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_lite.h>
  32 #include <linux/lustre_ha.h>
  33 #include <linux/lustre_ver.h>
  34 #include <linux/lustre_dlm.h>
  35 #include <linux/lprocfs_status.h>
  36 #include <linux/lustre_disk.h>
  37 #include "llite_internal.h"
  38
  39 kmem_cache_t *ll_file_data_slab;
  40
  41 LIST_HEAD(ll_super_blocks);
  42 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  43
  44 extern struct address_space_operations ll_aops;
  45 extern struct address_space_operations ll_dir_aops;
  46
  47 #ifndef log2
  48 #define log2(n) ffz(~(n))
  49 #endif
  50
  51 /* We need to have some extra twiddling here because some systems have
  52  * no random state when they start up. */
  53 static void
  54 lustre_generate_random_uuid(class_uuid_t uuid)
  55 {
  56         struct timeval t;
  57         int *i, j, k;
  58
  59         ENTRY;
  60         LASSERT(sizeof(class_uuid_t) % sizeof(*i) == 0);
  61
  62         j = jiffies;
  63         do_gettimeofday(&t);
  64         k = t.tv_usec;
  65
  66         generate_random_uuid(uuid);
  67
  68         for (i = (int *)uuid; (char *)i < (char *)uuid + sizeof(class_uuid_t); i++) {
  69                 *i ^= j ^ k;
  70                 j = ((j << 8) & 0xffffff00) | ((j >> 24) & 0x000000ff);
  71                 k = ((k >> 8) & 0x00ffffff) | ((k << 24) & 0xff000000);
  72         }
  73
  74         EXIT;
  75 }
  76
  77 struct ll_sb_info *ll_init_sbi(void)
  78 {
  79         struct ll_sb_info *sbi = NULL;
  80         class_uuid_t uuid;
  81         ENTRY;
  82
  83         OBD_ALLOC(sbi, sizeof(*sbi));
  84         if (!sbi)
  85                 RETURN(NULL);
  86
  87         spin_lock_init(&sbi->ll_lock);
  88         spin_lock_init(&sbi->ll_lco.lco_lock);
  89         INIT_LIST_HEAD(&sbi->ll_pglist);
  90         sbi->ll_pglist_gen = 0;
  91         if (num_physpages >> (20 - PAGE_SHIFT) < 512)
  92                 sbi->ll_async_page_max = num_physpages / 2;
  93         else
  94                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  95         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  96                                            SBI_DEFAULT_READAHEAD_MAX);
  97
  98         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  99         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
 100
 101         lustre_generate_random_uuid(uuid);
 102         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 103         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 104
 105         spin_lock(&ll_sb_lock);
 106         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 107         spin_unlock(&ll_sb_lock);
 108
 109         INIT_LIST_HEAD(&sbi->ll_deathrow);
 110         spin_lock_init(&sbi->ll_deathrow_lock);
 111         RETURN(sbi);
 112 }
 113
 114 void ll_free_sbi(struct super_block *sb)
 115 {
 116         struct ll_sb_info *sbi = ll_s2sbi(sb);
 117         ENTRY;
 118
 119         if (sbi != NULL) {
 120                 spin_lock(&ll_sb_lock);
 121                 list_del(&sbi->ll_list);
 122                 spin_unlock(&ll_sb_lock);
 123                 OBD_FREE(sbi, sizeof(*sbi));
 124         }
 125         EXIT;
 126 }
 127
 128 static struct dentry_operations ll_d_root_ops = {
 129         .d_compare = ll_dcompare,
 130 };
 131
 132 int client_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 133 {
 134         struct inode *root = 0;
 135         struct ll_sb_info *sbi = ll_s2sbi(sb);
 136         struct obd_device *obd;
 137         struct ll_fid rootfid;
 138         struct obd_statfs osfs;
 139         struct ptlrpc_request *request = NULL;
 140         struct lustre_handle osc_conn = {0, };
 141         struct lustre_handle mdc_conn = {0, };
 142         struct lustre_md md;
 143         struct obd_connect_data *data = NULL;
 144         int err;
 145         ENTRY;
 146
 147         obd = class_name2obd(mdc);
 148         if (!obd) {
 149                 CERROR("MDC %s: not setup or attached\n", mdc);
 150                 RETURN(-EINVAL);
 151         }
 152
 153         OBD_ALLOC(data, sizeof(*data));
 154         if (data == NULL)
 155                 RETURN(-ENOMEM);
 156
 157         if (proc_lustre_fs_root) {
 158                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 159                                                   osc, mdc);
 160                 if (err < 0)
 161                         CERROR("could not register mount in /proc/lustre");
 162         }
 163
 164         /* indicate that inodebits locking is supported by this client */
 165         data->ocd_connect_flags |= OBD_CONNECT_IBITS;
 166         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 167
 168         if (sb->s_flags & MS_RDONLY)
 169                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 170         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 171                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 172         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_JOIN;
 173
 174         if (sbi->ll_flags & LL_SBI_FLOCK) {
 175                 sbi->ll_fop = &ll_file_operations_flock;
 176         } else {
 177                 sbi->ll_fop = &ll_file_operations;
 178         }
 179
 180         data->ocd_connect_flags |= OBD_CONNECT_VERSION;
 181         data->ocd_version = LUSTRE_VERSION_CODE;
 182
 183         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data);
 184         if (err == -EBUSY) {
 185                 CERROR("An MDT (mdc %s) is performing recovery, of which this"
 186                        " client is not a part.  Please wait for recovery to "
 187                        "complete, abort, or time out.\n", mdc);
 188                 GOTO(out, err);
 189         } else if (err) {
 190                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 191                 GOTO(out, err);
 192         }
 193         sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
 194
 195         err = obd_statfs(obd, &osfs, jiffies - HZ);
 196         if (err)
 197                 GOTO(out_mdc, err);
 198
 199         /* async connect is surely finished by now */
 200         *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
 201
 202         LASSERT(osfs.os_bsize);
 203         sb->s_blocksize = osfs.os_bsize;
 204         sb->s_blocksize_bits = log2(osfs.os_bsize);
 205         sb->s_magic = LL_SUPER_MAGIC;
 206         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 207         sbi->ll_namelen = osfs.os_namelen;
 208
 209         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 210             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 211                 LCONSOLE_INFO("Disabling user_xattr feature because "
 212                               "it is not supported on the server\n");
 213                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 214         }
 215
 216         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 217 #ifdef MS_POSIXACL
 218                 sb->s_flags |= MS_POSIXACL;
 219 #endif
 220                 sbi->ll_flags |= LL_SBI_ACL;
 221         } else
 222                 sbi->ll_flags &= ~LL_SBI_ACL;
 223
 224         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 225                 sbi->ll_flags |= LL_SBI_JOIN;
 226
 227 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 228         /* We set sb->s_dev equal on all lustre clients in order to support
 229          * NFS export clustering.  NFSD requires that the FSID be the same
 230          * on all clients. */
