/* * Rewrite an existing directory entry to point at the inode * supplied. The parameters describing the directory entry are * set up by a call to namei. */ int ufs_dirrewrite(struct inode *dp, struct inode *oip, ufsino_t newinum, int newtype, int isrmdir) { struct buf *bp; struct direct *ep; struct vnode *vdp = ITOV(dp); int error; error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, (char **)&ep, &bp); if (error) return (error); ep->d_ino = newinum; if (vdp->v_mount->mnt_maxsymlinklen > 0) ep->d_type = newtype; oip->i_effnlink--; if (DOINGSOFTDEP(vdp)) { softdep_change_linkcnt(oip, 0); softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir); bdwrite(bp); } else { DIP_ADD(oip, nlink, -1); oip->i_flag |= IN_CHANGE; UFS_WAPBL_UPDATE(oip, MNT_WAIT); if (DOINGASYNC(vdp)) { bdwrite(bp); error = 0; } else { error = VOP_BWRITE(bp); } } dp->i_flag |= IN_CHANGE | IN_UPDATE; UFS_WAPBL_UPDATE(dp, MNT_WAIT); return (error); }
/* * Update the access, modified, and inode change times as specified by the * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode * to disk if the IN_MODIFIED flag is set (it may be set initially, or by * the timestamp update). The IN_LAZYMOD flag is set to force a write * later if not now. If we write now, then clear both IN_MODIFIED and * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is * set, then wait for the write to complete. */ int ext2_update(struct vnode *vp, int waitfor) { struct m_ext2fs *fs; struct buf *bp; struct inode *ip; int error; ASSERT_VOP_ELOCKED(vp, "ext2_update"); ext2_itimes(vp); ip = VTOI(vp); if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) return (0); ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED); fs = ip->i_e2fs; if(fs->e2fs_ronly) return (0); if ((error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) { brelse(bp); return (error); } ext2_i2ei(ip, (struct ext2fs_dinode *)((char *)bp->b_data + EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number))); if (waitfor && !DOINGASYNC(vp)) return (bwrite(bp)); else { bdwrite(bp); return (0); } }
/* * Update the access, modified, and inode change times as specified by the * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode * to disk if the IN_MODIFIED flag is set (it may be set initially, or by * the timestamp update). The IN_LAZYMOD flag is set to force a write * later if not now. If we write now, then clear both IN_MODIFIED and * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is * set, then wait for the write to complete. */ int ffs_update(struct vnode *vp, int waitfor) { struct fs *fs; struct buf *bp; struct inode *ip; int error; ufs_itimes(vp); ip = VTOI(vp); if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0) return (0); ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED); fs = ip->i_fs; if (fs->fs_ronly) return (0); /* * The vnode type is usually set to VBAD if an unrecoverable I/O * error has occured (such as when reading the inode). Clear the * modified bits but do not write anything out in this case. */ if (vp->v_type == VBAD) return (0); /* * Ensure that uid and gid are correct. This is a temporary * fix until fsck has been changed to do the update. */ if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */ ip->i_din.di_ouid = ip->i_uid; /* XXX */ ip->i_din.di_ogid = ip->i_gid; /* XXX */ } /* XXX */ error = bread(ip->i_devvp, fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, &bp); if (error) { brelse(bp); return (error); } if (DOINGSOFTDEP(vp)) softdep_update_inodeblock(ip, bp, waitfor); else if (ip->i_effnlink != ip->i_nlink) panic("ffs_update: bad link cnt"); *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = ip->i_din; if (waitfor && !DOINGASYNC(vp)) { return (bwrite(bp)); } else if (vm_page_count_severe() || buf_dirty_count_severe()) { return (bwrite(bp)); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); return (0); } }
/* * Extend the file described by dep to length specified by length. */ int deextend(struct denode *dep, u_long length, struct ucred *cred) { struct msdosfsmount *pmp = dep->de_pmp; u_long count; int error; /* * The root of a DOS filesystem cannot be extended. */ if ((DETOV(dep)->v_vflag & VV_ROOT) && !FAT32(pmp)) return (EINVAL); /* * Directories cannot be extended. */ if (dep->de_Attributes & ATTR_DIRECTORY) return (EISDIR); if (length <= dep->de_FileSize) panic("deextend: file too large"); /* * Compute the number of clusters to allocate. */ count = de_clcount(pmp, length) - de_clcount(pmp, dep->de_FileSize); if (count > 0) { if (count > pmp->pm_freeclustercount) return (ENOSPC); error = extendfile(dep, count, NULL, NULL, DE_CLEAR); if (error) { /* truncate the added clusters away again */ (void) detrunc(dep, dep->de_FileSize, 0, cred); return (error); } } dep->de_FileSize = length; dep->de_flag |= DE_UPDATE | DE_MODIFIED; return (deupdat(dep, !DOINGASYNC(DETOV(dep)))); }
/* * link vnode call */ static int ext2_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; int error; #ifdef DIAGNOSTIC if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_link: no name"); #endif if (tdvp->v_mount != vp->v_mount) { error = EXDEV; goto out; } ip = VTOI(vp); if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_nlink++; ip->i_flag |= IN_CHANGE; error = ext2_update(vp, !DOINGASYNC(vp)); if (!error) error = ext2_direnter(ip, tdvp, cnp); if (error) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); }
/* * Truncate the file described by dep to the length specified by length. */ int detrunc(struct denode *dep, u_long length, int flags, struct ucred *cred) { int error; int allerror; u_long eofentry; u_long chaintofree; daddr_t bn; int boff; int isadir = dep->de_Attributes & ATTR_DIRECTORY; struct buf *bp; struct msdosfsmount *pmp = dep->de_pmp; #ifdef MSDOSFS_DEBUG printf("detrunc(): file %s, length %lu, flags %x\n", dep->de_Name, length, flags); #endif /* * Disallow attempts to truncate the root directory since it is of * fixed size. That's just the way dos filesystems are. We use * the VROOT bit in the vnode because checking for the directory * bit and a startcluster of 0 in the denode is not adequate to * recognize the root directory at this point in a file or * directory's life. */ if ((DETOV(dep)->v_vflag & VV_ROOT) && !FAT32(pmp)) { #ifdef MSDOSFS_DEBUG printf("detrunc(): can't truncate root directory, clust %ld, offset %ld\n", dep->de_dirclust, dep->de_diroffset); #endif return (EINVAL); } if (dep->de_FileSize < length) { vnode_pager_setsize(DETOV(dep), length); return deextend(dep, length, cred); } /* * If the desired length is 0 then remember the starting cluster of * the file and set the StartCluster field in the directory entry * to 0. If the desired length is not zero, then get the number of * the last cluster in the shortened file. Then get the number of * the first cluster in the part of the file that is to be freed. * Then set the next cluster pointer in the last cluster of the * file to CLUST_EOFE. */ if (length == 0) { chaintofree = dep->de_StartCluster; dep->de_StartCluster = 0; eofentry = ~0; } else { error = pcbmap(dep, de_clcount(pmp, length) - 1, 0, &eofentry, 0); if (error) { #ifdef MSDOSFS_DEBUG printf("detrunc(): pcbmap fails %d\n", error); #endif return (error); } } fc_purge(dep, de_clcount(pmp, length)); /* * If the new length is not a multiple of the cluster size then we * must zero the tail end of the new last cluster in case it * becomes part of the file again because of a seek. */ if ((boff = length & pmp->pm_crbomask) != 0) { if (isadir) { bn = cntobn(pmp, eofentry); error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster, NOCRED, &bp); if (error) { brelse(bp); #ifdef MSDOSFS_DEBUG printf("detrunc(): bread fails %d\n", error); #endif return (error); } bzero(bp->b_data + boff, pmp->pm_bpcluster - boff); if (flags & IO_SYNC) bwrite(bp); else bdwrite(bp); } } /* * Write out the updated directory entry. Even if the update fails * we free the trailing clusters. */ dep->de_FileSize = length; if (!isadir) dep->de_flag |= DE_UPDATE | DE_MODIFIED; allerror = vtruncbuf(DETOV(dep), cred, length, pmp->pm_bpcluster); #ifdef MSDOSFS_DEBUG if (allerror) printf("detrunc(): vtruncbuf error %d\n", allerror); #endif error = deupdat(dep, !DOINGASYNC((DETOV(dep)))); if (error != 0 && allerror == 0) allerror = error; #ifdef MSDOSFS_DEBUG printf("detrunc(): allerror %d, eofentry %lu\n", allerror, eofentry); #endif /* * If we need to break the cluster chain for the file then do it * now. */ if (eofentry != ~0) { error = fatentry(FAT_GET_AND_SET, pmp, eofentry, &chaintofree, CLUST_EOFE); if (error) { #ifdef MSDOSFS_DEBUG printf("detrunc(): fatentry errors %d\n", error); #endif return (error); } fc_setcache(dep, FC_LASTFC, de_cluster(pmp, length - 1), eofentry); } /* * Now free the clusters removed from the file because of the * truncation. */ if (chaintofree != 0 && !MSDOSFSEOF(pmp, chaintofree)) freeclusterchain(pmp, chaintofree); return (allerror); }
/* * Update the access, modified, and inode change times as specified by the * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. The IN_MODIFIED * flag is used to specify that the inode needs to be updated but that the * times have already been set. The access and modified times are taken from * the second and third parameters; the inode change time is always taken * from the current time. If waitfor is set, then wait for the disk write * of the inode to complete. */ int ffs_update(struct inode *ip, struct timespec *atime, struct timespec *mtime, int waitfor) { struct vnode *vp; struct fs *fs; struct buf *bp; int error; struct timespec ts; vp = ITOV(ip); if (vp->v_mount->mnt_flag & MNT_RDONLY) { ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE); return (0); } if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && waitfor != MNT_WAIT) return (0); getnanotime(&ts); if (ip->i_flag & IN_ACCESS) { DIP_ASSIGN(ip, atime, atime ? atime->tv_sec : ts.tv_sec); DIP_ASSIGN(ip, atimensec, atime ? atime->tv_nsec : ts.tv_nsec); } if (ip->i_flag & IN_UPDATE) { DIP_ASSIGN(ip, mtime, mtime ? mtime->tv_sec : ts.tv_sec); DIP_ASSIGN(ip, mtimensec, mtime ? mtime->tv_nsec : ts.tv_nsec); ip->i_modrev++; } if (ip->i_flag & IN_CHANGE) { DIP_ASSIGN(ip, ctime, ts.tv_sec); DIP_ASSIGN(ip, ctimensec, ts.tv_nsec); } ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE); fs = ip->i_fs; /* * Ensure that uid and gid are correct. This is a temporary * fix until fsck has been changed to do the update. */ if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) { ip->i_din1->di_ouid = ip->i_ffs1_uid; ip->i_din1->di_ogid = ip->i_ffs1_gid; } error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, &bp); if (error) { brelse(bp); return (error); } if (DOINGSOFTDEP(vp)) softdep_update_inodeblock(ip, bp, waitfor); else if (ip->i_effnlink != DIP(ip, nlink)) panic("ffs_update: bad link cnt"); #ifdef FFS2 if (ip->i_ump->um_fstype == UM_UFS2) *((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2; else #endif *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1; if (waitfor && !DOINGASYNC(vp)) { return (bwrite(bp)); } else { bdwrite(bp); return (0); } }
