sys/vfs/gnu/ext2fs/ext2_inode.c

   1 /*
   2  *  modified for Lites 1.1
   3  *
   4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
   5  *  University of Utah, Department of Computer Science
   6  */
   7 /*
   8  * Copyright (c) 1982, 1986, 1989, 1993
   9  *      The Regents of the University of California.  All rights reserved.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. All advertising materials mentioning features or use of this software
  20  *    must display the following acknowledgement:
  21  *      This product includes software developed by the University of
  22  *      California, Berkeley and its contributors.
  23  * 4. Neither the name of the University nor the names of its contributors
  24  *    may be used to endorse or promote products derived from this software
  25  *    without specific prior written permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  37  * SUCH DAMAGE.
  38  *
  39  *      @(#)ext2_inode.c        8.5 (Berkeley) 12/30/93
  40  * $FreeBSD: src/sys/gnu/ext2fs/ext2_inode.c,v 1.24.2.1 2000/08/03 00:52:57 peter Exp $
  41  * $DragonFly: src/sys/vfs/gnu/ext2fs/ext2_inode.c,v 1.21 2007/05/06 19:23:33 dillon Exp $
  42  */
  43
  44 #include "opt_quota.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/mount.h>
  49 #include <sys/buf.h>
  50 #include <sys/vnode.h>
  51 #include <sys/malloc.h>
  52
  53 #include <vm/vm.h>
  54 #include <vm/vm_extern.h>
  55
  56 #include "quota.h"
  57 #include "inode.h"
  58 #include "ext2mount.h"
  59
  60 #include "ext2_fs.h"
  61 #include "ext2_fs_sb.h"
  62 #include "fs.h"
  63 #include "ext2_extern.h"
  64
  65 static int ext2_indirtrunc (struct inode *, daddr_t, off_t, daddr_t,
  66                             int, long *);
  67
  68 /*
  69  * Update the access, modified, and inode change times as specified by the
  70  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
  71  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
  72  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
  73  * later if not now.  If we write now, then clear both IN_MODIFIED and
  74  * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
  75  * set, then wait for the write to complete.
  76  */
  77 int
  78 ext2_update(struct vnode *vp, int waitfor)
  79 {
  80         struct ext2_sb_info *fs;
  81         struct buf *bp;
  82         struct inode *ip;
  83         int error;
  84
  85         ext2_itimes(vp);
  86         ip = VTOI(vp);
  87         if ((ip->i_flag & IN_MODIFIED) == 0)
  88                 return (0);
  89         ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
  90         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  91                 return (0);
  92         fs = ip->i_e2fs;
  93         error = bread(ip->i_devvp,
  94                       fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)),
  95                       (int)fs->s_blocksize, &bp);
  96         if (error) {
  97                 brelse(bp);
  98                 return (error);
  99         }
 100         ext2_di2ei( &ip->i_din, (struct ext2_inode *) ((char *)bp->b_data + EXT2_INODE_SIZE(fs) *
 101             ino_to_fsbo(fs, ip->i_number)));
 102 /*
 103         if (waitfor && (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
 104                 return (bwrite(bp));
 105         else {
 106 */
 107                 bdwrite(bp);
 108                 return (0);
 109 /*
 110         }
 111 */
 112 }
 113
 114 #define SINGLE  0       /* index of single indirect block */
 115 #define DOUBLE  1       /* index of double indirect block */
 116 #define TRIPLE  2       /* index of triple indirect block */
 117 /*
 118  * Truncate the inode oip to at most length size, freeing the
 119  * disk blocks.
 120  */
 121 int
 122 ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
 123 {
 124         struct vnode *ovp = vp;
 125         daddr_t lastblock;
 126         struct inode *oip;
 127         daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
 128         daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
 129         struct ext2_sb_info *fs;
 130         struct buf *bp;
 131         int offset, size, level;
 132         long count, nblocks, blocksreleased = 0;
 133         int i;
 134         int aflags, error, allerror;
 135         off_t osize;
 136 /*
 137 kprintf("ext2_truncate called %d to %d\n", VTOI(ovp)->i_number, length);
 138 */      /*
 139          * negative file sizes will totally break the code below and
 140          * are not meaningful anyways.
