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