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 <linux/lustre_idl.h>
  32 #include <linux/lustre_lite.h>
  33 #include <linux/lustre_ha.h>
  34 #include <linux/lustre_ver.h>
  35 #include <linux/lustre_dlm.h>
  36 #include <linux/lprocfs_status.h>
  37 #include <linux/lustre_disk.h>
  38 #include "llite_internal.h"
  39
  40 kmem_cache_t *ll_file_data_slab;
  41
  42 LIST_HEAD(ll_super_blocks);
  43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52
  53 struct ll_sb_info *ll_init_sbi(void)
  54 {
  55         struct ll_sb_info *sbi = NULL;
  56         class_uuid_t uuid;
  57         ENTRY;
  58
  59         OBD_ALLOC(sbi, sizeof(*sbi));
  60         if (!sbi)
  61                 RETURN(NULL);
  62
  63         spin_lock_init(&sbi->ll_lock);
  64         spin_lock_init(&sbi->ll_lco.lco_lock);
  65         INIT_LIST_HEAD(&sbi->ll_pglist);
  66         sbi->ll_pglist_gen = 0;
  67         if (num_physpages >> (20 - PAGE_SHIFT) < 512)
  68                 sbi->ll_async_page_max = num_physpages / 2;
  69         else
  70                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  71         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  72                                            SBI_DEFAULT_READAHEAD_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 int client_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 109 {
 110         struct inode *root = 0;
 111         struct ll_sb_info *sbi = ll_s2sbi(sb);
 112         struct obd_device *obd;
 113         struct lu_fid rootfid;
 114         struct obd_statfs osfs;
 115         struct ptlrpc_request *request = NULL;
 116         struct lustre_handle osc_conn = {0, };
 117         struct lustre_handle mdc_conn = {0, };
 118         struct lustre_md md;
 119         struct obd_connect_data *data = NULL;
 120         int err;
 121         ENTRY;
 122
 123         obd = class_name2obd(mdc);
 124         if (!obd) {
 125                 CERROR("MDC %s: not setup or attached\n", mdc);
 126                 RETURN(-EINVAL);
 127         }
 128
 129         OBD_ALLOC(data, sizeof(*data));
 130         if (data == NULL)
 131                 RETURN(-ENOMEM);
 132
 133         if (proc_lustre_fs_root) {
 134                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 135                                                   osc, mdc);
 136                 if (err < 0)
 137                         CERROR("could not register mount in /proc/lustre");
 138         }
 139
 140         /* indicate that inodebits locking is supported by this client */
 141         data->ocd_connect_flags |= OBD_CONNECT_IBITS;
 142         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 143
 144         if (sb->s_flags & MS_RDONLY)
 145                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 146         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 147                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 148         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_JOIN;
 149
 150         if (sbi->ll_flags & LL_SBI_FLOCK) {
 151                 sbi->ll_fop = &ll_file_operations_flock;
 152         } else {
 153                 sbi->ll_fop = &ll_file_operations;
 154         }
 155
 156         data->ocd_connect_flags |= OBD_CONNECT_VERSION;
 157         data->ocd_version = LUSTRE_VERSION_CODE;
 158
 159         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data);
 160         if (err == -EBUSY) {
 161                 CERROR("An MDT (mdc %s) is performing recovery, of which this"
 162                        " client is not a part.  Please wait for recovery to "
 163                        "complete, abort, or time out.\n", mdc);
 164                 GOTO(out, err);
 165         } else if (err) {
 166                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 167                 GOTO(out, err);
 168         }
 169         sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
 170
 171         err = obd_statfs(obd, &osfs, jiffies - HZ);
 172         if (err)
 173                 GOTO(out_mdc, err);
 174
 175         /* async connect is surely finished by now */
 176         *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
 177
 178         LASSERT(osfs.os_bsize);
 179         sb->s_blocksize = osfs.os_bsize;
 180         sb->s_blocksize_bits = log2(osfs.os_bsize);
 181         sb->s_magic = LL_SUPER_MAGIC;
 182         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 183         sbi->ll_namelen = osfs.os_namelen;
 184
 185         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 186             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 187                 LCONSOLE_INFO("Disabling user_xattr feature because "
 188                               "it is not supported on the server\n");
 189                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 190         }
 191
 192         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 193 #ifdef MS_POSIXACL
 194                 sb->s_flags |= MS_POSIXACL;
 195 #endif
 196                 sbi->ll_flags |= LL_SBI_ACL;
 197         } else
 198                 sbi->ll_flags &= ~LL_SBI_ACL;
 199
 200         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 201                 sbi->ll_flags |= LL_SBI_JOIN;
 202
 203 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 204         /* We set sb->s_dev equal on all lustre clients in order to support
 205          * NFS export clustering.  NFSD requires that the FSID be the same
 206          * on all clients. */
 207         /* s_dev is also used in lt_compare() to compare two fs, but that is
 208          * only a node-local comparison. */
 209         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
 210                          strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
 211 #endif
 212
 213         obd = class_name2obd(osc);
 214         if (!obd) {
 215                 CERROR("OSC %s: not setup or attached\n", osc);
 216                 GOTO(out_mdc, err);
 217         }
 218
 219         data->ocd_connect_flags =
 220                 OBD_CONNECT_GRANT|OBD_CONNECT_VERSION|OBD_CONNECT_REQPORTAL;
 221
 222         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 223                "ocd_grant: %d\n", data->ocd_connect_flags,
 224                data->ocd_version, data->ocd_grant);
 225
 226         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 227         obd->obd_upcall.onu_upcall = ll_ocd_update;
 228
 229         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data);
 230         if (err == -EBUSY) {
 231                 CERROR("An OST (osc %s) is performing recovery, of which this"
 232                        " client is not a part.  Please wait for recovery to "
 233                        "complete, abort, or time out.\n", osc);
 234                 GOTO(out, err);
 235         } else if (err) {
 236                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 237                 GOTO(out_mdc, err);
 238         }
 239         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 240         spin_lock(&sbi->ll_lco.lco_lock);
 241         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 242         spin_unlock(&sbi->ll_lco.lco_lock);
 243
 244         mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
 245
 246         err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
 247                                   0, NULL, NULL, NULL);
 248         if (err < 0) {
 249                 LCONSOLE_ERROR("There are no OST's in this filesystem. "
 250                                "There must be at least one active OST for "
 251                                "a client to start.\n");
 252                 GOTO(out_osc, err);
 253         }
 254
 255         if (!ll_async_page_slab) {
 256                 ll_async_page_slab_size =
