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_ag.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_inode.h"
  27 #include "xfs_trans.h"
  28 #include "xfs_inode_item.h"
  29 #include "xfs_error.h"
  30 #include "xfs_trace.h"
  31 #include "xfs_trans_priv.h"
  32 #include "xfs_dinode.h"
  33 #include "xfs_log.h"
  34
  35
  36 kmem_zone_t     *xfs_ili_zone;          /* inode log item zone */
  37
  38 static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
  39 {
  40         return container_of(lip, struct xfs_inode_log_item, ili_item);
  41 }
  42
  43 STATIC void
  44 xfs_inode_item_data_fork_size(
  45         struct xfs_inode_log_item *iip,
  46         int                     *nvecs,
  47         int                     *nbytes)
  48 {
  49         struct xfs_inode        *ip = iip->ili_inode;
  50
  51         switch (ip->i_d.di_format) {
  52         case XFS_DINODE_FMT_EXTENTS:
  53                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
  54                     ip->i_d.di_nextents > 0 &&
  55                     ip->i_df.if_bytes > 0) {
  56                         /* worst case, doesn't subtract delalloc extents */
  57                         *nbytes += XFS_IFORK_DSIZE(ip);
  58                         *nvecs += 1;
  59                 }
  60                 break;
  61         case XFS_DINODE_FMT_BTREE:
  62                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
  63                     ip->i_df.if_broot_bytes > 0) {
  64                         *nbytes += ip->i_df.if_broot_bytes;
  65                         *nvecs += 1;
  66                 }
  67                 break;
  68         case XFS_DINODE_FMT_LOCAL:
  69                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
  70                     ip->i_df.if_bytes > 0) {
  71                         *nbytes += roundup(ip->i_df.if_bytes, 4);
  72                         *nvecs += 1;
  73                 }
  74                 break;
  75
  76         case XFS_DINODE_FMT_DEV:
  77         case XFS_DINODE_FMT_UUID:
  78                 break;
  79         default:
  80                 ASSERT(0);
  81                 break;
  82         }
  83 }
  84
  85 STATIC void
  86 xfs_inode_item_attr_fork_size(
  87         struct xfs_inode_log_item *iip,
  88         int                     *nvecs,
  89         int                     *nbytes)
  90 {
  91         struct xfs_inode        *ip = iip->ili_inode;
  92
  93         switch (ip->i_d.di_aformat) {
  94         case XFS_DINODE_FMT_EXTENTS:
  95                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
  96                     ip->i_d.di_anextents > 0 &&
  97                     ip->i_afp->if_bytes > 0) {
  98                         /* worst case, doesn't subtract unused space */
  99                         *nbytes += XFS_IFORK_ASIZE(ip);
 100                         *nvecs += 1;
 101                 }
 102                 break;
 103         case XFS_DINODE_FMT_BTREE:
 104                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 105                     ip->i_afp->if_broot_bytes > 0) {
 106                         *nbytes += ip->i_afp->if_broot_bytes;
 107                         *nvecs += 1;
 108                 }
 109                 break;
 110         case XFS_DINODE_FMT_LOCAL:
 111                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 112                     ip->i_afp->if_bytes > 0) {
 113                         *nbytes += roundup(ip->i_afp->if_bytes, 4);
 114                         *nvecs += 1;
 115                 }
 116                 break;
 117         default:
 118                 ASSERT(0);
 119                 break;
 120         }
 121 }
 122
 123 /*
 124  * This returns the number of iovecs needed to log the given inode item.
 125  *
 126  * We need one iovec for the inode log format structure, one for the
 127  * inode core, and possibly one for the inode data/extents/b-tree root
 128  * and one for the inode attribute data/extents/b-tree root.
 129  */
 130 STATIC void
 131 xfs_inode_item_size(
 132         struct xfs_log_item     *lip,
 133         int                     *nvecs,
 134         int                     *nbytes)
 135 {
 136         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 137         struct xfs_inode        *ip = iip->ili_inode;
 138
 139         *nvecs += 2;
 140         *nbytes += sizeof(struct xfs_inode_log_format) +
 141                    xfs_icdinode_size(ip->i_d.di_version);
 142
 143         xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
 144         if (XFS_IFORK_Q(ip))
 145                 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
 146 }
 147
 148 /*
 149  * If this is a v1 format inode, then we need to log it as such.  This means
 150  * that we have to copy the link count from the new field to the old.  We
 151  * don't have to worry about the new fields, because nothing trusts them as
 152  * long as the old inode version number is there.
