Patchwork [U-Boot,v3,2/4] usb/gadget: fastboot: add eMMC support for flash command

login
register
mail settings
Submitter Steve Rae
Date June 26, 2014, 8:13 p.m.
Message ID <1403813604-31685-3-git-send-email-srae@broadcom.com>
Download mbox | patch
Permalink /patch/364706/
State Superseded
Delegated to: Marek Vasut
Headers show

Comments

Steve Rae - June 26, 2014, 8:13 p.m.
- add support for 'fastboot flash' command for eMMC devices

Signed-off-by: Steve Rae <srae@broadcom.com>
---
I suspect that the "sparse image" handling (ie. the "while (remaining_chunks)" loop)
has been implemented elsewhere -- I need help finding the original code to determine
any licensing issues....
Thanks, Steve

Changes in v3:
- remove most references to 'mmc',
  which leaves only one mmc specific function: mmc_get_dev()

Changes in v2:
- split large function into three
- improved handling of response messages
- additional partition size checking when writing sparse image

 common/Makefile  |   5 ++
 common/fb_mmc.c  | 190 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 include/fb_mmc.h |   8 +++
 3 files changed, 203 insertions(+)
 create mode 100644 common/fb_mmc.c
 create mode 100644 include/fb_mmc.h
Marek Vasut - July 31, 2014, 1:37 a.m.
On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
[...]
> +
> +#include <common.h>
> +#include <fb_mmc.h>
> +#include <part.h>
> +#include <sparse_format.h>
> +
> +/* The 64 defined bytes plus \0 */
> +#define RESPONSE_LEN	(64 + 1)
> +
> +static char *response_str;

I'd suggest to pass this "response_str" around instead of making it global.

> +static void fastboot_resp(const char *s)
> +{
> +	strncpy(response_str, s, RESPONSE_LEN);
> +	response_str[RESPONSE_LEN - 1] = '\0';

This could be shrunk to a single snprintf(response_str, RESPONSE_LENGTH, s); I 
think, but I'm not sure if the overhead won't grow.

> +}
> +
> +static int is_sparse_image(void *buf)
> +{
> +	sparse_header_t *s_header = (sparse_header_t *)buf;
> +
> +	if ((le32_to_cpu(s_header->magic) == SPARSE_HEADER_MAGIC) &&
> +	    (le16_to_cpu(s_header->major_version) == 1))
> +		return 1;
> +
> +	return 0;
> +}
> +
> +static void write_sparse_image(block_dev_desc_t *dev_desc,
> +		disk_partition_t *info, const char *part_name,
> +		void *buffer, unsigned int download_bytes)
> +{
> +	lbaint_t blk;
> +	lbaint_t blkcnt;
> +	lbaint_t blks;
> +	sparse_header_t *s_header = (sparse_header_t *)buffer;
> +	chunk_header_t *c_header;
> +	void *buf;
> +	uint32_t blk_sz;
> +	uint32_t remaining_chunks;
> +	uint32_t bytes_written = 0;
> +
> +	blk_sz = le32_to_cpu(s_header->blk_sz);
> +
> +	/* verify s_header->blk_sz is exact multiple of info->blksz */
> +	if (blk_sz != (blk_sz & ~(info->blksz - 1))) {
> +		printf("%s: Sparse image block size issue [%u]\n",
> +		       __func__, blk_sz);
> +		fastboot_resp("FAILsparse image block size issue");

Can't you just make the fastboot_resp() function a variadic one AND move the 
printf() into the fastboot_resp() function? You could then even get consistent 
output on both the device and in the response if you snprintf() into the 
response_str first and then printf() the response_str .

> +		return;
> +	}

[...]

