fs/xfs/xfs_rw.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_itable.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_attr.h"
  43 #include "xfs_bmap.h"
  44 #include "xfs_error.h"
  45 #include "xfs_buf_item.h"
  46 #include "xfs_rw.h"
  47
  48 /*
  49  * This is a subroutine for xfs_write() and other writers (xfs_ioctl)
  50  * which clears the setuid and setgid bits when a file is written.
  51  */
  52 int
  53 xfs_write_clear_setuid(
  54         xfs_inode_t     *ip)
  55 {
  56         xfs_mount_t     *mp;
  57         xfs_trans_t     *tp;
  58         int             error;
  59
  60         mp = ip->i_mount;
  61         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
  62         if ((error = xfs_trans_reserve(tp, 0,
  63                                       XFS_WRITEID_LOG_RES(mp),
  64                                       0, 0, 0))) {
  65                 xfs_trans_cancel(tp, 0);
  66                 return error;
  67         }
  68         xfs_ilock(ip, XFS_ILOCK_EXCL);
  69         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
  70         xfs_trans_ihold(tp, ip);
  71         ip->i_d.di_mode &= ~S_ISUID;
  72
  73         /*
  74          * Note that we don't have to worry about mandatory
  75          * file locking being disabled here because we only
  76          * clear the S_ISGID bit if the Group execute bit is
  77          * on, but if it was on then mandatory locking wouldn't
  78          * have been enabled.
  79          */
  80         if (ip->i_d.di_mode & S_IXGRP) {
  81                 ip->i_d.di_mode &= ~S_ISGID;
  82         }
  83         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  84         xfs_trans_set_sync(tp);
  85         error = xfs_trans_commit(tp, 0);
  86         xfs_iunlock(ip, XFS_ILOCK_EXCL);
  87         return 0;
  88 }
  89
  90 /*
  91  * Handle logging requirements of various synchronous types of write.
  92  */
  93 int
  94 xfs_write_sync_logforce(
  95         xfs_mount_t     *mp,
  96         xfs_inode_t     *ip)
  97 {
  98         int             error = 0;
  99
 100         /*
 101          * If we're treating this as O_DSYNC and we have not updated the
 102          * size, force the log.
 103          */
 104         if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC) &&
 105             !(ip->i_update_size)) {
 106                 xfs_inode_log_item_t    *iip = ip->i_itemp;
 107
 108                 /*
 109                  * If an allocation transaction occurred
 110                  * without extending the size, then we have to force
 111                  * the log up the proper point to ensure that the
 112                  * allocation is permanent.  We can't count on
 113                  * the fact that buffered writes lock out direct I/O
 114                  * writes - the direct I/O write could have extended
 115                  * the size nontransactionally, then finished before
 116                  * we started.  xfs_write_file will think that the file
 117                  * didn't grow but the update isn't safe unless the
 118                  * size change is logged.
 119                  *
 120                  * Force the log if we've committed a transaction
 121                  * against the inode or if someone else has and
 122                  * the commit record hasn't gone to disk (e.g.
 123                  * the inode is pinned).  This guarantees that
 124                  * all changes affecting the inode are permanent
 125                  * when we return.
 126                  */
 127                 if (iip && iip->ili_last_lsn) {
 128                         error = _xfs_log_force(mp, iip->ili_last_lsn,
 129                                         XFS_LOG_FORCE | XFS_LOG_SYNC, NULL);
 130                 } else if (xfs_ipincount(ip) > 0) {
 131                         error = _xfs_log_force(mp, (xfs_lsn_t)0,
 132                                         XFS_LOG_FORCE | XFS_LOG_SYNC, NULL);
 133                 }
 134
 135         } else {
 136                 xfs_trans_t     *tp;
 137
 138                 /*
 139                  * O_SYNC or O_DSYNC _with_ a size update are handled
 140                  * the same way.
 141                  *
 142                  * If the write was synchronous then we need to make
 143                  * sure that the inode modification time is permanent.
 144                  * We'll have updated the timestamp above, so here
 145                  * we use a synchronous transaction to log the inode.
 146                  * It's not fast, but it's necessary.
 147                  *
 148                  * If this a dsync write and the size got changed
 149                  * non-transactionally, then we need to ensure that
 150                  * the size change gets logged in a synchronous
 151                  * transaction.
