1139825Simp/*-
21541Srgrimes * Copyright (c) 1982, 1986, 1989, 1993
31541Srgrimes *	The Regents of the University of California.  All rights reserved.
41541Srgrimes *
51541Srgrimes * Redistribution and use in source and binary forms, with or without
61541Srgrimes * modification, are permitted provided that the following conditions
71541Srgrimes * are met:
81541Srgrimes * 1. Redistributions of source code must retain the above copyright
91541Srgrimes *    notice, this list of conditions and the following disclaimer.
101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright
111541Srgrimes *    notice, this list of conditions and the following disclaimer in the
121541Srgrimes *    documentation and/or other materials provided with the distribution.
131541Srgrimes * 4. Neither the name of the University nor the names of its contributors
141541Srgrimes *    may be used to endorse or promote products derived from this software
151541Srgrimes *    without specific prior written permission.
161541Srgrimes *
171541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
181541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
191541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
201541Srgrimes * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
211541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
221541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
231541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
241541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
251541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
261541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
271541Srgrimes * SUCH DAMAGE.
281541Srgrimes *
2922521Sdyson *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
301541Srgrimes */
311541Srgrimes
32116192Sobrien#include <sys/cdefs.h>
33116192Sobrien__FBSDID("$FreeBSD$");
34116192Sobrien
3513260Swollman#include "opt_quota.h"
3613260Swollman
371541Srgrimes#include <sys/param.h>
381541Srgrimes#include <sys/systm.h>
391541Srgrimes#include <sys/mount.h>
401541Srgrimes#include <sys/proc.h>
4160041Sphk#include <sys/bio.h>
421541Srgrimes#include <sys/buf.h>
431541Srgrimes#include <sys/vnode.h>
441541Srgrimes#include <sys/malloc.h>
451541Srgrimes#include <sys/resourcevar.h>
46251171Sjeff#include <sys/rwlock.h>
4768885Sdillon#include <sys/vmmeter.h>
4863897Smckusick#include <sys/stat.h>
491541Srgrimes
501541Srgrimes#include <vm/vm.h>
5112662Sdg#include <vm/vm_extern.h>
52187467Skib#include <vm/vm_object.h>
531541Srgrimes
5459241Srwatson#include <ufs/ufs/extattr.h>
551541Srgrimes#include <ufs/ufs/quota.h>
5630492Sphk#include <ufs/ufs/ufsmount.h>
571541Srgrimes#include <ufs/ufs/inode.h>
5837362Sbde#include <ufs/ufs/ufs_extern.h>
591541Srgrimes
601541Srgrimes#include <ufs/ffs/fs.h>
611541Srgrimes#include <ufs/ffs/ffs_extern.h>
621541Srgrimes
6398542Smckusickstatic int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
6498542Smckusick	    ufs2_daddr_t, int, ufs2_daddr_t *);
651541Srgrimes
661541Srgrimes/*
671541Srgrimes * Update the access, modified, and inode change times as specified by the
6837363Sbde * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
6937363Sbde * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
7037363Sbde * the timestamp update).  The IN_LAZYMOD flag is set to force a write
71163194Skib * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
72163194Skib * is currently being suspended (or is suspended) and vnode has been accessed.
73163194Skib * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
74163194Skib * reflect the presumably successful write, and if waitfor is set, then wait
75163194Skib * for the write to complete.
761541Srgrimes */
771541Srgrimesint
7842374Sbdeffs_update(vp, waitfor)
7930492Sphk	struct vnode *vp;
8030492Sphk	int waitfor;
811541Srgrimes{
8296506Sphk	struct fs *fs;
831541Srgrimes	struct buf *bp;
841541Srgrimes	struct inode *ip;
85232351Smckusick	int flags, error;
861541Srgrimes
87163194Skib	ASSERT_VOP_ELOCKED(vp, "ffs_update");
8842350Sbde	ufs_itimes(vp);
8930492Sphk	ip = VTOI(vp);
9042354Sbde	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
911541Srgrimes		return (0);
92163194Skib	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
9389384Smckusick	fs = ip->i_fs;
94224061Smckusick	if (fs->fs_ronly && ip->i_ump->um_fsckpid == 0)
951541Srgrimes		return (0);
96232351Smckusick	/*
97232351Smckusick	 * If we are updating a snapshot and another process is currently
98232351Smckusick	 * writing the buffer containing the inode for this snapshot then
99232351Smckusick	 * a deadlock can occur when it tries to check the snapshot to see
100232351Smckusick	 * if that block needs to be copied. Thus when updating a snapshot
101232351Smckusick	 * we check to see if the buffer is already locked, and if it is
102232351Smckusick	 * we drop the snapshot lock until the buffer has been written
103232351Smckusick	 * and is available to us. We have to grab a reference to the
104232351Smckusick	 * snapshot vnode to prevent it from being removed while we are
105232351Smckusick	 * waiting for the buffer.