> +static void write_raw_image(block_dev_desc_t *dev_desc, disk_partition_t
> *info, +		const char *part_name, void *buffer,
> +		unsigned int download_bytes)
> +{
> +	lbaint_t blkcnt;
> +	lbaint_t blks;
> +
> +	/* determine number of blocks to write */
> +	blkcnt = ((download_bytes + (info->blksz - 1)) & ~(info->blksz - 1));
> +	blkcnt = blkcnt / info->blksz;
> +
> +	if (blkcnt > info->size) {
> +		printf("%s: too large for partition: '%s'\n", __func__,
> +		       part_name);
> +		fastboot_resp("FAILtoo large for partition");
> +		return;
> +	}
> +
> +	printf("Flashing Raw Image\n");

Use puts() here and everywhere where printf() is not taking any args please.

> +	blks = dev_desc->block_write(dev_desc->dev, info->start, blkcnt,
> +				     buffer);
> +	if (blks != blkcnt) {
> +		printf("%s: failed writing to device %d\n", __func__,
> +		       dev_desc->dev);
> +		fastboot_resp("FAILfailed writing to device");
> +		return;
> +	}
> +
> +	printf("........ wrote " LBAFU " bytes to '%s'\n", blkcnt * info->blksz,
> +	       part_name);
> +	fastboot_resp("OKAY");
> +}
[...]
Steve Rae - Aug. 6, 2014, 11:48 p.m.
On 14-07-30 06:37 PM, Marek Vasut wrote:
> On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
> [...]
>> +
>> +#include <common.h>
>> +#include <fb_mmc.h>
>> +#include <part.h>
>> +#include <sparse_format.h>
>> +
>> +/* The 64 defined bytes plus \0 */
>> +#define RESPONSE_LEN	(64 + 1)
>> +
>> +static char *response_str;
>
> I'd suggest to pass this "response_str" around instead of making it global.
>

That would involve adding it to fastboot_resp(), which is called 11 
times in this code, from 3 different functions (so would need to add 
this to two of the functions...). And as these evolve, there will likely 
be more nested functions, which would all require "passing it around"....
I think that this "static global pointer" is a cleaner implementation.

>> +static void fastboot_resp(const char *s)
>> +{
>> +	strncpy(response_str, s, RESPONSE_LEN);
>> +	response_str[RESPONSE_LEN - 1] = '\0';
>
> This could be shrunk to a single snprintf(response_str, RESPONSE_LENGTH, s); I
> think, but I'm not sure if the overhead won't grow.
>

snprintf() is used very sparingling in U-Boot, and with the cautionary 
statements in README (line 852) and the fact that CONFIG_SYS_VSNPRINTF 
is not defined for armv7 builds, I am not going to use it....

>> +}
>> +
>> +static int is_sparse_image(void *buf)
>> +{
>> +	sparse_header_t *s_header = (sparse_header_t *)buf;
>> +
>> +	if ((le32_to_cpu(s_header->magic) == SPARSE_HEADER_MAGIC) &&
>> +	    (le16_to_cpu(s_header->major_version) == 1))
>> +		return 1;
>> +
>> +	return 0;
>> +}
>> +
>> +static void write_sparse_image(block_dev_desc_t *dev_desc,
>> +		disk_partition_t *info, const char *part_name,
>> +		void *buffer, unsigned int download_bytes)
>> +{
>> +	lbaint_t blk;
>> +	lbaint_t blkcnt;
>> +	lbaint_t blks;
>> +	sparse_header_t *s_header = (sparse_header_t *)buffer;
>> +	chunk_header_t *c_header;
>> +	void *buf;
>> +	uint32_t blk_sz;
>> +	uint32_t remaining_chunks;
>> +	uint32_t bytes_written = 0;
>> +
>> +	blk_sz = le32_to_cpu(s_header->blk_sz);
>> +
>> +	/* verify s_header->blk_sz is exact multiple of info->blksz */
>> +	if (blk_sz != (blk_sz & ~(info->blksz - 1))) {
>> +		printf("%s: Sparse image block size issue [%u]\n",
>> +		       __func__, blk_sz);
>> +		fastboot_resp("FAILsparse image block size issue");
>
> Can't you just make the fastboot_resp() function a variadic one AND move the
> printf() into the fastboot_resp() function? You could then even get consistent
> output on both the device and in the response if you snprintf() into the
> response_str first and then printf() the response_str .
>

Generally, the printf() statements which are sent to the console, and 
the fastboot_resp() statements which are sent to the host running the 
"fastboot" application are not the same....

