fs/xfs/xfs_utils.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_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_dir2_sf.h"
  32 #include "xfs_attr_sf.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_inode_item.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_error.h"
  38 #include "xfs_quota.h"
  39 #include "xfs_rw.h"
  40 #include "xfs_itable.h"
  41 #include "xfs_utils.h"
  42
  43
  44 int
  45 xfs_dir_lookup_int(
  46         xfs_inode_t     *dp,
  47         uint            lock_mode,
  48         struct xfs_name *name,
  49         xfs_ino_t       *inum,
  50         xfs_inode_t     **ipp)
  51 {
  52         int             error;
  53
  54         xfs_itrace_entry(dp);
  55
  56         error = xfs_dir_lookup(NULL, dp, name, inum);
  57         if (!error) {
  58                 /*
  59                  * Unlock the directory. We do this because we can't
  60                  * hold the directory lock while doing the vn_get()
  61                  * in xfs_iget().  Doing so could cause us to hold
  62                  * a lock while waiting for the inode to finish
  63                  * being inactive while it's waiting for a log
  64                  * reservation in the inactive routine.
  65                  */
  66                 xfs_iunlock(dp, lock_mode);
  67                 error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0);
  68                 xfs_ilock(dp, lock_mode);
  69
  70                 if (error) {
  71                         *ipp = NULL;
  72                 } else if ((*ipp)->i_d.di_mode == 0) {
  73                         /*
  74                          * The inode has been freed.  Something is
  75                          * wrong so just get out of here.
  76                          */
  77                         xfs_iunlock(dp, lock_mode);
  78                         xfs_iput_new(*ipp, 0);
  79                         *ipp = NULL;
  80                         xfs_ilock(dp, lock_mode);
  81                         error = XFS_ERROR(ENOENT);
  82                 }
  83         }
  84         return error;
  85 }
  86
  87 /*
  88  * Allocates a new inode from disk and return a pointer to the
  89  * incore copy. This routine will internally commit the current
  90  * transaction and allocate a new one if the Space Manager needed
  91  * to do an allocation to replenish the inode free-list.
  92  *
  93  * This routine is designed to be called from xfs_create and
  94  * xfs_create_dir.
  95  *
  96  */
  97 int
  98 xfs_dir_ialloc(
  99         xfs_trans_t     **tpp,          /* input: current transaction;
 100                                            output: may be a new transaction. */
 101         xfs_inode_t     *dp,            /* directory within whose allocate
 102                                            the inode. */
 103         mode_t          mode,
 104         xfs_nlink_t     nlink,
 105         xfs_dev_t       rdev,
 106         cred_t          *credp,
 107         prid_t          prid,           /* project id */
 108         int             okalloc,        /* ok to allocate new space */
 109         xfs_inode_t     **ipp,          /* pointer to inode; it will be
 110                                            locked. */
 111         int             *committed)
 112
 113 {
 114         xfs_trans_t     *tp;
 115         xfs_trans_t     *ntp;
 116         xfs_inode_t     *ip;
 117         xfs_buf_t       *ialloc_context = NULL;
 118         boolean_t       call_again = B_FALSE;
 119         int             code;
 120         uint            log_res;
 121         uint            log_count;
 122         void            *dqinfo;
 123         uint            tflags;
 124
 125         tp = *tpp;
 126         ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 127
 128         /*
 129          * xfs_ialloc will return a pointer to an incore inode if
 130          * the Space Manager has an available inode on the free
 131          * list. Otherwise, it will do an allocation and replenish
 132          * the freelist.  Since we can only do one allocation per
 133          * transaction without deadlocks, we will need to commit the
 134          * current transaction and start a new one.  We will then
 135          * need to call xfs_ialloc again to get the inode.
 136          *
 137          * If xfs_ialloc did an allocation to replenish the freelist,
 138          * it returns the bp containing the head of the freelist as
 139          * ialloc_context. We will hold a lock on it across the
 140          * transaction commit so that no other process can steal
 141          * the inode(s) that we've just allocated.
 142          */
 143         code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, okalloc,
 144                           &ialloc_context, &call_again, &ip);
 145
 146         /*
 147          * Return an error if we were unable to allocate a new inode.
 148          * This should only happen if we run out of space on disk or
 149          * encounter a disk error.
 150          */
 151         if (code) {
 152                 *ipp = NULL;
 153                 return code;
 154         }
 155         if (!call_again && (ip == NULL)) {
 156                 *ipp = NULL;
 157                 return XFS_ERROR(ENOSPC);
 158         }
 159
 160         /*
 161          * If call_again is set, then we were unable to get an
 162          * inode in one operation.  We need to commit the current
 163          * transaction and call xfs_ialloc() again.  It is guaranteed
 164          * to succeed the second time.
