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_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_trans_priv.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_inode_item.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35
  36
  37 kmem_zone_t     *xfs_ili_zone;          /* inode log item zone */
  38
  39 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
  40 {
  41         return container_of(lip, struct xfs_inode_log_item, ili_item);
  42 }
  43
  44
  45 /*
  46  * This returns the number of iovecs needed to log the given inode item.
  47  *
  48  * We need one iovec for the inode log format structure, one for the
  49  * inode core, and possibly one for the inode data/extents/b-tree root
  50  * and one for the inode attribute data/extents/b-tree root.
  51  */
  52 STATIC uint
  53 xfs_inode_item_size(
  54         struct xfs_log_item     *lip)
  55 {
  56         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
  57         struct xfs_inode        *ip = iip->ili_inode;
  58         uint                    nvecs = 2;
  59
  60         switch (ip->i_d.di_format) {
  61         case XFS_DINODE_FMT_EXTENTS:
  62                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
  63                     ip->i_d.di_nextents > 0 &&
  64                     ip->i_df.if_bytes > 0)
  65                         nvecs++;
  66                 break;
  67
  68         case XFS_DINODE_FMT_BTREE:
  69                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
  70                     ip->i_df.if_broot_bytes > 0)
  71                         nvecs++;
  72                 break;
  73
  74         case XFS_DINODE_FMT_LOCAL:
  75                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
  76                     ip->i_df.if_bytes > 0)
  77                         nvecs++;
  78                 break;
  79
  80         case XFS_DINODE_FMT_DEV:
  81         case XFS_DINODE_FMT_UUID:
  82                 break;
  83
  84         default:
  85                 ASSERT(0);
  86                 break;
  87         }
  88
  89         if (!XFS_IFORK_Q(ip))
  90                 return nvecs;
  91
  92
  93         /*
  94          * Log any necessary attribute data.
  95          */
  96         switch (ip->i_d.di_aformat) {
  97         case XFS_DINODE_FMT_EXTENTS:
  98                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
  99                     ip->i_d.di_anextents > 0 &&
 100                     ip->i_afp->if_bytes > 0)
 101                         nvecs++;
 102                 break;
 103
 104         case XFS_DINODE_FMT_BTREE:
 105                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 106                     ip->i_afp->if_broot_bytes > 0)
 107                         nvecs++;
 108                 break;
 109
 110         case XFS_DINODE_FMT_LOCAL:
 111                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 112                     ip->i_afp->if_bytes > 0)
 113                         nvecs++;
 114                 break;
 115
 116         default:
 117                 ASSERT(0);
 118                 break;
 119         }
 120
 121         return nvecs;
 122 }
 123
 124 /*
 125  * xfs_inode_item_format_extents - convert in-core extents to on-disk form
 126  *
 127  * For either the data or attr fork in extent format, we need to endian convert
 128  * the in-core extent as we place them into the on-disk inode. In this case, we
 129  * need to do this conversion before we write the extents into the log. Because
 130  * we don't have the disk inode to write into here, we allocate a buffer and
 131  * format the extents into it via xfs_iextents_copy(). We free the buffer in
 132  * the unlock routine after the copy for the log has been made.
 133  *
 134  * In the case of the data fork, the in-core and on-disk fork sizes can be
 135  * different due to delayed allocation extents. We only log on-disk extents
 136  * here, so always use the physical fork size to determine the size of the
 137  * buffer we need to allocate.
 138  */
 139 STATIC void
 140 xfs_inode_item_format_extents(
 141         struct xfs_inode        *ip,
 142         struct xfs_log_iovec    *vecp,
 143         int                     whichfork,
 144         int                     type)
 145 {
 146         xfs_bmbt_rec_t          *ext_buffer;
 147
 148         ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
 149         if (whichfork == XFS_DATA_FORK)
 150                 ip->i_itemp->ili_extents_buf = ext_buffer;
 151         else
 152                 ip->i_itemp->ili_aextents_buf = ext_buffer;
 153
 154         vecp->i_addr = ext_buffer;
 155         vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
 156         vecp->i_type = type;
 157 }
 158
 159 /*
 160  * This is called to fill in the vector of log iovecs for the
 161  * given inode log item.  It fills the first item with an inode
 162  * log format structure, the second with the on-disk inode structure,
 163  * and a possible third and/or fourth with the inode data/extents/b-tree
 164  * root and inode attributes data/extents/b-tree root.
