svn commit: r222958 - in head: sbin/fsck_ffs sys/sys sys/ufs/ffs
sys/ufs/ufs
Jeff Roberson
jroberson at jroberson.net
Fri Jun 10 23:24:39 UTC 2011
On Fri, 10 Jun 2011, Jeff Roberson wrote:
> Author: jeff
> Date: Fri Jun 10 22:48:35 2011
> New Revision: 222958
> URL: http://svn.freebsd.org/changeset/base/222958
>
> Log:
> Implement fully asynchronous partial truncation with softupdates journaling
> to resolve errors which can cause corruption on recovery with the old
> synchronous mechanism.
>
This diff is enormous and took months of work. I'm sorry to get it in so
close to 9.0, I had no idea it would take so long. pho has tested
multiple versions of the patch with and without journaling for days of
test time and it has probably racked up a week of machine time for me but
there may be problems given that it is so huge.
There is still a snapshot problem with SUJ that mckusick and I are working
on. Expect to see some checkins for that soon.
Thanks,
Jeff
> - Append partial truncation freework structures to indirdeps while
> truncation is proceeding. These prevent new block pointers from
> becoming valid until truncation completes and serialize truncations.
> - On completion of a partial truncate journal work waits for zeroed
> pointers to hit indirects.
> - softdep_journal_freeblocks() handles last frag allocation and last
> block zeroing.
> - vtruncbuf/ffs_page_remove moved into softdep_*_freeblocks() so it
> is only implemented in one place.
> - Block allocation failure handling moved up one level so it does not
> proceed with buf locks held. This permits us to do more extensive
> reclaims when filesystem space is exhausted.
> - softdep_sync_metadata() is broken into two parts, the first executes
> once at the start of ffs_syncvnode() and flushes truncations and
> inode dependencies. The second is called on each locked buf. This
> eliminates excessive looping and rollbacks.
> - Improve the mechanism in process_worklist_item() that handles
> acquiring vnode locks for handle_workitem_remove() so that it works
> more generally and does not loop excessively over the same worklist
> items on each call.
> - Don't corrupt directories by zeroing the tail in fsck. This is only
> done for regular files.
> - Push a fsync complete record for files that need it so the checker
> knows a truncation in the journal is no longer valid.
>
> Discussed with: mckusick, kib (ffs_pages_remove and ffs_truncate parts)
> Tested by: pho
>
> Modified:
> head/sbin/fsck_ffs/suj.c
> head/sys/sys/vnode.h
> head/sys/ufs/ffs/ffs_alloc.c
> head/sys/ufs/ffs/ffs_balloc.c
> head/sys/ufs/ffs/ffs_extern.h
> head/sys/ufs/ffs/ffs_inode.c
> head/sys/ufs/ffs/ffs_softdep.c
> head/sys/ufs/ffs/ffs_vfsops.c
> head/sys/ufs/ffs/ffs_vnops.c
> head/sys/ufs/ffs/fs.h
> head/sys/ufs/ffs/softdep.h
> head/sys/ufs/ufs/inode.h
> head/sys/ufs/ufs/ufsmount.h
>
> Modified: head/sbin/fsck_ffs/suj.c
> ==============================================================================
> --- head/sbin/fsck_ffs/suj.c Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sbin/fsck_ffs/suj.c Fri Jun 10 22:48:35 2011 (r222958)
> @@ -1604,7 +1604,7 @@ ino_trunc(ino_t ino, off_t size)
> * uninitialized space later.
> */
> off = blkoff(fs, size);
> - if (off) {
> + if (off && DIP(ip, di_mode) != IFDIR) {
> uint8_t *buf;
> long clrsize;
>
> @@ -1775,13 +1775,18 @@ cg_trunc(struct suj_cg *sc)
> struct suj_ino *sino;
> int i;
>
> - for (i = 0; i < SUJ_HASHSIZE; i++)
> - LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
> + for (i = 0; i < SUJ_HASHSIZE; i++) {
> + LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
> if (sino->si_trunc) {
> ino_trunc(sino->si_ino,
> sino->si_trunc->jt_size);
> + sino->si_blkadj = 0;
> sino->si_trunc = NULL;
> }
> + if (sino->si_blkadj)
> + ino_adjblks(sino);
> + }
> + }
> }
>
> /*
> @@ -1791,7 +1796,6 @@ cg_trunc(struct suj_cg *sc)
> static void
> cg_check_blk(struct suj_cg *sc)
> {
> - struct suj_ino *sino;
> struct suj_blk *sblk;
> int i;
>
> @@ -1799,15 +1803,6 @@ cg_check_blk(struct suj_cg *sc)
> for (i = 0; i < SUJ_HASHSIZE; i++)
> LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
> blk_check(sblk);
> - /*
> - * Now that we've freed blocks which are not referenced we
> - * make a second pass over all inodes to adjust their block
> - * counts.
> - */
> - for (i = 0; i < SUJ_HASHSIZE; i++)
> - LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
> - if (sino->si_blkadj)
> - ino_adjblks(sino);
> }
>
> /*
> @@ -1961,14 +1956,7 @@ ino_append(union jrec *rec)
> "parent %d, diroff %jd\n",
> refrec->jr_op, refrec->jr_ino, refrec->jr_nlink,
> refrec->jr_parent, refrec->jr_diroff);
> - /*
> - * Lookup the ino and clear truncate if one is found. Partial
> - * truncates are always done synchronously so if we discover
> - * an operation that requires a lock the truncation has completed
> - * and can be discarded.