 231         /* s_dev is also used in lt_compare() to compare two fs, but that is
 232          * only a node-local comparison. */
 233         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
 234                          strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
 235 #endif
 236
 237         obd = class_name2obd(osc);
 238         if (!obd) {
 239                 CERROR("OSC %s: not setup or attached\n", osc);
 240                 GOTO(out_mdc, err);
 241         }
 242
 243         data->ocd_connect_flags =
 244                 OBD_CONNECT_GRANT|OBD_CONNECT_VERSION|OBD_CONNECT_REQPORTAL;
 245
 246         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 247                "ocd_grant: %d\n", data->ocd_connect_flags,
 248                data->ocd_version, data->ocd_grant);
 249
 250         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 251         obd->obd_upcall.onu_upcall = ll_ocd_update;
 252
 253         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data);
 254         if (err == -EBUSY) {
 255                 CERROR("An OST (osc %s) is performing recovery, of which this"
 256                        " client is not a part.  Please wait for recovery to "
 257                        "complete, abort, or time out.\n", osc);
 258                 GOTO(out, err);
 259         } else if (err) {
 260                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 261                 GOTO(out_mdc, err);
 262         }
 263         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 264         spin_lock(&sbi->ll_lco.lco_lock);
 265         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 266         spin_unlock(&sbi->ll_lco.lco_lock);
 267
 268         mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
 269
 270         if (!ll_async_page_slab) {
 271                 ll_async_page_slab_size =
 272                         size_round(sizeof(struct ll_async_page)) +
 273                         obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
 274                                             0, NULL, NULL, NULL);
 275                 ll_async_page_slab = kmem_cache_create("ll_async_page",
 276                                                        ll_async_page_slab_size,
 277                                                        0, 0, NULL, NULL);
 278                 if (!ll_async_page_slab)
 279                         GOTO(out_osc, -ENOMEM);
 280         }
 281
 282         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
 283         if (err) {
 284                 CERROR("cannot mds_connect: rc = %d\n", err);
 285                 GOTO(out_osc, err);
 286         }
 287         CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
 288         sbi->ll_rootino = rootfid.id;
 289
 290         sb->s_op = &lustre_super_operations;
 291
 292         /* make root inode
 293          * XXX: move this to after cbd setup? */
 294         err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
 295                           OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
 296                           (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
 297                           0, &request);
 298         if (err) {
 299                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 300                 GOTO(out_osc, err);
 301         }
 302
 303         err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
 304         if (err) {
 305                 CERROR("failed to understand root inode md: rc = %d\n",err);
 306                 ptlrpc_req_finished (request);
 307                 GOTO(out_osc, err);
 308         }
 309
 310         LASSERT(sbi->ll_rootino != 0);
 311         root = ll_iget(sb, sbi->ll_rootino, &md);
 312
 313         ptlrpc_req_finished(request);
 314
 315         if (root == NULL || is_bad_inode(root)) {
 316                 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
 317                 CERROR("lustre_lite: bad iget4 for root\n");
 318                 GOTO(out_root, err = -EBADF);
 319         }
 320
 321         err = ll_close_thread_start(&sbi->ll_lcq);
 322         if (err) {
 323                 CERROR("cannot start close thread: rc %d\n", err);
 324                 GOTO(out_root, err);
 325         }
 326
 327         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 328            backing dev info assigned to inode mapping is used for
 329            determining maximal readahead. */
 330 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 331     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 332         /* bug 2805 - set VM readahead to zero */
 333         vm_max_readahead = vm_min_readahead = 0;
 334 #endif
 335
 336         sb->s_root = d_alloc_root(root);
 337         if (data != NULL)
 338                 OBD_FREE(data, sizeof(*data));
 339         sb->s_root->d_op = &ll_d_root_ops;
 340         RETURN(err);
 341
 342 out_root:
 343         if (root)
 344                 iput(root);
 345 out_osc:
 346         obd_disconnect(sbi->ll_osc_exp);
 347 out_mdc:
 348         obd_disconnect(sbi->ll_mdc_exp);
 349 out:
 350         if (data != NULL)
 351                 OBD_FREE(data, sizeof(*data));
 352         lprocfs_unregister_mountpoint(sbi);
 353         RETURN(err);
 354 }
 355
 356 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 357 {
 358         int size, rc;
 359
 360         *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
 361         size = sizeof(int);
 362         rc = obd_get_info(sbi->ll_mdc_exp, strlen("max_easize"), "max_easize",
 363                           &size, lmmsize);
 364         if (rc)
 365                 CERROR("Get max mdsize error rc %d \n", rc);
 366
 367         RETURN(rc);
 368 }
 369
 370 void ll_dump_inode(struct inode *inode)
 371 {
 372         struct list_head *tmp;
 373         int dentry_count = 0;
 374
 375         LASSERT(inode != NULL);
 376
 377         list_for_each(tmp, &inode->i_dentry)
 378                 dentry_count++;
 379
 380         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 381                inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
 382                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 383 }
 384
 385 void lustre_dump_dentry(struct dentry *dentry, int recur)
 386 {
 387         struct list_head *tmp;
 388         int subdirs = 0;
 389
 390         LASSERT(dentry != NULL);
 391
 392         list_for_each(tmp, &dentry->d_subdirs)
 393                 subdirs++;
 394
 395         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 396                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 397                dentry->d_name.len, dentry->d_name.name,
 398                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 399                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 400                dentry->d_flags, dentry->d_fsdata, subdirs);
 401         if (dentry->d_inode != NULL)
 402                 ll_dump_inode(dentry->d_inode);
 403
 404         if (recur == 0)
 405                 return;
 406
 407         list_for_each(tmp, &dentry->d_subdirs) {
 408                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 409                 lustre_dump_dentry(d, recur - 1);
 410         }
 411 }
 412
 413 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 414 void lustre_throw_orphan_dentries(struct super_block *sb)
 415 {
 416         struct hlist_node *tmp, *next;
 417         struct ll_sb_info *sbi = ll_s2sbi(sb);
 418
 419         /* Do this to get rid of orphaned dentries. That is not really trw. */