 257                         size_round(sizeof(struct ll_async_page)) + err;
 258                 ll_async_page_slab = kmem_cache_create("ll_async_page",
 259                                                        ll_async_page_slab_size,
 260                                                        0, 0, NULL, NULL);
 261                 if (!ll_async_page_slab)
 262                         GOTO(out_osc, -ENOMEM);
 263         }
 264
 265         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
 266         if (err) {
 267                 CERROR("cannot mds_connect: rc = %d\n", err);
 268                 GOTO(out_osc, err);
 269         }
 270         CDEBUG(D_SUPER, "rootfid "DFID3"\n", PFID3(&rootfid));
 271         sbi->ll_root_fid = rootfid;
 272
 273         sb->s_op = &lustre_super_operations;
 274
 275         /* make root inode
 276          * XXX: move this to after cbd setup? */
 277         err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
 278                           OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
 279                           (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
 280                           0, &request);
 281         if (err) {
 282                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 283                 GOTO(out_osc, err);
 284         }
 285
 286         err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
 287         if (err) {
 288                 CERROR("failed to understand root inode md: rc = %d\n", err);
 289                 ptlrpc_req_finished (request);
 290                 GOTO(out_osc, err);
 291         }
 292
 293         LASSERT(fid_oid(&sbi->ll_root_fid) != 0);
 294         root = ll_iget(sb, ll_fid2ino(sbi, &sbi->ll_root_fid), &md);
 295         ll_i2info(root)->lli_fid = sbi->ll_root_fid;
 296         ptlrpc_req_finished(request);
 297
 298         if (root == NULL || is_bad_inode(root)) {
 299                 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
 300                 CERROR("lustre_lite: bad iget4 for root\n");
 301                 GOTO(out_root, err = -EBADF);
 302         }
 303
 304         err = ll_close_thread_start(&sbi->ll_lcq);
 305         if (err) {
 306                 CERROR("cannot start close thread: rc %d\n", err);
 307                 GOTO(out_root, err);
 308         }
 309
 310         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 311            backing dev info assigned to inode mapping is used for
 312            determining maximal readahead. */
 313 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 314     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 315         /* bug 2805 - set VM readahead to zero */
 316         vm_max_readahead = vm_min_readahead = 0;
 317 #endif
 318
 319         sb->s_root = d_alloc_root(root);
 320         if (data != NULL)
 321                 OBD_FREE(data, sizeof(*data));
 322         sb->s_root->d_op = &ll_d_root_ops;
 323         RETURN(err);
 324
 325 out_root:
 326         if (root)
 327                 iput(root);
 328 out_osc:
 329         obd_disconnect(sbi->ll_osc_exp);
 330 out_mdc:
 331         obd_disconnect(sbi->ll_mdc_exp);
 332 out:
 333         if (data != NULL)
 334                 OBD_FREE(data, sizeof(*data));
 335         lprocfs_unregister_mountpoint(sbi);
 336         RETURN(err);
 337 }
 338
 339 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 340 {
 341         int size, rc;
 342
 343         *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
 344         size = sizeof(int);
 345         rc = obd_get_info(sbi->ll_mdc_exp, strlen("max_easize"), "max_easize",
 346                           &size, lmmsize);
 347         if (rc)
 348                 CERROR("Get max mdsize error rc %d \n", rc);
 349
 350         RETURN(rc);
 351 }
 352
 353 void ll_dump_inode(struct inode *inode)
 354 {
 355         struct list_head *tmp;
 356         int dentry_count = 0;
 357
 358         LASSERT(inode != NULL);
 359
 360         list_for_each(tmp, &inode->i_dentry)
 361                 dentry_count++;
 362
 363         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 364                inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
 365                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 366 }
 367
 368 void lustre_dump_dentry(struct dentry *dentry, int recur)
 369 {
 370         struct list_head *tmp;
 371         int subdirs = 0;
 372
 373         LASSERT(dentry != NULL);
 374
 375         list_for_each(tmp, &dentry->d_subdirs)
 376                 subdirs++;
 377
 378         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 379                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 380                dentry->d_name.len, dentry->d_name.name,
 381                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 382                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 383                dentry->d_flags, dentry->d_fsdata, subdirs);
 384         if (dentry->d_inode != NULL)
 385                 ll_dump_inode(dentry->d_inode);
 386
 387         if (recur == 0)
 388                 return;
 389
 390         list_for_each(tmp, &dentry->d_subdirs) {
 391                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 392                 lustre_dump_dentry(d, recur - 1);
 393         }
 394 }
 395
 396 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 397 void lustre_throw_orphan_dentries(struct super_block *sb)
 398 {
 399         struct hlist_node *tmp, *next;
 400         struct ll_sb_info *sbi = ll_s2sbi(sb);
 401
 402         /* Do this to get rid of orphaned dentries. That is not really trw. */
 403         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 404                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 405                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 406                       "before and after shrink_dcache_parent\n",
 407                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 408                 lustre_dump_dentry(dentry, 1);
 409                 shrink_dcache_parent(dentry);
 410                 lustre_dump_dentry(dentry, 1);
 411         }
 412 }
 413 #else
 414 #define lustre_throw_orphan_dentries(sb)
 415 #endif
 416
 417 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 418 {
 419         LIST_HEAD(throw_away);
 420         int locked = 0;
 421         ENTRY;
 422
 423         if (try) {
 424                 locked = spin_trylock(&sbi->ll_deathrow_lock);
 425         } else {
 426                 spin_lock(&sbi->ll_deathrow_lock);
 427                 locked = 1;
 428         }
 429
 430         if (!locked) {
 431                 EXIT;
 432                 return;
 433         }
 434
 435         list_splice_init(&sbi->ll_deathrow, &throw_away);
 436         spin_unlock(&sbi->ll_deathrow_lock);
 437
 438         while (!list_empty(&throw_away)) {
 439                 struct ll_inode_info *lli;
 440                 struct inode *inode;
 441
 442                 lli = list_entry(throw_away.next, struct ll_inode_info,
 443                                  lli_dead_list);
 444                 list_del_init(&lli->lli_dead_list);
 445
 446                 inode = ll_info2i(lli);
 447                 d_prune_aliases(inode);
 448
 449                 CDEBUG(D_INODE, "prune duplicate inode %p inum %lu count %u\n",
 450                        inode, inode->i_ino, atomic_read(&inode->i_count));
 451                 iput(inode);
 452         }
 453         EXIT;
 454 }
 455
 456 void client_common_put_super(struct super_block *sb)
 457 {
 458         struct ll_sb_info *sbi = ll_s2sbi(sb);
 459         ENTRY;
 460
 461         ll_close_thread_shutdown(sbi->ll_lcq);
 462
 463         /* destroy inodes in deathrow */
 464         prune_deathrow(sbi, 0);
 465
 466         list_del(&sbi->ll_conn_chain);
