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_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_dir_sf.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_bmap.h"
  39 #include "xfs_error.h"
  40 #include "xfs_quota.h"
  41 #include "xfs_rw.h"
  42 #include "xfs_itable.h"
  43 #include "xfs_utils.h"
  44
  45 /*
  46  * xfs_get_dir_entry is used to get a reference to an inode given
  47  * its parent directory inode and the name of the file.  It does
  48  * not lock the child inode, and it unlocks the directory before
  49  * returning.  The directory's generation number is returned for
  50  * use by a later call to xfs_lock_dir_and_entry.
  51  */
  52 int
  53 xfs_get_dir_entry(
  54         vname_t         *dentry,
  55         xfs_inode_t     **ipp)
  56 {
  57         vnode_t         *vp;
  58         bhv_desc_t      *bdp;
  59
  60         vp = VNAME_TO_VNODE(dentry);
  61         bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(vp), &xfs_vnodeops);
  62         if (!bdp) {
  63                 *ipp = NULL;
  64                 return XFS_ERROR(ENOENT);
  65         }
  66         VN_HOLD(vp);
  67         *ipp = XFS_BHVTOI(bdp);
  68         return 0;
  69 }
  70
  71 int
  72 xfs_dir_lookup_int(
  73         bhv_desc_t      *dir_bdp,
  74         uint            lock_mode,
  75         vname_t         *dentry,
  76         xfs_ino_t       *inum,
  77         xfs_inode_t     **ipp)
  78 {
  79         vnode_t         *dir_vp;
  80         xfs_inode_t     *dp;
  81         int             error;
  82
  83         dir_vp = BHV_TO_VNODE(dir_bdp);
  84         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
  85
  86         dp = XFS_BHVTOI(dir_bdp);
  87
  88         error = XFS_DIR_LOOKUP(dp->i_mount, NULL, dp,
  89                                 VNAME(dentry), VNAMELEN(dentry), inum);
  90         if (!error) {
  91                 /*
  92                  * Unlock the directory. We do this because we can't
  93                  * hold the directory lock while doing the vn_get()
  94                  * in xfs_iget().  Doing so could cause us to hold
  95                  * a lock while waiting for the inode to finish
  96                  * being inactive while it's waiting for a log
  97                  * reservation in the inactive routine.
  98                  */
  99                 xfs_iunlock(dp, lock_mode);
 100                 error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0);
 101                 xfs_ilock(dp, lock_mode);
 102
 103                 if (error) {
 104                         *ipp = NULL;
 105                 } else if ((*ipp)->i_d.di_mode == 0) {
 106                         /*
 107                          * The inode has been freed.  Something is
 108                          * wrong so just get out of here.
 109                          */
 110                         xfs_iunlock(dp, lock_mode);
 111                         xfs_iput_new(*ipp, 0);
 112                         *ipp = NULL;
 113                         xfs_ilock(dp, lock_mode);
 114                         error = XFS_ERROR(ENOENT);
 115                 }
 116         }
 117         return error;
 118 }
 119
 120 /*
 121  * Allocates a new inode from disk and return a pointer to the
 122  * incore copy. This routine will internally commit the current
 123  * transaction and allocate a new one if the Space Manager needed
 124  * to do an allocation to replenish the inode free-list.
 125  *
 126  * This routine is designed to be called from xfs_create and
 127  * xfs_create_dir.
 128  *
 129  */
 130 int
 131 xfs_dir_ialloc(
 132         xfs_trans_t     **tpp,          /* input: current transaction;
 133                                            output: may be a new transaction. */
 134         xfs_inode_t     *dp,            /* directory within whose allocate
 135                                            the inode. */
 136         mode_t          mode,
 137         xfs_nlink_t     nlink,
 138         xfs_dev_t       rdev,
 139         cred_t          *credp,
 140         prid_t          prid,           /* project id */
 141         int             okalloc,        /* ok to allocate new space */
 142         xfs_inode_t     **ipp,          /* pointer to inode; it will be
 143                                            locked. */
 144         int             *committed)
 145
 146 {
 147         xfs_trans_t     *tp;
 148         xfs_trans_t     *ntp;
 149         xfs_inode_t     *ip;
 150         xfs_buf_t       *ialloc_context = NULL;
 151         boolean_t       call_again = B_FALSE;
 152         int             code;
 153         uint            log_res;
 154         uint            log_count;
 155         void            *dqinfo;
 156         uint            tflags;
 157
 158         tp = *tpp;
 159         ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 160
 161         /*
 162          * xfs_ialloc will return a pointer to an incore inode if
 163          * the Space Manager has an available inode on the free
 164          * list. Otherwise, it will do an allocation and replenish
 165          * the freelist.  Since we can only do one allocation per
 166          * transaction without deadlocks, we will need to commit the
 167          * current transaction and start a new one.  We will then
 168          * need to call xfs_ialloc again to get the inode.
 169          *
 170          * If xfs_ialloc did an allocation to replenish the freelist,
 171          * it returns the bp containing the head of the freelist as
 172          * ialloc_context. We will hold a lock on it across the
 173          * transaction commit so that no other process can steal
 174          * the inode(s) that we've just allocated.