 153  */
 154 STATIC void
 155 xfs_inode_item_format_v1_inode(
 156         struct xfs_inode        *ip)
 157 {
 158         if (!xfs_sb_version_hasnlink(&ip->i_mount->m_sb)) {
 159                 /*
 160                  * Convert it back.
 161                  */
 162                 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
 163                 ip->i_d.di_onlink = ip->i_d.di_nlink;
 164         } else {
 165                 /*
 166                  * The superblock version has already been bumped,
 167                  * so just make the conversion to the new inode
 168                  * format permanent.
 169                  */
 170                 ip->i_d.di_version = 2;
 171                 ip->i_d.di_onlink = 0;
 172                 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 173         }
 174 }
 175
 176 STATIC void
 177 xfs_inode_item_format_data_fork(
 178         struct xfs_inode_log_item *iip,
 179         struct xfs_inode_log_format *ilf,
 180         struct xfs_log_vec      *lv,
 181         struct xfs_log_iovec    **vecp)
 182 {
 183         struct xfs_inode        *ip = iip->ili_inode;
 184         size_t                  data_bytes;
 185
 186         switch (ip->i_d.di_format) {
 187         case XFS_DINODE_FMT_EXTENTS:
 188                 iip->ili_fields &=
 189                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 190                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 191
 192                 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 193                     ip->i_d.di_nextents > 0 &&
 194                     ip->i_df.if_bytes > 0) {
 195                         struct xfs_bmbt_rec *p;
 196
 197                         ASSERT(ip->i_df.if_u1.if_extents != NULL);
 198                         ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
 199
 200                         p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
 201                         data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
 202                         xlog_finish_iovec(lv, *vecp, data_bytes);
 203
 204                         ASSERT(data_bytes <= ip->i_df.if_bytes);
 205
 206                         ilf->ilf_dsize = data_bytes;
 207                         ilf->ilf_size++;
 208                 } else {
 209                         iip->ili_fields &= ~XFS_ILOG_DEXT;
 210                 }
 211                 break;
 212         case XFS_DINODE_FMT_BTREE:
 213                 iip->ili_fields &=
 214                         ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
 215                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 216
 217                 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 218                     ip->i_df.if_broot_bytes > 0) {
 219                         ASSERT(ip->i_df.if_broot != NULL);
 220                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
 221                                         ip->i_df.if_broot,
 222                                         ip->i_df.if_broot_bytes);
 223                         ilf->ilf_dsize = ip->i_df.if_broot_bytes;
 224                         ilf->ilf_size++;
 225                 } else {
 226                         ASSERT(!(iip->ili_fields &
 227                                  XFS_ILOG_DBROOT));
 228                         iip->ili_fields &= ~XFS_ILOG_DBROOT;
 229                 }
 230                 break;
 231         case XFS_DINODE_FMT_LOCAL:
 232                 iip->ili_fields &=
 233                         ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
 234                           XFS_ILOG_DEV | XFS_ILOG_UUID);
 235                 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 236                     ip->i_df.if_bytes > 0) {
 237                         /*
 238                          * Round i_bytes up to a word boundary.
 239                          * The underlying memory is guaranteed to
 240                          * to be there by xfs_idata_realloc().
