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_acl.h"
  45 #include "xfs_error.h"
  46 #include "xfs_buf_item.h"
  47 #include "xfs_rw.h"
  48
  49 /*
  50  * This is a subroutine for xfs_write() and other writers (xfs_ioctl)
  51  * which clears the setuid and setgid bits when a file is written.
  52  */
  53 int
  54 xfs_write_clear_setuid(
  55         xfs_inode_t     *ip)
  56 {
  57         xfs_mount_t     *mp;
  58         xfs_trans_t     *tp;
  59         int             error;
  60
  61         mp = ip->i_mount;
  62         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
  63         if ((error = xfs_trans_reserve(tp, 0,
  64                                       XFS_WRITEID_LOG_RES(mp),
  65                                       0, 0, 0))) {
  66                 xfs_trans_cancel(tp, 0);
  67                 return error;
  68         }
  69         xfs_ilock(ip, XFS_ILOCK_EXCL);
  70         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
  71         xfs_trans_ihold(tp, ip);
  72         ip->i_d.di_mode &= ~S_ISUID;
  73
  74         /*
  75          * Note that we don't have to worry about mandatory
  76          * file locking being disabled here because we only
  77          * clear the S_ISGID bit if the Group execute bit is
  78          * on, but if it was on then mandatory locking wouldn't
  79          * have been enabled.
  80          */
  81         if (ip->i_d.di_mode & S_IXGRP) {
  82                 ip->i_d.di_mode &= ~S_ISGID;
  83         }
  84         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  85         xfs_trans_set_sync(tp);
  86         error = xfs_trans_commit(tp, 0);
  87         xfs_iunlock(ip, XFS_ILOCK_EXCL);
  88         return 0;
  89 }
  90
  91 /*
  92  * Handle logging requirements of various synchronous types of write.
  93  */
  94 int
  95 xfs_write_sync_logforce(
  96         xfs_mount_t     *mp,
  97         xfs_inode_t     *ip)
  98 {
  99         int             error = 0;
 100
 101         /*
 102          * If we're treating this as O_DSYNC and we have not updated the
 103          * size, force the log.
 104          */
 105         if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC) &&
 106             !(ip->i_update_size)) {
 107                 xfs_inode_log_item_t    *iip = ip->i_itemp;
 108
 109                 /*
 110                  * If an allocation transaction occurred
 111                  * without extending the size, then we have to force
 112                  * the log up the proper point to ensure that the
 113                  * allocation is permanent.  We can't count on
 114                  * the fact that buffered writes lock out direct I/O
 115                  * writes - the direct I/O write could have extended
 116                  * the size nontransactionally, then finished before
 117                  * we started.  xfs_write_file will think that the file
 118                  * didn't grow but the update isn't safe unless the
 119                  * size change is logged.
 120                  *
 121                  * Force the log if we've committed a transaction
 122                  * against the inode or if someone else has and
 123                  * the commit record hasn't gone to disk (e.g.
 124                  * the inode is pinned).  This guarantees that
 125                  * all changes affecting the inode are permanent
 126                  * when we return.
 127                  */
 128                 if (iip && iip->ili_last_lsn) {
 129                         xfs_log_force(mp, iip->ili_last_lsn,
 130                                         XFS_LOG_FORCE | XFS_LOG_SYNC);
 131                 } else if (xfs_ipincount(ip) > 0) {
 132                         xfs_log_force(mp, (xfs_lsn_t)0,
 133                                         XFS_LOG_FORCE | XFS_LOG_SYNC);
 134                 }
 135
 136         } else {
 137                 xfs_trans_t     *tp;
 138
 139                 /*
 140                  * O_SYNC or O_DSYNC _with_ a size update are handled
 141                  * the same way.
 142                  *
 143                  * If the write was synchronous then we need to make
 144                  * sure that the inode modification time is permanent.
 145                  * We'll have updated the timestamp above, so here
 146                  * we use a synchronous transaction to log the inode.
 147                  * It's not fast, but it's necessary.
 148                  *
 149                  * If this a dsync write and the size got changed
 150                  * non-transactionally, then we need to ensure that
 151                  * the size change gets logged in a synchronous
 152                  * transaction.