106232351Smckusick	 */
107232351Smckusick	flags = 0;
108232351Smckusick	if (IS_SNAPSHOT(ip))
109232351Smckusick		flags = GB_LOCK_NOWAIT;
110233629Smckusickloop:
111232351Smckusick	error = breadn_flags(ip->i_devvp,
112232351Smckusick	     fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
113232351Smckusick	     (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
114232351Smckusick	if (error != 0) {
115232351Smckusick		if (error != EBUSY) {
116232351Smckusick			brelse(bp);
117232351Smckusick			return (error);
118232351Smckusick		}
119232351Smckusick		KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
120233629Smckusick		/*
121233629Smckusick		 * Wait for our inode block to become available.
122233629Smckusick		 *
123233629Smckusick		 * Hold a reference to the vnode to protect against
124233629Smckusick		 * ffs_snapgone(). Since we hold a reference, it can only
125233629Smckusick		 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
126233629Smckusick		 * or unmount. For an unmount, the entire filesystem will be
127233629Smckusick		 * gone, so we cannot attempt to touch anything associated
128233629Smckusick		 * with it while the vnode is unlocked; all we can do is
129233629Smckusick		 * pause briefly and try again. If when we relock the vnode
130233629Smckusick		 * we discover that it has been reclaimed, updating it is no
131233629Smckusick		 * longer necessary and we can just return an error.
132233629Smckusick		 */
133233629Smckusick		vref(vp);
134232351Smckusick		VOP_UNLOCK(vp, 0);
135233629Smckusick		pause("ffsupd", 1);
136232351Smckusick		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
137232351Smckusick		vrele(vp);
138233629Smckusick		if ((vp->v_iflag & VI_DOOMED) != 0)
139233629Smckusick			return (ENOENT);
140233629Smckusick		goto loop;
1411541Srgrimes	}
14234266Sjulian	if (DOINGSOFTDEP(vp))
14334266Sjulian		softdep_update_inodeblock(ip, bp, waitfor);
14434266Sjulian	else if (ip->i_effnlink != ip->i_nlink)
14534266Sjulian		panic("ffs_update: bad link cnt");
14698542Smckusick	if (ip->i_ump->um_fstype == UFS1)
14798542Smckusick		*((struct ufs1_dinode *)bp->b_data +
14898542Smckusick		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
14998542Smckusick	else
15098542Smckusick		*((struct ufs2_dinode *)bp->b_data +
15198542Smckusick		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
152232837Skib	if (waitfor && !DOINGASYNC(vp))
153232351Smckusick		error = bwrite(bp);
154232835Skib	else if (vm_page_count_severe() || buf_dirty_count_severe()) {
155232835Skib		bawrite(bp);
156232835Skib		error = 0;
15734266Sjulian	} else {
15832286Sdyson		if (bp->b_bufsize == fs->fs_bsize)
15932286Sdyson			bp->b_flags |= B_CLUSTEROK;
1601541Srgrimes		bdwrite(bp);
161232351Smckusick		error = 0;
1621541Srgrimes	}
163232351Smckusick	return (error);
1641541Srgrimes}
1651541Srgrimes
1661541Srgrimes#define	SINGLE	0	/* index of single indirect block */
1671541Srgrimes#define	DOUBLE	1	/* index of double indirect block */
1681541Srgrimes#define	TRIPLE	2	/* index of triple indirect block */
1691541Srgrimes/*
170144659Sjeff * Truncate the inode ip to at most length size, freeing the
1711541Srgrimes * disk blocks.
