CURRENT_TIME;
 920                 attr->ia_valid |= ATTR_MTIME_SET;
 921         }
 922
 923         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 924                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
 925                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 926                        CURRENT_SECONDS);
 927
 928         /* NB: ATTR_SIZE will only be set after this point if the size
 929          * resides on the MDS, ie, this file has no objects. */
 930         if (lsm)
 931                 attr->ia_valid &= ~ATTR_SIZE;
 932
 933         /* If only OST attributes being set on objects, don't do MDS RPC.
 934          * In that case, we need to check permissions and update the local
 935          * inode ourselves so we can call obdo_from_inode() always. */
 936         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 937                 struct lustre_md md;
 938                 ll_prepare_md_op_data(&op_data, inode, NULL, NULL, 0, 0);
 939
 940                 rc = md_setattr(sbi->ll_md_exp, &op_data,
 941                                 attr, NULL, 0, NULL, 0, &request);
 942
 943                 if (rc) {
 944                         ptlrpc_req_finished(request);
 945                         if (rc == -ENOENT) {
 946                                 inode->i_nlink = 0;
 947                                 /* Unlinked special device node? Or just a race?
 948                                  * Pretend we done everything. */
 949                                 if (!S_ISREG(inode->i_mode) &&
 950                                     !S_ISDIR(inode->i_mode) &&
 951                                     !S_ISDIR(inode->i_mode))
 952                                         rc = inode_setattr(inode, attr);
 953                         } else if (rc != -EPERM && rc != -EACCES) {
 954                                 CERROR("mdcsetattr fails: rc = %d\n", rc);
 955                         }
 956                         RETURN(rc);
 957                 }
 958
 959                 rc = md_get_lustre_md(sbi->ll_md_exp, request, 0, sbi->ll_dt_exp, &md);
 960                 if (rc) {
 961                         ptlrpc_req_finished(request);
 962                         RETURN(rc);
 963                 }
 964
 965                 /* We call inode_setattr to adjust timestamps.
 966                  * If there is at least some data in file, we cleared ATTR_SIZE
 967                  * above to avoid invoking vmtruncate, otherwise it is important
 968                  * to call vmtruncate in inode_setattr to update inode->i_size
 969                  * (bug 6196) */
 970                 rc = inode_setattr(inode, attr);
 971
 972                 ll_update_inode(inode, &md);
 973                 ptlrpc_req_finished(request);
 974
 975                 if (!lsm || !S_ISREG(inode->i_mode)) {
 976                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 977                         RETURN(rc);
 978                 }
 979         } else {
 980                 /* The OST doesn't check permissions, but the alternative is
 981                  * a gratuitous RPC to the MDS.  We already rely on the client
 982                  * to do read/write/truncate permission checks, so is mtime OK?
 983                  */
 984                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 985                         /* from sys_utime() */
 986                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 987                                 if (current->fsuid != inode->i_uid &&
 988                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
 989                                         RETURN(rc);
 990                         } else {
 991                                 /* from inode_change_ok() */
 992                                 if (current->fsuid != inode->i_uid &&
 993                                     !capable(CAP_FOWNER))
 994                                         RETURN(-EPERM);
 995                         }
 996                 }
 997
 998                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
 999                 rc = inode_setattr(inode, attr);
1000         }
1001
1002         /* We really need to get our PW lock before we change inode->i_size.