/* * Balloc defines the structure of file system storage * by allocating the physical blocks on a device given * the inode and the logical block number in a file. * This is the allocation strategy for UFS2. Above is * the allocation strategy for UFS1. */ int ffs_balloc_ufs2(vnode *vp, off_t startoffset, int size, Ucred *cred, int flags, Buf **bpp) { int error = 0; print("HARVEY TODO: %s\n", __func__); #if 0 struct inode *ip; struct ufs2_dinode *dp; ufs_lbn_t lbn, lastlbn; struct fs *fs; struct buf *bp, *nbp; struct ufsmount *ump; struct indir indirs[UFS_NIADDR + 2]; ufs2_daddr_t nb, newb, *bap, pref; ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1]; ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1]; int deallocated, osize, nsize, num, i, error; int unwindidx = -1; int saved_inbdflush; static struct timeval lastfail; static int curfail; int gbflags, reclaimed; ip = VTOI(vp); dp = ip->i_din2; fs = ITOFS(ip); ump = ITOUMP(ip); lbn = lblkno(fs, startoffset); size = blkoff(fs, startoffset) + size; reclaimed = 0; if (size > fs->fs_bsize) panic("ffs_balloc_ufs2: blk too big"); *bpp = nil; if (lbn < 0) return (EFBIG); gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; if (DOINGSOFTDEP(vp)) softdep_prealloc(vp, MNT_WAIT); /* * Check for allocating external data. */ if (flags & IO_EXT) { if (lbn >= UFS_NXADDR) return (EFBIG); /* * If the next write will extend the data into a new block, * and the data is currently composed of a fragment * this fragment has to be extended to be a full block. */ lastlbn = lblkno(fs, dp->di_extsize); if (lastlbn < lbn) { nb = lastlbn; osize = sblksize(fs, dp->di_extsize, nb); if (osize < fs->fs_bsize && osize > 0) { UFS_LOCK(ump); error = ffs_realloccg(ip, -1 - nb, dp->di_extb[nb], ffs_blkpref_ufs2(ip, lastlbn, (int)nb, &dp->di_extb[0]), osize, (int)fs->fs_bsize, flags, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocext(ip, nb, dbtofsb(fs, bp->b_blkno), dp->di_extb[nb], fs->fs_bsize, osize, bp); dp->di_extsize = smalllblktosize(fs, nb + 1); dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno); bp->b_xflags |= BX_ALTDATA; ip->i_flag |= IN_CHANGE; if (flags & IO_SYNC) bwrite(bp); else bawrite(bp); } } /* * All blocks are direct blocks */ if (flags & BA_METAONLY) panic("ffs_balloc_ufs2: BA_METAONLY for ext block"); nb = dp->di_extb[lbn]; if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) { error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED, gbflags, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); bp->b_xflags |= BX_ALTDATA; *bpp = bp; return (0); } if (nb != 0) { /* * Consider need to reallocate a fragment. */ osize = fragroundup(fs, blkoff(fs, dp->di_extsize)); nsize = fragroundup(fs, size); if (nsize <= osize) { error = bread_gb(vp, -1 - lbn, osize, NOCRED, gbflags, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); bp->b_xflags |= BX_ALTDATA; } else { UFS_LOCK(ump); error = ffs_realloccg(ip, -1 - lbn, dp->di_extb[lbn], ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]), osize, nsize, flags, cred, &bp); if (error) return (error); bp->b_xflags |= BX_ALTDATA; if (DOINGSOFTDEP(vp)) softdep_setup_allocext(ip, lbn, dbtofsb(fs, bp->b_blkno), nb, nsize, osize, bp); } } else { if (dp->di_extsize < smalllblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; UFS_LOCK(ump); error = ffs_alloc(ip, lbn, ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]), nsize, flags, cred, &newb); if (error) return (error); bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags); bp->b_blkno = fsbtodb(fs, newb); bp->b_xflags |= BX_ALTDATA; if (flags & BA_CLRBUF) vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) softdep_setup_allocext(ip, lbn, newb, 0, nsize, 0, bp); } dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno); ip->i_flag |= IN_CHANGE; *bpp = bp; return (0); } /* * If the next write will extend the file into a new block, * and the file is currently composed of a fragment * this fragment has to be extended to be a full block. */ lastlbn = lblkno(fs, ip->i_size); if (lastlbn < UFS_NDADDR && lastlbn < lbn) { nb = lastlbn; osize = blksize(fs, ip, nb); if (osize < fs->fs_bsize && osize > 0) { UFS_LOCK(ump); error = ffs_realloccg(ip, nb, dp->di_db[nb], ffs_blkpref_ufs2(ip, lastlbn, (int)nb, &dp->di_db[0]), osize, (int)fs->fs_bsize, flags, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, nb, dbtofsb(fs, bp->b_blkno), dp->di_db[nb], fs->fs_bsize, osize, bp); ip->i_size = smalllblktosize(fs, nb + 1); dp->di_size = ip->i_size; dp->di_db[nb] = dbtofsb(fs, bp->b_blkno); ip->i_flag |= IN_CHANGE | IN_UPDATE; if (flags & IO_SYNC) bwrite(bp); else bawrite(bp); } } /* * The first UFS_NDADDR blocks are direct blocks */ if (lbn < UFS_NDADDR) { if (flags & BA_METAONLY) panic("ffs_balloc_ufs2: BA_METAONLY for direct block"); nb = dp->di_db[lbn]; if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED, gbflags, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); *bpp = bp; return (0); } if (nb != 0) { /* * Consider need to reallocate a fragment. */ osize = fragroundup(fs, blkoff(fs, ip->i_size)); nsize = fragroundup(fs, size); if (nsize <= osize) { error = bread_gb(vp, lbn, osize, NOCRED, gbflags, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); } else { UFS_LOCK(ump); error = ffs_realloccg(ip, lbn, dp->di_db[lbn], ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_db[0]), osize, nsize, flags, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, dbtofsb(fs, bp->b_blkno), nb, nsize, osize, bp); } } else { if (ip->i_size < smalllblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; UFS_LOCK(ump); error = ffs_alloc(ip, lbn, ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_db[0]), nsize, flags, cred, &newb); if (error) return (error); bp = getblk(vp, lbn, nsize, 0, 0, gbflags); bp->b_blkno = fsbtodb(fs, newb); if (flags & BA_CLRBUF) vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, 0, nsize, 0, bp); } dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno); ip->i_flag |= IN_CHANGE | IN_UPDATE; *bpp = bp; return (0); } /* * Determine the number of levels of indirection. */ pref = 0; if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) return(error); #ifdef INVARIANTS if (num < 1) panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block"); #endif saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH); /* * Fetch the first indirect block allocating if necessary. */ --num; nb = dp->di_ib[indirs[0].in_off]; allocib = nil; allocblk = allociblk; lbns_remfree = lbns; if (nb == 0) { UFS_LOCK(ump); pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1, (ufs2_daddr_t *)0); if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags, cred, &newb)) != 0) { curthread_pflags_restore(saved_inbdflush); return (error); } pref = newb + fs->fs_frag; nb = newb; MPASS(allocblk < allociblk + nitems(allociblk)); MPASS(lbns_remfree < lbns + nitems(lbns)); *allocblk++ = nb; *lbns_remfree++ = indirs[1].in_lbn; bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, GB_UNMAPPED); bp->b_blkno = fsbtodb(fs, nb); vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocdirect(ip, UFS_NDADDR + indirs[0].in_off, newb, 0, fs->fs_bsize, 0, bp); bdwrite(bp); } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } else { if ((error = bwrite(bp)) != 0) goto fail; } allocib = &dp->di_ib[indirs[0].in_off]; *allocib = nb; ip->i_flag |= IN_CHANGE | IN_UPDATE; } /* * Fetch through the indirect blocks, allocating as necessary. */ retry: for (i = 1;;) { error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); if (error) { brelse(bp); goto fail; } bap = (ufs2_daddr_t *)bp->b_data; nb = bap[indirs[i].in_off]; if (i == num) break; i += 1; if (nb != 0) { bqrelse(bp); continue; } UFS_LOCK(ump); /* * If parent indirect has just been allocated, try to cluster * immediately following it. */ if (pref == 0) pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1, (ufs2_daddr_t *)0); if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb)) != 0) { brelse(bp); if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); UFS_UNLOCK(ump); goto retry; } if (ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem " "is full\n", fs->fs_fsmnt); } goto fail; } pref = newb + fs->fs_frag; nb = newb; MPASS(allocblk < allociblk + nitems(allociblk)); MPASS(lbns_remfree < lbns + nitems(lbns)); *allocblk++ = nb; *lbns_remfree++ = indirs[i].in_lbn; nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, GB_UNMAPPED); nbp->b_blkno = fsbtodb(fs, nb); vfs_bio_clrbuf(nbp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocindir_meta(nbp, ip, bp, indirs[i - 1].in_off, nb); bdwrite(nbp); } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) { if (nbp->b_bufsize == fs->fs_bsize) nbp->b_flags |= B_CLUSTEROK; bdwrite(nbp); } else { if ((error = bwrite(nbp)) != 0) { brelse(bp); goto fail; } } bap[indirs[i - 1].in_off] = nb; if (allocib == nil && unwindidx < 0) unwindidx = i - 1; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & IO_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } } /* * If asked only for the indirect block, then return it. */ if (flags & BA_METAONLY) { curthread_pflags_restore(saved_inbdflush); *bpp = bp; return (0); } /* * Get the data block, allocating if necessary. */ if (nb == 0) { UFS_LOCK(ump); /* * If allocating metadata at the front of the cylinder * group and parent indirect block has just been allocated, * then cluster next to it if it is the first indirect in * the file. Otherwise it has been allocated in the metadata * area, so we want to find our own place out in the data area. */ if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0)) pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off, &bap[0]); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb); if (error) { brelse(bp); if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); UFS_UNLOCK(ump); goto retry; } if (ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem " "is full\n", fs->fs_fsmnt); } goto fail; } nb = newb; MPASS(allocblk < allociblk + nitems(allociblk)); MPASS(lbns_remfree < lbns + nitems(lbns)); *allocblk++ = nb; *lbns_remfree++ = lbn; nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags); nbp->b_blkno = fsbtodb(fs, nb); if (flags & BA_CLRBUF) vfs_bio_clrbuf(nbp); if (DOINGSOFTDEP(vp)) softdep_setup_allocindir_page(ip, lbn, bp, indirs[i].in_off, nb, 0, nbp); bap[indirs[i].in_off] = nb; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & IO_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } curthread_pflags_restore(saved_inbdflush); *bpp = nbp; return (0); } brelse(bp); /* * If requested clear invalid portions of the buffer. If we * have to do a read-before-write (typical if BA_CLRBUF is set), * try to do some read-ahead in the sequential case to reduce * the number of I/O transactions. */ if (flags & BA_CLRBUF) { int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT; if (seqcount != 0 && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 && !(vm_page_count_severe() || buf_dirty_count_severe())) { error = cluster_read(vp, ip->i_size, lbn, (int)fs->fs_bsize, NOCRED, MAXBSIZE, seqcount, gbflags, &nbp); } else { error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED, gbflags, &nbp); } if (error) { brelse(nbp); goto fail; } } else {
/* * Allocate a new inode. */ static int ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct inode *ip, *pdir; struct vnode *tvp; int error; pdir = VTOI(dvp); #ifdef DIAGNOSTIC if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_makeinode: no name"); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); if (error) { return (error); } ip = VTOI(tvp); ip->i_gid = pdir->i_gid; #ifdef SUIDDIR { /* * if we are * not the owner of the directory, * and we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * Note that this drops off the execute bits for security. */ if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (pdir->i_mode & ISUID) && (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { ip->i_uid = pdir->i_uid; mode &= ~07111; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = mode; tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_nlink = 1; if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0)) ip->i_mode &= ~ISGID; } if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; /* * Make sure inode goes to disk before directory entry. */ error = ext2_update(tvp, !DOINGASYNC(tvp)); if (error) goto bad; error = ext2_direnter(ip, dvp, cnp); if (error) goto bad; *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); return (error); }
