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_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dir2_sf.h"
  34 #include "xfs_attr_sf.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_inode_item.h"
  38 #include "xfs_bmap.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_itable.h"
  42 #include "xfs_dfrag.h"
  43 #include "xfs_error.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_vnodeops.h"
  46 #include "xfs_trace.h"
  47
  48
  49 static int xfs_swap_extents(
  50         xfs_inode_t     *ip,    /* target inode */
  51         xfs_inode_t     *tip,   /* tmp inode */
  52         xfs_swapext_t   *sxp);
  53
  54 /*
  55  * ioctl interface for swapext
  56  */
  57 int
  58 xfs_swapext(
  59         xfs_swapext_t   *sxp)
  60 {
  61         xfs_inode_t     *ip, *tip;
  62         struct file     *file, *tmp_file;
  63         int             error = 0;
  64
  65         /* Pull information for the target fd */
  66         file = fget((int)sxp->sx_fdtarget);
  67         if (!file) {
  68                 error = XFS_ERROR(EINVAL);
  69                 goto out;
  70         }
  71
  72         if (!(file->f_mode & FMODE_WRITE) ||
  73             !(file->f_mode & FMODE_READ) ||
  74             (file->f_flags & O_APPEND)) {
  75                 error = XFS_ERROR(EBADF);
  76                 goto out_put_file;
  77         }
  78
  79         tmp_file = fget((int)sxp->sx_fdtmp);
  80         if (!tmp_file) {
  81                 error = XFS_ERROR(EINVAL);
  82                 goto out_put_file;
  83         }
  84
  85         if (!(tmp_file->f_mode & FMODE_WRITE) ||
  86             !(tmp_file->f_mode & FMODE_READ) ||
  87             (tmp_file->f_flags & O_APPEND)) {
  88                 error = XFS_ERROR(EBADF);
  89                 goto out_put_tmp_file;
  90         }
  91
  92         if (IS_SWAPFILE(file->f_path.dentry->d_inode) ||
  93             IS_SWAPFILE(tmp_file->f_path.dentry->d_inode)) {
  94                 error = XFS_ERROR(EINVAL);
  95                 goto out_put_tmp_file;
  96         }
  97
  98         ip = XFS_I(file->f_path.dentry->d_inode);
  99         tip = XFS_I(tmp_file->f_path.dentry->d_inode);
 100
 101         if (ip->i_mount != tip->i_mount) {
 102                 error = XFS_ERROR(EINVAL);
 103                 goto out_put_tmp_file;
 104         }
 105
 106         if (ip->i_ino == tip->i_ino) {
 107                 error = XFS_ERROR(EINVAL);
 108                 goto out_put_tmp_file;
 109         }
 110
 111         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 112                 error = XFS_ERROR(EIO);
 113                 goto out_put_tmp_file;
 114         }
 115
 116         error = xfs_swap_extents(ip, tip, sxp);
 117
 118  out_put_tmp_file:
 119         fput(tmp_file);
 120  out_put_file:
 121         fput(file);
 122  out:
 123         return error;
 124 }
 125
 126 /*
 127  * We need to check that the format of the data fork in the temporary inode is
 128  * valid for the target inode before doing the swap. This is not a problem with
 129  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 130  * data fork depending on the space the attribute fork is taking so we can get
 131  * invalid formats on the target inode.
 132  *
 133  * E.g. target has space for 7 extents in extent format, temp inode only has
 134  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 135  * btree, but when swapped it needs to be in extent format. Hence we can't just
 136  * blindly swap data forks on attr2 filesystems.
 137  *
 138  * Note that we check the swap in both directions so that we don't end up with
 139  * a corrupt temporary inode, either.
 140  *
 141  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 142  * inode will prevent this situation from occurring, so all we do here is
 143  * reject and log the attempt. basically we are putting the responsibility on
 144  * userspace to get this right.
 145  */
 146 static int
 147 xfs_swap_extents_check_format(
 148         xfs_inode_t     *ip,    /* target inode */
 149         xfs_inode_t     *tip)   /* tmp inode */
 150 {
 151
 152         /* Should never get a local format */
 153         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 154             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 155                 return EINVAL;
 156
 157         /*
 158          * if the target inode has less extents that then temporary inode then
 159          * why did userspace call us?