 141          */
 142         if (length < 0)
 143             return EFBIG;
 144
 145         oip = VTOI(ovp);
 146         if (ovp->v_type == VLNK &&
 147             oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
 148 #if DIAGNOSTIC
 149                 if (length != 0)
 150                         panic("ext2_truncate: partial truncate of symlink");
 151 #endif
 152                 bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
 153                 oip->i_size = 0;
 154                 oip->i_flag |= IN_CHANGE | IN_UPDATE;
 155                 return (EXT2_UPDATE(ovp, 1));
 156         }
 157         if (oip->i_size == length) {
 158                 oip->i_flag |= IN_CHANGE | IN_UPDATE;
 159                 return (EXT2_UPDATE(ovp, 0));
 160         }
 161 #if QUOTA
 162         if ((error = ext2_getinoquota(oip)) != 0)
 163                 return (error);
 164 #endif
 165         fs = oip->i_e2fs;
 166         osize = oip->i_size;
 167         ext2_discard_prealloc(oip);
 168         /*
 169          * Lengthen the size of the file. We must ensure that the
 170          * last byte of the file is allocated. Since the smallest
 171          * value of osize is 0, length will be at least 1.
 172          */
 173         if (osize < length) {
 174                 offset = blkoff(fs, length - 1);
 175                 lbn = lblkno(fs, length - 1);
 176                 aflags = B_CLRBUF;
 177                 if (flags & IO_SYNC)
 178                         aflags |= B_SYNC;
 179                 vnode_pager_setsize(ovp, length);
 180                 error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, aflags);
 181                 if (error) {
 182                         vnode_pager_setsize(ovp, osize);
 183                         return (error);
 184                 }
 185                 oip->i_size = length;
 186                 if (aflags & IO_SYNC)
 187                         bwrite(bp);
 188                 else
 189                         bawrite(bp);
 190                 oip->i_flag |= IN_CHANGE | IN_UPDATE;
 191                 return (EXT2_UPDATE(ovp, 1));
 192         }
 193         /*
 194          * Shorten the size of the file. If the file is not being
 195          * truncated to a block boundry, the contents of the
 196          * partial block following the end of the file must be
 197          * zero'ed in case it ever become accessable again because
 198          * of subsequent file growth.
 199          */
 200         /* I don't understand the comment above */
 201         offset = blkoff(fs, length);
 202         if (offset == 0) {
 203                 oip->i_size = length;
 204         } else {
 205                 lbn = lblkno(fs, length);
 206                 aflags = B_CLRBUF;
 207                 if (flags & IO_SYNC)
 208                         aflags |= B_SYNC;
 209                 error = ext2_balloc(oip, lbn, offset, cred, &bp, aflags);
 210                 if (error)
 211                         return (error);
 212                 oip->i_size = length;
 213                 size = blksize(fs, oip, lbn);
 214                 bzero((char *)bp->b_data + offset, (u_int)(size - offset));
 215                 allocbuf(bp, size);
 216                 if (aflags & IO_SYNC)
 217                         bwrite(bp);
 218                 else
 219                         bawrite(bp);
 220         }
 221         /*
 222          * Calculate index into inode's block list of
 223          * last direct and indirect blocks (if any)
 224          * which we want to keep.  Lastblock is -1 when
 225          * the file is truncated to 0.
 226          */
 227         lastblock = lblkno(fs, length + fs->s_blocksize - 1) - 1;
 228         lastiblock[SINGLE] = lastblock - NDADDR;
 229         lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
 230         lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
 231         nblocks = btodb(fs->s_blocksize);
 232         /*
 233          * Update file and block pointers on disk before we start freeing
 234          * blocks.  If we crash before free'ing blocks below, the blocks
 235          * will be returned to the free list.  lastiblock values are also
 236          * normalized to -1 for calls to ext2_indirtrunc below.
 237          */
 238         bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
 239         for (level = TRIPLE; level >= SINGLE; level--)
 240                 if (lastiblock[level] < 0) {
 241                         oip->i_ib[level] = 0;
 242                         lastiblock[level] = -1;
 243                 }
 244         for (i = NDADDR - 1; i > lastblock; i--)
 245                 oip->i_db[i] = 0;
 246         oip->i_flag |= IN_CHANGE | IN_UPDATE;
 247         allerror = EXT2_UPDATE(ovp, 1);
 248
 249         /*
 250          * Having written the new inode to disk, save its new configuration
 251          * and put back the old block pointers long enough to process them.