1721541Srgrimes */
1731549Srgrimesint
174234605Straszffs_truncate(vp, length, flags, cred)
17530474Sphk	struct vnode *vp;
17630474Sphk	off_t length;
17730474Sphk	int flags;
17830474Sphk	struct ucred *cred;
1791541Srgrimes{
180144659Sjeff	struct inode *ip;
18198542Smckusick	ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
18298542Smckusick	ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
183100344Smckusick	ufs2_daddr_t count, blocksreleased = 0, datablocks;
184177493Sjeff	struct bufobj *bo;
18596506Sphk	struct fs *fs;
1861541Srgrimes	struct buf *bp;
187140703Sjeff	struct ufsmount *ump;
188222958Sjeff	int softdeptrunc, journaltrunc;
189222958Sjeff	int needextclean, extblocks;
19098542Smckusick	int offset, size, level, nblocks;
191100344Smckusick	int i, error, allerror;
1921541Srgrimes	off_t osize;
1931541Srgrimes
194144659Sjeff	ip = VTOI(vp);
195144659Sjeff	fs = ip->i_fs;
196144659Sjeff	ump = ip->i_ump;
197177493Sjeff	bo = &vp->v_bufobj;
198140703Sjeff
199142263Sjeff	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
200142263Sjeff
20122521Sdyson	if (length < 0)
2023768Sdg		return (EINVAL);
203207141Sjeff	if (length > fs->fs_maxfilesize)
204207141Sjeff		return (EFBIG);
205222958Sjeff#ifdef QUOTA
206222958Sjeff	error = getinoquota(ip);
207222958Sjeff	if (error)
208222958Sjeff		return (error);
209222958Sjeff#endif
210100344Smckusick	/*
211100344Smckusick	 * Historically clients did not have to specify which data
212100344Smckusick	 * they were truncating. So, if not specified, we assume
213100344Smckusick	 * traditional behavior, e.g., just the normal data.
214100344Smckusick	 */
215100344Smckusick	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
216100344Smckusick		flags |= IO_NORMAL;
217209717Sjeff	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
218209717Sjeff		flags |= IO_SYNC;
219100344Smckusick	/*
220100344Smckusick	 * If we are truncating the extended-attributes, and cannot
221100344Smckusick	 * do it with soft updates, then do it slowly here. If we are
222100344Smckusick	 * truncating both the extended attributes and the file contents
223100344Smckusick	 * (e.g., the file is being unlinked), then pick it off with
224100344Smckusick	 * soft updates below.
225100344Smckusick	 */
226207141Sjeff	allerror = 0;
227100344Smckusick	needextclean = 0;
228222958Sjeff	softdeptrunc = 0;
229222958Sjeff	journaltrunc = DOINGSUJ(vp);
230222958Sjeff	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
231222958Sjeff		softdeptrunc = !softdep_slowdown(vp);
232100344Smckusick	extblocks = 0;
233144659Sjeff	datablocks = DIP(ip, i_blocks);
234144659Sjeff	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
235144659Sjeff		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
236100344Smckusick		datablocks -= extblocks;
237100344Smckusick	}
238100344Smckusick	if ((flags & IO_EXT) && extblocks > 0) {
239222958Sjeff		if (length != 0)
240222958Sjeff			panic("ffs_truncate: partial trunc of extdata");
241222958Sjeff		if (softdeptrunc || journaltrunc) {
242222958Sjeff			if ((flags & IO_NORMAL) == 0)
243222958Sjeff				goto extclean;
244100344Smckusick			needextclean = 1;
245100344Smckusick		} else {
246233438Smckusick			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
247100344Smckusick				return (error);
248100344Smckusick#ifdef QUOTA
249144659Sjeff			(void) chkdq(ip, -extblocks, NOCRED, 0);
250100344Smckusick#endif
251183754Sattilio			vinvalbuf(vp, V_ALT, 0, 0);
252225166Smm			vn_pages_remove(vp,
253187490Skib			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
254222958Sjeff			osize = ip->i_din2->di_extsize;
255222958Sjeff			ip->i_din2->di_blocks -= extblocks;
256144659Sjeff			ip->i_din2->di_extsize = 0;
257100344Smckusick			for (i = 0; i < NXADDR; i++) {
258144659Sjeff				oldblks[i] = ip->i_din2->di_extb[i];
259144659Sjeff				ip->i_din2->di_extb[i] = 0;
260100344Smckusick			}
261187790Srwatson			ip->i_flag |= IN_CHANGE;
262232836Skib			if ((error = ffs_update(vp, !DOINGASYNC(vp))))
263222958Sjeff				return (error);
264100344Smckusick			for (i = 0; i < NXADDR; i++) {
265100344Smckusick				if (oldblks[i] == 0)
266100344Smckusick					continue;
267144659Sjeff				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
