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   __user *sxu)
  53 {
  54         xfs_swapext_t   *sxp;
  55         xfs_inode_t     *ip, *tip;
  56         struct file     *file, *target_file;
  57         int             error = 0;
  58
  59         sxp = kmem_alloc(sizeof(xfs_swapext_t), KM_MAYFAIL);
  60         if (!sxp) {
  61                 error = XFS_ERROR(ENOMEM);
  62                 goto out;
  63         }
  64
  65         if (copy_from_user(sxp, sxu, sizeof(xfs_swapext_t))) {
  66                 error = XFS_ERROR(EFAULT);
  67                 goto out_free_sxp;
  68         }
  69
  70         /* Pull information for the target fd */
  71         file = fget((int)sxp->sx_fdtarget);
  72         if (!file) {
  73                 error = XFS_ERROR(EINVAL);
  74                 goto out_free_sxp;
  75         }
  76
  77         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) {
  78                 error = XFS_ERROR(EBADF);
  79                 goto out_put_file;
  80         }
  81
  82         target_file = fget((int)sxp->sx_fdtmp);
  83         if (!target_file) {
  84                 error = XFS_ERROR(EINVAL);
  85                 goto out_put_file;
  86         }
  87
  88         if (!(target_file->f_mode & FMODE_WRITE) ||
  89             (target_file->f_flags & O_APPEND)) {
  90                 error = XFS_ERROR(EBADF);
  91                 goto out_put_target_file;
  92         }
  93
  94         ip = XFS_I(file->f_path.dentry->d_inode);
  95         tip = XFS_I(target_file->f_path.dentry->d_inode);
  96
  97         if (ip->i_mount != tip->i_mount) {
  98                 error = XFS_ERROR(EINVAL);
  99                 goto out_put_target_file;
 100         }
 101
 102         if (ip->i_ino == tip->i_ino) {
 103                 error = XFS_ERROR(EINVAL);
 104                 goto out_put_target_file;
 105         }
 106
 107         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 108                 error = XFS_ERROR(EIO);
 109                 goto out_put_target_file;
 110         }
 111
 112         error = xfs_swap_extents(ip, tip, sxp);
 113
 114  out_put_target_file:
 115         fput(target_file);
 116  out_put_file:
 117         fput(file);
 118  out_free_sxp:
 119         kmem_free(sxp, sizeof(xfs_swapext_t));
 120  out:
 121         return error;
 122 }
 123
 124 int
 125 xfs_swap_extents(
 126         xfs_inode_t     *ip,
 127         xfs_inode_t     *tip,
 128         xfs_swapext_t   *sxp)
 129 {
 130         xfs_mount_t     *mp;
 131         xfs_inode_t     *ips[2];
 132         xfs_trans_t     *tp;
 133         xfs_bstat_t     *sbp = &sxp->sx_stat;
 134         bhv_vnode_t     *vp, *tvp;
 135         xfs_ifork_t     *tempifp, *ifp, *tifp;
 136         int             ilf_fields, tilf_fields;
 137         static uint     lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
 138         int             error = 0;
 139         int             aforkblks = 0;
 140         int             taforkblks = 0;
 141         __uint64_t      tmp;
 142         char            locked = 0;
 143
 144         mp = ip->i_mount;
 145
 146         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 147         if (!tempifp) {
 148                 error = XFS_ERROR(ENOMEM);
 149                 goto error0;
 150         }
 151
 152         sbp = &sxp->sx_stat;
 153         vp = XFS_ITOV(ip);
 154         tvp = XFS_ITOV(tip);
 155
 156         /* Lock in i_ino order */
 157         if (ip->i_ino < tip->i_ino) {
 158                 ips[0] = ip;
 159                 ips[1] = tip;
 160         } else {
 161                 ips[0] = tip;
 162                 ips[1] = ip;
 163         }
 164
 165         xfs_lock_inodes(ips, 2, 0, lock_flags);
 166         locked = 1;
 167
 168         /* Verify that both files have the same format */
 169         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 170                 error = XFS_ERROR(EINVAL);
 171                 goto error0;
 172         }
 173
 174         /* Verify both files are either real-time or non-realtime */
 175         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 176                 error = XFS_ERROR(EINVAL);
 177                 goto error0;
 178         }
 179
 180         /* Should never get a local format */
 181         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 182             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
 183                 error = XFS_ERROR(EINVAL);
 184                 goto error0;
 185         }
 186
 187         if (VN_CACHED(tvp) != 0) {
 188                 xfs_inval_cached_trace(tip, 0, -1, 0, -1);
 189                 error = xfs_flushinval_pages(tip, 0, -1,
 190                                 FI_REMAPF_LOCKED);
 191                 if (error)
 192                         goto error0;
 193         }
 194
 195         /* Verify O_DIRECT for ftmp */
 196         if (VN_CACHED(tvp) != 0) {
 197                 error = XFS_ERROR(EINVAL);
 198                 goto error0;
 199         }
 200
 201         /* Verify all data are being swapped */
 202         if (sxp->sx_offset != 0 ||
 203             sxp->sx_length != ip->i_d.di_size ||
 204             sxp->sx_length != tip->i_d.di_size) {
 205                 error = XFS_ERROR(EFAULT);
 206                 goto error0;
 207         }
 208
 209         /*
 210          * If the target has extended attributes, the tmp file
 211          * must also in order to ensure the correct data fork
 212          * format.
