| Message ID | 20190830102411.519-2-hch@lst.de (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [1/3] xfs: add a xfs_valid_startblock helper | expand |
On Fri, Aug 30, 2019 at 12:24:09PM +0200, Christoph Hellwig wrote: > Add a helper that validates the startblock is valid. This checks for a > non-zero block on the main device, but skips that check for blocks on > the realtime device. > > Signed-off-by: Christoph Hellwig <hch@lst.de> > --- > fs/xfs/libxfs/xfs_bmap.c | 2 +- > fs/xfs/libxfs/xfs_bmap.h | 3 +++ > fs/xfs/xfs_iomap.c | 6 +++--- > 3 files changed, 7 insertions(+), 4 deletions(-) > > diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c > index 05aedf4a538c..80b25e21e708 100644 > --- a/fs/xfs/libxfs/xfs_bmap.c > +++ b/fs/xfs/libxfs/xfs_bmap.c > @@ -4519,7 +4519,7 @@ xfs_bmapi_convert_delalloc( > if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK)) > goto out_finish; > error = -EFSCORRUPTED; > - if (WARN_ON_ONCE(!bma.got.br_startblock && !XFS_IS_REALTIME_INODE(ip))) > + if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock))) > goto out_finish; > > XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length)); > diff --git a/fs/xfs/libxfs/xfs_bmap.h b/fs/xfs/libxfs/xfs_bmap.h > index c409871a096e..7efa56e8750f 100644 > --- a/fs/xfs/libxfs/xfs_bmap.h > +++ b/fs/xfs/libxfs/xfs_bmap.h > @@ -171,6 +171,9 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec) > !isnullstartblock(irec->br_startblock); > } > > +#define xfs_valid_startblock(ip, startblock) \ > + ((startblock) != 0 || XFS_IS_REALTIME_INODE(ip)) We have more robust validators for data/rtdev fsblock_t, so why not: #define xfs_valid_startblock(ip, startblock) \ (XFS_IS_REALTIME_INODE(ip) ? xfs_verify_rtbno(startblock) : \ xfs_verify_fsbno(startblock)) and why not make it a static inline function too? --D > + > void xfs_trim_extent(struct xfs_bmbt_irec *irec, xfs_fileoff_t bno, > xfs_filblks_t len); > int xfs_bmap_add_attrfork(struct xfs_inode *ip, int size, int rsvd); > diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c > index 3a4310d7cb59..f780e223b118 100644 > --- a/fs/xfs/xfs_iomap.c > +++ b/fs/xfs/xfs_iomap.c > @@ -58,7 +58,7 @@ xfs_bmbt_to_iomap( > { > struct xfs_mount *mp = ip->i_mount; > > - if (unlikely(!imap->br_startblock && !XFS_IS_REALTIME_INODE(ip))) > + if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) > return xfs_alert_fsblock_zero(ip, imap); > > if (imap->br_startblock == HOLESTARTBLOCK) { > @@ -297,7 +297,7 @@ xfs_iomap_write_direct( > goto out_unlock; > } > > - if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) > + if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) > error = xfs_alert_fsblock_zero(ip, imap); > > out_unlock: > @@ -814,7 +814,7 @@ xfs_iomap_write_unwritten( > if (error) > return error; > > - if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) > + if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) > return xfs_alert_fsblock_zero(ip, &imap); > > if ((numblks_fsb = imap.br_blockcount) == 0) { > -- > 2.20.1 >
On Fri, Aug 30, 2019 at 08:06:50AM -0700, Darrick J. Wong wrote: > > --- a/fs/xfs/libxfs/xfs_bmap.h > > +++ b/fs/xfs/libxfs/xfs_bmap.h > > @@ -171,6 +171,9 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec) > > !isnullstartblock(irec->br_startblock); > > } > > > > +#define xfs_valid_startblock(ip, startblock) \ > > + ((startblock) != 0 || XFS_IS_REALTIME_INODE(ip)) > > We have more robust validators for data/rtdev fsblock_t, so why not: > > #define xfs_valid_startblock(ip, startblock) \ > (XFS_IS_REALTIME_INODE(ip) ? xfs_verify_rtbno(startblock) : \ > xfs_verify_fsbno(startblock)) > > and why not make it a static inline function too? I tried an inline function, but I could not find a header to place it that would actually easily compile everywhere... Maybe we should just make that a xfs_verify_bno(mp, startblock) and move that out of line such in a way that a smart compiler avoids the function call overhead for xfs_verify_rtbno / xfs_verify_fsbno. I'll take another stab at this.