 252          * Note that we save the new block configuration so we can check it
 253          * when we are done.
 254          */
 255         bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
 256         bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
 257         oip->i_size = osize;
 258         error = vtruncbuf(ovp, length, (int)fs->s_blocksize);
 259         if (error && (allerror == 0))
 260                 allerror = error;
 261
 262         /*
 263          * Indirect blocks first.
 264          */
 265         indir_lbn[SINGLE] = -NDADDR;
 266         indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
 267         indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
 268         for (level = TRIPLE; level >= SINGLE; level--) {
 269                 bn = oip->i_ib[level];
 270                 if (bn != 0) {
 271                         error = ext2_indirtrunc(oip, indir_lbn[level],
 272                             fsbtodoff(fs, bn), lastiblock[level], level, &count);
 273                         if (error)
 274                                 allerror = error;
 275                         blocksreleased += count;
 276                         if (lastiblock[level] < 0) {
 277                                 oip->i_ib[level] = 0;
 278                                 ext2_blkfree(oip, bn, fs->s_frag_size);
 279                                 blocksreleased += nblocks;
 280                         }
 281                 }
 282                 if (lastiblock[level] >= 0)
 283                         goto done;
 284         }
 285
 286         /*
 287          * All whole direct blocks or frags.
 288          */
 289         for (i = NDADDR - 1; i > lastblock; i--) {
 290                 long bsize;
 291
 292                 bn = oip->i_db[i];
 293                 if (bn == 0)
 294                         continue;
 295                 oip->i_db[i] = 0;
 296                 bsize = blksize(fs, oip, i);
 297                 ext2_blkfree(oip, bn, bsize);
 298                 blocksreleased += btodb(bsize);
 299         }
 300         if (lastblock < 0)
 301                 goto done;
 302
 303         /*
 304          * Finally, look for a change in size of the
 305          * last direct block; release any frags.
 306          */
 307         bn = oip->i_db[lastblock];
 308         if (bn != 0) {
 309                 long oldspace, newspace;
 310
 311                 /*
 312                  * Calculate amount of space we're giving
 313                  * back as old block size minus new block size.
 314                  */
 315                 oldspace = blksize(fs, oip, lastblock);
 316                 oip->i_size = length;
 317                 newspace = blksize(fs, oip, lastblock);
 318                 if (newspace == 0)
 319                         panic("itrunc: newspace");
 320                 if (oldspace - newspace > 0) {
 321                         /*
 322                          * Block number of space to be free'd is
 323                          * the old block # plus the number of frags
 324                          * required for the storage we're keeping.
 325                          */
 326                         bn += numfrags(fs, newspace);
 327                         ext2_blkfree(oip, bn, oldspace - newspace);
 328                         blocksreleased += btodb(oldspace - newspace);
 329                 }
 330         }
 331 done:
 332 #if DIAGNOSTIC
 333         for (level = SINGLE; level <= TRIPLE; level++)
 334                 if (newblks[NDADDR + level] != oip->i_ib[level])
 335                         panic("itrunc1");
 336         for (i = 0; i < NDADDR; i++)
 337                 if (newblks[i] != oip->i_db[i])
 338                         panic("itrunc2");
 339         if (length == 0 && (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
 340                             !RB_EMPTY(&ovp->v_rbclean_tree)))
 341                 panic("itrunc3");
 342 #endif /* DIAGNOSTIC */
 343         /*
 344          * Put back the real size.
 345          */
 346         oip->i_size = length;
 347         oip->i_blocks -= blocksreleased;
 348         if (oip->i_blocks < 0)                  /* sanity */
 349                 oip->i_blocks = 0;
 350         oip->i_flag |= IN_CHANGE;
 351         vnode_pager_setsize(ovp, length);
 352 #if QUOTA
 353         ext2_chkdq(oip, -blocksreleased, NOCRED, 0);
 354 #endif
 355         return (allerror);
 356 }
 357
 358 /*
 359  * Release blocks associated with the inode ip and stored in the indirect
 360  * block bn.  Blocks are free'd in LIFO order up to (but not including)
 361  * lastbn.  If level is greater than SINGLE, the block is an indirect block
 362  * and recursive calls to indirtrunc must be used to cleanse other indirect
 363  * blocks.
