linux/fs/xfs/xfs_dfrag.c
<<
>>
Prefs
   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_log.h"
  22#include "xfs_trans.h"
  23#include "xfs_sb.h"
  24#include "xfs_ag.h"
  25#include "xfs_mount.h"
  26#include "xfs_bmap_btree.h"
  27#include "xfs_dinode.h"
  28#include "xfs_inode.h"
  29#include "xfs_inode_item.h"
  30#include "xfs_bmap.h"
  31#include "xfs_itable.h"
  32#include "xfs_dfrag.h"
  33#include "xfs_error.h"
  34#include "xfs_vnodeops.h"
  35#include "xfs_trace.h"
  36
  37
  38static int xfs_swap_extents(
  39        xfs_inode_t     *ip,    /* target inode */
  40        xfs_inode_t     *tip,   /* tmp inode */
  41        xfs_swapext_t   *sxp);
  42
  43/*
  44 * ioctl interface for swapext
  45 */
  46int
  47xfs_swapext(
  48        xfs_swapext_t   *sxp)
  49{
  50        xfs_inode_t     *ip, *tip;
  51        struct fd       f, tmp;
  52        int             error = 0;
  53
  54        /* Pull information for the target fd */
  55        f = fdget((int)sxp->sx_fdtarget);
  56        if (!f.file) {
  57                error = XFS_ERROR(EINVAL);
  58                goto out;
  59        }
  60
  61        if (!(f.file->f_mode & FMODE_WRITE) ||
  62            !(f.file->f_mode & FMODE_READ) ||
  63            (f.file->f_flags & O_APPEND)) {
  64                error = XFS_ERROR(EBADF);
  65                goto out_put_file;
  66        }
  67
  68        tmp = fdget((int)sxp->sx_fdtmp);
  69        if (!tmp.file) {
  70                error = XFS_ERROR(EINVAL);
  71                goto out_put_file;
  72        }
  73
  74        if (!(tmp.file->f_mode & FMODE_WRITE) ||
  75            !(tmp.file->f_mode & FMODE_READ) ||
  76            (tmp.file->f_flags & O_APPEND)) {
  77                error = XFS_ERROR(EBADF);
  78                goto out_put_tmp_file;
  79        }
  80
  81        if (IS_SWAPFILE(f.file->f_path.dentry->d_inode) ||
  82            IS_SWAPFILE(tmp.file->f_path.dentry->d_inode)) {
  83                error = XFS_ERROR(EINVAL);
  84                goto out_put_tmp_file;
  85        }
  86
  87        ip = XFS_I(f.file->f_path.dentry->d_inode);
  88        tip = XFS_I(tmp.file->f_path.dentry->d_inode);
  89
  90        if (ip->i_mount != tip->i_mount) {
  91                error = XFS_ERROR(EINVAL);
  92                goto out_put_tmp_file;
  93        }
  94
  95        if (ip->i_ino == tip->i_ino) {
  96                error = XFS_ERROR(EINVAL);
  97                goto out_put_tmp_file;
  98        }
  99
 100        if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 101                error = XFS_ERROR(EIO);
 102                goto out_put_tmp_file;
 103        }
 104
 105        error = xfs_swap_extents(ip, tip, sxp);
 106
 107 out_put_tmp_file:
 108        fdput(tmp);
 109 out_put_file:
 110        fdput(f);
 111 out:
 112        return error;
 113}
 114
 115/*
 116 * We need to check that the format of the data fork in the temporary inode is
 117 * valid for the target inode before doing the swap. This is not a problem with
 118 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 119 * data fork depending on the space the attribute fork is taking so we can get
 120 * invalid formats on the target inode.
 121 *
 122 * E.g. target has space for 7 extents in extent format, temp inode only has
 123 * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 124 * btree, but when swapped it needs to be in extent format. Hence we can't just
 125 * blindly swap data forks on attr2 filesystems.
 126 *
 127 * Note that we check the swap in both directions so that we don't end up with
 128 * a corrupt temporary inode, either.
 129 *
 130 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 131 * inode will prevent this situation from occurring, so all we do here is
 132 * reject and log the attempt. basically we are putting the responsibility on
 133 * userspace to get this right.
 134 */
 135static int
 136xfs_swap_extents_check_format(
 137        xfs_inode_t     *ip,    /* target inode */
 138        xfs_inode_t     *tip)   /* tmp inode */
 139{
 140
 141        /* Should never get a local format */
 142        if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 143            tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 144                return EINVAL;
 145
 146        /*
 147         * if the target inode has less extents that then temporary inode then
 148         * why did userspace call us?