On Fri, Aug 30, 2019 at 05:32:53PM +0200, Christoph Hellwig wrote: > On Fri, Aug 30, 2019 at 08:06:50AM -0700, Darrick J. Wong wrote: > > > --- a/fs/xfs/libxfs/xfs_bmap.h > > > +++ b/fs/xfs/libxfs/xfs_bmap.h > > > @@ -171,6 +171,9 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec) > > > !isnullstartblock(irec->br_startblock); > > > } > > > > > > +#define xfs_valid_startblock(ip, startblock) \ > > > + ((startblock) != 0 || XFS_IS_REALTIME_INODE(ip)) > > > > We have more robust validators for data/rtdev fsblock_t, so why not: > > > > #define xfs_valid_startblock(ip, startblock) \ > > (XFS_IS_REALTIME_INODE(ip) ? xfs_verify_rtbno(startblock) : \ > > xfs_verify_fsbno(startblock)) > > > > and why not make it a static inline function too? > > I tried an inline function, but I could not find a header to place > it that would actually easily compile everywhere... Maybe we should > just make that a xfs_verify_bno(mp, startblock) and move that out of > line such in a way that a smart compiler avoids the function call > overhead for xfs_verify_rtbno / xfs_verify_fsbno. I'll take another > stab at this. So I looked into your suggestion, but xfs_verify_rtbno / xfs_verify_fsbno do a lot of validity checking, but they don't actually contain the check that was in the existing code. The bmap code just checks that there is a startblock of 0 for non-rt devices, probably this was added to find some old bug where a irec structure that was zeroed was returned. So replacing it with xfs_verify_rtbno / xfs_verify_fsbno would not help in any way. But the big question is if keeping the 0 check is even worth it.
On Sun, Sep 01, 2019 at 09:36:34AM +0200, Christoph Hellwig wrote: > On Fri, Aug 30, 2019 at 05:32:53PM +0200, Christoph Hellwig wrote: > > On Fri, Aug 30, 2019 at 08:06:50AM -0700, Darrick J. Wong wrote: > > > > --- a/fs/xfs/libxfs/xfs_bmap.h > > > > +++ b/fs/xfs/libxfs/xfs_bmap.h > > > > @@ -171,6 +171,9 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec) > > > > !isnullstartblock(irec->br_startblock); > > > > } > > > > > > > > +#define xfs_valid_startblock(ip, startblock) \ > > > > + ((startblock) != 0 || XFS_IS_REALTIME_INODE(ip)) > > > > > > We have more robust validators for data/rtdev fsblock_t, so why not: > > > > > > #define xfs_valid_startblock(ip, startblock) \ > > > (XFS_IS_REALTIME_INODE(ip) ? xfs_verify_rtbno(startblock) : \ > > > xfs_verify_fsbno(startblock)) > > > > > > and why not make it a static inline function too? > > > > I tried an inline function, but I could not find a header to place > > it that would actually easily compile everywhere... Maybe we should > > just make that a xfs_verify_bno(mp, startblock) and move that out of > > line such in a way that a smart compiler avoids the function call > > overhead for xfs_verify_rtbno / xfs_verify_fsbno. I'll take another > > stab at this. > > So I looked into your suggestion, but xfs_verify_rtbno / xfs_verify_fsbno > do a lot of validity checking, but they don't actually contain the > check that was in the existing code. The bmap code just checks that > there is a startblock of 0 for non-rt devices, probably this was added > to find some old bug where a irec structure that was zeroed was returned. > > So replacing it with xfs_verify_rtbno / xfs_verify_fsbno would not help > in any way. But the big question is if keeping the 0 check is even > worth it. It's been mildly helpful for noticing when my online/offline repair prototype code totally screws up, but at that point so much magic smoke is already pouring out everywhere that it's hard not to notice. :) --D
On Sun, Sep 01, 2019 at 01:31:40PM -0700, Darrick J. Wong wrote: > It's been mildly helpful for noticing when my online/offline repair > prototype code totally screws up, but at that point so much magic smoke > is already pouring out everywhere that it's hard not to notice. :) That suggests to just keep the macro as I submitted it, maybe with a big fat comment explaining the usage.
