fs/xfs/xfs_inode_item.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_trans_priv.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_error.h"
  33 #include "xfs_trace.h"
  34
  35
  36 kmem_zone_t     *xfs_ili_zone;          /* inode log item zone */
  37
  38 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
  39 {
  40         return container_of(lip, struct xfs_inode_log_item, ili_item);
  41 }
  42
  43
  44 /*
  45  * This returns the number of iovecs needed to log the given inode item.
  46  *
  47  * We need one iovec for the inode log format structure, one for the
  48  * inode core, and possibly one for the inode data/extents/b-tree root
  49  * and one for the inode attribute data/extents/b-tree root.
  50  */
  51 STATIC uint
  52 xfs_inode_item_size(
  53         struct xfs_log_item     *lip)
  54 {
  55         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
  56         struct xfs_inode        *ip = iip->ili_inode;
  57         uint                    nvecs = 2;
  58
  59         switch (ip->i_d.di_format) {
  60         case XFS_DINODE_FMT_EXTENTS:
  61                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
  62                     ip->i_d.di_nextents > 0 &&
  63                     ip->i_df.if_bytes > 0)
  64                         nvecs++;
  65                 break;
  66
  67         case XFS_DINODE_FMT_BTREE:
  68                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
  69                     ip->i_df.if_broot_bytes > 0)
  70                         nvecs++;
  71                 break;
  72
  73         case XFS_DINODE_FMT_LOCAL:
  74                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
  75                     ip->i_df.if_bytes > 0)
  76                         nvecs++;
  77                 break;
  78
  79         case XFS_DINODE_FMT_DEV:
  80         case XFS_DINODE_FMT_UUID:
  81                 break;
  82
  83         default:
  84                 ASSERT(0);
  85                 break;
  86         }
  87
  88         if (!XFS_IFORK_Q(ip))
  89                 return nvecs;
  90
  91
  92         /*
  93          * Log any necessary attribute data.
  94          */
  95         switch (ip->i_d.di_aformat) {
  96         case XFS_DINODE_FMT_EXTENTS:
  97                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
  98                     ip->i_d.di_anextents > 0 &&
  99                     ip->i_afp->if_bytes > 0)
 100                         nvecs++;
 101                 break;
 102
 103         case XFS_DINODE_FMT_BTREE:
 104                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 105                     ip->i_afp->if_broot_bytes > 0)
 106                         nvecs++;
 107                 break;
 108
 109         case XFS_DINODE_FMT_LOCAL:
 110                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 111                     ip->i_afp->if_bytes > 0)
 112                         nvecs++;
 113                 break;
 114
 115         default:
 116                 ASSERT(0);
 117                 break;
 118         }
 119
 120         return nvecs;
 121 }
 122
 123 /*
 124  * xfs_inode_item_format_extents - convert in-core extents to on-disk form
 125  *
 126  * For either the data or attr fork in extent format, we need to endian convert
 127  * the in-core extent as we place them into the on-disk inode. In this case, we
 128  * need to do this conversion before we write the extents into the log. Because
 129  * we don't have the disk inode to write into here, we allocate a buffer and
 130  * format the extents into it via xfs_iextents_copy(). We free the buffer in
 131  * the unlock routine after the copy for the log has been made.
 132  *
 133  * In the case of the data fork, the in-core and on-disk fork sizes can be
 134  * different due to delayed allocation extents. We only log on-disk extents
 135  * here, so always use the physical fork size to determine the size of the
 136  * buffer we need to allocate.