>> +		return;
>> +	}
>
> [...]
>
>> +static void write_raw_image(block_dev_desc_t *dev_desc, disk_partition_t
>> *info, +		const char *part_name, void *buffer,
>> +		unsigned int download_bytes)
>> +{
>> +	lbaint_t blkcnt;
>> +	lbaint_t blks;
>> +
>> +	/* determine number of blocks to write */
>> +	blkcnt = ((download_bytes + (info->blksz - 1)) & ~(info->blksz - 1));
>> +	blkcnt = blkcnt / info->blksz;
>> +
>> +	if (blkcnt > info->size) {
>> +		printf("%s: too large for partition: '%s'\n", __func__,
>> +		       part_name);
>> +		fastboot_resp("FAILtoo large for partition");
>> +		return;
>> +	}
>> +
>> +	printf("Flashing Raw Image\n");
>
> Use puts() here and everywhere where printf() is not taking any args please.

done in v4 - Thanks!

>
>> +	blks = dev_desc->block_write(dev_desc->dev, info->start, blkcnt,
>> +				     buffer);
>> +	if (blks != blkcnt) {
>> +		printf("%s: failed writing to device %d\n", __func__,
>> +		       dev_desc->dev);
>> +		fastboot_resp("FAILfailed writing to device");
>> +		return;
>> +	}
>> +
>> +	printf("........ wrote " LBAFU " bytes to '%s'\n", blkcnt * info->blksz,
>> +	       part_name);
>> +	fastboot_resp("OKAY");
>> +}
> [...]
>

Thanks, Steve
Marek Vasut - Aug. 7, 2014, 12:13 a.m.
On Thursday, August 07, 2014 at 01:48:06 AM, Steve Rae wrote:
> On 14-07-30 06:37 PM, Marek Vasut wrote:
> > On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
> > [...]
> > 
> >> +
> >> +#include <common.h>
> >> +#include <fb_mmc.h>
> >> +#include <part.h>
> >> +#include <sparse_format.h>
> >> +
> >> +/* The 64 defined bytes plus \0 */
> >> +#define RESPONSE_LEN	(64 + 1)
> >> +
> >> +static char *response_str;
> > 
> > I'd suggest to pass this "response_str" around instead of making it
> > global.
> 
> That would involve adding it to fastboot_resp(), which is called 11
> times in this code, from 3 different functions (so would need to add
> this to two of the functions...). And as these evolve, there will likely
> be more nested functions, which would all require "passing it around"....
> I think that this "static global pointer" is a cleaner implementation.

Eventually, the amount of these static variables in the code will grow and it 
will become increasingly difficult to weed them out. I believe it would be even 
better to pass around some kind of a structure with "private data" of the 
fastboot, which would cater for all possible variables which might come in the 
future. What do you think ?

> >> +static void fastboot_resp(const char *s)
> >> +{
> >> +	strncpy(response_str, s, RESPONSE_LEN);
> >> +	response_str[RESPONSE_LEN - 1] = '\0';
> > 
> > This could be shrunk to a single snprintf(response_str, RESPONSE_LENGTH,
> > s); I think, but I'm not sure if the overhead won't grow.
> 
> snprintf() is used very sparingling in U-Boot

This is not a reason to avoid it.

> , and with the cautionary statements in README (line 852)

Which statements? Can you please point them out? I fail to see them, sorry.

> and the fact that CONFIG_SYS_VSNPRINTF is not defined for armv7 builds, I am 
not going to use it....

Is it a problem to define it? Also, even without CONFIG_SYS_VSNPRINTF , the 
functions are still available, see the README:
 857                 If this option is not given then these functions will
 858                 silently discard their buffer size argument - this means
 859                 you are not getting any overflow checking in this case.