1003          * If we don't we can race with other i_size updaters on our node, like
1004          * ll_file_read.  We can also race with i_size propogation to other
1005          * nodes through dirtying and writeback of final cached pages.  This
1006          * last one is especially bad for racing o_append users on other
1007          * nodes. */
1008         if (ia_valid & ATTR_SIZE) {
1009                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1010                                                            OBD_OBJECT_EOF } };
1011                 struct lustre_handle lockh = { 0 };
1012                 int err, ast_flags = 0;
1013                 /* XXX when we fix the AST intents to pass the discard-range
1014                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1015                  * XXX here. */
1016                 if (attr->ia_size == 0)
1017                         ast_flags = LDLM_AST_DISCARD_DATA;
1018
1019                 UNLOCK_INODE_MUTEX(inode);
1020                 UP_WRITE_I_ALLOC_SEM(inode);
1021                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1022                                     ast_flags);
1023 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1024                 DOWN_WRITE_I_ALLOC_SEM(inode);
1025                 LOCK_INODE_MUTEX(inode);
1026 #else
1027                 LOCK_INODE_MUTEX(inode);
1028                 DOWN_WRITE_I_ALLOC_SEM(inode);
1029 #endif
1030                 if (rc != 0)
1031                         RETURN(rc);
1032
1033                 /* Only ll_inode_size_lock is taken at this level.
1034                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1035                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1036                  * ll_truncate dropped ll_inode_size_lock() */
1037                 ll_inode_size_lock(inode, 0);
1038                 rc = vmtruncate(inode, attr->ia_size);
1039                 if (rc != 0) {
1040                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1041                         ll_inode_size_unlock(inode, 0);
1042                 }
1043
1044                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1045                 if (err) {
1046                         CERROR("ll_extent_unlock failed: %d\n", err);
1047                         if (!rc)
1048                                 rc = err;
1049                 }
1050         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1051                 obd_flag flags;
1052                 struct obdo oa;
1053
1054                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1055                        inode->i_ino, LTIME_S(attr->ia_mtime));
1056
1057                 oa.o_id = lsm->lsm_object_id;
1058                 oa.o_valid = OBD_MD_FLID;
1059
1060                 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1061                         OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1062                         OBD_MD_FLFID | OBD_MD_FLGENER;
1063
1064                 obdo_from_inode(&oa, inode, flags);
1065                 rc = obd_setattr(sbi->ll_dt_exp, &oa, lsm, NULL);
1066                 if (rc)
1067                         CERROR("obd_setattr fails: rc=%d\n", rc);
1068         }
1069         RETURN(rc);
1070 }
1071
1072 int ll_setattr(struct dentry *de, struct iattr *attr)
1073 {
1074         return ll_setattr_raw(de->d_inode, attr);
1075 }
1076
1077 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1078                        unsigned long max_age)
1079 {
1080         struct ll_sb_info *sbi = ll_s2sbi(sb);
1081         struct obd_statfs obd_osfs;
1082         int rc;
1083         ENTRY;
1084
1085         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1086         if (rc) {
1087                 CERROR("md_statfs fails: rc = %d\n", rc);
1088                 RETURN(rc);
1089         }
1090
1091         osfs->os_type = sb->s_magic;
1092
1093         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1094                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1095
1096         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1097         if (rc) {
1098                 CERROR("obd_statfs fails: rc = %d\n", rc);
1099                 RETURN(rc);
1100         }
1101
1102         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1103                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1104                obd_osfs.os_files);
1105
1106         osfs->os_blocks = obd_osfs.os_blocks;
1107         osfs->os_bfree = obd_osfs.os_bfree;
1108         osfs->os_bavail = obd_osfs.os_bavail;
1109
1110         /* If we don't have as many objects free on the OST as inodes
1111          * on the MDS, we reduce the total number of inodes to
1112          * compensate, so that the "inodes in use" number is correct.
1113          */
1114         if (obd_osfs.os_ffree < osfs->os_ffree) {
1115                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1116                         obd_osfs.os_ffree;
1117                 osfs->os_ffree = obd_osfs.os_ffree;
1118         }
1119
1120         RETURN(rc);
1121 }
1122
1123 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1124 {
1125         struct obd_statfs osfs;
1126         int rc;
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1129         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1130
1131         /* For now we will always get up-to-date statfs values, but in the
1132          * future we may allow some amount of caching on the client (e.g.
