fs/xfs/xfs_dfrag.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_bmap.h"
  33 #include "xfs_itable.h"
  34 #include "xfs_dfrag.h"
  35 #include "xfs_error.h"
  36 #include "xfs_vnodeops.h"
  37 #include "xfs_trace.h"
  38
  39
  40 static int xfs_swap_extents(
  41         xfs_inode_t     *ip,    /* target inode */
  42         xfs_inode_t     *tip,   /* tmp inode */
  43         xfs_swapext_t   *sxp);
  44
  45 /*
  46  * ioctl interface for swapext
  47  */
  48 int
  49 xfs_swapext(
  50         xfs_swapext_t   *sxp)
  51 {
  52         xfs_inode_t     *ip, *tip;
  53         struct file     *file, *tmp_file;
  54         int             error = 0;
  55
  56         /* Pull information for the target fd */
  57         file = fget((int)sxp->sx_fdtarget);
  58         if (!file) {
  59                 error = XFS_ERROR(EINVAL);
  60                 goto out;
  61         }
  62
  63         if (!(file->f_mode & FMODE_WRITE) ||
  64             !(file->f_mode & FMODE_READ) ||
  65             (file->f_flags & O_APPEND)) {
  66                 error = XFS_ERROR(EBADF);
  67                 goto out_put_file;
  68         }
  69
  70         tmp_file = fget((int)sxp->sx_fdtmp);
  71         if (!tmp_file) {
  72                 error = XFS_ERROR(EINVAL);
  73                 goto out_put_file;
  74         }
  75
  76         if (!(tmp_file->f_mode & FMODE_WRITE) ||
  77             !(tmp_file->f_mode & FMODE_READ) ||
  78             (tmp_file->f_flags & O_APPEND)) {
  79                 error = XFS_ERROR(EBADF);
  80                 goto out_put_tmp_file;
  81         }
  82
  83         if (IS_SWAPFILE(file->f_path.dentry->d_inode) ||
  84             IS_SWAPFILE(tmp_file->f_path.dentry->d_inode)) {
  85                 error = XFS_ERROR(EINVAL);
  86                 goto out_put_tmp_file;
  87         }
  88
  89         ip = XFS_I(file->f_path.dentry->d_inode);
  90         tip = XFS_I(tmp_file->f_path.dentry->d_inode);
  91
  92         if (ip->i_mount != tip->i_mount) {
  93                 error = XFS_ERROR(EINVAL);
  94                 goto out_put_tmp_file;
  95         }
  96
  97         if (ip->i_ino == tip->i_ino) {
  98                 error = XFS_ERROR(EINVAL);
  99                 goto out_put_tmp_file;
 100         }
 101
 102         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 103                 error = XFS_ERROR(EIO);
 104                 goto out_put_tmp_file;
 105         }
 106
 107         error = xfs_swap_extents(ip, tip, sxp);
 108
 109  out_put_tmp_file:
 110         fput(tmp_file);
 111  out_put_file:
 112         fput(file);
 113  out:
 114         return error;
 115 }
 116
 117 /*
 118  * We need to check that the format of the data fork in the temporary inode is
 119  * valid for the target inode before doing the swap. This is not a problem with
 120  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 121  * data fork depending on the space the attribute fork is taking so we can get
 122  * invalid formats on the target inode.
 123  *
 124  * E.g. target has space for 7 extents in extent format, temp inode only has
 125  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 126  * btree, but when swapped it needs to be in extent format. Hence we can't just
 127  * blindly swap data forks on attr2 filesystems.
 128  *
 129  * Note that we check the swap in both directions so that we don't end up with
 130  * a corrupt temporary inode, either.
 131  *
 132  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 133  * inode will prevent this situation from occurring, so all we do here is
 134  * reject and log the attempt. basically we are putting the responsibility on
 135  * userspace to get this right.
 136  */
 137 static int
 138 xfs_swap_extents_check_format(
 139         xfs_inode_t     *ip,    /* target inode */
 140         xfs_inode_t     *tip)   /* tmp inode */
 141 {
 142
 143         /* Should never get a local format */
 144         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 145             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 146                 return EINVAL;
 147
 148         /*
 149          * if the target inode has less extents that then temporary inode then
 150          * why did userspace call us?
 151          */
 152         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 153                 return EINVAL;
 154
 155         /*
 156          * if the target inode is in extent form and the temp inode is in btree
 157          * form then we will end up with the target inode in the wrong format
 158          * as we already know there are less extents in the temp inode.
 159          */
 160         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 161             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 162                 return EINVAL;
 163
 164         /* Check temp in extent form to max in target */
 165         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 166             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > ip->i_df.if_ext_max)
 167                 return EINVAL;
 168
 169         /* Check target in extent form to max in temp */
 170         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 171             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > tip->i_df.if_ext_max)
 172                 return EINVAL;
 173
 174         /*
 175          * If we are in a btree format, check that the temp root block will fit
 176          * in the target and that it has enough extents to be in btree format
 177          * in the target.
