/* * Truncate the inode oip to at most length size, freeing the * disk blocks. */ int ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred) { struct vnode *ovp = vp; ufs_daddr_t lastblock; struct inode *oip; ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR]; struct fs *fs; struct buf *bp; int offset, size, level; long count, nblocks, blocksreleased = 0; int i; int aflags, error, allerror; off_t osize; oip = VTOI(ovp); fs = oip->i_fs; if (length < 0) return (EINVAL); if (length > fs->fs_maxfilesize) return (EFBIG); if (ovp->v_type == VLNK && (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) { #ifdef DIAGNOSTIC if (length != 0) panic("ffs_truncate: partial truncate of symlink"); #endif /* DIAGNOSTIC */ bzero((char *)&oip->i_shortlink, (uint)oip->i_size); oip->i_size = 0; oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, 1)); } if (oip->i_size == length) { oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, 0)); } if (fs->fs_ronly) panic("ffs_truncate: read-only filesystem"); #ifdef QUOTA error = ufs_getinoquota(oip); if (error) return (error); #endif ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0; if (DOINGSOFTDEP(ovp)) { if (length > 0 || softdep_slowdown(ovp)) { /* * If a file is only partially truncated, then * we have to clean up the data structures * describing the allocation past the truncation * point. Finding and deallocating those structures * is a lot of work. Since partial truncation occurs * rarely, we solve the problem by syncing the file * so that it will have no data structures left. */ if ((error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0) return (error); } else { #ifdef QUOTA (void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0); #endif softdep_setup_freeblocks(oip, length); vinvalbuf(ovp, 0, 0, 0); nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, 0)); } } 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. * * nvextendbuf() only breads the old buffer. The blocksize * of the new buffer must be specified so it knows how large * to make the VM object. */ if (osize < length) { nvextendbuf(vp, osize, length, blkoffsize(fs, oip, osize), /* oblksize */ blkoffresize(fs, length), /* nblksize */ blkoff(fs, osize), blkoff(fs, length), 0); aflags = B_CLRBUF; if (flags & IO_SYNC) aflags |= B_SYNC; /* BALLOC will reallocate the fragment at the old EOF */ error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp); if (error) return (error); oip->i_size = length; if (bp->b_bufsize == fs->fs_bsize) bp->b_flags |= B_CLUSTEROK; if (aflags & B_SYNC) bwrite(bp); else bawrite(bp); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, 1)); } /* * Shorten the size of the file. * * NOTE: The block size specified in nvtruncbuf() is the blocksize * of the buffer containing length prior to any reallocation * of the block. */ allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length), blkoff(fs, length), 0); offset = blkoff(fs, length); if (offset == 0) { oip->i_size = length; } else { lbn = lblkno(fs, length); aflags = B_CLRBUF; if (flags & IO_SYNC) aflags |= B_SYNC; error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &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. * * nvtruncbuf() may have re-dirtied the underlying block * as part of its truncation zeroing code. To avoid a * 'locking against myself' panic in the second fsync we * can simply undirty the bp since the redirtying was * related to areas of the buffer that we are going to * throw away anyway, and we will b*write() the remainder * anyway down below. */ if (DOINGSOFTDEP(ovp) && lbn < NDADDR && fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) { bundirty(bp); error = VOP_FSYNC(ovp, MNT_WAIT, 0); if (error) { bdwrite(bp); return (error); } } oip->i_size = length; size = blksize(fs, oip, lbn); #if 0 /* remove - nvtruncbuf deals with this */ if (ovp->v_type != VDIR) bzero((char *)bp->b_data + offset, (uint)(size - offset)); #endif /* 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 (aflags & B_SYNC) bwrite(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 - 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. */ bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks); for (level = TRIPLE; level >= SINGLE; level--) if (lastiblock[level] < 0) { oip->i_ib[level] = 0; lastiblock[level] = -1; } for (i = NDADDR - 1; i > lastblock; i--) oip->i_db[i] = 0; oip->i_flag |= IN_CHANGE | IN_UPDATE; error = ffs_update(ovp, 1); if (error && allerror == 0) allerror = error; /* * 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. */ bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks); bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks); oip->i_size = osize; if (error && allerror == 0) allerror = error; /* * 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 = ffs_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; ffs_blkfree(oip, bn, fs->fs_bsize); 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); ffs_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("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(oip, bn, oldspace - newspace); blocksreleased += btodb(oldspace - newspace); } } done: #ifdef DIAGNOSTIC for (level = SINGLE; level <= TRIPLE; level++) if (newblks[NDADDR + level] != oip->i_ib[level]) panic("ffs_truncate1"); for (i = 0; i < NDADDR; i++) if (newblks[i] != oip->i_db[i]) panic("ffs_truncate2"); if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree)) panic("ffs_truncate3"); #endif /* DIAGNOSTIC */ /* * Put back the real size. */ oip->i_size = length; oip->i_blocks -= blocksreleased; if (oip->i_blocks < 0) /* sanity */ oip->i_blocks = 0; oip->i_flag |= IN_CHANGE; #ifdef QUOTA (void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0); #endif return (allerror); }
/* * 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); }