268223127Smckusick				    sblksize(fs, osize, i), ip->i_number,
269223127Smckusick				    vp->v_type, NULL);
270100344Smckusick			}
271100344Smckusick		}
272100344Smckusick	}
273222958Sjeff	if ((flags & IO_NORMAL) == 0)
274222958Sjeff		return (0);
275144659Sjeff	if (vp->v_type == VLNK &&
276144659Sjeff	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
277100344Smckusick	     datablocks == 0)) {
278173464Sobrien#ifdef INVARIANTS
2791541Srgrimes		if (length != 0)
2801541Srgrimes			panic("ffs_truncate: partial truncate of symlink");
2811541Srgrimes#endif
282144659Sjeff		bzero(SHORTLINK(ip), (u_int)ip->i_size);
283144659Sjeff		ip->i_size = 0;
284144659Sjeff		DIP_SET(ip, i_size, 0);
285144659Sjeff		ip->i_flag |= IN_CHANGE | IN_UPDATE;
286100344Smckusick		if (needextclean)
287222958Sjeff			goto extclean;
288232836Skib		return (ffs_update(vp, !DOINGASYNC(vp)));
2891541Srgrimes	}
290144659Sjeff	if (ip->i_size == length) {
291144659Sjeff		ip->i_flag |= IN_CHANGE | IN_UPDATE;
292100344Smckusick		if (needextclean)
293222958Sjeff			goto extclean;
294232836Skib		return (ffs_update(vp, 0));
2951541Srgrimes	}
29689413Smckusick	if (fs->fs_ronly)
29789413Smckusick		panic("ffs_truncate: read-only filesystem");
298232351Smckusick	if (IS_SNAPSHOT(ip))
299144659Sjeff		ffs_snapremove(vp);
300144659Sjeff	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
301207141Sjeff	osize = ip->i_size;
302207141Sjeff	/*
303207141Sjeff	 * Lengthen the size of the file. We must ensure that the
304207141Sjeff	 * last byte of the file is allocated. Since the smallest
305207141Sjeff	 * value of osize is 0, length will be at least 1.
306207141Sjeff	 */
307207141Sjeff	if (osize < length) {
308207141Sjeff		vnode_pager_setsize(vp, length);
309207141Sjeff		flags |= BA_CLRBUF;
310207141Sjeff		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
311207141Sjeff		if (error) {
312207141Sjeff			vnode_pager_setsize(vp, osize);
313222958Sjeff			return (error);
314207141Sjeff		}
315207141Sjeff		ip->i_size = length;
316207141Sjeff		DIP_SET(ip, i_size, length);
317207141Sjeff		if (bp->b_bufsize == fs->fs_bsize)
318207141Sjeff			bp->b_flags |= B_CLUSTEROK;
319207141Sjeff		if (flags & IO_SYNC)
320207141Sjeff			bwrite(bp);
321232836Skib		else if (DOINGASYNC(vp))
322232836Skib			bdwrite(bp);
323207141Sjeff		else
324207141Sjeff			bawrite(bp);
325207141Sjeff		ip->i_flag |= IN_CHANGE | IN_UPDATE;
326232836Skib		return (ffs_update(vp, !DOINGASYNC(vp)));
327207141Sjeff	}
328144659Sjeff	if (DOINGSOFTDEP(vp)) {
329222958Sjeff		if (softdeptrunc == 0 && journaltrunc == 0) {
33034266Sjulian			/*
33134266Sjulian			 * If a file is only partially truncated, then
33234266Sjulian			 * we have to clean up the data structures
33334266Sjulian			 * describing the allocation past the truncation
33434266Sjulian			 * point. Finding and deallocating those structures
33534266Sjulian			 * is a lot of work. Since partial truncation occurs
33634266Sjulian			 * rarely, we solve the problem by syncing the file
33734266Sjulian			 * so that it will have no data structures left.
33834266Sjulian			 */
339233438Smckusick			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
340222958Sjeff				return (error);
34134266Sjulian		} else {
342222958Sjeff			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
343222958Sjeff			if (journaltrunc)
344222958Sjeff				softdep_journal_freeblocks(ip, cred, length,
345222958Sjeff				    flags);
346222958Sjeff			else
347222958Sjeff				softdep_setup_freeblocks(ip, length, flags);
348142263Sjeff			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
349222958Sjeff			if (journaltrunc == 0) {
350222958Sjeff				ip->i_flag |= IN_CHANGE | IN_UPDATE;
351222958Sjeff				error = ffs_update(vp, 0);
352222958Sjeff			}
353222958Sjeff			return (error);
35434266Sjulian		}
35534266Sjulian	}
3561541Srgrimes	/*
3571541Srgrimes	 * Shorten the size of the file. If the file is not being
35836990Sjulian	 * truncated to a block boundary, the contents of the
3591541Srgrimes	 * partial block following the end of the file must be
36036990Sjulian	 * zero'ed in case it ever becomes accessible again because
36136990Sjulian	 * of subsequent file growth. Directories however are not
36236990Sjulian	 * zero'ed as they should grow back initialized to empty.