 165  */
 166 STATIC void
 167 xfs_inode_item_format(
 168         struct xfs_log_item     *lip,
 169         struct xfs_log_iovec    *vecp)
 170 {
 171         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 172         struct xfs_inode        *ip = iip->ili_inode;
 173         uint                    nvecs;
 174         size_t                  data_bytes;
 175         xfs_mount_t             *mp;
 176
 177         vecp->i_addr = &iip->ili_format;
 178         vecp->i_len  = sizeof(xfs_inode_log_format_t);
 179         vecp->i_type = XLOG_REG_TYPE_IFORMAT;
 180         vecp++;
 181         nvecs        = 1;
 182
 183         vecp->i_addr = &ip->i_d;
 184         vecp->i_len  = sizeof(struct xfs_icdinode);
 185         vecp->i_type = XLOG_REG_TYPE_ICORE;
 186         vecp++;
 187         nvecs++;
 188
 189         /*
 190          * If this is really an old format inode, then we need to
 191          * log it as such.  This means that we have to copy the link
 192          * count from the new field to the old.  We don't have to worry
 193          * about the new fields, because nothing trusts them as long as
 194          * the old inode version number is there.  If the superblock already
 195          * has a new version number, then we don't bother converting back.
 196          */
 197         mp = ip->i_mount;
 198         ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
 199         if (ip->i_d.di_version == 1) {
 200                 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
 201                         /*
 202                          * Convert it back.
 203                          */
 204                         ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
 205                         ip->i_d.di_onlink = ip->i_d.di_nlink;
 206                 } else {
 207                         /*
 208                          * The superblock version has already been bumped,
 209                          * so just make the conversion to the new inode
 210                          * format permanent.
 211                          */
 212                         ip->i_d.di_version = 2;
 213                         ip->i_d.di_onlink = 0;
 214                         memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 215                 }
 216         }
 217
 218         switch (ip->i_d.di_format) {
 219         case XFS_DINODE_FMT_EXTENTS:
 220                 iip->ili_fields &=
 221                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 222                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 223
 224                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 225                     ip->i_d.di_nextents > 0 &&
 226                     ip->i_df.if_bytes > 0) {
 227                         ASSERT(ip->i_df.if_u1.if_extents != NULL);
 228                         ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
 229                         ASSERT(iip->ili_extents_buf == NULL);
 230
 231 #ifdef XFS_NATIVE_HOST
 232                        if (ip->i_d.di_nextents == ip->i_df.if_bytes /
 233                                                (uint)sizeof(xfs_bmbt_rec_t)) {
 234                                 /*
 235                                  * There are no delayed allocation
 236                                  * extents, so just point to the
 237                                  * real extents array.
 238                                  */
 239                                 vecp->i_addr = ip->i_df.if_u1.if_extents;
 240                                 vecp->i_len = ip->i_df.if_bytes;
 241                                 vecp->i_type = XLOG_REG_TYPE_IEXT;
 242                         } else
 243 #endif
 244                         {
 245                                 xfs_inode_item_format_extents(ip, vecp,
 246                                         XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
 247                         }
 248                         ASSERT(vecp->i_len <= ip->i_df.if_bytes);
 249                         iip->ili_format.ilf_dsize = vecp->i_len;
 250                         vecp++;
 251                         nvecs++;
 252                 } else {
 253                         iip->ili_fields &= ~XFS_ILOG_DEXT;
 254                 }
 255                 break;
 256
 257         case XFS_DINODE_FMT_BTREE:
 258                 iip->ili_fields &=
 259                         ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
 260                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 261
 262                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 263                     ip->i_df.if_broot_bytes > 0) {
 264                         ASSERT(ip->i_df.if_broot != NULL);
 265                         vecp->i_addr = ip->i_df.if_broot;
 266                         vecp->i_len = ip->i_df.if_broot_bytes;
 267                         vecp->i_type = XLOG_REG_TYPE_IBROOT;
 268                         vecp++;
 269                         nvecs++;
 270                         iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
 271                 } else {
 272                         ASSERT(!(iip->ili_fields &
 273                                  XFS_ILOG_DBROOT));
 274 #ifdef XFS_TRANS_DEBUG
 275                         if (iip->ili_root_size > 0) {
 276                                 ASSERT(iip->ili_root_size ==
 277                                        ip->i_df.if_broot_bytes);
 278                                 ASSERT(memcmp(iip->ili_orig_root,
 279                                             ip->i_df.if_broot,
 280                                             iip->ili_root_size) == 0);
 281                         } else {
 282                                 ASSERT(ip->i_df.if_broot_bytes == 0);
 283                         }
 284 #endif
 285                         iip->ili_fields &= ~XFS_ILOG_DBROOT;
 286                 }
 287                 break;
 288
 289         case XFS_DINODE_FMT_LOCAL:
 290                 iip->ili_fields &=
 291                         ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
 292                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 293                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 294                     ip->i_df.if_bytes > 0) {
 295                         ASSERT(ip->i_df.if_u1.if_data != NULL);
 296                         ASSERT(ip->i_d.di_size > 0);
 297
 298                         vecp->i_addr = ip->i_df.if_u1.if_data;
 299                         /*
 300                          * Round i_bytes up to a word boundary.
