| Message ID | 2cafa0607ca171ebd00ac6c7e073b46808e24f00.1640537669.git.christophe.jaillet@wanadoo.fr (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fsi: Aspeed: Fix a potential double free | expand |
On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: > 'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or > there will be a double free(). A struct device can never be devm_alloced for obvious reasons. Perhaps that is the real problem here? > Remove the 'release' function that is wrong and unneeded. > > Fixes: 606397d67f41 ("fsi: Add ast2600 master driver") > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > --- > This patch is completely theoretical. It looks good to me, but there is a > little too much indirections for me. I'm also not that familiar with > fixing issue related to 'release' function... > > ... So review with care :) > --- > drivers/fsi/fsi-master-aspeed.c | 9 --------- > 1 file changed, 9 deletions(-) > > diff --git a/drivers/fsi/fsi-master-aspeed.c b/drivers/fsi/fsi-master-aspeed.c > index 8606e55c1721..4a745ccb60cf 100644 > --- a/drivers/fsi/fsi-master-aspeed.c > +++ b/drivers/fsi/fsi-master-aspeed.c > @@ -373,14 +373,6 @@ static int aspeed_master_break(struct fsi_master *master, int link) > return aspeed_master_write(master, link, 0, addr, &cmd, 4); > } > > -static void aspeed_master_release(struct device *dev) > -{ > - struct fsi_master_aspeed *aspeed = > - to_fsi_master_aspeed(dev_to_fsi_master(dev)); > - > - kfree(aspeed); > -} > - > /* mmode encoders */ > static inline u32 fsi_mmode_crs0(u32 x) > { > @@ -603,7 +595,6 @@ static int fsi_master_aspeed_probe(struct platform_device *pdev) > dev_info(&pdev->dev, "hub version %08x (%d links)\n", reg, links); > > aspeed->master.dev.parent = &pdev->dev; > - aspeed->master.dev.release = aspeed_master_release; Odd, then what deletes this device structure when the release function wants to be called? You should have gotten a big warning from the kernel when removing the device from the system at runtime, did you test this somehow? This does not look correct at all. greg k-h
Le 27/12/2021 à 07:29, Greg KH a écrit : > On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: >> 'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or >> there will be a double free(). > > A struct device can never be devm_alloced for obvious reasons. Perhaps > that is the real problem here? Thanks for the feed-back. This goes beyond my knowledge of how this should work. As I can not test myself, I won't be of any help. I'll let you or anyone else check if something needs to be fixed, and how to fix it properly. Just take my patch as a "Hey! Looks strange to have a kfree() in a driver that only call devm_kzalloc() to allocate memory. S.o. should give a deeper look at it". :) CJ > >> Remove the 'release' function that is wrong and unneeded. >> >> Fixes: 606397d67f41 ("fsi: Add ast2600 master driver") >> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> >> --- >> This patch is completely theoretical. It looks good to me, but there is a >> little too much indirections for me. I'm also not that familiar with >> fixing issue related to 'release' function... >> >> ... So review with care :) >> --- >> drivers/fsi/fsi-master-aspeed.c | 9 --------- >> 1 file changed, 9 deletions(-) >> >> diff --git a/drivers/fsi/fsi-master-aspeed.c b/drivers/fsi/fsi-master-aspeed.c >> index 8606e55c1721..4a745ccb60cf 100644 >> --- a/drivers/fsi/fsi-master-aspeed.c >> +++ b/drivers/fsi/fsi-master-aspeed.c >> @@ -373,14 +373,6 @@ static int aspeed_master_break(struct fsi_master *master, int link) >> return aspeed_master_write(master, link, 0, addr, &cmd, 4); >> } >> >> -static void aspeed_master_release(struct device *dev) >> -{ >> - struct fsi_master_aspeed *aspeed = >> - to_fsi_master_aspeed(dev_to_fsi_master(dev)); >> - >> - kfree(aspeed); >> -} >> - >> /* mmode encoders */ >> static inline u32 fsi_mmode_crs0(u32 x) >> { >> @@ -603,7 +595,6 @@ static int fsi_master_aspeed_probe(struct platform_device *pdev) >> dev_info(&pdev->dev, "hub version %08x (%d links)\n", reg, links); >> >> aspeed->master.dev.parent = &pdev->dev; >> - aspeed->master.dev.release = aspeed_master_release; > > Odd, then what deletes this device structure when the release function > wants to be called? You should have gotten a big warning from the > kernel when removing the device from the system at runtime, did you test > this somehow? > > This does not look correct at all. > > greg k-h >
On Mon, Dec 27, 2021 at 07:29:07AM +0100, Greg KH wrote: > On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: > > 'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or > > there will be a double free(). > > A struct device can never be devm_alloced for obvious reasons. Perhaps > that is the real problem here? > I don't understand how "aspeed" is a struct device. I've been working on understanding device managed memory recently for Smatch. It's really complicated. There are a bunch of rules/heuristics that I'm slowly creating to generate new warnings but I'm a long way from understanding it well myself. regards, dan carpenter