 467         obd_disconnect(sbi->ll_osc_exp);
 468
 469         lprocfs_unregister_mountpoint(sbi);
 470         if (sbi->ll_proc_root) {
 471                 lprocfs_remove(sbi->ll_proc_root);
 472                 sbi->ll_proc_root = NULL;
 473         }
 474
 475         obd_disconnect(sbi->ll_mdc_exp);
 476
 477         lustre_throw_orphan_dentries(sb);
 478         EXIT;
 479 }
 480
 481 char *ll_read_opt(const char *opt, char *data)
 482 {
 483         char *value;
 484         char *retval;
 485         ENTRY;
 486
 487         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 488         if (strncmp(opt, data, strlen(opt)))
 489                 RETURN(NULL);
 490         if ((value = strchr(data, '=')) == NULL)
 491                 RETURN(NULL);
 492
 493         value++;
 494         OBD_ALLOC(retval, strlen(value) + 1);
 495         if (!retval) {
 496                 CERROR("out of memory!\n");
 497                 RETURN(NULL);
 498         }
 499
 500         memcpy(retval, value, strlen(value)+1);
 501         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 502         RETURN(retval);
 503 }
 504
 505 static inline int ll_set_opt(const char *opt, char *data, int fl)
 506 {
 507         if (strncmp(opt, data, strlen(opt)) != 0)
 508                 return(0);
 509         else
 510                 return(fl);
 511 }
 512
 513 /* non-client-specific mount options are parsed in lmd_parse */
 514 void ll_options(char *options, int *flags)
 515 {
 516         int tmp;
 517         char *s1 = options, *s2;
 518         ENTRY;
 519
 520         if (!options) {
 521                 EXIT;
 522                 return;
 523         }
 524
 525         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 526
 527         while (*s1) {
 528                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 529                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 530                 if (tmp) {
 531                         *flags |= tmp;
 532                         goto next;
 533                 }
 534                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 535                 if (tmp) {
 536                         *flags |= tmp;
 537                         goto next;
 538                 }
 539                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK);
 540                 if (tmp) {
 541                         *flags &= ~tmp;
 542                         goto next;
 543                 }
 544                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 545                 if (tmp) {
 546                         *flags |= tmp;
 547                         goto next;
 548                 }
 549                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 550                 if (tmp) {
 551                         *flags &= ~tmp;
 552                         goto next;
 553                 }
 554                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 555                 if (tmp) {
 556                         /* Ignore deprecated mount option.  The client will
 557                          * always try to mount with ACL support, whether this
 558                          * is used depends on whether server supports it. */
 559                         goto next;
 560                 }
 561                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 562                 if (tmp) {
 563                         goto next;
 564                 }
 565
 566 next:
 567                 /* Find next opt */
 568                 s2 = strchr(s1, ',');
 569                 if (s2 == NULL)
 570                         break;
 571                 s1 = s2 + 1;
 572         }
 573         EXIT;
 574 }
 575
 576 void ll_lli_init(struct ll_inode_info *lli)
 577 {
 578         sema_init(&lli->lli_open_sem, 1);
 579         sema_init(&lli->lli_size_sem, 1);
 580         lli->lli_flags = 0;
 581         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 582         spin_lock_init(&lli->lli_lock);
 583         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 584         lli->lli_inode_magic = LLI_INODE_MAGIC;
 585         INIT_LIST_HEAD(&lli->lli_dead_list);
 586 }
 587
 588 int ll_fill_super(struct super_block *sb)
 589 {
 590         struct lustre_profile *lprof;
 591         struct lustre_sb_info *lsi = s2lsi(sb);
 592         struct ll_sb_info *sbi;
 593         char  *osc = NULL;
 594         char  *mdc = NULL;
 595         char  *profilenm = get_profile_name(sb);
 596         struct config_llog_instance cfg;
 597         char   ll_instance[sizeof(sb) * 2 + 1];
 598         int    err;
 599         ENTRY;
 600
 601         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 602
 603         /* client additional sb info */
 604         lsi->lsi_llsbi = sbi = ll_init_sbi();
 605         if (!sbi)
 606                 RETURN(-ENOMEM);
 607
 608         ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 609
 610         /* Generate a string unique to this super, in case some joker tries
 611            to mount the same fs at two mount points.
 612            Use the address of the super itself.*/
 613         sprintf(ll_instance, "%p", sb);
 614         cfg.cfg_instance = ll_instance;
 615         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 616         cfg.cfg_last_idx = 0;
 617
 618         /* set up client obds */
 619         err = lustre_process_log(sb, profilenm, &cfg);
 620         if (err < 0) {
 621                 CERROR("Unable to process log: %d\n", err);
 622                 GOTO(out_free, err);
 623         }
 624
 625         lprof = class_get_profile(profilenm);
 626         if (lprof == NULL) {
 627                 CERROR("No profile found: %s\n", profilenm);
 628                 GOTO(out_free, err = -EINVAL);
 629         }
 630         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 631                lprof->lp_mdc, lprof->lp_osc);
 632
 633         OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 634                   strlen(ll_instance) + 2);
 635         if (!osc)
 636                 GOTO(out_free, err = -ENOMEM);
 637         sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
 638
 639         OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 640                   strlen(ll_instance) + 2);
 641         if (!mdc)
 642                 GOTO(out_free, err = -ENOMEM);
 643         sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
 644
 645         /* connections, registrations, sb setup */
 646         err = client_common_fill_super(sb, mdc, osc);
 647
 648 out_free:
 649         if (mdc)
 650                 OBD_FREE(mdc, strlen(mdc) + 1);
 651         if (osc)
 652                 OBD_FREE(osc, strlen(osc) + 1);
 653         if (err) {
 654                 struct obd_device *obd;
 655                 int next = 0;
 656                 /* like client_put_super below */
 657                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 658                        != NULL) {
 659                         class_manual_cleanup(obd);
 660                 }
 661                 class_del_profile(profilenm);
 662                 ll_free_sbi(sb);
 663                 lsi->lsi_llsbi = NULL;
 664         }
 665         RETURN(err);
 666 } /* ll_fill_super */
 667
 668
 669 void ll_put_super(struct super_block *sb)
 670 {
 671         struct config_llog_instance cfg;
 672         char   ll_instance[sizeof(sb) * 2 + 1];
 673         struct obd_device *obd;
 674         struct lustre_sb_info *lsi = s2lsi(sb);
 675         struct ll_sb_info *sbi = ll_s2sbi(sb);
 676         char *profilenm = get_profile_name(sb);
 677         int next = 0;
 678         ENTRY;
 679
 680         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 681
 682         sprintf(ll_instance, "%p", sb);
 683         cfg.cfg_instance = ll_instance;
