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) || (file->f_flags & O_APPEND)) {
  73                 error = XFS_ERROR(EBADF);
  74                 goto out_put_file;
  75         }
  76
  77         tmp_file = fget((int)sxp->sx_fdtmp);
  78         if (!tmp_file) {
  79                 error = XFS_ERROR(EINVAL);
  80                 goto out_put_file;
  81         }
  82
  83         if (!(tmp_file->f_mode & FMODE_WRITE) ||
  84             (tmp_file->f_flags & O_APPEND)) {
  85                 error = XFS_ERROR(EBADF);
  86                 goto out_put_tmp_file;
  87         }
  88
  89         if (IS_SWAPFILE(file->f_path.dentry->d_inode) ||
  90             IS_SWAPFILE(tmp_file->f_path.dentry->d_inode)) {
  91                 error = XFS_ERROR(EINVAL);
  92                 goto out_put_tmp_file;
  93         }
  94
  95         ip = XFS_I(file->f_path.dentry->d_inode);
  96         tip = XFS_I(tmp_file->f_path.dentry->d_inode);
  97
  98         if (ip->i_mount != tip->i_mount) {
  99                 error = XFS_ERROR(EINVAL);
 100                 goto out_put_tmp_file;
 101         }
 102
 103         if (ip->i_ino == tip->i_ino) {
 104                 error = XFS_ERROR(EINVAL);
 105                 goto out_put_tmp_file;
 106         }
 107
 108         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 109                 error = XFS_ERROR(EIO);
 110                 goto out_put_tmp_file;
 111         }
 112
 113         error = xfs_swap_extents(ip, tip, sxp);
 114
 115  out_put_tmp_file:
 116         fput(tmp_file);
 117  out_put_file:
 118         fput(file);
 119  out:
 120         return error;
 121 }
 122
 123 /*
 124  * We need to check that the format of the data fork in the temporary inode is
 125  * valid for the target inode before doing the swap. This is not a problem with
 126  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 127  * data fork depending on the space the attribute fork is taking so we can get
 128  * invalid formats on the target inode.
 129  *
 130  * E.g. target has space for 7 extents in extent format, temp inode only has
 131  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 132  * btree, but when swapped it needs to be in extent format. Hence we can't just
 133  * blindly swap data forks on attr2 filesystems.
 134  *
 135  * Note that we check the swap in both directions so that we don't end up with
 136  * a corrupt temporary inode, either.
 137  *
 138  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 139  * inode will prevent this situation from occurring, so all we do here is
 140  * reject and log the attempt. basically we are putting the responsibility on
 141  * userspace to get this right.
 142  */
 143 static int
 144 xfs_swap_extents_check_format(
 145         xfs_inode_t     *ip,    /* target inode */
 146         xfs_inode_t     *tip)   /* tmp inode */
 147 {
 148
 149         /* Should never get a local format */
 150         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 151             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 152                 return EINVAL;
 153
 154         /*
 155          * if the target inode has less extents that then temporary inode then
 156          * why did userspace call us?
 157          */
 158         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 159                 return EINVAL;
 160
 161         /*
 162          * if the target inode is in extent form and the temp inode is in btree
 163          * form then we will end up with the target inode in the wrong format
 164          * as we already know there are less extents in the temp inode.
 165          */
 166         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 167             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 168                 return EINVAL;
 169
 170         /* Check temp in extent form to max in target */
 171         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 172             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > ip->i_df.if_ext_max)
 173                 return EINVAL;
 174
 175         /* Check target in extent form to max in temp */
 176         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 177             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > tip->i_df.if_ext_max)
 178                 return EINVAL;
 179
 180         /*
 181          * If we are in a btree format, check that the temp root block will fit
 182          * in the target and that it has enough extents to be in btree format
 183          * in the target.
 184          *
 185          * Note that we have to be careful to allow btree->extent conversions
 186          * (a common defrag case) which will occur when the temp inode is in
 187          * extent format...
