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