 420         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 421                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 422                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 423                       "before and after shrink_dcache_parent\n",
 424                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 425                 lustre_dump_dentry(dentry, 1);
 426                 shrink_dcache_parent(dentry);
 427                 lustre_dump_dentry(dentry, 1);
 428         }
 429 }
 430 #else
 431 #define lustre_throw_orphan_dentries(sb)
 432 #endif
 433
 434 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 435 {
 436         LIST_HEAD(throw_away);
 437         int locked = 0;
 438         ENTRY;
 439
 440         if (try) {
 441                 locked = spin_trylock(&sbi->ll_deathrow_lock);
 442         } else {
 443                 spin_lock(&sbi->ll_deathrow_lock);
 444                 locked = 1;
 445         }
 446
 447         if (!locked) {
 448                 EXIT;
 449                 return;
 450         }
 451
 452         list_splice_init(&sbi->ll_deathrow, &throw_away);
 453         spin_unlock(&sbi->ll_deathrow_lock);
 454
 455         while (!list_empty(&throw_away)) {
 456                 struct ll_inode_info *lli;
 457                 struct inode *inode;
 458
 459                 lli = list_entry(throw_away.next, struct ll_inode_info,
 460                                  lli_dead_list);
 461                 list_del_init(&lli->lli_dead_list);
 462
 463                 inode = ll_info2i(lli);
 464                 d_prune_aliases(inode);
 465
 466                 CDEBUG(D_INODE, "prune duplicate inode %p inum %lu count %u\n",
 467                        inode, inode->i_ino, atomic_read(&inode->i_count));
 468                 iput(inode);
 469         }
 470         EXIT;
 471 }
 472
 473 void client_common_put_super(struct super_block *sb)
 474 {
 475         struct ll_sb_info *sbi = ll_s2sbi(sb);
 476         ENTRY;
 477
 478         ll_close_thread_shutdown(sbi->ll_lcq);
 479
 480         /* destroy inodes in deathrow */
 481         prune_deathrow(sbi, 0);
 482
 483         list_del(&sbi->ll_conn_chain);
 484         obd_disconnect(sbi->ll_osc_exp);
 485
 486         lprocfs_unregister_mountpoint(sbi);
 487         if (sbi->ll_proc_root) {
 488                 lprocfs_remove(sbi->ll_proc_root);
 489                 sbi->ll_proc_root = NULL;
 490         }
 491
 492         obd_disconnect(sbi->ll_mdc_exp);
 493
 494         lustre_throw_orphan_dentries(sb);
 495
 496         EXIT;
 497 }
 498
 499 char *ll_read_opt(const char *opt, char *data)
 500 {
 501         char *value;
 502         char *retval;
 503         ENTRY;
 504
 505         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 506         if (strncmp(opt, data, strlen(opt)))
 507                 RETURN(NULL);
 508         if ((value = strchr(data, '=')) == NULL)
 509                 RETURN(NULL);
 510
 511         value++;
 512         OBD_ALLOC(retval, strlen(value) + 1);
 513         if (!retval) {
 514                 CERROR("out of memory!\n");
 515                 RETURN(NULL);
 516         }
 517
 518         memcpy(retval, value, strlen(value)+1);
 519         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 520         RETURN(retval);
 521 }
 522
 523 int ll_set_opt(const char *opt, char *data, int fl)
 524 {
 525         ENTRY;
 526
 527         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 528         if (strncmp(opt, data, strlen(opt)) != 0)
 529                 RETURN(0);
 530         else
 531                 RETURN(fl);
 532 }
 533
 534 /* non-client-specific mount options are parsed in lmd_parse */
 535 void ll_options(char *options, int *flags)
 536 {
 537         int tmp;
 538         char *this_char;
 539 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 540         char *opt_ptr = options;
 541 #endif
 542         ENTRY;
 543
 544         if (!options) {
 545                 EXIT;
 546                 return;
 547         }
 548
 549         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 550 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 551         for (this_char = strtok (options, ",");
 552              this_char != NULL;
 553              this_char = strtok (NULL, ","))
 554 #else
 555         while ((this_char = strsep (&opt_ptr, ",")) != NULL)
 556 #endif
 557         {
 558                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 559                 tmp = ll_set_opt("nolock", this_char, LL_SBI_NOLCK);
 560                 if (tmp) {
 561                         *flags |= tmp;
 562                         continue;
 563                 }
 564                 tmp = ll_set_opt("flock", this_char, LL_SBI_FLOCK);
 565                 if (tmp) {
 566                         *flags |= tmp;
 567                         continue;
 568                 }
 569                 tmp = ll_set_opt("noflock", this_char, LL_SBI_FLOCK);
 570                 if (tmp) {
 571                         *flags &= ~tmp;
 572                         continue;
 573                 }
 574                 tmp = ll_set_opt("user_xattr", this_char, LL_SBI_USER_XATTR);
 575                 if (tmp) {
 576                         *flags |= tmp;
 577                         continue;
 578                 }
 579                 tmp = ll_set_opt("nouser_xattr", this_char, LL_SBI_USER_XATTR);
 580                 if (tmp) {
 581                         *flags &= ~tmp;
 582                         continue;
 583                 }
 584                 tmp = ll_set_opt("acl", this_char, LL_SBI_ACL);
 585                 if (tmp) {
 586                         /* Ignore deprecated mount option.  The client will
 587                          * always try to mount with ACL support, whether this
 588                          * is used depends on whether server supports it. */
 589                         continue;
 590                 }
 591                 tmp = ll_set_opt("noacl", this_char, LL_SBI_ACL);
 592                 if (tmp) {
 593                         continue;
 594                 }
 595         }
 596         CERROR("flags %#x\n", *flags);
 597
 598         EXIT;
 599 }
 600
 601 void ll_lli_init(struct ll_inode_info *lli)
 602 {
 603         sema_init(&lli->lli_open_sem, 1);
 604         sema_init(&lli->lli_size_sem, 1);
 605         lli->lli_flags = 0;
 606         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 607         spin_lock_init(&lli->lli_lock);
 608         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 609         lli->lli_inode_magic = LLI_INODE_MAGIC;
 610         INIT_LIST_HEAD(&lli->lli_dead_list);
 611 }
 612
 613 int ll_fill_super(struct super_block *sb)
 614 {
 615         struct lustre_profile *lprof;
 616         struct lustre_sb_info *lsi = s2lsi(sb);
 617         struct ll_sb_info *sbi;
 618         char  *osc = NULL;
 619         char  *mdc = NULL;
 620         char  *profilenm = get_profile_name(sb);
 621         struct config_llog_instance cfg;
 622         char   ll_instance[sizeof(sb) * 2 + 1];
 623         int    err;
 624         ENTRY;
 625
 626         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 627
 628         /* client additional sb info */
 629         lsi->lsi_llsbi = sbi = ll_init_sbi();
 630         if (!sbi)
 631                 RETURN(-ENOMEM);
 632
 633         ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 634
 635         /* Generate a string unique to this super, in case some joker tries
 636            to mount the same fs at two mount points.