/* * Truncate the inode oip to at most length size, freeing the * disk blocks. */ int ffs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred) { struct vnode *ovp; daddr64_t lastblock; daddr64_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR]; daddr64_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR]; struct fs *fs; struct buf *bp; int offset, size, level; long count, nblocks, vflags, blocksreleased = 0; int i, aflags, error, allerror, indirect = 0; off_t osize; extern int num_indirdep; extern int max_indirdep; if (length < 0) return (EINVAL); ovp = ITOV(oip); if (ovp->v_type != VREG && ovp->v_type != VDIR && ovp->v_type != VLNK) return (0); if (DIP(oip, size) == length) return (0); if (ovp->v_type == VLNK && (DIP(oip, size) < ovp->v_mount->mnt_maxsymlinklen || (ovp->v_mount->mnt_maxsymlinklen == 0 && oip->i_din1->di_blocks == 0))) { #ifdef DIAGNOSTIC if (length != 0) panic("ffs_truncate: partial truncate of symlink"); #endif memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size)); DIP_ASSIGN(oip, size, 0); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (UFS_UPDATE(oip, MNT_WAIT)); } if ((error = getinoquota(oip)) != 0) return (error); uvm_vnp_setsize(ovp, length); oip->i_ci.ci_lasta = oip->i_ci.ci_clen = oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0; if (DOINGSOFTDEP(ovp)) { if (length > 0 || softdep_slowdown(ovp)) { /* * If a file is only partially truncated, then * we have to clean up the data structures * describing the allocation past the truncation * point. Finding and deallocating those structures * is a lot of work. Since partial truncation occurs * rarely, we solve the problem by syncing the file * so that it will have no data structures left. */ if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0) return (error); } else { (void)ufs_quota_free_blocks(oip, DIP(oip, blocks), NOCRED); softdep_setup_freeblocks(oip, length); (void) vinvalbuf(ovp, 0, cred, curproc, 0, 0); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (UFS_UPDATE(oip, 0)); } } fs = oip->i_fs; osize = DIP(oip, 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 > fs->fs_maxfilesize) return (EFBIG); aflags = B_CLRBUF; if (flags & IO_SYNC) aflags |= B_SYNC; error = UFS_BUF_ALLOC(oip, length - 1, 1, cred, aflags, &bp); if (error) return (error); if (bp->b_lblkno >= NDADDR) indirect = 1; DIP_ASSIGN(oip, size, length); uvm_vnp_setsize(ovp, length); (void) uvm_vnp_uncache(ovp); if (aflags & B_SYNC) bwrite(bp); else bawrite(bp); oip->i_flag |= IN_CHANGE | IN_UPDATE; error = UFS_UPDATE(oip, MNT_WAIT); if (DOINGSOFTDEP(ovp) && num_indirdep > max_indirdep) if (indirect) { /* * If the number of pending indirect block * dependencies is sufficiently close to the * maximum number of simultaneously mappable * buffers force a sync on the vnode to prevent * buffer cache exhaustion. */ VOP_FSYNC(ovp, curproc->p_ucred, MNT_WAIT); } return (error); } uvm_vnp_setsize(ovp, length); /* * Shorten the size of the file. 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 (offset == 0) { DIP_ASSIGN(oip, size, length); } else { lbn = lblkno(fs, length); aflags = B_CLRBUF; if (flags & IO_SYNC) aflags |= B_SYNC; error = UFS_BUF_ALLOC(oip, length - 1, 1, cred, aflags, &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(ovp) && lbn < NDADDR && fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize && (error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0) return (error); DIP_ASSIGN(oip, size, length); size = blksize(fs, oip, lbn); (void) uvm_vnp_uncache(ovp); if (ovp->v_type != VDIR) bzero((char *)bp->b_data + offset, (u_int)(size - offset)); bp->b_bcount = size; if (aflags & B_SYNC) bwrite(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 - 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[NDADDR + level] = DIP(oip, ib[level]); if (lastiblock[level] < 0) { DIP_ASSIGN(oip, ib[level], 0); lastiblock[level] = -1; } } for (i = 0; i < NDADDR; i++) { oldblks[i] = DIP(oip, db[i]); if (i > lastblock) DIP_ASSIGN(oip, db[i], 0); } oip->i_flag |= IN_CHANGE | IN_UPDATE; if ((error = UFS_UPDATE(oip, MNT_WAIT)) != 0) allerror = error; /* * 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] = DIP(oip, db[i]); DIP_ASSIGN(oip, db[i], oldblks[i]); } for (i = 0; i < NIADDR; i++) { newblks[NDADDR + i] = DIP(oip, ib[i]); DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]); } DIP_ASSIGN(oip, size, osize); vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA; allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0); /* * 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 = DIP(oip, ib[level]); if (bn != 0) { error = ffs_indirtrunc(oip, indir_lbn[level], fsbtodb(fs, bn), lastiblock[level], level, &count); if (error) allerror = error; blocksreleased += count; if (lastiblock[level] < 0) { DIP_ASSIGN(oip, ib[level], 0); ffs_blkfree(oip, bn, fs->fs_bsize); blocksreleased += nblocks; } } if (lastiblock[level] >= 0) goto done; } /* * All whole direct blocks or frags. */ for (i = NDADDR - 1; i > lastblock; i--) { long bsize; bn = DIP(oip, db[i]); if (bn == 0) continue; DIP_ASSIGN(oip, db[i], 0); bsize = blksize(fs, oip, i); ffs_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 = DIP(oip, 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); DIP_ASSIGN(oip, size, length); newspace = blksize(fs, oip, 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(oip, bn, oldspace - newspace); blocksreleased += btodb(oldspace - newspace); } } done: #ifdef DIAGNOSTIC for (level = SINGLE; level <= TRIPLE; level++) if (newblks[NDADDR + level] != DIP(oip, ib[level])) panic("ffs_truncate1"); for (i = 0; i < NDADDR; i++) if (newblks[i] != DIP(oip, db[i])) panic("ffs_truncate2"); #endif /* DIAGNOSTIC */ /* * Put back the real size. */ DIP_ASSIGN(oip, size, length); DIP_ADD(oip, blocks, -blocksreleased); oip->i_flag |= IN_CHANGE; (void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED); return (allerror); }
/* * 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); }
