| Message ID | 20200824211027.11543-1-trix@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] usb: storage: initialize variable | expand |
On Mon, Aug 24, 2020 at 02:10:27PM -0700, trix@redhat.com wrote: > From: Tom Rix <trix@redhat.com> > > clang static analysis reports this representative problem > > transport.c:495:15: warning: Assigned value is garbage or > undefined > length_left -= partial; > ^ ~~~~~~~ > partial is set only when usb_stor_bulk_transfer_sglist() > is successful. > > So set partial on entry to 0. > > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") > Signed-off-by: Tom Rix <trix@redhat.com> > --- > drivers/usb/storage/transport.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/usb/storage/transport.c b/drivers/usb/storage/transport.c > index 238a8088e17f..044429717dcc 100644 > --- a/drivers/usb/storage/transport.c > +++ b/drivers/usb/storage/transport.c > @@ -414,6 +414,9 @@ static int usb_stor_bulk_transfer_sglist(struct us_data *us, unsigned int pipe, > { > int result; > > + if (act_len) > + *act_len = 0; > + > /* don't submit s-g requests during abort processing */ > if (test_bit(US_FLIDX_ABORTING, &us->dflags)) > return USB_STOR_XFER_ERROR; At a glance this seems odd to me. If the caller insists on ignoring the return value, shouldn't it just initialize partial to zero? In my experience it's generally frowned upon for functions to store results in error paths. Regards, Vito Caputo
On 8/24/20 2:18 PM, Vito Caputo wrote: > On Mon, Aug 24, 2020 at 02:10:27PM -0700, trix@redhat.com wrote: >> From: Tom Rix <trix@redhat.com> >> >> clang static analysis reports this representative problem >> >> transport.c:495:15: warning: Assigned value is garbage or >> undefined >> length_left -= partial; >> ^ ~~~~~~~ >> partial is set only when usb_stor_bulk_transfer_sglist() >> is successful. >> >> So set partial on entry to 0. >> >> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") >> Signed-off-by: Tom Rix <trix@redhat.com> >> --- >> drivers/usb/storage/transport.c | 3 +++ >> 1 file changed, 3 insertions(+) >> >> diff --git a/drivers/usb/storage/transport.c b/drivers/usb/storage/transport.c >> index 238a8088e17f..044429717dcc 100644 >> --- a/drivers/usb/storage/transport.c >> +++ b/drivers/usb/storage/transport.c >> @@ -414,6 +414,9 @@ static int usb_stor_bulk_transfer_sglist(struct us_data *us, unsigned int pipe, >> { >> int result; >> >> + if (act_len) >> + *act_len = 0; >> + >> /* don't submit s-g requests during abort processing */ >> if (test_bit(US_FLIDX_ABORTING, &us->dflags)) >> return USB_STOR_XFER_ERROR; > At a glance this seems odd to me. If the caller insists on ignoring > the return value, shouldn't it just initialize partial to zero? > > In my experience it's generally frowned upon for functions to store > results in error paths. Then maybe v1 is more appropriate. Else i can spin a v3. My preference is v1 as it doesn't add any runtime if-checks. Tom > Regards, > Vito Caputo >
On Mon, Aug 24, 2020 at 02:18:39PM -0700, Vito Caputo wrote: > On Mon, Aug 24, 2020 at 02:10:27PM -0700, trix@redhat.com wrote: > > From: Tom Rix <trix@redhat.com> > > > > clang static analysis reports this representative problem > > > > transport.c:495:15: warning: Assigned value is garbage or > > undefined > > length_left -= partial; > > ^ ~~~~~~~ > > partial is set only when usb_stor_bulk_transfer_sglist() > > is successful. > > > > So set partial on entry to 0. > > > > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") > > Signed-off-by: Tom Rix <trix@redhat.com> > > --- > > drivers/usb/storage/transport.c | 3 +++ > > 1 file changed, 3 insertions(+) > > > > diff --git a/drivers/usb/storage/transport.c b/drivers/usb/storage/transport.c > > index 238a8088e17f..044429717dcc 100644 > > --- a/drivers/usb/storage/transport.c > > +++ b/drivers/usb/storage/transport.c > > @@ -414,6 +414,9 @@ static int usb_stor_bulk_transfer_sglist(struct us_data *us, unsigned int pipe, > > { > > int result; > > > > + if (act_len) > > + *act_len = 0; > > + > > /* don't submit s-g requests during abort processing */ > > if (test_bit(US_FLIDX_ABORTING, &us->dflags)) > > return USB_STOR_XFER_ERROR; > > At a glance this seems odd to me. If the caller insists on ignoring > the return value, shouldn't it just initialize partial to zero? In this case, the callers are not the final consumers of the return value or of partial. They merely copy those values back up to _their_ callers, and those copy operations are what the static analyzer objects to. > In my experience it's generally frowned upon for functions to store > results in error paths. I don't see any reason for such an attitude, at least not here. It makes perfectly good sense, if an error prevents transmission of an entire data buffer, to store the amount of data that did get transmitted. Alan Stern
On Mon, Aug 24, 2020 at 02:31:01PM -0700, Tom Rix wrote: > > On 8/24/20 2:18 PM, Vito Caputo wrote: > > On Mon, Aug 24, 2020 at 02:10:27PM -0700, trix@redhat.com wrote: > >> From: Tom Rix <trix@redhat.com> > >> > >> clang static analysis reports this representative problem > >> > >> transport.c:495:15: warning: Assigned value is garbage or > >> undefined > >> length_left -= partial; > >> ^ ~~~~~~~ > >> partial is set only when usb_stor_bulk_transfer_sglist() > >> is successful. > >> > >> So set partial on entry to 0. > >> > >> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") > >> Signed-off-by: Tom Rix <trix@redhat.com> > >> --- > >> drivers/usb/storage/transport.c | 3 +++ > >> 1 file changed, 3 insertions(+) > >> > >> diff --git a/drivers/usb/storage/transport.c b/drivers/usb/storage/transport.c > >> index 238a8088e17f..044429717dcc 100644 > >> --- a/drivers/usb/storage/transport.c > >> +++ b/drivers/usb/storage/transport.c > >> @@ -414,6 +414,9 @@ static int usb_stor_bulk_transfer_sglist(struct us_data *us, unsigned int pipe, > >> { > >> int result; > >> > >> + if (act_len) > >> + *act_len = 0; > >> + > >> /* don't submit s-g requests during abort processing */ > >> if (test_bit(US_FLIDX_ABORTING, &us->dflags)) > >> return USB_STOR_XFER_ERROR; > > At a glance this seems odd to me. If the caller insists on ignoring > > the return value, shouldn't it just initialize partial to zero? > > > > In my experience it's generally frowned upon for functions to store > > results in error paths. > > Then maybe v1 is more appropriate. > > Else i can spin a v3. > > My preference is v1 as it doesn't add any runtime if-checks. If you really want to get rid of the runtime check (both the one you added and the one already present), you can audit all the callers of this routine to make certain that none of them pass a NULL pointer for act_len. Alan Stern
diff --git a/drivers/usb/storage/transport.c b/drivers/usb/storage/transport.c index 238a8088e17f..044429717dcc 100644 --- a/drivers/usb/storage/transport.c +++ b/drivers/usb/storage/transport.c @@ -414,6 +414,9 @@ static int usb_stor_bulk_transfer_sglist(struct us_data *us, unsigned int pipe, { int result; + if (act_len) + *act_len = 0; + /* don't submit s-g requests during abort processing */ if (test_bit(US_FLIDX_ABORTING, &us->dflags)) return USB_STOR_XFER_ERROR;