 301                          * The underlying memory is guaranteed to
 302                          * to be there by xfs_idata_realloc().
 303                          */
 304                         data_bytes = roundup(ip->i_df.if_bytes, 4);
 305                         ASSERT((ip->i_df.if_real_bytes == 0) ||
 306                                (ip->i_df.if_real_bytes == data_bytes));
 307                         vecp->i_len = (int)data_bytes;
 308                         vecp->i_type = XLOG_REG_TYPE_ILOCAL;
 309                         vecp++;
 310                         nvecs++;
 311                         iip->ili_format.ilf_dsize = (unsigned)data_bytes;
 312                 } else {
 313                         iip->ili_fields &= ~XFS_ILOG_DDATA;
 314                 }
 315                 break;
 316
 317         case XFS_DINODE_FMT_DEV:
 318                 iip->ili_fields &=
 319                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 320                           XFS_ILOG_DEXT | XFS_ILOG_UUID);
 321                 if (iip->ili_fields & XFS_ILOG_DEV) {
 322                         iip->ili_format.ilf_u.ilfu_rdev =
 323                                 ip->i_df.if_u2.if_rdev;
 324                 }
 325                 break;
 326
 327         case XFS_DINODE_FMT_UUID:
 328                 iip->ili_fields &=
 329                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 330                           XFS_ILOG_DEXT | XFS_ILOG_DEV);
 331                 if (iip->ili_fields & XFS_ILOG_UUID) {
 332                         iip->ili_format.ilf_u.ilfu_uuid =
 333                                 ip->i_df.if_u2.if_uuid;
 334                 }
 335                 break;
 336
 337         default:
 338                 ASSERT(0);
 339                 break;
 340         }
 341
 342         /*
 343          * If there are no attributes associated with the file, then we're done.
 344          */
 345         if (!XFS_IFORK_Q(ip)) {
 346                 iip->ili_fields &=
 347                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
 348                 goto out;
 349         }
 350
 351         switch (ip->i_d.di_aformat) {
 352         case XFS_DINODE_FMT_EXTENTS:
 353                 iip->ili_fields &=
 354                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
 355
 356                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 357                     ip->i_d.di_anextents > 0 &&
 358                     ip->i_afp->if_bytes > 0) {
 359                         ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
 360                                 ip->i_d.di_anextents);
 361                         ASSERT(ip->i_afp->if_u1.if_extents != NULL);
 362 #ifdef XFS_NATIVE_HOST
 363                         /*
 364                          * There are not delayed allocation extents
 365                          * for attributes, so just point at the array.
 366                          */
 367                         vecp->i_addr = ip->i_afp->if_u1.if_extents;
 368                         vecp->i_len = ip->i_afp->if_bytes;
 369                         vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
 370 #else
 371                         ASSERT(iip->ili_aextents_buf == NULL);
 372                         xfs_inode_item_format_extents(ip, vecp,
 373                                         XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
 374 #endif
 375                         iip->ili_format.ilf_asize = vecp->i_len;
 376                         vecp++;
 377                         nvecs++;
 378                 } else {
 379                         iip->ili_fields &= ~XFS_ILOG_AEXT;
 380                 }
 381                 break;
 382
 383         case XFS_DINODE_FMT_BTREE:
 384                 iip->ili_fields &=
 385                         ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
 386
 387                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 388                     ip->i_afp->if_broot_bytes > 0) {
 389                         ASSERT(ip->i_afp->if_broot != NULL);
 390
 391                         vecp->i_addr = ip->i_afp->if_broot;
 392                         vecp->i_len = ip->i_afp->if_broot_bytes;
 393                         vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
 394                         vecp++;
 395                         nvecs++;
 396                         iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
 397                 } else {
 398                         iip->ili_fields &= ~XFS_ILOG_ABROOT;
 399                 }
 400                 break;
 401
 402         case XFS_DINODE_FMT_LOCAL:
 403                 iip->ili_fields &=
 404                         ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 405
 406                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 407                     ip->i_afp->if_bytes > 0) {
 408                         ASSERT(ip->i_afp->if_u1.if_data != NULL);
 409
 410                         vecp->i_addr = ip->i_afp->if_u1.if_data;
 411                         /*
 412                          * Round i_bytes up to a word boundary.