 188          */
 189         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 190             ((XFS_IFORK_BOFF(ip) &&
 191               tip->i_df.if_broot_bytes > XFS_IFORK_BOFF(ip)) ||
 192              XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= ip->i_df.if_ext_max))
 193                 return EINVAL;
 194
 195         /* Reciprocal target->temp btree format checks */
 196         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 197             ((XFS_IFORK_BOFF(tip) &&
 198               ip->i_df.if_broot_bytes > XFS_IFORK_BOFF(tip)) ||
 199              XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= tip->i_df.if_ext_max))
 200                 return EINVAL;
 201
 202         return 0;
 203 }
 204
 205 static int
 206 xfs_swap_extents(
 207         xfs_inode_t     *ip,    /* target inode */
 208         xfs_inode_t     *tip,   /* tmp inode */
 209         xfs_swapext_t   *sxp)
 210 {
 211         xfs_mount_t     *mp;
 212         xfs_trans_t     *tp;
 213         xfs_bstat_t     *sbp = &sxp->sx_stat;
 214         xfs_ifork_t     *tempifp, *ifp, *tifp;
 215         int             ilf_fields, tilf_fields;
 216         int             error = 0;
 217         int             aforkblks = 0;
 218         int             taforkblks = 0;
 219         __uint64_t      tmp;
 220
 221         mp = ip->i_mount;
 222
 223         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 224         if (!tempifp) {
 225                 error = XFS_ERROR(ENOMEM);
 226                 goto out;
 227         }
 228
 229         sbp = &sxp->sx_stat;
 230
 231         /*
 232          * we have to do two separate lock calls here to keep lockdep
 233          * happy. If we try to get all the locks in one call, lock will
 234          * report false positives when we drop the ILOCK and regain them
 235          * below.
 236          */
 237         xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 238         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 239
 240         /* Verify that both files have the same format */
 241         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 242                 error = XFS_ERROR(EINVAL);
 243                 goto out_unlock;
 244         }
 245
 246         /* Verify both files are either real-time or non-realtime */
 247         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 248                 error = XFS_ERROR(EINVAL);
 249                 goto out_unlock;
 250         }
 251
 252         if (VN_CACHED(VFS_I(tip)) != 0) {
 253                 error = xfs_flushinval_pages(tip, 0, -1,
 254                                 FI_REMAPF_LOCKED);
 255                 if (error)
 256                         goto out_unlock;
 257         }
 258
 259         /* Verify O_DIRECT for ftmp */
 260         if (VN_CACHED(VFS_I(tip)) != 0) {
 261                 error = XFS_ERROR(EINVAL);
 262                 goto out_unlock;
 263         }
 264
 265         /* Verify all data are being swapped */
 266         if (sxp->sx_offset != 0 ||
 267             sxp->sx_length != ip->i_d.di_size ||
 268             sxp->sx_length != tip->i_d.di_size) {
 269                 error = XFS_ERROR(EFAULT);
 270                 goto out_unlock;
 271         }
 272
 273         trace_xfs_swap_extent_before(ip, 0);
 274         trace_xfs_swap_extent_before(tip, 1);
 275
 276         /* check inode formats now that data is flushed */
 277         error = xfs_swap_extents_check_format(ip, tip);
 278         if (error) {
 279                 xfs_fs_cmn_err(CE_NOTE, mp,
 280                     "%s: inode 0x%llx format is incompatible for exchanging.",
 281                                 __FILE__, ip->i_ino);
 282                 goto out_unlock;
 283         }
 284
 285         /*
 286          * Compare the current change & modify times with that
 287          * passed in.  If they differ, we abort this swap.
 288          * This is the mechanism used to ensure the calling
 289          * process that the file was not changed out from
 290          * under it.
 291          */
 292         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 293             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 294             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 295             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 296                 error = XFS_ERROR(EBUSY);
 297                 goto out_unlock;
 298         }
 299
 300         /* We need to fail if the file is memory mapped.  Once we have tossed
 301          * all existing pages, the page fault will have no option
 302          * but to go to the filesystem for pages. By making the page fault call
 303          * vop_read (or write in the case of autogrow) they block on the iolock
 304          * until we have switched the extents.