 364  *
 365  * NB: triple indirect blocks are untested.
 366  */
 367
 368 static int
 369 ext2_indirtrunc(struct inode *ip, daddr_t lbn, off_t doffset, daddr_t lastbn,
 370                 int level, long *countp)
 371 {
 372         int i;
 373         struct buf *bp;
 374         struct ext2_sb_info *fs = ip->i_e2fs;
 375         daddr_t *bap;
 376         struct vnode *vp;
 377         daddr_t *copy, nb, nlbn, last;
 378         long blkcount, factor;
 379         int nblocks, blocksreleased = 0;
 380         int error = 0, allerror = 0;
 381
 382         /*
 383          * Calculate index in current block of last
 384          * block to be kept.  -1 indicates the entire
 385          * block so we need not calculate the index.
 386          */
 387         factor = 1;
 388         for (i = SINGLE; i < level; i++)
 389                 factor *= NINDIR(fs);
 390         last = lastbn;
 391         if (lastbn > 0)
 392                 last /= factor;
 393         nblocks = btodb(fs->s_blocksize);
 394         /*
 395          * Get buffer of block pointers, zero those entries corresponding
 396          * to blocks to be free'd, and update on disk copy first.  Since
 397          * double(triple) indirect before single(double) indirect, calls
 398          * to bmap on these blocks will fail.  However, we already have
 399          * the on disk address, so we have to set the bio_offset field
 400          * explicitly instead of letting bread do everything for us.
 401          */
 402         vp = ITOV(ip);
 403         bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->s_blocksize, 0, 0);
 404         if ((bp->b_flags & B_CACHE) == 0) {
 405                 bp->b_flags &= ~(B_ERROR | B_INVAL);
 406                 bp->b_cmd = BUF_CMD_READ;
 407                 if (bp->b_bcount > bp->b_bufsize)
 408                         panic("ext2_indirtrunc: bad buffer size");
 409                 bp->b_bio2.bio_offset = doffset;
 410                 bp->b_bio1.bio_done = biodone_sync;
 411                 bp->b_bio1.bio_flags |= BIO_SYNC;
 412                 vfs_busy_pages(bp->b_vp, bp);
 413                 vn_strategy(vp, &bp->b_bio1);
 414                 error = biowait(&bp->b_bio1, "biord");
 415         }
 416         if (error) {
 417                 brelse(bp);
 418                 *countp = 0;
 419                 return (error);
 420         }
 421
 422         bap = (daddr_t *)bp->b_data;
 423         MALLOC(copy, daddr_t *, fs->s_blocksize, M_TEMP, M_WAITOK);
 424         bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->s_blocksize);
 425         bzero((caddr_t)&bap[last + 1],
 426           (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
 427         if (last == -1)
 428                 bp->b_flags |= B_INVAL;
 429         error = bwrite(bp);
 430         if (error)
 431                 allerror = error;
 432         bap = copy;
 433
 434         /*
 435          * Recursively free totally unused blocks.
 436          */
 437         for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
 438             i--, nlbn += factor) {
 439                 nb = bap[i];
 440                 if (nb == 0)
 441                         continue;
 442                 if (level > SINGLE) {
 443                         if ((error = ext2_indirtrunc(ip, nlbn,
 444                             fsbtodoff(fs, nb), (daddr_t)-1, level - 1, &blkcount)) != 0)
 445                                 allerror = error;
 446                         blocksreleased += blkcount;
 447                 }
 448                 ext2_blkfree(ip, nb, fs->s_blocksize);
 449                 blocksreleased += nblocks;
 450         }
 451
 452         /*
 453          * Recursively free last partial block.
