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
  47 /*
  48  * Syssgi interface for swapext
  49  */
  50 int
  51 xfs_swapext(
  52         xfs_swapext_t   *sxp)
  53 {
  54         xfs_inode_t     *ip, *tip;
  55         struct file     *file, *target_file;
  56         int             error = 0;
  57
  58         /* Pull information for the target fd */
  59         file = fget((int)sxp->sx_fdtarget);
  60         if (!file) {
  61                 error = XFS_ERROR(EINVAL);
  62                 goto out;
  63         }
  64
  65         if (!(file->f_mode & FMODE_WRITE) ||
  66             !(file->f_mode & FMODE_READ) ||
  67             (file->f_flags & O_APPEND)) {
  68                 error = XFS_ERROR(EBADF);
  69                 goto out_put_file;
  70         }
  71
  72         target_file = fget((int)sxp->sx_fdtmp);
  73         if (!target_file) {
  74                 error = XFS_ERROR(EINVAL);
  75                 goto out_put_file;
  76         }
  77
  78         if (!(target_file->f_mode & FMODE_WRITE) ||
  79             !(target_file->f_mode & FMODE_READ) ||
  80             (target_file->f_flags & O_APPEND)) {
  81                 error = XFS_ERROR(EBADF);
  82                 goto out_put_target_file;
  83         }
  84
  85         if (IS_SWAPFILE(file->f_path.dentry->d_inode) ||
  86             IS_SWAPFILE(target_file->f_path.dentry->d_inode)) {
  87                 error = XFS_ERROR(EINVAL);
  88                 goto out_put_target_file;
  89         }
  90
  91         ip = XFS_I(file->f_path.dentry->d_inode);
  92         tip = XFS_I(target_file->f_path.dentry->d_inode);
  93
  94         if (ip->i_mount != tip->i_mount) {
  95                 error = XFS_ERROR(EINVAL);
  96                 goto out_put_target_file;
  97         }
  98
  99         if (ip->i_ino == tip->i_ino) {
 100                 error = XFS_ERROR(EINVAL);
 101                 goto out_put_target_file;
 102         }
 103
 104         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 105                 error = XFS_ERROR(EIO);
 106                 goto out_put_target_file;
 107         }
 108
 109         error = xfs_swap_extents(ip, tip, sxp);
 110
 111  out_put_target_file:
 112         fput(target_file);
 113  out_put_file:
 114         fput(file);
 115  out:
 116         return error;
 117 }
 118
 119 /*
 120  * We need to check that the format of the data fork in the temporary inode is
 121  * valid for the target inode before doing the swap. This is not a problem with
 122  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 123  * data fork depending on the space the attribute fork is taking so we can get
 124  * invalid formats on the target inode.
 125  *
 126  * E.g. target has space for 7 extents in extent format, temp inode only has
 127  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 128  * btree, but when swapped it needs to be in extent format. Hence we can't just
 129  * blindly swap data forks on attr2 filesystems.
 130  *
 131  * Note that we check the swap in both directions so that we don't end up with
 132  * a corrupt temporary inode, either.
 133  *
 134  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 135  * inode will prevent this situation from occurring, so all we do here is
 136  * reject and log the attempt. basically we are putting the responsibility on
 137  * userspace to get this right.
 138  */
 139 static int
 140 xfs_swap_extents_check_format(
 141         xfs_inode_t     *ip,    /* target inode */
 142         xfs_inode_t     *tip)   /* tmp inode */
 143 {
 144
 145         /* Should never get a local format */
 146         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 147             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 148                 return EINVAL;
 149
 150         /*
 151          * if the target inode has less extents that then temporary inode then
 152          * why did userspace call us?
 153          */
 154         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 155                 return EINVAL;
 156
 157         /*
 158          * if the target inode is in extent form and the temp inode is in btree
 159          * form then we will end up with the target inode in the wrong format
 160          * as we already know there are less extents in the temp inode.