 684         lustre_end_log(sb, NULL, &cfg);
 685
 686         obd = class_exp2obd(sbi->ll_mdc_exp);
 687         if (obd) {
 688                 int next = 0;
 689                 int force = obd->obd_no_recov;
 690                 /* We need to set force before the lov_disconnect in
 691                 lustre_common_put_super, since l_d cleans up osc's as well. */
 692                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 693                        != NULL) {
 694                         obd->obd_force = force;
 695                 }
 696         }
 697
 698         client_common_put_super(sb);
 699
 700         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 701                 class_manual_cleanup(obd);
 702         }
 703
 704         if (profilenm)
 705                 class_del_profile(profilenm);
 706
 707         ll_free_sbi(sb);
 708         lsi->lsi_llsbi = NULL;
 709
 710         lustre_common_put_super(sb);
 711
 712         CDEBUG(D_WARNING, "client umount done\n");
 713         EXIT;
 714 } /* client_put_super */
 715
 716 #ifdef HAVE_REGISTER_CACHE
 717 #include <linux/cache_def.h>
 718 #ifdef HAVE_CACHE_RETURN_INT
 719 static int
 720 #else
 721 static void
 722 #endif
 723 ll_shrink_cache(int priority, unsigned int gfp_mask)
 724 {
 725         struct ll_sb_info *sbi;
 726         int count = 0;
 727
 728         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
 729                 count += llap_shrink_cache(sbi, priority);
 730
 731 #ifdef HAVE_CACHE_RETURN_INT
 732         return count;
 733 #endif
 734 }
 735
 736 struct cache_definition ll_cache_definition = {
 737         .name = "llap_cache",
 738         .shrink = ll_shrink_cache
 739 };
 740 #endif /* HAVE_REGISTER_CACHE */
 741
 742 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 743 {
 744         struct inode *inode = NULL;
 745         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 746         if (lock->l_ast_data) {
 747                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 748                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 749                         inode = igrab(lock->l_ast_data);
 750                 } else {
 751                         inode = lock->l_ast_data;
 752                         __LDLM_DEBUG(inode->i_state & I_FREEING ?
 753                                      D_INFO : D_WARNING, lock,
 754                                      "l_ast_data %p is bogus: magic %08x",
 755                                      lock->l_ast_data, lli->lli_inode_magic);
 756                         inode = NULL;
 757                 }
 758         }
 759         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 760         return inode;
 761 }
 762
 763 static int null_if_equal(struct ldlm_lock *lock, void *data)
 764 {
 765         if (data == lock->l_ast_data) {
 766                 lock->l_ast_data = NULL;
 767
 768                 if (lock->l_req_mode != lock->l_granted_mode)
 769                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
 770         }
 771
 772         return LDLM_ITER_CONTINUE;
 773 }
 774
 775 void ll_clear_inode(struct inode *inode)
 776 {
 777         struct lu_fid fid;
 778         struct ll_inode_info *lli = ll_i2info(inode);
 779         struct ll_sb_info *sbi = ll_i2sbi(inode);
 780         ENTRY;
 781
 782         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 783                inode->i_generation, inode);
 784
 785         ll_inode2fid(&fid, inode);
 786         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 787         mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 788
 789         if (lli->lli_smd) {
 790                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
 791                                   null_if_equal, inode);
 792
 793                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
 794                 lli->lli_smd = NULL;
 795         }
 796
 797         if (lli->lli_symlink_name) {
 798                 OBD_FREE(lli->lli_symlink_name,
 799                          strlen(lli->lli_symlink_name) + 1);
 800                 lli->lli_symlink_name = NULL;
 801         }
 802
 803 #ifdef CONFIG_FS_POSIX_ACL
 804         if (lli->lli_posix_acl) {
 805                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
 806                 posix_acl_release(lli->lli_posix_acl);
 807                 lli->lli_posix_acl = NULL;
 808         }
 809 #endif
 810
 811         lli->lli_inode_magic = LLI_INODE_DEAD;
 812
 813         spin_lock(&sbi->ll_deathrow_lock);
 814         list_del_init(&lli->lli_dead_list);
 815         spin_unlock(&sbi->ll_deathrow_lock);
 816
 817         EXIT;
 818 }
 819
 820 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 821  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 822  * keep these values until such a time that objects are allocated for it.
 823  * We do the MDS operations first, as it is checking permissions for us.
 824  * We don't to the MDS RPC if there is nothing that we want to store there,
 825  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 826  * going to do an RPC anyways.
 827  *
 828  * If we are doing a truncate, we will send the mtime and ctime updates
 829  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 830  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 831  * at the same time.
 832  */
 833 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 834 {
 835         struct ll_inode_info *lli = ll_i2info(inode);
 836         struct lov_stripe_md *lsm = lli->lli_smd;
 837         struct ll_sb_info *sbi = ll_i2sbi(inode);
 838         struct ptlrpc_request *request = NULL;
 839         struct mdc_op_data op_data;
 840         int ia_valid = attr->ia_valid;
 841         int rc = 0;
 842         ENTRY;
 843
 844         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
 845                attr->ia_valid);
 846         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
 847
 848         if (ia_valid & ATTR_SIZE) {
 849                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 850                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 851                                attr->ia_size, ll_file_maxbytes(inode));
 852                         RETURN(-EFBIG);
 853                 }
 854
 855                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 856         }
 857
 858         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
 859         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
 860                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 861                         RETURN(-EPERM);
 862         }
 863
 864         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 865         if (attr->ia_valid & ATTR_CTIME) {
 866                 attr->ia_ctime = CURRENT_TIME;
 867                 attr->ia_valid |= ATTR_CTIME_SET;
 868         }
 869         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 870                 attr->ia_atime = CURRENT_TIME;
 871                 attr->ia_valid |= ATTR_ATIME_SET;
 872         }
 873         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 874                 attr->ia_mtime = CURRENT_TIME;
 875                 attr->ia_valid |= ATTR_MTIME_SET;
 876         }
 877
 878         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 879                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
 880                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 881                        CURRENT_SECONDS);
 882
 883
 884         /* NB: ATTR_SIZE will only be set after this point if the size
 885          * resides on the MDS, ie, this file has no objects. */
 886         if (lsm)
 887                 attr->ia_valid &= ~ATTR_SIZE;
 888
 889         /* If only OST attributes being set on objects, don't do MDS RPC.