 637            Use the address of the super itself.*/
 638         sprintf(ll_instance, "%p", sb);
 639         cfg.cfg_instance = ll_instance;
 640         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 641         cfg.cfg_last_idx = 0;
 642
 643         /* set up client obds */
 644         err = lustre_process_log(sb, profilenm, &cfg);
 645         if (err < 0) {
 646                 CERROR("Unable to process log: %d\n", err);
 647                 GOTO(out_free, err);
 648         }
 649
 650         lprof = class_get_profile(profilenm);
 651         if (lprof == NULL) {
 652                 CERROR("No profile found: %s\n", profilenm);
 653                 GOTO(out_free, err = -EINVAL);
 654         }
 655         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 656                lprof->lp_mdc, lprof->lp_osc);
 657
 658         OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 659                   strlen(ll_instance) + 2);
 660         if (!osc)
 661                 GOTO(out_free, err = -ENOMEM);
 662         sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
 663
 664         OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 665                   strlen(ll_instance) + 2);
 666         if (!mdc)
 667                 GOTO(out_free, err = -ENOMEM);
 668         sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
 669
 670         /* connections, registrations, sb setup */
 671         err = client_common_fill_super(sb, mdc, osc);
 672
 673 out_free:
 674         if (mdc)
 675                 OBD_FREE(mdc, strlen(mdc) + 1);
 676         if (osc)
 677                 OBD_FREE(osc, strlen(osc) + 1);
 678         if (err) {
 679                 struct obd_device *obd;
 680                 int next = 0;
 681                 /* like client_put_super below */
 682                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 683                        != NULL) {
 684                         class_manual_cleanup(obd);
 685                 }
 686                 class_del_profile(profilenm);
 687                 ll_free_sbi(sb);
 688                 lsi->lsi_llsbi = NULL;
 689         }
 690         RETURN(err);
 691 } /* ll_fill_super */
 692
 693
 694 void ll_put_super(struct super_block *sb)
 695 {
 696         struct config_llog_instance cfg;
 697         char   ll_instance[sizeof(sb) * 2 + 1];
 698         struct obd_device *obd;
 699         struct lustre_sb_info *lsi = s2lsi(sb);
 700         struct ll_sb_info *sbi = ll_s2sbi(sb);
 701         char *profilenm = get_profile_name(sb);
 702         int next = 0;
 703         ENTRY;
 704
 705         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 706
 707         sprintf(ll_instance, "%p", sb);
 708         cfg.cfg_instance = ll_instance;
 709         lustre_end_log(sb, NULL, &cfg);
 710
 711         obd = class_exp2obd(sbi->ll_mdc_exp);
 712         if (obd) {
 713                 int next = 0;
 714                 int force = obd->obd_no_recov;
 715                 /* We need to set force before the lov_disconnect in
 716                 lustre_common_put_super, since l_d cleans up osc's as well. */
 717                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next))
 718                        != NULL) {
 719                         obd->obd_force = force;
 720                 }
 721         }
 722
 723         client_common_put_super(sb);
 724
 725         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 726                 class_manual_cleanup(obd);
 727         }
 728
 729         if (profilenm)
 730                 class_del_profile(profilenm);
 731
 732         ll_free_sbi(sb);
 733         lsi->lsi_llsbi = NULL;
 734
 735         lustre_common_put_super(sb);
 736
 737         EXIT;
 738 } /* client_put_super */
 739
 740 #ifdef HAVE_REGISTER_CACHE
 741 #include <linux/cache_def.h>
 742 #ifdef HAVE_CACHE_RETURN_INT
 743 static int
 744 #else
 745 static void
 746 #endif
 747 ll_shrink_cache(int priority, unsigned int gfp_mask)
 748 {
 749         struct ll_sb_info *sbi;
 750         int count = 0;
 751
 752         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
 753                 count += llap_shrink_cache(sbi, priority);
 754
 755 #ifdef HAVE_CACHE_RETURN_INT
 756         return count;
 757 #endif
 758 }
 759
 760 struct cache_definition ll_cache_definition = {
 761         .name = "llap_cache",
 762         .shrink = ll_shrink_cache
 763 };
 764 #endif /* HAVE_REGISTER_CACHE */
 765
 766 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 767 {
 768         struct inode *inode = NULL;
 769         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 770         if (lock->l_ast_data) {
 771                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 772                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 773                         inode = igrab(lock->l_ast_data);
 774                 } else {
 775                         inode = lock->l_ast_data;
 776                         __LDLM_DEBUG(inode->i_state & I_FREEING ?
 777                                      D_INFO : D_WARNING, lock,
 778                                      "l_ast_data %p is bogus: magic %08x",
 779                                      lock->l_ast_data, lli->lli_inode_magic);
 780                         inode = NULL;
 781                 }
 782         }
 783         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 784         return inode;
 785 }
 786
 787 static int null_if_equal(struct ldlm_lock *lock, void *data)
 788 {
 789         if (data == lock->l_ast_data) {
 790                 lock->l_ast_data = NULL;
 791
 792                 if (lock->l_req_mode != lock->l_granted_mode)
 793                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
 794         }
 795
 796         return LDLM_ITER_CONTINUE;
 797 }
 798
 799 void ll_clear_inode(struct inode *inode)
 800 {
 801         struct ll_fid fid;
 802         struct ll_inode_info *lli = ll_i2info(inode);
 803         struct ll_sb_info *sbi = ll_i2sbi(inode);
 804         ENTRY;
 805
 806         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 807                inode->i_generation, inode);
 808
 809         ll_inode2fid(&fid, inode);
 810         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 811         mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 812
 813         if (lli->lli_smd) {
 814                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
 815                                   null_if_equal, inode);
 816
 817                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
 818                 lli->lli_smd = NULL;
 819         }
 820
 821         if (lli->lli_symlink_name) {
 822                 OBD_FREE(lli->lli_symlink_name,
 823                          strlen(lli->lli_symlink_name) + 1);
 824                 lli->lli_symlink_name = NULL;
 825         }
 826
 827 #ifdef CONFIG_FS_POSIX_ACL
 828         if (lli->lli_posix_acl) {
 829                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
 830                 posix_acl_release(lli->lli_posix_acl);
 831                 lli->lli_posix_acl = NULL;
 832         }
 833 #endif
 834
 835         lli->lli_inode_magic = LLI_INODE_DEAD;
 836
 837         spin_lock(&sbi->ll_deathrow_lock);
 838         list_del_init(&lli->lli_dead_list);
 839         spin_unlock(&sbi->ll_deathrow_lock);
 840
 841         EXIT;
 842 }
 843
 844 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 845  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 846  * keep these values until such a time that objects are allocated for it.