 165          */
 166         if (call_again) {
 167
 168                 /*
 169                  * Normally, xfs_trans_commit releases all the locks.
 170                  * We call bhold to hang on to the ialloc_context across
 171                  * the commit.  Holding this buffer prevents any other
 172                  * processes from doing any allocations in this
 173                  * allocation group.
 174                  */
 175                 xfs_trans_bhold(tp, ialloc_context);
 176                 /*
 177                  * Save the log reservation so we can use
 178                  * them in the next transaction.
 179                  */
 180                 log_res = xfs_trans_get_log_res(tp);
 181                 log_count = xfs_trans_get_log_count(tp);
 182
 183                 /*
 184                  * We want the quota changes to be associated with the next
 185                  * transaction, NOT this one. So, detach the dqinfo from this
 186                  * and attach it to the next transaction.
 187                  */
 188                 dqinfo = NULL;
 189                 tflags = 0;
 190                 if (tp->t_dqinfo) {
 191                         dqinfo = (void *)tp->t_dqinfo;
 192                         tp->t_dqinfo = NULL;
 193                         tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
 194                         tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
 195                 }
 196
 197                 ntp = xfs_trans_dup(tp);
 198                 code = xfs_trans_commit(tp, 0);
 199                 tp = ntp;
 200                 if (committed != NULL) {
 201                         *committed = 1;
 202                 }
 203                 /*
 204                  * If we get an error during the commit processing,
 205                  * release the buffer that is still held and return
 206                  * to the caller.
 207                  */
 208                 if (code) {
 209                         xfs_buf_relse(ialloc_context);
 210                         if (dqinfo) {
 211                                 tp->t_dqinfo = dqinfo;
 212                                 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
 213                         }
 214                         *tpp = ntp;
 215                         *ipp = NULL;
 216                         return code;
 217                 }
 218                 code = xfs_trans_reserve(tp, 0, log_res, 0,
 219                                          XFS_TRANS_PERM_LOG_RES, log_count);
 220                 /*
 221                  * Re-attach the quota info that we detached from prev trx.
 222                  */
 223                 if (dqinfo) {
 224                         tp->t_dqinfo = dqinfo;
 225                         tp->t_flags |= tflags;
 226                 }
 227
 228                 if (code) {
 229                         xfs_buf_relse(ialloc_context);
 230                         *tpp = ntp;
 231                         *ipp = NULL;
 232                         return code;
 233                 }
 234                 xfs_trans_bjoin(tp, ialloc_context);
 235
 236                 /*
 237                  * Call ialloc again. Since we've locked out all
 238                  * other allocations in this allocation group,
 239                  * this call should always succeed.
 240                  */
 241                 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid,
 242                                   okalloc, &ialloc_context, &call_again, &ip);
 243
 244                 /*
 245                  * If we get an error at this point, return to the caller
 246                  * so that the current transaction can be aborted.
 247                  */
 248                 if (code) {
 249                         *tpp = tp;
 250                         *ipp = NULL;
 251                         return code;
 252                 }
 253                 ASSERT ((!call_again) && (ip != NULL));
 254
 255         } else {
 256                 if (committed != NULL) {
 257                         *committed = 0;
 258                 }
 259         }
 260
 261         *ipp = ip;
 262         *tpp = tp;
 263
 264         return 0;
 265 }
 266
 267 /*
 268  * Decrement the link count on an inode & log the change.
 269  * If this causes the link count to go to zero, initiate the
 270  * logging activity required to truncate a file.
 271  */
 272 int                             /* error */
 273 xfs_droplink(
 274         xfs_trans_t *tp,
 275         xfs_inode_t *ip)
 276 {
 277         int     error;
 278
 279         xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
 280
 281         ASSERT (ip->i_d.di_nlink > 0);
 282         ip->i_d.di_nlink--;
 283         drop_nlink(ip->i_vnode);
 284         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 285
 286         error = 0;
 287         if (ip->i_d.di_nlink == 0) {
 288                 /*
 289                  * We're dropping the last link to this file.
 290                  * Move the on-disk inode to the AGI unlinked list.
 291                  * From xfs_inactive() we will pull the inode from
 292                  * the list and free it.
 293                  */
 294                 error = xfs_iunlink(tp, ip);
 295         }
 296         return error;
 297 }
 298
 299 /*
 300  * This gets called when the inode's version needs to be changed from 1 to 2.
 301  * Currently this happens when the nlink field overflows the old 16-bit value
 302  * or when chproj is called to change the project for the first time.
 303  * As a side effect the superblock version will also get rev'd
 304  * to contain the NLINK bit.