 137  */
 138 STATIC void
 139 xfs_inode_item_format_extents(
 140         struct xfs_inode        *ip,
 141         struct xfs_log_iovec    *vecp,
 142         int                     whichfork,
 143         int                     type)
 144 {
 145         xfs_bmbt_rec_t          *ext_buffer;
 146
 147         ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
 148         if (whichfork == XFS_DATA_FORK)
 149                 ip->i_itemp->ili_extents_buf = ext_buffer;
 150         else
 151                 ip->i_itemp->ili_aextents_buf = ext_buffer;
 152
 153         vecp->i_addr = ext_buffer;
 154         vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
 155         vecp->i_type = type;
 156 }
 157
 158 /*
 159  * This is called to fill in the vector of log iovecs for the
 160  * given inode log item.  It fills the first item with an inode
 161  * log format structure, the second with the on-disk inode structure,
 162  * and a possible third and/or fourth with the inode data/extents/b-tree
 163  * root and inode attributes data/extents/b-tree root.
 164  */
 165 STATIC void
 166 xfs_inode_item_format(
 167         struct xfs_log_item     *lip,
 168         struct xfs_log_iovec    *vecp)
 169 {
 170         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 171         struct xfs_inode        *ip = iip->ili_inode;
 172         uint                    nvecs;
 173         size_t                  data_bytes;
 174         xfs_mount_t             *mp;
 175
 176         vecp->i_addr = &iip->ili_format;
 177         vecp->i_len  = sizeof(xfs_inode_log_format_t);
 178         vecp->i_type = XLOG_REG_TYPE_IFORMAT;
 179         vecp++;
 180         nvecs        = 1;
 181
 182         vecp->i_addr = &ip->i_d;
 183         vecp->i_len  = sizeof(struct xfs_icdinode);
 184         vecp->i_type = XLOG_REG_TYPE_ICORE;
 185         vecp++;
 186         nvecs++;
 187
 188         /*
 189          * If this is really an old format inode, then we need to
 190          * log it as such.  This means that we have to copy the link
 191          * count from the new field to the old.  We don't have to worry
 192          * about the new fields, because nothing trusts them as long as
 193          * the old inode version number is there.  If the superblock already
 194          * has a new version number, then we don't bother converting back.
 195          */
 196         mp = ip->i_mount;
 197         ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
 198         if (ip->i_d.di_version == 1) {
 199                 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
 200                         /*
 201                          * Convert it back.
 202                          */
 203                         ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
 204                         ip->i_d.di_onlink = ip->i_d.di_nlink;
 205                 } else {
 206                         /*
 207                          * The superblock version has already been bumped,
 208                          * so just make the conversion to the new inode
 209                          * format permanent.
 210                          */
 211                         ip->i_d.di_version = 2;
 212                         ip->i_d.di_onlink = 0;
 213                         memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 214                 }
 215         }
 216
 217         switch (ip->i_d.di_format) {
 218         case XFS_DINODE_FMT_EXTENTS:
 219                 iip->ili_fields &=
 220                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 221                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 222
 223                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 224                     ip->i_d.di_nextents > 0 &&
 225                     ip->i_df.if_bytes > 0) {
 226                         ASSERT(ip->i_df.if_u1.if_extents != NULL);
 227                         ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
 228                         ASSERT(iip->ili_extents_buf == NULL);
 229
 230 #ifdef XFS_NATIVE_HOST
 231                        if (ip->i_d.di_nextents == ip->i_df.if_bytes /
 232                                                (uint)sizeof(xfs_bmbt_rec_t)) {
 233                                 /*
 234                                  * There are no delayed allocation
 235                                  * extents, so just point to the
 236                                  * real extents array.