 847  * We do the MDS operations first, as it is checking permissions for us.
 848  * We don't to the MDS RPC if there is nothing that we want to store there,
 849  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 850  * going to do an RPC anyways.
 851  *
 852  * If we are doing a truncate, we will send the mtime and ctime updates
 853  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 854  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 855  * at the same time.
 856  */
 857 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 858 {
 859         struct ll_inode_info *lli = ll_i2info(inode);
 860         struct lov_stripe_md *lsm = lli->lli_smd;
 861         struct ll_sb_info *sbi = ll_i2sbi(inode);
 862         struct ptlrpc_request *request = NULL;
 863         struct mdc_op_data op_data;
 864         int ia_valid = attr->ia_valid;
 865         int rc = 0;
 866         ENTRY;
 867
 868         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
 869                attr->ia_valid);
 870         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
 871
 872         if (ia_valid & ATTR_SIZE) {
 873                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 874                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 875                                attr->ia_size, ll_file_maxbytes(inode));
 876                         RETURN(-EFBIG);
 877                 }
 878
 879                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 880         }
 881
 882         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
 883         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
 884                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 885                         RETURN(-EPERM);
 886         }
 887
 888         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 889         if (attr->ia_valid & ATTR_CTIME) {
 890                 attr->ia_ctime = CURRENT_TIME;
 891                 attr->ia_valid |= ATTR_CTIME_SET;
 892         }
 893         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 894                 attr->ia_atime = CURRENT_TIME;
 895                 attr->ia_valid |= ATTR_ATIME_SET;
 896         }
 897         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 898                 attr->ia_mtime = CURRENT_TIME;
 899                 attr->ia_valid |= ATTR_MTIME_SET;
 900         }
 901
 902         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 903                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
 904                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 905                        CURRENT_SECONDS);
 906
 907
 908         /* NB: ATTR_SIZE will only be set after this point if the size
 909          * resides on the MDS, ie, this file has no objects. */
 910         if (lsm)
 911                 attr->ia_valid &= ~ATTR_SIZE;
 912
 913         /* If only OST attributes being set on objects, don't do MDS RPC.
 914          * In that case, we need to check permissions and update the local
 915          * inode ourselves so we can call obdo_from_inode() always. */
 916         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 917                 struct lustre_md md;
 918                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 919
 920                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
 921                                  attr, NULL, 0, NULL, 0, &request);
 922
 923                 if (rc) {
 924                         ptlrpc_req_finished(request);
 925                         if (rc != -EPERM && rc != -EACCES)
 926                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
 927                         RETURN(rc);
 928                 }
 929
 930                 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
 931                 if (rc) {
 932                         ptlrpc_req_finished(request);
 933                         RETURN(rc);
 934                 }
 935
 936                 /* We call inode_setattr to adjust timestamps.
 937                  * If there is at least some data in file, we cleared ATTR_SIZE
 938                  * above to avoid invoking vmtruncate, otherwise it is important
 939                  * to call vmtruncate in inode_setattr to update inode->i_size
 940                  * (bug 6196) */
 941                 rc = inode_setattr(inode, attr);
 942
 943                 ll_update_inode(inode, &md);
 944                 ptlrpc_req_finished(request);
 945
 946                 if (!lsm || !S_ISREG(inode->i_mode)) {
 947                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 948                         RETURN(rc);
 949                 }
 950         } else {
 951                 /* The OST doesn't check permissions, but the alternative is
 952                  * a gratuitous RPC to the MDS.  We already rely on the client
 953                  * to do read/write/truncate permission checks, so is mtime OK?
 954                  */
 955                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 956                         /* from sys_utime() */
 957                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 958                                 if (current->fsuid != inode->i_uid &&
 959                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
 960                                         RETURN(rc);
 961                         } else {
 962                                 /* from inode_change_ok() */
 963                                 if (current->fsuid != inode->i_uid &&
 964                                     !capable(CAP_FOWNER))
 965                                         RETURN(-EPERM);
 966                         }
 967                 }
 968
 969                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
 970                 rc = inode_setattr(inode, attr);
 971         }
 972
 973         /* We really need to get our PW lock before we change inode->i_size.
 974          * If we don't we can race with other i_size updaters on our node, like
 975          * ll_file_read.  We can also race with i_size propogation to other
 976          * nodes through dirtying and writeback of final cached pages.  This
 977          * last one is especially bad for racing o_append users on other
 978          * nodes. */
 979         if (ia_valid & ATTR_SIZE) {
 980                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
 981                                                            OBD_OBJECT_EOF } };
 982                 struct lustre_handle lockh = { 0 };
 983                 int err, ast_flags = 0;
 984                 /* XXX when we fix the AST intents to pass the discard-range
 985                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
 986                  * XXX here. */
 987                 if (attr->ia_size == 0)
 988                         ast_flags = LDLM_AST_DISCARD_DATA;
 989
 990                 up(&inode->i_sem);
 991                 UP_WRITE_I_ALLOC_SEM(inode);
 992                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
 993                                     ast_flags);
 994 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 995                 DOWN_WRITE_I_ALLOC_SEM(inode);
 996                 down(&inode->i_sem);
 997 #else
 998                 down(&inode->i_sem);
 999                 DOWN_WRITE_I_ALLOC_SEM(inode);
1000 #endif
1001                 if (rc != 0)
1002                         RETURN(rc);
1003
1004                 /* Only ll_inode_size_lock is taken at this level.