I have yet to see some hard-evidence against using safe printing functions here.

> >> +}
> >> +
> >> +static int is_sparse_image(void *buf)
> >> +{
> >> +	sparse_header_t *s_header = (sparse_header_t *)buf;
> >> +
> >> +	if ((le32_to_cpu(s_header->magic) == SPARSE_HEADER_MAGIC) &&
> >> +	    (le16_to_cpu(s_header->major_version) == 1))
> >> +		return 1;
> >> +
> >> +	return 0;
> >> +}
> >> +
> >> +static void write_sparse_image(block_dev_desc_t *dev_desc,
> >> +		disk_partition_t *info, const char *part_name,
> >> +		void *buffer, unsigned int download_bytes)
> >> +{
> >> +	lbaint_t blk;
> >> +	lbaint_t blkcnt;
> >> +	lbaint_t blks;
> >> +	sparse_header_t *s_header = (sparse_header_t *)buffer;
> >> +	chunk_header_t *c_header;
> >> +	void *buf;
> >> +	uint32_t blk_sz;
> >> +	uint32_t remaining_chunks;
> >> +	uint32_t bytes_written = 0;
> >> +
> >> +	blk_sz = le32_to_cpu(s_header->blk_sz);
> >> +
> >> +	/* verify s_header->blk_sz is exact multiple of info->blksz */
> >> +	if (blk_sz != (blk_sz & ~(info->blksz - 1))) {
> >> +		printf("%s: Sparse image block size issue [%u]\n",
> >> +		       __func__, blk_sz);
> >> +		fastboot_resp("FAILsparse image block size issue");
> > 
> > Can't you just make the fastboot_resp() function a variadic one AND move
> > the printf() into the fastboot_resp() function? You could then even get
> > consistent output on both the device and in the response if you
> > snprintf() into the response_str first and then printf() the
> > response_str .
> 
> Generally, the printf() statements which are sent to the console, and
> the fastboot_resp() statements which are sent to the host running the
> "fastboot" application are not the same....

OK, thanks!
Steve Rae - Aug. 7, 2014, 12:28 a.m.
On 14-08-06 05:13 PM, Marek Vasut wrote:
> On Thursday, August 07, 2014 at 01:48:06 AM, Steve Rae wrote:
>> On 14-07-30 06:37 PM, Marek Vasut wrote:
>>> On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
>>> [...]
>>>
>>>> +
>>>> +#include <common.h>
>>>> +#include <fb_mmc.h>
>>>> +#include <part.h>
>>>> +#include <sparse_format.h>
>>>> +
>>>> +/* The 64 defined bytes plus \0 */
>>>> +#define RESPONSE_LEN	(64 + 1)
>>>> +
>>>> +static char *response_str;
>>>
>>> I'd suggest to pass this "response_str" around instead of making it
>>> global.
>>
>> That would involve adding it to fastboot_resp(), which is called 11
>> times in this code, from 3 different functions (so would need to add
>> this to two of the functions...). And as these evolve, there will likely
>> be more nested functions, which would all require "passing it around"....
>> I think that this "static global pointer" is a cleaner implementation.
>
> Eventually, the amount of these static variables in the code will grow and it
> will become increasingly difficult to weed them out. I believe it would be even
> better to pass around some kind of a structure with "private data" of the
> fastboot, which would cater for all possible variables which might come in the
> future. What do you think ?
>

Yes -- if there is private data that the fastboot implementation 
requires, then a data structure is the way to go. However, I still think
that this "fastboot response string" would even be an exception to that 
private data....

>>>> +static void fastboot_resp(const char *s)
>>>> +{
>>>> +	strncpy(response_str, s, RESPONSE_LEN);
>>>> +	response_str[RESPONSE_LEN - 1] = '\0';
>>>
>>> This could be shrunk to a single snprintf(response_str, RESPONSE_LENGTH,
>>> s); I think, but I'm not sure if the overhead won't grow.
>>
>> snprintf() is used very sparingling in U-Boot
>
> This is not a reason to avoid it.
true....
>
>> , and with the cautionary statements in README (line 852)
>
> Which statements? Can you please point them out? I fail to see them, sorry.

