xfs_db: release ip resource before returning from get_next_unlinked()
PATCH v3 1/1 xfs_db: release ip resource before returning from get_next_unlinked()
Fix potential memory leak in function get_next_unlinked(). Call libxfs_irele(ip) before exiting.
Details:
Error: RESOURCE_LEAK (CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2: alloc_arg: "libxfs_iget" allocates memory that is stored into "ip". xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp". xfsprogs-6.5.0/db/iunlink.c:76:2: leaked_storage: Variable "ip" going out of scope leaks the storage it points to.
# 74| libxfs_buf_relse(ino_bp);
# 75|
# 76|-> return ret;
# 77| bad:
# 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));---
v2: cover error case.
v3: fix coverage to not release unitialized variable.
--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
db/iunlink.c +5 -2
@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp); goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);return ret; bad: +libxfs_irele (ip);dbprintf (_ ("AG %u agino %u: %s\n" ), agno, agino,strerror (error));return NULLAGINO ; }
Signature · 1 line
2.45.2
Looks good:
Quoted 27 lines · Bill O'Donnell
Fix potential memory leak in function
get_next_unlinked(). Calllibxfs_irele(ip)before exiting.Details:
Error:RESOURCE_LEAK(CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2:alloc_arg: "libxfs_iget" allocates memory that is stored into "ip".xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp".xfsprogs-6.5.0/db/iunlink.c:76:2:leaked_storage: Variable "ip" going out of scope leaks the storage it points to.# 74| libxfs_buf_relse(ino_bp); # 75| # 76|-> return ret; # 77| bad: # 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));Signed-off-by Bill O'Donnell <[email protected]>---
v2: cover error case.
v3: fix coverage to not release unitialized variable.--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)db/iunlink.c +0
-0
db/iunlink.c get_next_unlinked( · line 70
Quoted 6 lines
db/iunlink.c +2
-1@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp);
Sorry, I think I've led you astray -- it's not necessary to libxfs_buf_relse in any of the bailouts.
--D
Quoted 18 lines
db/iunlink.c +3
-1goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);return ret; bad: +libxfs_irele (ip);dbprintf (_ ("AG %u agino %u: %s\n" ), agno, agino,strerror (error));return NULLAGINO ; }Signature · 1 line
2.45.2
Quoted 48 lines · Darrick J. Wong
Quoted 27 lines · Bill O'Donnell
Fix potential memory leak in function
get_next_unlinked(). Calllibxfs_irele(ip)before exiting.Details:
Error:RESOURCE_LEAK(CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2:alloc_arg: "libxfs_iget" allocates memory that is stored into "ip".xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp".xfsprogs-6.5.0/db/iunlink.c:76:2:leaked_storage: Variable "ip" going out of scope leaks the storage it points to.# 74| libxfs_buf_relse(ino_bp); # 75| # 76|-> return ret; # 77| bad: # 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));Signed-off-by Bill O'Donnell <[email protected]>---
v2: cover error case.
v3: fix coverage to not release unitialized variable.--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)db/iunlink.c +0
-0db/iunlink.c get_next_unlinked( · line 70
Quoted 6 lines
db/iunlink.c +2
-1@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp);Sorry, I think I've led you astray -- it's not necessary to
libxfs_buf_relsein any of the bailouts.--D
Quoted 8 lines
db/iunlink.c +2
-1goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);
Quoted 3 lines
db/iunlink.c get_next_unlinked( · line 80
Quoted 3 lines
db/iunlink.c +1
-0return ret; bad: +libxfs_irele (ip);
And this addition results in a libxfs_irele of an ip() which failed iget() via the first goto bad, so you're releasing a thing which was never obtained, which doesn't make sense.
There are 2 relevant actions here. The libxfs_iget, and the libxfs_imap_to_bp. Only after libxfs_iget(ip) /succeeds/ does it need a libxfs_irele(ip), on either error paths or normal exit. The fact that it does neither leads to the two leaks noted in CID 1554242.
libxfs_imap_to_bp needs a corresponding libxfs_buf_relse() (thanks for clarifying djwong) but that libxfs_buf_relse() is already present if libxfs_imap_to_bp succeeds. It's not needed if it fails, because there's nothing to release.
When Darrick said
Quoted 2 lines
I think this needs to cover the error return for
libxfs_imap_to_bptoo, doesn't it?