> - */
> sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
> - sino->si_trunc = NULL;
> sino->si_hasrecs = 1;
> srec = errmalloc(sizeof(*srec));
> srec->sr_rec = rec;
> @@ -2174,9 +2162,7 @@ blk_build(struct jblkrec *blkrec)
> struct suj_rec *srec;
> struct suj_blk *sblk;
> struct jblkrec *blkrn;
> - struct suj_ino *sino;
> ufs2_daddr_t blk;
> - off_t foff;
> int frag;
>
> if (debug)
> @@ -2185,17 +2171,6 @@ blk_build(struct jblkrec *blkrec)
> blkrec->jb_op, blkrec->jb_blkno, blkrec->jb_frags,
> blkrec->jb_oldfrags, blkrec->jb_ino, blkrec->jb_lbn);
>
> - /*
> - * Look up the inode and clear the truncate if any lbns after the
> - * truncate lbn are freed or allocated.
> - */
> - sino = ino_lookup(blkrec->jb_ino, 0);
> - if (sino && sino->si_trunc) {
> - foff = lblktosize(fs, blkrec->jb_lbn);
> - foff += lfragtosize(fs, blkrec->jb_frags);
> - if (foff > sino->si_trunc->jt_size)
> - sino->si_trunc = NULL;
> - }
> blk = blknum(fs, blkrec->jb_blkno);
> frag = fragnum(fs, blkrec->jb_blkno);
> sblk = blk_lookup(blk, 1);
> @@ -2242,10 +2217,15 @@ ino_build_trunc(struct jtrncrec *rec)
> struct suj_ino *sino;
>
> if (debug)
> - printf("ino_build_trunc: ino %d, size %jd\n",
> - rec->jt_ino, rec->jt_size);
> + printf("ino_build_trunc: op %d ino %d, size %jd\n",
> + rec->jt_op, rec->jt_ino, rec->jt_size);
> sino = ino_lookup(rec->jt_ino, 1);
> - sino->si_trunc = rec;
> + if (rec->jt_op == JOP_SYNC) {
> + sino->si_trunc = NULL;
> + return;
> + }
> + if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
> + sino->si_trunc = rec;
> }
>
> /*
>
> Modified: head/sys/sys/vnode.h
> ==============================================================================
> --- head/sys/sys/vnode.h Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/sys/vnode.h Fri Jun 10 22:48:35 2011 (r222958)
> @@ -302,6 +302,7 @@ struct vattr {
> #define IO_EXT 0x0400 /* operate on external attributes */
> #define IO_NORMAL 0x0800 /* operate on regular data */
> #define IO_NOMACCHECK 0x1000 /* MAC checks unnecessary */
> +#define IO_BUFLOCKED 0x2000 /* ffs flag; indir buf is locked */
>
> #define IO_SEQMAX 0x7F /* seq heuristic max value */
> #define IO_SEQSHIFT 16 /* seq heuristic in upper 16 bits */
>
> Modified: head/sys/ufs/ffs/ffs_alloc.c
> ==============================================================================
> --- head/sys/ufs/ffs/ffs_alloc.c Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/ufs/ffs/ffs_alloc.c Fri Jun 10 22:48:35 2011 (r222958)
> @@ -217,7 +217,7 @@ nospace:
> (void) chkdq(ip, -btodb(size), cred, FORCE);
> UFS_LOCK(ump);
> #endif
> - if (reclaimed == 0) {
> + if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
> reclaimed = 1;
> softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT);
> goto retry;
> @@ -418,7 +418,7 @@ nospace:
> /*
> * no space available
> */
> - if (reclaimed == 0) {
> + if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
> reclaimed = 1;
> UFS_UNLOCK(ump);
> if (bp) {
>
> Modified: head/sys/ufs/ffs/ffs_balloc.c
> ==============================================================================
> --- head/sys/ufs/ffs/ffs_balloc.c Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/ufs/ffs/ffs_balloc.c Fri Jun 10 22:48:35 2011 (r222958)
> @@ -105,6 +105,7 @@ ffs_balloc_ufs1(struct vnode *vp, off_t
> ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
> int unwindidx = -1;
> int saved_inbdflush;
> + int reclaimed;
>
> ip = VTOI(vp);
> dp = ip->i_din1;
> @@ -112,6 +113,7 @@ ffs_balloc_ufs1(struct vnode *vp, off_t
> ump = ip->i_ump;
> lbn = lblkno(fs, startoffset);
> size = blkoff(fs, startoffset) + size;
> + reclaimed = 0;
> if (size > fs->fs_bsize)
> panic("ffs_balloc_ufs1: blk too big");
> *bpp = NULL;
> @@ -276,6 +278,7 @@ ffs_balloc_ufs1(struct vnode *vp, off_t
> /*
> * Fetch through the indirect blocks, allocating as necessary.
> */
> +retry:
> for (i = 1;;) {
> error = bread(vp,
> indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
> @@ -296,8 +299,15 @@ ffs_balloc_ufs1(struct vnode *vp, off_t
> if (pref == 0)
> pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
> if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
> - flags, cred, &newb)) != 0) {
> + flags | IO_BUFLOCKED, cred, &newb)) != 0) {
> brelse(bp);
> + if (++reclaimed == 1) {
> + UFS_LOCK(ump);
> + softdep_request_cleanup(fs, vp, cred,
> + FLUSH_BLOCKS_WAIT);
> + UFS_UNLOCK(ump);
> + goto retry;
> + }
> goto fail;
> }
> nb = newb;
> @@ -349,10 +359,17 @@ ffs_balloc_ufs1(struct vnode *vp, off_t
> if (nb == 0) {
> UFS_LOCK(ump);
> pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]);
> - error = ffs_alloc(ip,
> - lbn, pref, (int)fs->fs_bsize, flags, cred, &newb);
> + error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
> + flags | IO_BUFLOCKED, cred, &newb);
> if (error) {
> brelse(bp);
> + if (++reclaimed == 1) {
> + UFS_LOCK(ump);
> + softdep_request_cleanup(fs, vp, cred,
> + FLUSH_BLOCKS_WAIT);
> + UFS_UNLOCK(ump);
> + goto retry;
> + }
> goto fail;
> }
> nb = newb;
> @@ -506,6 +523,7 @@ ffs_balloc_ufs2(struct vnode *vp, off_t
> int deallocated, osize, nsize, num, i, error;
> int unwindidx = -1;
> int saved_inbdflush;
> + int reclaimed;
>
> ip = VTOI(vp);
> dp = ip->i_din2;
> @@ -513,6 +531,7 @@ ffs_balloc_ufs2(struct vnode *vp, off_t
> ump = ip->i_ump;
> lbn = lblkno(fs, startoffset);
> size = blkoff(fs, startoffset) + size;
> + reclaimed = 0;
> if (size > fs->fs_bsize)
> panic("ffs_balloc_ufs2: blk too big");
> *bpp = NULL;
> @@ -787,6 +806,7 @@ ffs_balloc_ufs2(struct vnode *vp, off_t
> /*
> * Fetch through the indirect blocks, allocating as necessary.