 413                          * The underlying memory is guaranteed to
 414                          * to be there by xfs_idata_realloc().
 415                          */
 416                         data_bytes = roundup(ip->i_afp->if_bytes, 4);
 417                         ASSERT((ip->i_afp->if_real_bytes == 0) ||
 418                                (ip->i_afp->if_real_bytes == data_bytes));
 419                         vecp->i_len = (int)data_bytes;
 420                         vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
 421                         vecp++;
 422                         nvecs++;
 423                         iip->ili_format.ilf_asize = (unsigned)data_bytes;
 424                 } else {
 425                         iip->ili_fields &= ~XFS_ILOG_ADATA;
 426                 }
 427                 break;
 428
 429         default:
 430                 ASSERT(0);
 431                 break;
 432         }
 433
 434 out:
 435         /*
 436          * Now update the log format that goes out to disk from the in-core
 437          * values.  We always write the inode core to make the arithmetic
 438          * games in recovery easier, which isn't a big deal as just about any
 439          * transaction would dirty it anyway.
 440          */
 441         iip->ili_format.ilf_fields = XFS_ILOG_CORE |
 442                 (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
 443         iip->ili_format.ilf_size = nvecs;
 444 }
 445
 446
 447 /*
 448  * This is called to pin the inode associated with the inode log
 449  * item in memory so it cannot be written out.
 450  */
 451 STATIC void
 452 xfs_inode_item_pin(
 453         struct xfs_log_item     *lip)
 454 {
 455         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 456
 457         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 458
 459         trace_xfs_inode_pin(ip, _RET_IP_);
 460         atomic_inc(&ip->i_pincount);
 461 }
 462
 463
 464 /*
 465  * This is called to unpin the inode associated with the inode log
 466  * item which was previously pinned with a call to xfs_inode_item_pin().
 467  *
 468  * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
 469  */
 470 STATIC void
 471 xfs_inode_item_unpin(
 472         struct xfs_log_item     *lip,
 473         int                     remove)
 474 {
 475         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 476
 477         trace_xfs_inode_unpin(ip, _RET_IP_);
 478         ASSERT(atomic_read(&ip->i_pincount) > 0);
 479         if (atomic_dec_and_test(&ip->i_pincount))
 480                 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 481 }
 482
 483 /*
 484  * This is called to attempt to lock the inode associated with this
 485  * inode log item, in preparation for the push routine which does the actual
 486  * iflush.  Don't sleep on the inode lock or the flush lock.
 487  *
 488  * If the flush lock is already held, indicating that the inode has
 489  * been or is in the process of being flushed, then (ideally) we'd like to
 490  * see if the inode's buffer is still incore, and if so give it a nudge.
 491  * We delay doing so until the pushbuf routine, though, to avoid holding
 492  * the AIL lock across a call to the blackhole which is the buffer cache.
 493  * Also we don't want to sleep in any device strategy routines, which can happen
 494  * if we do the subsequent bawrite in here.
 495  */
 496 STATIC uint
 497 xfs_inode_item_trylock(
 498         struct xfs_log_item     *lip)
 499 {
 500         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 501         struct xfs_inode        *ip = iip->ili_inode;
 502
 503         if (xfs_ipincount(ip) > 0)
 504                 return XFS_ITEM_PINNED;
 505
 506         if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
 507                 return XFS_ITEM_LOCKED;
 508
 509         /*
 510          * Re-check the pincount now that we stabilized the value by
 511          * taking the ilock.
 512          */
 513         if (xfs_ipincount(ip) > 0) {
 514                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 515                 return XFS_ITEM_PINNED;
 516         }
 517
 518         if (!xfs_iflock_nowait(ip)) {
 519                 /*
 520                  * inode has already been flushed to the backing buffer,
 521                  * leave it locked in shared mode, pushbuf routine will
 522                  * unlock it.