/* * Truncate the inode oip to at most length size, freeing the * disk blocks. */ int ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred) { daddr_t lastblock; struct inode *oip = VTOI(ovp); daddr_t bn, lastiblock[UFS_NIADDR], indir_lbn[UFS_NIADDR]; daddr_t blks[UFS_NDADDR + UFS_NIADDR]; struct fs *fs; int offset, pgoffset, level; int64_t count, blocksreleased = 0; int i, aflag, nblocks; int error, allerror = 0; off_t osize; int sync; struct ufsmount *ump = oip->i_ump; if (ovp->v_type == VCHR || ovp->v_type == VBLK || ovp->v_type == VFIFO || ovp->v_type == VSOCK) { KASSERT(oip->i_size == 0); return 0; } if (length < 0) return (EINVAL); if (ovp->v_type == VLNK && (oip->i_size < ump->um_maxsymlinklen || (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) { KDASSERT(length == 0); memset(SHORTLINK(oip), 0, (size_t)oip->i_size); oip->i_size = 0; DIP_ASSIGN(oip, size, 0); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, NULL, NULL, 0)); } if (oip->i_size == length) { /* still do a uvm_vnp_setsize() as writesize may be larger */ uvm_vnp_setsize(ovp, length); oip->i_flag |= IN_CHANGE | IN_UPDATE; return (ffs_update(ovp, NULL, NULL, 0)); } fs = oip->i_fs; if (length > ump->um_maxfilesize) return (EFBIG); if ((oip->i_flags & SF_SNAPSHOT) != 0) ffs_snapremove(ovp); osize = oip->i_size; aflag = ioflag & IO_SYNC ? B_SYNC : 0; /* * 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 (ffs_lblkno(fs, osize) < UFS_NDADDR && ffs_lblkno(fs, osize) != ffs_lblkno(fs, length) && ffs_blkroundup(fs, osize) != osize) { off_t eob; eob = ffs_blkroundup(fs, osize); uvm_vnp_setwritesize(ovp, eob); error = ufs_balloc_range(ovp, osize, eob - osize, cred, aflag); if (error) { (void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred); return error; } if (ioflag & IO_SYNC) { mutex_enter(ovp->v_interlock); VOP_PUTPAGES(ovp, trunc_page(osize & fs->fs_bmask), round_page(eob), PGO_CLEANIT | PGO_SYNCIO | PGO_JOURNALLOCKED); } } uvm_vnp_setwritesize(ovp, length); error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag); if (error) { (void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred); return (error); } uvm_vnp_setsize(ovp, length); oip->i_flag |= IN_CHANGE | IN_UPDATE; KASSERT(ovp->v_size == oip->i_size); return (ffs_update(ovp, NULL, NULL, 0)); } /* * When truncating a regular file down to a non-block-aligned size, * we must zero the part of last block which is past the new EOF. * We must synchronously flush the zeroed pages to disk * since the new pages will be invalidated as soon as we * inform the VM system of the new, smaller size. * We must do this before acquiring the GLOCK, since fetching * the pages will acquire the GLOCK internally. * So there is a window where another thread could see a whole * zeroed page past EOF, but that's life. */ offset = ffs_blkoff(fs, length); pgoffset = length & PAGE_MASK; if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) && osize > length) { daddr_t lbn; voff_t eoz; int size; if (offset != 0) { error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag); if (error) return error; } lbn = ffs_lblkno(fs, length); size = ffs_blksize(fs, oip, lbn); eoz = MIN(MAX(ffs_lblktosize(fs, lbn) + size, round_page(pgoffset)), osize); ubc_zerorange(&ovp->v_uobj, length, eoz - length, UBC_UNMAP_FLAG(ovp)); if (round_page(eoz) > round_page(length)) { mutex_enter(ovp->v_interlock); error = VOP_PUTPAGES(ovp, round_page(length), round_page(eoz), PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED | ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0)); if (error) return error; } } genfs_node_wrlock(ovp); oip->i_size = length; DIP_ASSIGN(oip, size, length); uvm_vnp_setsize(ovp, length); /* * 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 = ffs_lblkno(fs, length + fs->fs_bsize - 1) - 1; lastiblock[SINGLE] = lastblock - UFS_NDADDR; lastiblock[DOUBLE] = lastiblock[SINGLE] - FFS_NINDIR(fs); lastiblock[TRIPLE] = lastiblock[DOUBLE] - FFS_NINDIR(fs) * FFS_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. */ sync = 0; for (level = TRIPLE; level >= SINGLE; level--) { blks[UFS_NDADDR + level] = DIP(oip, ib[level]); if (lastiblock[level] < 0 && blks[UFS_NDADDR + level] != 0) { sync = 1; DIP_ASSIGN(oip, ib[level], 0); lastiblock[level] = -1; } } for (i = 0; i < UFS_NDADDR; i++) { blks[i] = DIP(oip, db[i]); if (i > lastblock && blks[i] != 0) { sync = 1; DIP_ASSIGN(oip, db[i], 0); } } oip->i_flag |= IN_CHANGE | IN_UPDATE; if (sync) { error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT); if (error && !allerror) allerror = error; } /* * 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++) { bn = DIP(oip, db[i]); DIP_ASSIGN(oip, db[i], blks[i]); blks[i] = bn; } for (i = 0; i < UFS_NIADDR; i++) { bn = DIP(oip, ib[i]); DIP_ASSIGN(oip, ib[i], blks[UFS_NDADDR + i]); blks[UFS_NDADDR + i] = bn; } oip->i_size = osize; DIP_ASSIGN(oip, size, osize); error = vtruncbuf(ovp, lastblock + 1, 0, 0); if (error && !allerror) allerror = error; /* * Indirect blocks first. */ indir_lbn[SINGLE] = -UFS_NDADDR; indir_lbn[DOUBLE] = indir_lbn[SINGLE] - FFS_NINDIR(fs) - 1; indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - FFS_NINDIR(fs) * FFS_NINDIR(fs) - 1; for (level = TRIPLE; level >= SINGLE; level--) { if (oip->i_ump->um_fstype == UFS1) bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs)); else bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs)); if (bn != 0) { error = ffs_indirtrunc(oip, indir_lbn[level], FFS_FSBTODB(fs, bn), lastiblock[level], level, &count); if (error) allerror = error; blocksreleased += count; if (lastiblock[level] < 0) { DIP_ASSIGN(oip, ib[level], 0); if (oip->i_ump->um_mountp->mnt_wapbl) { UFS_WAPBL_REGISTER_DEALLOCATION( oip->i_ump->um_mountp, FFS_FSBTODB(fs, bn), fs->fs_bsize); } else ffs_blkfree(fs, oip->i_devvp, bn, fs->fs_bsize, oip->i_number); blocksreleased += nblocks; } } if (lastiblock[level] >= 0) goto done; } /* * All whole direct blocks or frags. */ for (i = UFS_NDADDR - 1; i > lastblock; i--) { long bsize; if (oip->i_ump->um_fstype == UFS1) bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs)); else bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs)); if (bn == 0) continue; DIP_ASSIGN(oip, db[i], 0); bsize = ffs_blksize(fs, oip, i); if ((oip->i_ump->um_mountp->mnt_wapbl) && (ovp->v_type != VREG)) { UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp, FFS_FSBTODB(fs, bn), bsize); } else ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number); blocksreleased += btodb(bsize); } if (lastblock < 0) goto done; /* * Finally, look for a change in size of the * last direct block; release any frags. */ if (oip->i_ump->um_fstype == UFS1) bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs)); else bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs)); if (bn != 0) { long oldspace, newspace; /* * Calculate amount of space we're giving * back as old block size minus new block size. */ oldspace = ffs_blksize(fs, oip, lastblock); oip->i_size = length; DIP_ASSIGN(oip, size, length); newspace = ffs_blksize(fs, oip, lastblock); if (newspace == 0) panic("itrunc: 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 += ffs_numfrags(fs, newspace); if ((oip->i_ump->um_mountp->mnt_wapbl) && (ovp->v_type != VREG)) { UFS_WAPBL_REGISTER_DEALLOCATION( oip->i_ump->um_mountp, FFS_FSBTODB(fs, bn), oldspace - newspace); } else ffs_blkfree(fs, oip->i_devvp, bn, oldspace - newspace, oip->i_number); blocksreleased += btodb(oldspace - newspace); } } done: #ifdef DIAGNOSTIC for (level = SINGLE; level <= TRIPLE; level++) if (blks[UFS_NDADDR + level] != DIP(oip, ib[level])) panic("itrunc1"); for (i = 0; i < UFS_NDADDR; i++) if (blks[i] != DIP(oip, db[i])) panic("itrunc2"); if (length == 0 && (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd))) panic("itrunc3"); #endif /* DIAGNOSTIC */ /* * Put back the real size. */ oip->i_size = length; DIP_ASSIGN(oip, size, length); DIP_ADD(oip, blocks, -blocksreleased); genfs_node_unlock(ovp); oip->i_flag |= IN_CHANGE; UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0); #if defined(QUOTA) || defined(QUOTA2) (void) chkdq(oip, -blocksreleased, NOCRED, 0); #endif KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size); return (allerror); }
/* * 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, daddr_t lbn, daddr_t dbn, daddr_t lastbn, int level, int64_t *countp) { int i; struct buf *bp; struct fs *fs = ip->i_fs; int32_t *bap1 = NULL; int64_t *bap2 = NULL; struct vnode *vp; daddr_t nb, nlbn, last; char *copy = NULL; int64_t blkcount, factor, blocksreleased = 0; int nblocks; int error = 0, allerror = 0; const int needswap = UFS_FSNEEDSWAP(fs); #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \ ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap)) #define BAP_ASSIGN(ip, i, value) \ do { \ if ((ip)->i_ump->um_fstype == UFS1) \ bap1[i] = (value); \ else \ bap2[i] = (value); \ } while(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 *= FFS_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); error = ffs_getblk(vp, lbn, FFS_NOBLK, fs->fs_bsize, false, &bp); if (error) { *countp = 0; return error; } if (bp->b_oflags & (BO_DONE | BO_DELWRI)) { /* Braces must be here in case trace evaluates to nothing. */ trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn); } else { trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn); curlwp->l_ru.ru_inblock++; /* pay for read */ bp->b_flags |= B_READ; bp->b_flags &= ~B_COWDONE; /* we change blkno below */ if (bp->b_bcount > bp->b_bufsize) panic("ffs_indirtrunc: bad buffer size"); bp->b_blkno = dbn; BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); VOP_STRATEGY(vp, bp); error = biowait(bp); if (error == 0) error = fscow_run(bp, true); } if (error) { brelse(bp, 0); *countp = 0; return (error); } if (ip->i_ump->um_fstype == UFS1) bap1 = (int32_t *)bp->b_data; else bap2 = (int64_t *)bp->b_data; if (lastbn >= 0) { copy = kmem_alloc(fs->fs_bsize, KM_SLEEP); memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize); for (i = last + 1; i < FFS_NINDIR(fs); i++) BAP_ASSIGN(ip, i, 0); error = bwrite(bp); if (error) allerror = error; if (ip->i_ump->um_fstype == UFS1) bap1 = (int32_t *)copy; else bap2 = (int64_t *)copy; } /* * Recursively free totally unused blocks. */ for (i = FFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last; i--, nlbn += factor) { nb = RBAP(ip, i); if (nb == 0) continue; if (level > SINGLE) { error = ffs_indirtrunc(ip, nlbn, FFS_FSBTODB(fs, nb), (daddr_t)-1, level - 1, &blkcount); if (error) allerror = error; blocksreleased += blkcount; } if ((ip->i_ump->um_mountp->mnt_wapbl) && ((level > SINGLE) || (ITOV(ip)->v_type != VREG))) { UFS_WAPBL_REGISTER_DEALLOCATION(ip->i_ump->um_mountp, FFS_FSBTODB(fs, nb), fs->fs_bsize); } else ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize, ip->i_number); blocksreleased += nblocks; } /* * Recursively free last partial block. */ if (level > SINGLE && lastbn >= 0) { last = lastbn % factor; nb = RBAP(ip, i); if (nb != 0) { error = ffs_indirtrunc(ip, nlbn, FFS_FSBTODB(fs, nb), last, level - 1, &blkcount); if (error) allerror = error; blocksreleased += blkcount; } } if (copy != NULL) { kmem_free(copy, fs->fs_bsize); } else { brelse(bp, BC_INVAL); } *countp = blocksreleased; return (allerror); }
/* * 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; }
/* * 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. */ ffs_balloc( register struct inode *ip, register ufs_daddr_t lbn, int size, kauth_cred_t cred, struct buf **bpp, int flags, int * blk_alloc) { register struct fs *fs; register ufs_daddr_t nb; struct buf *bp, *nbp; struct vnode *vp = ITOV(ip); 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]; int devBlockSize=0; int alloc_buffer = 1; struct mount *mp=vp->v_mount; #if REV_ENDIAN_FS int rev_endian=(mp->mnt_flag & MNT_REVEND); #endif /* REV_ENDIAN_FS */ *bpp = NULL; if (lbn < 0) return (EFBIG); fs = ip->i_fs; if (flags & B_NOBUFF) alloc_buffer = 0; if (blk_alloc) *blk_alloc = 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. */ nb = lblkno(fs, ip->i_size); if (nb < NDADDR && nb < lbn) { /* the filesize prior to this write can fit in direct * blocks (ie. fragmentaion is possibly done) * we are now extending the file write beyond * the block which has end of 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) { /* few fragments are already allocated,since the * current extends beyond this block * allocate the complete block as fragments are only * 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); /* adjust the inode size we just grew */ /* it is in nb+1 as nb starts from 0 */ ip->i_size = (nb + 1) * fs->fs_bsize; ubc_setsize(vp, (off_t)ip->i_size); ip->i_db[nb] = dbtofsb(fs, (ufs_daddr_t)buf_blkno(bp)); ip->i_flag |= IN_CHANGE | IN_UPDATE; if ((flags & B_SYNC) || (!alloc_buffer)) { if (!alloc_buffer) buf_setflags(bp, B_NOCACHE); buf_bwrite(bp); } else buf_bdwrite(bp); /* note that 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 >= (lbn + 