> */
> +retry:
> for (i = 1;;) {
> error = bread(vp,
> indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
> @@ -807,8 +827,15 @@ ffs_balloc_ufs2(struct vnode *vp, off_t
> if (pref == 0)
> pref = ffs_blkpref_ufs2(ip, lbn, 0, (ufs2_daddr_t *)0);
> if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
> - flags, cred, &newb)) != 0) {
> + flags | IO_BUFLOCKED, cred, &newb)) != 0) {
> brelse(bp);
> + if (++reclaimed == 1) {
> + UFS_LOCK(ump);
> + softdep_request_cleanup(fs, vp, cred,
> + FLUSH_BLOCKS_WAIT);
> + UFS_UNLOCK(ump);
> + goto retry;
> + }
> goto fail;
> }
> nb = newb;
> @@ -860,10 +887,17 @@ ffs_balloc_ufs2(struct vnode *vp, off_t
> if (nb == 0) {
> UFS_LOCK(ump);
> pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off, &bap[0]);
> - error = ffs_alloc(ip,
> - lbn, pref, (int)fs->fs_bsize, flags, cred, &newb);
> + error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
> + flags | IO_BUFLOCKED, cred, &newb);
> if (error) {
> brelse(bp);
> + if (++reclaimed == 1) {
> + UFS_LOCK(ump);
> + softdep_request_cleanup(fs, vp, cred,
> + FLUSH_BLOCKS_WAIT);
> + UFS_UNLOCK(ump);
> + goto retry;
> + }
> goto fail;
> }
> nb = newb;
>
> Modified: head/sys/ufs/ffs/ffs_extern.h
> ==============================================================================
> --- head/sys/ufs/ffs/ffs_extern.h Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/ufs/ffs/ffs_extern.h Fri Jun 10 22:48:35 2011 (r222958)
> @@ -74,6 +74,7 @@ int ffs_isfreeblock(struct fs *, u_char
> void ffs_load_inode(struct buf *, struct inode *, struct fs *, ino_t);
> int ffs_mountroot(void);
> void ffs_oldfscompat_write(struct fs *, struct ufsmount *);
> +void ffs_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end);
> int ffs_reallocblks(struct vop_reallocblks_args *);
> int ffs_realloccg(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
> ufs2_daddr_t, int, int, int, struct ucred *, struct buf **);
> @@ -107,7 +108,6 @@ extern struct vop_vector ffs_fifoops2;
>
> int softdep_check_suspend(struct mount *, struct vnode *,
> int, int, int, int);
> -int softdep_complete_trunc(struct vnode *, void *);
> void softdep_get_depcounts(struct mount *, int *, int *);
> void softdep_initialize(void);
> void softdep_uninitialize(void);
> @@ -139,14 +139,17 @@ void softdep_setup_blkfree(struct mount
> void softdep_setup_inofree(struct mount *, struct buf *, ino_t,
> struct workhead *);
> void softdep_setup_sbupdate(struct ufsmount *, struct fs *, struct buf *);
> -void *softdep_setup_trunc(struct vnode *vp, off_t length, int flags);
> void softdep_fsync_mountdev(struct vnode *);
> int softdep_sync_metadata(struct vnode *);
> +int softdep_sync_buf(struct vnode *, struct buf *, int);
> int softdep_process_worklist(struct mount *, int);
> int softdep_fsync(struct vnode *);
> int softdep_waitidle(struct mount *);
> int softdep_prealloc(struct vnode *, int);
> int softdep_journal_lookup(struct mount *, struct vnode **);
> +void softdep_journal_freeblocks(struct inode *, struct ucred *, off_t, int);
> +void softdep_journal_fsync(struct inode *);
> +
>
> /*
> * Things to request flushing in softdep_request_cleanup()
>
> Modified: head/sys/ufs/ffs/ffs_inode.c
> ==============================================================================
> --- head/sys/ufs/ffs/ffs_inode.c Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/ufs/ffs/ffs_inode.c Fri Jun 10 22:48:35 2011 (r222958)
> @@ -120,7 +120,7 @@ ffs_update(vp, waitfor)
> }
> }
>
> -static void
> +void
> ffs_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
> {
> vm_object_t object;
> @@ -151,12 +151,12 @@ ffs_truncate(vp, length, flags, cred, td
> ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
> ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
> ufs2_daddr_t count, blocksreleased = 0, datablocks;
> - void *cookie;
> struct bufobj *bo;
> struct fs *fs;
> struct buf *bp;
> struct ufsmount *ump;
> - int needextclean, softdepslowdown, extblocks;
> + int softdeptrunc, journaltrunc;
> + int needextclean, extblocks;
> int offset, size, level, nblocks;