 523                  */
 524                 return XFS_ITEM_PUSHBUF;
 525         }
 526
 527         /* Stale items should force out the iclog */
 528         if (ip->i_flags & XFS_ISTALE) {
 529                 xfs_ifunlock(ip);
 530                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 531                 return XFS_ITEM_PINNED;
 532         }
 533
 534 #ifdef DEBUG
 535         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 536                 ASSERT(iip->ili_fields != 0);
 537                 ASSERT(iip->ili_logged == 0);
 538                 ASSERT(lip->li_flags & XFS_LI_IN_AIL);
 539         }
 540 #endif
 541         return XFS_ITEM_SUCCESS;
 542 }
 543
 544 /*
 545  * Unlock the inode associated with the inode log item.
 546  * Clear the fields of the inode and inode log item that
 547  * are specific to the current transaction.  If the
 548  * hold flags is set, do not unlock the inode.
 549  */
 550 STATIC void
 551 xfs_inode_item_unlock(
 552         struct xfs_log_item     *lip)
 553 {
 554         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 555         struct xfs_inode        *ip = iip->ili_inode;
 556         unsigned short          lock_flags;
 557
 558         ASSERT(ip->i_itemp != NULL);
 559         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 560
 561         /*
 562          * If the inode needed a separate buffer with which to log
 563          * its extents, then free it now.
 564          */
 565         if (iip->ili_extents_buf != NULL) {
 566                 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
 567                 ASSERT(ip->i_d.di_nextents > 0);
 568                 ASSERT(iip->ili_fields & XFS_ILOG_DEXT);
 569                 ASSERT(ip->i_df.if_bytes > 0);
 570                 kmem_free(iip->ili_extents_buf);
 571                 iip->ili_extents_buf = NULL;
 572         }
 573         if (iip->ili_aextents_buf != NULL) {
 574                 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
 575                 ASSERT(ip->i_d.di_anextents > 0);
 576                 ASSERT(iip->ili_fields & XFS_ILOG_AEXT);
 577                 ASSERT(ip->i_afp->if_bytes > 0);
 578                 kmem_free(iip->ili_aextents_buf);
 579                 iip->ili_aextents_buf = NULL;
 580         }
 581
 582         lock_flags = iip->ili_lock_flags;
 583         iip->ili_lock_flags = 0;
 584         if (lock_flags)
 585                 xfs_iunlock(ip, lock_flags);
 586 }
 587
 588 /*
 589  * This is called to find out where the oldest active copy of the inode log
 590  * item in the on disk log resides now that the last log write of it completed
 591  * at the given lsn.  Since we always re-log all dirty data in an inode, the
 592  * latest copy in the on disk log is the only one that matters.  Therefore,
 593  * simply return the given lsn.
 594  *
 595  * If the inode has been marked stale because the cluster is being freed, we
 596  * don't want to (re-)insert this inode into the AIL. There is a race condition
 597  * where the cluster buffer may be unpinned before the inode is inserted into
 598  * the AIL during transaction committed processing. If the buffer is unpinned
 599  * before the inode item has been committed and inserted, then it is possible
 600  * for the buffer to be written and IO completes before the inode is inserted
 601  * into the AIL. In that case, we'd be inserting a clean, stale inode into the
 602  * AIL which will never get removed. It will, however, get reclaimed which
 603  * triggers an assert in xfs_inode_free() complaining about freein an inode
 604  * still in the AIL.
 605  *
 606  * To avoid this, just unpin the inode directly and return a LSN of -1 so the
 607  * transaction committed code knows that it does not need to do any further
 608  * processing on the item.
 609  */
 610 STATIC xfs_lsn_t
 611 xfs_inode_item_committed(
 612         struct xfs_log_item     *lip,
 613         xfs_lsn_t               lsn)
 614 {
 615         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 616         struct xfs_inode        *ip = iip->ili_inode;
 617
 618         if (xfs_iflags_test(ip, XFS_ISTALE)) {
 619                 xfs_inode_item_unpin(lip, 0);
 620                 return -1;
 621         }
 622         return lsn;
 623 }
 624
 625 /*
 626  * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
 627  * failed to get the inode flush lock but did get the inode locked SHARED.