 237                                  */
 238                                 vecp->i_addr = ip->i_df.if_u1.if_extents;
 239                                 vecp->i_len = ip->i_df.if_bytes;
 240                                 vecp->i_type = XLOG_REG_TYPE_IEXT;
 241                         } else
 242 #endif
 243                         {
 244                                 xfs_inode_item_format_extents(ip, vecp,
 245                                         XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
 246                         }
 247                         ASSERT(vecp->i_len <= ip->i_df.if_bytes);
 248                         iip->ili_format.ilf_dsize = vecp->i_len;
 249                         vecp++;
 250                         nvecs++;
 251                 } else {
 252                         iip->ili_fields &= ~XFS_ILOG_DEXT;
 253                 }
 254                 break;
 255
 256         case XFS_DINODE_FMT_BTREE:
 257                 iip->ili_fields &=
 258                         ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
 259                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 260
 261                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 262                     ip->i_df.if_broot_bytes > 0) {
 263                         ASSERT(ip->i_df.if_broot != NULL);
 264                         vecp->i_addr = ip->i_df.if_broot;
 265                         vecp->i_len = ip->i_df.if_broot_bytes;
 266                         vecp->i_type = XLOG_REG_TYPE_IBROOT;
 267                         vecp++;
 268                         nvecs++;
 269                         iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
 270                 } else {
 271                         ASSERT(!(iip->ili_fields &
 272                                  XFS_ILOG_DBROOT));
 273 #ifdef XFS_TRANS_DEBUG
 274                         if (iip->ili_root_size > 0) {
 275                                 ASSERT(iip->ili_root_size ==
 276                                        ip->i_df.if_broot_bytes);
 277                                 ASSERT(memcmp(iip->ili_orig_root,
 278                                             ip->i_df.if_broot,
 279                                             iip->ili_root_size) == 0);
 280                         } else {
 281                                 ASSERT(ip->i_df.if_broot_bytes == 0);
 282                         }
 283 #endif
 284                         iip->ili_fields &= ~XFS_ILOG_DBROOT;
 285                 }
 286                 break;
 287
 288         case XFS_DINODE_FMT_LOCAL:
 289                 iip->ili_fields &=
 290                         ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
 291                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 292                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 293                     ip->i_df.if_bytes > 0) {
 294                         ASSERT(ip->i_df.if_u1.if_data != NULL);
 295                         ASSERT(ip->i_d.di_size > 0);
 296
 297                         vecp->i_addr = ip->i_df.if_u1.if_data;
 298                         /*
 299                          * Round i_bytes up to a word boundary.
 300                          * The underlying memory is guaranteed to
 301                          * to be there by xfs_idata_realloc().
 302                          */
 303                         data_bytes = roundup(ip->i_df.if_bytes, 4);
 304                         ASSERT((ip->i_df.if_real_bytes == 0) ||
 305                                (ip->i_df.if_real_bytes == data_bytes));
 306                         vecp->i_len = (int)data_bytes;
 307                         vecp->i_type = XLOG_REG_TYPE_ILOCAL;
 308                         vecp++;
 309                         nvecs++;
 310                         iip->ili_format.ilf_dsize = (unsigned)data_bytes;
 311                 } else {
 312                         iip->ili_fields &= ~XFS_ILOG_DDATA;
 313                 }
 314                 break;
 315
 316         case XFS_DINODE_FMT_DEV:
 317                 iip->ili_fields &=
 318                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 319                           XFS_ILOG_DEXT | XFS_ILOG_UUID);
 320                 if (iip->ili_fields & XFS_ILOG_DEV) {
 321                         iip->ili_format.ilf_u.ilfu_rdev =
 322                                 ip->i_df.if_u2.if_rdev;
 323                 }
 324                 break;
 325
 326         case XFS_DINODE_FMT_UUID:
 327                 iip->ili_fields &=
 328                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 329                           XFS_ILOG_DEXT | XFS_ILOG_DEV);
 330                 if (iip->ili_fields & XFS_ILOG_UUID) {
 331                         iip->ili_format.ilf_u.ilfu_uuid =
 332                                 ip->i_df.if_u2.if_uuid;
 333                 }
 334                 break;
 335
 336         default:
 337                 ASSERT(0);
 338                 break;
 339         }
 340
 341         /*
 342          * If there are no attributes associated with the file, then we're done.
 343          */
 344         if (!XFS_IFORK_Q(ip)) {
 345                 iip->ili_fields &=
 346                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
 347                 goto out;
 348         }
 349
 350         switch (ip->i_d.di_aformat) {
 351         case XFS_DINODE_FMT_EXTENTS:
 352                 iip->ili_fields &=
 353                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
 354
 355                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 356                     ip->i_d.di_anextents > 0 &&
 357                     ip->i_afp->if_bytes > 0) {
 358                         ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
 359                                 ip->i_d.di_anextents);
 360                         ASSERT(ip->i_afp->if_u1.if_extents != NULL);
 361 #ifdef XFS_NATIVE_HOST
 362                         /*
 363                          * There are not delayed allocation extents
 364                          * for attributes, so just point at the array.