 161          */
 162         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 163             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 164                 return EINVAL;
 165
 166         /* Check temp in extent form to max in target */
 167         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 168             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > ip->i_df.if_ext_max)
 169                 return EINVAL;
 170
 171         /* Check target in extent form to max in temp */
 172         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 173             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > tip->i_df.if_ext_max)
 174                 return EINVAL;
 175
 176         /* Check root block of temp in btree form to max in target */
 177         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 178             XFS_IFORK_BOFF(ip) &&
 179             tip->i_df.if_broot_bytes > XFS_IFORK_BOFF(ip))
 180                 return EINVAL;
 181
 182         /* Check root block of target in btree form to max in temp */
 183         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 184             XFS_IFORK_BOFF(tip) &&
 185             ip->i_df.if_broot_bytes > XFS_IFORK_BOFF(tip))
 186                 return EINVAL;
 187
 188         return 0;
 189 }
 190
 191 int
 192 xfs_swap_extents(
 193         xfs_inode_t     *ip,    /* target inode */
 194         xfs_inode_t     *tip,   /* tmp inode */
 195         xfs_swapext_t   *sxp)
 196 {
 197         xfs_mount_t     *mp;
 198         xfs_trans_t     *tp;
 199         xfs_bstat_t     *sbp = &sxp->sx_stat;
 200         xfs_ifork_t     *tempifp, *ifp, *tifp;
 201         int             ilf_fields, tilf_fields;
 202         int             error = 0;
 203         int             aforkblks = 0;
 204         int             taforkblks = 0;
 205         __uint64_t      tmp;
 206
 207         mp = ip->i_mount;
 208
 209         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 210         if (!tempifp) {
 211                 error = XFS_ERROR(ENOMEM);
 212                 goto out;
 213         }
 214
 215         sbp = &sxp->sx_stat;
 216
 217         /*
 218          * we have to do two separate lock calls here to keep lockdep
 219          * happy. If we try to get all the locks in one call, lock will
 220          * report false positives when we drop the ILOCK and regain them
 221          * below.
 222          */
 223         xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 224         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 225
 226         /* Verify that both files have the same format */
 227         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 228                 error = XFS_ERROR(EINVAL);
 229                 goto out_unlock;
 230         }
 231
 232         /* Verify both files are either real-time or non-realtime */
 233         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 234                 error = XFS_ERROR(EINVAL);
 235                 goto out_unlock;
 236         }
 237
 238         if (VN_CACHED(VFS_I(tip)) != 0) {
 239                 xfs_inval_cached_trace(tip, 0, -1, 0, -1);
 240                 error = xfs_flushinval_pages(tip, 0, -1,
 241                                 FI_REMAPF_LOCKED);
 242                 if (error)
 243                         goto out_unlock;
 244         }
 245
 246         /* Verify O_DIRECT for ftmp */
 247         if (VN_CACHED(VFS_I(tip)) != 0) {
 248                 error = XFS_ERROR(EINVAL);
 249                 goto out_unlock;
 250         }
 251
 252         /* Verify all data are being swapped */
 253         if (sxp->sx_offset != 0 ||
 254             sxp->sx_length != ip->i_d.di_size ||
 255             sxp->sx_length != tip->i_d.di_size) {
 256                 error = XFS_ERROR(EFAULT);
 257                 goto out_unlock;
 258         }
 259
 260         /* check inode formats now that data is flushed */
 261         error = xfs_swap_extents_check_format(ip, tip);
 262         if (error) {
 263                 xfs_fs_cmn_err(CE_NOTE, mp,
 264                     "%s: inode 0x%llx format is incompatible for exchanging.",
 265                                 __FILE__, ip->i_ino);
 266                 goto out_unlock;
 267         }
 268
 269         /*
 270          * Compare the current change & modify times with that
 271          * passed in.  If they differ, we abort this swap.
 272          * This is the mechanism used to ensure the calling
 273          * process that the file was not changed out from
 274          * under it.
 275          */
 276         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 277             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 278             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 279             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 280                 error = XFS_ERROR(EBUSY);
 281                 goto out_unlock;
 282         }
 283
 284         /* We need to fail if the file is memory mapped.  Once we have tossed
 285          * all existing pages, the page fault will have no option
 286          * but to go to the filesystem for pages. By making the page fault call
 287          * vop_read (or write in the case of autogrow) they block on the iolock
 288          * until we have switched the extents.
 289          */
 290         if (VN_MAPPED(VFS_I(ip))) {
 291                 error = XFS_ERROR(EBUSY);
 292                 goto out_unlock;
 293         }
 294
 295         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 296         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 297
 298         /*
 299          * There is a race condition here since we gave up the
 300          * ilock.  However, the data fork will not change since
 301          * we have the iolock (locked for truncation too) so we
 302          * are safe.  We don't really care if non-io related
 303          * fields change.