 160          */
 161         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 162                 return EINVAL;
 163
 164         /*
 165          * if the target inode is in extent form and the temp inode is in btree
 166          * form then we will end up with the target inode in the wrong format
 167          * as we already know there are less extents in the temp inode.
 168          */
 169         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 170             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 171                 return EINVAL;
 172
 173         /* Check temp in extent form to max in target */
 174         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 175             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > ip->i_df.if_ext_max)
 176                 return EINVAL;
 177
 178         /* Check target in extent form to max in temp */
 179         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 180             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > tip->i_df.if_ext_max)
 181                 return EINVAL;
 182
 183         /*
 184          * If we are in a btree format, check that the temp root block will fit
 185          * in the target and that it has enough extents to be in btree format
 186          * in the target.
 187          *
 188          * Note that we have to be careful to allow btree->extent conversions
 189          * (a common defrag case) which will occur when the temp inode is in
 190          * extent format...
 191          */
 192         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 193             ((XFS_IFORK_BOFF(ip) &&
 194               tip->i_df.if_broot_bytes > XFS_IFORK_BOFF(ip)) ||
 195              XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= ip->i_df.if_ext_max))
 196                 return EINVAL;
 197
 198         /* Reciprocal target->temp btree format checks */
 199         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 200             ((XFS_IFORK_BOFF(tip) &&
 201               ip->i_df.if_broot_bytes > XFS_IFORK_BOFF(tip)) ||
 202              XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= tip->i_df.if_ext_max))
 203                 return EINVAL;
 204
 205         return 0;
 206 }
 207
 208 static int
 209 xfs_swap_extents(
 210         xfs_inode_t     *ip,    /* target inode */
 211         xfs_inode_t     *tip,   /* tmp inode */
 212         xfs_swapext_t   *sxp)
 213 {
 214         xfs_mount_t     *mp;
 215         xfs_trans_t     *tp;
 216         xfs_bstat_t     *sbp = &sxp->sx_stat;
 217         xfs_ifork_t     *tempifp, *ifp, *tifp;
 218         int             ilf_fields, tilf_fields;
 219         int             error = 0;
 220         int             aforkblks = 0;
 221         int             taforkblks = 0;
 222         __uint64_t      tmp;
 223
 224         mp = ip->i_mount;
 225
 226         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 227         if (!tempifp) {
 228                 error = XFS_ERROR(ENOMEM);
 229                 goto out;
 230         }
 231
 232         sbp = &sxp->sx_stat;
 233
 234         /*
 235          * we have to do two separate lock calls here to keep lockdep
 236          * happy. If we try to get all the locks in one call, lock will
 237          * report false positives when we drop the ILOCK and regain them
 238          * below.
 239          */
 240         xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 241         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 242
 243         /* Verify that both files have the same format */
 244         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 245                 error = XFS_ERROR(EINVAL);
 246                 goto out_unlock;
 247         }
 248
 249         /* Verify both files are either real-time or non-realtime */
 250         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 251                 error = XFS_ERROR(EINVAL);
 252                 goto out_unlock;
 253         }
 254
 255         if (VN_CACHED(VFS_I(tip)) != 0) {
 256                 error = xfs_flushinval_pages(tip, 0, -1,
 257                                 FI_REMAPF_LOCKED);
 258                 if (error)
 259                         goto out_unlock;
 260         }
 261
 262         /* Verify O_DIRECT for ftmp */
 263         if (VN_CACHED(VFS_I(tip)) != 0) {
 264                 error = XFS_ERROR(EINVAL);
 265                 goto out_unlock;
 266         }
 267
 268         /* Verify all data are being swapped */
 269         if (sxp->sx_offset != 0 ||
 270             sxp->sx_length != ip->i_d.di_size ||
 271             sxp->sx_length != tip->i_d.di_size) {
 272                 error = XFS_ERROR(EFAULT);
 273                 goto out_unlock;
 274         }
 275
 276         trace_xfs_swap_extent_before(ip, 0);
 277         trace_xfs_swap_extent_before(tip, 1);
 278
 279         /* check inode formats now that data is flushed */
 280         error = xfs_swap_extents_check_format(ip, tip);
 281         if (error) {
 282                 xfs_fs_cmn_err(CE_NOTE, mp,
 283                     "%s: inode 0x%llx format is incompatible for exchanging.",
 284                                 __FILE__, ip->i_ino);
 285                 goto out_unlock;
 286         }
 287
 288         /*
 289          * Compare the current change & modify times with that
 290          * passed in.  If they differ, we abort this swap.