1133          * from QOS or lprocfs updates). */
1134         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1135         if (rc)
1136                 return rc;
1137
1138         statfs_unpack(sfs, &osfs);
1139
1140         if (sizeof(sfs->f_blocks) == 4) {
1141                 while (osfs.os_blocks > ~0UL) {
1142                         sfs->f_bsize <<= 1;
1143
1144                         osfs.os_blocks >>= 1;
1145                         osfs.os_bfree >>= 1;
1146                         osfs.os_bavail >>= 1;
1147                 }
1148         }
1149
1150         sfs->f_blocks = osfs.os_blocks;
1151         sfs->f_bfree = osfs.os_bfree;
1152         sfs->f_bavail = osfs.os_bavail;
1153
1154         return 0;
1155 }
1156
1157 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1158 {
1159         struct ll_inode_info *lli;
1160         struct lov_stripe_md *lsm;
1161
1162         lli = ll_i2info(inode);
1163         LASSERT(lli->lli_size_sem_owner != current);
1164         down(&lli->lli_size_sem);
1165         LASSERT(lli->lli_size_sem_owner == NULL);
1166         lli->lli_size_sem_owner = current;
1167         lsm = lli->lli_smd;
1168         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1169                  lsm, lock_lsm);
1170         if (lock_lsm)
1171                 lov_stripe_lock(lsm);
1172 }
1173
1174 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1175 {
1176         struct ll_inode_info *lli;
1177         struct lov_stripe_md *lsm;
1178
1179         lli = ll_i2info(inode);
1180         lsm = lli->lli_smd;
1181         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1182                  lsm, unlock_lsm);
1183         if (unlock_lsm)
1184                 lov_stripe_unlock(lsm);
1185         LASSERT(lli->lli_size_sem_owner == current);
1186         lli->lli_size_sem_owner = NULL;
1187         up(&lli->lli_size_sem);
1188 }
1189
1190 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1191 {
1192         struct ll_inode_info *lli = ll_i2info(inode);
1193
1194         dump_lsm(D_INODE, lsm);
1195         dump_lsm(D_INODE, lli->lli_smd);
1196         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1197                  "lsm must be joined lsm %p\n", lsm);
1198         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1199         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1200                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1201         lli->lli_smd = lsm;
1202         lli->lli_maxbytes = lsm->lsm_maxbytes;
1203         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1204                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1205 }
1206
1207 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1208 {
1209         struct ll_inode_info *lli = ll_i2info(inode);
1210         struct mdt_body *body = md->body;
1211         struct lov_stripe_md *lsm = md->lsm;
1212
1213         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1214         if (lsm != NULL) {
1215                 if (lli->lli_smd == NULL) {
1216                         if (lsm->lsm_magic != LOV_MAGIC &&
1217                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1218                                 dump_lsm(D_ERROR, lsm);
1219                                 LBUG();
1220                         }
1221                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1222                                lsm, inode->i_ino, inode->i_generation, inode);
1223                         /* ll_inode_size_lock() requires it is only called
1224                          * with lli_smd != NULL or lock_lsm == 0 or we can
1225                          * race between lock/unlock.  bug 9547 */
1226                         lli->lli_smd = lsm;
1227                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1228                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1229                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1230                 } else {
1231                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1232                              lli->lli_smd->lsm_stripe_count ==
1233                                         lsm->lsm_stripe_count) {
1234                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1235                                         CERROR("lsm mismatch for inode %ld\n",
1236                                                 inode->i_ino);