/* * Balloc defines the structure of filesystem storage * by allocating the physical blocks on a device given * the inode and the logical block number in a file. * This is the allocation strategy for UFS1. Below is * the allocation strategy for UFS2. */ int ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size, struct ucred *cred, int flags, struct buf **bpp) { struct inode *ip; struct ufs1_dinode *dp; ufs_lbn_t lbn, lastlbn; struct fs *fs; ufs1_daddr_t nb; struct buf *bp, *nbp; struct ufsmount *ump; struct indir indirs[NIADDR + 2]; int deallocated, osize, nsize, num, i, error; ufs2_daddr_t newb; ufs1_daddr_t *bap, pref; ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1]; ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1]; int unwindidx = -1; int saved_inbdflush; static struct timeval lastfail; static int curfail; int reclaimed; ip = VTOI(vp); dp = ip->i_din1; fs = ip->i_fs; ump = ip->i_ump; lbn = lblkno(fs, startoffset); size = blkoff(fs, startoffset) + size; reclaimed = 0; if (size > fs->fs_bsize) panic("ffs_balloc_ufs1: blk too big"); *bpp = NULL; if (flags & IO_EXT) return (EOPNOTSUPP); if (lbn < 0) return (EFBIG); if (DOINGSOFTDEP(vp)) softdep_prealloc(vp, MNT_WAIT); /* * If the next write will extend the file into a new block, * and the file is currently composed of a fragment * this fragment has to be extended to be a full block. */ lastlbn = lblkno(fs, ip->i_size); if (lastlbn < NDADDR && lastlbn < lbn) { nb = lastlbn; osize = blksize(fs, ip, nb); if (osize < fs->fs_bsize && osize > 0) { UFS_LOCK(ump); error = ffs_realloccg(ip, nb, dp->di_db[nb], ffs_blkpref_ufs1(ip, lastlbn, (int)nb, &dp->di_db[0]), osize, (int)fs->fs_bsize, flags, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, nb, dbtofsb(fs, bp->b_blkno), dp->di_db[nb], fs->fs_bsize, osize, bp); ip->i_size = smalllblktosize(fs, nb + 1); dp->di_size = ip->i_size; dp->di_db[nb] = dbtofsb(fs, bp->b_blkno); ip->i_flag |= IN_CHANGE | IN_UPDATE; if (flags & IO_SYNC) bwrite(bp); else bawrite(bp); } } /* * The first NDADDR blocks are direct blocks */ if (lbn < NDADDR) { if (flags & BA_METAONLY) panic("ffs_balloc_ufs1: BA_METAONLY for direct block"); nb = dp->di_db[lbn]; if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); *bpp = bp; return (0); } if (nb != 0) { /* * Consider need to reallocate a fragment. */ osize = fragroundup(fs, blkoff(fs, ip->i_size)); nsize = fragroundup(fs, size); if (nsize <= osize) { error = bread(vp, lbn, osize, NOCRED, &bp); if (error) { brelse(bp); return (error); } bp->b_blkno = fsbtodb(fs, nb); } else { UFS_LOCK(ump); error = ffs_realloccg(ip, lbn, dp->di_db[lbn], ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]), osize, nsize, flags, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, dbtofsb(fs, bp->b_blkno), nb, nsize, osize, bp); } } else { if (ip->i_size < smalllblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; UFS_LOCK(ump); error = ffs_alloc(ip, lbn, ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]), nsize, flags, cred, &newb); if (error) return (error); bp = getblk(vp, lbn, nsize, 0, 0, 0); bp->b_blkno = fsbtodb(fs, newb); if (flags & BA_CLRBUF) vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, 0, nsize, 0, bp); } dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno); ip->i_flag |= IN_CHANGE | IN_UPDATE; *bpp = bp; return (0); } /* * Determine the number of levels of indirection. */ pref = 0; if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) return(error); #ifdef INVARIANTS if (num < 1) panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block"); #endif saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH); /* * Fetch the first indirect block allocating if necessary. */ --num; nb = dp->di_ib[indirs[0].in_off]; allocib = NULL; allocblk = allociblk; lbns_remfree = lbns; if (nb == 0) { UFS_LOCK(ump); pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0); if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags, cred, &newb)) != 0) { curthread_pflags_restore(saved_inbdflush); return (error); } nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[1].in_lbn; bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0); bp->b_blkno = fsbtodb(fs, nb); vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, newb, 0, fs->fs_bsize, 0, bp); bdwrite(bp); } else { /* * Write synchronously so that indirect blocks * never point at garbage. */ if (DOINGASYNC(vp)) bdwrite(bp); else if ((error = bwrite(bp)) != 0) goto fail; } allocib = &dp->di_ib[indirs[0].in_off]; *allocib = nb; ip->i_flag |= IN_CHANGE | IN_UPDATE; } /* * Fetch through the indirect blocks, allocating as necessary. */ retry: for (i = 1;;) { error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); if (error) { brelse(bp); goto fail; } bap = (ufs1_daddr_t *)bp->b_data; nb = bap[indirs[i].in_off]; if (i == num) break; i += 1; if (nb != 0) { bqrelse(bp); continue; } UFS_LOCK(ump); if (pref == 0) pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0); if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb)) != 0) { brelse(bp); if (++reclaimed == 1) { UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); UFS_UNLOCK(ump); goto retry; } if (ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem " "is full\n", fs->fs_fsmnt); } goto fail; } nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[i].in_lbn; nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); vfs_bio_clrbuf(nbp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocindir_meta(nbp, ip, bp, indirs[i - 1].in_off, nb); bdwrite(nbp); } else { /* * Write synchronously so that indirect blocks * never point at garbage. */ if ((error = bwrite(nbp)) != 0) { brelse(bp); goto fail; } } bap[indirs[i - 1].in_off] = nb; if (allocib == NULL && unwindidx < 0) unwindidx = i - 1; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & IO_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } } /* * If asked only for the indirect block, then return it. */ if (flags & BA_METAONLY) { curthread_pflags_restore(saved_inbdflush); *bpp = bp; return (0); } /* * Get the data block, allocating if necessary. */ if (nb == 0) { UFS_LOCK(ump); pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb); if (error) { brelse(bp); if (++reclaimed == 1) { UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); UFS_UNLOCK(ump); goto retry; } if (ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem " "is full\n", fs->fs_fsmnt); } goto fail; } nb = newb; *allocblk++ = nb; *lbns_remfree++ = lbn; nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); if (flags & BA_CLRBUF) vfs_bio_clrbuf(nbp); if (DOINGSOFTDEP(vp)) softdep_setup_allocindir_page(ip, lbn, bp, indirs[i].in_off, nb, 0, nbp); bap[indirs[i].in_off] = nb; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & IO_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } curthread_pflags_restore(saved_inbdflush); *bpp = nbp; return (0); } brelse(bp); if (flags & BA_CLRBUF) { int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT; if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, ip->i_size, lbn, (int)fs->fs_bsize, NOCRED, MAXBSIZE, seqcount, &nbp); } else { error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp); } if (error) { brelse(nbp); goto fail; } } else {
/* * Truncate the inode ip to at most length size, freeing the * disk blocks. */ int ffs_truncate(vnode *vp, off_t length, int flags, Ucred *cred) { print("HARVEY TODO: %s\n", __func__); #if 0 struct inode *ip; ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR]; ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR]; ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR]; ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno; struct bufobj *bo; struct fs *fs; struct buf *bp; struct ufsmount *ump; int softdeptrunc, journaltrunc; int needextclean, extblocks; int offset, size, level, nblocks; int i, error, allerror, indiroff, waitforupdate; off_t osize; ip = VTOI(vp); ump = VFSTOUFS(vp->v_mount); fs = ump->um_fs; bo = &vp->v_bufobj; ASSERT_VOP_LOCKED(vp, "ffs_truncate"); if (length < 0) return (EINVAL); if (length > fs->fs_maxfilesize) return (EFBIG); #ifdef QUOTA error = getinoquota(ip); if (error) return (error); #endif /* * Historically clients did not have to specify which data * they were truncating. So, if not specified, we assume * traditional behavior, e.g., just the normal data. */ if ((flags & (IO_EXT | IO_NORMAL)) == 0) flags |= IO_NORMAL; if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp)) flags |= IO_SYNC; waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp); /* * If we are truncating the extended-attributes, and cannot * do it with soft updates, then do it slowly here. If we are * truncating both the extended attributes and the file contents * (e.g., the file is being unlinked), then pick it off with * soft updates below. */ allerror = 0; needextclean = 0; softdeptrunc = 0; journaltrunc = DOINGSUJ(vp); if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0) softdeptrunc = !softdep_slowdown(vp); extblocks = 0; datablocks = DIP(ip, i_blocks); if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) { extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize)); datablocks -= extblocks; } if ((flags & IO_EXT) && extblocks > 0) { if (length != 0) panic("ffs_truncate: partial trunc of extdata"); if (softdeptrunc || journaltrunc) { if ((flags & IO_NORMAL) == 0) goto extclean; needextclean = 1; } else { if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) return (error); #ifdef QUOTA (void) chkdq(ip, -extblocks, NOCRED, 0); #endif vinvalbuf(vp, V_ALT, 0, 0); vn_pages_remove(vp, OFF_TO_IDX(lblktosize(fs, -extblocks)), 0); osize = ip->i_din2->di_extsize; ip->i_din2->di_blocks -= extblocks; ip->i_din2->di_extsize = 0; for (i = 0; i < UFS_NXADDR; i++) { oldblks[i] = ip->i_din2->di_extb[i]; ip->i_din2->di_extb[i] = 0; } ip->i_flag |= IN_CHANGE; if ((error = ffs_update(vp, waitforupdate))) return (error); for (i = 0; i < UFS_NXADDR; i++) { if (oldblks[i] == 0) continue; ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i], sblksize(fs, osize, i), ip->i_number, vp->v_type, nil); } } } if ((flags & IO_NORMAL) == 0) return (0); if (vp->v_type == VLNK && (ip->i_size < vp->v_mount->mnt_maxsymlinklen || datablocks == 0)) { #ifdef INVARIANTS if (length != 0) panic("ffs_truncate: partial truncate of symlink"); #endif bzero(SHORTLINK(ip), (uint)ip->i_size); ip->i_size = 0; DIP_SET(ip, i_size, 0); ip->i_flag |= IN_CHANGE | IN_UPDATE; if (needextclean) goto extclean; return (ffs_update(vp, waitforupdate)); } if (ip->i_size == length) { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (needextclean) goto extclean; return (ffs_update(vp, 0)); } if (fs->fs_ronly) panic("ffs_truncate: read-only filesystem"); if (IS_SNAPSHOT(ip)) ffs_snapremove(vp); vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; osize = ip->i_size; /* * Lengthen the size of the file. We must ensure that the * last byte of the file is allocated. Since the smallest * value of osize is 0, length will be at least 1. */ if (osize < length) { vnode_pager_setsize(vp, length); flags |= BA_CLRBUF; error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); if (error) { vnode_pager_setsize(vp, osize); return (error); } ip->i_size = length; DIP_SET(ip, i_size, length); if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; if (flags & IO_SYNC) bwrite(bp); else if (DOINGASYNC(vp)) bdwrite(bp); else bawrite(bp); ip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(vp, waitforupdate)); } /* * Lookup block number for a given offset. Zero length files * have no blocks, so return a blkno of -1. */ lbn = lblkno(fs, length - 1); if (length == 0) { blkno = -1; } else if (lbn < UFS_NDADDR) { blkno = DIP(ip, i_db[lbn]); } else { error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize, cred, BA_METAONLY, &bp); if (error) return (error); indiroff = (lbn - UFS_NDADDR) % NINDIR(fs); if (I_IS_UFS1(ip)) blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff]; else blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff]; /* * If the block number is non-zero, then the indirect block * must have been previously allocated and need not be written. * If the block number is zero, then we may have allocated * the indirect block and hence need to write it out. */ if (blkno != 0) brelse(bp); else if (flags & IO_SYNC) bwrite(bp); else bdwrite(bp); } /* * If