 153                  */
 154                 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
 155                 if ((error = xfs_trans_reserve(tp, 0,
 156                                                 XFS_SWRITE_LOG_RES(mp),
 157                                                 0, 0, 0))) {
 158                         /* Transaction reserve failed */
 159                         xfs_trans_cancel(tp, 0);
 160                 } else {
 161                         /* Transaction reserve successful */
 162                         xfs_ilock(ip, XFS_ILOCK_EXCL);
 163                         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 164                         xfs_trans_ihold(tp, ip);
 165                         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 166                         xfs_trans_set_sync(tp);
 167                         error = xfs_trans_commit(tp, 0);
 168                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 169                 }
 170         }
 171
 172         return error;
 173 }
 174
 175 /*
 176  * Force a shutdown of the filesystem instantly while keeping
 177  * the filesystem consistent. We don't do an unmount here; just shutdown
 178  * the shop, make sure that absolutely nothing persistent happens to
 179  * this filesystem after this point.
 180  */
 181
 182 void
 183 xfs_do_force_shutdown(
 184         bhv_desc_t      *bdp,
 185         int             flags,
 186         char            *fname,
 187         int             lnnum)
 188 {
 189         int             logerror;
 190         xfs_mount_t     *mp;
 191
 192         mp = XFS_BHVTOM(bdp);
 193         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 194
 195         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 196                 cmn_err(CE_NOTE, "xfs_force_shutdown(%s,0x%x) called from "
 197                                  "line %d of file %s.  Return address = 0x%p",
 198                         mp->m_fsname, flags, lnnum, fname, __return_address);
 199         }
 200         /*
 201          * No need to duplicate efforts.
 202          */
 203         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 204                 return;
 205
 206         /*
 207          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 208          * queue up anybody new on the log reservations, and wakes up
 209          * everybody who's sleeping on log reservations to tell them
 210          * the bad news.
 211          */
 212         if (xfs_log_force_umount(mp, logerror))
 213                 return;
 214
 215         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 216                 xfs_cmn_err(XFS_PTAG_SHUTDOWN_CORRUPT, CE_ALERT, mp,
 217     "Corruption of in-memory data detected.  Shutting down filesystem: %s",
 218                         mp->m_fsname);
 219                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level) {
 220                         xfs_stack_trace();
 221                 }
 222         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 223                 if (logerror) {
 224                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_LOGERROR, CE_ALERT, mp,
 225                 "Log I/O Error Detected.  Shutting down filesystem: %s",
 226                                 mp->m_fsname);
 227                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 228                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
 229                 "All device paths lost.  Shutting down filesystem: %s",
 230                                 mp->m_fsname);
 231                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 232                         xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
 233                 "I/O Error Detected.  Shutting down filesystem: %s",
 234                                 mp->m_fsname);
 235                 }
 236         }
 237         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 238                 cmn_err(CE_ALERT, "Please umount the filesystem, "
 239                                   "and rectify the problem(s)");
 240         }
 241 }
 242
 243
 244 /*
 245  * Called when we want to stop a buffer from getting written or read.
 246  * We attach the EIO error, muck with its flags, and call biodone
 247  * so that the proper iodone callbacks get called.
 248  */
 249 int
 250 xfs_bioerror(
 251         xfs_buf_t *bp)
 252 {
 253
 254 #ifdef XFSERRORDEBUG
 255         ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
 256 #endif
 257
 258         /*
 259          * No need to wait until the buffer is unpinned.
 260          * We aren't flushing it.
 261          */
 262         xfs_buftrace("XFS IOERROR", bp);
 263         XFS_BUF_ERROR(bp, EIO);
 264         /*
 265          * We're calling biodone, so delete B_DONE flag. Either way
 266          * we have to call the iodone callback, and calling biodone
 267          * probably is the best way since it takes care of
 268          * GRIO as well.
 269          */
 270         XFS_BUF_UNREAD(bp);
 271         XFS_BUF_UNDELAYWRITE(bp);
 272         XFS_BUF_UNDONE(bp);
 273         XFS_BUF_STALE(bp);
 274
 275         XFS_BUF_CLR_BDSTRAT_FUNC(bp);
 276         xfs_biodone(bp);
 277
 278         return (EIO);
 279 }
 280
 281 /*
 282  * Same as xfs_bioerror, except that we are releasing the buffer
 283  * here ourselves, and avoiding the biodone call.
 284  * This is meant for userdata errors; metadata bufs come with
 285  * iodone functions attached, so that we can track down errors.
 286  */
 287 int
 288 xfs_bioerror_relse(
 289         xfs_buf_t *bp)
 290 {
 291         int64_t fl;
 292
 293         ASSERT(XFS_BUF_IODONE_FUNC(bp) != xfs_buf_iodone_callbacks);
 294         ASSERT(XFS_BUF_IODONE_FUNC(bp) != xlog_iodone);
 295
 296         xfs_buftrace("XFS IOERRELSE", bp);
 297         fl = XFS_BUF_BFLAGS(bp);
 298         /*
 299          * No need to wait until the buffer is unpinned.
 300          * We aren't flushing it.