1237                                         CERROR("lli_smd:\n");
1238                                         dump_lsm(D_ERROR, lli->lli_smd);
1239                                         CERROR("lsm:\n");
1240                                         dump_lsm(D_ERROR, lsm);
1241                                         LBUG();
1242                                 }
1243                         } else
1244                                 ll_replace_lsm(inode, lsm);
1245                 }
1246                 /* bug 2844 - limit i_blksize for broken user-space apps */
1247                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1248                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1249                 if (lli->lli_smd != lsm)
1250                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1251         } else {
1252                 inode->i_blksize = max(inode->i_blksize,
1253                                        inode->i_sb->s_blocksize);
1254         }
1255
1256 #ifdef CONFIG_FS_POSIX_ACL
1257         LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1258         if (body->valid & OBD_MD_FLACL) {
1259                 spin_lock(&lli->lli_lock);
1260                 if (lli->lli_posix_acl)
1261                         posix_acl_release(lli->lli_posix_acl);
1262                 lli->lli_posix_acl = md->posix_acl;
1263                 spin_unlock(&lli->lli_lock);
1264         }
1265 #endif
1266
1267         if (body->valid & OBD_MD_FLATIME &&
1268             body->atime > LTIME_S(inode->i_atime))
1269                 LTIME_S(inode->i_atime) = body->atime;
1270         if (body->valid & OBD_MD_FLMTIME &&
1271             body->mtime > LTIME_S(inode->i_mtime)) {
1272                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
1273                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1274                 LTIME_S(inode->i_mtime) = body->mtime;
1275         }
1276         if (body->valid & OBD_MD_FLCTIME &&
1277             body->ctime > LTIME_S(inode->i_ctime))
1278                 LTIME_S(inode->i_ctime) = body->ctime;
1279         if (body->valid & OBD_MD_FLMODE)
1280                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1281         if (body->valid & OBD_MD_FLTYPE)
1282                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1283         if (body->valid & OBD_MD_FLUID)
1284                 inode->i_uid = body->uid;
1285         if (body->valid & OBD_MD_FLGID)
1286                 inode->i_gid = body->gid;
1287         if (body->valid & OBD_MD_FLFLAGS)
1288                 inode->i_flags = body->flags;
1289         if (body->valid & OBD_MD_FLNLINK)
1290                 inode->i_nlink = body->nlink;
1291         if (body->valid & OBD_MD_FLRDEV)
1292 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1293                 inode->i_rdev = body->rdev;
1294 #else
1295                 inode->i_rdev = old_decode_dev(body->rdev);
1296 #endif
1297         if (body->valid & OBD_MD_FLSIZE)
1298                 inode->i_size = body->size;
1299         if (body->valid & OBD_MD_FLBLOCKS)
1300                 inode->i_blocks = body->blocks;
1301
1302         if (body->valid & OBD_MD_FLSIZE)
1303                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1304
1305         if (body->valid & OBD_MD_FLID)
1306                 lli->lli_fid = body->fid1;
1307
1308         LASSERT(fid_seq(&lli->lli_fid) != 0);
1309 }
1310
1311 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1312 static struct backing_dev_info ll_backing_dev_info = {
1313         .ra_pages       = 0,    /* No readahead */
1314 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1315         .capabilities   = 0,    /* Does contribute to dirty memory */
1316 #else
1317         .memory_backed  = 0,    /* Does contribute to dirty memory */
1318 #endif
1319 };
1320 #endif
1321
1322 void ll_read_inode2(struct inode *inode, void *opaque)
1323 {
1324         struct lustre_md *md = opaque;
1325         struct ll_inode_info *lli = ll_i2info(inode);
1326         ENTRY;
1327
1328         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1329                inode->i_ino, inode->i_generation, inode);
1330
1331         ll_lli_init(lli);
1332
1333         LASSERT(!lli->lli_smd);
1334
1335         /* Core attributes from the MDS first.  This is a new inode, and
1336          * the VFS doesn't zero times in the core inode so we have to do