 628  * Here we're trying to see if the inode buffer is incore, and if so whether it's
 629  * marked delayed write. If that's the case, we'll promote it and that will
 630  * allow the caller to write the buffer by triggering the xfsbufd to run.
 631  */
 632 STATIC bool
 633 xfs_inode_item_pushbuf(
 634         struct xfs_log_item     *lip)
 635 {
 636         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 637         struct xfs_inode        *ip = iip->ili_inode;
 638         struct xfs_buf          *bp;
 639         bool                    ret = true;
 640
 641         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
 642
 643         /*
 644          * If a flush is not in progress anymore, chances are that the
 645          * inode was taken off the AIL. So, just get out.
 646          */
 647         if (!xfs_isiflocked(ip) ||
 648             !(lip->li_flags & XFS_LI_IN_AIL)) {
 649                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 650                 return true;
 651         }
 652
 653         bp = xfs_incore(ip->i_mount->m_ddev_targp, iip->ili_format.ilf_blkno,
 654                         iip->ili_format.ilf_len, XBF_TRYLOCK);
 655
 656         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 657         if (!bp)
 658                 return true;
 659         if (XFS_BUF_ISDELAYWRITE(bp))
 660                 xfs_buf_delwri_promote(bp);
 661         if (xfs_buf_ispinned(bp))
 662                 ret = false;
 663         xfs_buf_relse(bp);
 664         return ret;
 665 }
 666
 667 /*
 668  * This is called to asynchronously write the inode associated with this
 669  * inode log item out to disk. The inode will already have been locked by
 670  * a successful call to xfs_inode_item_trylock().
 671  */
 672 STATIC void
 673 xfs_inode_item_push(
 674         struct xfs_log_item     *lip)
 675 {
 676         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 677         struct xfs_inode        *ip = iip->ili_inode;
 678         struct xfs_buf          *bp = NULL;
 679         int                     error;
 680
 681         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
 682         ASSERT(xfs_isiflocked(ip));
 683
 684         /*
 685          * Since we were able to lock the inode's flush lock and
 686          * we found it on the AIL, the inode must be dirty.  This
 687          * is because the inode is removed from the AIL while still
 688          * holding the flush lock in xfs_iflush_done().  Thus, if
 689          * we found it in the AIL and were able to obtain the flush
 690          * lock without sleeping, then there must not have been
 691          * anyone in the process of flushing the inode.
 692          */
 693         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || iip->ili_fields != 0);
 694
 695         /*
 696          * Push the inode to it's backing buffer. This will not remove the
 697          * inode from the AIL - a further push will be required to trigger a
 698          * buffer push. However, this allows all the dirty inodes to be pushed
 699          * to the buffer before it is pushed to disk. The buffer IO completion
 700          * will pull the inode from the AIL, mark it clean and unlock the flush
 701          * lock.
 702          */
 703         error = xfs_iflush(ip, &bp);
 704         if (!error) {
 705                 xfs_buf_delwri_queue(bp);
 706                 xfs_buf_relse(bp);
 707         }
 708         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 709 }
 710
 711 /*
 712  * XXX rcc - this one really has to do something.  Probably needs
 713  * to stamp in a new field in the incore inode.
 714  */
 715 STATIC void
 716 xfs_inode_item_committing(
 717         struct xfs_log_item     *lip,
 718         xfs_lsn_t               lsn)
 719 {
 720         INODE_ITEM(lip)->ili_last_lsn = lsn;
 721 }
 722
 723 /*
 724  * This is the ops vector shared by all buf log items.
 725  */
 726 static const struct xfs_item_ops xfs_inode_item_ops = {
 727         .iop_size       = xfs_inode_item_size,
 728         .iop_format     = xfs_inode_item_format,
 729         .iop_pin        = xfs_inode_item_pin,
 730         .iop_unpin      = xfs_inode_item_unpin,
 731         .iop_trylock    = xfs_inode_item_trylock,
 732         .iop_unlock     = xfs_inode_item_unlock,
 733         .iop_committed  = xfs_inode_item_committed,
 734         .iop_push       = xfs_inode_item_push,
 735         .iop_pushbuf    = xfs_inode_item_pushbuf,
 736         .iop_committing = xfs_inode_item_committing
 737 };
 738
 739
 740 /*
 741  * Initialize the inode log item for a newly allocated (in-core) inode.