I think he meant that in the error case where libxfs_imap_to_bp fails, libxfs_irele is also needed. (In addition to being needed on a normal return.)
-Eric
Quoted 58 lines · Eric Sandeen
Quoted 48 lines · Darrick J. Wong
Quoted 27 lines · Bill O'Donnell
Fix potential memory leak in function
get_next_unlinked(). Calllibxfs_irele(ip)before exiting.Details:
Error:RESOURCE_LEAK(CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2:alloc_arg: "libxfs_iget" allocates memory that is stored into "ip".xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp".xfsprogs-6.5.0/db/iunlink.c:76:2:leaked_storage: Variable "ip" going out of scope leaks the storage it points to.# 74| libxfs_buf_relse(ino_bp); # 75| # 76|-> return ret; # 77| bad: # 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));Signed-off-by Bill O'Donnell <[email protected]>---
v2: cover error case.
v3: fix coverage to not release unitialized variable.--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)db/iunlink.c +0
-0db/iunlink.c get_next_unlinked( · line 70
Quoted 6 lines
db/iunlink.c +2
-1@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp);Sorry, I think I've led you astray -- it's not necessary to
libxfs_buf_relsein any of the bailouts.--D
Quoted 8 lines
db/iunlink.c +2
-1goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);Quoted 3 lines
db/iunlink.c get_next_unlinked( · line 80
Quoted 3 lines
db/iunlink.c +1
-0return ret; bad: +libxfs_irele (ip);And this addition results in a
libxfs_ireleof anip()which failediget()via the first goto bad, so you're releasing a thing which was never obtained, which doesn't make sense.
Quoted 4 lines
There are 2 relevant actions here. The
libxfs_iget, and thelibxfs_imap_to_bp. Only afterlibxfs_iget(ip)/succeeds/ does it need alibxfs_irele(ip), on either error paths or normal exit. The fact that it does neither leads to the two leaks noted in CID 1554242.
In libxfs_iget, -ENOMEM is returned when kmem_cache_zalloc() fails. For all other error cases in that function, kmem_cache_free() releases the memory that was presumably successfully allocated. I had wondered if we need to use libxfs_irele() at all in get_next_unlinked() (except for the success case)?
Quoted 14 lines
libxfs_imap_to_bpneeds a correspondinglibxfs_buf_relse()(thanks for clarifying djwong) but thatlibxfs_buf_relse()is already present iflibxfs_imap_to_bpsucceeds. It's not needed if it fails, because there's nothing to release.When Darrick said
Quoted 2 lines
I think this needs to cover the error return for
libxfs_imap_to_bptoo, doesn't it?I think he meant that in the error case where
libxfs_imap_to_bpfails,libxfs_ireleis also needed. (In addition to being needed on a normal return.)-Eric
Quoted 64 lines · Bill O'Donnell
Quoted 58 lines · Eric Sandeen
Quoted 48 lines · Darrick J. Wong
Quoted 27 lines · Bill O'Donnell
Fix potential memory leak in function
get_next_unlinked(). Calllibxfs_irele(ip)before exiting.Details:
Error:RESOURCE_LEAK(CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2:alloc_arg: "libxfs_iget" allocates memory that is stored into "ip".xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp".xfsprogs-6.5.0/db/iunlink.c:76:2:leaked_storage: Variable "ip" going out of scope leaks the storage it points to.# 74| libxfs_buf_relse(ino_bp); # 75| # 76|-> return ret; # 77| bad: # 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));Signed-off-by Bill O'Donnell <[email protected]>---
v2: cover error case.
v3: fix coverage to not release unitialized variable.--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)db/iunlink.c +0
-0db/iunlink.c get_next_unlinked( · line 70
Quoted 6 lines
db/iunlink.c +2
-1@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp);Sorry, I think I've led you astray -- it's not necessary to
libxfs_buf_relsein any of the bailouts.--D
Quoted 8 lines
db/iunlink.c +2
-1goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);Quoted 3 lines
db/iunlink.c get_next_unlinked( · line 80
Quoted 3 lines
db/iunlink.c +1
-0return ret; bad: +libxfs_irele (ip);And this addition results in a
libxfs_ireleof anip()which failediget()via the first goto bad, so you're releasing a thing which was never obtained, which doesn't make sense.Quoted 4 lines
There are 2 relevant actions here. The
libxfs_iget, and thelibxfs_imap_to_bp. Only afterlibxfs_iget(ip)/succeeds/ does it need alibxfs_irele(ip), on either error paths or normal exit. The fact that it does neither leads to the two leaks noted in CID 1554242.