I was referring to what you mention below...
  852 - Safe printf() functions
  853      Define CONFIG_SYS_VSNPRINTF to compile in safe versions of
  854      the printf() functions. These are defined in
  855      include/vsprintf.h and include snprintf(), vsnprintf() and
  856      so on. Code size increase is approximately 300-500 bytes.
  857      If this option is not given then these functions will
  858      silently discard their buffer size argument - this means
  859      you are not getting any overflow checking in this case.

>
>> and the fact that CONFIG_SYS_VSNPRINTF is not defined for armv7 builds, I am
> not going to use it....
>
> Is it a problem to define it? Also, even without CONFIG_SYS_VSNPRINTF , the
> functions are still available, see the README:
>   857                 If this option is not given then these functions will
>   858                 silently discard their buffer size argument - this means
>   859                 you are not getting any overflow checking in this case.
>
> I have yet to see some hard-evidence against using safe printing functions here.
>

I don't want to be the first to defined it for all of armv7....
And I really don't want to define it only only my boards running so that 
they can run 'fastboot'
What do you suggest?

>>>> +}
>>>> +
>>>> +static int is_sparse_image(void *buf)
>>>> +{
>>>> +	sparse_header_t *s_header = (sparse_header_t *)buf;
>>>> +
>>>> +	if ((le32_to_cpu(s_header->magic) == SPARSE_HEADER_MAGIC) &&
>>>> +	    (le16_to_cpu(s_header->major_version) == 1))
>>>> +		return 1;
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static void write_sparse_image(block_dev_desc_t *dev_desc,
>>>> +		disk_partition_t *info, const char *part_name,
>>>> +		void *buffer, unsigned int download_bytes)
>>>> +{
>>>> +	lbaint_t blk;
>>>> +	lbaint_t blkcnt;
>>>> +	lbaint_t blks;
>>>> +	sparse_header_t *s_header = (sparse_header_t *)buffer;
>>>> +	chunk_header_t *c_header;
>>>> +	void *buf;
>>>> +	uint32_t blk_sz;
>>>> +	uint32_t remaining_chunks;
>>>> +	uint32_t bytes_written = 0;
>>>> +
>>>> +	blk_sz = le32_to_cpu(s_header->blk_sz);
>>>> +
>>>> +	/* verify s_header->blk_sz is exact multiple of info->blksz */
>>>> +	if (blk_sz != (blk_sz & ~(info->blksz - 1))) {
>>>> +		printf("%s: Sparse image block size issue [%u]\n",
>>>> +		       __func__, blk_sz);
>>>> +		fastboot_resp("FAILsparse image block size issue");
>>>
>>> Can't you just make the fastboot_resp() function a variadic one AND move
>>> the printf() into the fastboot_resp() function? You could then even get
>>> consistent output on both the device and in the response if you
>>> snprintf() into the response_str first and then printf() the
>>> response_str .
>>
>> Generally, the printf() statements which are sent to the console, and
>> the fastboot_resp() statements which are sent to the host running the
>> "fastboot" application are not the same....
>
> OK, thanks!
>
Thanks, Steve
Marek Vasut - Aug. 7, 2014, 1:23 p.m.
On Thursday, August 07, 2014 at 02:28:13 AM, Steve Rae wrote:
> On 14-08-06 05:13 PM, Marek Vasut wrote:
> > On Thursday, August 07, 2014 at 01:48:06 AM, Steve Rae wrote:
> >> On 14-07-30 06:37 PM, Marek Vasut wrote:
> >>> On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
> >>> [...]
> >>> 
> >>>> +
> >>>> +#include <common.h>
> >>>> +#include <fb_mmc.h>
> >>>> +#include <part.h>
> >>>> +#include <sparse_format.h>
> >>>> +
> >>>> +/* The 64 defined bytes plus \0 */
> >>>> +#define RESPONSE_LEN	(64 + 1)
> >>>> +
> >>>> +static char *response_str;
> >>> 
> >>> I'd suggest to pass this "response_str" around instead of making it
> >>> global.
> >> 
> >> That would involve adding it to fastboot_resp(), which is called 11
> >> times in this code, from 3 different functions (so would need to add
> >> this to two of the functions...). And as these evolve, there will likely
> >> be more nested functions, which would all require "passing it
> >> around".... I think that this "static global pointer" is a cleaner
> >> implementation.
> > 
> > Eventually, the amount of these static variables in the code will grow
> > and it will become increasingly difficult to weed them out. I believe it
> > would be even better to pass around some kind of a structure with
> > "private data" of the fastboot, which would cater for all possible
> > variables which might come in the future. What do you think ?
> 
> Yes -- if there is private data that the fastboot implementation
> requires, then a data structure is the way to go. However, I still think
> that this "fastboot response string" would even be an exception to that
> private data....