On Mon, Sep 02, 2019 at 09:59:46AM +0200, Christoph Hellwig wrote: > On Sun, Sep 01, 2019 at 01:31:40PM -0700, Darrick J. Wong wrote: > > It's been mildly helpful for noticing when my online/offline repair > > prototype code totally screws up, but at that point so much magic smoke > > is already pouring out everywhere that it's hard not to notice. :) > > That suggests to just keep the macro as I submitted it, maybe with > a big fat comment explaining the usage. Ok. Do you want to resubmit with a comment of your choosing, or let me write in whatever: /* * Check the mapping for obviously garbage allocations that could trash * the filesystem immediately. */ ? --D
On Mon, Sep 02, 2019 at 10:04:40AM -0700, Darrick J. Wong wrote: > On Mon, Sep 02, 2019 at 09:59:46AM +0200, Christoph Hellwig wrote: > > On Sun, Sep 01, 2019 at 01:31:40PM -0700, Darrick J. Wong wrote: > > > It's been mildly helpful for noticing when my online/offline repair > > > prototype code totally screws up, but at that point so much magic smoke > > > is already pouring out everywhere that it's hard not to notice. :) > > > > That suggests to just keep the macro as I submitted it, maybe with > > a big fat comment explaining the usage. > > Ok. Do you want to resubmit with a comment of your choosing, or let me > write in whatever: > > /* > * Check the mapping for obviously garbage allocations that could trash > * the filesystem immediately. > */ I was going to resend it, but now that you've written the comment for me feel free to just apply it with that added.
diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c index 05aedf4a538c..80b25e21e708 100644 --- a/fs/xfs/libxfs/xfs_bmap.c +++ b/fs/xfs/libxfs/xfs_bmap.c @@ -4519,7 +4519,7 @@ xfs_bmapi_convert_delalloc( if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK)) goto out_finish; error = -EFSCORRUPTED; - if (WARN_ON_ONCE(!bma.got.br_startblock && !XFS_IS_REALTIME_INODE(ip))) + if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock))) goto out_finish; XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length)); diff --git a/fs/xfs/libxfs/xfs_bmap.h b/fs/xfs/libxfs/xfs_bmap.h index c409871a096e..7efa56e8750f 100644 --- a/fs/xfs/libxfs/xfs_bmap.h +++ b/fs/xfs/libxfs/xfs_bmap.h @@ -171,6 +171,9 @@ static inline bool xfs_bmap_is_real_extent(struct xfs_bmbt_irec *irec) !isnullstartblock(irec->br_startblock); } +#define xfs_valid_startblock(ip, startblock) \ + ((startblock) != 0 || XFS_IS_REALTIME_INODE(ip)) + void xfs_trim_extent(struct xfs_bmbt_irec *irec, xfs_fileoff_t bno, xfs_filblks_t len); int xfs_bmap_add_attrfork(struct xfs_inode *ip, int size, int rsvd); diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index 3a4310d7cb59..f780e223b118 100644 --- a/fs/xfs/xfs_iomap.c +++ b/fs/xfs/xfs_iomap.c @@ -58,7 +58,7 @@ xfs_bmbt_to_iomap( { struct xfs_mount *mp = ip->i_mount; - if (unlikely(!imap->br_startblock && !XFS_IS_REALTIME_INODE(ip))) + if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) return xfs_alert_fsblock_zero(ip, imap); if (imap->br_startblock == HOLESTARTBLOCK) { @@ -297,7 +297,7 @@ xfs_iomap_write_direct( goto out_unlock; } - if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) + if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) error = xfs_alert_fsblock_zero(ip, imap); out_unlock: @@ -814,7 +814,7 @@ xfs_iomap_write_unwritten( if (error) return error; - if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) + if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) return xfs_alert_fsblock_zero(ip, &imap); if ((numblks_fsb = imap.br_blockcount) == 0) {
Add a helper that validates the startblock is valid. This checks for a non-zero block on the main device, but skips that check for blocks on the realtime device. Signed-off-by: Christoph Hellwig <hch@lst.de> --- fs/xfs/libxfs/xfs_bmap.c | 2 +- fs/xfs/libxfs/xfs_bmap.h | 3 +++ fs/xfs/xfs_iomap.c | 6 +++--- 3 files changed, 7 insertions(+), 4 deletions(-)