 291          * This is the mechanism used to ensure the calling
 292          * process that the file was not changed out from
 293          * under it.
 294          */
 295         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 296             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 297             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 298             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 299                 error = XFS_ERROR(EBUSY);
 300                 goto out_unlock;
 301         }
 302
 303         /* We need to fail if the file is memory mapped.  Once we have tossed
 304          * all existing pages, the page fault will have no option
 305          * but to go to the filesystem for pages. By making the page fault call
 306          * vop_read (or write in the case of autogrow) they block on the iolock
 307          * until we have switched the extents.
 308          */
 309         if (VN_MAPPED(VFS_I(ip))) {
 310                 error = XFS_ERROR(EBUSY);
 311                 goto out_unlock;
 312         }
 313
 314         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 315         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 316
 317         /*
 318          * There is a race condition here since we gave up the
 319          * ilock.  However, the data fork will not change since
 320          * we have the iolock (locked for truncation too) so we
 321          * are safe.  We don't really care if non-io related
 322          * fields change.
 323          */
 324
 325         xfs_tosspages(ip, 0, -1, FI_REMAPF);
 326
 327         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 328         if ((error = xfs_trans_reserve(tp, 0,
 329                                      XFS_ICHANGE_LOG_RES(mp), 0,
 330                                      0, 0))) {
 331                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 332                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 333                 xfs_trans_cancel(tp, 0);
 334                 goto out;
 335         }
 336         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 337
 338         /*
 339          * Count the number of extended attribute blocks
 340          */
 341         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 342              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 343                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 344                 if (error)
 345                         goto out_trans_cancel;
 346         }
 347         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 348              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 349                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 350                         &taforkblks);
 351                 if (error)
 352                         goto out_trans_cancel;
 353         }
 354
 355         /*
 356          * Swap the data forks of the inodes
 357          */
 358         ifp = &ip->i_df;
 359         tifp = &tip->i_df;
 360         *tempifp = *ifp;        /* struct copy */
 361         *ifp = *tifp;           /* struct copy */
 362         *tifp = *tempifp;       /* struct copy */
 363
 364         /*
 365          * Fix the in-memory data fork values that are dependent on the fork
 366          * offset in the inode. We can't assume they remain the same as attr2
 367          * has dynamic fork offsets.
 368          */
 369         ifp->if_ext_max = XFS_IFORK_SIZE(ip, XFS_DATA_FORK) /
 370                                         (uint)sizeof(xfs_bmbt_rec_t);
 371         tifp->if_ext_max = XFS_IFORK_SIZE(tip, XFS_DATA_FORK) /
 372                                         (uint)sizeof(xfs_bmbt_rec_t);
 373
 374         /*
 375          * Fix the on-disk inode values
 376          */
 377         tmp = (__uint64_t)ip->i_d.di_nblocks;
 378         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 379         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 380
 381         tmp = (__uint64_t) ip->i_d.di_nextents;
 382         ip->i_d.di_nextents = tip->i_d.di_nextents;
 383         tip->i_d.di_nextents = tmp;
 384
 385         tmp = (__uint64_t) ip->i_d.di_format;
 386         ip->i_d.di_format = tip->i_d.di_format;
 387         tip->i_d.di_format = tmp;
 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         IHOLD(ip);
 429         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 430
 431         IHOLD(tip);
 432         xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 433
 434         xfs_trans_log_inode(tp, ip,  ilf_fields);
 435         xfs_trans_log_inode(tp, tip, tilf_fields);
 436
 437         /*
 438          * If this is a synchronous mount, make sure that the
 439          * transaction goes to disk before returning to the user.
 440          */
 441         if (mp->m_flags & XFS_MOUNT_WSYNC)
 442                 xfs_trans_set_sync(tp);
 443
 444         error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
 445
 446         trace_xfs_swap_extent_after(ip, 0);
 447         trace_xfs_swap_extent_after(tip, 1);
 448 out:
 449         kmem_free(tempifp);
 450         return error;
 451
 452 out_unlock:
 453         xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 454         xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 455         goto out;
 456
 457 out_trans_cancel:
 458         xfs_trans_cancel(tp, 0);
 459         goto out_unlock;
 460 }