 742  */
 743 void
 744 xfs_inode_item_init(
 745         struct xfs_inode        *ip,
 746         struct xfs_mount        *mp)
 747 {
 748         struct xfs_inode_log_item *iip;
 749
 750         ASSERT(ip->i_itemp == NULL);
 751         iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
 752
 753         iip->ili_inode = ip;
 754         xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
 755                                                 &xfs_inode_item_ops);
 756         iip->ili_format.ilf_type = XFS_LI_INODE;
 757         iip->ili_format.ilf_ino = ip->i_ino;
 758         iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
 759         iip->ili_format.ilf_len = ip->i_imap.im_len;
 760         iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
 761 }
 762
 763 /*
 764  * Free the inode log item and any memory hanging off of it.
 765  */
 766 void
 767 xfs_inode_item_destroy(
 768         xfs_inode_t     *ip)
 769 {
 770 #ifdef XFS_TRANS_DEBUG
 771         if (ip->i_itemp->ili_root_size != 0) {
 772                 kmem_free(ip->i_itemp->ili_orig_root);
 773         }
 774 #endif
 775         kmem_zone_free(xfs_ili_zone, ip->i_itemp);
 776 }
 777
 778
 779 /*
 780  * This is the inode flushing I/O completion routine.  It is called
 781  * from interrupt level when the buffer containing the inode is
 782  * flushed to disk.  It is responsible for removing the inode item
 783  * from the AIL if it has not been re-logged, and unlocking the inode's
 784  * flush lock.
 785  *
 786  * To reduce AIL lock traffic as much as possible, we scan the buffer log item
 787  * list for other inodes that will run this function. We remove them from the
 788  * buffer list so we can process all the inode IO completions in one AIL lock
 789  * traversal.
 790  */
 791 void
 792 xfs_iflush_done(
 793         struct xfs_buf          *bp,
 794         struct xfs_log_item     *lip)
 795 {
 796         struct xfs_inode_log_item *iip;
 797         struct xfs_log_item     *blip;
 798         struct xfs_log_item     *next;
 799         struct xfs_log_item     *prev;
 800         struct xfs_ail          *ailp = lip->li_ailp;
 801         int                     need_ail = 0;
 802
 803         /*
 804          * Scan the buffer IO completions for other inodes being completed and
 805          * attach them to the current inode log item.
 806          */
 807         blip = bp->b_fspriv;
 808         prev = NULL;
 809         while (blip != NULL) {
 810                 if (lip->li_cb != xfs_iflush_done) {
 811                         prev = blip;
 812                         blip = blip->li_bio_list;
 813                         continue;
 814                 }
 815
 816                 /* remove from list */
 817                 next = blip->li_bio_list;
 818                 if (!prev) {
 819                         bp->b_fspriv = next;
 820                 } else {
 821                         prev->li_bio_list = next;
 822                 }
 823
 824                 /* add to current list */
 825                 blip->li_bio_list = lip->li_bio_list;
 826                 lip->li_bio_list = blip;
 827
 828                 /*
 829                  * while we have the item, do the unlocked check for needing
 830                  * the AIL lock.
 831                  */
 832                 iip = INODE_ITEM(blip);
 833                 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
 834                         need_ail++;
 835
 836                 blip = next;
 837         }
 838
 839         /* make sure we capture the state of the initial inode. */
 840         iip = INODE_ITEM(lip);
 841         if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
 842                 need_ail++;
 843
 844         /*
 845          * We only want to pull the item from the AIL if it is
 846          * actually there and its location in the log has not
 847          * changed since we started the flush.  Thus, we only bother
 848          * if the ili_logged flag is set and the inode's lsn has not
 849          * changed.  First we check the lsn outside
 850          * the lock since it's cheaper, and then we recheck while
 851          * holding the lock before removing the inode from the AIL.
 852          */
 853         if (need_ail) {
 854                 struct xfs_log_item *log_items[need_ail];
 855                 int i = 0;
 856                 spin_lock(&ailp->xa_lock);
 857                 for (blip = lip; blip; blip = blip->li_bio_list) {
 858                         iip = INODE_ITEM(blip);
 859                         if (iip->ili_logged &&
 860                             blip->li_lsn == iip->ili_flush_lsn) {
 861                                 log_items[i++] = blip;
 862                         }
 863                         ASSERT(i <= need_ail);
 864                 }
 865                 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
 866                 xfs_trans_ail_delete_bulk(ailp, log_items, i);
 867         }
 868
 869
 870         /*
 871          * clean up and unlock the flush lock now we are done. We can clear the
 872          * ili_last_fields bits now that we know that the data corresponding to
 873          * them is safely on disk.