 304          */
 305
 306         xfs_tosspages(ip, 0, -1, FI_REMAPF);
 307
 308         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 309         if ((error = xfs_trans_reserve(tp, 0,
 310                                      XFS_ICHANGE_LOG_RES(mp), 0,
 311                                      0, 0))) {
 312                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 313                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 314                 xfs_trans_cancel(tp, 0);
 315                 goto out;
 316         }
 317         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 318
 319         /*
 320          * Count the number of extended attribute blocks
 321          */
 322         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 323              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 324                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 325                 if (error)
 326                         goto out_trans_cancel;
 327         }
 328         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 329              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 330                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 331                         &taforkblks);
 332                 if (error)
 333                         goto out_trans_cancel;
 334         }
 335
 336         /*
 337          * Swap the data forks of the inodes
 338          */
 339         ifp = &ip->i_df;
 340         tifp = &tip->i_df;
 341         *tempifp = *ifp;        /* struct copy */
 342         *ifp = *tifp;           /* struct copy */
 343         *tifp = *tempifp;       /* struct copy */
 344
 345         /*
 346          * Fix the in-memory data fork values that are dependent on the fork
 347          * offset in the inode. We can't assume they remain the same as attr2
 348          * has dynamic fork offsets.
 349          */
 350         ifp->if_ext_max = XFS_IFORK_SIZE(ip, XFS_DATA_FORK) /
 351                                         (uint)sizeof(xfs_bmbt_rec_t);
 352         tifp->if_ext_max = XFS_IFORK_SIZE(tip, XFS_DATA_FORK) /
 353                                         (uint)sizeof(xfs_bmbt_rec_t);
 354
 355         /*
 356          * Fix the on-disk inode values
 357          */
 358         tmp = (__uint64_t)ip->i_d.di_nblocks;
 359         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 360         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 361
 362         tmp = (__uint64_t) ip->i_d.di_nextents;
 363         ip->i_d.di_nextents = tip->i_d.di_nextents;
 364         tip->i_d.di_nextents = tmp;
 365
 366         tmp = (__uint64_t) ip->i_d.di_format;
 367         ip->i_d.di_format = tip->i_d.di_format;
 368         tip->i_d.di_format = tmp;
 369
 370         ilf_fields = XFS_ILOG_CORE;
 371
 372         switch(ip->i_d.di_format) {
 373         case XFS_DINODE_FMT_EXTENTS:
 374                 /* If the extents fit in the inode, fix the
 375                  * pointer.  Otherwise it's already NULL or
 376                  * pointing to the extent.
 377                  */
 378                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 379                         ifp->if_u1.if_extents =
 380                                 ifp->if_u2.if_inline_ext;
 381                 }
 382                 ilf_fields |= XFS_ILOG_DEXT;
 383                 break;
 384         case XFS_DINODE_FMT_BTREE:
 385                 ilf_fields |= XFS_ILOG_DBROOT;
 386                 break;
 387         }
 388
 389         tilf_fields = XFS_ILOG_CORE;
 390
 391         switch(tip->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 (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 398                         tifp->if_u1.if_extents =
 399                                 tifp->if_u2.if_inline_ext;
 400                 }
 401                 tilf_fields |= XFS_ILOG_DEXT;
 402                 break;
 403         case XFS_DINODE_FMT_BTREE:
 404                 tilf_fields |= XFS_ILOG_DBROOT;
 405                 break;
 406         }
 407
 408
 409         IHOLD(ip);
 410         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 411
 412         IHOLD(tip);
 413         xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 414
 415         xfs_trans_log_inode(tp, ip,  ilf_fields);
 416         xfs_trans_log_inode(tp, tip, tilf_fields);
 417
 418         /*
 419          * If this is a synchronous mount, make sure that the
 420          * transaction goes to disk before returning to the user.
 421          */
 422         if (mp->m_flags & XFS_MOUNT_WSYNC)
 423                 xfs_trans_set_sync(tp);
 424
 425         error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
 426
 427 out:
 428         kmem_free(tempifp);
 429         return error;
 430
 431 out_unlock:
 432         xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 433         xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 434         goto out;
 435
 436 out_trans_cancel:
 437         xfs_trans_cancel(tp, 0);
 438         goto out_unlock;
 439 }