 890          * In that case, we need to check permissions and update the local
 891          * inode ourselves so we can call obdo_from_inode() always. */
 892         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 893                 struct lustre_md md;
 894                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 895
 896                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
 897                                  attr, NULL, 0, NULL, 0, &request);
 898
 899                 if (rc) {
 900                         ptlrpc_req_finished(request);
 901                         if (rc != -EPERM && rc != -EACCES)
 902                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
 903                         RETURN(rc);
 904                 }
 905
 906                 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
 907                 if (rc) {
 908                         ptlrpc_req_finished(request);
 909                         RETURN(rc);
 910                 }
 911
 912                 /* We call inode_setattr to adjust timestamps.
 913                  * If there is at least some data in file, we cleared ATTR_SIZE
 914                  * above to avoid invoking vmtruncate, otherwise it is important
 915                  * to call vmtruncate in inode_setattr to update inode->i_size
 916                  * (bug 6196) */
 917                 rc = inode_setattr(inode, attr);
 918
 919                 ll_update_inode(inode, &md);
 920                 ptlrpc_req_finished(request);
 921
 922                 if (!lsm || !S_ISREG(inode->i_mode)) {
 923                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 924                         RETURN(rc);
 925                 }
 926         } else {
 927                 /* The OST doesn't check permissions, but the alternative is
 928                  * a gratuitous RPC to the MDS.  We already rely on the client
 929                  * to do read/write/truncate permission checks, so is mtime OK?
 930                  */
 931                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 932                         /* from sys_utime() */
 933                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 934                                 if (current->fsuid != inode->i_uid &&
 935                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
 936                                         RETURN(rc);
 937                         } else {
 938                                 /* from inode_change_ok() */
 939                                 if (current->fsuid != inode->i_uid &&
 940                                     !capable(CAP_FOWNER))
 941                                         RETURN(-EPERM);
 942                         }
 943                 }
 944
 945                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
 946                 rc = inode_setattr(inode, attr);
 947         }
 948
 949         /* We really need to get our PW lock before we change inode->i_size.
 950          * If we don't we can race with other i_size updaters on our node, like
 951          * ll_file_read.  We can also race with i_size propogation to other
 952          * nodes through dirtying and writeback of final cached pages.  This
 953          * last one is especially bad for racing o_append users on other
 954          * nodes. */
 955         if (ia_valid & ATTR_SIZE) {
 956                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
 957                                                            OBD_OBJECT_EOF } };
 958                 struct lustre_handle lockh = { 0 };
 959                 int err, ast_flags = 0;
 960                 /* XXX when we fix the AST intents to pass the discard-range
 961                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
 962                  * XXX here. */
 963                 if (attr->ia_size == 0)
 964                         ast_flags = LDLM_AST_DISCARD_DATA;
 965
 966                 up(&inode->i_sem);
 967                 UP_WRITE_I_ALLOC_SEM(inode);
 968                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
 969                                     ast_flags);
 970 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 971                 DOWN_WRITE_I_ALLOC_SEM(inode);
 972                 down(&inode->i_sem);
 973 #else
 974                 down(&inode->i_sem);
 975                 DOWN_WRITE_I_ALLOC_SEM(inode);
 976 #endif
 977                 if (rc != 0)
 978                         RETURN(rc);
 979
 980                 /* Only ll_inode_size_lock is taken at this level.
 981                  * lov_stripe_lock() is grabbed by ll_truncate() only over
 982                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
 983                  * ll_truncate dropped ll_inode_size_lock() */
 984                 ll_inode_size_lock(inode, 0);
 985                 rc = vmtruncate(inode, attr->ia_size);
 986                 if (rc != 0) {
 987                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
 988                         ll_inode_size_unlock(inode, 0);
 989                 }
 990
 991                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
 992                 if (err) {
 993                         CERROR("ll_extent_unlock failed: %d\n", err);
 994                         if (!rc)
 995                                 rc = err;
 996                 }
 997         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
 998                 obd_flag flags;
 999                 struct obdo oa;
1000
1001                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1002                        inode->i_ino, LTIME_S(attr->ia_mtime));
1003
1004                 oa.o_id = lsm->lsm_object_id;
1005                 oa.o_valid = OBD_MD_FLID;
1006
1007                 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1008                         OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1009                         OBD_MD_FLFID | OBD_MD_FLGENER;
1010
1011                 obdo_from_inode(&oa, inode, flags);
1012                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1013                 if (rc)
1014                         CERROR("obd_setattr fails: rc=%d\n", rc);
1015         }
1016         RETURN(rc);
1017 }
1018
1019 int ll_setattr(struct dentry *de, struct iattr *attr)
1020 {
1021         LBUG(); /* code is unused, but leave this in case of VFS changes */
1022         RETURN(-ENOSYS);
1023 }
1024
1025 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1026                        unsigned long max_age)
1027 {
1028         struct ll_sb_info *sbi = ll_s2sbi(sb);
1029         struct obd_statfs obd_osfs;
1030         int rc;
1031         ENTRY;
1032
1033         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1034         if (rc) {
1035                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1036                 RETURN(rc);
1037         }
1038
1039         osfs->os_type = sb->s_magic;
1040
1041         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1042                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1043
1044         rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1045         if (rc) {
1046                 CERROR("obd_statfs fails: rc = %d\n", rc);
1047                 RETURN(rc);
1048         }
1049
1050         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1051                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1052                obd_osfs.os_files);
1053
1054         osfs->os_blocks = obd_osfs.os_blocks;
1055         osfs->os_bfree = obd_osfs.os_bfree;
1056         osfs->os_bavail = obd_osfs.os_bavail;
1057
1058         /* If we don't have as many objects free on the OST as inodes
1059          * on the MDS, we reduce the total number of inodes to
1060          * compensate, so that the "inodes in use" number is correct.
1061          */
1062         if (obd_osfs.os_ffree < osfs->os_ffree) {
1063                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1064                         obd_osfs.os_ffree;
1065                 osfs->os_ffree = obd_osfs.os_ffree;
1066         }
1067
1068         RETURN(rc);
1069 }
1070
1071 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1072 {
1073         struct obd_statfs osfs;
1074         int rc;
1075
1076         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1077         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1078
1079         /* For now we will always get up-to-date statfs values, but in the
1080          * future we may allow some amount of caching on the client (e.g.