 874          */
 875         for (blip = lip; blip; blip = next) {
 876                 next = blip->li_bio_list;
 877                 blip->li_bio_list = NULL;
 878
 879                 iip = INODE_ITEM(blip);
 880                 iip->ili_logged = 0;
 881                 iip->ili_last_fields = 0;
 882                 xfs_ifunlock(iip->ili_inode);
 883         }
 884 }
 885
 886 /*
 887  * This is the inode flushing abort routine.  It is called
 888  * from xfs_iflush when the filesystem is shutting down to clean
 889  * up the inode state.
 890  * It is responsible for removing the inode item
 891  * from the AIL if it has not been re-logged, and unlocking the inode's
 892  * flush lock.
 893  */
 894 void
 895 xfs_iflush_abort(
 896         xfs_inode_t             *ip)
 897 {
 898         xfs_inode_log_item_t    *iip = ip->i_itemp;
 899
 900         if (iip) {
 901                 struct xfs_ail  *ailp = iip->ili_item.li_ailp;
 902                 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 903                         spin_lock(&ailp->xa_lock);
 904                         if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 905                                 /* xfs_trans_ail_delete() drops the AIL lock. */
 906                                 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip);
 907                         } else
 908                                 spin_unlock(&ailp->xa_lock);
 909                 }
 910                 iip->ili_logged = 0;
 911                 /*
 912                  * Clear the ili_last_fields bits now that we know that the
 913                  * data corresponding to them is safely on disk.
 914                  */
 915                 iip->ili_last_fields = 0;
 916                 /*
 917                  * Clear the inode logging fields so no more flushes are
 918                  * attempted.
 919                  */
 920                 iip->ili_fields = 0;
 921         }
 922         /*
 923          * Release the inode's flush lock since we're done with it.
 924          */
 925         xfs_ifunlock(ip);
 926 }
 927
 928 void
 929 xfs_istale_done(
 930         struct xfs_buf          *bp,
 931         struct xfs_log_item     *lip)
 932 {
 933         xfs_iflush_abort(INODE_ITEM(lip)->ili_inode);
 934 }
 935
 936 /*
 937  * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
 938  * (which can have different field alignments) to the native version
 939  */
 940 int
 941 xfs_inode_item_format_convert(
 942         xfs_log_iovec_t         *buf,
 943         xfs_inode_log_format_t  *in_f)
 944 {
 945         if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
 946                 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
 947
 948                 in_f->ilf_type = in_f32->ilf_type;
 949                 in_f->ilf_size = in_f32->ilf_size;
 950                 in_f->ilf_fields = in_f32->ilf_fields;
 951                 in_f->ilf_asize = in_f32->ilf_asize;
 952                 in_f->ilf_dsize = in_f32->ilf_dsize;
 953                 in_f->ilf_ino = in_f32->ilf_ino;
 954                 /* copy biggest field of ilf_u */
 955                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 956                        in_f32->ilf_u.ilfu_uuid.__u_bits,
 957                        sizeof(uuid_t));
 958                 in_f->ilf_blkno = in_f32->ilf_blkno;
 959                 in_f->ilf_len = in_f32->ilf_len;
 960                 in_f->ilf_boffset = in_f32->ilf_boffset;
 961                 return 0;
 962         } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
 963                 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
 964
 965                 in_f->ilf_type = in_f64->ilf_type;
 966                 in_f->ilf_size = in_f64->ilf_size;
 967                 in_f->ilf_fields = in_f64->ilf_fields;
 968                 in_f->ilf_asize = in_f64->ilf_asize;
 969                 in_f->ilf_dsize = in_f64->ilf_dsize;
 970                 in_f->ilf_ino = in_f64->ilf_ino;
 971                 /* copy biggest field of ilf_u */
 972                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 973                        in_f64->ilf_u.ilfu_uuid.__u_bits,
 974                        sizeof(uuid_t));
 975                 in_f->ilf_blkno = in_f64->ilf_blkno;
 976                 in_f->ilf_len = in_f64->ilf_len;
 977                 in_f->ilf_boffset = in_f64->ilf_boffset;
 978                 return 0;
 979         }
 980         return EFSCORRUPTED;
 981 }