 241                          */
 242                         data_bytes = roundup(ip->i_df.if_bytes, 4);
 243                         ASSERT(ip->i_df.if_real_bytes == 0 ||
 244                                ip->i_df.if_real_bytes == data_bytes);
 245                         ASSERT(ip->i_df.if_u1.if_data != NULL);
 246                         ASSERT(ip->i_d.di_size > 0);
 247                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
 248                                         ip->i_df.if_u1.if_data, data_bytes);
 249                         ilf->ilf_dsize = (unsigned)data_bytes;
 250                         ilf->ilf_size++;
 251                 } else {
 252                         iip->ili_fields &= ~XFS_ILOG_DDATA;
 253                 }
 254                 break;
 255         case XFS_DINODE_FMT_DEV:
 256                 iip->ili_fields &=
 257                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 258                           XFS_ILOG_DEXT | XFS_ILOG_UUID);
 259                 if (iip->ili_fields & XFS_ILOG_DEV)
 260                         ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
 261                 break;
 262         case XFS_DINODE_FMT_UUID:
 263                 iip->ili_fields &=
 264                         ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 265                           XFS_ILOG_DEXT | XFS_ILOG_DEV);
 266                 if (iip->ili_fields & XFS_ILOG_UUID)
 267                         ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
 268                 break;
 269         default:
 270                 ASSERT(0);
 271                 break;
 272         }
 273 }
 274
 275 STATIC void
 276 xfs_inode_item_format_attr_fork(
 277         struct xfs_inode_log_item *iip,
 278         struct xfs_inode_log_format *ilf,
 279         struct xfs_log_vec      *lv,
 280         struct xfs_log_iovec    **vecp)
 281 {
 282         struct xfs_inode        *ip = iip->ili_inode;
 283         size_t                  data_bytes;
 284
 285         switch (ip->i_d.di_aformat) {
 286         case XFS_DINODE_FMT_EXTENTS:
 287                 iip->ili_fields &=
 288                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
 289
 290                 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 291                     ip->i_d.di_anextents > 0 &&
 292                     ip->i_afp->if_bytes > 0) {
 293                         struct xfs_bmbt_rec *p;
 294
 295                         ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
 296                                 ip->i_d.di_anextents);
 297                         ASSERT(ip->i_afp->if_u1.if_extents != NULL);
 298
 299                         p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
 300                         data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
 301                         xlog_finish_iovec(lv, *vecp, data_bytes);
 302
 303                         ilf->ilf_asize = data_bytes;
 304                         ilf->ilf_size++;
 305                 } else {
 306                         iip->ili_fields &= ~XFS_ILOG_AEXT;
 307                 }
 308                 break;
 309         case XFS_DINODE_FMT_BTREE:
 310                 iip->ili_fields &=
 311                         ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
 312
 313                 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
 314                     ip->i_afp->if_broot_bytes > 0) {
 315                         ASSERT(ip->i_afp->if_broot != NULL);
 316
 317                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
 318                                         ip->i_afp->if_broot,
 319                                         ip->i_afp->if_broot_bytes);
 320                         ilf->ilf_asize = ip->i_afp->if_broot_bytes;
 321                         ilf->ilf_size++;
 322                 } else {
 323                         iip->ili_fields &= ~XFS_ILOG_ABROOT;
 324                 }
 325                 break;
 326         case XFS_DINODE_FMT_LOCAL:
 327                 iip->ili_fields &=
 328                         ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
 329
 330                 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
 331                     ip->i_afp->if_bytes > 0) {
 332                         /*
 333                          * Round i_bytes up to a word boundary.
 334                          * The underlying memory is guaranteed to
 335                          * to be there by xfs_idata_realloc().
 336                          */
 337                         data_bytes = roundup(ip->i_afp->if_bytes, 4);
 338                         ASSERT(ip->i_afp->if_real_bytes == 0 ||
 339                                ip->i_afp->if_real_bytes == data_bytes);
 340                         ASSERT(ip->i_afp->if_u1.if_data != NULL);
 341                         xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
 342                                         ip->i_afp->if_u1.if_data,
 343                                         data_bytes);
 344                         ilf->ilf_asize = (unsigned)data_bytes;
 345                         ilf->ilf_size++;
 346                 } else {
 347                         iip->ili_fields &= ~XFS_ILOG_ADATA;
 348                 }
 349                 break;
 350         default:
 351                 ASSERT(0);
 352                 break;
 353         }
 354 }
 355
 356 /*
 357  * This is called to fill in the vector of log iovecs for the given inode
 358  * log item.  It fills the first item with an inode log format structure,
 359  * the second with the on-disk inode structure, and a possible third and/or
 360  * fourth with the inode data/extents/b-tree root and inode attributes
 361  * data/extents/b-tree root.