1081          * from QOS or lprocfs updates). */
1082         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1083         if (rc)
1084                 return rc;
1085
1086         statfs_unpack(sfs, &osfs);
1087
1088         if (sizeof(sfs->f_blocks) == 4) {
1089                 while (osfs.os_blocks > ~0UL) {
1090                         sfs->f_bsize <<= 1;
1091
1092                         osfs.os_blocks >>= 1;
1093                         osfs.os_bfree >>= 1;
1094                         osfs.os_bavail >>= 1;
1095                 }
1096         }
1097
1098         sfs->f_blocks = osfs.os_blocks;
1099         sfs->f_bfree = osfs.os_bfree;
1100         sfs->f_bavail = osfs.os_bavail;
1101
1102         return 0;
1103 }
1104
1105 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1106 {
1107         struct ll_inode_info *lli;
1108         struct lov_stripe_md *lsm;
1109
1110         lli = ll_i2info(inode);
1111         LASSERT(lli->lli_size_sem_owner != current);
1112         down(&lli->lli_size_sem);
1113         LASSERT(lli->lli_size_sem_owner == NULL);
1114         lli->lli_size_sem_owner = current;
1115         lsm = lli->lli_smd;
1116         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1117                  lsm, lock_lsm);
1118         if (lock_lsm)
1119                 lov_stripe_lock(lsm);
1120 }
1121
1122 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1123 {
1124         struct ll_inode_info *lli;
1125         struct lov_stripe_md *lsm;
1126
1127         lli = ll_i2info(inode);
1128         lsm = lli->lli_smd;
1129         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1130                  lsm, unlock_lsm);
1131         if (unlock_lsm)
1132                 lov_stripe_unlock(lsm);
1133         LASSERT(lli->lli_size_sem_owner == current);
1134         lli->lli_size_sem_owner = NULL;
1135         up(&lli->lli_size_sem);
1136 }
1137
1138 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1139 {
1140         struct ll_inode_info *lli = ll_i2info(inode);
1141
1142         dump_lsm(D_INODE, lsm);
1143         dump_lsm(D_INODE, lli->lli_smd);
1144         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1145                  "lsm must be joined lsm %p\n", lsm);
1146         obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1147         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1148                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1149         lli->lli_smd = lsm;
1150         lli->lli_maxbytes = lsm->lsm_maxbytes;
1151         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1152                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1153 }
1154
1155 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1156 {
1157         struct ll_inode_info *lli = ll_i2info(inode);
1158         struct mdt_body *body = md->body;
1159         struct lov_stripe_md *lsm = md->lsm;
1160
1161         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1162         if (lsm != NULL) {
1163                 if (lli->lli_smd == NULL) {
1164                         if (lsm->lsm_magic != LOV_MAGIC &&
1165                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1166                                 dump_lsm(D_ERROR, lsm);
1167                                 LBUG();
1168                         }
1169                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1170                                lsm, inode->i_ino, inode->i_generation, inode);
1171                         /* ll_inode_size_lock() requires it is only called
1172                          * with lli_smd != NULL or lock_lsm == 0 or we can
1173                          * race between lock/unlock.  bug 9547 */
1174                         lli->lli_smd = lsm;
1175                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1176                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1177                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1178                 } else {
1179                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1180                              lli->lli_smd->lsm_stripe_count ==
1181                                         lsm->lsm_stripe_count) {
1182                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1183                                         CERROR("lsm mismatch for inode %ld\n",
1184                                                 inode->i_ino);
1185                                         CERROR("lli_smd:\n");
1186                                         dump_lsm(D_ERROR, lli->lli_smd);
1187                                         CERROR("lsm:\n");
1188                                         dump_lsm(D_ERROR, lsm);
1189                                         LBUG();
1190                                 }
1191                         } else
1192                                 ll_replace_lsm(inode, lsm);
1193                 }
1194                 /* bug 2844 - limit i_blksize for broken user-space apps */
1195                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1196                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1197                 if (lli->lli_smd != lsm)
1198                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1199         } else {
1200                 inode->i_blksize = max(inode->i_blksize,
1201                                        inode->i_sb->s_blocksize);
1202         }
1203
1204 #ifdef CONFIG_FS_POSIX_ACL
1205         LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1206         if (body->valid & OBD_MD_FLACL) {
1207                 spin_lock(&lli->lli_lock);
1208                 if (lli->lli_posix_acl)
1209                         posix_acl_release(lli->lli_posix_acl);
1210                 lli->lli_posix_acl = md->posix_acl;
1211                 spin_unlock(&lli->lli_lock);
1212         }
1213 #endif
1214
1215         if (body->valid & OBD_MD_FLID)
1216                 inode->i_ino = body->ino;
1217         if (body->valid & OBD_MD_FLATIME &&
1218             body->atime > LTIME_S(inode->i_atime))
1219                 LTIME_S(inode->i_atime) = body->atime;
1220         if (body->valid & OBD_MD_FLMTIME &&
1221             body->mtime > LTIME_S(inode->i_mtime)) {
1222                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
1223                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1224                 LTIME_S(inode->i_mtime) = body->mtime;
1225         }
1226         if (body->valid & OBD_MD_FLCTIME &&
1227             body->ctime > LTIME_S(inode->i_ctime))
1228                 LTIME_S(inode->i_ctime) = body->ctime;
1229         if (body->valid & OBD_MD_FLMODE)
1230                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1231         if (body->valid & OBD_MD_FLTYPE)
1232                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1233         if (body->valid & OBD_MD_FLUID)
1234                 inode->i_uid = body->uid;
1235         if (body->valid & OBD_MD_FLGID)
1236                 inode->i_gid = body->gid;
1237         if (body->valid & OBD_MD_FLFLAGS)
1238                 inode->i_flags = body->flags;
1239         if (body->valid & OBD_MD_FLNLINK)
1240                 inode->i_nlink = body->nlink;
1241         if (body->valid & OBD_MD_FLGENER)
1242                 inode->i_generation = body->generation;
1243         if (body->valid & OBD_MD_FLRDEV)
1244 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1245                 inode->i_rdev = body->rdev;
1246 #else
1247                 inode->i_rdev = old_decode_dev(body->rdev);
1248 #endif
1249         if (body->valid & OBD_MD_FLSIZE)
1250                 inode->i_size = body->size;
1251         if (body->valid & OBD_MD_FLBLOCKS)
1252                 inode->i_blocks = body->blocks;
1253
1254         if (body->valid & OBD_MD_FLSIZE)
1255                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1256 }
1257
1258 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1259 static struct backing_dev_info ll_backing_dev_info = {
1260         .ra_pages       = 0,    /* No readahead */
1261 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1262         .capabilities   = 0,    /* Does contribute to dirty memory */
1263 #else
1264         .memory_backed  = 0,    /* Does contribute to dirty memory */
1265 #endif