> int i, error, allerror;
> off_t osize;
> @@ -165,7 +165,6 @@ ffs_truncate(vp, length, flags, cred, td
> fs = ip->i_fs;
> ump = ip->i_ump;
> bo = &vp->v_bufobj;
> - cookie = NULL;
>
> ASSERT_VOP_LOCKED(vp, "ffs_truncate");
>
> @@ -173,6 +172,11 @@ ffs_truncate(vp, length, flags, cred, td
> 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
> @@ -191,7 +195,10 @@ ffs_truncate(vp, length, flags, cred, td
> */
> allerror = 0;
> needextclean = 0;
> - softdepslowdown = DOINGSOFTDEP(vp) && softdep_slowdown(vp);
> + 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) {
> @@ -199,27 +206,23 @@ ffs_truncate(vp, length, flags, cred, td
> datablocks -= extblocks;
> }
> if ((flags & IO_EXT) && extblocks > 0) {
> - if (DOINGSOFTDEP(vp) && softdepslowdown == 0 && length == 0) {
> - if ((flags & IO_NORMAL) == 0) {
> - softdep_setup_freeblocks(ip, length, IO_EXT);
> - return (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 (length != 0)
> - panic("ffs_truncate: partial trunc of extdata");
> if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
> return (error);
> - if (DOINGSUJ(vp))
> - cookie = softdep_setup_trunc(vp, length, flags);
> - osize = ip->i_din2->di_extsize;
> - ip->i_din2->di_blocks -= extblocks;
> #ifdef QUOTA
> (void) chkdq(ip, -extblocks, NOCRED, 0);
> #endif
> vinvalbuf(vp, V_ALT, 0, 0);
> ffs_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 < NXADDR; i++) {
> oldblks[i] = ip->i_din2->di_extb[i];
> @@ -227,7 +230,7 @@ ffs_truncate(vp, length, flags, cred, td
> }
> ip->i_flag |= IN_CHANGE;
> if ((error = ffs_update(vp, 1)))
> - goto out;
> + return (error);
> for (i = 0; i < NXADDR; i++) {
> if (oldblks[i] == 0)
> continue;
> @@ -236,10 +239,8 @@ ffs_truncate(vp, length, flags, cred, td
> }
> }
> }
> - if ((flags & IO_NORMAL) == 0) {
> - error = 0;
> - goto out;
> - }
> + if ((flags & IO_NORMAL) == 0)
> + return (0);
> if (vp->v_type == VLNK &&
> (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
> datablocks == 0)) {
> @@ -252,24 +253,17 @@ ffs_truncate(vp, length, flags, cred, td
> DIP_SET(ip, i_size, 0);
> ip->i_flag |= IN_CHANGE | IN_UPDATE;
> if (needextclean)
> - softdep_setup_freeblocks(ip, length, IO_EXT);
> - error = ffs_update(vp, 1);
> - goto out;
> + goto extclean;
> + return ffs_update(vp, 1);
> }
> if (ip->i_size == length) {
> ip->i_flag |= IN_CHANGE | IN_UPDATE;
> if (needextclean)
> - softdep_setup_freeblocks(ip, length, IO_EXT);
> - error = ffs_update(vp, 0);
> - goto out;
> + goto extclean;
> + return ffs_update(vp, 0);
> }
> if (fs->fs_ronly)
> panic("ffs_truncate: read-only filesystem");
> -#ifdef QUOTA
> - error = getinoquota(ip);
> - if (error)
> - goto out;
> -#endif
> if ((ip->i_flags & SF_SNAPSHOT) != 0)
> ffs_snapremove(vp);
> vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
> @@ -285,7 +279,7 @@ ffs_truncate(vp, length, flags, cred, td
> error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
> if (error) {
> vnode_pager_setsize(vp, osize);
> - goto out;
> + return (error);
> }
> ip->i_size = length;
> DIP_SET(ip, i_size, length);
> @@ -296,11 +290,10 @@ ffs_truncate(vp, length, flags, cred, td
> else
> bawrite(bp);
> ip->i_flag |= IN_CHANGE | IN_UPDATE;
> - error = ffs_update(vp, 1);
> - goto out;
> + return ffs_update(vp, 1);
> }
> if (DOINGSOFTDEP(vp)) {
> - if (length > 0 || softdepslowdown) {
> + if (softdeptrunc == 0 && journaltrunc == 0) {
> /*
> * If a file is only partially truncated, then
> * we have to clean up the data structures
> @@ -311,29 +304,20 @@ ffs_truncate(vp, length, flags, cred, td
> * so that it will have no data structures left.
> */
> if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
> - goto out;
> - /*
> - * We have to journal the truncation before we change
> - * any blocks so we don't leave the file partially
> - * truncated.
> - */
> - if (DOINGSUJ(vp) && cookie == NULL)
> - cookie = softdep_setup_trunc(vp, length, flags);
> + return (error);
> } else {
> -#ifdef QUOTA
> - (void) chkdq(ip, -datablocks, NOCRED, 0);
> -#endif
> - softdep_setup_freeblocks(ip, length, needextclean ?