1) * fs->fs_bsize) { if (alloc_buffer) { error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, NOCRED, &bp); if (error) { buf_brelse(bp); return (error); } *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) { if (alloc_buffer) { error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), osize, NOCRED, &bp); if (error) { buf_brelse(bp); return (error); } ip->i_flag |= IN_CHANGE | IN_UPDATE; *bpp = bp; return (0); } else { ip->i_flag |= IN_CHANGE | IN_UPDATE; return (0); } } else { error = ffs_realloccg(ip, lbn, ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), osize, nsize, cred, &bp); if (error) return (error); ip->i_db[lbn] = dbtofsb(fs, (ufs_daddr_t)buf_blkno(bp)); ip->i_flag |= IN_CHANGE | IN_UPDATE; /* adjust the inode size we just grew */ ip->i_size = (lbn * fs->fs_bsize) + size; ubc_setsize(vp, (off_t)ip->i_size); if (!alloc_buffer) { buf_setflags(bp, B_NOCACHE); if (flags & B_SYNC) buf_bwrite(bp); else buf_bdwrite(bp); } else *bpp = bp; return (0); } } else { if (ip->i_size < (lbn + 1) * fs->fs_bsize) 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); if (alloc_buffer) { bp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), nsize, 0, 0, BLK_WRITE); buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, newb))); if (flags & B_CLRBUF) buf_clear(bp); } ip->i_db[lbn] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; if (blk_alloc) { *blk_alloc = nsize; } if (alloc_buffer) *bpp = bp; return (0); } } /* * Determine the number of levels of indirection. */ pref = 0; if (error = ufs_getlbns(vp, lbn, indirs, &num)) return(error); #if DIAGNOSTIC if (num < 1) panic ("ffs_balloc: ufs_bmaparray returned indirect block"); #endif /* * Fetch the first indirect block allocating if necessary. */ --num; nb = ip->i_ib[indirs[0].in_off]; allocib = NULL; allocblk = allociblk; if (nb == 0) { pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) return (error); nb = newb; *allocblk++ = nb; bp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[1].in_lbn)), fs->fs_bsize, 0, 0, BLK_META); buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); buf_clear(bp); /* * Write synchronously conditional on mount flags. */ if ((vp)->v_mount->mnt_flag & MNT_ASYNC) { error = 0; buf_bdwrite(bp); } else if ((error = buf_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 = (int)buf_meta_bread(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { buf_brelse(bp); goto fail; } bap = (ufs_daddr_t *)buf_dataptr(bp); #if REV_ENDIAN_FS if (rev_endian) nb = OSSwapInt32(bap[indirs[i].in_off]); else { #endif /* REV_ENDIAN_FS */ nb = bap[indirs[i].in_off]; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ if (i == num) break; i += 1; if (nb != 0) { buf_brelse(bp); continue; } if (pref == 0) pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) { buf_brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; nbp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), fs->fs_bsize, 0, 0, BLK_META); buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); buf_clear(nbp); /* * Write synchronously conditional on mount flags. */ if ((vp)->v_mount->mnt_flag & MNT_ASYNC) { error = 0; buf_bdwrite(nbp); } else if (error = buf_bwrite(nbp)) { buf_brelse(bp); goto fail; } #if REV_ENDIAN_FS if (rev_endian) bap[indirs[i - 1].in_off] = OSSwapInt32(nb); else { #endif /* REV_ENDIAN_FS */ bap[indirs[i - 1].in_off] = nb; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { buf_bwrite(bp); } else { buf_bdwrite(bp); } } /* * Get the data block, allocating if necessary. */ if (nb == 0) { pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) { buf_brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; #if REV_ENDIAN_FS if (rev_endian) bap[indirs[i].in_off] = OSSwapInt32(nb); else { #endif /* REV_ENDIAN_FS */ bap[indirs[i].in_off] = nb; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ /* * If required, write synchronously, otherwise use * delayed write. */ if ((flags & B_SYNC)) { buf_bwrite(bp); } else { buf_bdwrite(bp); } if(alloc_buffer ) { nbp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, 0, 0, BLK_WRITE); buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); if (flags & B_CLRBUF) buf_clear(nbp); } if (blk_alloc) { *blk_alloc = fs->fs_bsize; } if(alloc_buffer) *bpp = nbp; return (0); } buf_brelse(bp); if (alloc_buffer) { if (flags & B_CLRBUF) { error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), (int)fs->fs_bsize, NOCRED, &nbp); if (error) { buf_brelse(nbp); goto fail; } } else { nbp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, 0, 0, BLK_WRITE); buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); } *bpp = nbp; } return (0); fail: /* * If we have failed part way through block allocation, we * have to deallocate any indirect blocks that we have allocated. */ for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { ffs_blkfree(ip, *blkp, fs->fs_bsize); deallocated += fs->fs_bsize; } if (allocib != NULL) *allocib = 0; if (deallocated) { devBlockSize = vfs_devblocksize(mp); #if QUOTA /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, (int64_t)-deallocated, cred, FORCE); #endif /* QUOTA */ ip->i_blocks -= btodb(deallocated, devBlockSize); ip->i_flag |= IN_CHANGE | IN_UPDATE; } return (error); }
/* * ffs_blkalloc allocates a disk block for ffs_pageout(), as a consequence * it does no buf_breads (that could lead to deadblock as the page may be already * marked busy as it is being paged out. Also important to note that we are not * growing the file in pageouts. So ip->i_size cannot increase by this call * due to the way UBC works. * This code is derived from ffs_balloc and many cases of that are dealt * in ffs_balloc are not applicable here * Do not call with B_CLRBUF flags as this should only be called only * from pageouts */ ffs_blkalloc( struct inode *ip, ufs_daddr_t lbn, int size, kauth_cred_t cred, int flags) { register struct fs *fs; register ufs_daddr_t nb; struct buf *bp, *nbp; struct vnode *vp = ITOV(ip); 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]; int devBlockSize=0; struct mount *mp=vp->v_mount; #if REV_ENDIAN_FS int rev_endian=(mp->mnt_flag & MNT_REVEND); #endif /* REV_ENDIAN_FS */ fs = ip->i_fs; if(size > fs->fs_bsize) panic("ffs_blkalloc: too large for allocation"); /* * 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) { panic("ffs_blkalloc():cannot extend file: i_size %d, lbn %d", ip->i_size, lbn); } /* * The first NDADDR blocks are direct blocks */ if (lbn < NDADDR) { nb = ip->i_db[lbn]; if (nb != 0 && ip->i_size >= (lbn + 1) * fs->fs_bsize) { /* TBD: trivial case; the block is already allocated */ 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) { panic("ffs_allocblk: trying to extend a fragment"); } return(0); } else { if (ip->i_size < (lbn + 1) * fs->fs_bsize) 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); ip->i_db[lbn] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; return (0); } } /* * Determine the number of levels of indirection. */ pref = 0; if (error = ufs_getlbns(vp, lbn, indirs, &num)) return(error); if(num == 0) { panic("ffs_blkalloc: file with direct blocks only"); } /* * Fetch the first indirect block allocating if necessary. */ --num; nb = ip->i_ib[indirs[0].in_off]; allocib = NULL; allocblk = allociblk; if (nb == 0) { pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) return (error); nb = newb; *allocblk++ = nb; bp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[1].in_lbn)), fs->fs_bsize, 0, 0, BLK_META); buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); buf_clear(bp); /* * Write synchronously conditional on mount flags. */ if ((vp)->v_mount->mnt_flag & MNT_ASYNC) { error = 0; buf_bdwrite(bp); } else if (error = buf_bwrite(bp)) { 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 = (int)buf_meta_bread(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { buf_brelse(bp); goto fail; } bap = (ufs_daddr_t *)buf_dataptr(bp); #if REV_ENDIAN_FS if (rev_endian) nb = OSSwapInt32(bap[indirs[i].in_off]); else { #endif /* REV_ENDIAN_FS */ nb = bap[indirs[i].in_off]; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ if (i == num) break; i += 1; if (nb != 0) { buf_brelse(bp); continue; } if (pref == 0) pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) { buf_brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; nbp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), fs->fs_bsize, 0, 0, BLK_META); buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb))); buf_clear(nbp); /* * Write synchronously conditional on mount flags. */ if ((vp)->v_mount->mnt_flag & MNT_ASYNC) { error = 0; buf_bdwrite(nbp); } else if (error = buf_bwrite(nbp)) { buf_brelse(bp); goto fail; } #if REV_ENDIAN_FS if (rev_endian) bap[indirs[i - 1].in_off] = OSSwapInt32(nb); else { #endif /* REV_ENDIAN_FS */ bap[indirs[i - 1].in_off] = nb; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { buf_bwrite(bp); } else { buf_bdwrite(bp); } } /* * Get the data block, allocating if necessary. */ if (nb == 0) { pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) { buf_brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; #if REV_ENDIAN_FS if (rev_endian) bap[indirs[i].in_off] = OSSwapInt32(nb); else { #endif /* REV_ENDIAN_FS */ bap[indirs[i].in_off] = nb; #if REV_ENDIAN_FS } #endif /* REV_ENDIAN_FS */ /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { buf_bwrite(bp); } else { buf_bdwrite(bp); } return (0); } buf_brelse(bp); return (0); fail: /* * If we have failed part way through block allocation, we * have to deallocate any indirect blocks that we have allocated. */ for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { ffs_blkfree(ip, *blkp, fs->fs_bsize); deallocated += fs->fs_bsize; } if (allocib != NULL) *allocib = 0; if (deallocated) { devBlockSize = vfs_devblocksize(mp); #if QUOTA /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, (int64_t)-deallocated, cred, FORCE); #endif /* QUOTA */ ip->i_blocks -= btodb(deallocated, devBlockSize); ip->i_flag |= IN_CHANGE | IN_UPDATE; } return (error); }
/* * 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. */ int ffs1_balloc(struct inode *ip, off_t startoffset, int size, struct ucred *cred, int flags, struct buf **bpp) { daddr_t lbn, nb, newb, pref; struct fs *fs; struct buf *bp, *nbp; struct vnode *vp; struct proc *p; struct indir indirs[NIADDR + 2]; int32_t *bap; int deallocated, osize, nsize, num, i, error; int32_t *allocib, *blkp, *allocblk, allociblk[NIADDR+1]; int unwindidx = -1; vp = ITOV(ip); fs = ip->i_fs; p = curproc; lbn = lblkno(fs, startoffset); size = blkoff(fs, startoffset) + size; if (size > fs->fs_bsize) panic("ffs1_balloc: blk too big"); if (bpp != NULL) *bpp = NULL; if (lbn < 0) return (EFBIG); /* * 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_ffs1_size); if (nb < NDADDR && nb < lbn) { osize = blksize(fs, ip, nb); if (osize < fs->fs_bsize && osize > 0) { error = ffs_realloccg(ip, nb, ffs1_blkpref(ip, nb, (int)nb, &ip->i_ffs1_db[0]), osize, (int)fs->fs_bsize, cred, bpp, &newb); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, nb, newb, ip->i_ffs1_db[nb], fs->fs_bsize, osize, bpp ? *bpp : NULL); ip->i_ffs1_size = lblktosize(fs, nb + 1); uvm_vnp_setsize(vp, ip->i_ffs1_size); ip->i_ffs1_db[nb] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; if (bpp != NULL) { if (flags & B_SYNC) bwrite(*bpp); else bawrite(*bpp); } } } /* * The first NDADDR blocks are direct blocks */ if (lbn < NDADDR) { nb = ip->i_ffs1_db[lbn]; if (nb != 0 && ip->i_ffs1_size >= lblktosize(fs, lbn + 1)) { /* * The block is an already-allocated direct block * and the file already extends past this block, * thus this must be a whole block. * Just read the block (if requested). */ if (bpp != NULL) { error = bread(vp, lbn, fs->fs_bsize, bpp); if (error) { brelse(*bpp); return (error); } } return (0); } if (nb != 0) { /* * Consider need to reallocate a fragment. */ osize = fragroundup(fs, blkoff(fs, ip->i_ffs1_size)); nsize = fragroundup(fs, size); if (nsize <= osize) { /* * The existing block is already * at least as big as we want. * Just read the block (if requested). */ if (bpp != NULL) { error = bread(vp, lbn, fs->fs_bsize, bpp); if (error) { brelse(*bpp); return (error); } (*bpp)->b_bcount = osize; } return (0); } else { /* * The existing block is smaller than we * want, grow it. */ error = ffs_realloccg(ip, lbn, ffs1_blkpref(ip, lbn, (int)lbn, &ip->i_ffs1_db[0]), osize, nsize, cred, bpp, &newb); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, nb, nsize, osize, bpp ? *bpp : NULL); } } else { /* * The block was not previously allocated, * allocate a new block or fragment. */ if (ip->i_ffs1_size < lblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; error = ffs_alloc(ip, lbn, ffs1_blkpref(ip, lbn, (int)lbn, &ip->i_ffs1_db[0]), nsize, cred, &newb); if (error) return (error); if (bpp != NULL) { *bpp = getblk(vp, lbn, fs->fs_bsize, 0, 0); if (nsize < fs->fs_bsize) (*bpp)->b_bcount = nsize; (*bpp)->b_blkno = fsbtodb(fs, newb); if (flags & B_CLRBUF) clrbuf(*bpp); } if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, 0, nsize, 0, bpp ? *bpp : NULL); } ip->i_ffs1_db[lbn] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; 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 ("ffs1_balloc: ufs_bmaparray returned indirect block"); #endif /* * Fetch the first indirect block allocating if necessary. */ --num; nb = ip->i_ffs1_ib[indirs[0].in_off]; allocib = NULL; allocblk = allociblk; if (nb == 0) { pref = ffs1_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb); if (error) goto fail; nb = newb; *allocblk++ = nb; bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); bp->b_blkno = fsbtodb(fs, nb); 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 ((error = bwrite(bp)) != 0) goto fail; } allocib = &ip->i_ffs1_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, indirs[i].in_lbn, (int)fs->fs_bsize, &bp); if (error) { brelse(bp); goto fail; } bap = (int32_t *)bp->b_data; nb = bap[indirs[i].in_off]; if (i == num) break; i++; if (nb != 0) { brelse(bp); continue; } if (pref == 0) pref = ffs1_blkpref(ip, lbn, i - num - 1, NULL); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb); if (error) { brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); 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 { bdwrite(bp); } } /* * Get the data block, allocating if necessary. */ if (nb == 0) { pref = ffs1_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; if (bpp != NULL) { nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); if (flags & B_CLRBUF) clrbuf(nbp); *bpp = nbp; } if (DOINGSOFTDEP(vp)) softdep_setup_allocindir_page(ip, lbn, bp, indirs[i].in_off, nb, 0, bpp ? *bpp : NULL); bap[indirs[i].in_off] = nb; /* * If required, write synchronously, otherwise use * delayed write. */ if (flags & B_SYNC) { bwrite(bp); } else { bdwrite(bp); } return (0); } brelse(bp); if (bpp != NULL) { if (flags & B_CLRBUF) { error = bread(vp, lbn, (int)fs->fs_bsize, &nbp); if (error) { brelse(nbp); goto fail; } } else { nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); } *bpp = nbp; } return (0); fail: /* * If we have failed to allocate any blocks, simply return the error. * This is the usual case and avoids the need to fsync the file. */ if (allocblk == allociblk && allocib == NULL && unwindidx == -1) return (error); /* * If we have failed part way through block allocation, we have to * deallocate any indirect blocks that we have allocated. We have to * fsync the file before we start to get rid of all of its * dependencies so that we do not leave them dangling. We have to sync * it at the end so that the softdep code does not find any untracked * changes. Although this is really slow, running out of disk space is * not expected to be a common occurrence. The error return from fsync * is ignored as we already have an error to return to the user. */ VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p); for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { ffs_blkfree(ip, *blkp, fs->fs_bsize); deallocated += fs->fs_bsize; } if (allocib != NULL) { *allocib = 0; } else if (unwindidx >= 0) { int r; r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp); if (r) panic("Could not unwind indirect block, error %d", r); bap = (int32_t *)bp->b_data; bap[indirs[unwindidx].in_off] = 0; if (flags & B_SYNC) { bwrite(bp); } else { bdwrite(bp); } } if (deallocated) { /* * Restore user's disk quota because allocation failed. */ (void)ufs_quota_free_blocks(ip, btodb(deallocated), cred); ip->i_ffs1_blocks -= btodb(deallocated); ip->i_flag |= IN_CHANGE | IN_UPDATE; } VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p); return (error); }
int ffs2_balloc(struct inode *ip, off_t off, int size, struct ucred *cred, int flags, struct buf **bpp) { daddr_t lbn, lastlbn, nb, newb, *blkp; daddr_t pref, *allocblk, allociblk[NIADDR + 1]; daddr_t *bap, *allocib; int deallocated, osize, nsize, num, i, error, unwindidx, r; struct buf *bp, *nbp; struct indir indirs[NIADDR + 2]; struct fs *fs; struct vnode *vp; struct proc *p; vp = ITOV(ip); fs = ip->i_fs; p = curproc; unwindidx = -1; lbn = lblkno(fs, off); size = blkoff(fs, off) + size; if (size > fs->fs_bsize) panic("ffs2_balloc: block too big"); if (bpp != NULL) *bpp = NULL; if (lbn < 0) return (EFBIG); /* * 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_ffs2_size); if (lastlbn < NDADDR && lastlbn < lbn) { nb = lastlbn; osize = blksize(fs, ip, nb); if (osize < fs->fs_bsize && osize > 0) { error = ffs_realloccg(ip, nb, ffs2_blkpref(ip, lastlbn, nb, &ip->i_ffs2_db[0]), osize, (int) fs->fs_bsize, cred, bpp, &newb); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, nb, newb, ip->i_ffs2_db[nb], fs->fs_bsize, osize, bpp ? *bpp : NULL); ip->i_ffs2_size = lblktosize(fs, nb + 1); uvm_vnp_setsize(vp, ip->i_ffs2_size); ip->i_ffs2_db[nb] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; if (bpp) { if (flags & B_SYNC) bwrite(*bpp); else bawrite(*bpp); } } } /* * The first NDADDR blocks are direct. */ if (lbn < NDADDR) { nb = ip->i_ffs2_db[lbn]; if (nb != 0 && ip->i_ffs2_size >= lblktosize(fs, lbn + 1)) { /* * The direct block is already allocated and the file * extends past this block, thus this must be a whole * block. Just read it, if requested. */ if (bpp != NULL) { error = bread(vp, lbn, fs->fs_bsize, bpp); if (error) { brelse(*bpp); return (error); } } return (0); } if (nb != 0) { /* * Consider the need to allocate a fragment. */ osize = fragroundup(fs, blkoff(fs, ip->i_ffs2_size)); nsize = fragroundup(fs, size); if (nsize <= osize) { /* * The existing block is already at least as * big as we want. Just read