 362  */
 363 STATIC void
 364 xfs_inode_item_format(
 365         struct xfs_log_item     *lip,
 366         struct xfs_log_vec      *lv)
 367 {
 368         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 369         struct xfs_inode        *ip = iip->ili_inode;
 370         struct xfs_inode_log_format *ilf;
 371         struct xfs_log_iovec    *vecp = NULL;
 372
 373         ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
 374         ilf->ilf_type = XFS_LI_INODE;
 375         ilf->ilf_ino = ip->i_ino;
 376         ilf->ilf_blkno = ip->i_imap.im_blkno;
 377         ilf->ilf_len = ip->i_imap.im_len;
 378         ilf->ilf_boffset = ip->i_imap.im_boffset;
 379         ilf->ilf_fields = XFS_ILOG_CORE;
 380         ilf->ilf_size = 2; /* format + core */
 381         xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
 382
 383         if (ip->i_d.di_version == 1)
 384                 xfs_inode_item_format_v1_inode(ip);
 385         xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICORE,
 386                         &ip->i_d,
 387                         xfs_icdinode_size(ip->i_d.di_version));
 388
 389         xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
 390         if (XFS_IFORK_Q(ip)) {
 391                 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
 392         } else {
 393                 iip->ili_fields &=
 394                         ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
 395         }
 396
 397         /* update the format with the exact fields we actually logged */
 398         ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
 399 }
 400
 401 /*
 402  * This is called to pin the inode associated with the inode log
 403  * item in memory so it cannot be written out.
 404  */
 405 STATIC void
 406 xfs_inode_item_pin(
 407         struct xfs_log_item     *lip)
 408 {
 409         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 410
 411         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 412
 413         trace_xfs_inode_pin(ip, _RET_IP_);
 414         atomic_inc(&ip->i_pincount);
 415 }
 416
 417
 418 /*
 419  * This is called to unpin the inode associated with the inode log
 420  * item which was previously pinned with a call to xfs_inode_item_pin().
 421  *
 422  * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
 423  */
 424 STATIC void
 425 xfs_inode_item_unpin(
 426         struct xfs_log_item     *lip,
 427         int                     remove)
 428 {
 429         struct xfs_inode        *ip = INODE_ITEM(lip)->ili_inode;
 430
 431         trace_xfs_inode_unpin(ip, _RET_IP_);
 432         ASSERT(atomic_read(&ip->i_pincount) > 0);
 433         if (atomic_dec_and_test(&ip->i_pincount))
 434                 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 435 }
 436
 437 STATIC uint
 438 xfs_inode_item_push(
 439         struct xfs_log_item     *lip,
 440         struct list_head        *buffer_list)
 441 {
 442         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 443         struct xfs_inode        *ip = iip->ili_inode;
 444         struct xfs_buf          *bp = NULL;
 445         uint                    rval = XFS_ITEM_SUCCESS;
 446         int                     error;
 447
 448         if (xfs_ipincount(ip) > 0)
 449                 return XFS_ITEM_PINNED;
 450
 451         if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
 452                 return XFS_ITEM_LOCKED;
 453
 454         /*
 455          * Re-check the pincount now that we stabilized the value by
 456          * taking the ilock.
 457          */
 458         if (xfs_ipincount(ip) > 0) {
 459                 rval = XFS_ITEM_PINNED;
 460                 goto out_unlock;
 461         }
 462
 463         /*
 464          * Stale inode items should force out the iclog.
 465          */
 466         if (ip->i_flags & XFS_ISTALE) {
 467                 rval = XFS_ITEM_PINNED;
 468                 goto out_unlock;
 469         }
 470
 471         /*
 472          * Someone else is already flushing the inode.  Nothing we can do
 473          * here but wait for the flush to finish and remove the item from
 474          * the AIL.
 475          */
 476         if (!xfs_iflock_nowait(ip)) {
 477                 rval = XFS_ITEM_FLUSHING;
 478                 goto out_unlock;
 479         }
 480
 481         ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
 482         ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
 483
 484         spin_unlock(&lip->li_ailp->xa_lock);
 485
 486         error = xfs_iflush(ip, &bp);
 487         if (!error) {
 488                 if (!xfs_buf_delwri_queue(bp, buffer_list))
 489                         rval = XFS_ITEM_FLUSHING;
 490                 xfs_buf_relse(bp);
 491         }
 492
 493         spin_lock(&lip->li_ailp->xa_lock);
 494 out_unlock:
 495         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 496         return rval;
 497 }
 498
 499 /*
 500  * Unlock the inode associated with the inode log item.
