fs/xfs/xfs_dfrag.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 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_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir_sf.h"
  35 #include "xfs_dir2_sf.h"
  36 #include "xfs_attr_sf.h"
  37 #include "xfs_dinode.h"
  38 #include "xfs_inode.h"
  39 #include "xfs_inode_item.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_ialloc.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_dfrag.h"
  45 #include "xfs_error.h"
  46 #include "xfs_mac.h"
  47 #include "xfs_rw.h"
  48
  49 /*
  50  * Syssgi interface for swapext
  51  */
  52 int
  53 xfs_swapext(
  54         xfs_swapext_t   __user *sxu)
  55 {
  56         xfs_swapext_t   *sxp;
  57         xfs_inode_t     *ip=NULL, *tip=NULL, *ips[2];
  58         xfs_trans_t     *tp;
  59         xfs_mount_t     *mp;
  60         xfs_bstat_t     *sbp;
  61         struct file     *fp = NULL, *tfp = NULL;
  62         vnode_t         *vp, *tvp;
  63         static uint     lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
  64         int             ilf_fields, tilf_fields;
  65         int             error = 0;
  66         xfs_ifork_t     *tempifp, *ifp, *tifp;
  67         __uint64_t      tmp;
  68         int             aforkblks = 0;
  69         int             taforkblks = 0;
  70         char            locked = 0;
  71
  72         sxp = kmem_alloc(sizeof(xfs_swapext_t), KM_MAYFAIL);
  73         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
  74         if (!sxp || !tempifp) {
  75                 error = XFS_ERROR(ENOMEM);
  76                 goto error0;
  77         }
  78
  79         if (copy_from_user(sxp, sxu, sizeof(xfs_swapext_t))) {
  80                 error = XFS_ERROR(EFAULT);
  81                 goto error0;
  82         }
  83
  84         /* Pull information for the target fd */
  85         if (((fp = fget((int)sxp->sx_fdtarget)) == NULL) ||
  86             ((vp = LINVFS_GET_VP(fp->f_dentry->d_inode)) == NULL))  {
  87                 error = XFS_ERROR(EINVAL);
  88                 goto error0;
  89         }
  90
  91         ip = xfs_vtoi(vp);
  92         if (ip == NULL) {
  93                 error = XFS_ERROR(EBADF);
  94                 goto error0;
  95         }
  96
  97         if (((tfp = fget((int)sxp->sx_fdtmp)) == NULL) ||
  98             ((tvp = LINVFS_GET_VP(tfp->f_dentry->d_inode)) == NULL)) {
  99                 error = XFS_ERROR(EINVAL);
 100                 goto error0;
 101         }
 102
 103         tip = xfs_vtoi(tvp);
 104         if (tip == NULL) {
 105                 error = XFS_ERROR(EBADF);
 106                 goto error0;
 107         }
 108
 109         if (ip->i_mount != tip->i_mount) {
 110                 error =  XFS_ERROR(EINVAL);
 111                 goto error0;
 112         }
 113
 114         if (ip->i_ino == tip->i_ino) {
 115                 error =  XFS_ERROR(EINVAL);
 116                 goto error0;
 117         }
 118
 119         mp = ip->i_mount;
 120
 121         sbp = &sxp->sx_stat;
 122
 123         if (XFS_FORCED_SHUTDOWN(mp)) {
 124                 error =  XFS_ERROR(EIO);
 125                 goto error0;
 126         }
 127
 128         locked = 1;
 129
 130         /* Lock in i_ino order */
 131         if (ip->i_ino < tip->i_ino) {
 132                 ips[0] = ip;
 133                 ips[1] = tip;
 134         } else {
 135                 ips[0] = tip;
 136                 ips[1] = ip;
 137         }
 138
 139         xfs_lock_inodes(ips, 2, 0, lock_flags);
 140
 141         /* Check permissions */
 142         error = xfs_iaccess(ip, S_IWUSR, NULL);
 143         if (error)
 144                 goto error0;
 145
 146         error = xfs_iaccess(tip, S_IWUSR, NULL);
 147         if (error)
 148                 goto error0;
 149
 150         /* Verify that both files have the same format */
 151         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 152                 error = XFS_ERROR(EINVAL);
 153                 goto error0;
 154         }
 155
 156         /* Verify both files are either real-time or non-realtime */
 157         if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
 158             (tip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
 159                 error = XFS_ERROR(EINVAL);
 160                 goto error0;
 161         }
 162
 163         /* Should never get a local format */
 164         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 165             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
 166                 error = XFS_ERROR(EINVAL);
 167                 goto error0;
 168         }
 169
 170         if (VN_CACHED(tvp) != 0) {
 171                 xfs_inval_cached_trace(&tip->i_iocore, 0, -1, 0, -1);
 172                 VOP_FLUSHINVAL_PAGES(tvp, 0, -1, FI_REMAPF_LOCKED);
 173         }
 174
 175         /* Verify O_DIRECT for ftmp */
 176         if (VN_CACHED(tvp) != 0) {
 177                 error = XFS_ERROR(EINVAL);
 178                 goto error0;
 179         }
 180
 181         /* Verify all data are being swapped */
 182         if (sxp->sx_offset != 0 ||
 183             sxp->sx_length != ip->i_d.di_size ||
 184             sxp->sx_length != tip->i_d.di_size) {
 185                 error = XFS_ERROR(EFAULT);
 186                 goto error0;
 187         }
 188
 189         /*
 190          * If the target has extended attributes, the tmp file
 191          * must also in order to ensure the correct data fork
 192          * format.
 193          */
 194         if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
 195                 error = XFS_ERROR(EINVAL);
 196                 goto error0;
 197         }
 198
 199         /*
 200          * Compare the current change & modify times with that
 201          * passed in.  If they differ, we abort this swap.