Quoted 4 lines
In
libxfs_iget,-ENOMEMis returned whenkmem_cache_zalloc()fails. For all other error cases in that function,kmem_cache_free()releases the memory that was presumably successfully allocated. I had wondered if we need to uselibxfs_irele()at all inget_next_unlinked()(except for the success case)?
So, if that's the case, I'm back to v1 of this patch. -Bill
Quoted 18 lines
Quoted 16 lines
libxfs_imap_to_bpneeds a correspondinglibxfs_buf_relse()(thanks for clarifying djwong) but thatlibxfs_buf_relse()is already present iflibxfs_imap_to_bpsucceeds. It's not needed if it fails, because there's nothing to release.When Darrick said
Quoted 2 lines
I think this needs to cover the error return for
libxfs_imap_to_bptoo, doesn't it?I think he meant that in the error case where
libxfs_imap_to_bpfails,libxfs_ireleis also needed. (In addition to being needed on a normal return.)-Eric
Quoted 65 lines · Bill O'Donnell
Quoted 64 lines · Bill O'Donnell
Quoted 62 lines · Eric Sandeen
Quoted 48 lines · Darrick J. Wong
Quoted 27 lines · Bill O'Donnell
Fix potential memory leak in function
get_next_unlinked(). Calllibxfs_irele(ip)before exiting.Details:
Error:RESOURCE_LEAK(CWE-772):xfsprogs-6.5.0/db/iunlink.c:51:2:alloc_arg: "libxfs_iget" allocates memory that is stored into "ip".xfsprogs-6.5.0/db/iunlink.c:68:2: noescape: Resource "&ip->i_imap" is not freed or pointed-to in "libxfs_imap_to_bp".xfsprogs-6.5.0/db/iunlink.c:76:2:leaked_storage: Variable "ip" going out of scope leaks the storage it points to.# 74| libxfs_buf_relse(ino_bp); # 75| # 76|-> return ret; # 77| bad: # 78| dbprintf(_("AG %u agino %u: %s\n"), agno, agino, strerror(error));Signed-off-by Bill O'Donnell <[email protected]>---
v2: cover error case.
v3: fix coverage to not release unitialized variable.--- db/iunlink.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)db/iunlink.c +0
-0db/iunlink.c get_next_unlinked( · line 70
Quoted 6 lines
db/iunlink.c +2
-1@@ -66,15 +66,18 @@ get_next_unlinked( } error = -libxfs_imap_to_bp (mp,NULL , &ip->i_imap, &ino_bp); -if (error) +if (error) { +libxfs_buf_relse (ino_bp);Sorry, I think I've led you astray -- it's not necessary to
libxfs_buf_relsein any of the bailouts.--D
Quoted 8 lines
db/iunlink.c +2
-1goto bad; - + } dip =xfs_buf_offset (ino_bp, ip->i_imap.im_boffset); ret =be32_to_cpu (dip->di_next_unlinked);libxfs_buf_relse (ino_bp); +libxfs_irele (ip);Quoted 3 lines
db/iunlink.c get_next_unlinked( · line 80
Quoted 3 lines
db/iunlink.c +1
-0return ret; bad: +libxfs_irele (ip);And this addition results in a
libxfs_ireleof anip()which failediget()via the first goto bad, so you're releasing a thing which was never obtained, which doesn't make sense.There are 2 relevant actions here. The
libxfs_iget, and thelibxfs_imap_to_bp. Only afterlibxfs_iget(ip)/succeeds/ does it need alibxfs_irele(ip), on either error paths or normal exit. The fact that it does neither leads to the two leaks noted in CID 1554242.
Quoted 7 lines
Quoted 4 lines
In
libxfs_iget,-ENOMEMis returned whenkmem_cache_zalloc()fails. For all other error cases in that function,kmem_cache_free()releases the memory that was presumably successfully allocated. I had wondered if we need to uselibxfs_irele()at all inget_next_unlinked()(except for the success case)?So, if that's the case, I'm back to v1 of this patch. -Bill
when libxfs_iget succeeds, it has obtained an inode.
After that has happened, libxfs_irele needs to be called either in the normal return path, or any subsequent error path, to free that inode in this function.
As Darrick pointed out in his first reply, V1 missed irele on the error path, so it was not sufficient.
-Eric