On 1/6/22 12:14 AM, Dan Carpenter wrote: > On Mon, Dec 27, 2021 at 07:29:07AM +0100, Greg KH wrote: >> On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: >>> 'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or >>> there will be a double free(). >> >> A struct device can never be devm_alloced for obvious reasons. Perhaps >> that is the real problem here? >> > > I don't understand how "aspeed" is a struct device. > -static void aspeed_master_release(struct device *dev) -{ - struct fsi_master_aspeed *aspeed = - to_fsi_master_aspeed(dev_to_fsi_master(dev)); - - kfree(aspeed); -} So "dev" is embedded in struct fsi_master, and struct fsi_master is embedded in struct fsi_master_aspeed. Since "struct device" is embedded, the data structure embedding it must be released with the release function, as is done here. The problem is indeed that the data structure is allocated with devm_kzalloc(), which as Greg points out must not be devm_ allocated (because its lifetime does not match the lifetime of devm_ allocated memory). > I've been working on understanding device managed memory recently for > Smatch. It's really complicated. There are a bunch of rules/heuristics > that I'm slowly creating to generate new warnings but I'm a long way > from understanding it well myself. > A data structure embedding struct device must not be devm_ allocated, and it must be released with the release callback. Maybe there is a means to flag that somehow ? Guenter
Le 06/01/2022 à 18:25, Guenter Roeck a écrit : > On 1/6/22 12:14 AM, Dan Carpenter wrote: >> On Mon, Dec 27, 2021 at 07:29:07AM +0100, Greg KH wrote: >>> On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: >>>> 'aspeed' is a devm_alloc'ed, so there is no need to free it >>>> explicitly or >>>> there will be a double free(). >>> >>> A struct device can never be devm_alloced for obvious reasons. Perhaps >>> that is the real problem here? >>> >> >> I don't understand how "aspeed" is a struct device. >> > > -static void aspeed_master_release(struct device *dev) > -{ > - struct fsi_master_aspeed *aspeed = > - to_fsi_master_aspeed(dev_to_fsi_master(dev)); > - > - kfree(aspeed); > -} > > So "dev" is embedded in struct fsi_master, and struct fsi_master is > embedded > in struct fsi_master_aspeed. Since "struct device" is embedded, the data > structure embedding it must be released with the release function, as is > done > here. The problem is indeed that the data structure is allocated with > devm_kzalloc(), which as Greg points out must not be devm_ allocated > (because its lifetime does not match the lifetime of devm_ allocated > memory). Thanks a lot for the detailed explanation. Crystal clear for me now. Do you want me to send a patch to remove the devm_ or will you? CJ > >> I've been working on understanding device managed memory recently for >> Smatch. It's really complicated. There are a bunch of rules/heuristics >> that I'm slowly creating to generate new warnings but I'm a long way >> from understanding it well myself. >> > > A data structure embedding struct device must not be devm_ allocated, > and it must be released with the release callback. Maybe there is > a means to flag that somehow ? > > Guenter >
On 1/6/22 10:35 AM, Christophe JAILLET wrote: > Le 06/01/2022 à 18:25, Guenter Roeck a écrit : >> On 1/6/22 12:14 AM, Dan Carpenter wrote: >>> On Mon, Dec 27, 2021 at 07:29:07AM +0100, Greg KH wrote: >>>> On Sun, Dec 26, 2021 at 05:56:02PM +0100, Christophe JAILLET wrote: >>>>> 'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or >>>>> there will be a double free(). >>>> >>>> A struct device can never be devm_alloced for obvious reasons. Perhaps >>>> that is the real problem here? >>>> >>> >>> I don't understand how "aspeed" is a struct device. >>> >> >> -static void aspeed_master_release(struct device *dev) >> -{ >> - struct fsi_master_aspeed *aspeed = >> - to_fsi_master_aspeed(dev_to_fsi_master(dev)); >> - >> - kfree(aspeed); >> -} >> >> So "dev" is embedded in struct fsi_master, and struct fsi_master is embedded >> in struct fsi_master_aspeed. Since "struct device" is embedded, the data >> structure embedding it must be released with the release function, as is done >> here. The problem is indeed that the data structure is allocated with >> devm_kzalloc(), which as Greg points out must not be devm_ allocated >> (because its lifetime does not match the lifetime of devm_ allocated >> memory). > > Thanks a lot for the detailed explanation. > Crystal clear for me now. > > Do you want me to send a patch to remove the devm_ or will you? > Sorry, I am way behind with code reviews. I won't have time to submit a patch. Guenter
diff --git a/drivers/fsi/fsi-master-aspeed.c b/drivers/fsi/fsi-master-aspeed.c index 8606e55c1721..4a745ccb60cf 100644 --- a/drivers/fsi/fsi-master-aspeed.c +++ b/drivers/fsi/fsi-master-aspeed.c @@ -373,14 +373,6 @@ static int aspeed_master_break(struct fsi_master *master, int link) return aspeed_master_write(master, link, 0, addr, &cmd, 4); } -static void aspeed_master_release(struct device *dev) -{ - struct fsi_master_aspeed *aspeed = - to_fsi_master_aspeed(dev_to_fsi_master(dev)); - - kfree(aspeed); -} - /* mmode encoders */ static inline u32 fsi_mmode_crs0(u32 x) { @@ -603,7 +595,6 @@ static int fsi_master_aspeed_probe(struct platform_device *pdev) dev_info(&pdev->dev, "hub version %08x (%d links)\n", reg, links); aspeed->master.dev.parent = &pdev->dev; - aspeed->master.dev.release = aspeed_master_release; aspeed->master.dev.of_node = of_node_get(dev_of_node(&pdev->dev)); aspeed->master.n_links = links;
'aspeed' is a devm_alloc'ed, so there is no need to free it explicitly or there will be a double free(). Remove the 'release' function that is wrong and unneeded. Fixes: 606397d67f41 ("fsi: Add ast2600 master driver") Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> --- This patch is completely theoretical. It looks good to me, but there is a little too much indirections for me. I'm also not that familiar with fixing issue related to 'release' function... ... So review with care :) --- drivers/fsi/fsi-master-aspeed.c | 9 --------- 1 file changed, 9 deletions(-)