 202          * This is the mechanism used to ensure the calling
 203          * process that the file was not changed out from
 204          * under it.
 205          */
 206         if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
 207             (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
 208             (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
 209             (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
 210                 error = XFS_ERROR(EBUSY);
 211                 goto error0;
 212         }
 213
 214         /* We need to fail if the file is memory mapped.  Once we have tossed
 215          * all existing pages, the page fault will have no option
 216          * but to go to the filesystem for pages. By making the page fault call
 217          * VOP_READ (or write in the case of autogrow) they block on the iolock
 218          * until we have switched the extents.
 219          */
 220         if (VN_MAPPED(vp)) {
 221                 error = XFS_ERROR(EBUSY);
 222                 goto error0;
 223         }
 224
 225         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 226         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 227
 228         /*
 229          * There is a race condition here since we gave up the
 230          * ilock.  However, the data fork will not change since
 231          * we have the iolock (locked for truncation too) so we
 232          * are safe.  We don't really care if non-io related
 233          * fields change.
 234          */
 235
 236         VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
 237
 238         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 239         if ((error = xfs_trans_reserve(tp, 0,
 240                                      XFS_ICHANGE_LOG_RES(mp), 0,
 241                                      0, 0))) {
 242                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 243                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 244                 xfs_trans_cancel(tp, 0);
 245                 locked = 0;
 246                 goto error0;
 247         }
 248         xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
 249
 250         /*
 251          * Count the number of extended attribute blocks
 252          */
 253         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 254              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 255                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 256                 if (error) {
 257                         xfs_trans_cancel(tp, 0);
 258                         goto error0;
 259                 }
 260         }
 261         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 262              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 263                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 264                         &taforkblks);
 265                 if (error) {
 266                         xfs_trans_cancel(tp, 0);
 267                         goto error0;
 268                 }
 269         }
 270
 271         /*
 272          * Swap the data forks of the inodes
 273          */
 274         ifp = &ip->i_df;
 275         tifp = &tip->i_df;
 276         *tempifp = *ifp;        /* struct copy */
 277         *ifp = *tifp;           /* struct copy */
 278         *tifp = *tempifp;       /* struct copy */
 279
 280         /*
 281          * Fix the on-disk inode values
 282          */
 283         tmp = (__uint64_t)ip->i_d.di_nblocks;
 284         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 285         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 286
 287         tmp = (__uint64_t) ip->i_d.di_nextents;
 288         ip->i_d.di_nextents = tip->i_d.di_nextents;
 289         tip->i_d.di_nextents = tmp;
 290
 291         tmp = (__uint64_t) ip->i_d.di_format;
 292         ip->i_d.di_format = tip->i_d.di_format;
 293         tip->i_d.di_format = tmp;
 294
 295         ilf_fields = XFS_ILOG_CORE;
 296
 297         switch(ip->i_d.di_format) {
 298         case XFS_DINODE_FMT_EXTENTS:
 299                 /* If the extents fit in the inode, fix the
 300                  * pointer.  Otherwise it's already NULL or
 301                  * pointing to the extent.
 302                  */
 303                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 304                         ifp->if_u1.if_extents =
 305                                 ifp->if_u2.if_inline_ext;
 306                 }
 307                 ilf_fields |= XFS_ILOG_DEXT;
 308                 break;
 309         case XFS_DINODE_FMT_BTREE:
 310                 ilf_fields |= XFS_ILOG_DBROOT;
 311                 break;
 312         }
 313
 314         tilf_fields = XFS_ILOG_CORE;
 315
 316         switch(tip->i_d.di_format) {
 317         case XFS_DINODE_FMT_EXTENTS:
 318                 /* If the extents fit in the inode, fix the
 319                  * pointer.  Otherwise it's already NULL or
 320                  * pointing to the extent.
 321                  */
 322                 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 323                         tifp->if_u1.if_extents =
 324                                 tifp->if_u2.if_inline_ext;
 325                 }
 326                 tilf_fields |= XFS_ILOG_DEXT;
 327                 break;
 328         case XFS_DINODE_FMT_BTREE:
 329                 tilf_fields |= XFS_ILOG_DBROOT;
 330                 break;
 331         }
 332
 333         /*
 334          * Increment vnode ref counts since xfs_trans_commit &
 335          * xfs_trans_cancel will both unlock the inodes and
 336          * decrement the associated ref counts.
 337          */
 338         VN_HOLD(vp);
 339         VN_HOLD(tvp);
 340
 341         xfs_trans_ijoin(tp, ip, lock_flags);
 342         xfs_trans_ijoin(tp, tip, lock_flags);
 343
 344         xfs_trans_log_inode(tp, ip,  ilf_fields);
 345         xfs_trans_log_inode(tp, tip, tilf_fields);
 346
 347         /*
 348          * If this is a synchronous mount, make sure that the
 349          * transaction goes to disk before returning to the user.
 350          */
 351         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 352                 xfs_trans_set_sync(tp);
 353         }
 354
 355         error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT, NULL);
 356         locked = 0;
 357
 358  error0:
 359         if (locked) {
 360                 xfs_iunlock(ip,  lock_flags);
 361                 xfs_iunlock(tip, lock_flags);
 362         }
 363
 364         if (fp != NULL)
 365                 fput(fp);
 366         if (tfp != NULL)
 367                 fput(tfp);
 368
 369         if (sxp != NULL)
 370                 kmem_free(sxp, sizeof(xfs_swapext_t));
 371         if (tempifp != NULL)
 372                 kmem_free(tempifp, sizeof(xfs_ifork_t));
 373
 374         return error;
 375 }