3631541Srgrimes	 */
3641541Srgrimes	offset = blkoff(fs, length);
3651541Srgrimes	if (offset == 0) {
366144659Sjeff		ip->i_size = length;
367144659Sjeff		DIP_SET(ip, i_size, length);
3681541Srgrimes	} else {
3691541Srgrimes		lbn = lblkno(fs, length);
370100344Smckusick		flags |= BA_CLRBUF;
371144659Sjeff		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
372207141Sjeff		if (error)
373222958Sjeff			return (error);
37487782Smckusick		/*
37587782Smckusick		 * When we are doing soft updates and the UFS_BALLOC
37687782Smckusick		 * above fills in a direct block hole with a full sized
37787782Smckusick		 * block that will be truncated down to a fragment below,
37887782Smckusick		 * we must flush out the block dependency with an FSYNC
37987782Smckusick		 * so that we do not get a soft updates inconsistency
38087782Smckusick		 * when we create the fragment below.
38187782Smckusick		 */
382144659Sjeff		if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
38387782Smckusick		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
384233438Smckusick		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
385222958Sjeff			return (error);
386144659Sjeff		ip->i_size = length;
387144659Sjeff		DIP_SET(ip, i_size, length);
388144659Sjeff		size = blksize(fs, ip, lbn);
389144659Sjeff		if (vp->v_type != VDIR)
39036990Sjulian			bzero((char *)bp->b_data + offset,
39136990Sjulian			    (u_int)(size - offset));
39236990Sjulian		/* Kirk's code has reallocbuf(bp, size, 1) here */
3937399Sdg		allocbuf(bp, size);
39432286Sdyson		if (bp->b_bufsize == fs->fs_bsize)
39532286Sdyson			bp->b_flags |= B_CLUSTEROK;
396100344Smckusick		if (flags & IO_SYNC)
3971541Srgrimes			bwrite(bp);
398232836Skib		else if (DOINGASYNC(vp))
399232836Skib			bdwrite(bp);
4001541Srgrimes		else
4011541Srgrimes			bawrite(bp);
4021541Srgrimes	}
4031541Srgrimes	/*
4041541Srgrimes	 * Calculate index into inode's block list of
4051541Srgrimes	 * last direct and indirect blocks (if any)
4061541Srgrimes	 * which we want to keep.  Lastblock is -1 when
4071541Srgrimes	 * the file is truncated to 0.
4081541Srgrimes	 */
4091541Srgrimes	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
4101541Srgrimes	lastiblock[SINGLE] = lastblock - NDADDR;
4111541Srgrimes	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
4121541Srgrimes	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
4131541Srgrimes	nblocks = btodb(fs->fs_bsize);
4141541Srgrimes	/*
4151541Srgrimes	 * Update file and block pointers on disk before we start freeing
4161541Srgrimes	 * blocks.  If we crash before free'ing blocks below, the blocks
4171541Srgrimes	 * will be returned to the free list.  lastiblock values are also
4181541Srgrimes	 * normalized to -1 for calls to ffs_indirtrunc below.
4191541Srgrimes	 */
42098542Smckusick	for (level = TRIPLE; level >= SINGLE; level--) {
421144659Sjeff		oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
4221541Srgrimes		if (lastiblock[level] < 0) {
423144659Sjeff			DIP_SET(ip, i_ib[level], 0);
4241541Srgrimes			lastiblock[level] = -1;
4251541Srgrimes		}
42698542Smckusick	}
42798542Smckusick	for (i = 0; i < NDADDR; i++) {
428144659Sjeff		oldblks[i] = DIP(ip, i_db[i]);
42998542Smckusick		if (i > lastblock)
430144659Sjeff			DIP_SET(ip, i_db[i], 0);
43198542Smckusick	}
432144659Sjeff	ip->i_flag |= IN_CHANGE | IN_UPDATE;
433232836Skib	allerror = ffs_update(vp, !DOINGASYNC(vp));
43434696Sdyson
4351541Srgrimes	/*
4361541Srgrimes	 * Having written the new inode to disk, save its new configuration
4371541Srgrimes	 * and put back the old block pointers long enough to process them.
4381541Srgrimes	 * Note that we save the new block configuration so we can check it
4391541Srgrimes	 * when we are done.