1337          * it ourselves.  They will be overwritten by either MDS or OST
1338          * attributes - we just need to make sure they aren't newer. */
1339         LTIME_S(inode->i_mtime) = 0;
1340         LTIME_S(inode->i_atime) = 0;
1341         LTIME_S(inode->i_ctime) = 0;
1342         inode->i_rdev = 0;
1343         ll_update_inode(inode, md);
1344
1345         /* OIDEBUG(inode); */
1346
1347         if (S_ISREG(inode->i_mode)) {
1348                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1349                 inode->i_op = &ll_file_inode_operations;
1350                 inode->i_fop = sbi->ll_fop;
1351                 inode->i_mapping->a_ops = &ll_aops;
1352                 EXIT;
1353         } else if (S_ISDIR(inode->i_mode)) {
1354                 inode->i_op = &ll_dir_inode_operations;
1355                 inode->i_fop = &ll_dir_operations;
1356                 inode->i_mapping->a_ops = &ll_dir_aops;
1357                 EXIT;
1358         } else if (S_ISLNK(inode->i_mode)) {
1359                 inode->i_op = &ll_fast_symlink_inode_operations;
1360                 EXIT;
1361         } else {
1362                 inode->i_op = &ll_special_inode_operations;
1363
1364 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1365                 init_special_inode(inode, inode->i_mode,
1366                                    kdev_t_to_nr(inode->i_rdev));
1367
1368                 /* initializing backing dev info. */
1369                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1370 #else
1371                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1372 #endif
1373                 EXIT;
1374         }
1375 }
1376
1377 void ll_delete_inode(struct inode *inode)
1378 {
1379         struct ll_sb_info *sbi = ll_i2sbi(inode);
1380         int rc;
1381         ENTRY;
1382
1383         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1384         if (rc) {
1385                 CERROR("fid_delete() failed, rc %d\n", rc);
1386         }
1387
1388         EXIT;
1389 }
1390
1391 int ll_iocontrol(struct inode *inode, struct file *file,
1392                  unsigned int cmd, unsigned long arg)
1393 {
1394         struct ll_sb_info *sbi = ll_i2sbi(inode);
1395         struct ptlrpc_request *req = NULL;
1396         int rc, flags = 0;
1397         ENTRY;
1398
1399         switch(cmd) {
1400         case EXT3_IOC_GETFLAGS: {
1401                 struct mdt_body *body;
1402
1403                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
1404                                 OBD_MD_FLFLAGS, 0, &req);
1405                 if (rc) {
1406                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1407                         RETURN(-abs(rc));
1408                 }
1409
1410                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1411
1412                 if (body->flags & S_APPEND)
1413                         flags |= EXT3_APPEND_FL;
1414                 if (body->flags & S_IMMUTABLE)
1415                         flags |= EXT3_IMMUTABLE_FL;
1416                 if (body->flags & S_NOATIME)
1417                         flags |= EXT3_NOATIME_FL;
1418
1419                 ptlrpc_req_finished (req);
1420
1421                 RETURN(put_user(flags, (int *)arg));
1422         }
1423         case EXT3_IOC_SETFLAGS: {
1424                 struct md_op_data op_data = { { 0 } };
1425                 struct ll_iattr_struct attr;
1426                 struct obdo *oa;
1427                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1428
1429                 if (get_user(flags, (int *)arg))
1430                         RETURN(-EFAULT);
1431
1432                 oa = obdo_alloc();
1433                 if (!oa)
1434                         RETURN(-ENOMEM);
1435
1436                 ll_prepare_md_op_data(&op_data, inode, NULL, NULL, 0, 0);
1437
1438                 memset(&attr, 0x0, sizeof(attr));
1439                 attr.ia_attr_flags = flags;
1440                 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
1441
1442                 rc = md_setattr(sbi->ll_md_exp, &op_data,
1443                                 (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
1444                 if (rc || lsm == NULL) {
1445                         ptlrpc_req_finished(req);
1446                         obdo_free(oa);
1447                         RETURN(rc);
1448                 }
1449                 ptlrpc_req_finished(req);
1450