1005                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1006                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1007                  * ll_truncate dropped ll_inode_size_lock() */
1008                 ll_inode_size_lock(inode, 0);
1009                 rc = vmtruncate(inode, attr->ia_size);
1010                 if (rc != 0) {
1011                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1012                         ll_inode_size_unlock(inode, 0);
1013                 }
1014
1015                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1016                 if (err) {
1017                         CERROR("ll_extent_unlock failed: %d\n", err);
1018                         if (!rc)
1019                                 rc = err;
1020                 }
1021         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1022                 obd_flag flags;
1023                 struct obdo oa;
1024
1025                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1026                        inode->i_ino, LTIME_S(attr->ia_mtime));
1027
1028                 oa.o_id = lsm->lsm_object_id;
1029                 oa.o_valid = OBD_MD_FLID;
1030
1031                 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1032                         OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1033                         OBD_MD_FLFID | OBD_MD_FLGENER;
1034
1035                 obdo_from_inode(&oa, inode, flags);
1036                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1037                 if (rc)
1038                         CERROR("obd_setattr fails: rc=%d\n", rc);
1039         }
1040         RETURN(rc);
1041 }
1042
1043 int ll_setattr(struct dentry *de, struct iattr *attr)
1044 {
1045         LBUG(); /* code is unused, but leave this in case of VFS changes */
1046         RETURN(-ENOSYS);
1047 }
1048
1049 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1050                        unsigned long max_age)
1051 {
1052         struct ll_sb_info *sbi = ll_s2sbi(sb);
1053         struct obd_statfs obd_osfs;
1054         int rc;
1055         ENTRY;
1056
1057         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1058         if (rc) {
1059                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1060                 RETURN(rc);
1061         }
1062
1063         osfs->os_type = sb->s_magic;
1064
1065         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1066                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1067
1068         rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1069         if (rc) {
1070                 CERROR("obd_statfs fails: rc = %d\n", rc);
1071                 RETURN(rc);
1072         }
1073
1074         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1075                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1076                obd_osfs.os_files);
1077
1078         osfs->os_blocks = obd_osfs.os_blocks;
1079         osfs->os_bfree = obd_osfs.os_bfree;
1080         osfs->os_bavail = obd_osfs.os_bavail;
1081
1082         /* If we don't have as many objects free on the OST as inodes
1083          * on the MDS, we reduce the total number of inodes to
1084          * compensate, so that the "inodes in use" number is correct.
1085          */
1086         if (obd_osfs.os_ffree < osfs->os_ffree) {
1087                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1088                         obd_osfs.os_ffree;
1089                 osfs->os_ffree = obd_osfs.os_ffree;
1090         }
1091
1092         RETURN(rc);
1093 }
1094
1095 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1096 {
1097         struct obd_statfs osfs;
1098         int rc;
1099
1100         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1101         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1102
1103         /* For now we will always get up-to-date statfs values, but in the
1104          * future we may allow some amount of caching on the client (e.g.
1105          * from QOS or lprocfs updates). */
1106         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1107         if (rc)
1108                 return rc;
1109
1110         statfs_unpack(sfs, &osfs);
1111
1112         if (sizeof(sfs->f_blocks) == 4) {
1113                 while (osfs.os_blocks > ~0UL) {
1114                         sfs->f_bsize <<= 1;
1115
1116                         osfs.os_blocks >>= 1;
1117                         osfs.os_bfree >>= 1;
1118                         osfs.os_bavail >>= 1;
1119                 }
1120         }
1121
1122         sfs->f_blocks = osfs.os_blocks;
1123         sfs->f_bfree = osfs.os_bfree;
1124         sfs->f_bavail = osfs.os_bavail;
1125
1126         return 0;
1127 }
1128
1129 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1130 {
1131         struct ll_inode_info *lli;
1132         struct lov_stripe_md *lsm;
1133
1134         lli = ll_i2info(inode);
1135         LASSERT(lli->lli_size_sem_owner != current);
1136         down(&lli->lli_size_sem);
1137         LASSERT(lli->lli_size_sem_owner == NULL);
1138         lli->lli_size_sem_owner = current;
1139         lsm = lli->lli_smd;
1140         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1141                  lsm, lock_lsm);
1142         if (lock_lsm)
1143                 lov_stripe_lock(lsm);
1144 }
1145
1146 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1147 {
1148         struct ll_inode_info *lli;
1149         struct lov_stripe_md *lsm;
1150
1151         lli = ll_i2info(inode);
1152         lsm = lli->lli_smd;
1153         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1154                  lsm, unlock_lsm);
1155         if (unlock_lsm)
1156                 lov_stripe_unlock(lsm);
1157         LASSERT(lli->lli_size_sem_owner == current);
1158         lli->lli_size_sem_owner = NULL;
1159         up(&lli->lli_size_sem);
1160 }
1161
1162 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1163 {
1164         struct ll_inode_info *lli = ll_i2info(inode);
1165
1166         dump_lsm(D_INODE, lsm);
1167         dump_lsm(D_INODE, lli->lli_smd);
1168         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1169                  "lsm must be joined lsm %p\n", lsm);
1170         obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1171         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1172                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1173         lli->lli_smd = lsm;
1174         lli->lli_maxbytes = lsm->lsm_maxbytes;
1175         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1176                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1177 }
1178
1179 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1180 {
1181         struct ll_inode_info *lli = ll_i2info(inode);
1182         struct mds_body *body = md->body;
1183         struct lov_stripe_md *lsm = md->lsm;
1184
1185         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1186         if (lsm != NULL) {
1187                 if (lli->lli_smd == NULL) {
1188                         if (lsm->lsm_magic != LOV_MAGIC &&
1189                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1190                                 dump_lsm(D_ERROR, lsm);
1191                                 LBUG();
1192                         }
1193                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1194                                lsm, inode->i_ino, inode->i_generation, inode);
1195                         /* ll_inode_size_lock() requires it is only called
1196                          * with lli_smd != NULL or lock_lsm == 0 or we can
1197                          * race between lock/unlock.  bug 9547 */
1198                         lli->lli_smd = lsm;
1199                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1200                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1201                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1202                 } else {