the block number at the new end of the file is zero, * then we must allocate it to ensure that the last block of * the file is allocated. Soft updates does not handle this * case, so here we have to clean up the soft updates data * structures describing the allocation past the truncation * point. Finding and deallocating those structures is a lot of * work. Since partial truncation with a hole at the end occurs * rarely, we solve the problem by syncing the file so that it * will have no soft updates data structures left. */ if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) return (error); if (blkno != 0 && DOINGSOFTDEP(vp)) { if (softdeptrunc == 0 && journaltrunc == 0) { /* * If soft updates cannot handle this truncation, * clean up soft dependency data structures and * fall through to the synchronous truncation. */ if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) return (error); } else { flags = IO_NORMAL | (needextclean ? IO_EXT: 0); if (journaltrunc) softdep_journal_freeblocks(ip, cred, length, flags); else softdep_setup_freeblocks(ip, length, flags); ASSERT_VOP_LOCKED(vp, "ffs_truncate1"); if (journaltrunc == 0) { ip->i_flag |= IN_CHANGE | IN_UPDATE; error = ffs_update(vp, 0); } return (error); } } /* * Shorten the size of the file. If the last block of the * shortened file is unallocated, we must allocate it. * Additionally, if the file is not being truncated to a * block boundary, the contents of the partial block * following the end of the file must be zero'ed in * case it ever becomes accessible again because of * subsequent file growth. Directories however are not * zero'ed as they should grow back initialized to empty. */ offset = blkoff(fs, length); if (blkno != 0 && offset == 0) { ip->i_size = length; DIP_SET(ip, i_size, length); } else { lbn = lblkno(fs, length); flags |= BA_CLRBUF; error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp); if (error) return (error); /* * When we are doing soft updates and the UFS_BALLOC * above fills in a direct block hole with a full sized * block that will be truncated down to a fragment below, * we must flush out the block dependency with an FSYNC * so that we do not get a soft updates inconsistency * when we create the fragment below. */ if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR && fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0) return (error); ip->i_size = length; DIP_SET(ip, i_size, length); size = blksize(fs, ip, lbn); if (vp->v_type != VDIR && offset != 0) bzero((char *)bp->b_data + offset, (uint)(size - offset)); /* Kirk's code has reallocbuf(bp, size, 1) here */ allocbuf(bp, size); if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; if (flags & IO_SYNC) bwrite(bp); else if (DOINGASYNC(vp)) bdwrite(bp); else bawrite(bp); } /* * Calculate index into inode's block list of * last direct and indirect blocks (if any) * which we want to keep. Lastblock is -1 when * the file is truncated to 0. */ lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1; lastiblock[SINGLE] = lastblock - UFS_NDADDR; lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); nblocks = btodb(fs->fs_bsize); /* * Update file and block pointers on disk before we start freeing * blocks. If we crash before free'ing blocks below, the blocks * will be returned to the free list. lastiblock values are also * normalized to -1 for calls to ffs_indirtrunc below. */ for (level = TRIPLE; level >= SINGLE; level--) { oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]); if (lastiblock[level] < 0) { DIP_SET(ip, i_ib[level], 0); lastiblock[level] = -1; } } for (i = 0; i < UFS_NDADDR; i++) { oldblks[i] = DIP(ip, i_db[i]); if (i > lastblock) DIP_SET(ip, i_db[i], 0); } ip->i_flag |= IN_CHANGE | IN_UPDATE; allerror = ffs_update(vp, waitforupdate); /* * Having written the new inode to disk, save its new configuration * and put back the old block pointers long enough to process them. * Note that we save the new block configuration so we can check it * when we are done. */ for (i = 0; i < UFS_NDADDR; i++) { newblks[i] = DIP(ip, i_db[i]); DIP_SET(ip, i_db[i], oldblks[i]); } for (i = 0; i < UFS_NIADDR; i++) { newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]); DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]); } ip->i_size = osize; DIP_SET(ip, i_size, osize); error = vtruncbuf(vp, cred, length, fs->fs_bsize); if (error && (allerror == 0)) allerror = error; /* * Indirect blocks first. */ indir_lbn[SINGLE] = -UFS_NDADDR; indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; for (level = TRIPLE; level >= SINGLE; level--) { bn = DIP(ip, i_ib[level]); if (bn != 0) { error = ffs_indirtrunc(ip, indir_lbn[level], fsbtodb(fs, bn), lastiblock[level], level, &count); if (error) allerror = error; blocksreleased += count; if (lastiblock[level] < 0) { DIP_SET(ip, i_ib[level], 0); ffs_blkfree(ump, fs, ump->um_devvp, bn, fs->fs_bsize, ip->i_number, vp->v_type, nil); blocksreleased += nblocks; } } if (lastiblock[level] >= 0) goto done; } /* * All whole direct blocks or frags. */ for (i = UFS_NDADDR - 1; i > lastblock; i--) { long bsize; bn = DIP(ip, i_db[i]); if (bn == 0) continue; DIP_SET(ip, i_db[i], 0); bsize = blksize(fs, ip, i); ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number, vp->v_type, nil); blocksreleased += btodb(bsize); } if (lastblock < 0) goto done; /* * Finally, look for a change in size of the * last direct block; release any frags. */ bn = DIP(ip, i_db[lastblock]); if (bn != 0) { long oldspace, newspace; /* * Calculate amount of space we're giving * back as old block size minus new block size. */ oldspace = blksize(fs, ip, lastblock); ip->i_size = length; DIP_SET(ip, i_size, length); newspace = blksize(fs, ip, lastblock); if (newspace == 0) panic("ffs_truncate: newspace"); if (oldspace - newspace > 0) { /* * Block number of space to be free'd is * the old block # plus the number of frags * required for the storage we're keeping. */ bn += numfrags(fs, newspace); ffs_blkfree(ump, fs, ump->um_devvp, bn, oldspace - newspace, ip->i_number, vp->v_type, nil); blocksreleased += btodb(oldspace - newspace); } } done: #ifdef INVARIANTS for (level = SINGLE; level <= TRIPLE; level++) if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level])) panic("ffs_truncate1"); for (i = 0; i < UFS_NDADDR; i++) if (newblks[i] != DIP(ip, i_db[i])) panic("ffs_truncate2"); BO_LOCK(bo); if (length == 0 && (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("ffs_truncate3"); BO_UNLOCK(bo); #endif /* INVARIANTS */ /* * Put back the real size. */ ip->i_size = length; DIP_SET(ip, i_size, length); if (DIP(ip, i_blocks) >= blocksreleased) DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased); else /* sanity */ DIP_SET(ip, i_blocks, 0); ip->i_flag |= IN_CHANGE; #ifdef QUOTA (void) chkdq(ip, -blocksreleased, NOCRED, 0); #endif return (allerror); extclean: if (journaltrunc) softdep_journal_freeblocks(ip, cred, length, IO_EXT); else softdep_setup_freeblocks(ip, length, IO_EXT); return (ffs_update(vp, waitforupdate)); #endif // 0 return 0; }
/* * Rmdir system call. */ int ufs_rmdir(void *v) { struct vop_rmdir_args *ap = v; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; int error; ip = VTOI(vp); dp = VTOI(dvp); /* * No rmdir "." or of mounted on directories. */ if (dp == ip || vp->v_mountedhere != NULL) { if (dp == ip) vrele(dvp); else vput(dvp); vput(vp); return (EINVAL); } /* * Do not remove a directory that is in the process of being renamed. * Verify the directory is empty (and valid). Rmdir ".." will not be * valid since ".." will contain a reference to the current directory * and thus be non-empty. */ error = 0; if (ip->i_flag & IN_RENAME) { error = EINVAL; goto out; } if (ip->i_effnlink != 2 || !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) { error = ENOTEMPTY; goto out; } if ((DIP(dp, flags) & APPEND) || (DIP(ip, flags) & (IMMUTABLE | APPEND))) { error = EPERM; goto out; } /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ dp->i_effnlink--; ip->i_effnlink--; if (DOINGSOFTDEP(vp)) { softdep_change_linkcnt(dp, 0); softdep_change_linkcnt(ip, 0); } if ((error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1)) != 0) { dp->i_effnlink++; ip->i_effnlink++; if (DOINGSOFTDEP(vp)) { softdep_change_linkcnt(dp, 0); softdep_change_linkcnt(ip, 0); } goto out; } VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK); cache_purge(dvp); /* * Truncate inode. The only stuff left in the directory is "." and * "..". The "." reference is inconsequential since we are quashing * it. The soft dependency code will arrange to do these operations * after the parent directory entry has been deleted on disk, so * when running with that code we avoid doing them now. */ if (!DOINGSOFTDEP(vp)) { int ioflag; DIP_ADD(dp, nlink, -1); dp->i_flag |= IN_CHANGE; DIP_ADD(ip, nlink, -1); ip->i_flag |= IN_CHANGE; ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC; error = UFS_TRUNCATE(ip, (off_t)0, ioflag, cnp->cn_cred); } cache_purge(vp); #ifdef UFS_DIRHASH /* Kill any active hash; i_effnlink == 0, so it will not come back. */ if (ip->i_dirhash != NULL) ufsdirhash_free(ip); #endif out: VN_KNOTE(vp, NOTE_DELETE); vput(dvp); vput(vp); return (error); }
/* * Extended attribute area writing. */ int ffs_ea_write(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) { struct inode *ip; struct ufs2_dinode *dp; struct fs *fs; struct buf *bp; off_t osize; size_t resid; int error, flags, xfersize; ip = VTOI(vp); fs = ip->i_fs; dp = ip->i_din2; error = 0; /* Don't cross the limit for the extended attribute area length. */ if ((unsigned)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize) return (EFBIG); /* * Save the original amount of data to be written ('uio->uio_resid') as * well as the extended attribute area size ('dp->di_extsize') in case * we need to truncate in the end due to an unpredicted error. */ resid = uio->uio_resid; osize = dp->di_extsize; /* Prepare flags to be passed to UFS_BUF_ALLOC(). */ flags = IO_EXT | B_CLRBUF; if (!DOINGASYNC(vp)) flags |= B_SYNC; /* * Loop over the amount of data requested by the caller, stopping only * if an error occurs. By default, we always try to write a file system * block worth of bytes per iteration ('xfersize'). Check this value * against what is left to be copied ('uio->uio_resid'). If we are * writing less than a full block, ask the buffer to be cleaned first. */ while (uio->uio_resid > 0) { xfersize = fs->fs_bsize; if (uio->uio_resid < xfersize) { xfersize = uio->uio_resid; flags |= B_CLRBUF; } /* Ask the ffs_balloc.c code for the block. */ error = UFS_BUF_ALLOC(ip, uio->uio_offset, xfersize, ucred, flags, &bp); if (error) break; /* Check if we are growing the extended attributes area. */ if (uio->uio_offset + xfersize > dp->di_extsize) dp->di_extsize = uio->uio_offset + xfersize; error = uiomove(bp->b_data, xfersize, uio); if (error) { clrbuf(bp); /* Get rid of stray contents. */ error = EIO; break; } /* * We use the same criteria for calling bwrite(), bawrite(), or * bdwrite() as the rest of the FFS code does. */ if (ioflag & IO_SYNC) bwrite(bp); else if (xfersize == fs->fs_bsize) bawrite(bp); else bdwrite(bp); if (error || xfersize == 0) break; ip->i_flag |= IN_CHANGE | IN_UPDATE; } /* * If we successfully wrote any data, and we are not the superuser we * clear the setuid and setgid bits as a precaution against tampering. */ if ((DIP(ip, mode) & (ISUID | ISGID)) && resid > uio->uio_resid) if (ucred && suser_ucred(ucred)) DIP(ip, mode) &= ~(ISUID | ISGID); if (error) { /* Release blocks if we failed. */ if (ioflag & IO_UNIT) { ffs_truncate(ip, osize, flags, ucred); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } return (error); } /* If needed, sync the inode now. */ if (resid > uio->uio_resid && (ioflag & IO_SYNC)) return (ffs_update(ip, NULL, NULL, MNT_WAIT)); return (0); }