 175          */
 176         code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, okalloc,
 177                           &ialloc_context, &call_again, &ip);
 178
 179         /*
 180          * Return an error if we were unable to allocate a new inode.
 181          * This should only happen if we run out of space on disk or
 182          * encounter a disk error.
 183          */
 184         if (code) {
 185                 *ipp = NULL;
 186                 return code;
 187         }
 188         if (!call_again && (ip == NULL)) {
 189                 *ipp = NULL;
 190                 return XFS_ERROR(ENOSPC);
 191         }
 192
 193         /*
 194          * If call_again is set, then we were unable to get an
 195          * inode in one operation.  We need to commit the current
 196          * transaction and call xfs_ialloc() again.  It is guaranteed
 197          * to succeed the second time.
 198          */
 199         if (call_again) {
 200
 201                 /*
 202                  * Normally, xfs_trans_commit releases all the locks.
 203                  * We call bhold to hang on to the ialloc_context across
 204                  * the commit.  Holding this buffer prevents any other
 205                  * processes from doing any allocations in this
 206                  * allocation group.
 207                  */
 208                 xfs_trans_bhold(tp, ialloc_context);
 209                 /*
 210                  * Save the log reservation so we can use
 211                  * them in the next transaction.
 212                  */
 213                 log_res = xfs_trans_get_log_res(tp);
 214                 log_count = xfs_trans_get_log_count(tp);
 215
 216                 /*
 217                  * We want the quota changes to be associated with the next
 218                  * transaction, NOT this one. So, detach the dqinfo from this
 219                  * and attach it to the next transaction.
 220                  */
 221                 dqinfo = NULL;
 222                 tflags = 0;
 223                 if (tp->t_dqinfo) {
 224                         dqinfo = (void *)tp->t_dqinfo;
 225                         tp->t_dqinfo = NULL;
 226                         tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
 227                         tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
 228                 }
 229
 230                 ntp = xfs_trans_dup(tp);
 231                 code = xfs_trans_commit(tp, 0, NULL);
 232                 tp = ntp;
 233                 if (committed != NULL) {
 234                         *committed = 1;
 235                 }
 236                 /*
 237                  * If we get an error during the commit processing,
 238                  * release the buffer that is still held and return
 239                  * to the caller.
 240                  */
 241                 if (code) {
 242                         xfs_buf_relse(ialloc_context);
 243                         if (dqinfo) {
 244                                 tp->t_dqinfo = dqinfo;
 245                                 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
 246                         }
 247                         *tpp = ntp;
 248                         *ipp = NULL;
 249                         return code;
 250                 }
 251                 code = xfs_trans_reserve(tp, 0, log_res, 0,
 252                                          XFS_TRANS_PERM_LOG_RES, log_count);
 253                 /*
 254                  * Re-attach the quota info that we detached from prev trx.
 255                  */
 256                 if (dqinfo) {
 257                         tp->t_dqinfo = dqinfo;
 258                         tp->t_flags |= tflags;
 259                 }
 260
 261                 if (code) {
 262                         xfs_buf_relse(ialloc_context);
 263                         *tpp = ntp;
 264                         *ipp = NULL;
 265                         return code;
 266                 }
 267                 xfs_trans_bjoin(tp, ialloc_context);
 268
 269                 /*
 270                  * Call ialloc again. Since we've locked out all
 271                  * other allocations in this allocation group,
 272                  * this call should always succeed.
 273                  */
 274                 code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid,
 275                                   okalloc, &ialloc_context, &call_again, &ip);
 276
 277                 /*
 278                  * If we get an error at this point, return to the caller
 279                  * so that the current transaction can be aborted.
 280                  */
 281                 if (code) {
 282                         *tpp = tp;
 283                         *ipp = NULL;
 284                         return code;
 285                 }
 286                 ASSERT ((!call_again) && (ip != NULL));
 287
 288         } else {
 289                 if (committed != NULL) {
 290                         *committed = 0;
 291                 }
 292         }
 293
 294         *ipp = ip;
 295         *tpp = tp;
 296
 297         return 0;
 298 }
 299
 300 /*
 301  * Decrement the link count on an inode & log the change.
 302  * If this causes the link count to go to zero, initiate the
 303  * logging activity required to truncate a file.
 304  */
 305 int                             /* error */
 306 xfs_droplink(
 307         xfs_trans_t *tp,
 308         xfs_inode_t *ip)
 309 {
 310         int     error;
 311
 312         xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
 313
 314         ASSERT (ip->i_d.di_nlink > 0);
 315         ip->i_d.di_nlink--;
 316         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 317
 318         error = 0;
 319         if (ip->i_d.di_nlink == 0) {
 320                 /*
 321                  * We're dropping the last link to this file.
 322                  * Move the on-disk inode to the AGI unlinked list.
 323                  * From xfs_inactive() we will pull the inode from
 324                  * the list and free it.