 305          */
 306         if (VN_MAPPED(VFS_I(ip))) {
 307                 error = XFS_ERROR(EBUSY);
 308                 goto out_unlock;
 309         }
 310
 311         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 312         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 313
 314         /*
 315          * There is a race condition here since we gave up the
 316          * ilock.  However, the data fork will not change since
 317          * we have the iolock (locked for truncation too) so we
 318          * are safe.  We don't really care if non-io related
 319          * fields change.
 320          */
 321
 322         xfs_tosspages(ip, 0, -1, FI_REMAPF);
 323
 324         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 325         if ((error = xfs_trans_reserve(tp, 0,
 326                                      XFS_ICHANGE_LOG_RES(mp), 0,
 327                                      0, 0))) {
 328                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 329                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 330                 xfs_trans_cancel(tp, 0);
 331                 goto out;
 332         }
 333         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 334
 335         /*
 336          * Count the number of extended attribute blocks
 337          */
 338         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 339              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 340                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 341                 if (error)
 342                         goto out_trans_cancel;
 343         }
 344         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 345              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 346                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 347                         &taforkblks);
 348                 if (error)
 349                         goto out_trans_cancel;
 350         }
 351
 352         /*
 353          * Swap the data forks of the inodes
 354          */
 355         ifp = &ip->i_df;
 356         tifp = &tip->i_df;
 357         *tempifp = *ifp;        /* struct copy */
 358         *ifp = *tifp;           /* struct copy */
 359         *tifp = *tempifp;       /* struct copy */
 360
 361         /*
 362          * Fix the in-memory data fork values that are dependent on the fork
 363          * offset in the inode. We can't assume they remain the same as attr2
 364          * has dynamic fork offsets.
 365          */
 366         ifp->if_ext_max = XFS_IFORK_SIZE(ip, XFS_DATA_FORK) /
 367                                         (uint)sizeof(xfs_bmbt_rec_t);
 368         tifp->if_ext_max = XFS_IFORK_SIZE(tip, XFS_DATA_FORK) /
 369                                         (uint)sizeof(xfs_bmbt_rec_t);
 370
 371         /*
 372          * Fix the on-disk inode values
 373          */
 374         tmp = (__uint64_t)ip->i_d.di_nblocks;
 375         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 376         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 377
 378         tmp = (__uint64_t) ip->i_d.di_nextents;
 379         ip->i_d.di_nextents = tip->i_d.di_nextents;
 380         tip->i_d.di_nextents = tmp;
 381
 382         tmp = (__uint64_t) ip->i_d.di_format;
 383         ip->i_d.di_format = tip->i_d.di_format;
 384         tip->i_d.di_format = tmp;
 385
 386         ilf_fields = XFS_ILOG_CORE;
 387
 388         switch(ip->i_d.di_format) {
 389         case XFS_DINODE_FMT_EXTENTS:
 390                 /* If the extents fit in the inode, fix the
 391                  * pointer.  Otherwise it's already NULL or
 392                  * pointing to the extent.
 393                  */
 394                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 395                         ifp->if_u1.if_extents =
 396                                 ifp->if_u2.if_inline_ext;
 397                 }
 398                 ilf_fields |= XFS_ILOG_DEXT;
 399                 break;
 400         case XFS_DINODE_FMT_BTREE:
 401                 ilf_fields |= XFS_ILOG_DBROOT;
 402                 break;
 403         }
 404
 405         tilf_fields = XFS_ILOG_CORE;
 406
 407         switch(tip->i_d.di_format) {
 408         case XFS_DINODE_FMT_EXTENTS:
 409                 /* If the extents fit in the inode, fix the
 410                  * pointer.  Otherwise it's already NULL or
 411                  * pointing to the extent.
 412                  */
 413                 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 414                         tifp->if_u1.if_extents =
 415                                 tifp->if_u2.if_inline_ext;
 416                 }
 417                 tilf_fields |= XFS_ILOG_DEXT;
 418                 break;
 419         case XFS_DINODE_FMT_BTREE:
 420                 tilf_fields |= XFS_ILOG_DBROOT;
 421                 break;
 422         }
 423
 424
 425         IHOLD(ip);
 426         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 427
 428         IHOLD(tip);
 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, XFS_TRANS_SWAPEXT);
 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 }