 305  */
 306 void
 307 xfs_bump_ino_vers2(
 308         xfs_trans_t     *tp,
 309         xfs_inode_t     *ip)
 310 {
 311         xfs_mount_t     *mp;
 312
 313         ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
 314         ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1);
 315
 316         ip->i_d.di_version = XFS_DINODE_VERSION_2;
 317         ip->i_d.di_onlink = 0;
 318         memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 319         mp = tp->t_mountp;
 320         if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
 321                 spin_lock(&mp->m_sb_lock);
 322                 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
 323                         xfs_sb_version_addnlink(&mp->m_sb);
 324                         spin_unlock(&mp->m_sb_lock);
 325                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
 326                 } else {
 327                         spin_unlock(&mp->m_sb_lock);
 328                 }
 329         }
 330         /* Caller must log the inode */
 331 }
 332
 333 /*
 334  * Increment the link count on an inode & log the change.
 335  */
 336 int
 337 xfs_bumplink(
 338         xfs_trans_t *tp,
 339         xfs_inode_t *ip)
 340 {
 341         if (ip->i_d.di_nlink >= XFS_MAXLINK)
 342                 return XFS_ERROR(EMLINK);
 343         xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
 344
 345         ASSERT(ip->i_d.di_nlink > 0);
 346         ip->i_d.di_nlink++;
 347         inc_nlink(ip->i_vnode);
 348         if ((ip->i_d.di_version == XFS_DINODE_VERSION_1) &&
 349             (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
 350                 /*
 351                  * The inode has increased its number of links beyond
 352                  * what can fit in an old format inode.  It now needs
 353                  * to be converted to a version 2 inode with a 32 bit
 354                  * link count.  If this is the first inode in the file
 355                  * system to do this, then we need to bump the superblock
 356                  * version number as well.
 357                  */
 358                 xfs_bump_ino_vers2(tp, ip);
 359         }
 360
 361         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 362         return 0;
 363 }
 364
 365 /*
 366  * Try to truncate the given file to 0 length.  Currently called
 367  * only out of xfs_remove when it has to truncate a file to free
 368  * up space for the remove to proceed.
 369  */
 370 int
 371 xfs_truncate_file(
 372         xfs_mount_t     *mp,
 373         xfs_inode_t     *ip)
 374 {
 375         xfs_trans_t     *tp;
 376         int             error;
 377
 378 #ifdef QUOTADEBUG
 379         /*
 380          * This is called to truncate the quotainodes too.
 381          */
 382         if (XFS_IS_UQUOTA_ON(mp)) {
 383                 if (ip->i_ino != mp->m_sb.sb_uquotino)
 384                         ASSERT(ip->i_udquot);
 385         }
 386         if (XFS_IS_OQUOTA_ON(mp)) {
 387                 if (ip->i_ino != mp->m_sb.sb_gquotino)
 388                         ASSERT(ip->i_gdquot);
 389         }
 390 #endif
 391         /*
 392          * Make the call to xfs_itruncate_start before starting the
 393          * transaction, because we cannot make the call while we're
 394          * in a transaction.
 395          */
 396         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 397         error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
 398         if (error) {
 399                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 400                 return error;
 401         }
 402
 403         tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE);
 404         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
 405                                       XFS_TRANS_PERM_LOG_RES,
 406                                       XFS_ITRUNCATE_LOG_COUNT))) {
 407                 xfs_trans_cancel(tp, 0);
 408                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 409                 return error;
 410         }
 411
 412         /*
 413          * Follow the normal truncate locking protocol.  Since we
 414          * hold the inode in the transaction, we know that it's number
 415          * of references will stay constant.
 416          */
 417         xfs_ilock(ip, XFS_ILOCK_EXCL);
 418         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 419         xfs_trans_ihold(tp, ip);
 420         /*
 421          * Signal a sync xaction.  The only case where that isn't
 422          * the case is if we're truncating an already unlinked file
 423          * on a wsync fs.  In that case, we know the blocks can't
 424          * reappear in the file because the links to file are
 425          * permanently toast.  Currently, we're always going to
 426          * want a sync transaction because this code is being
 427          * called from places where nlink is guaranteed to be 1
 428          * but I'm leaving the tests in to protect against future
 429          * changes -- rcc.
 430          */
 431         error = xfs_itruncate_finish(&tp, ip, (xfs_fsize_t)0,
 432                                      XFS_DATA_FORK,
 433                                      ((ip->i_d.di_nlink != 0 ||
 434                                        !(mp->m_flags & XFS_MOUNT_WSYNC))
 435                                       ? 1 : 0));
 436         if (error) {
 437                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
 438                                  XFS_TRANS_ABORT);
 439         } else {
 440                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 441                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 442         }
 443         xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 444
 445         return error;
 446 }