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