 213          */
 214         if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
 215                 error = XFS_ERROR(EINVAL);
 216                 goto error0;
 217         }
 218
 219         /*
 220          * Compare the current change & modify times with that
 221          * passed in.  If they differ, we abort this swap.
 222          * This is the mechanism used to ensure the calling
 223          * process that the file was not changed out from
 224          * under it.
 225          */
 226         if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
 227             (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
 228             (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
 229             (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
 230                 error = XFS_ERROR(EBUSY);
 231                 goto error0;
 232         }
 233
 234         /* We need to fail if the file is memory mapped.  Once we have tossed
 235          * all existing pages, the page fault will have no option
 236          * but to go to the filesystem for pages. By making the page fault call
 237          * vop_read (or write in the case of autogrow) they block on the iolock
 238          * until we have switched the extents.
 239          */
 240         if (VN_MAPPED(vp)) {
 241                 error = XFS_ERROR(EBUSY);
 242                 goto error0;
 243         }
 244
 245         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 246         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 247
 248         /*
 249          * There is a race condition here since we gave up the
 250          * ilock.  However, the data fork will not change since
 251          * we have the iolock (locked for truncation too) so we
 252          * are safe.  We don't really care if non-io related
 253          * fields change.
 254          */
 255
 256         xfs_tosspages(ip, 0, -1, FI_REMAPF);
 257
 258         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 259         if ((error = xfs_trans_reserve(tp, 0,
 260                                      XFS_ICHANGE_LOG_RES(mp), 0,
 261                                      0, 0))) {
 262                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 263                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 264                 xfs_trans_cancel(tp, 0);
 265                 locked = 0;
 266                 goto error0;
 267         }
 268         xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
 269
 270         /*
 271          * Count the number of extended attribute blocks
 272          */
 273         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 274              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 275                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 276                 if (error) {
 277                         xfs_trans_cancel(tp, 0);
 278                         goto error0;
 279                 }
 280         }
 281         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 282              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 283                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 284                         &taforkblks);
 285                 if (error) {
 286                         xfs_trans_cancel(tp, 0);
 287                         goto error0;
 288                 }
 289         }
 290
 291         /*
 292          * Swap the data forks of the inodes
 293          */
 294         ifp = &ip->i_df;
 295         tifp = &tip->i_df;
 296         *tempifp = *ifp;        /* struct copy */
 297         *ifp = *tifp;           /* struct copy */
 298         *tifp = *tempifp;       /* struct copy */
 299
 300         /*
 301          * Fix the on-disk inode values
 302          */
 303         tmp = (__uint64_t)ip->i_d.di_nblocks;
 304         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 305         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 306
 307         tmp = (__uint64_t) ip->i_d.di_nextents;
 308         ip->i_d.di_nextents = tip->i_d.di_nextents;
 309         tip->i_d.di_nextents = tmp;
 310
 311         tmp = (__uint64_t) ip->i_d.di_format;
 312         ip->i_d.di_format = tip->i_d.di_format;
 313         tip->i_d.di_format = tmp;
 314
 315         ilf_fields = XFS_ILOG_CORE;
 316
 317         switch(ip->i_d.di_format) {
 318         case XFS_DINODE_FMT_EXTENTS:
 319                 /* If the extents fit in the inode, fix the
 320                  * pointer.  Otherwise it's already NULL or
 321                  * pointing to the extent.
 322                  */
 323                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 324                         ifp->if_u1.if_extents =
 325                                 ifp->if_u2.if_inline_ext;
 326                 }
 327                 ilf_fields |= XFS_ILOG_DEXT;
 328                 break;
 329         case XFS_DINODE_FMT_BTREE:
 330                 ilf_fields |= XFS_ILOG_DBROOT;
 331                 break;
 332         }
 333
 334         tilf_fields = XFS_ILOG_CORE;
 335
 336         switch(tip->i_d.di_format) {
 337         case XFS_DINODE_FMT_EXTENTS:
 338                 /* If the extents fit in the inode, fix the
 339                  * pointer.  Otherwise it's already NULL or
 340                  * pointing to the extent.
 341                  */
 342                 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 343                         tifp->if_u1.if_extents =
 344                                 tifp->if_u2.if_inline_ext;
 345                 }
 346                 tilf_fields |= XFS_ILOG_DEXT;
 347                 break;
 348         case XFS_DINODE_FMT_BTREE:
 349                 tilf_fields |= XFS_ILOG_DBROOT;
 350                 break;
 351         }
 352
 353         /*
 354          * Increment vnode ref counts since xfs_trans_commit &
 355          * xfs_trans_cancel will both unlock the inodes and
 356          * decrement the associated ref counts.
 357          */
 358         VN_HOLD(vp);
 359         VN_HOLD(tvp);
 360
 361         xfs_trans_ijoin(tp, ip, lock_flags);
 362         xfs_trans_ijoin(tp, tip, lock_flags);
 363
 364         xfs_trans_log_inode(tp, ip,  ilf_fields);
 365         xfs_trans_log_inode(tp, tip, tilf_fields);
 366
 367         /*
 368          * If this is a synchronous mount, make sure that the
 369          * transaction goes to disk before returning to the user.
 370          */
 371         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 372                 xfs_trans_set_sync(tp);
 373         }
 374
 375         error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
 376         locked = 0;
 377
 378  error0:
 379         if (locked) {
 380                 xfs_iunlock(ip,  lock_flags);
 381                 xfs_iunlock(tip, lock_flags);
 382         }
 383         if (tempifp != NULL)
 384                 kmem_free(tempifp, sizeof(xfs_ifork_t));
 385         return error;
 386 }