 365                          */
 366                         vecp->i_addr = ip->i_afp->if_u1.if_extents;
 367                         vecp->i_len = ip->i_afp->if_bytes;
 368                         vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
 369 #else
 370                         ASSERT(iip->ili_aextents_buf == NULL);
 371                         xfs_inode_item_format_extents(ip, vecp,
 372                                         XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
 373 #endif
 374                         iip->ili_format.ilf_asize = vecp->i_len;
 375                         vecp++;
 376                         nvecs++;
 377                 } else {
 378                         iip->ili_fields &= ~XFS_ILOG_AEXT;
 379                 }
 380                 break;
 381
 382         case XFS_DINODE_FMT_BTREE:
 383                 iip->ili_fields &=
 384                         ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
 385
 386                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 387                     ip->i_afp->if_broot_bytes > 0) {
 388                         ASSERT(ip->i_afp->if_broot != NULL);
 389
 390                         vecp->i_addr = ip->i_afp->if_broot;
 391                         vecp->i_len = ip->i_afp->if_broot_bytes;
 392                         vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
 393                         vecp++;
 394                         nvecs++;
 395                         iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
 396                 } else {
 397                         iip->ili_fields &= ~XFS_ILOG_ABROOT;
 398                 }
 399                 break;
 400
 401         case XFS_DINODE_FMT_LOCAL:
 402                 iip->ili_fields &=
 403                         ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 404
 405                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 406                     ip->i_afp->if_bytes > 0) {
 407                         ASSERT(ip->i_afp->if_u1.if_data != NULL);
 408
 409                         vecp->i_addr = ip->i_afp->if_u1.if_data;
 410                         /*
 411                          * Round i_bytes up to a word boundary.
 412                          * The underlying memory is guaranteed to
 413                          * to be there by xfs_idata_realloc().
 414                          */
 415                         data_bytes = roundup(ip->i_afp->if_bytes, 4);
 416                         ASSERT((ip->i_afp->if_real_bytes == 0) ||
 417                                (ip->i_afp->if_real_bytes == data_bytes));
 418                         vecp->i_len = (int)data_bytes;
 419                         vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
 420                         vecp++;
 421                         nvecs++;
 422                         iip->ili_format.ilf_asize = (unsigned)data_bytes;
 423                 } else {
 424                         iip->ili_fields &= ~XFS_ILOG_ADATA;
 425                 }
 426                 break;
 427
 428         default:
 429                 ASSERT(0);
 430                 break;
 431         }
 432
 433 out:
 434         /*
 435          * Now update the log format that goes out to disk from the in-core
 436          * values.  We always write the inode core to make the arithmetic
 437          * games in recovery easier, which isn't a big deal as just about any
 438          * transaction would dirty it anyway.
 439          */
 440         iip->ili_format.ilf_fields = XFS_ILOG_CORE |
 441                 (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
 442         iip->ili_format.ilf_size = nvecs;
 443 }
 444
 445
 446 /*
 447  * This is called to pin the inode associated with the inode log
 448  * item in memory so it cannot be written out.
 449  */
 450 STATIC void
 451 xfs_inode_item_pin(
 452         struct xfs_log_item     *lip)
 453 {
 454         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 455
 456         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 457
 458         trace_xfs_inode_pin(ip, _RET_IP_);
 459         atomic_inc(&ip->i_pincount);
 460 }
 461
 462
 463 /*
 464  * This is called to unpin the inode associated with the inode log
 465  * item which was previously pinned with a call to xfs_inode_item_pin().