 501  * Clear the fields of the inode and inode log item that
 502  * are specific to the current transaction.  If the
 503  * hold flags is set, do not unlock the inode.
 504  */
 505 STATIC void
 506 xfs_inode_item_unlock(
 507         struct xfs_log_item     *lip)
 508 {
 509         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 510         struct xfs_inode        *ip = iip->ili_inode;
 511         unsigned short          lock_flags;
 512
 513         ASSERT(ip->i_itemp != NULL);
 514         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 515
 516         lock_flags = iip->ili_lock_flags;
 517         iip->ili_lock_flags = 0;
 518         if (lock_flags)
 519                 xfs_iunlock(ip, lock_flags);
 520 }
 521
 522 /*
 523  * This is called to find out where the oldest active copy of the inode log
 524  * item in the on disk log resides now that the last log write of it completed
 525  * at the given lsn.  Since we always re-log all dirty data in an inode, the
 526  * latest copy in the on disk log is the only one that matters.  Therefore,
 527  * simply return the given lsn.
 528  *
 529  * If the inode has been marked stale because the cluster is being freed, we
 530  * don't want to (re-)insert this inode into the AIL. There is a race condition
 531  * where the cluster buffer may be unpinned before the inode is inserted into
 532  * the AIL during transaction committed processing. If the buffer is unpinned
 533  * before the inode item has been committed and inserted, then it is possible
 534  * for the buffer to be written and IO completes before the inode is inserted
 535  * into the AIL. In that case, we'd be inserting a clean, stale inode into the
 536  * AIL which will never get removed. It will, however, get reclaimed which
 537  * triggers an assert in xfs_inode_free() complaining about freein an inode
 538  * still in the AIL.
 539  *
 540  * To avoid this, just unpin the inode directly and return a LSN of -1 so the
 541  * transaction committed code knows that it does not need to do any further
 542  * processing on the item.
 543  */
 544 STATIC xfs_lsn_t
 545 xfs_inode_item_committed(
 546         struct xfs_log_item     *lip,
 547         xfs_lsn_t               lsn)
 548 {
 549         struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 550         struct xfs_inode        *ip = iip->ili_inode;
 551
 552         if (xfs_iflags_test(ip, XFS_ISTALE)) {
 553                 xfs_inode_item_unpin(lip, 0);
 554                 return -1;
 555         }
 556         return lsn;
 557 }
 558
 559 /*
 560  * XXX rcc - this one really has to do something.  Probably needs
 561  * to stamp in a new field in the incore inode.
 562  */
 563 STATIC void
 564 xfs_inode_item_committing(
 565         struct xfs_log_item     *lip,
 566         xfs_lsn_t               lsn)
 567 {
 568         INODE_ITEM(lip)->ili_last_lsn = lsn;
 569 }
 570
 571 /*
 572  * This is the ops vector shared by all buf log items.
 573  */
 574 static const struct xfs_item_ops xfs_inode_item_ops = {
 575         .iop_size       = xfs_inode_item_size,
 576         .iop_format     = xfs_inode_item_format,
 577         .iop_pin        = xfs_inode_item_pin,
 578         .iop_unpin      = xfs_inode_item_unpin,
 579         .iop_unlock     = xfs_inode_item_unlock,
 580         .iop_committed  = xfs_inode_item_committed,
 581         .iop_push       = xfs_inode_item_push,
 582         .iop_committing = xfs_inode_item_committing
 583 };
 584
 585
 586 /*
 587  * Initialize the inode log item for a newly allocated (in-core) inode.
 588  */
 589 void
 590 xfs_inode_item_init(
 591         struct xfs_inode        *ip,
 592         struct xfs_mount        *mp)
 593 {
 594         struct xfs_inode_log_item *iip;
 595
 596         ASSERT(ip->i_itemp == NULL);
 597         iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
 598
 599         iip->ili_inode = ip;
 600         xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
 601                                                 &xfs_inode_item_ops);
 602 }
 603
 604 /*
 605  * Free the inode log item and any memory hanging off of it.