4401541Srgrimes	 */
44198542Smckusick	for (i = 0; i < NDADDR; i++) {
442144659Sjeff		newblks[i] = DIP(ip, i_db[i]);
443144659Sjeff		DIP_SET(ip, i_db[i], oldblks[i]);
44498542Smckusick	}
44598542Smckusick	for (i = 0; i < NIADDR; i++) {
446144659Sjeff		newblks[NDADDR + i] = DIP(ip, i_ib[i]);
447144659Sjeff		DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
44898542Smckusick	}
449144659Sjeff	ip->i_size = osize;
450144659Sjeff	DIP_SET(ip, i_size, osize);
4511541Srgrimes
452234605Strasz	error = vtruncbuf(vp, cred, length, fs->fs_bsize);
45334696Sdyson	if (error && (allerror == 0))
45434696Sdyson		allerror = error;
45534696Sdyson
4561541Srgrimes	/*
4571541Srgrimes	 * Indirect blocks first.
4581541Srgrimes	 */
4591541Srgrimes	indir_lbn[SINGLE] = -NDADDR;
4601541Srgrimes	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
4611541Srgrimes	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
4621541Srgrimes	for (level = TRIPLE; level >= SINGLE; level--) {
463144659Sjeff		bn = DIP(ip, i_ib[level]);
4641541Srgrimes		if (bn != 0) {
465144659Sjeff			error = ffs_indirtrunc(ip, indir_lbn[level],
4661541Srgrimes			    fsbtodb(fs, bn), lastiblock[level], level, &count);
4671541Srgrimes			if (error)
4681541Srgrimes				allerror = error;
4691541Srgrimes			blocksreleased += count;
4701541Srgrimes			if (lastiblock[level] < 0) {
471144659Sjeff				DIP_SET(ip, i_ib[level], 0);
472144659Sjeff				ffs_blkfree(ump, fs, ip->i_devvp, bn,
473223127Smckusick				    fs->fs_bsize, ip->i_number,
474223127Smckusick				    vp->v_type, NULL);
4751541Srgrimes				blocksreleased += nblocks;
4761541Srgrimes			}
4771541Srgrimes		}
4781541Srgrimes		if (lastiblock[level] >= 0)
4791541Srgrimes			goto done;
4801541Srgrimes	}
4811541Srgrimes
4821541Srgrimes	/*
4831541Srgrimes	 * All whole direct blocks or frags.
4841541Srgrimes	 */
4851541Srgrimes	for (i = NDADDR - 1; i > lastblock; i--) {
48696506Sphk		long bsize;
4871541Srgrimes
488144659Sjeff		bn = DIP(ip, i_db[i]);
4891541Srgrimes		if (bn == 0)
4901541Srgrimes			continue;
491144659Sjeff		DIP_SET(ip, i_db[i], 0);
492144659Sjeff		bsize = blksize(fs, ip, i);
493207141Sjeff		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
494223127Smckusick		    vp->v_type, NULL);
4951541Srgrimes		blocksreleased += btodb(bsize);
4961541Srgrimes	}
4971541Srgrimes	if (lastblock < 0)
4981541Srgrimes		goto done;
4991541Srgrimes
5001541Srgrimes	/*
5011541Srgrimes	 * Finally, look for a change in size of the
5021541Srgrimes	 * last direct block; release any frags.
5031541Srgrimes	 */
504144659Sjeff	bn = DIP(ip, i_db[lastblock]);
5051541Srgrimes	if (bn != 0) {
5061541Srgrimes		long oldspace, newspace;
5071541Srgrimes
5081541Srgrimes		/*
5091541Srgrimes		 * Calculate amount of space we're giving
5101541Srgrimes		 * back as old block size minus new block size.
5111541Srgrimes		 */
512144659Sjeff		oldspace = blksize(fs, ip, lastblock);
513144659Sjeff		ip->i_size = length;
514144659Sjeff		DIP_SET(ip, i_size, length);
515144659Sjeff		newspace = blksize(fs, ip, lastblock);
5161541Srgrimes		if (newspace == 0)
5171826Sdg			panic("ffs_truncate: newspace");
5181541Srgrimes		if (oldspace - newspace > 0) {
5191541Srgrimes			/*
5201541Srgrimes			 * Block number of space to be free'd is
5211541Srgrimes			 * the old block # plus the number of frags
5221541Srgrimes			 * required for the storage we're keeping.