OK, let's leave it this way for now.

> >>>> +static void fastboot_resp(const char *s)
> >>>> +{
> >>>> +	strncpy(response_str, s, RESPONSE_LEN);
> >>>> +	response_str[RESPONSE_LEN - 1] = '\0';
> >>> 
> >>> This could be shrunk to a single snprintf(response_str,
> >>> RESPONSE_LENGTH, s); I think, but I'm not sure if the overhead won't
> >>> grow.
> >> 
> >> snprintf() is used very sparingling in U-Boot
> > 
> > This is not a reason to avoid it.
> 
> true....
> 
> >> , and with the cautionary statements in README (line 852)
> > 
> > Which statements? Can you please point them out? I fail to see them,
> > sorry.
> 
> I was referring to what you mention below...
>   852 - Safe printf() functions
>   853      Define CONFIG_SYS_VSNPRINTF to compile in safe versions of
>   854      the printf() functions. These are defined in
>   855      include/vsprintf.h and include snprintf(), vsnprintf() and
>   856      so on. Code size increase is approximately 300-500 bytes.
>   857      If this option is not given then these functions will
>   858      silently discard their buffer size argument - this means
>   859      you are not getting any overflow checking in this case.

I really don't see the "cautionary statements" here , no . I see that it 
discards the size checking if this CONFIG_SYS_VSNPRINTF is not enabled, but that 
does not obstruct the operation of those functions.

> >> and the fact that CONFIG_SYS_VSNPRINTF is not defined for armv7 builds,
> >> I am
> > 
> > not going to use it....
> > 
> > Is it a problem to define it? Also, even without CONFIG_SYS_VSNPRINTF ,
> > the
> > 
> > functions are still available, see the README:
> >   857                 If this option is not given then these functions
> >   will 858                 silently discard their buffer size argument -
> >   this means 859                 you are not getting any overflow
> >   checking in this case.
> > 
> > I have yet to see some hard-evidence against using safe printing
> > functions here.
> 
> I don't want to be the first to defined it for all of armv7....

Honestly, we should just enable this CONFIG_SYS_VSNPRINTF by default for the 
good of humanity and all the things, since this unbounded string handling is 
just evil (see how OpenSSL ended up, partly because of that ... and I am just 
starting to see the pattern in all the security code). I don't want to go down 
that road with U-Boot.

So, would you please cook a separate patch to enable this by default, so it 
would spur the right kind of discussion on this matter ?

> And I really don't want to define it only only my boards running so that
> they can run 'fastboot'
> What do you suggest?

See above, thanks !
Pantelis Antoniou - Aug. 7, 2014, 1:28 p.m.
Hi Marek,

[snip]

>> I don't want to be the first to defined it for all of armv7....
> 
> Honestly, we should just enable this CONFIG_SYS_VSNPRINTF by default for the 
> good of humanity and all the things, since this unbounded string handling is 
> just evil (see how OpenSSL ended up, partly because of that ... and I am just 
> starting to see the pattern in all the security code). I don't want to go down 
> that road with U-Boot.
> 
> So, would you please cook a separate patch to enable this by default, so it 
> would spur the right kind of discussion on this matter ?
> 

We should enable this by default. Unbounded string handling scares me.