 466  *
 467  * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
 468  */
 469 STATIC void
 470 xfs_inode_item_unpin(
 471         struct xfs_log_item     *lip,
 472         int                     remove)
 473 {
 474         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 475
 476         trace_xfs_inode_unpin(ip, _RET_IP_);
 477         ASSERT(atomic_read(&ip->i_pincount) > 0);
 478         if (atomic_dec_and_test(&ip->i_pincount))
 479                 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 480 }
 481
 482 STATIC uint
 483 xfs_inode_item_push(
 484         struct xfs_log_item     *lip,
 485         struct list_head        *buffer_list)
 486 {
 487         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 488         struct xfs_inode        *ip = iip->ili_inode;
 489         struct xfs_buf          *bp = NULL;
 490         uint                    rval = XFS_ITEM_SUCCESS;
 491         int                     error;
 492
 493         if (xfs_ipincount(ip) > 0)
 494                 return XFS_ITEM_PINNED;
 495
 496         if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
 497                 return XFS_ITEM_LOCKED;
 498
 499         /*
 500          * Re-check the pincount now that we stabilized the value by
 501          * taking the ilock.
 502          */
 503         if (xfs_ipincount(ip) > 0) {
 504                 rval = XFS_ITEM_PINNED;
 505                 goto out_unlock;
 506         }
 507
 508         /*
 509          * Someone else is already flushing the inode.  Nothing we can do
 510          * here but wait for the flush to finish and remove the item from
 511          * the AIL.
 512          */
 513         if (!xfs_iflock_nowait(ip)) {
 514                 rval = XFS_ITEM_FLUSHING;
 515                 goto out_unlock;
 516         }
 517
 518         /*
 519          * Stale inode items should force out the iclog.
 520          */
 521         if (ip->i_flags & XFS_ISTALE) {
 522                 xfs_ifunlock(ip);
 523                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 524                 return XFS_ITEM_PINNED;
 525         }
 526
 527         ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
 528         ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
 529
 530         spin_unlock(&lip->li_ailp->xa_lock);
 531
 532         error = xfs_iflush(ip, &bp);
 533         if (!error) {
 534                 if (!xfs_buf_delwri_queue(bp, buffer_list))
 535                         rval = XFS_ITEM_FLUSHING;
 536                 xfs_buf_relse(bp);
 537         }
 538
 539         spin_lock(&lip->li_ailp->xa_lock);
 540 out_unlock:
 541         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 542         return rval;
 543 }
 544
 545 /*
 546  * Unlock the inode associated with the inode log item.
 547  * Clear the fields of the inode and inode log item that
 548  * are specific to the current transaction.  If the
 549  * hold flags is set, do not unlock the inode.
 550  */
 551 STATIC void
 552 xfs_inode_item_unlock(
 553         struct xfs_log_item     *lip)
 554 {
 555         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 556         struct xfs_inode        *ip = iip->ili_inode;
 557         unsigned short          lock_flags;
 558
 559         ASSERT(ip->i_itemp != NULL);
 560         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 561
 562         /*
 563          * If the inode needed a separate buffer with which to log
 564          * its extents, then free it now.
 565          */
 566         if (iip->ili_extents_buf != NULL) {
 567                 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
 568                 ASSERT(ip->i_d.di_nextents > 0);
 569                 ASSERT(iip->ili_fields & XFS_ILOG_DEXT);
 570                 ASSERT(ip->i_df.if_bytes > 0);
 571                 kmem_free(iip->ili_extents_buf);
 572                 iip->ili_extents_buf = NULL;
 573         }
 574         if (iip->ili_aextents_buf != NULL) {
 575                 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
 576                 ASSERT(ip->i_d.di_anextents > 0);
 577                 ASSERT(iip->ili_fields & XFS_ILOG_AEXT);
 578                 ASSERT(ip->i_afp->if_bytes > 0);
 579                 kmem_free(iip->ili_aextents_buf);
 580                 iip->ili_aextents_buf = NULL;
 581         }
 582
 583         lock_flags = iip->ili_lock_flags;
 584         iip->ili_lock_flags = 0;
 585         if (lock_flags)
 586                 xfs_iunlock(ip, lock_flags);
 587 }
 588
 589 /*
 590  * This is called to find out where the oldest active copy of the inode log
 591  * item in the on disk log resides now that the last log write of it completed
 592  * at the given lsn.  Since we always re-log all dirty data in an inode, the
 593  * latest copy in the on disk log is the only one that matters.  Therefore,
 594  * simply return the given lsn.