> - IO_EXT | IO_NORMAL : IO_NORMAL);
> + 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");
> - vinvalbuf(vp, needextclean ? 0 : V_NORMAL, 0, 0);
> - if (!needextclean)
> - ffs_pages_remove(vp, 0,
> - OFF_TO_IDX(lblktosize(fs, -extblocks)));
> - vnode_pager_setsize(vp, 0);
> - ip->i_flag |= IN_CHANGE | IN_UPDATE;
> - error = ffs_update(vp, 0);
> - goto out;
> + if (journaltrunc == 0) {
> + ip->i_flag |= IN_CHANGE | IN_UPDATE;
> + error = ffs_update(vp, 0);
> + }
> + return (error);
> }
> }
> /*
> @@ -353,7 +337,7 @@ ffs_truncate(vp, length, flags, cred, td
> flags |= BA_CLRBUF;
> error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
> if (error)
> - goto out;
> + return (error);
> /*
> * When we are doing soft updates and the UFS_BALLOC
> * above fills in a direct block hole with a full sized
> @@ -365,7 +349,7 @@ ffs_truncate(vp, length, flags, cred, td
> if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
> fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
> (error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
> - goto out;
> + return (error);
> ip->i_size = length;
> DIP_SET(ip, i_size, length);
> size = blksize(fs, ip, lbn);
> @@ -411,13 +395,7 @@ ffs_truncate(vp, length, flags, cred, td
> DIP_SET(ip, i_db[i], 0);
> }
> ip->i_flag |= IN_CHANGE | IN_UPDATE;
> - /*
> - * When doing softupdate journaling we must preserve the size along
> - * with the old pointers until they are freed or we might not
> - * know how many fragments remain.
> - */
> - if (!DOINGSUJ(vp))
> - allerror = ffs_update(vp, 1);
> + allerror = ffs_update(vp, 1);
>
> /*
> * Having written the new inode to disk, save its new configuration
> @@ -541,14 +519,14 @@ done:
> #ifdef QUOTA
> (void) chkdq(ip, -blocksreleased, NOCRED, 0);
> #endif
> - error = allerror;
> -out:
> - if (cookie) {
> - allerror = softdep_complete_trunc(vp, cookie);
> - if (allerror != 0 && error == 0)
> - error = allerror;
> - }
> - return (error);
> + 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, MNT_WAIT);
> }
>
> /*
>
> Modified: head/sys/ufs/ffs/ffs_softdep.c
> ==============================================================================
> --- head/sys/ufs/ffs/ffs_softdep.c Fri Jun 10 22:42:00 2011 (r222957)
> +++ head/sys/ufs/ffs/ffs_softdep.c Fri Jun 10 22:48:35 2011 (r222958)
> @@ -59,6 +59,7 @@ __FBSDID("$FreeBSD$");
> #include <sys/buf.h>
> #include <sys/kdb.h>
> #include <sys/kthread.h>
> +#include <sys/limits.h>
> #include <sys/lock.h>
> #include <sys/malloc.h>
> #include <sys/mount.h>
> @@ -71,6 +72,7 @@ __FBSDID("$FreeBSD$");
> #include <sys/syslog.h>
> #include <sys/vnode.h>
> #include <sys/conf.h>
> +
> #include <ufs/ufs/dir.h>
> #include <ufs/ufs/extattr.h>
> #include <ufs/ufs/quota.h>
> @@ -82,6 +84,8 @@ __FBSDID("$FreeBSD$");
> #include <ufs/ufs/ufs_extern.h>
>
> #include <vm/vm.h>
> +#include <vm/vm_extern.h>
> +#include <vm/vm_object.h>
>
> #include <ddb/ddb.h>
>
> @@ -214,6 +218,25 @@ softdep_setup_allocindir_meta(nbp, ip, b
> }
>
> void
> +softdep_journal_freeblocks(ip, cred, length, flags)
> + struct inode *ip;
> + struct ucred *cred;
> + off_t length;
> + int flags;
> +{
> +
> + panic("softdep_journal_freeblocks called");
> +}
> +
> +void
> +softdep_journal_fsync(ip)
> + struct inode *ip;
> +{
> +
> + panic("softdep_journal_fsync called");
> +}
> +
> +void
> softdep_setup_freeblocks(ip, length, flags)
> struct inode *ip;
> off_t length;
> @@ -282,29 +305,6 @@ softdep_setup_directory_change(bp, dp, i
> panic("softdep_setup_directory_change called");
> }
>
> -void *
> -softdep_setup_trunc(vp, length, flags)
> - struct vnode *vp;
> - off_t length;
> - int flags;
> -{
> -
> - panic("%s called", __FUNCTION__);
> -
> - return (NULL);
> -}
> -
> -int
> -softdep_complete_trunc(vp, cookie)
> - struct vnode *vp;
> - void *cookie;
> -{
> -
> - panic("%s called", __FUNCTION__);
> -
> - return (0);
> -}
> -
> void
> softdep_setup_blkfree(mp, bp, blkno, frags, wkhd)
> struct mount *mp;
> @@ -499,6 +499,13 @@ softdep_sync_metadata(struct vnode *vp)
> }
>
> int
> +softdep_sync_buf(struct vnode *vp, struct buf *bp, int waitfor)
> +{
> +
> + return (0);
> +}
> +
> +int
> softdep_slowdown(vp)
> struct vnode *vp;
> {
> @@ -614,10 +621,13 @@ FEATURE(softupdates, "FFS soft-updates s
> #define D_JSEGDEP 23
> #define D_SBDEP 24
> #define D_JTRUNC 25
> -#define D_LAST D_JTRUNC
> +#define D_JFSYNC 26
> +#define D_SENTINAL 27
> +#define D_LAST D_SENTINAL
>
> unsigned long dep_current[D_LAST + 1];
> unsigned long dep_total[D_LAST + 1];
> +unsigned long dep_write[D_LAST + 1];
>
>
> SYSCTL_NODE(_debug, OID_AUTO, softdep, CTLFLAG_RW, 0, "soft updates stats");
> @@ -625,13 +635,17 @@ SYSCTL_NODE(_debug_softdep, OID_AUTO, to
> "total dependencies allocated");
> SYSCTL_NODE(_debug_softdep, OID_AUTO, current, CTLFLAG_RW, 0,