5231541Srgrimes			 */
5241541Srgrimes			bn += numfrags(fs, newspace);
525144659Sjeff			ffs_blkfree(ump, fs, ip->i_devvp, bn,
526223127Smckusick			   oldspace - newspace, ip->i_number, vp->v_type, NULL);
5271541Srgrimes			blocksreleased += btodb(oldspace - newspace);
5281541Srgrimes		}
5291541Srgrimes	}
5301541Srgrimesdone:
531173464Sobrien#ifdef INVARIANTS
5321541Srgrimes	for (level = SINGLE; level <= TRIPLE; level++)
533144659Sjeff		if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
5341826Sdg			panic("ffs_truncate1");
5351541Srgrimes	for (i = 0; i < NDADDR; i++)
536144659Sjeff		if (newblks[i] != DIP(ip, i_db[i]))
5371826Sdg			panic("ffs_truncate2");
538177493Sjeff	BO_LOCK(bo);
5391541Srgrimes	if (length == 0 &&
540144659Sjeff	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
541177493Sjeff	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
5421826Sdg		panic("ffs_truncate3");
543177493Sjeff	BO_UNLOCK(bo);
544173464Sobrien#endif /* INVARIANTS */
5451541Srgrimes	/*
5461541Srgrimes	 * Put back the real size.
5471541Srgrimes	 */
548144659Sjeff	ip->i_size = length;
549144659Sjeff	DIP_SET(ip, i_size, length);
550246289Smckusick	if (DIP(ip, i_blocks) >= blocksreleased)
551246289Smckusick		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
552246289Smckusick	else	/* sanity */
553144659Sjeff		DIP_SET(ip, i_blocks, 0);
554144659Sjeff	ip->i_flag |= IN_CHANGE;
5551541Srgrimes#ifdef QUOTA
556144659Sjeff	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
5571541Srgrimes#endif
558222958Sjeff	return (allerror);
559222958Sjeff
560222958Sjeffextclean:
561222958Sjeff	if (journaltrunc)
562222958Sjeff		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
563222958Sjeff	else
564222958Sjeff		softdep_setup_freeblocks(ip, length, IO_EXT);
565300600Skib	return (ffs_update(vp, (flags & IO_SYNC) != 0 || !DOINGASYNC(vp)));
5661541Srgrimes}
5671541Srgrimes
5681541Srgrimes/*
5691541Srgrimes * Release blocks associated with the inode ip and stored in the indirect
5701541Srgrimes * block bn.  Blocks are free'd in LIFO order up to (but not including)
5711541Srgrimes * lastbn.  If level is greater than SINGLE, the block is an indirect block
5721541Srgrimes * and recursive calls to indirtrunc must be used to cleanse other indirect
5731541Srgrimes * blocks.
5741541Srgrimes */
5751541Srgrimesstatic int
5761541Srgrimesffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
57796506Sphk	struct inode *ip;
57898542Smckusick	ufs2_daddr_t lbn, lastbn;
57998542Smckusick	ufs2_daddr_t dbn;
5801541Srgrimes	int level;
58198542Smckusick	ufs2_daddr_t *countp;
5821541Srgrimes{
5831541Srgrimes	struct buf *bp;
58496506Sphk	struct fs *fs = ip->i_fs;
5851541Srgrimes	struct vnode *vp;
58698542Smckusick	caddr_t copy = NULL;
58798542Smckusick	int i, nblocks, error = 0, allerror = 0;
58898542Smckusick	ufs2_daddr_t nb, nlbn, last;
58998542Smckusick	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
59098542Smckusick	ufs1_daddr_t *bap1 = NULL;
59198542Smckusick	ufs2_daddr_t *bap2 = NULL;
59298542Smckusick#	define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
5931541Srgrimes
5941541Srgrimes	/*
5951541Srgrimes	 * Calculate index in current block of last
5961541Srgrimes	 * block to be kept.  -1 indicates the entire
5971541Srgrimes	 * block so we need not calculate the index.
5981541Srgrimes	 */
599215113Skib	factor = lbn_offset(fs, level);
6001541Srgrimes	last = lastbn;
6011541Srgrimes	if (lastbn > 0)
6021541Srgrimes		last /= factor;
6031541Srgrimes	nblocks = btodb(fs->fs_bsize);
6041541Srgrimes	/*
6051541Srgrimes	 * Get buffer of block pointers, zero those entries corresponding
6061541Srgrimes	 * to blocks to be free'd, and update on disk copy first.  Since
6071541Srgrimes	 * double(triple) indirect before single(double) indirect, calls
6081541Srgrimes	 * to bmap on these blocks will fail.  However, we already have
6091541Srgrimes	 * the on disk address, so we have to set the b_blkno field
6101541Srgrimes	 * explicitly instead of letting bread do everything for us.