 595  *
 596  * If the inode has been marked stale because the cluster is being freed, we
 597  * don't want to (re-)insert this inode into the AIL. There is a race condition
 598  * where the cluster buffer may be unpinned before the inode is inserted into
 599  * the AIL during transaction committed processing. If the buffer is unpinned
 600  * before the inode item has been committed and inserted, then it is possible
 601  * for the buffer to be written and IO completes before the inode is inserted
 602  * into the AIL. In that case, we'd be inserting a clean, stale inode into the
 603  * AIL which will never get removed. It will, however, get reclaimed which
 604  * triggers an assert in xfs_inode_free() complaining about freein an inode
 605  * still in the AIL.
 606  *
 607  * To avoid this, just unpin the inode directly and return a LSN of -1 so the
 608  * transaction committed code knows that it does not need to do any further
 609  * processing on the item.
 610  */
 611 STATIC xfs_lsn_t
 612 xfs_inode_item_committed(
 613         struct xfs_log_item     *lip,
 614         xfs_lsn_t               lsn)
 615 {
 616         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 617         struct xfs_inode        *ip = iip->ili_inode;
 618
 619         if (xfs_iflags_test(ip, XFS_ISTALE)) {
 620                 xfs_inode_item_unpin(lip, 0);
 621                 return -1;
 622         }
 623         return lsn;
 624 }
 625
 626 /*
 627  * XXX rcc - this one really has to do something.  Probably needs
 628  * to stamp in a new field in the incore inode.
 629  */
 630 STATIC void
 631 xfs_inode_item_committing(
 632         struct xfs_log_item     *lip,
 633         xfs_lsn_t               lsn)
 634 {
 635         INODE_ITEM(lip)->ili_last_lsn = lsn;
 636 }
 637
 638 /*
 639  * This is the ops vector shared by all buf log items.
 640  */
 641 static const struct xfs_item_ops xfs_inode_item_ops = {
 642         .iop_size       = xfs_inode_item_size,
 643         .iop_format     = xfs_inode_item_format,
 644         .iop_pin        = xfs_inode_item_pin,
 645         .iop_unpin      = xfs_inode_item_unpin,
 646         .iop_unlock     = xfs_inode_item_unlock,
 647         .iop_committed  = xfs_inode_item_committed,
 648         .iop_push       = xfs_inode_item_push,
 649         .iop_committing = xfs_inode_item_committing
 650 };
 651
 652
 653 /*
 654  * Initialize the inode log item for a newly allocated (in-core) inode.
 655  */
 656 void
 657 xfs_inode_item_init(
 658         struct xfs_inode        *ip,
 659         struct xfs_mount        *mp)
 660 {
 661         struct xfs_inode_log_item *iip;
 662
 663         ASSERT(ip->i_itemp == NULL);
 664         iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
 665
 666         iip->ili_inode = ip;
 667         xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
 668                                                 &xfs_inode_item_ops);
 669         iip->ili_format.ilf_type = XFS_LI_INODE;
 670         iip->ili_format.ilf_ino = ip->i_ino;
 671         iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
 672         iip->ili_format.ilf_len = ip->i_imap.im_len;
 673         iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
 674 }
 675
 676 /*
 677  * Free the inode log item and any memory hanging off of it.