/* * Remove a directory entry. At this point the file represented by the * directory entry to be removed is still full length until no one has it * open. When the file no longer being used msdosfs_inactive() is called * and will truncate the file to 0 length. When the vnode containing the * denode is needed for some other purpose by VFS it will call * msdosfs_reclaim() which will remove the denode from the denode cache. * * pdep directory where the entry is removed * dep file to be removed */ int removede(struct denode *pdep, struct denode *dep) { int error; struct direntry *ep; struct buf *bp; daddr_t bn; int blsize; struct msdosfsmount *pmp = pdep->de_pmp; u_long offset = pdep->de_fndoffset; #ifdef MSDOSFS_DEBUG printf("removede(): filename %s, dep %p, offset %08lx\n", dep->de_Name, dep, offset); #endif dep->de_refcnt--; offset += sizeof(struct direntry); do { offset -= sizeof(struct direntry); error = pcbmap(pdep, de_cluster(pmp, offset), &bn, 0, &blsize); if (error) return error; error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { brelse(bp); return error; } ep = bptoep(pmp, bp, offset); /* * Check whether, if we came here the second time, i.e. * when underflowing into the previous block, the last * entry in this block is a longfilename entry, too. */ if (ep->deAttributes != ATTR_WIN95 && offset != pdep->de_fndoffset) { brelse(bp); break; } offset += sizeof(struct direntry); while (1) { /* * We are a bit aggressive here in that we delete any Win95 * entries preceding this entry, not just the ones we "own". * Since these presumably aren't valid anyway, * there should be no harm. */ offset -= sizeof(struct direntry); ep--->deName[0] = SLOT_DELETED; if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) || !(offset & pmp->pm_crbomask) || ep->deAttributes != ATTR_WIN95) break; } if (DOINGASYNC(DETOV(pdep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; } while (!(pmp->pm_flags & MSDOSFSMNT_NOWIN95) && !(offset & pmp->pm_crbomask) && offset); return 0; }
/* * dep - directory entry to copy into the directory * ddep - directory to add to * depp - return the address of the denode for the created directory entry * if depp != 0 * cnp - componentname needed for Win95 long filenames */ int createde(struct denode *dep, struct denode *ddep, struct denode **depp, struct componentname *cnp) { int error; u_long dirclust, diroffset; struct direntry *ndep; struct msdosfsmount *pmp = ddep->de_pmp; struct buf *bp; daddr_t bn; int blsize; #ifdef MSDOSFS_DEBUG printf("createde(dep %p, ddep %p, depp %p, cnp %p)\n", dep, ddep, depp, cnp); #endif /* * If no space left in the directory then allocate another cluster * and chain it onto the end of the file. There is one exception * to this. That is, if the root directory has no more space it * can NOT be expanded. extendfile() checks for and fails attempts * to extend the root directory. We just return an error in that * case. */ if (ddep->de_fndoffset >= ddep->de_FileSize) { diroffset = ddep->de_fndoffset + sizeof(struct direntry) - ddep->de_FileSize; dirclust = de_clcount(pmp, diroffset); error = extendfile(ddep, dirclust, 0, 0, DE_CLEAR); if (error) { (void)detrunc(ddep, ddep->de_FileSize, 0, NOCRED); return error; } /* * Update the size of the directory */ ddep->de_FileSize += de_cn2off(pmp, dirclust); } /* * We just read in the cluster with space. Copy the new directory * entry in. Then write it to disk. NOTE: DOS directories * do not get smaller as clusters are emptied. */ error = pcbmap(ddep, de_cluster(pmp, ddep->de_fndoffset), &bn, &dirclust, &blsize); if (error) return error; diroffset = ddep->de_fndoffset; if (dirclust != MSDOSFSROOT) diroffset &= pmp->pm_crbomask; if ((error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp)) != 0) { brelse(bp); return error; } ndep = bptoep(pmp, bp, ddep->de_fndoffset); DE_EXTERNALIZE(ndep, dep); /* * Now write the Win95 long name */ if (ddep->de_fndcnt > 0) { u_int8_t chksum = winChksum(ndep->deName); const u_char *un = (const u_char *)cnp->cn_nameptr; int unlen = cnp->cn_namelen; int cnt = 1; while (--ddep->de_fndcnt >= 0) { if (!(ddep->de_fndoffset & pmp->pm_crbomask)) { if (DOINGASYNC(DETOV(ddep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; ddep->de_fndoffset -= sizeof(struct direntry); error = pcbmap(ddep, de_cluster(pmp, ddep->de_fndoffset), &bn, 0, &blsize); if (error) return error; error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp); if (error) { brelse(bp); return error; } ndep = bptoep(pmp, bp, ddep->de_fndoffset); } else { ndep--; ddep->de_fndoffset -= sizeof(struct direntry); } if (!unix2winfn(un, unlen, (struct winentry *)ndep, cnt++, chksum, pmp)) break; } } if (DOINGASYNC(DETOV(ddep))) bdwrite(bp); else if ((error = bwrite(bp)) != 0) return error; /* * If they want us to return with the denode gotten. */ if (depp) { if (dep->de_Attributes & ATTR_DIRECTORY) { dirclust = dep->de_StartCluster; if (FAT32(pmp) && dirclust == pmp->pm_rootdirblk) dirclust = MSDOSFSROOT; if (dirclust == MSDOSFSROOT) diroffset = MSDOSFSROOT_OFS; else diroffset = 0; } return deget(pmp, dirclust, diroffset, depp); } return 0; }
/* * Release blocks associated with the inode ip and stored in the indirect * block bn. Blocks are free'd in LIFO order up to (but not including) * lastbn. If level is greater than SINGLE, the block is an indirect block * and recursive calls to indirtrunc must be used to cleanse other indirect * blocks. * * NB: triple indirect blocks are untested. */ static int ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn, ufs_daddr_t lastbn, int level, long *countp) { int i; struct buf *bp; struct fs *fs = ip->i_fs; ufs_daddr_t *bap; struct vnode *vp; ufs_daddr_t *copy = NULL, nb, nlbn, last; long blkcount, factor; int nblocks, blocksreleased = 0; int error = 0, allerror = 0; /* * Calculate index in current block of last * block to be kept. -1 indicates the entire * block so we need not calculate the index. */ factor = 1; for (i = SINGLE; i < level; i++) factor *= NINDIR(fs); last = lastbn; if (lastbn > 0) last /= factor; nblocks = btodb(fs->fs_bsize); /* * Get buffer of block pointers, zero those entries corresponding * to blocks to be free'd, and update on disk copy first. Since * double(triple) indirect before single(double) indirect, calls * to bmap on these blocks will fail. However, we already have * the on disk address, so we have to set the bio_offset field * explicitly instead of letting bread do everything for us. */ vp = ITOV(ip); bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0); if ((bp->b_flags & B_CACHE) == 0) { bp->b_flags &= ~(B_ERROR|B_INVAL); bp->b_cmd = BUF_CMD_READ; if (bp->b_bcount > bp->b_bufsize) panic("ffs_indirtrunc: bad buffer size"); /* * BIO is bio2 which chains back to bio1. We wait * on bio1. */ bp->b_bio2.bio_offset = dbtodoff(fs, dbn); bp->b_bio1.bio_done = biodone_sync; bp->b_bio1.bio_flags |= BIO_SYNC; vfs_busy_pages(vp, bp); /* * Access the block device layer using the device vnode * and the translated block number (bio2) instead of the * file vnode (vp) and logical block number (bio1). * * Even though we are bypassing the vnode layer, we still * want the vnode state to indicate that an I/O on its behalf * is in progress. */ bio_start_transaction(&bp->b_bio1, &vp->v_track_read); vn_strategy(ip->i_devvp, &bp->b_bio2); error = biowait(&bp->b_bio1, "biord"); } if (error) { brelse(bp); *countp = 0; return (error); } bap = (ufs_daddr_t *)bp->b_data; if (lastbn != -1) { copy = kmalloc(fs->fs_bsize, M_TEMP, M_WAITOK); bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize); bzero((caddr_t)&bap[last + 1], (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t)); if (DOINGASYNC(vp)) { bawrite(bp); } else { error = bwrite(bp); if (error) allerror = error; } bap = copy; } /* * Recursively free totally unused blocks. */ for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; i--, nlbn += factor) { nb = bap[i]; if (nb == 0) continue; if (level > SINGLE) { if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), (ufs_daddr_t)-1, level - 1, &blkcount)) != 0) allerror = error; blocksreleased += blkcount; } ffs_blkfree(ip, nb, fs->fs_bsize); blocksreleased += nblocks; } /* * Recursively free last partial block. */ if (level > SINGLE && lastbn >= 0) { last = lastbn % factor; nb = bap[i]; if (nb != 0) { error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), last, level - 1, &blkcount); if (error) allerror = error; blocksreleased += blkcount; } } if (copy != NULL) { kfree(copy, M_TEMP); } else { bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); } *countp = blocksreleased; return (allerror); }
/* * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size, * struct ucred *a_cred, int a_flags, struct buf *a_bpp) * * Balloc defines the structure of filesystem storage by allocating * the physical blocks on a device given the inode and the logical * block number in a file. * * NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions * of the buffer. However, any dirty bits will override missing * valid bits. This case occurs when writable mmaps are truncated * and then extended. */ int ffs_balloc(struct vop_balloc_args *ap) { struct inode *ip; ufs_daddr_t lbn; int size; struct ucred *cred; int flags; struct fs *fs; ufs_daddr_t nb; struct buf *bp, *nbp, *dbp; struct vnode *vp; struct indir indirs[NIADDR + 2]; ufs_daddr_t newb, *bap, pref; int deallocated, osize, nsize, num, i, error; ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1]; ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1]; int unwindidx; int seqcount; vp = ap->a_vp; ip = VTOI(vp); fs = ip->i_fs; lbn = lblkno(fs, ap->a_startoffset); size = blkoff(fs, ap->a_startoffset) + ap->a_size; if (size > fs->fs_bsize) panic("ffs_balloc: blk too big"); *ap->a_bpp = NULL; if (lbn < 0) return (EFBIG); cred = ap->a_cred; flags = ap->a_flags; /* * The vnode must be locked for us to be able to safely mess * around with the inode. */ if (vn_islocked(vp) != LK_EXCLUSIVE) { panic("ffs_balloc: vnode %p not exclusively locked!", vp); } /* * If the next write will extend the file into a new block, * and the file is currently composed of a fragment * this fragment has to be extended to be a full block. */ nb = lblkno(fs, ip->i_size); if (nb < NDADDR && nb < lbn) { /* * The filesize prior to this write can fit in direct * blocks (ex. fragmentation is possibly done) * we are now extending the file write beyond * the block which has end of the file prior to this write. */ osize = blksize(fs, ip, nb); /* * osize gives disk allocated size in the last block. It is * either in fragments or a file system block size. */ if (osize < fs->fs_bsize && osize > 0) { /* A few fragments are already allocated, since the * current extends beyond this block allocated the * complete block as fragments are on in last block. */ error = ffs_realloccg(ip, nb, ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]), osize, (int)fs->fs_bsize, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, nb, dofftofsb(fs, bp->b_bio2.bio_offset), ip->i_db[nb], fs->fs_bsize, osize, bp); /* adjust the inode size, we just grew */ ip->i_size = smalllblktosize(fs, nb + 1); ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset); ip->i_flag |= IN_CHANGE | IN_UPDATE; if (flags & B_SYNC) bwrite(bp); else bawrite(bp); /* bp is already released here */ } } /* * The first NDADDR blocks are direct blocks */ if (lbn < NDADDR) { nb = ip->i_db[lbn]; if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp); if (error) { brelse(bp); return (error); } bp->b_bio2.bio_offset = fsbtodoff(fs, nb); *ap->a_bpp = bp; return (0); } if (nb != 0) { /* * Consider need to reallocate a fragment. */ osize = fragroundup(fs, blkoff(fs, ip->i_size)); nsize = fragroundup(fs, size); if (nsize <= osize) { error = bread(vp, lblktodoff(fs, lbn), osize, &bp); if (error) { brelse(bp); return (error); } bp->b_bio2.bio_offset = fsbtodoff(fs, nb); } else { /* * NOTE: ffs_realloccg() issues a bread(). */ error = ffs_realloccg(ip, lbn, ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), osize, nsize, cred, &bp); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, dofftofsb(fs, bp->b_bio2.bio_offset), nb, nsize, osize, bp); } } else { if (ip->i_size < smalllblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; error = ffs_alloc(ip, lbn, ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), nsize, cred, &newb); if (error) return (error); bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0); bp->b_bio2.bio_offset = fsbtodoff(fs, newb); if (flags & B_CLRBUF) vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, 0, nsize, 0, bp); } ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset); ip->i_flag |= IN_CHANGE | IN_UPDATE; *ap->a_bpp = bp; return (0); } /* * Determine the number of levels of indirection. */ pref = 0; if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) return(error); #ifdef DIAGNOSTIC if (num < 1) panic ("ffs_balloc: ufs_bmaparray returned indirect block"); #endif /* * Get a handle on the data block buffer before working through * indirect blocks to avoid a deadlock between the VM system holding * a locked VM page and issuing a BMAP (which tries to lock the * indirect blocks), and the filesystem holding a locked indirect * block and then trying to read a data block (which tries to lock * the underlying VM pages). */ dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0); /* * Setup undo history */ allocib = NULL; allocblk = allociblk; lbns_remfree = lbns; unwindidx = -1; /* * Fetch the first indirect block directly from the inode, allocating * one if necessary. */ --num; nb = ip->i_ib[indirs[0].in_off]; if (nb == 0) { pref = ffs_blkpref(ip, lbn, 0, NULL); /* * If the filesystem has run out of space we can skip the * full fsync/undo of the main [fail] case since no undo * history has been built yet. Hence the goto fail2. */ if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) goto fail2; nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[1].in_lbn; bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn), fs->fs_bsize, 0, 0); bp->b_bio2.bio_offset = fsbtodoff(fs, nb); vfs_bio_clrbuf(bp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, newb, 0, fs->fs_bsize, 0, bp); bdwrite(bp); } else { /* * Write synchronously so that indirect blocks * never point at garbage. */ if (DOINGASYNC(vp)) bdwrite(bp); else if ((error = bwrite(bp)) != 0) goto fail; } allocib = &ip->i_ib[indirs[0].in_off]; *allocib = nb; ip->i_flag |= IN_CHANGE | IN_UPDATE; } /* * Fetch through the indirect blocks, allocating as necessary. */ for (i = 1;;) { error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp); if (error) { brelse(bp); goto fail; } bap = (ufs_daddr_t *)bp->b_data; nb = bap[indirs[i].in_off]; if (i == num) break; i += 1; if (nb != 0) { bqrelse(bp); continue; } if (pref == 0) pref = ffs_blkpref(ip, lbn, 0, NULL); if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) { brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[i].in_lbn; nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn), fs->fs_bsize, 0, 0); nbp->b_bio2.bio_offset = fsbtodoff(fs, nb); vfs_bio_clrbuf(nbp); if (DOINGSOFTDEP(vp)) { softdep_setup_allocindir_meta(nbp, ip, bp, indirs[i - 1].in_off, nb); bdwrite(nbp); } else { /* * Write synchronously so that indirect blocks * never point at garbage. */ if ((error = bwrite(nbp)) != 0) { brelse(bp); goto fail; } } bap[indirs[i - 1].in_off] = nb; if (allocib == NULL && unwindidx < 0) unwindidx = i - 1; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } } /* * Get the data block, allocating if necessary. We have already * called getblk() on the data block buffer, dbp. If we have to * allocate it and B_CLRBUF has been set the inference is an intention * to zero out the related disk blocks, so we do not have to issue * a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is * not set the caller intends to overwrite the entire contents of the * buffer and we don't waste time trying to clean up the contents. * * bp references the current indirect block. When allocating, * the block must be updated. */ if (nb == 0) { pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb); if (error) { brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; *lbns_remfree++ = lbn; dbp->b_bio2.bio_offset = fsbtodoff(fs, nb); if (flags & B_CLRBUF) vfs_bio_clrbuf(dbp); if (DOINGSOFTDEP(vp)) softdep_setup_allocindir_page(ip, lbn, bp, indirs[i].in_off, nb, 0, dbp); bap[indirs[i].in_off] = nb; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { bwrite(bp); } else { if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } *ap->a_bpp = dbp; return (0); } brelse(bp); /* * At this point all related indirect blocks have been allocated * if necessary and released. bp is no longer valid. dbp holds * our getblk()'d data block. * * XXX we previously performed a cluster_read operation here. */ if (flags & B_CLRBUF) { /* * If B_CLRBUF is set we must validate the invalid portions * of the buffer. This typically requires a read-before- * write. The strategy call will fill in bio_offset in that * case. * * If we hit this case we do a cluster read if possible * since nearby data blocks are likely to be accessed soon * too. */ if ((dbp->b_flags & B_CACHE) == 0) { bqrelse(dbp); seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT; if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, (off_t)ip->i_size, lblktodoff(fs, lbn), (int)fs->fs_bsize, fs->fs_bsize, seqcount * BKVASIZE, &dbp); } else { error = bread(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, &dbp); } if (error) goto fail; } else {
/* * Mkdir system call */ static int ext2_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; struct vnode *tvp; struct dirtemplate dirtemplate, *dtp; int error, dmode; #ifdef DIAGNOSTIC if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_mkdir: no name"); #endif dp = VTOI(dvp); if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto out; } dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ext2_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; ip = VTOI(tvp); ip->i_gid = dp->i_gid; #ifdef SUIDDIR { /* * if we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * The new directory also inherits the SUID bit. * If user's UID and dir UID are the same, * 'give it away' so that the SUID is still forced on. */ if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (dp->i_mode & ISUID) && dp->i_uid) { dmode |= ISUID; ip->i_uid = dp->i_uid; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = dmode; tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_nlink = 2; if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; error = ext2_update(tvp, 1); /* * Bump link count in parent directory * to reflect work done below. Should * be done before reference is created * so reparation is possible if we crash. */ dp->i_nlink++; dp->i_flag |= IN_CHANGE; error = ext2_update(dvp, !DOINGASYNC(dvp)); if (error) goto bad; /* Initialize directory with "." and ".." from static template. */ if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) dtp = &mastertemplate; else dtp = &omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = ip->i_number; dirtemplate.dotdot_ino = dp->i_number; /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE * so let's just redefine it - for this function only */ #undef DIRBLKSIZ #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate, sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, NULL, NULL); if (error) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; goto bad; } if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) /* XXX should grow with balloc() */ panic("ext2_mkdir: blksize"); else { ip->i_size = DIRBLKSIZ; ip->i_flag |= IN_CHANGE; } /* Directory set up, now install its entry in the parent directory. */ error = ext2_direnter(ip, dvp, cnp); if (error) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; } bad: /* * No need to do an explicit VOP_TRUNCATE here, vrele will do this * for us because we set the link count to 0. */ if (error) { ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); } else *ap->a_vpp = tvp; out: return (error); #undef DIRBLKSIZ #define DIRBLKSIZ DEV_BSIZE }
/* * Truncate the inode oip to at most length size, freeing the * disk blocks. */ int ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred, struct thread *td) { struct vnode *ovp = vp; int32_t lastblock; struct inode *oip; int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; uint32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR]; struct m_ext2fs *fs; struct buf *bp; int offset, size, level; e4fs_daddr_t count, nblocks, blocksreleased = 0; int error, i, allerror; off_t osize; #ifdef INVARIANTS struct bufobj *bo; #endif oip = VTOI(ovp); #ifdef INVARIANTS bo = &ovp->v_bufobj; #endif ASSERT_VOP_LOCKED(vp, "ext2_truncate"); if (length < 0) return (EINVAL); if (ovp->v_type == VLNK && oip->i_size < ovp->v_mount->mnt_maxsymlinklen) { #ifdef INVARIANTS if (length != 0) panic("ext2_truncate: partial truncate of symlink"); #endif bzero((char *)&oip->i_shortlink, (u_int)oip->i_size); oip->i_size = 0; oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ext2_update(ovp, 1)); } if (oip->i_size == length) { oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ext2_update(ovp, 0)); } fs = oip->i_e2fs; osize = oip->i_size; /* * Lengthen the size of the file. We must ensure that the * last byte of the file is allocated. Since the smallest * value of osize is 0, length will be at least 1. */ if (osize < length) { if (length > oip->i_e2fs->e2fs_maxfilesize) return (EFBIG); vnode_pager_setsize(ovp, length); offset = blkoff(fs, length - 1); lbn = lblkno(fs, length - 1); flags |= BA_CLRBUF; error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, flags); if (error) { vnode_pager_setsize(vp, osize); return (error); } oip->i_size = length; if (bp->b_bufsize == fs->e2fs_bsize) bp->b_flags |= B_CLUSTEROK; if (flags & IO_SYNC) bwrite(bp); else if (DOINGASYNC(ovp)) bdwrite(bp); else bawrite(bp); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ext2_update(ovp, !DOINGASYNC(ovp))); } /* * Shorten the size of the file. If the file is not being * truncated to a block boundry, the contents of the * partial block following the end of the file must be * zero'ed in case it ever become accessible again because * of subsequent file growth. */ /* I don't understand the comment above */ offset = blkoff(fs, length); if (offset == 0) { oip->i_size = length; } else { lbn = lblkno(fs, length); flags |= BA_CLRBUF; error = ext2_balloc(oip, lbn, offset, cred, &bp, flags); if (error) return (error); oip->i_size = length; size = blksize(fs, oip, lbn); bzero((char *)bp->b_data + offset, (u_int)(size - offset)); allocbuf(bp, size); if (bp->b_bufsize == fs->e2fs_bsize) bp->b_flags |= B_CLUSTEROK; if (flags & IO_SYNC) bwrite(bp); else if (DOINGASYNC(ovp)) bdwrite(bp); else bawrite(bp); } /* * Calculate index into inode's block list of * last direct and indirect blocks (if any) * which we want to keep. Lastblock is -1 when * the file is truncated to 0. */ lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1; lastiblock[SINGLE] = lastblock - NDADDR; lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs); lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs); nblocks = btodb(fs->e2fs_bsize); /* * Update file and block pointers on disk before we start freeing * blocks. If we crash before free'ing blocks below, the blocks * will be returned to the free list. lastiblock values are also * normalized to -1 for calls to ext2_indirtrunc below. */ for (level = TRIPLE; level >= SINGLE; level--) { oldblks[NDADDR + level] = oip->i_ib[level]; if (lastiblock[level] < 0) { oip->i_ib[level] = 0; lastiblock[level] = -1; } } for (i = 0; i < NDADDR; i++) { oldblks[i] = oip->i_db[i]; if (i > lastblock) oip->i_db[i] = 0; } oip->i_flag |= IN_CHANGE | IN_UPDATE; allerror = ext2_update(ovp, !DOINGASYNC(ovp)); /* * Having written the new inode to disk, save its new configuration * and put back the old block pointers long enough to process them. * Note that we save the new block configuration so we can check it * when we are done. */ for (i = 0; i < NDADDR; i++) { newblks[i] = oip->i_db[i]; oip->i_db[i] = oldblks[i]; } for (i = 0; i < NIADDR; i++) { newblks[NDADDR + i] = oip->i_ib[i]; oip->i_ib[i] = oldblks[NDADDR + i]; } oip->i_size = osize; error = vtruncbuf(ovp, cred, length, (int)fs->e2fs_bsize); if (error && (allerror == 0)) allerror = error; vnode_pager_setsize(ovp, length); /* * Indirect blocks first. */ indir_lbn[SINGLE] = -NDADDR; indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1; indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1; for (level = TRIPLE; level >= SINGLE; level--) { bn = oip->i_ib[level]; if (bn != 0) { error = ext2_indirtrunc(oip, indir_lbn[level], fsbtodb(fs, bn), lastiblock[level], level, &count); if (error) allerror = error; blocksreleased += count; if (lastiblock[level] < 0) { oip->i_ib[level] = 0; ext2_blkfree(oip, bn, fs->e2fs_fsize); blocksreleased += nblocks; } } if (lastiblock[level] >= 0) goto done; } /* * All whole direct blocks or frags. */ for (i = NDADDR - 1; i > lastblock; i--) { long bsize; bn = oip->i_db[i]; if (bn == 0) continue; oip->i_db[i] = 0; bsize = blksize(fs, oip, i); ext2_blkfree(oip, bn, bsize); blocksreleased += btodb(bsize); } if (lastblock < 0) goto done; /* * Finally, look for a change in size of the * last direct block; release any frags. */ bn = oip->i_db[lastblock]; if (bn != 0) { long oldspace, newspace; /* * Calculate amount of space we're giving * back as old block size minus new block size. */ oldspace = blksize(fs, oip, lastblock); oip->i_size = length; newspace = blksize(fs, oip, lastblock); if (newspace == 0) panic("ext2_truncate: newspace"); if (oldspace - newspace > 0) { /* * Block number of space to be free'd is * the old block # plus the number of frags * required for the storage we're keeping. */ bn += numfrags(fs, newspace); ext2_blkfree(oip, bn, oldspace - newspace); blocksreleased += btodb(oldspace - newspace); } } done: #ifdef INVARIANTS for (level = SINGLE; level <= TRIPLE; level++) if (newblks[NDADDR + level] != oip->i_ib[level]) panic("itrunc1"); for (i = 0; i < NDADDR; i++) if (newblks[i] != oip->i_db[i]) panic("itrunc2"); BO_LOCK(bo); if (length == 0 && (bo->bo_dirty.bv_cnt != 0 || bo->bo_clean.bv_cnt != 0)) panic("itrunc3"); BO_UNLOCK(bo); #endif /* INVARIANTS */ /* * Put back the real size. */ oip->i_size = length; if (oip->i_blocks >= blocksreleased) oip->i_blocks -= blocksreleased; else /* sanity */ oip->i_blocks = 0; oip->i_flag |= IN_CHANGE; vnode_pager_setsize(ovp, length); return (allerror); }
/* * Vnode op for writing. */ static int ext2_write(struct vop_write_args *ap) { struct vnode *vp; struct uio *uio; struct inode *ip; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn; off_t osize; int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; ioflag = ap->a_ioflag; uio = ap->a_uio; vp = ap->a_vp; seqcount = ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE) panic("%s: mode", "ext2_write"); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = ip->i_size; if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) return (EPERM); /* FALLTHROUGH */ case VLNK: break; case VDIR: /* XXX differs from ffs -- this is called from ext2_mkdir(). */ if ((ioflag & IO_SYNC) == 0) panic("ext2_write: nonsync dir write"); break; default: panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, vp->v_type, (intmax_t)uio->uio_offset, (intmax_t)uio->uio_resid); } KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); fs = ip->i_e2fs; if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) return (EFBIG); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, I don't think it matters. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); resid = uio->uio_resid; osize = ip->i_size; if (seqcount > BA_SEQMAX) flags = BA_SEQMAX << BA_SEQSHIFT; else flags = seqcount << BA_SEQSHIFT; if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) flags |= IO_SYNC; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (uio->uio_offset + xfersize > ip->i_size) vnode_pager_setsize(vp, uio->uio_offset + xfersize); /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->e2fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; error = ext2_balloc(ip, lbn, blkoffset + xfersize, ap->a_cred, &bp, flags); if (error != 0) break; /* * If the buffer is not valid and we did not clear garbage * out above, we have to do so here even though the write * covers the entire buffer in order to avoid a mmap()/write * race where another process may see the garbage prior to * the uiomove() for a write replacing it. */ if ((bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize <= xfersize) vfs_bio_clrbuf(bp); if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) bp->b_flags |= B_NOCACHE; if (uio->uio_offset + xfersize > ip->i_size) ip->i_size = uio->uio_offset + xfersize; size = blksize(fs, ip, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); if (ioflag & (IO_VMIO|IO_DIRECT)) { bp->b_flags |= B_RELBUF; } /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || (ioflag & IO_ASYNC)) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else if (xfersize + blkoffset == fs->e2fs_fsize) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { bp->b_flags |= B_CLUSTEROK; cluster_write(vp, bp, ip->i_size, seqcount, 0); } else { bawrite(bp); } } else if (ioflag & IO_DIRECT) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else { bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } if (error || xfersize == 0) break; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ap->a_cred) { if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) ip->i_mode &= ~(ISUID | ISGID); } if (error) { if (ioflag & IO_UNIT) { (void)ext2_truncate(vp, osize, ioflag & IO_SYNC, ap->a_cred, uio->uio_td); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } if (uio->uio_resid != resid) { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (ioflag & IO_SYNC) error = ext2_update(vp, 1); } return (error); }
static int ext2_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn, int level, e4fs_daddr_t *countp) { struct buf *bp; struct m_ext2fs *fs = ip->i_e2fs; struct vnode *vp; e2fs_daddr_t *bap, *copy; int i, nblocks, error = 0, allerror = 0; e2fs_lbn_t nb, nlbn, last; e4fs_daddr_t blkcount, factor, blocksreleased = 0; /* * Calculate index in current block of last * block to be kept. -1 indicates the entire * block so we need not calculate the index. */ factor = 1; for (i = SINGLE; i < level; i++) factor *= NINDIR(fs); last = lastbn; if (lastbn > 0) last /= factor; nblocks = btodb(fs->e2fs_bsize); /* * Get buffer of block pointers, zero those entries corresponding * to blocks to be free'd, and update on disk copy first. Since * double(triple) indirect before single(double) indirect, calls * to bmap on these blocks will fail. However, we already have * the on disk address, so we have to set the b_blkno field * explicitly instead of letting bread do everything for us. */ vp = ITOV(ip); bp = getblk(vp, lbn, (int)fs->e2fs_bsize, 0, 0, 0); if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) { bp->b_iocmd = BIO_READ; if (bp->b_bcount > bp->b_bufsize) panic("ext2_indirtrunc: bad buffer size"); bp->b_blkno = dbn; vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); error = bufwait(bp); } if (error) { brelse(bp); *countp = 0; return (error); } bap = (e2fs_daddr_t *)bp->b_data; copy = malloc(fs->e2fs_bsize, M_TEMP, M_WAITOK); bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->e2fs_bsize); bzero((caddr_t)&bap[last + 1], (NINDIR(fs) - (last + 1)) * sizeof(e2fs_daddr_t)); if (last == -1) bp->b_flags |= B_INVAL; if (DOINGASYNC(vp)) { bdwrite(bp); } else { error = bwrite(bp); if (error) allerror = error; } bap = copy; /* * Recursively free totally unused blocks. */ for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; i--, nlbn += factor) { nb = bap[i]; if (nb == 0) continue; if (level > SINGLE) { if ((error = ext2_indirtrunc(ip, nlbn, fsbtodb(fs, nb), (int32_t)-1, level - 1, &blkcount)) != 0) allerror = error; blocksreleased += blkcount; } ext2_blkfree(ip, nb, fs->e2fs_bsize); blocksreleased += nblocks; } /* * Recursively free last partial block. */ if (level > SINGLE && lastbn >= 0) { last = lastbn % factor; nb = bap[i]; if (nb != 0) { if ((error = ext2_indirtrunc(ip, nlbn, fsbtodb(fs, nb), last, level - 1, &blkcount)) != 0) allerror = error; blocksreleased += blkcount; } } free(copy, M_TEMP); *countp = blocksreleased; return (allerror); }
/* * Rename system call. * rename("foo", "bar"); * is essentially * unlink("bar"); * link("foo", "bar"); * unlink("foo"); * but ``atomically''. Can't do full commit without saving state in the * inode on disk which isn't feasible at this time. Best we can do is * always guarantee the target exists. * * Basic algorithm is: * * 1) Bump link count on source while we're linking it to the * target. This also ensure the inode won't be deleted out * from underneath us while we work (it may be truncated by * a concurrent `trunc' or `open' for creation). * 2) Link source to destination. If destination already exists, * delete it first. * 3) Unlink source reference to inode if still around. If a * directory was moved and the parent of the destination * is different from the source, patch the ".." entry in the * directory. */ static int ext2_rename(struct vop_rename_args *ap) { struct vnode *tvp = ap->a_tvp; struct vnode *tdvp = ap->a_tdvp; struct vnode *fvp = ap->a_fvp; struct vnode *fdvp = ap->a_fdvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct inode *ip, *xp, *dp; struct dirtemplate dirbuf; int doingdirectory = 0, oldparent = 0, newparent = 0; int error = 0; u_char namlen; #ifdef DIAGNOSTIC if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ext2_rename: no name"); #endif /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; abortit: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); return (error); } if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto abortit; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. Temporarily just warn if they do. */ if (fvp == tvp) { printf("ext2_rename: fvp == tvp (can't happen)\n"); error = 0; goto abortit; } if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) goto abortit; dp = VTOI(fdvp); ip = VTOI(fvp); if (ip->i_nlink >= EXT2_LINK_MAX) { VOP_UNLOCK(fvp, 0); error = EMLINK; goto abortit; } if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) { VOP_UNLOCK(fvp, 0); error = EPERM; goto abortit; } if ((ip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || (ip->i_flag & IN_RENAME)) { VOP_UNLOCK(fvp, 0); error = EINVAL; goto abortit; } ip->i_flag |= IN_RENAME; oldparent = dp->i_number; doingdirectory++; } vrele(fdvp); /* * When the target exists, both the directory * and target vnodes are returned locked. */ dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); /* * 1) Bump link count while we're moving stuff * around. If we crash somewhere before * completing our work, the link count * may be wrong, but correctable. */ ip->i_nlink++; ip->i_flag |= IN_CHANGE; if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { VOP_UNLOCK(fvp, 0); goto bad; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". We must repeat the call * to namei, as the parent directory is unlocked by the * call to checkpath(). */ error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); VOP_UNLOCK(fvp, 0); if (oldparent != dp->i_number) newparent = dp->i_number; if (doingdirectory && newparent) { if (error) /* write access check above */ goto bad; if (xp != NULL) vput(tvp); error = ext2_checkpath(ip, dp, tcnp->cn_cred); if (error) goto out; VREF(tdvp); error = relookup(tdvp, &tvp, tcnp); if (error) goto out; vrele(tdvp); dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); } /* * 2) If target doesn't exist, link the target * to the source and unlink the source. * Otherwise, rewrite the target directory * entry to reference the source inode and * expunge the original entry's existence. */ if (xp == NULL) { if (dp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Account