6111541Srgrimes	 */
6121541Srgrimes	vp = ITOV(ip);
613111856Sjeff	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
6146875Sdg	if ((bp->b_flags & B_CACHE) == 0) {
615170174Sjeff		curthread->td_ru.ru_inblock++;	/* pay for read */
61658345Sphk		bp->b_iocmd = BIO_READ;
61758934Sphk		bp->b_flags &= ~B_INVAL;
61858934Sphk		bp->b_ioflags &= ~BIO_ERROR;
6191541Srgrimes		if (bp->b_bcount > bp->b_bufsize)
6201541Srgrimes			panic("ffs_indirtrunc: bad buffer size");
6211541Srgrimes		bp->b_blkno = dbn;
6225455Sdg		vfs_busy_pages(bp, 0);
623121205Sphk		bp->b_iooffset = dbtob(bp->b_blkno);
624136927Sphk		bstrategy(bp);
62559762Sphk		error = bufwait(bp);
6261541Srgrimes	}
6271541Srgrimes	if (error) {
6281541Srgrimes		brelse(bp);
6291541Srgrimes		*countp = 0;
6301541Srgrimes		return (error);
6311541Srgrimes	}
6321541Srgrimes
63398542Smckusick	if (ip->i_ump->um_fstype == UFS1)
63498542Smckusick		bap1 = (ufs1_daddr_t *)bp->b_data;
63598542Smckusick	else
63698542Smckusick		bap2 = (ufs2_daddr_t *)bp->b_data;
63719424Sdg	if (lastbn != -1) {
638184205Sdes		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
63998542Smckusick		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
64098542Smckusick		for (i = last + 1; i < NINDIR(fs); i++)
641132775Skan			if (ip->i_ump->um_fstype == UFS1)
642132775Skan				bap1[i] = 0;
643132775Skan			else
644132775Skan				bap2[i] = 0;
64548801Smckusick		if (DOINGASYNC(vp)) {
646232836Skib			bdwrite(bp);
64748801Smckusick		} else {
64819424Sdg			error = bwrite(bp);
64919424Sdg			if (error)
65019424Sdg				allerror = error;
65119424Sdg		}
65298542Smckusick		if (ip->i_ump->um_fstype == UFS1)
65398542Smckusick			bap1 = (ufs1_daddr_t *)copy;
65498542Smckusick		else
65598542Smckusick			bap2 = (ufs2_daddr_t *)copy;
65612767Sdyson	}
6571541Srgrimes
6581541Srgrimes	/*
6591541Srgrimes	 * Recursively free totally unused blocks.
6601541Srgrimes	 */
6611541Srgrimes	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
6621541Srgrimes	    i--, nlbn += factor) {
66398542Smckusick		nb = BAP(ip, i);
6641541Srgrimes		if (nb == 0)
6651541Srgrimes			continue;
6661541Srgrimes		if (level > SINGLE) {
66743311Sdillon			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
66898542Smckusick			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
6691541Srgrimes				allerror = error;
6701541Srgrimes			blocksreleased += blkcount;
6711541Srgrimes		}
672140703Sjeff		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
673223127Smckusick		    ip->i_number, vp->v_type, NULL);
6741541Srgrimes		blocksreleased += nblocks;
6751541Srgrimes	}
6761541Srgrimes
6771541Srgrimes	/*
6781541Srgrimes	 * Recursively free last partial block.
6791541Srgrimes	 */
6801541Srgrimes	if (level > SINGLE && lastbn >= 0) {
6811541Srgrimes		last = lastbn % factor;
68298542Smckusick		nb = BAP(ip, i);
6831541Srgrimes		if (nb != 0) {
6843487Sphk			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
6853487Sphk			    last, level - 1, &blkcount);
6863487Sphk			if (error)
6871541Srgrimes				allerror = error;
6881541Srgrimes			blocksreleased += blkcount;
6891541Srgrimes		}
6901541Srgrimes	}
69119424Sdg	if (copy != NULL) {
692184205Sdes		free(copy, M_TEMP);
69319424Sdg	} else {
69419424Sdg		bp->b_flags |= B_INVAL | B_NOCACHE;
69519424Sdg		brelse(bp);
69619424Sdg	}
69798542Smckusick
6981541Srgrimes	*countp = blocksreleased;
6991541Srgrimes	return (allerror);
7001541Srgrimes}
701183070Skib
702183070Skibint
703183070Skibffs_rdonly(struct inode *ip)
704183070Skib{
705183070Skib
706183070Skib	return (ip->i_ump->um_fs->fs_ronly != 0);
707183070Skib}
708183070Skib
709