 178          *
 179          * Note that we have to be careful to allow btree->extent conversions
 180          * (a common defrag case) which will occur when the temp inode is in
 181          * extent format...
 182          */
 183         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 184             ((XFS_IFORK_BOFF(ip) &&
 185               tip->i_df.if_broot_bytes > XFS_IFORK_BOFF(ip)) ||
 186              XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= ip->i_df.if_ext_max))
 187                 return EINVAL;
 188
 189         /* Reciprocal target->temp btree format checks */
 190         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 191             ((XFS_IFORK_BOFF(tip) &&
 192               ip->i_df.if_broot_bytes > XFS_IFORK_BOFF(tip)) ||
 193              XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= tip->i_df.if_ext_max))
 194                 return EINVAL;
 195
 196         return 0;
 197 }
 198
 199 static int
 200 xfs_swap_extents(
 201         xfs_inode_t     *ip,    /* target inode */
 202         xfs_inode_t     *tip,   /* tmp inode */
 203         xfs_swapext_t   *sxp)
 204 {
 205         xfs_mount_t     *mp = ip->i_mount;
 206         xfs_trans_t     *tp;
 207         xfs_bstat_t     *sbp = &sxp->sx_stat;
 208         xfs_ifork_t     *tempifp, *ifp, *tifp;
 209         int             ilf_fields, tilf_fields;
 210         int             error = 0;
 211         int             aforkblks = 0;
 212         int             taforkblks = 0;
 213         __uint64_t      tmp;
 214
 215         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 216         if (!tempifp) {
 217                 error = XFS_ERROR(ENOMEM);
 218                 goto out;
 219         }
 220
 221         /*
 222          * we have to do two separate lock calls here to keep lockdep
 223          * happy. If we try to get all the locks in one call, lock will
 224          * report false positives when we drop the ILOCK and regain them
 225          * below.
 226          */
 227         xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 228         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 229
 230         /* Verify that both files have the same format */
 231         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 232                 error = XFS_ERROR(EINVAL);
 233                 goto out_unlock;
 234         }
 235
 236         /* Verify both files are either real-time or non-realtime */
 237         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 238                 error = XFS_ERROR(EINVAL);
 239                 goto out_unlock;
 240         }
 241
 242         if (VN_CACHED(VFS_I(tip)) != 0) {
 243                 error = xfs_flushinval_pages(tip, 0, -1,
 244                                 FI_REMAPF_LOCKED);
 245                 if (error)
 246                         goto out_unlock;
 247         }
 248
 249         /* Verify O_DIRECT for ftmp */
 250         if (VN_CACHED(VFS_I(tip)) != 0) {
 251                 error = XFS_ERROR(EINVAL);
 252                 goto out_unlock;
 253         }
 254
 255         /* Verify all data are being swapped */
 256         if (sxp->sx_offset != 0 ||
 257             sxp->sx_length != ip->i_d.di_size ||
 258             sxp->sx_length != tip->i_d.di_size) {
 259                 error = XFS_ERROR(EFAULT);
 260                 goto out_unlock;
 261         }
 262
 263         trace_xfs_swap_extent_before(ip, 0);
 264         trace_xfs_swap_extent_before(tip, 1);
 265
 266         /* check inode formats now that data is flushed */
 267         error = xfs_swap_extents_check_format(ip, tip);
 268         if (error) {
 269                 xfs_notice(mp,
 270                     "%s: inode 0x%llx format is incompatible for exchanging.",
 271                                 __func__, ip->i_ino);
 272                 goto out_unlock;
 273         }
 274
 275         /*
 276          * Compare the current change & modify times with that
 277          * passed in.  If they differ, we abort this swap.
 278          * This is the mechanism used to ensure the calling
 279          * process that the file was not changed out from
 280          * under it.
 281          */
 282         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 283             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 284             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 285             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 286                 error = XFS_ERROR(EBUSY);
 287                 goto out_unlock;
 288         }
 289
 290         /* We need to fail if the file is memory mapped.  Once we have tossed
 291          * all existing pages, the page fault will have no option
 292          * but to go to the filesystem for pages. By making the page fault call
 293          * vop_read (or write in the case of autogrow) they block on the iolock
 294          * until we have switched the extents.
 295          */
 296         if (VN_MAPPED(VFS_I(ip))) {
 297                 error = XFS_ERROR(EBUSY);
 298                 goto out_unlock;
 299         }
 300
 301         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 302         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 303
 304         /*
 305          * There is a race condition here since we gave up the
 306          * ilock.  However, the data fork will not change since
 307          * we have the iolock (locked for truncation too) so we
 308          * are safe.  We don't really care if non-io related
 309          * fields change.