it, if requested. */ if (bpp != NULL) { error = bread(vp, lbn, fs->fs_bsize, bpp); if (error) { brelse(*bpp); return (error); } (*bpp)->b_bcount = osize; } return (0); } else { /* * The existing block is smaller than we want, * grow it. */ error = ffs_realloccg(ip, lbn, ffs2_blkpref(ip, lbn, (int) lbn, &ip->i_ffs2_db[0]), osize, nsize, cred, bpp, &newb); if (error) return (error); if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, nb, nsize, osize, bpp ? *bpp : NULL); } } else { /* * The block was not previously allocated, allocate a * new block or fragment. */ if (ip->i_ffs2_size < lblktosize(fs, lbn + 1)) nsize = fragroundup(fs, size); else nsize = fs->fs_bsize; error = ffs_alloc(ip, lbn, ffs2_blkpref(ip, lbn, (int) lbn, &ip->i_ffs2_db[0]), nsize, cred, &newb); if (error) return (error); if (bpp != NULL) { bp = getblk(vp, lbn, fs->fs_bsize, 0, 0); if (nsize < fs->fs_bsize) bp->b_bcount = nsize; bp->b_blkno = fsbtodb(fs, newb); if (flags & B_CLRBUF) clrbuf(bp); *bpp = bp; } if (DOINGSOFTDEP(vp)) softdep_setup_allocdirect(ip, lbn, newb, 0, nsize, 0, bpp ? *bpp : NULL); } ip->i_ffs2_db[lbn] = newb; ip->i_flag |= IN_CHANGE | IN_UPDATE; return (0); } /* * Determine the number of levels of indirection. */ pref = 0; error = ufs_getlbns(vp, lbn, indirs, &num); if (error) return (error); #ifdef DIAGNOSTIC if (num < 1) panic("ffs2_balloc: ufs_bmaparray returned indirect block"); #endif /* * Fetch the first indirect block allocating it necessary. */ --num; nb = ip->i_ffs2_ib[indirs[0].in_off]; allocib = NULL; allocblk = allociblk; if (nb == 0) { pref = ffs2_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL); error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred, &newb); if (error) goto fail; nb = newb; *allocblk++ = nb; bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); bp->b_blkno = fsbtodb(fs, nb); 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. */ error = bwrite(bp); if (error) goto fail; } unwindidx = 0; allocib = &ip->i_ffs2_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, indirs[i].in_lbn, (int)fs->fs_bsize, &bp); if (error) { brelse(bp); goto fail; } bap = (int64_t *) bp->b_data; nb = bap[indirs[i].in_off]; if (i == num) break; i++; if (nb != 0) { brelse(bp); continue; } if (pref == 0) pref = ffs2_blkpref(ip, lbn, i - num - 1, NULL); error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred, &newb); if (error) { brelse(bp); goto fail; } nb = newb; *allocblk++ = nb; nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); 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. */ error = bwrite(nbp); if (error) { brelse(bp); goto fail; } } if (unwindidx < 0) unwindidx = i - 1; bap[indirs[i - 1].in_off] = nb; /* * If required, write synchronously, otherwise use delayed * write. */ if (flags & B_SYNC) bwrite(bp); else bdwrite(bp); } /* * Get the data block, allocating if necessary. */ if (nb == 0) { pref = ffs2_blkpref(ip, lbn, indirs[num].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; if (bpp != NULL) { nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); if (flags & B_CLRBUF) clrbuf(nbp); *bpp = nbp; } if (DOINGSOFTDEP(vp)) softdep_setup_allocindir_page(ip, lbn, bp, indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); bap[indirs[num].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 bdwrite(bp); return (0); } brelse(bp); if (bpp != NULL) { if (flags & B_CLRBUF) { error = bread(vp, lbn, (int)fs->fs_bsize, &nbp); if (error) { brelse(nbp); goto fail; } } else { nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); nbp->b_blkno = fsbtodb(fs, nb); clrbuf(nbp); } *bpp = nbp; } return (0); fail: /* * If we have failed to allocate any blocks, simply return the error. * This is the usual case and avoids the need to fsync the file. */ if (allocblk == allociblk && allocib == NULL && unwindidx == -1) return (error); /* * If we have failed part way through block allocation, we have to * deallocate any indirect blocks that we have allocated. We have to * fsync the file before we start to get rid of all of its * dependencies so that we do not leave them dangling. We have to sync * it at the end so that the softdep code does not find any untracked * changes. Although this is really slow, running out of disk space is * not expected to be a common occurrence. The error return from fsync * is ignored as we already have an error to return to the user. */ VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p); if (unwindidx >= 0) { /* * First write out any buffers we've created to resolve their * softdeps. This must be done in reverse order of creation so * that we resolve the dependencies in one pass. * Write the cylinder group buffers for these buffers too. */ for (i = num; i >= unwindidx; i--) { if (i == 0) break; bp = getblk(vp, indirs[i].in_lbn, (int) fs->fs_bsize, 0, 0); if (bp->b_flags & B_DELWRI) { nb = fsbtodb(fs, cgtod(fs, dtog(fs, dbtofsb(fs, bp->b_blkno)))); bwrite(bp); bp = getblk(ip->i_devvp, nb, (int) fs->fs_cgsize, 0, 0); if (bp->b_flags & B_DELWRI) bwrite(bp); else { bp->b_flags |= B_INVAL; brelse(bp); } } else { bp->b_flags |= B_INVAL; brelse(bp); } } if (DOINGSOFTDEP(vp) && unwindidx == 0) { ip->i_flag |= IN_CHANGE | IN_UPDATE; ffs_update(ip, 1); } /* * Now that any dependencies that we created have been * resolved, we can undo the partial allocation. */ if (unwindidx == 0) { *allocib = 0; ip->i_flag |= IN_CHANGE | IN_UPDATE; if (DOINGSOFTDEP(vp)) ffs_update(ip, 1); } else { r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp); if (r) panic("ffs2_balloc: unwind failed"); bap = (int64_t *) bp->b_data; bap[indirs[unwindidx].in_off] = 0; bwrite(bp); } for (i = unwindidx + 1; i <= num; i++) { bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 0); bp->b_flags |= B_INVAL; brelse(bp); } } for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { ffs_blkfree(ip, *blkp, fs->fs_bsize); deallocated += fs->fs_bsize; } if (deallocated) { /* * Restore user's disk quota because allocation failed. */ (void) ufs_quota_free_blocks(ip, btodb(deallocated), cred); ip->i_ffs2_blocks -= btodb(deallocated); ip->i_flag |= IN_CHANGE | IN_UPDATE; } VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p); return (error); }