 678  */
 679 void
 680 xfs_inode_item_destroy(
 681         xfs_inode_t     *ip)
 682 {
 683 #ifdef XFS_TRANS_DEBUG
 684         if (ip->i_itemp->ili_root_size != 0) {
 685                 kmem_free(ip->i_itemp->ili_orig_root);
 686         }
 687 #endif
 688         kmem_zone_free(xfs_ili_zone, ip->i_itemp);
 689 }
 690
 691
 692 /*
 693  * This is the inode flushing I/O completion routine.  It is called
 694  * from interrupt level when the buffer containing the inode is
 695  * flushed to disk.  It is responsible for removing the inode item
 696  * from the AIL if it has not been re-logged, and unlocking the inode's
 697  * flush lock.
 698  *
 699  * To reduce AIL lock traffic as much as possible, we scan the buffer log item
 700  * list for other inodes that will run this function. We remove them from the
 701  * buffer list so we can process all the inode IO completions in one AIL lock
 702  * traversal.
 703  */
 704 void
 705 xfs_iflush_done(
 706         struct xfs_buf          *bp,
 707         struct xfs_log_item     *lip)
 708 {
 709         struct xfs_inode_log_item *iip;
 710         struct xfs_log_item     *blip;
 711         struct xfs_log_item     *next;
 712         struct xfs_log_item     *prev;
 713         struct xfs_ail          *ailp = lip->li_ailp;
 714         int                     need_ail = 0;
 715
 716         /*
 717          * Scan the buffer IO completions for other inodes being completed and
 718          * attach them to the current inode log item.
 719          */
 720         blip = bp->b_fspriv;
 721         prev = NULL;
 722         while (blip != NULL) {
 723                 if (lip->li_cb != xfs_iflush_done) {
 724                         prev = blip;
 725                         blip = blip->li_bio_list;
 726                         continue;
 727                 }
 728
 729                 /* remove from list */
 730                 next = blip->li_bio_list;
 731                 if (!prev) {
 732                         bp->b_fspriv = next;
 733                 } else {
 734                         prev->li_bio_list = next;
 735                 }
 736
 737                 /* add to current list */
 738                 blip->li_bio_list = lip->li_bio_list;
 739                 lip->li_bio_list = blip;
 740
 741                 /*
 742                  * while we have the item, do the unlocked check for needing
 743                  * the AIL lock.
 744                  */
 745                 iip = INODE_ITEM(blip);
 746                 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
 747                         need_ail++;
 748
 749                 blip = next;
 750         }
 751
 752         /* make sure we capture the state of the initial inode. */
 753         iip = INODE_ITEM(lip);
 754         if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
 755                 need_ail++;
 756
 757         /*
 758          * We only want to pull the item from the AIL if it is
 759          * actually there and its location in the log has not
 760          * changed since we started the flush.  Thus, we only bother
 761          * if the ili_logged flag is set and the inode's lsn has not
 762          * changed.  First we check the lsn outside
 763          * the lock since it's cheaper, and then we recheck while
 764          * holding the lock before removing the inode from the AIL.
 765          */
 766         if (need_ail) {
 767                 struct xfs_log_item *log_items[need_ail];
 768                 int i = 0;
 769                 spin_lock(&ailp->xa_lock);
 770                 for (blip = lip; blip; blip = blip->li_bio_list) {
 771                         iip = INODE_ITEM(blip);
 772                         if (iip->ili_logged &&
 773                             blip->li_lsn == iip->ili_flush_lsn) {
 774                                 log_items[i++] = blip;
 775                         }
 776                         ASSERT(i <= need_ail);
 777                 }
 778                 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
 779                 xfs_trans_ail_delete_bulk(ailp, log_items, i,
 780                                           SHUTDOWN_CORRUPT_INCORE);
 781         }
 782
 783
 784         /*
 785          * clean up and unlock the flush lock now we are done. We can clear the
 786          * ili_last_fields bits now that we know that the data corresponding to
 787          * them is safely on disk.
 788          */
 789         for (blip = lip; blip; blip = next) {
 790                 next = blip->li_bio_list;
 791                 blip->li_bio_list = NULL;
 792
 793                 iip = INODE_ITEM(blip);
 794                 iip->ili_logged = 0;
 795                 iip->ili_last_fields = 0;
 796                 xfs_ifunlock(iip->ili_inode);
 797         }
 798 }
 799
 800 /*
 801  * This is the inode flushing abort routine.  It is called from xfs_iflush when
 802  * the filesystem is shutting down to clean up the inode state.  It is
 803  * responsible for removing the inode item from the AIL if it has not been
 804  * re-logged, and unlocking the inode's flush lock.