 325                  */
 326                 error = xfs_iunlink(tp, ip);
 327         }
 328         return error;
 329 }
 330
 331 /*
 332  * This gets called when the inode's version needs to be changed from 1 to 2.
 333  * Currently this happens when the nlink field overflows the old 16-bit value
 334  * or when chproj is called to change the project for the first time.
 335  * As a side effect the superblock version will also get rev'd
 336  * to contain the NLINK bit.
 337  */
 338 void
 339 xfs_bump_ino_vers2(
 340         xfs_trans_t     *tp,
 341         xfs_inode_t     *ip)
 342 {
 343         xfs_mount_t     *mp;
 344         unsigned long           s;
 345
 346         ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
 347         ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1);
 348
 349         ip->i_d.di_version = XFS_DINODE_VERSION_2;
 350         ip->i_d.di_onlink = 0;
 351         memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 352         mp = tp->t_mountp;
 353         if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
 354                 s = XFS_SB_LOCK(mp);
 355                 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
 356                         XFS_SB_VERSION_ADDNLINK(&mp->m_sb);
 357                         XFS_SB_UNLOCK(mp, s);
 358                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
 359                 } else {
 360                         XFS_SB_UNLOCK(mp, s);
 361                 }
 362         }
 363         /* Caller must log the inode */
 364 }
 365
 366 /*
 367  * Increment the link count on an inode & log the change.
 368  */
 369 int
 370 xfs_bumplink(
 371         xfs_trans_t *tp,
 372         xfs_inode_t *ip)
 373 {
 374         if (ip->i_d.di_nlink >= XFS_MAXLINK)
 375                 return XFS_ERROR(EMLINK);
 376         xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
 377
 378         ASSERT(ip->i_d.di_nlink > 0);
 379         ip->i_d.di_nlink++;
 380         if ((ip->i_d.di_version == XFS_DINODE_VERSION_1) &&
 381             (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
 382                 /*
 383                  * The inode has increased its number of links beyond
 384                  * what can fit in an old format inode.  It now needs
 385                  * to be converted to a version 2 inode with a 32 bit
 386                  * link count.  If this is the first inode in the file
 387                  * system to do this, then we need to bump the superblock
 388                  * version number as well.
 389                  */
 390                 xfs_bump_ino_vers2(tp, ip);
 391         }
 392
 393         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 394         return 0;
 395 }
 396
 397 /*
 398  * Try to truncate the given file to 0 length.  Currently called
 399  * only out of xfs_remove when it has to truncate a file to free
 400  * up space for the remove to proceed.
 401  */
 402 int
 403 xfs_truncate_file(
 404         xfs_mount_t     *mp,
 405         xfs_inode_t     *ip)
 406 {
 407         xfs_trans_t     *tp;
 408         int             error;
 409
 410 #ifdef QUOTADEBUG
 411         /*
 412          * This is called to truncate the quotainodes too.
 413          */
 414         if (XFS_IS_UQUOTA_ON(mp)) {
 415                 if (ip->i_ino != mp->m_sb.sb_uquotino)
 416                         ASSERT(ip->i_udquot);
 417         }
 418         if (XFS_IS_OQUOTA_ON(mp)) {
 419                 if (ip->i_ino != mp->m_sb.sb_gquotino)
 420                         ASSERT(ip->i_gdquot);
 421         }
 422 #endif
 423         /*
 424          * Make the call to xfs_itruncate_start before starting the
 425          * transaction, because we cannot make the call while we're
 426          * in a transaction.
 427          */
 428         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 429         xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
 430
 431         tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE);
 432         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
 433                                       XFS_TRANS_PERM_LOG_RES,
 434                                       XFS_ITRUNCATE_LOG_COUNT))) {
 435                 xfs_trans_cancel(tp, 0);
 436                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 437                 return error;
 438         }
 439
 440         /*
 441          * Follow the normal truncate locking protocol.  Since we
 442          * hold the inode in the transaction, we know that it's number
 443          * of references will stay constant.
 444          */
 445         xfs_ilock(ip, XFS_ILOCK_EXCL);
 446         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 447         xfs_trans_ihold(tp, ip);
 448         /*
 449          * Signal a sync xaction.  The only case where that isn't
 450          * the case is if we're truncating an already unlinked file
 451          * on a wsync fs.  In that case, we know the blocks can't
 452          * reappear in the file because the links to file are
 453          * permanently toast.  Currently, we're always going to
 454          * want a sync transaction because this code is being
 455          * called from places where nlink is guaranteed to be 1
 456          * but I'm leaving the tests in to protect against future
 457          * changes -- rcc.
 458          */
 459         error = xfs_itruncate_finish(&tp, ip, (xfs_fsize_t)0,
 460                                      XFS_DATA_FORK,
 461                                      ((ip->i_d.di_nlink != 0 ||
 462                                        !(mp->m_flags & XFS_MOUNT_WSYNC))
 463                                       ? 1 : 0));
 464         if (error) {
 465                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
 466                                  XFS_TRANS_ABORT);
 467         } else {
 468                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 469                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
 470                                          NULL);
 471         }
 472         xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 473
 474         return error;
 475 }