1203                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1204                              lli->lli_smd->lsm_stripe_count ==
1205                                         lsm->lsm_stripe_count) {
1206                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1207                                         CERROR("lsm mismatch for inode %ld\n",
1208                                                 inode->i_ino);
1209                                         CERROR("lli_smd:\n");
1210                                         dump_lsm(D_ERROR, lli->lli_smd);
1211                                         CERROR("lsm:\n");
1212                                         dump_lsm(D_ERROR, lsm);
1213                                         LBUG();
1214                                 }
1215                         } else
1216                                 ll_replace_lsm(inode, lsm);
1217                 }
1218                 /* bug 2844 - limit i_blksize for broken user-space apps */
1219                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1220                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1221                 if (lli->lli_smd != lsm)
1222                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1223         } else {
1224                 inode->i_blksize = max(inode->i_blksize,
1225                                        inode->i_sb->s_blocksize);
1226         }
1227
1228 #ifdef CONFIG_FS_POSIX_ACL
1229         LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1230         if (body->valid & OBD_MD_FLACL) {
1231                 spin_lock(&lli->lli_lock);
1232                 if (lli->lli_posix_acl)
1233                         posix_acl_release(lli->lli_posix_acl);
1234                 lli->lli_posix_acl = md->posix_acl;
1235                 spin_unlock(&lli->lli_lock);
1236         }
1237 #endif
1238
1239         if (body->valid & OBD_MD_FLID)
1240                 inode->i_ino = body->ino;
1241         if (body->valid & OBD_MD_FLATIME)
1242                 LTIME_S(inode->i_atime) = body->atime;
1243         if (body->valid & OBD_MD_FLMTIME &&
1244             body->mtime > LTIME_S(inode->i_mtime)) {
1245                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
1246                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1247                 LTIME_S(inode->i_mtime) = body->mtime;
1248         }
1249         if (body->valid & OBD_MD_FLCTIME &&
1250             body->ctime > LTIME_S(inode->i_ctime))
1251                 LTIME_S(inode->i_ctime) = body->ctime;
1252         if (body->valid & OBD_MD_FLMODE)
1253                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1254         if (body->valid & OBD_MD_FLTYPE)
1255                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1256         if (body->valid & OBD_MD_FLUID)
1257                 inode->i_uid = body->uid;
1258         if (body->valid & OBD_MD_FLGID)
1259                 inode->i_gid = body->gid;
1260         if (body->valid & OBD_MD_FLFLAGS)
1261                 inode->i_flags = body->flags;
1262         if (body->valid & OBD_MD_FLNLINK)
1263                 inode->i_nlink = body->nlink;
1264         if (body->valid & OBD_MD_FLGENER)
1265                 inode->i_generation = body->generation;
1266         if (body->valid & OBD_MD_FLRDEV)
1267 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1268                 inode->i_rdev = body->rdev;
1269 #else
1270                 inode->i_rdev = old_decode_dev(body->rdev);
1271 #endif
1272         if (body->valid & OBD_MD_FLSIZE)
1273                 inode->i_size = body->size;
1274         if (body->valid & OBD_MD_FLBLOCKS)
1275                 inode->i_blocks = body->blocks;
1276
1277         if (body->valid & OBD_MD_FLSIZE)
1278                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1279 }
1280
1281 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1282 static struct backing_dev_info ll_backing_dev_info = {
1283         .ra_pages       = 0,    /* No readahead */
1284 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1285         .capabilities   = 0,    /* Does contribute to dirty memory */
1286 #else
1287         .memory_backed  = 0,    /* Does contribute to dirty memory */
1288 #endif
1289 };
1290 #endif
1291
1292 void ll_read_inode2(struct inode *inode, void *opaque)
1293 {
1294         struct lustre_md *md = opaque;
1295         struct ll_inode_info *lli = ll_i2info(inode);
1296         ENTRY;
1297
1298         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1299                inode->i_generation, inode);
1300
1301         ll_lli_init(lli);
1302
1303         LASSERT(!lli->lli_smd);
1304
1305         /* Core attributes from the MDS first.  This is a new inode, and
1306          * the VFS doesn't zero times in the core inode so we have to do
1307          * it ourselves.  They will be overwritten by either MDS or OST
1308          * attributes - we just need to make sure they aren't newer. */
1309         LTIME_S(inode->i_mtime) = 0;
1310         LTIME_S(inode->i_atime) = 0;
1311         LTIME_S(inode->i_ctime) = 0;
1312         inode->i_rdev = 0;
1313         ll_update_inode(inode, md);
1314
1315         /* OIDEBUG(inode); */
1316
1317         if (S_ISREG(inode->i_mode)) {
1318                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1319                 inode->i_op = &ll_file_inode_operations;
1320                 inode->i_fop = sbi->ll_fop;
1321                 inode->i_mapping->a_ops = &ll_aops;
1322                 EXIT;
1323         } else if (S_ISDIR(inode->i_mode)) {
1324                 inode->i_op = &ll_dir_inode_operations;
1325                 inode->i_fop = &ll_dir_operations;
1326                 inode->i_mapping->a_ops = &ll_dir_aops;
1327                 EXIT;
1328         } else if (S_ISLNK(inode->i_mode)) {
1329                 inode->i_op = &ll_fast_symlink_inode_operations;
1330                 EXIT;
1331         } else {
1332                 inode->i_op = &ll_special_inode_operations;
1333
1334 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1335                 init_special_inode(inode, inode->i_mode,
1336                                    kdev_t_to_nr(inode->i_rdev));
1337
1338                 /* initializing backing dev info. */
1339                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1340 #else
1341                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1342 #endif
1343                 lli->ll_save_ifop = inode->i_fop;
1344
1345                 if (S_ISCHR(inode->i_mode))
1346                         inode->i_fop = &ll_special_chr_inode_fops;
1347                 else if (S_ISBLK(inode->i_mode))
1348                         inode->i_fop = &ll_special_blk_inode_fops;
1349                 else if (S_ISFIFO(inode->i_mode))
1350                         inode->i_fop = &ll_special_fifo_inode_fops;
1351                 else if (S_ISSOCK(inode->i_mode))
1352                         inode->i_fop = &ll_special_sock_inode_fops;
1353                 EXIT;
1354         }
1355 }
1356
1357 int ll_iocontrol(struct inode *inode, struct file *file,
1358                  unsigned int cmd, unsigned long arg)
1359 {
1360         struct ll_sb_info *sbi = ll_i2sbi(inode);
1361         struct ptlrpc_request *req = NULL;
1362         int rc, flags = 0;
1363         ENTRY;
1364
1365         switch(cmd) {
1366         case EXT3_IOC_GETFLAGS: {
1367                 struct ll_fid fid;
1368                 struct mds_body *body;
1369
1370                 ll_inode2fid(&fid, inode);
1371                 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
1372                 if (rc) {
1373                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1374                         RETURN(-abs(rc));
1375                 }
1376