1266 };
1267 #endif
1268
1269 void ll_read_inode2(struct inode *inode, void *opaque)
1270 {
1271         struct lustre_md *md = opaque;
1272         struct ll_inode_info *lli = ll_i2info(inode);
1273         ENTRY;
1274
1275         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1276                inode->i_generation, inode);
1277
1278         ll_lli_init(lli);
1279
1280         LASSERT(!lli->lli_smd);
1281
1282         /* Core attributes from the MDS first.  This is a new inode, and
1283          * the VFS doesn't zero times in the core inode so we have to do
1284          * it ourselves.  They will be overwritten by either MDS or OST
1285          * attributes - we just need to make sure they aren't newer. */
1286         LTIME_S(inode->i_mtime) = 0;
1287         LTIME_S(inode->i_atime) = 0;
1288         LTIME_S(inode->i_ctime) = 0;
1289         inode->i_rdev = 0;
1290         ll_update_inode(inode, md);
1291
1292         /* OIDEBUG(inode); */
1293
1294         if (S_ISREG(inode->i_mode)) {
1295                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1296                 inode->i_op = &ll_file_inode_operations;
1297                 inode->i_fop = sbi->ll_fop;
1298                 inode->i_mapping->a_ops = &ll_aops;
1299                 EXIT;
1300         } else if (S_ISDIR(inode->i_mode)) {
1301                 inode->i_op = &ll_dir_inode_operations;
1302                 inode->i_fop = &ll_dir_operations;
1303                 inode->i_mapping->a_ops = &ll_dir_aops;
1304                 EXIT;
1305         } else if (S_ISLNK(inode->i_mode)) {
1306                 inode->i_op = &ll_fast_symlink_inode_operations;
1307                 EXIT;
1308         } else {
1309                 inode->i_op = &ll_special_inode_operations;
1310
1311 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1312                 init_special_inode(inode, inode->i_mode,
1313                                    kdev_t_to_nr(inode->i_rdev));
1314
1315                 /* initializing backing dev info. */
1316                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1317 #else
1318                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1319 #endif
1320                 lli->ll_save_ifop = inode->i_fop;
1321
1322                 if (S_ISCHR(inode->i_mode))
1323                         inode->i_fop = &ll_special_chr_inode_fops;
1324                 else if (S_ISBLK(inode->i_mode))
1325                         inode->i_fop = &ll_special_blk_inode_fops;
1326                 else if (S_ISFIFO(inode->i_mode))
1327                         inode->i_fop = &ll_special_fifo_inode_fops;
1328                 else if (S_ISSOCK(inode->i_mode))
1329                         inode->i_fop = &ll_special_sock_inode_fops;
1330                 EXIT;
1331         }
1332 }
1333
1334 int ll_iocontrol(struct inode *inode, struct file *file,
1335                  unsigned int cmd, unsigned long arg)
1336 {
1337         struct ll_sb_info *sbi = ll_i2sbi(inode);
1338         struct ptlrpc_request *req = NULL;
1339         int rc, flags = 0;
1340         ENTRY;
1341
1342         switch(cmd) {
1343         case EXT3_IOC_GETFLAGS: {
1344                 struct lu_fid fid;
1345                 struct mdt_body *body;
1346
1347                 ll_inode2fid(&fid, inode);
1348                 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
1349                 if (rc) {
1350                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1351                         RETURN(-abs(rc));
1352                 }
1353
1354                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1355
1356                 if (body->flags & S_APPEND)
1357                         flags |= EXT3_APPEND_FL;
1358                 if (body->flags & S_IMMUTABLE)
1359                         flags |= EXT3_IMMUTABLE_FL;
1360                 if (body->flags & S_NOATIME)
1361                         flags |= EXT3_NOATIME_FL;
1362
1363                 ptlrpc_req_finished (req);
1364
1365                 RETURN(put_user(flags, (int *)arg));
1366         }
1367         case EXT3_IOC_SETFLAGS: {
1368                 struct mdc_op_data op_data;
1369                 struct iattr attr;
1370                 struct obdo *oa;
1371                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1372
1373                 if (get_user(flags, (int *)arg))
1374                         RETURN(-EFAULT);
1375
1376                 oa = obdo_alloc();
1377                 if (!oa)
1378                         RETURN(-ENOMEM);
1379
1380                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1381
1382                 memset(&attr, 0x0, sizeof(attr));
1383                 attr.ia_attr_flags = flags;
1384                 attr.ia_valid |= ATTR_ATTR_FLAG;
1385
1386                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1387                                  &attr, NULL, 0, NULL, 0, &req);
1388                 if (rc) {
1389                         ptlrpc_req_finished(req);
1390                         if (rc != -EPERM && rc != -EACCES)
1391                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
1392                         obdo_free(oa);
1393                         RETURN(rc);
1394                 }
1395                 ptlrpc_req_finished(req);
1396
1397                 oa->o_id = lsm->lsm_object_id;
1398                 oa->o_flags = flags;
1399                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1400
1401                 obdo_from_inode(oa, inode, OBD_MD_FLFID | OBD_MD_FLGENER);
1402                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1403                 obdo_free(oa);
1404                 if (rc) {
1405                         if (rc != -EPERM && rc != -EACCES)
1406                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
1407                         RETURN(rc);
1408                 }
1409
1410                 if (flags & EXT3_APPEND_FL)
1411                         inode->i_flags |= S_APPEND;
1412                 else
1413                         inode->i_flags &= ~S_APPEND;
1414                 if (flags & EXT3_IMMUTABLE_FL)
1415                         inode->i_flags |= S_IMMUTABLE;
1416                 else
1417                         inode->i_flags &= ~S_IMMUTABLE;
1418                 if (flags & EXT3_NOATIME_FL)
1419                         inode->i_flags |= S_NOATIME;
1420                 else
1421                         inode->i_flags &= ~S_NOATIME;
1422
1423                 RETURN(0);
1424         }
1425         default:
1426                 RETURN(-ENOSYS);
1427         }
1428
1429         RETURN(0);
1430 }
1431
1432 /* umount -f client means force down, don't save state */
1433 void ll_umount_begin(struct super_block *sb)
1434 {
1435         struct lustre_sb_info *lsi = s2lsi(sb);
1436         struct ll_sb_info *sbi = ll_s2sbi(sb);
1437         struct obd_device *obd;
1438         struct obd_ioctl_data ioc_data = { 0 };
1439         ENTRY;
1440
1441         /* Tell the MGC we got umount -f */
1442         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1443
1444         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1445                sb->s_count, atomic_read(&sb->s_active));
1446
1447         obd = class_exp2obd(sbi->ll_mdc_exp);
1448         if (obd == NULL) {
1449                 CERROR("Invalid MDC connection handle "LPX64"\n",
1450                        sbi->ll_mdc_exp->exp_handle.h_cookie);
1451                 EXIT;
1452                 return;
1453         }
1454         obd->obd_no_recov = 1;
1455         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1456                       &ioc_data, NULL);
1457
1458         obd = class_exp2obd(sbi->ll_osc_exp);
1459         if (obd == NULL) {
1460                 CERROR("Invalid LOV connection handle "LPX64"\n",
1461                        sbi->ll_osc_exp->exp_handle.h_cookie);
1462                 EXIT;
1463                 return;
1464         }
1465         obd->obd_no_recov = 1;
1466         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1467                       &ioc_data, NULL);
1468
1469         /* Really, we'd like to wait until there are no requests outstanding,
1470          * and then continue.  For now, we just invalidate the requests,
1471          * schedule, and hope.