 149         */
 150        if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 151                return EINVAL;
 152
 153        /*
 154         * if the target inode is in extent form and the temp inode is in btree
 155         * form then we will end up with the target inode in the wrong format
 156         * as we already know there are less extents in the temp inode.
 157         */
 158        if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 159            tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 160                return EINVAL;
 161
 162        /* Check temp in extent form to max in target */
 163        if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 164            XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
 165                        XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
 166                return EINVAL;
 167
 168        /* Check target in extent form to max in temp */
 169        if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 170            XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
 171                        XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
 172                return EINVAL;
 173
 174        /*
 175         * If we are in a btree format, check that the temp root block will fit
 176         * in the target and that it has enough extents to be in btree format
 177         * in the target.
 178         *
 179         * Note that we have to be careful to allow btree->extent conversions
 180         * (a common defrag case) which will occur when the temp inode is in
 181         * extent format...
 182         */
 183        if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
 184                if (XFS_IFORK_BOFF(ip) &&
 185                    tip->i_df.if_broot_bytes > XFS_IFORK_BOFF(ip))
 186                        return EINVAL;
 187                if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
 188                    XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
 189                        return EINVAL;
 190        }
 191
 192        /* Reciprocal target->temp btree format checks */
 193        if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
 194                if (XFS_IFORK_BOFF(tip) &&
 195                    ip->i_df.if_broot_bytes > XFS_IFORK_BOFF(tip))
 196                        return EINVAL;
 197
 198                if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
 199                    XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
 200                        return EINVAL;
 201        }
 202
 203        return 0;
 204}
 205
 206static int
 207xfs_swap_extents(
 208        xfs_inode_t     *ip,    /* target inode */
 209        xfs_inode_t     *tip,   /* tmp inode */
 210        xfs_swapext_t   *sxp)
 211{
 212        xfs_mount_t     *mp = ip->i_mount;
 213        xfs_trans_t     *tp;
 214        xfs_bstat_t     *sbp = &sxp->sx_stat;
 215        xfs_ifork_t     *tempifp, *ifp, *tifp;
 216        int             src_log_flags, target_log_flags;
 217        int             error = 0;
 218        int             aforkblks = 0;
 219        int             taforkblks = 0;
 220        __uint64_t      tmp;
 221
 222        tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 223        if (!tempifp) {
 224                error = XFS_ERROR(ENOMEM);
 225                goto out;
 226        }
 227
 228        /*
 229         * we have to do two separate lock calls here to keep lockdep
 230         * happy. If we try to get all the locks in one call, lock will
 231         * report false positives when we drop the ILOCK and regain them
 232         * below.
 233         */
 234        xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 235        xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 236
 237        /* Verify that both files have the same format */
 238        if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 239                error = XFS_ERROR(EINVAL);
 240                goto out_unlock;
 241        }
 242
 243        /* Verify both files are either real-time or non-realtime */
 244        if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 245                error = XFS_ERROR(EINVAL);
 246                goto out_unlock;
 247        }
 248
 249        error = -filemap_write_and_wait(VFS_I(tip)->i_mapping);
 250        if (error)
 251                goto out_unlock;
 252        truncate_pagecache_range(VFS_I(tip), 0, -1);
 253
 254        /* Verify O_DIRECT for ftmp */
 255        if (VN_CACHED(VFS_I(tip)) != 0) {
 256                error = XFS_ERROR(EINVAL);
 257                goto out_unlock;
 258        }
 259
 260        /* Verify all data are being swapped */
 261        if (sxp->sx_offset != 0 ||
 262            sxp->sx_length != ip->i_d.di_size ||
 263            sxp->sx_length != tip->i_d.di_size) {
 264                error = XFS_ERROR(EFAULT);
 265                goto out_unlock;
 266        }
 267
 268        trace_xfs_swap_extent_before(ip, 0);
 269        trace_xfs_swap_extent_before(tip, 1);
 270
 271        /* check inode formats now that data is flushed */
 272        error = xfs_swap_extents_check_format(ip, tip);
 273        if (error) {
 274                xfs_notice(mp,
 275                    "%s: inode 0x%llx format is incompatible for exchanging.",
 276                                __func__, ip->i_ino);
 277                goto out_unlock;
 278        }
 279
 280        /*
 281         * Compare the current change & modify times with that
 282         * passed in.  If they differ, we abort this swap.
 283         * This is the mechanism used to ensure the calling
 284         * process that the file was not changed out from
 285         * under it.
 286         */
 287        if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 288            (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 289            (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 290            (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 291                error = XFS_ERROR(EBUSY);
 292                goto out_unlock;
 293        }
 294
 295        /* We need to fail if the file is memory mapped.  Once we have tossed
 296         * all existing pages, the page fault will have no option
 297         * but to go to the filesystem for pages. By making the page fault call
 298         * vop_read (or write in the case of autogrow) they block on the iolock
 299         * until we have switched the extents.