If we have problems with blowing over SPL size restrictions, perhaps have it
disabled only on those cases (that are known to have a problem).

>> And I really don't want to define it only only my boards running so that
>> they can run 'fastboot'
>> What do you suggest?
> 
> See above, thanks !

Regards

-- Pantelis
Marek Vasut - Aug. 7, 2014, 1:43 p.m.
On Thursday, August 07, 2014 at 03:28:14 PM, Pantelis Antoniou wrote:
> Hi Marek,
> 
> [snip]
> 
> >> I don't want to be the first to defined it for all of armv7....
> > 
> > Honestly, we should just enable this CONFIG_SYS_VSNPRINTF by default for
> > the good of humanity and all the things, since this unbounded string
> > handling is just evil (see how OpenSSL ended up, partly because of that
> > ... and I am just starting to see the pattern in all the security code).
> > I don't want to go down that road with U-Boot.
> > 
> > So, would you please cook a separate patch to enable this by default, so
> > it would spur the right kind of discussion on this matter ?
> 
> We should enable this by default. Unbounded string handling scares me.
> 
> If we have problems with blowing over SPL size restrictions, perhaps have
> it disabled only on those cases (that are known to have a problem).

Right, I fully agree with what you said. The SPL and TPL might have issues with 
this being enabled, but then this can be enabled for full-blown U-Boot only.

But this discussion should happen in a thread associated with patch enabling 
this. ;-)

Best regards,
Marek Vasut
Steve Rae - Aug. 7, 2014, 4:52 p.m.
On 14-08-07 06:23 AM, Marek Vasut wrote:
> On Thursday, August 07, 2014 at 02:28:13 AM, Steve Rae wrote:
>> On 14-08-06 05:13 PM, Marek Vasut wrote:
>>> On Thursday, August 07, 2014 at 01:48:06 AM, Steve Rae wrote:
>>>> On 14-07-30 06:37 PM, Marek Vasut wrote:
>>>>> On Thursday, June 26, 2014 at 10:13:22 PM, Steve Rae wrote:
>>>>> [...]
>>>>>
>>>>>> +
>>>>>> +#include <common.h>
>>>>>> +#include <fb_mmc.h>
>>>>>> +#include <part.h>
>>>>>> +#include <sparse_format.h>
>>>>>> +
>>>>>> +/* The 64 defined bytes plus \0 */
>>>>>> +#define RESPONSE_LEN	(64 + 1)
>>>>>> +
>>>>>> +static char *response_str;
>>>>>
>>>>> I'd suggest to pass this "response_str" around instead of making it
>>>>> global.
>>>>
>>>> That would involve adding it to fastboot_resp(), which is called 11
>>>> times in this code, from 3 different functions (so would need to add
>>>> this to two of the functions...). And as these evolve, there will likely
>>>> be more nested functions, which would all require "passing it
>>>> around".... I think that this "static global pointer" is a cleaner
>>>> implementation.
>>>
>>> Eventually, the amount of these static variables in the code will grow
>>> and it will become increasingly difficult to weed them out. I believe it
>>> would be even better to pass around some kind of a structure with
>>> "private data" of the fastboot, which would cater for all possible
>>> variables which might come in the future. What do you think ?
>>
>> Yes -- if there is private data that the fastboot implementation
>> requires, then a data structure is the way to go. However, I still think
>> that this "fastboot response string" would even be an exception to that
>> private data....
>
> OK, let's leave it this way for now.
>
>>>>>> +static void fastboot_resp(const char *s)
>>>>>> +{
>>>>>> +	strncpy(response_str, s, RESPONSE_LEN);
>>>>>> +	response_str[RESPONSE_LEN - 1] = '\0';
>>>>>
>>>>> This could be shrunk to a single snprintf(response_str,
>>>>> RESPONSE_LENGTH, s); I think, but I'm not sure if the overhead won't
>>>>> grow.
>>>>
>>>> snprintf() is used very sparingling in U-Boot
>>>
>>> This is not a reason to avoid it.
>>
>> true....
>>
>>>> , and with the cautionary statements in README (line 852)
>>>
>>> Which statements? Can you please point them out? I fail to see them,
>>> sorry.
>>
>> I was referring to what you mention below...
>>    852 - Safe printf() functions
>>    853      Define CONFIG_SYS_VSNPRINTF to compile in safe versions of
>>    854      the printf() functions. These are defined in
>>    855      include/vsprintf.h and include snprintf(), vsnprintf() and
>>    856      so on. Code size increase is approximately 300-500 bytes.
>>    857      If this option is not given then these functions will
>>    858      silently discard their buffer size argument - this means
>>    859      you are not getting any overflow checking in this case.
>
> I really don't see the "cautionary statements" here , no . I see that it
> discards the size checking if this CONFIG_SYS_VSNPRINTF is not enabled, but that
> does not obstruct the operation of those functions.
>