> "current dependencies allocated");
> +SYSCTL_NODE(_debug_softdep, OID_AUTO, write, CTLFLAG_RW, 0,
> + "current dependencies written");
>
> #define SOFTDEP_TYPE(type, str, long) \
> static MALLOC_DEFINE(M_ ## type, #str, long); \
> SYSCTL_ULONG(_debug_softdep_total, OID_AUTO, str, CTLFLAG_RD, \
> &dep_total[D_ ## type], 0, ""); \
> SYSCTL_ULONG(_debug_softdep_current, OID_AUTO, str, CTLFLAG_RD, \
> - &dep_current[D_ ## type], 0, "");
> + &dep_current[D_ ## type], 0, ""); \
> + SYSCTL_ULONG(_debug_softdep_write, OID_AUTO, str, CTLFLAG_RD, \
> + &dep_write[D_ ## type], 0, "");
>
> SOFTDEP_TYPE(PAGEDEP, pagedep, "File page dependencies");
> SOFTDEP_TYPE(INODEDEP, inodedep, "Inode dependencies");
> @@ -660,6 +674,7 @@ SOFTDEP_TYPE(JSEG, jseg, "Journal segmen
> SOFTDEP_TYPE(JSEGDEP, jsegdep, "Journal segment complete");
> SOFTDEP_TYPE(SBDEP, sbdep, "Superblock write dependency");
> SOFTDEP_TYPE(JTRUNC, jtrunc, "Journal inode truncation");
> +SOFTDEP_TYPE(JFSYNC, jfsync, "Journal fsync complete");
>
> static MALLOC_DEFINE(M_SAVEDINO, "savedino", "Saved inodes");
> static MALLOC_DEFINE(M_JBLOCKS, "jblocks", "Journal block locations");
> @@ -694,7 +709,8 @@ static struct malloc_type *memtype[] = {
> M_JSEG,
> M_JSEGDEP,
> M_SBDEP,
> - M_JTRUNC
> + M_JTRUNC,
> + M_JFSYNC
> };
>
> static LIST_HEAD(mkdirlist, mkdir) mkdirlisthd;
> @@ -734,10 +750,11 @@ static void clear_unlinked_inodedep(stru
> static struct inodedep *first_unlinked_inodedep(struct ufsmount *);
> static int flush_pagedep_deps(struct vnode *, struct mount *,
> struct diraddhd *);
> -static void free_pagedep(struct pagedep *);
> +static int free_pagedep(struct pagedep *);
> static int flush_newblk_dep(struct vnode *, struct mount *, ufs_lbn_t);
> -static int flush_inodedep_deps(struct mount *, ino_t);
> +static int flush_inodedep_deps(struct vnode *, struct mount *, ino_t);
> static int flush_deplist(struct allocdirectlst *, int, int *);
> +static int sync_cgs(struct mount *, int);
> static int handle_written_filepage(struct pagedep *, struct buf *);
> static int handle_written_sbdep(struct sbdep *, struct buf *);
> static void initiate_write_sbdep(struct sbdep *);
> @@ -750,7 +767,7 @@ static void handle_written_jaddref(struc
> static void handle_written_jremref(struct jremref *);
> static void handle_written_jseg(struct jseg *, struct buf *);
> static void handle_written_jnewblk(struct jnewblk *);
> -static void handle_written_jfreeblk(struct jfreeblk *);
> +static void handle_written_jblkdep(struct jblkdep *);
> static void handle_written_jfreefrag(struct jfreefrag *);
> static void complete_jseg(struct jseg *);
> static void jseg_write(struct ufsmount *ump, struct jseg *, uint8_t *);
> @@ -758,6 +775,7 @@ static void jaddref_write(struct jaddref
> static void jremref_write(struct jremref *, struct jseg *, uint8_t *);
> static void jmvref_write(struct jmvref *, struct jseg *, uint8_t *);
> static void jtrunc_write(struct jtrunc *, struct jseg *, uint8_t *);
> +static void jfsync_write(struct jfsync *, struct jseg *, uint8_t *data);
> static void jnewblk_write(struct jnewblk *, struct jseg *, uint8_t *);
> static void jfreeblk_write(struct jfreeblk *, struct jseg *, uint8_t *);
> static void jfreefrag_write(struct jfreefrag *, struct jseg *, uint8_t *);
> @@ -768,7 +786,9 @@ static void handle_allocdirect_partdone(
> static struct jnewblk *cancel_newblk(struct newblk *, struct worklist *,
> struct workhead *);
> static void indirdep_complete(struct indirdep *);
> -static int indirblk_inseg(struct mount *, ufs2_daddr_t);
> +static int indirblk_lookup(struct mount *, ufs2_daddr_t);
> +static void indirblk_insert(struct freework *);
> +static void indirblk_remove(struct freework *);
> static void handle_allocindir_partdone(struct allocindir *);
> static void initiate_write_filepage(struct pagedep *, struct buf *);
> static void initiate_write_indirdep(struct indirdep*, struct buf *);
> @@ -777,10 +797,12 @@ static void initiate_write_bmsafemap(str
> static void initiate_write_inodeblock_ufs1(struct inodedep *, struct buf *);
> static void initiate_write_inodeblock_ufs2(struct inodedep *, struct buf *);
> static void handle_workitem_freefile(struct freefile *);
> -static void handle_workitem_remove(struct dirrem *, struct vnode *);
> +static int handle_workitem_remove(struct dirrem *, int);
> static struct dirrem *newdirrem(struct buf *, struct inode *,
> struct inode *, int, struct dirrem **);
> -static void cancel_indirdep(struct indirdep *, struct buf *, struct inodedep *,
> +static struct indirdep *indirdep_lookup(struct mount *, struct inode *,
> + struct buf *);
> +static void cancel_indirdep(struct indirdep *, struct buf *,
> struct freeblks *);
> static void free_indirdep(struct indirdep *);
> static void free_diradd(struct diradd *, struct workhead *);