 454          */
 455         if (level > SINGLE && lastbn >= 0) {
 456                 last = lastbn % factor;
 457                 nb = bap[i];
 458                 if (nb != 0) {
 459                         error = ext2_indirtrunc(ip, nlbn, fsbtodoff(fs, nb),
 460                                                 last, level - 1, &blkcount);
 461                         if (error)
 462                                 allerror = error;
 463                         blocksreleased += blkcount;
 464                 }
 465         }
 466         FREE(copy, M_TEMP);
 467         *countp = blocksreleased;
 468         return (allerror);
 469 }
 470
 471 /*
 472  * Last reference to an inode.  If necessary, write or delete it.
 473  *
 474  * ext2_inactive(struct vnode *a_vp)
 475  */
 476 int
 477 ext2_inactive(struct vop_inactive_args *ap)
 478 {
 479         struct vnode *vp = ap->a_vp;
 480         struct inode *ip = VTOI(vp);
 481         int mode, error = 0;
 482
 483         ext2_discard_prealloc(ip);
 484         if (prtactive && vp->v_sysref.refcnt > 1)
 485                 vprint("ext2_inactive: pushing active", vp);
 486
 487         /*
 488          * Ignore inodes related to stale file handles.
 489          */
 490         if (ip == NULL || ip->i_mode == 0)
 491                 goto out;
 492         if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 493 #ifdef QUOTA
 494                 if (!ext2_getinoquota(ip))
 495                         (void)ext2_chkiq(ip, -1, NOCRED, FORCE);
 496 #endif
 497                 error = EXT2_TRUNCATE(vp, (off_t)0, 0, NOCRED);
 498                 ip->i_rdev = 0;
 499                 mode = ip->i_mode;
 500                 ip->i_mode = 0;
 501                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
 502                 EXT2_VFREE(vp, ip->i_number, mode);
 503         }
 504         if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE))
 505                 EXT2_UPDATE(vp, 0);
 506 out:
 507         /*
 508          * If we are done with the inode, reclaim it
 509          * so that it can be reused immediately.
 510          */
 511         if (ip == NULL || ip->i_mode == 0)
 512                 vrecycle(vp);
 513         return (error);
 514 }
 515
 516 /*
 517  * Reclaim an inode so that it can be used for other purposes.
 518  *
 519  * ext2_reclaim(struct vnode *a_vp)
 520  */
 521 int
 522 ext2_reclaim(struct vop_reclaim_args *ap)
 523 {
 524         struct inode *ip;
 525         struct vnode *vp = ap->a_vp;
 526 #ifdef QUOTA
 527         int i;
 528 #endif
 529
 530         if (prtactive && vp->v_sysref.refcnt > 1)
 531                 vprint("ext2_reclaim: pushing active", vp);
 532         ip = VTOI(vp);
 533
 534         /*
 535          * Lazy updates.
 536          */
 537         if (ip) {
 538                 if (ip->i_flag & IN_LAZYMOD) {
 539                         ip->i_flag |= IN_MODIFIED;
 540                         EXT2_UPDATE(vp, 0);
 541                 }
 542         }
 543 #ifdef INVARIANTS
 544         if (ip && (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE))) {
 545                 kprintf("WARNING: INODE %ld flags %08x: modified inode being released!\n", (long)ip->i_number, (int)ip->i_flag);
 546                 ip->i_flag |= IN_MODIFIED;
 547                 EXT2_UPDATE(vp, 0);
 548         }
 549 #endif
 550         /*
 551          * Remove the inode from its hash chain and purge namecache
 552          * data associated with the vnode.
 553          */
 554         vp->v_data = NULL;
 555         if (ip) {
 556                 ext2_ihashrem(ip);
 557                 if (ip->i_devvp) {
 558                         vrele(ip->i_devvp);
 559                         ip->i_devvp = 0;
 560                 }
 561 #ifdef QUOTA
 562                 for (i = 0; i < MAXQUOTAS; i++) {
 563                         if (ip->i_dquot[i] != NODQUOT) {
 564                                 ext2_dqrele(vp, ip->i_dquot[i]);
 565                                 ip->i_dquot[i] = NODQUOT;
 566                         }
 567                 }
 568 #endif
 569 #ifdef UFS_DIRHASH
 570                 if (ip->i_dirhash != NULL)
 571                         ext2dirhash_free(ip);
 572 #endif
 573                 kfree(ip, VFSTOEXT2(vp->v_mount)->um_malloctype);
 574         }
 575         return (0);
 576 }