 152                  */
 153                 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
 154                 if ((error = xfs_trans_reserve(tp, 0,
 155                                                 XFS_SWRITE_LOG_RES(mp),
 156                                                 0, 0, 0))) {
 157                         /* Transaction reserve failed */
 158                         xfs_trans_cancel(tp, 0);
 159                 } else {
 160                         /* Transaction reserve successful */
 161                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 162                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 163                         xfs_trans_ihold(tp, ip);
 164                         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 165                         xfs_trans_set_sync(tp);
 166                         error = xfs_trans_commit(tp, 0);
 167                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 168                 }
 169         }
 170
 171         return error;
 172 }
 173
 174 /*
 175  * Force a shutdown of the filesystem instantly while keeping
 176  * the filesystem consistent. We don't do an unmount here; just shutdown
 177  * the shop, make sure that absolutely nothing persistent happens to
 178  * this filesystem after this point.
 179  */
 180 void
 181 xfs_do_force_shutdown(
 182         xfs_mount_t     *mp,
 183         int             flags,
 184         char            *fname,
 185         int             lnnum)
 186 {
 187         int             logerror;
 188
 189         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 190
 191         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 192                 cmn_err(CE_NOTE, "xfs_force_shutdown(%s,0x%x) called from "
 193                                  "line %d of file %s.  Return address = 0x%p",
 194                         mp->m_fsname, flags, lnnum, fname, __return_address);
 195         }
 196         /*
 197          * No need to duplicate efforts.
 198          */
 199         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 200                 return;
 201
 202         /*
 203          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 204          * queue up anybody new on the log reservations, and wakes up
 205          * everybody who's sleeping on log reservations to tell them
 206          * the bad news.
 207          */
 208         if (xfs_log_force_umount(mp, logerror))
 209                 return;
 210
 211         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 212                 xfs_cmn_err(XFS_PTAG_SHUTDOWN_CORRUPT, CE_ALERT, mp,
 213     "Corruption of in-memory data detected.  Shutting down filesystem: %s",
 214                         mp->m_fsname);
 215                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level) {
 216                         xfs_stack_trace();
 217                 }
 218         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 219                 if (logerror) {
 220                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_LOGERROR, CE_ALERT, mp,
 221                 "Log I/O Error Detected.  Shutting down filesystem: %s",
 222                                 mp->m_fsname);
 223                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 224                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
 225                 "All device paths lost.  Shutting down filesystem: %s",
 226                                 mp->m_fsname);
 227                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 228                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
 229                 "I/O Error Detected.  Shutting down filesystem: %s",
 230                                 mp->m_fsname);
 231                 }
 232         }
 233         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 234                 cmn_err(CE_ALERT, "Please umount the filesystem, "
 235                                   "and rectify the problem(s)");
 236         }
 237 }
 238
 239
 240 /*
 241  * Called when we want to stop a buffer from getting written or read.
 242  * We attach the EIO error, muck with its flags, and call biodone
 243  * so that the proper iodone callbacks get called.
 244  */
 245 int
 246 xfs_bioerror(
 247         xfs_buf_t *bp)
 248 {
 249
 250 #ifdef XFSERRORDEBUG
 251         ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
 252 #endif
 253
 254         /*
 255          * No need to wait until the buffer is unpinned.
 256          * We aren't flushing it.
 257          */
 258         xfs_buftrace("XFS IOERROR", bp);
 259         XFS_BUF_ERROR(bp, EIO);
 260         /*
 261          * We're calling biodone, so delete B_DONE flag. Either way
 262          * we have to call the iodone callback, and calling biodone
 263          * probably is the best way since it takes care of
 264          * GRIO as well.
 265          */
 266         XFS_BUF_UNREAD(bp);
 267         XFS_BUF_UNDELAYWRITE(bp);
 268         XFS_BUF_UNDONE(bp);
 269         XFS_BUF_STALE(bp);
 270
 271         XFS_BUF_CLR_BDSTRAT_FUNC(bp);
 272         xfs_biodone(bp);
 273
 274         return (EIO);
 275 }
 276
 277 /*
 278  * Same as xfs_bioerror, except that we are releasing the buffer
 279  * here ourselves, and avoiding the biodone call.
 280  * This is meant for userdata errors; metadata bufs come with
 281  * iodone functions attached, so that we can track down errors.
 282  */
 283 int
 284 xfs_bioerror_relse(
 285         xfs_buf_t *bp)
 286 {
 287         int64_t fl;
 288
 289         ASSERT(XFS_BUF_IODONE_FUNC(bp) != xfs_buf_iodone_callbacks);
 290         ASSERT(XFS_BUF_IODONE_FUNC(bp) != xlog_iodone);
 291
 292         xfs_buftrace("XFS IOERRELSE", bp);
 293         fl = XFS_BUF_BFLAGS(bp);
 294         /*
 295          * No need to wait until the buffer is unpinned.
 296          * We aren't flushing it.