1451                 oa->o_id = lsm->lsm_object_id;
1452                 oa->o_flags = flags;
1453                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1454
1455                 obdo_from_inode(oa, inode, OBD_MD_FLFID | OBD_MD_FLGENER);
1456                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1457                 obdo_free(oa);
1458                 if (rc) {
1459                         if (rc != -EPERM && rc != -EACCES)
1460                                 CERROR("md_setattr fails: rc = %d\n", rc);
1461                         RETURN(rc);
1462                 }
1463
1464                 if (flags & EXT3_APPEND_FL)
1465                         inode->i_flags |= S_APPEND;
1466                 else
1467                         inode->i_flags &= ~S_APPEND;
1468                 if (flags & EXT3_IMMUTABLE_FL)
1469                         inode->i_flags |= S_IMMUTABLE;
1470                 else
1471                         inode->i_flags &= ~S_IMMUTABLE;
1472                 if (flags & EXT3_NOATIME_FL)
1473                         inode->i_flags |= S_NOATIME;
1474                 else
1475                         inode->i_flags &= ~S_NOATIME;
1476
1477                 RETURN(0);
1478         }
1479         default:
1480                 RETURN(-ENOSYS);
1481         }
1482
1483         RETURN(0);
1484 }
1485
1486 /* umount -f client means force down, don't save state */
1487 void ll_umount_begin(struct super_block *sb)
1488 {
1489         struct lustre_sb_info *lsi = s2lsi(sb);
1490         struct ll_sb_info *sbi = ll_s2sbi(sb);
1491         struct obd_device *obd;
1492         struct obd_ioctl_data ioc_data = { 0 };
1493         ENTRY;
1494
1495         /* Tell the MGC we got umount -f */
1496         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1497
1498         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1499                sb->s_count, atomic_read(&sb->s_active));
1500
1501         obd = class_exp2obd(sbi->ll_md_exp);
1502         if (obd == NULL) {
1503                 CERROR("Invalid MDC connection handle "LPX64"\n",
1504                        sbi->ll_md_exp->exp_handle.h_cookie);
1505                 EXIT;
1506                 return;
1507         }
1508         obd->obd_no_recov = 1;
1509         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1510                       &ioc_data, NULL);
1511
1512         obd = class_exp2obd(sbi->ll_dt_exp);
1513         if (obd == NULL) {
1514                 CERROR("Invalid LOV connection handle "LPX64"\n",
1515                        sbi->ll_dt_exp->exp_handle.h_cookie);
1516                 EXIT;
1517                 return;
1518         }
1519
1520         obd->obd_no_recov = 1;
1521         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1522                       &ioc_data, NULL);
1523
1524         /* Really, we'd like to wait until there are no requests outstanding,
1525          * and then continue.  For now, we just invalidate the requests,
1526          * schedule, and hope.
1527          */
1528         schedule();
1529
1530         EXIT;
1531 }
1532
1533 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1534 {
1535         struct ll_sb_info *sbi = ll_s2sbi(sb);
1536         int err;
1537         __u32 read_only;
1538
1539         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1540                 read_only = *flags & MS_RDONLY;
1541                 err = obd_set_info_async(sbi->ll_md_exp, strlen("read-only"),
1542                                          "read-only", sizeof(read_only),
1543                                          &read_only, NULL);
1544                 if (err) {
1545                         CERROR("Failed to change the read-only flag during "
1546                                "remount: %d\n", err);
1547                         return err;
1548                 }
1549
1550                 if (read_only)
1551                         sb->s_flags |= MS_RDONLY;
1552                 else
1553                         sb->s_flags &= ~MS_RDONLY;
1554         }
1555         return 0;
1556 }
1557
1558 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1559                   int offset, struct super_block *sb)
1560 {
1561         struct ll_sb_info *sbi = NULL;
1562         struct lustre_md md;
1563         int rc = 0;
1564         ENTRY;
1565
1566         LASSERT(*inode || sb);
1567         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1568         prune_deathrow(sbi, 1);