> @@ -795,8 +817,13 @@ static void cancel_diradd(struct diradd
> struct jremref *, struct jremref *);
> static void dirrem_journal(struct dirrem *, struct jremref *, struct jremref *,
> struct jremref *);
> -static void cancel_allocindir(struct allocindir *, struct inodedep *,
> - struct freeblks *);
> +static void cancel_allocindir(struct allocindir *, struct buf *bp,
> + struct freeblks *, int);
> +static int setup_trunc_indir(struct freeblks *, struct inode *,
> + ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t);
> +static void complete_trunc_indir(struct freework *);
> +static void trunc_indirdep(struct indirdep *, struct freeblks *, struct buf *,
> + int);
> static void complete_mkdir(struct mkdir *);
> static void free_newdirblk(struct newdirblk *);
> static void free_jremref(struct jremref *);
> @@ -806,7 +833,7 @@ static void free_jsegs(struct jblocks *)
> static void rele_jseg(struct jseg *);
> static void free_jseg(struct jseg *, struct jblocks *);
> static void free_jnewblk(struct jnewblk *);
> -static void free_jfreeblk(struct jfreeblk *);
> +static void free_jblkdep(struct jblkdep *);
> static void free_jfreefrag(struct jfreefrag *);
> static void free_freedep(struct freedep *);
> static void journal_jremref(struct dirrem *, struct jremref *,
> @@ -818,30 +845,33 @@ static void cancel_jfreefrag(struct jfre
> static inline void setup_freedirect(struct freeblks *, struct inode *,
> int, int);
> static inline void setup_freeext(struct freeblks *, struct inode *, int, int);
> -static inline void setup_freeindir(struct freeblks *, struct inode *, int i,
> +static inline void setup_freeindir(struct freeblks *, struct inode *, int,
> ufs_lbn_t, int);
> static inline struct freeblks *newfreeblks(struct mount *, struct inode *);
> static void indir_trunc(struct freework *, ufs2_daddr_t, ufs_lbn_t);
> -static void softdep_trunc_deps(struct vnode *, struct freeblks *, ufs_lbn_t,
> +ufs2_daddr_t blkcount(struct fs *, ufs2_daddr_t, off_t);
> +static int trunc_check_buf(struct buf *, int *, ufs_lbn_t, int, int);
> +static void trunc_dependencies(struct inode *, struct freeblks *, ufs_lbn_t,
> int, int);
> -static int cancel_pagedep(struct pagedep *, struct inodedep *,
> - struct freeblks *);
> -static int deallocate_dependencies(struct buf *, struct inodedep *,
> - struct freeblks *, int off);
> +static void trunc_pages(struct inode *, off_t, ufs2_daddr_t, int);
> +static int cancel_pagedep(struct pagedep *, struct freeblks *, int);
> +static int deallocate_dependencies(struct buf *, struct freeblks *, int);
> +static void newblk_freefrag(struct newblk*);
> static void free_newblk(struct newblk *);
> static void cancel_allocdirect(struct allocdirectlst *,
> - struct allocdirect *, struct freeblks *, int);
> + struct allocdirect *, struct freeblks *);
> static int check_inode_unwritten(struct inodedep *);
> static int free_inodedep(struct inodedep *);
> static void freework_freeblock(struct freework *);
> -static void handle_workitem_freeblocks(struct freeblks *, int);
> -static void handle_complete_freeblocks(struct freeblks *);
> +static void freework_enqueue(struct freework *);
> +static int handle_workitem_freeblocks(struct freeblks *, int);
> +static int handle_complete_freeblocks(struct freeblks *, int);
> static void handle_workitem_indirblk(struct freework *);
> -static void handle_written_freework(struct freework *);
> +static void handle_written_freework(struct freework *, int);
> static void merge_inode_lists(struct allocdirectlst *,struct allocdirectlst *);
> static struct worklist *jnewblk_merge(struct worklist *, struct worklist *,
> struct workhead *);
> -static void setup_allocindir_phase2(struct buf *, struct inode *,
> +static struct freefrag *setup_allocindir_phase2(struct buf *, struct inode *,
> struct inodedep *, struct allocindir *, ufs_lbn_t);
> static struct allocindir *newallocindir(struct inode *, int, ufs2_daddr_t,
> ufs2_daddr_t, ufs_lbn_t);
> @@ -862,16 +892,20 @@ static int newblk_lookup(struct mount *,
> static int inodedep_find(struct inodedep_hashhead *, struct fs *, ino_t,
> struct inodedep **);
> static int inodedep_lookup(struct mount *, ino_t, int, struct inodedep **);
> -static int pagedep_lookup(struct mount *, ino_t, ufs_lbn_t, int,
> - struct pagedep **);
> +static int pagedep_lookup(struct mount *, struct buf *bp, ino_t, ufs_lbn_t,
> + int, struct pagedep **);
> static int pagedep_find(struct pagedep_hashhead *, ino_t, ufs_lbn_t,
> struct mount *mp, int, struct pagedep **);
> static void pause_timer(void *);
> static int request_cleanup(struct mount *, int);
> -static int process_worklist_item(struct mount *, int);
> +static int process_worklist_item(struct mount *, int, int);
> static void process_removes(struct vnode *);