 805  */
 806 void
 807 xfs_iflush_abort(
 808         xfs_inode_t             *ip,
 809         bool                    stale)
 810 {
 811         xfs_inode_log_item_t    *iip = ip->i_itemp;
 812
 813         if (iip) {
 814                 struct xfs_ail  *ailp = iip->ili_item.li_ailp;
 815                 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 816                         spin_lock(&ailp->xa_lock);
 817                         if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 818                                 /* xfs_trans_ail_delete() drops the AIL lock. */
 819                                 xfs_trans_ail_delete(ailp, &iip->ili_item,
 820                                                 stale ?
 821                                                      SHUTDOWN_LOG_IO_ERROR :
 822                                                      SHUTDOWN_CORRUPT_INCORE);
 823                         } else
 824                                 spin_unlock(&ailp->xa_lock);
 825                 }
 826                 iip->ili_logged = 0;
 827                 /*
 828                  * Clear the ili_last_fields bits now that we know that the
 829                  * data corresponding to them is safely on disk.
 830                  */
 831                 iip->ili_last_fields = 0;
 832                 /*
 833                  * Clear the inode logging fields so no more flushes are
 834                  * attempted.
 835                  */
 836                 iip->ili_fields = 0;
 837         }
 838         /*
 839          * Release the inode's flush lock since we're done with it.
 840          */
 841         xfs_ifunlock(ip);
 842 }
 843
 844 void
 845 xfs_istale_done(
 846         struct xfs_buf          *bp,
 847         struct xfs_log_item     *lip)
 848 {
 849         xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
 850 }
 851
 852 /*
 853  * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
 854  * (which can have different field alignments) to the native version
 855  */
 856 int
 857 xfs_inode_item_format_convert(
 858         xfs_log_iovec_t         *buf,
 859         xfs_inode_log_format_t  *in_f)
 860 {
 861         if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
 862                 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
 863
 864                 in_f->ilf_type = in_f32->ilf_type;
 865                 in_f->ilf_size = in_f32->ilf_size;
 866                 in_f->ilf_fields = in_f32->ilf_fields;
 867                 in_f->ilf_asize = in_f32->ilf_asize;
 868                 in_f->ilf_dsize = in_f32->ilf_dsize;
 869                 in_f->ilf_ino = in_f32->ilf_ino;
 870                 /* copy biggest field of ilf_u */
 871                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 872                        in_f32->ilf_u.ilfu_uuid.__u_bits,
 873                        sizeof(uuid_t));
 874                 in_f->ilf_blkno = in_f32->ilf_blkno;
 875                 in_f->ilf_len = in_f32->ilf_len;
 876                 in_f->ilf_boffset = in_f32->ilf_boffset;
 877                 return 0;
 878         } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
 879                 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
 880
 881                 in_f->ilf_type = in_f64->ilf_type;
 882                 in_f->ilf_size = in_f64->ilf_size;
 883                 in_f->ilf_fields = in_f64->ilf_fields;
 884                 in_f->ilf_asize = in_f64->ilf_asize;
 885                 in_f->ilf_dsize = in_f64->ilf_dsize;
 886                 in_f->ilf_ino = in_f64->ilf_ino;
 887                 /* copy biggest field of ilf_u */
 888                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 889                        in_f64->ilf_u.ilfu_uuid.__u_bits,
 890                        sizeof(uuid_t));
 891                 in_f->ilf_blkno = in_f64->ilf_blkno;
 892                 in_f->ilf_len = in_f64->ilf_len;
 893                 in_f->ilf_boffset = in_f64->ilf_boffset;
 894                 return 0;
 895         }
 896         return EFSCORRUPTED;
 897 }