 606  */
 607 void
 608 xfs_inode_item_destroy(
 609         xfs_inode_t     *ip)
 610 {
 611         kmem_zone_free(xfs_ili_zone, ip->i_itemp);
 612 }
 613
 614
 615 /*
 616  * This is the inode flushing I/O completion routine.  It is called
 617  * from interrupt level when the buffer containing the inode is
 618  * flushed to disk.  It is responsible for removing the inode item
 619  * from the AIL if it has not been re-logged, and unlocking the inode's
 620  * flush lock.
 621  *
 622  * To reduce AIL lock traffic as much as possible, we scan the buffer log item
 623  * list for other inodes that will run this function. We remove them from the
 624  * buffer list so we can process all the inode IO completions in one AIL lock
 625  * traversal.
 626  */
 627 void
 628 xfs_iflush_done(
 629         struct xfs_buf          *bp,
 630         struct xfs_log_item     *lip)
 631 {
 632         struct xfs_inode_log_item *iip;
 633         struct xfs_log_item     *blip;
 634         struct xfs_log_item     *next;
 635         struct xfs_log_item     *prev;
 636         struct xfs_ail          *ailp = lip->li_ailp;
 637         int                     need_ail = 0;
 638
 639         /*
 640          * Scan the buffer IO completions for other inodes being completed and
 641          * attach them to the current inode log item.
 642          */
 643         blip = bp->b_fspriv;
 644         prev = NULL;
 645         while (blip != NULL) {
 646                 if (lip->li_cb != xfs_iflush_done) {
 647                         prev = blip;
 648                         blip = blip->li_bio_list;
 649                         continue;
 650                 }
 651
 652                 /* remove from list */
 653                 next = blip->li_bio_list;
 654                 if (!prev) {
 655                         bp->b_fspriv = next;
 656                 } else {
 657                         prev->li_bio_list = next;
 658                 }
 659
 660                 /* add to current list */
 661                 blip->li_bio_list = lip->li_bio_list;
 662                 lip->li_bio_list = blip;
 663
 664                 /*
 665                  * while we have the item, do the unlocked check for needing
 666                  * the AIL lock.
 667                  */
 668                 iip = INODE_ITEM(blip);
 669                 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
 670                         need_ail++;
 671
 672                 blip = next;
 673         }
 674
 675         /* make sure we capture the state of the initial inode. */
 676         iip = INODE_ITEM(lip);
 677         if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
 678                 need_ail++;
 679
 680         /*
 681          * We only want to pull the item from the AIL if it is
 682          * actually there and its location in the log has not
 683          * changed since we started the flush.  Thus, we only bother
 684          * if the ili_logged flag is set and the inode's lsn has not
 685          * changed.  First we check the lsn outside
 686          * the lock since it's cheaper, and then we recheck while
 687          * holding the lock before removing the inode from the AIL.
 688          */
 689         if (need_ail) {
 690                 struct xfs_log_item *log_items[need_ail];
 691                 int i = 0;
 692                 spin_lock(&ailp->xa_lock);
 693                 for (blip = lip; blip; blip = blip->li_bio_list) {
 694                         iip = INODE_ITEM(blip);
 695                         if (iip->ili_logged &&
 696                             blip->li_lsn == iip->ili_flush_lsn) {
 697                                 log_items[i++] = blip;
 698                         }
 699                         ASSERT(i <= need_ail);
 700                 }
 701                 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
 702                 xfs_trans_ail_delete_bulk(ailp, log_items, i,
 703                                           SHUTDOWN_CORRUPT_INCORE);
 704         }
 705
 706
 707         /*
 708          * clean up and unlock the flush lock now we are done. We can clear the
 709          * ili_last_fields bits now that we know that the data corresponding to
 710          * them is safely on disk.