1377                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1378
1379                 if (body->flags & S_APPEND)
1380                         flags |= EXT3_APPEND_FL;
1381                 if (body->flags & S_IMMUTABLE)
1382                         flags |= EXT3_IMMUTABLE_FL;
1383                 if (body->flags & S_NOATIME)
1384                         flags |= EXT3_NOATIME_FL;
1385
1386                 ptlrpc_req_finished (req);
1387
1388                 RETURN(put_user(flags, (int *)arg));
1389         }
1390         case EXT3_IOC_SETFLAGS: {
1391                 struct mdc_op_data op_data;
1392                 struct iattr attr;
1393                 struct obdo *oa;
1394                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1395
1396                 if (get_user(flags, (int *)arg))
1397                         RETURN(-EFAULT);
1398
1399                 oa = obdo_alloc();
1400                 if (!oa)
1401                         RETURN(-ENOMEM);
1402
1403                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1404
1405                 memset(&attr, 0x0, sizeof(attr));
1406                 attr.ia_attr_flags = flags;
1407                 attr.ia_valid |= ATTR_ATTR_FLAG;
1408
1409                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1410                                  &attr, NULL, 0, NULL, 0, &req);
1411                 if (rc) {
1412                         ptlrpc_req_finished(req);
1413                         if (rc != -EPERM && rc != -EACCES)
1414                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
1415                         obdo_free(oa);
1416                         RETURN(rc);
1417                 }
1418                 ptlrpc_req_finished(req);
1419
1420                 oa->o_id = lsm->lsm_object_id;
1421                 oa->o_flags = flags;
1422                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1423
1424                 obdo_from_inode(oa, inode, OBD_MD_FLFID | OBD_MD_FLGENER);
1425                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1426                 obdo_free(oa);
1427                 if (rc) {
1428                         if (rc != -EPERM && rc != -EACCES)
1429                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
1430                         RETURN(rc);
1431                 }
1432
1433                 if (flags & EXT3_APPEND_FL)
1434                         inode->i_flags |= S_APPEND;
1435                 else
1436                         inode->i_flags &= ~S_APPEND;
1437                 if (flags & EXT3_IMMUTABLE_FL)
1438                         inode->i_flags |= S_IMMUTABLE;
1439                 else
1440                         inode->i_flags &= ~S_IMMUTABLE;
1441                 if (flags & EXT3_NOATIME_FL)
1442                         inode->i_flags |= S_NOATIME;
1443                 else
1444                         inode->i_flags &= ~S_NOATIME;
1445
1446                 RETURN(0);
1447         }
1448         default:
1449                 RETURN(-ENOSYS);
1450         }
1451
1452         RETURN(0);
1453 }
1454
1455 void ll_umount_begin(struct super_block *sb)
1456 {
1457         struct lustre_sb_info *lsi = s2lsi(sb);
1458         struct ll_sb_info *sbi = ll_s2sbi(sb);
1459         struct obd_device *obd;
1460         struct obd_ioctl_data ioc_data = { 0 };
1461         ENTRY;
1462
1463         /* Make the MGC not try to cancel locks */
1464         lsi->lsi_flags |= LSI_UMOUNT_FAILOVER;
1465
1466         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1467                sb->s_count, atomic_read(&sb->s_active));
1468
1469         obd = class_exp2obd(sbi->ll_mdc_exp);
1470         if (obd == NULL) {
1471                 CERROR("Invalid MDC connection handle "LPX64"\n",
1472                        sbi->ll_mdc_exp->exp_handle.h_cookie);
1473                 EXIT;
1474                 return;
1475         }
1476         obd->obd_no_recov = 1;
1477         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1478                       &ioc_data, NULL);
1479
1480         obd = class_exp2obd(sbi->ll_osc_exp);
1481         if (obd == NULL) {
1482                 CERROR("Invalid LOV connection handle "LPX64"\n",
1483                        sbi->ll_osc_exp->exp_handle.h_cookie);
1484                 EXIT;
1485                 return;
1486         }
1487         obd->obd_no_recov = 1;
1488         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1489                       &ioc_data, NULL);
1490
1491         /* Really, we'd like to wait until there are no requests outstanding,
1492          * and then continue.  For now, we just invalidate the requests,
1493          * schedule, and hope.
1494          */
1495         schedule();
1496
1497         EXIT;
1498 }
1499
1500 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1501 {
1502         struct ll_sb_info *sbi = ll_s2sbi(sb);
1503         int err;
1504         __u32 read_only;
1505
1506         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1507                 read_only = *flags & MS_RDONLY;
1508                 err = obd_set_info(sbi->ll_mdc_exp, strlen("read-only"),
1509                                    "read-only", sizeof(read_only), &read_only);
1510                 if (err) {
1511                         CERROR("Failed to change the read-only flag during "
1512                                "remount: %d\n", err);
1513                         return err;
1514                 }
1515
1516                 if (read_only)
1517                         sb->s_flags |= MS_RDONLY;
1518                 else
1519                         sb->s_flags &= ~MS_RDONLY;
1520         }
1521         return 0;
1522 }
1523
1524 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1525                   struct ptlrpc_request *req, int offset,struct super_block *sb)
1526 {
1527         struct lustre_md md;
1528         struct ll_sb_info *sbi = NULL;
1529         int rc = 0;
1530         ENTRY;
1531
1532         LASSERT(*inode || sb);
1533         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1534         prune_deathrow(sbi, 1);
1535
1536         rc = mdc_req2lustre_md(req, offset, exp, &md);
1537         if (rc)
1538                 RETURN(rc);
1539
1540         if (*inode) {
1541                 ll_update_inode(*inode, &md);
1542         } else {
1543                 LASSERT(sb);
1544                 *inode = ll_iget(sb, md.body->ino, &md);
1545                 if (*inode == NULL || is_bad_inode(*inode)) {
1546                         mdc_free_lustre_md(exp, &md);
1547                         rc = -ENOMEM;
1548                         CERROR("new_inode -fatal: rc %d\n", rc);
1549                         GOTO(out, rc);
1550                 }
1551         }
1552
1553         rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
1554                          ll_i2info(*inode)->lli_smd);
1555 out:
1556         RETURN(rc);
1557 }
1558
1559 char *llap_origins[] = {
1560         [LLAP_ORIGIN_UNKNOWN] = "--",
1561         [LLAP_ORIGIN_READPAGE] = "rp",
1562         [LLAP_ORIGIN_READAHEAD] = "ra",
1563         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
1564         [LLAP_ORIGIN_WRITEPAGE] = "wp",
1565 };
1566
1567 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
1568                                              struct list_head *list)
1569 {
1570         struct ll_async_page *llap;
1571         struct list_head *pos;
1572
1573         list_for_each(pos, list) {
1574                 if (pos == &sbi->ll_pglist)
1575                         return NULL;
1576                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
1577                 if (llap->llap_page == NULL)
1578                         continue;
1579                 return llap;
1580         }
1581         LBUG();
1582         return NULL;
1583 }
1584
1585 EXPORT_SYMBOL(ll_fill_super);
1586 EXPORT_SYMBOL(ll_put_super);
1587 EXPORT_SYMBOL(ll_remount_fs);
1588 EXPORT_SYMBOL(ll_umount_begin);
1589