1569
1570         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
1571                               sbi->ll_dt_exp, &md);
1572         if (rc)
1573                 RETURN(rc);
1574
1575         if (*inode) {
1576                 ll_update_inode(*inode, &md);
1577         } else {
1578                 LASSERT(sb != NULL);
1579
1580                 /* at this point server answers to client's RPC with same fid as
1581                  * client generated for creating some inode. So using ->fid1 is
1582                  * okay here. */
1583                 LASSERT(fid_num(&md.body->fid1) != 0);
1584
1585                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
1586                 if (*inode == NULL || is_bad_inode(*inode)) {
1587                         md_free_lustre_md(sbi->ll_dt_exp, &md);
1588                         rc = -ENOMEM;
1589                         CERROR("new_inode -fatal: rc %d\n", rc);
1590                         GOTO(out, rc);
1591                 }
1592         }
1593
1594         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
1595                          ll_i2info(*inode)->lli_smd);
1596 out:
1597         RETURN(rc);
1598 }
1599
1600 char *llap_origins[] = {
1601         [LLAP_ORIGIN_UNKNOWN] = "--",
1602         [LLAP_ORIGIN_READPAGE] = "rp",
1603         [LLAP_ORIGIN_READAHEAD] = "ra",
1604         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
1605         [LLAP_ORIGIN_WRITEPAGE] = "wp",
1606 };
1607
1608 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
1609                                              struct list_head *list)
1610 {
1611         struct ll_async_page *llap;
1612         struct list_head *pos;
1613
1614         list_for_each(pos, list) {
1615                 if (pos == &sbi->ll_pglist)
1616                         return NULL;
1617                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
1618                 if (llap->llap_page == NULL)
1619                         continue;
1620                 return llap;
1621         }
1622         LBUG();
1623         return NULL;
1624 }
1625
1626 int ll_obd_statfs(struct inode *inode, void *arg)
1627 {
1628         struct ll_sb_info *sbi = NULL;
1629         struct obd_device *client_obd = NULL, *lov_obd = NULL;
1630         struct lov_obd *lov = NULL;
1631         struct obd_statfs stat_buf = {0};
1632         char *buf = NULL;
1633         struct obd_ioctl_data *data = NULL;
1634         __u32 type, index;
1635         int len, rc;
1636
1637         if (!inode || !(sbi = ll_i2sbi(inode)))
1638                 GOTO(out_statfs, rc = -EINVAL);
1639
1640         rc = obd_ioctl_getdata(&buf, &len, arg);
1641         if (rc)
1642                 GOTO(out_statfs, rc);
1643
1644         data = (void*)buf;
1645         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1646             !data->ioc_pbuf1 || !data->ioc_pbuf2)
1647                 GOTO(out_statfs, rc = -EINVAL);
1648
1649         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1650         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1651
1652         if (type == LL_STATFS_MDC) {
1653                 if (index > 0)
1654                         GOTO(out_statfs, rc = -ENODEV);
1655                 client_obd = class_exp2obd(sbi->ll_md_exp);
1656         } else if (type == LL_STATFS_LOV) {
1657                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
1658                 lov = &lov_obd->u.lov;
1659
1660                 if (index >= lov->desc.ld_tgt_count)
1661                         GOTO(out_statfs, rc = -ENODEV);
1662
1663                 client_obd = class_exp2obd(lov->tgts[index].ltd_exp);
1664                 if (!lov->tgts[index].active)
1665                         GOTO(out_uuid, rc = -ENODATA);
1666         }
1667
1668         if (!client_obd)
1669                 GOTO(out_statfs, rc = -EINVAL);
1670
1671         rc = obd_statfs(client_obd, &stat_buf, jiffies - 1);
1672         if (rc)
1673                 GOTO(out_statfs, rc);
1674
1675         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1676                 GOTO(out_statfs, rc = -EFAULT);
1677
1678 out_uuid:
1679         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
1680                          data->ioc_plen2))
1681                 rc = -EFAULT;
1682
1683 out_statfs:
1684         if (buf)
1685                 obd_ioctl_freedata(buf, len);
1686         return rc;
1687 }