 711          */
 712         for (blip = lip; blip; blip = next) {
 713                 next = blip->li_bio_list;
 714                 blip->li_bio_list = NULL;
 715
 716                 iip = INODE_ITEM(blip);
 717                 iip->ili_logged = 0;
 718                 iip->ili_last_fields = 0;
 719                 xfs_ifunlock(iip->ili_inode);
 720         }
 721 }
 722
 723 /*
 724  * This is the inode flushing abort routine.  It is called from xfs_iflush when
 725  * the filesystem is shutting down to clean up the inode state.  It is
 726  * responsible for removing the inode item from the AIL if it has not been
 727  * re-logged, and unlocking the inode's flush lock.
 728  */
 729 void
 730 xfs_iflush_abort(
 731         xfs_inode_t             *ip,
 732         bool                    stale)
 733 {
 734         xfs_inode_log_item_t    *iip = ip->i_itemp;
 735
 736         if (iip) {
 737                 struct xfs_ail  *ailp = iip->ili_item.li_ailp;
 738                 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 739                         spin_lock(&ailp->xa_lock);
 740                         if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
 741                                 /* xfs_trans_ail_delete() drops the AIL lock. */
 742                                 xfs_trans_ail_delete(ailp, &iip->ili_item,
 743                                                 stale ?
 744                                                      SHUTDOWN_LOG_IO_ERROR :
 745                                                      SHUTDOWN_CORRUPT_INCORE);
 746                         } else
 747                                 spin_unlock(&ailp->xa_lock);
 748                 }
 749                 iip->ili_logged = 0;
 750                 /*
 751                  * Clear the ili_last_fields bits now that we know that the
 752                  * data corresponding to them is safely on disk.
 753                  */
 754                 iip->ili_last_fields = 0;
 755                 /*
 756                  * Clear the inode logging fields so no more flushes are
 757                  * attempted.
 758                  */
 759                 iip->ili_fields = 0;
 760         }
 761         /*
 762          * Release the inode's flush lock since we're done with it.
 763          */
 764         xfs_ifunlock(ip);
 765 }
 766
 767 void
 768 xfs_istale_done(
 769         struct xfs_buf          *bp,
 770         struct xfs_log_item     *lip)
 771 {
 772         xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
 773 }
 774
 775 /*
 776  * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
 777  * (which can have different field alignments) to the native version
 778  */
 779 int
 780 xfs_inode_item_format_convert(
 781         xfs_log_iovec_t         *buf,
 782         xfs_inode_log_format_t  *in_f)
 783 {
 784         if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
 785                 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
 786
 787                 in_f->ilf_type = in_f32->ilf_type;
 788                 in_f->ilf_size = in_f32->ilf_size;
 789                 in_f->ilf_fields = in_f32->ilf_fields;
 790                 in_f->ilf_asize = in_f32->ilf_asize;
 791                 in_f->ilf_dsize = in_f32->ilf_dsize;
 792                 in_f->ilf_ino = in_f32->ilf_ino;
 793                 /* copy biggest field of ilf_u */
 794                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 795                        in_f32->ilf_u.ilfu_uuid.__u_bits,
 796                        sizeof(uuid_t));
 797                 in_f->ilf_blkno = in_f32->ilf_blkno;
 798                 in_f->ilf_len = in_f32->ilf_len;
 799                 in_f->ilf_boffset = in_f32->ilf_boffset;
 800                 return 0;
 801         } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
 802                 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
 803
 804                 in_f->ilf_type = in_f64->ilf_type;
 805                 in_f->ilf_size = in_f64->ilf_size;
 806                 in_f->ilf_fields = in_f64->ilf_fields;
 807                 in_f->ilf_asize = in_f64->ilf_asize;
 808                 in_f->ilf_dsize = in_f64->ilf_dsize;
 809                 in_f->ilf_ino = in_f64->ilf_ino;
 810                 /* copy biggest field of ilf_u */
 811                 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
 812                        in_f64->ilf_u.ilfu_uuid.__u_bits,
 813                        sizeof(uuid_t));
 814                 in_f->ilf_blkno = in_f64->ilf_blkno;
 815                 in_f->ilf_len = in_f64->ilf_len;
 816                 in_f->ilf_boffset = in_f64->ilf_boffset;
 817                 return 0;
 818         }
 819         return EFSCORRUPTED;
 820 }