 310          */
 311
 312         xfs_tosspages(ip, 0, -1, FI_REMAPF);
 313
 314         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 315         if ((error = xfs_trans_reserve(tp, 0,
 316                                      XFS_ICHANGE_LOG_RES(mp), 0,
 317                                      0, 0))) {
 318                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 319                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 320                 xfs_trans_cancel(tp, 0);
 321                 goto out;
 322         }
 323         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 324
 325         /*
 326          * Count the number of extended attribute blocks
 327          */
 328         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 329              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 330                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 331                 if (error)
 332                         goto out_trans_cancel;
 333         }
 334         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 335              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 336                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 337                         &taforkblks);
 338                 if (error)
 339                         goto out_trans_cancel;
 340         }
 341
 342         /*
 343          * Swap the data forks of the inodes
 344          */
 345         ifp = &ip->i_df;
 346         tifp = &tip->i_df;
 347         *tempifp = *ifp;        /* struct copy */
 348         *ifp = *tifp;           /* struct copy */
 349         *tifp = *tempifp;       /* struct copy */
 350
 351         /*
 352          * Fix the in-memory data fork values that are dependent on the fork
 353          * offset in the inode. We can't assume they remain the same as attr2
 354          * has dynamic fork offsets.
 355          */
 356         ifp->if_ext_max = XFS_IFORK_SIZE(ip, XFS_DATA_FORK) /
 357                                         (uint)sizeof(xfs_bmbt_rec_t);
 358         tifp->if_ext_max = XFS_IFORK_SIZE(tip, XFS_DATA_FORK) /
 359                                         (uint)sizeof(xfs_bmbt_rec_t);
 360
 361         /*
 362          * Fix the on-disk inode values
 363          */
 364         tmp = (__uint64_t)ip->i_d.di_nblocks;
 365         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 366         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 367
 368         tmp = (__uint64_t) ip->i_d.di_nextents;
 369         ip->i_d.di_nextents = tip->i_d.di_nextents;
 370         tip->i_d.di_nextents = tmp;
 371
 372         tmp = (__uint64_t) ip->i_d.di_format;
 373         ip->i_d.di_format = tip->i_d.di_format;
 374         tip->i_d.di_format = tmp;
 375
 376         /*
 377          * The extents in the source inode could still contain speculative
 378          * preallocation beyond EOF (e.g. the file is open but not modified
 379          * while defrag is in progress). In that case, we need to copy over the
 380          * number of delalloc blocks the data fork in the source inode is
 381          * tracking beyond EOF so that when the fork is truncated away when the
 382          * temporary inode is unlinked we don't underrun the i_delayed_blks
 383          * counter on that inode.
 384          */
 385         ASSERT(tip->i_delayed_blks == 0);
 386         tip->i_delayed_blks = ip->i_delayed_blks;
 387         ip->i_delayed_blks = 0;
 388
 389         ilf_fields = XFS_ILOG_CORE;
 390
 391         switch(ip->i_d.di_format) {
 392         case XFS_DINODE_FMT_EXTENTS:
 393                 /* If the extents fit in the inode, fix the
 394                  * pointer.  Otherwise it's already NULL or
 395                  * pointing to the extent.
 396                  */
 397                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 398                         ifp->if_u1.if_extents =
 399                                 ifp->if_u2.if_inline_ext;
 400                 }
 401                 ilf_fields |= XFS_ILOG_DEXT;
 402                 break;
 403         case XFS_DINODE_FMT_BTREE:
 404                 ilf_fields |= XFS_ILOG_DBROOT;
 405                 break;
 406         }
 407
 408         tilf_fields = XFS_ILOG_CORE;
 409
 410         switch(tip->i_d.di_format) {
 411         case XFS_DINODE_FMT_EXTENTS:
 412                 /* If the extents fit in the inode, fix the
 413                  * pointer.  Otherwise it's already NULL or
 414                  * pointing to the extent.
 415                  */
 416                 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 417                         tifp->if_u1.if_extents =
 418                                 tifp->if_u2.if_inline_ext;
 419                 }
 420                 tilf_fields |= XFS_ILOG_DEXT;
 421                 break;
 422         case XFS_DINODE_FMT_BTREE:
 423                 tilf_fields |= XFS_ILOG_DBROOT;
 424                 break;
 425         }
 426
 427
 428         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 429         xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 430
 431         xfs_trans_log_inode(tp, ip,  ilf_fields);
 432         xfs_trans_log_inode(tp, tip, tilf_fields);
 433
 434         /*
 435          * If this is a synchronous mount, make sure that the
 436          * transaction goes to disk before returning to the user.
 437          */
 438         if (mp->m_flags & XFS_MOUNT_WSYNC)
 439                 xfs_trans_set_sync(tp);
 440
 441         error = xfs_trans_commit(tp, 0);
 442
 443         trace_xfs_swap_extent_after(ip, 0);
 444         trace_xfs_swap_extent_after(tip, 1);
 445 out:
 446         kmem_free(tempifp);
 447         return error;
 448
 449 out_unlock:
 450         xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 451         xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 452         goto out;
 453
 454 out_trans_cancel:
 455         xfs_trans_cancel(tp, 0);
 456         goto out_unlock;
 457 }