for ".." in new directory. * When source and destination have the same * parent we don't fool with the link count. */ if (doingdirectory && newparent) { if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto bad; } dp->i_nlink++; dp->i_flag |= IN_CHANGE; error = ext2_update(tdvp, !DOINGASYNC(tdvp)); if (error) goto bad; } error = ext2_direnter(ip, tdvp, tcnp); if (error) { if (doingdirectory && newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; (void)ext2_update(tdvp, 1); } goto bad; } vput(tdvp); } else { if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (xp->i_number == ip->i_number) panic("ext2_rename: same file"); /* * If the parent directory is "sticky", then the user must * own the parent directory, or the destination of the rename, * otherwise the destination may not be changed (except by * root). This implements append-only directories. */ if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && tcnp->cn_cred->cr_uid != dp->i_uid && xp->i_uid != tcnp->cn_cred->cr_uid) { error = EPERM; goto bad; } /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((xp->i_mode&IFMT) == IFDIR) { if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) || xp->i_nlink > 2) { error = ENOTEMPTY; goto bad; } if (!doingdirectory) { error = ENOTDIR; goto bad; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto bad; } error = ext2_dirrewrite(dp, ip, tcnp); if (error) goto bad; /* * If the target directory is in the same * directory as the source directory, * decrement the link count on the parent * of the target directory. */ if (doingdirectory && !newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; } vput(tdvp); /* * Adjust the link count of the target to * reflect the dirrewrite above. If this is * a directory it is empty and there are * no links to it, so we can squash the inode and * any space associated with it. We disallowed * renaming over top of a directory with links to * it above, as the remaining link would point to * a directory without "." or ".." entries. */ xp->i_nlink--; if (doingdirectory) { if (--xp->i_nlink != 0) panic("ext2_rename: linked directory"); error = ext2_truncate(tvp, (off_t)0, IO_SYNC, tcnp->cn_cred, tcnp->cn_thread); } xp->i_flag |= IN_CHANGE; vput(tvp); xp = NULL; } /* * 3) Unlink the source. */ fcnp->cn_flags &= ~MODMASK; fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; VREF(fdvp); error = relookup(fdvp, &fvp, fcnp); if (error == 0) vrele(fdvp); if (fvp != NULL) { xp = VTOI(fvp); dp = VTOI(fdvp); } else { /* * From name has disappeared. */ if (doingdirectory) panic("ext2_rename: lost dir entry"); vrele(ap->a_fvp); return (0); } /* * Ensure that the directory entry still exists and has not * changed while the new name has been entered. If the source is * a file then the entry may have been unlinked or renamed. In * either case there is no further work to be done. If the source * is a directory then it cannot have been rmdir'ed; its link * count of three would cause a rmdir to fail with ENOTEMPTY. * The IN_RENAME flag ensures that it cannot be moved by another * rename. */ if (xp != ip) { if (doingdirectory) panic("ext2_rename: lost dir entry"); } else { /* * If the source is a directory with a * new parent, the link count of the old * parent directory must be decremented * and ".." set to point to the new parent. */ if (doingdirectory && newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf, sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); if (error == 0) { /* Like ufs little-endian: */ namlen = dirbuf.dotdot_type; if (namlen != 2 || dirbuf.dotdot_name[0] != '.' || dirbuf.dotdot_name[1] != '.') { ext2_dirbad(xp, (doff_t)12, "rename: mangled dir"); } else { dirbuf.dotdot_ino = newparent; (void) vn_rdwr(UIO_WRITE, fvp, (caddr_t)&dirbuf, sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); cache_purge(fdvp); } } } error = ext2_dirremove(fdvp, fcnp); if (!error) { xp->i_nlink--; xp->i_flag |= IN_CHANGE; } xp->i_flag &= ~IN_RENAME; } if (dp) vput(fdvp); if (xp) vput(fvp); vrele(ap->a_fvp); return (error); bad: if (xp) vput(ITOV(xp)); vput(ITOV(dp)); out: if (doingdirectory) ip->i_flag &= ~IN_RENAME; if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; ip->i_flag &= ~IN_RENAME; vput(fvp); } else vrele(fvp); return (error); }
/* * Release blocks associated with the inode ip and stored in the indirect * block bn. Blocks are free'd in LIFO order up to (but not including) * lastbn. If level is greater than SINGLE, the block is an indirect block * and recursive calls to indirtrunc must be used to cleanse other indirect * blocks. * * NB: triple indirect blocks are untested. */ int ffs_indirtrunc(struct inode *ip, daddr64_t lbn, daddr64_t dbn, daddr64_t lastbn, int level, long *countp) { int i; struct buf *bp; struct fs *fs = ip->i_fs; struct vnode *vp; void *copy = NULL; daddr64_t nb, nlbn, last; long blkcount, factor; int nblocks, blocksreleased = 0; int error = 0, allerror = 0; int32_t *bap1 = NULL; #ifdef FFS2 int64_t *bap2 = NULL; #endif /* * Calculate index in current block of last * block to be kept. -1 indicates the entire * block so we need not calculate the index. */ factor = 1; for (i = SINGLE; i < level; i++) factor *= NINDIR(fs); last = lastbn; if (lastbn > 0) last /= factor; nblocks = btodb(fs->fs_bsize); /* * Get buffer of block pointers, zero those entries corresponding * to blocks to be free'd, and update on disk copy first. Since * double(triple) indirect before single(double) indirect, calls * to bmap on these blocks will fail. However, we already have * the on disk address, so we have to set the b_blkno field * explicitly instead of letting bread do everything for us. */ vp = ITOV(ip); bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0); if (!(bp->b_flags & (B_DONE | B_DELWRI))) { curproc->p_ru.ru_inblock++; /* pay for read */ bcstats.pendingreads++; bcstats.numreads++; bp->b_flags |= B_READ; if (bp->b_bcount > bp->b_bufsize) panic("ffs_indirtrunc: bad buffer size"); bp->b_blkno = dbn; VOP_STRATEGY(bp); error = biowait(bp); } if (error) { brelse(bp); *countp = 0; return (error); } #ifdef FFS2 if (ip->i_ump->um_fstype == UM_UFS2) bap2 = (int64_t *)bp->b_data; else #endif bap1 = (int32_t *)bp->b_data; if (lastbn != -1) { copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK); bcopy(bp->b_data, copy, (u_int) fs->fs_bsize); for (i = last + 1; i < NINDIR(fs); i++) BAP_ASSIGN(ip, i, 0); if (!DOINGASYNC(vp)) { error = bwrite(bp); if (error) allerror = error; } else { bawrite(bp); } #ifdef FFS2 if (ip->i_ump->um_fstype == UM_UFS2) bap2 = (int64_t *)copy; else #endif bap1 = (int32_t *)copy; } /* * Recursively free totally unused blocks. */ for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; i--, nlbn += factor) { nb = BAP(ip, i); if (nb == 0) continue; if (level > SINGLE) { error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), (daddr64_t)-1, level - 1, &blkcount); if (error) allerror = error; blocksreleased += blkcount; } ffs_blkfree(ip, nb, fs->fs_bsize); blocksreleased += nblocks; } /* * Recursively free last partial block. */ if (level > SINGLE && lastbn >= 0) { last = lastbn % factor; nb = BAP(ip, i); if (nb != 0) { error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb), last, level - 1, &blkcount); if (error) allerror = error; blocksreleased += blkcount; } } if (copy != NULL) { free(copy, M_TEMP); } else { bp->b_flags |= B_INVAL; brelse(bp); } *countp = blocksreleased; return (allerror); }
/* * Remove a directory entry after a call to namei, using * the parameters which it left in nameidata. The entry * dp->i_offset contains the offset into the directory of the * entry to be eliminated. The dp->i_count field contains the * size of the previous record in the directory. If this * is 0, the first entry is being deleted, so we need only * zero the inode number to mark the entry as free. If the * entry is not the first in the directory, we must reclaim * the space of the now empty record by adding the record size * to the size of the previous entry. */ int ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir) { struct inode *dp; struct direct *ep; struct buf *bp; int error; UFS_WAPBL_JLOCK_ASSERT(dvp->v_mount); dp = VTOI(dvp); if ((error = UFS_BUFATOFF(dp, (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0) return (error); #ifdef UFS_DIRHASH /* * Remove the dirhash entry. This is complicated by the fact * that `ep' is the previous entry when dp->i_count != 0. */ if (dp->i_dirhash != NULL) ufsdirhash_remove(dp, (dp->i_count == 0) ? ep : (struct direct *)((char *)ep + ep->d_reclen), dp->i_offset); #endif if (dp->i_count == 0) { /* * First entry in block: set d_ino to zero. */ ep->d_ino = 0; } else { /* * Collapse new free space into previous entry. */ ep->d_reclen += dp->i_reclen; } #ifdef UFS_DIRHASH if (dp->i_dirhash != NULL) ufsdirhash_checkblock(dp, (char *)ep - ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)), dp->i_offset & ~(DIRBLKSIZ - 1)); #endif if (DOINGSOFTDEP(dvp)) { if (ip) { ip->i_effnlink--; softdep_change_linkcnt(ip, 0); softdep_setup_remove(bp, dp, ip, isrmdir); } if (softdep_slowdown(dvp)) { error = bwrite(bp); } else { bdwrite(bp); error = 0; } } else { if (ip) { ip->i_effnlink--; DIP_ADD(ip, nlink, -1); ip->i_flag |= IN_CHANGE; UFS_WAPBL_UPDATE(ip, 0); } if (DOINGASYNC(dvp) && dp->i_count != 0) { bdwrite(bp); error = 0; } else error = bwrite(bp); } dp->i_flag |= IN_CHANGE | IN_UPDATE; UFS_WAPBL_UPDATE(dp, 0); return (error); }
/* * Create an HTree index for a directory */ int ext2_htree_create_index(struct vnode *vp, struct componentname *cnp, struct ext2fs_direct_2 *new_entry) { struct buf *bp = NULL; struct inode *dp; struct ext2fs *fs; struct m_ext2fs *m_fs; struct ext2fs_direct_2 *ep, *dotdot; struct ext2fs_htree_root *root; struct ext2fs_htree_lookup_info info; uint32_t blksize, dirlen, split_hash; uint8_t hash_version; char *buf1 = NULL; char *buf2 = NULL; int error = 0; dp = VTOI(vp); fs = dp->i_e2fs->e2fs; m_fs = dp->i_e2fs; blksize = m_fs->e2fs_bsize; buf1 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO); buf2 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO); if ((error = ext2_blkatoff(vp, 0, NULL, &bp)) != 0) goto out; root = (struct ext2fs_htree_root *)bp->b_data; dotdot = (struct ext2fs_direct_2 *)((char *)&(root->h_dotdot)); ep = (struct ext2fs_direct_2 *)((char *)dotdot + dotdot->e2d_reclen); dirlen = (char *)root + blksize - (char *)ep; memcpy(buf1, ep, dirlen); ep = (struct ext2fs_direct_2 *)buf1; while ((char *)ep < buf1 + dirlen) ep = (struct ext2fs_direct_2 *) ((char *)ep + ep->e2d_reclen); ep->e2d_reclen = buf1 + blksize - (char *)ep; dp->i_flag |= IN_E4INDEX; /* * Initialize index root. */ dotdot->e2d_reclen = blksize - EXT2_DIR_REC_LEN(1); memset(&root->h_info, 0, sizeof(root->h_info)); root->h_info.h_hash_version = fs->e3fs_def_hash_version; root->h_info.h_info_len = sizeof(root->h_info); ext2_htree_set_block(root->h_entries, 1); ext2_htree_set_count(root->h_entries, 1); ext2_htree_set_limit(root->h_entries, ext2_htree_root_limit(dp, sizeof(root->h_info))); memset(&info, 0, sizeof(info)); info.h_levels_num = 1; info.h_levels[0].h_entries = root->h_entries; info.h_levels[0].h_entry = root->h_entries; hash_version = root->h_info.h_hash_version; if (hash_version <= EXT2_HTREE_TEA) hash_version += m_fs->e2fs_uhash; ext2_htree_split_dirblock(buf1, buf2, blksize, fs->e3fs_hash_seed, hash_version, &split_hash, new_entry); ext2_htree_insert_entry(&info, split_hash, 2); /* * Write directory block 0. */ if (DOINGASYNC(vp)) { bdwrite(bp); error = 0; } else { error = bwrite(bp); } dp->i_flag |= IN_CHANGE | IN_UPDATE; if (error) goto out; /* * Write directory block 1. */ error = ext2_htree_append_block(vp, buf1, cnp, blksize); if (error) goto out1; /* * Write directory block 2. */ error = ext2_htree_append_block(vp, buf2, cnp, blksize); free(buf1, M_TEMP); free(buf2, M_TEMP); return (error); out: if (bp != NULL) brelse(bp); out1: free(buf1, M_TEMP); free(buf2, M_TEMP); return (error); }