 301          *
 302          * chunkhold expects B_DONE to be set, whether
 303          * we actually finish the I/O or not. We don't want to
 304          * change that interface.
 305          */
 306         XFS_BUF_UNREAD(bp);
 307         XFS_BUF_UNDELAYWRITE(bp);
 308         XFS_BUF_DONE(bp);
 309         XFS_BUF_STALE(bp);
 310         XFS_BUF_CLR_IODONE_FUNC(bp);
 311         XFS_BUF_CLR_BDSTRAT_FUNC(bp);
 312         if (!(fl & XFS_B_ASYNC)) {
 313                 /*
 314                  * Mark b_error and B_ERROR _both_.
 315                  * Lot's of chunkcache code assumes that.
 316                  * There's no reason to mark error for
 317                  * ASYNC buffers.
 318                  */
 319                 XFS_BUF_ERROR(bp, EIO);
 320                 XFS_BUF_V_IODONESEMA(bp);
 321         } else {
 322                 xfs_buf_relse(bp);
 323         }
 324         return (EIO);
 325 }
 326
 327 /*
 328  * Prints out an ALERT message about I/O error.
 329  */
 330 void
 331 xfs_ioerror_alert(
 332         char                    *func,
 333         struct xfs_mount        *mp,
 334         xfs_buf_t               *bp,
 335         xfs_daddr_t             blkno)
 336 {
 337         cmn_err(CE_ALERT,
 338  "I/O error in filesystem (\"%s\") meta-data dev %s block 0x%llx"
 339  "       (\"%s\") error %d buf count %zd",
 340                 (!mp || !mp->m_fsname) ? "(fs name not set)" : mp->m_fsname,
 341                 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
 342                 (__uint64_t)blkno, func,
 343                 XFS_BUF_GETERROR(bp), XFS_BUF_COUNT(bp));
 344 }
 345
 346 /*
 347  * This isn't an absolute requirement, but it is
 348  * just a good idea to call xfs_read_buf instead of
 349  * directly doing a read_buf call. For one, we shouldn't
 350  * be doing this disk read if we are in SHUTDOWN state anyway,
 351  * so this stops that from happening. Secondly, this does all
 352  * the error checking stuff and the brelse if appropriate for
 353  * the caller, so the code can be a little leaner.
 354  */
 355
 356 int
 357 xfs_read_buf(
 358         struct xfs_mount *mp,
 359         xfs_buftarg_t    *target,
 360         xfs_daddr_t      blkno,
 361         int              len,
 362         uint             flags,
 363         xfs_buf_t        **bpp)
 364 {
 365         xfs_buf_t        *bp;
 366         int              error;
 367
 368         if (flags)
 369                 bp = xfs_buf_read_flags(target, blkno, len, flags);
 370         else
 371                 bp = xfs_buf_read(target, blkno, len, flags);
 372         if (!bp)
 373                 return XFS_ERROR(EIO);
 374         error = XFS_BUF_GETERROR(bp);
 375         if (bp && !error && !XFS_FORCED_SHUTDOWN(mp)) {
 376                 *bpp = bp;
 377         } else {
 378                 *bpp = NULL;
 379                 if (error) {
 380                         xfs_ioerror_alert("xfs_read_buf", mp, bp, XFS_BUF_ADDR(bp));
 381                 } else {
 382                         error = XFS_ERROR(EIO);
 383                 }
 384                 if (bp) {
 385                         XFS_BUF_UNDONE(bp);
 386                         XFS_BUF_UNDELAYWRITE(bp);
 387                         XFS_BUF_STALE(bp);
 388                         /*
 389                          * brelse clears B_ERROR and b_error
 390                          */
 391                         xfs_buf_relse(bp);
 392                 }
 393         }
 394         return (error);
 395 }
 396
 397 /*
 398  * Wrapper around bwrite() so that we can trap
 399  * write errors, and act accordingly.
 400  */
 401 int
 402 xfs_bwrite(
 403         struct xfs_mount *mp,
 404         struct xfs_buf   *bp)
 405 {
 406         int     error;
 407
 408         /*
 409          * XXXsup how does this work for quotas.
 410          */
 411         XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
 412         XFS_BUF_SET_FSPRIVATE3(bp, mp);
 413         XFS_BUF_WRITE(bp);
 414
 415         if ((error = XFS_bwrite(bp))) {
 416                 ASSERT(mp);
 417                 /*
 418                  * Cannot put a buftrace here since if the buffer is not
 419                  * B_HOLD then we will brelse() the buffer before returning
 420                  * from bwrite and we could be tracing a buffer that has
 421                  * been reused.
 422                  */
 423                 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 424         }
 425         return (error);
 426 }