> +static void process_truncates(struct vnode *);
> static void jwork_move(struct workhead *, struct workhead *);
> +static void jwork_insert(struct workhead *, struct jsegdep *);
> static void add_to_worklist(struct worklist *, int);
> +static void wake_worklist(struct worklist *);
> +static void wait_worklist(struct worklist *, char *);
> static void remove_from_worklist(struct worklist *);
> static void softdep_flush(void);
> static int softdep_speedup(void);
> @@ -889,17 +923,20 @@ static struct jremref *newjremref(struct
> struct inode *ip, off_t, nlink_t);
> static struct jaddref *newjaddref(struct inode *, ino_t, off_t, int16_t,
> uint16_t);
> -static inline void newinoref(struct inoref *, ino_t, ino_t, off_t, nlink_t,
> +static inline void newinoref(struct inoref *, ino_t, ino_t, off_t, nlink_t,
> uint16_t);
> -static inline struct jsegdep *inoref_jseg(struct inoref *);
> +static inline struct jsegdep *inoref_jseg(struct inoref *);
> static struct jmvref *newjmvref(struct inode *, ino_t, off_t, off_t);
> static struct jfreeblk *newjfreeblk(struct freeblks *, ufs_lbn_t,
> ufs2_daddr_t, int);
> +static struct jtrunc *newjtrunc(struct freeblks *, off_t, int);
> +static void move_newblock_dep(struct jaddref *, struct inodedep *);
> +static void cancel_jfreeblk(struct freeblks *, ufs2_daddr_t);
> static struct jfreefrag *newjfreefrag(struct freefrag *, struct inode *,
> ufs2_daddr_t, long, ufs_lbn_t);
> static struct freework *newfreework(struct ufsmount *, struct freeblks *,
> - struct freework *, ufs_lbn_t, ufs2_daddr_t, int, int);
> -static void jwait(struct worklist *wk);
> + struct freework *, ufs_lbn_t, ufs2_daddr_t, int, int, int);
> +static int jwait(struct worklist *, int);
> static struct inodedep *inodedep_lookup_ip(struct inode *);
> static int bmsafemap_rollbacks(struct bmsafemap *);
> static struct freefile *handle_bufwait(struct inodedep *, struct workhead *);
> @@ -1064,6 +1101,30 @@ jwork_move(dst, src)
> }
> }
>
> +static void
> +jwork_insert(dst, jsegdep)
> + struct workhead *dst;
> + struct jsegdep *jsegdep;
> +{
> + struct jsegdep *jsegdepn;
> + struct worklist *wk;
> +
> + LIST_FOREACH(wk, dst, wk_list)
> + if (wk->wk_type == D_JSEGDEP)
> + break;
> + if (wk == NULL) {
> + WORKLIST_INSERT(dst, &jsegdep->jd_list);
> + return;
> + }
> + jsegdepn = WK_JSEGDEP(wk);
> + if (jsegdep->jd_seg->js_seq < jsegdepn->jd_seg->js_seq) {
> + WORKLIST_REMOVE(wk);
> + free_jsegdep(jsegdepn);
> + WORKLIST_INSERT(dst, &jsegdep->jd_list);
> + } else
> + free_jsegdep(jsegdep);
> +}
> +
> /*
> * Routines for tracking and managing workitems.
> */
> @@ -1088,6 +1149,8 @@ workitem_free(item, type)
> panic("workitem_free: type mismatch %s != %s",
> TYPENAME(item->wk_type), TYPENAME(type));
> #endif
> + if (item->wk_state & IOWAITING)
> + wakeup(item);
> ump = VFSTOUFS(item->wk_mp);
> if (--ump->softdep_deps == 0 && ump->softdep_req)
> wakeup(&ump->softdep_deps);
> @@ -1101,14 +1164,18 @@ workitem_alloc(item, type, mp)
> int type;
> struct mount *mp;
> {
> + struct ufsmount *ump;
> +
> item->wk_type = type;
> item->wk_mp = mp;
> item->wk_state = 0;
> +
> + ump = VFSTOUFS(mp);
> ACQUIRE_LOCK(&lk);
> dep_current[type]++;
> dep_total[type]++;
> - VFSTOUFS(mp)->softdep_deps++;
> - VFSTOUFS(mp)->softdep_accdeps++;
> + ump->softdep_deps++;
> + ump->softdep_accdeps++;
> FREE_LOCK(&lk);
> }
>
> @@ -1270,8 +1337,7 @@ softdep_flush(void)
> vfslocked = VFS_LOCK_GIANT(mp);
> progress += softdep_process_worklist(mp, 0);
> ump = VFSTOUFS(mp);
> - remaining += ump->softdep_on_worklist -
> - ump->softdep_on_worklist_inprogress;
> + remaining += ump->softdep_on_worklist;
> VFS_UNLOCK_GIANT(vfslocked);
> mtx_lock(&mountlist_mtx);
> nmp = TAILQ_NEXT(mp, mnt_list);
> @@ -1314,10 +1380,14 @@ softdep_speedup(void)
> * The following routine is the only one that removes items
> * and does so in order from first to last.
> */
> +
> +#define WK_HEAD 0x0001 /* Add to HEAD. */
> +#define WK_NODELAY 0x0002 /* Process immediately. */
> +
> static void
> -add_to_worklist(wk, nodelay)
> +add_to_worklist(wk, flags)
> struct worklist *wk;
> - int nodelay;
> + int flags;
> {
> struct ufsmount *ump;
>
> @@ -1327,13 +1397,17 @@ add_to_worklist(wk, nodelay)
> panic("add_to_worklist: %s(0x%X) already on list",
> TYPENAME(wk->wk_type), wk->wk_state);
> wk->wk_state |= ONWORKLIST;
> - if (LIST_EMPTY(&ump->softdep_workitem_pending))
> + if (ump->softdep_on_worklist == 0) {
> LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list);
> - else
> + ump->softdep_worklist_tail = wk;
> + } else if (flags & WK_HEAD) {
> + LIST_INSERT_HEAD(&ump->softdep_workitem_pending, wk, wk_list);
> + } else {
> LIST_INSERT_AFTER(ump->softdep_worklist_tail, wk, wk_list);
>
> *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
>
More information about the svn-src-head
mailing list