 300         */
 301        if (VN_MAPPED(VFS_I(ip))) {
 302                error = XFS_ERROR(EBUSY);
 303                goto out_unlock;
 304        }
 305
 306        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 307        xfs_iunlock(tip, XFS_ILOCK_EXCL);
 308
 309        /*
 310         * There is a race condition here since we gave up the
 311         * ilock.  However, the data fork will not change since
 312         * we have the iolock (locked for truncation too) so we
 313         * are safe.  We don't really care if non-io related
 314         * fields change.
 315         */
 316        truncate_pagecache_range(VFS_I(ip), 0, -1);
 317
 318        tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 319        if ((error = xfs_trans_reserve(tp, 0,
 320                                     XFS_ICHANGE_LOG_RES(mp), 0,
 321                                     0, 0))) {
 322                xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 323                xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 324                xfs_trans_cancel(tp, 0);
 325                goto out;
 326        }
 327        xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 328
 329        /*
 330         * Count the number of extended attribute blocks
 331         */
 332        if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 333             (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 334                error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 335                if (error)
 336                        goto out_trans_cancel;
 337        }
 338        if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 339             (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 340                error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 341                        &taforkblks);
 342                if (error)
 343                        goto out_trans_cancel;
 344        }
 345
 346        /*
 347         * Swap the data forks of the inodes
 348         */
 349        ifp = &ip->i_df;
 350        tifp = &tip->i_df;
 351        *tempifp = *ifp;        /* struct copy */
 352        *ifp = *tifp;           /* struct copy */
 353        *tifp = *tempifp;       /* struct copy */
 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        /*
 371         * The extents in the source inode could still contain speculative
 372         * preallocation beyond EOF (e.g. the file is open but not modified
 373         * while defrag is in progress). In that case, we need to copy over the
 374         * number of delalloc blocks the data fork in the source inode is
 375         * tracking beyond EOF so that when the fork is truncated away when the
 376         * temporary inode is unlinked we don't underrun the i_delayed_blks
 377         * counter on that inode.
 378         */
 379        ASSERT(tip->i_delayed_blks == 0);
 380        tip->i_delayed_blks = ip->i_delayed_blks;
 381        ip->i_delayed_blks = 0;
 382
 383        src_log_flags = XFS_ILOG_CORE;
 384        switch (ip->i_d.di_format) {
 385        case XFS_DINODE_FMT_EXTENTS:
 386                /* If the extents fit in the inode, fix the
 387                 * pointer.  Otherwise it's already NULL or
 388                 * pointing to the extent.
 389                 */
 390                if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 391                        ifp->if_u1.if_extents =
 392                                ifp->if_u2.if_inline_ext;
 393                }
 394                src_log_flags |= XFS_ILOG_DEXT;
 395                break;
 396        case XFS_DINODE_FMT_BTREE:
 397                src_log_flags |= XFS_ILOG_DBROOT;
 398                break;
 399        }
 400
 401        target_log_flags = XFS_ILOG_CORE;
 402        switch (tip->i_d.di_format) {
 403        case XFS_DINODE_FMT_EXTENTS:
 404                /* If the extents fit in the inode, fix the
 405                 * pointer.  Otherwise it's already NULL or
 406                 * pointing to the extent.
 407                 */
 408                if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 409                        tifp->if_u1.if_extents =
 410                                tifp->if_u2.if_inline_ext;
 411                }
 412                target_log_flags |= XFS_ILOG_DEXT;
 413                break;
 414        case XFS_DINODE_FMT_BTREE:
 415                target_log_flags |= XFS_ILOG_DBROOT;
 416                break;
 417        }
 418
 419
 420        xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 421        xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 422
 423        xfs_trans_log_inode(tp, ip,  src_log_flags);
 424        xfs_trans_log_inode(tp, tip, target_log_flags);
 425
 426        /*
 427         * If this is a synchronous mount, make sure that the
 428         * transaction goes to disk before returning to the user.
 429         */
 430        if (mp->m_flags & XFS_MOUNT_WSYNC)
 431                xfs_trans_set_sync(tp);
 432
 433        error = xfs_trans_commit(tp, 0);
 434
 435        trace_xfs_swap_extent_after(ip, 0);
 436        trace_xfs_swap_extent_after(tip, 1);
 437out:
 438        kmem_free(tempifp);
 439        return error;
 440
 441out_unlock:
 442        xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 443        xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 444        goto out;
 445
 446out_trans_cancel:
 447        xfs_trans_cancel(tp, 0);
 448        goto out_unlock;
 449}
 450
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.