1472          */
1473         schedule();
1474
1475         EXIT;
1476 }
1477
1478 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1479 {
1480         struct ll_sb_info *sbi = ll_s2sbi(sb);
1481         int err;
1482         __u32 read_only;
1483
1484         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1485                 read_only = *flags & MS_RDONLY;
1486                 err = obd_set_info(sbi->ll_mdc_exp, strlen("read-only"),
1487                                    "read-only", sizeof(read_only), &read_only);
1488                 if (err) {
1489                         CERROR("Failed to change the read-only flag during "
1490                                "remount: %d\n", err);
1491                         return err;
1492                 }
1493
1494                 if (read_only)
1495                         sb->s_flags |= MS_RDONLY;
1496                 else
1497                         sb->s_flags &= ~MS_RDONLY;
1498         }
1499         return 0;
1500 }
1501
1502 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1503                   struct ptlrpc_request *req, int offset,
1504                   struct super_block *sb)
1505 {
1506         struct ll_sb_info *sbi = NULL;
1507         struct lustre_md md;
1508         int rc = 0;
1509         ENTRY;
1510
1511         LASSERT(*inode || sb);
1512         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1513         prune_deathrow(sbi, 1);
1514
1515         rc = mdc_req2lustre_md(req, offset, exp, &md);
1516         if (rc)
1517                 RETURN(rc);
1518
1519         if (*inode) {
1520                 ll_update_inode(*inode, &md);
1521         } else {
1522                 struct lu_fid fid;
1523
1524                 LASSERT(sb != NULL);
1525
1526                 rc = ll_fid_alloc(sbi, &fid);
1527                 if (rc) {
1528                         CERROR("cannot allocate new fid, rc %d\n",
1529                                rc);
1530                         mdc_free_lustre_md(exp, &md);
1531                         GOTO(out, rc);
1532                 }
1533
1534                 *inode = ll_iget(sb, ll_fid2ino(sbi, &fid), &md);
1535                 if (*inode == NULL || is_bad_inode(*inode)) {
1536                         mdc_free_lustre_md(exp, &md);
1537                         rc = -ENOMEM;
1538                         CERROR("new_inode -fatal: rc %d\n", rc);
1539                         GOTO(out, rc);
1540                 }
1541                 ll_i2info(*inode)->lli_fid = fid;
1542         }
1543
1544         rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
1545                          ll_i2info(*inode)->lli_smd);
1546 out:
1547         RETURN(rc);
1548 }
1549
1550 char *llap_origins[] = {
1551         [LLAP_ORIGIN_UNKNOWN] = "--",
1552         [LLAP_ORIGIN_READPAGE] = "rp",
1553         [LLAP_ORIGIN_READAHEAD] = "ra",
1554         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
1555         [LLAP_ORIGIN_WRITEPAGE] = "wp",
1556 };
1557
1558 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
1559                                              struct list_head *list)
1560 {
1561         struct ll_async_page *llap;
1562         struct list_head *pos;
1563
1564         list_for_each(pos, list) {
1565                 if (pos == &sbi->ll_pglist)
1566                         return NULL;
1567                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
1568                 if (llap->llap_page == NULL)
1569                         continue;
1570                 return llap;
1571         }
1572         LBUG();
1573         return NULL;
1574 }
1575
1576 int ll_obd_statfs(struct inode *inode, void *arg)
1577 {
1578         struct ll_sb_info *sbi = NULL;
1579         struct obd_device *client_obd = NULL, *lov_obd = NULL;
1580         struct lov_obd *lov = NULL;
1581         struct obd_import *client_imp = NULL;
1582         struct obd_statfs stat_buf = {0};
1583         char *buf = NULL;
1584         struct obd_ioctl_data *data = NULL;
1585         __u32 type, index;
1586         int len, rc;
1587
1588         if (!inode || !(sbi = ll_i2sbi(inode)))
1589                 GOTO(out_statfs, rc = -EINVAL);
1590
1591         rc = obd_ioctl_getdata(&buf, &len, arg);
1592         if (rc)
1593                 GOTO(out_statfs, rc);
1594
1595         data = (void*)buf;
1596         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1597             !data->ioc_pbuf1 || !data->ioc_pbuf2)
1598                 GOTO(out_statfs, rc = -EINVAL);
1599
1600         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1601         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1602
1603         if (type == LL_STATFS_MDC) {
1604                 if (index > 0)
1605                         GOTO(out_statfs, rc = -ENODEV);
1606                 client_obd = class_exp2obd(sbi->ll_mdc_exp);
1607                 client_imp = class_exp2cliimp(sbi->ll_mdc_exp);
1608         } else if (type == LL_STATFS_LOV) {
1609                 lov_obd = class_exp2obd(sbi->ll_osc_exp);
1610                 lov = &lov_obd->u.lov;
1611
1612                 if (index >= lov->desc.ld_tgt_count)
1613                         GOTO(out_statfs, rc = -ENODEV);
1614
1615                 client_obd = class_exp2obd(lov->tgts[index].ltd_exp);
1616                 client_imp = class_exp2cliimp(lov->tgts[index].ltd_exp);
1617                 if (!lov->tgts[index].active)
1618                         GOTO(out_uuid, rc = -ENODATA);
1619         }
1620
1621         if (!client_obd || !client_imp)
1622                 GOTO(out_statfs, rc = -EINVAL);
1623
1624         rc = obd_statfs(client_obd, &stat_buf, jiffies - 1);
1625         if (rc)
1626                 GOTO(out_statfs, rc);
1627
1628         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1629                 GOTO(out_statfs, rc = -EFAULT);
1630
1631 out_uuid:
1632         if (copy_to_user(data->ioc_pbuf2, &client_imp->imp_target_uuid,
1633                          data->ioc_plen2))
1634                 rc = -EFAULT;
1635
1636 out_statfs:
1637         if (buf)
1638                 obd_ioctl_freedata(buf, len);
1639         return rc;
1640 }
1641
1642 EXPORT_SYMBOL(ll_fill_super);
1643 EXPORT_SYMBOL(ll_put_super);
1644 EXPORT_SYMBOL(ll_remount_fs);
1645 EXPORT_SYMBOL(ll_umount_begin);
1646