 297          *
 298          * chunkhold expects B_DONE to be set, whether
 299          * we actually finish the I/O or not. We don't want to
 300          * change that interface.
 301          */
 302         XFS_BUF_UNREAD(bp);
 303         XFS_BUF_UNDELAYWRITE(bp);
 304         XFS_BUF_DONE(bp);
 305         XFS_BUF_STALE(bp);
 306         XFS_BUF_CLR_IODONE_FUNC(bp);
 307         XFS_BUF_CLR_BDSTRAT_FUNC(bp);
 308         if (!(fl & XFS_B_ASYNC)) {
 309                 /*
 310                  * Mark b_error and B_ERROR _both_.
 311                  * Lot's of chunkcache code assumes that.
 312                  * There's no reason to mark error for
 313                  * ASYNC buffers.
 314                  */
 315                 XFS_BUF_ERROR(bp, EIO);
 316                 XFS_BUF_FINISH_IOWAIT(bp);
 317         } else {
 318                 xfs_buf_relse(bp);
 319         }
 320         return (EIO);
 321 }
 322
 323 /*
 324  * Prints out an ALERT message about I/O error.
 325  */
 326 void
 327 xfs_ioerror_alert(
 328         char                    *func,
 329         struct xfs_mount        *mp,
 330         xfs_buf_t               *bp,
 331         xfs_daddr_t             blkno)
 332 {
 333         cmn_err(CE_ALERT,
 334  "I/O error in filesystem (\"%s\") meta-data dev %s block 0x%llx"
 335  "       (\"%s\") error %d buf count %zd",
 336                 (!mp || !mp->m_fsname) ? "(fs name not set)" : mp->m_fsname,
 337                 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
 338                 (__uint64_t)blkno, func,
 339                 XFS_BUF_GETERROR(bp), XFS_BUF_COUNT(bp));
 340 }
 341
 342 /*
 343  * This isn't an absolute requirement, but it is
 344  * just a good idea to call xfs_read_buf instead of
 345  * directly doing a read_buf call. For one, we shouldn't
 346  * be doing this disk read if we are in SHUTDOWN state anyway,
 347  * so this stops that from happening. Secondly, this does all
 348  * the error checking stuff and the brelse if appropriate for
 349  * the caller, so the code can be a little leaner.
 350  */
 351
 352 int
 353 xfs_read_buf(
 354         struct xfs_mount *mp,
 355         xfs_buftarg_t    *target,
 356         xfs_daddr_t      blkno,
 357         int              len,
 358         uint             flags,
 359         xfs_buf_t        **bpp)
 360 {
 361         xfs_buf_t        *bp;
 362         int              error;
 363
 364         if (flags)
 365                 bp = xfs_buf_read_flags(target, blkno, len, flags);
 366         else
 367                 bp = xfs_buf_read(target, blkno, len, flags);
 368         if (!bp)
 369                 return XFS_ERROR(EIO);
 370         error = XFS_BUF_GETERROR(bp);
 371         if (bp && !error && !XFS_FORCED_SHUTDOWN(mp)) {
 372                 *bpp = bp;
 373         } else {
 374                 *bpp = NULL;
 375                 if (error) {
 376                         xfs_ioerror_alert("xfs_read_buf", mp, bp, XFS_BUF_ADDR(bp));
 377                 } else {
 378                         error = XFS_ERROR(EIO);
 379                 }
 380                 if (bp) {
 381                         XFS_BUF_UNDONE(bp);
 382                         XFS_BUF_UNDELAYWRITE(bp);
 383                         XFS_BUF_STALE(bp);
 384                         /*
 385                          * brelse clears B_ERROR and b_error
 386                          */
 387                         xfs_buf_relse(bp);
 388                 }
 389         }
 390         return (error);
 391 }
 392
 393 /*
 394  * Wrapper around bwrite() so that we can trap
 395  * write errors, and act accordingly.
 396  */
 397 int
 398 xfs_bwrite(
 399         struct xfs_mount *mp,
 400         struct xfs_buf   *bp)
 401 {
 402         int     error;
 403
 404         /*
 405          * XXXsup how does this work for quotas.
 406          */
 407         XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
 408         bp->b_mount = mp;
 409         XFS_BUF_WRITE(bp);
 410
 411         if ((error = XFS_bwrite(bp))) {
 412                 ASSERT(mp);
 413                 /*
 414                  * Cannot put a buftrace here since if the buffer is not
 415                  * B_HOLD then we will brelse() the buffer before returning
 416                  * from bwrite and we could be tracing a buffer that has
 417                  * been reused.
 418                  */
 419                 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 420         }
 421         return (error);
 422 }