I'm really confused: my code ensures that the buffer is not overflowed 
and that it is terminated properly. If snprintf() (without 
CONFIG_SYS_VSNPRINTF defined) doesn't provide "any overflow checking", 
then why would I use it?

>>>> and the fact that CONFIG_SYS_VSNPRINTF is not defined for armv7 builds,
>>>> I am
>>>
>>> not going to use it....
>>>
>>> Is it a problem to define it? Also, even without CONFIG_SYS_VSNPRINTF ,
>>> the
>>>
>>> functions are still available, see the README:
>>>    857                 If this option is not given then these functions
>>>    will 858                 silently discard their buffer size argument -
>>>    this means 859                 you are not getting any overflow
>>>    checking in this case.
>>>
>>> I have yet to see some hard-evidence against using safe printing
>>> functions here.
>>
>> I don't want to be the first to defined it for all of armv7....
>
> Honestly, we should just enable this CONFIG_SYS_VSNPRINTF by default for the
> good of humanity and all the things, since this unbounded string handling is
> just evil (see how OpenSSL ended up, partly because of that ... and I am just
> starting to see the pattern in all the security code). I don't want to go down
> that road with U-Boot.
>
> So, would you please cook a separate patch to enable this by default, so it
> would spur the right kind of discussion on this matter ?


I will apologize in advance, but I just don't know anything about SPL or 
TPL or any other boards (outside of my very limited armv7 and armv8 
scope)....
I would be happy to review and test this suggested patch (on our 
boards), but would be uncomfortable with proposing this patch.
Please go ahead and submit a patch, and I'll check it!
Thanks, Steve

>
>> And I really don't want to define it only only my boards running so that
>> they can run 'fastboot'
>> What do you suggest?
>
> See above, thanks !
>
Marek Vasut - Aug. 7, 2014, 5:12 p.m.
On Thursday, August 07, 2014 at 06:52:44 PM, Steve Rae wrote:

[...]

> >> I was referring to what you mention below...
> >> 
> >>    852 - Safe printf() functions
> >>    853      Define CONFIG_SYS_VSNPRINTF to compile in safe versions of
> >>    854      the printf() functions. These are defined in
> >>    855      include/vsprintf.h and include snprintf(), vsnprintf() and
> >>    856      so on. Code size increase is approximately 300-500 bytes.
> >>    857      If this option is not given then these functions will
> >>    858      silently discard their buffer size argument - this means
> >>    859      you are not getting any overflow checking in this case.
> > 
> > I really don't see the "cautionary statements" here , no . I see that it
> > discards the size checking if this CONFIG_SYS_VSNPRINTF is not enabled,
> > but that does not obstruct the operation of those functions.
> 
> I'm really confused: my code ensures that the buffer is not overflowed
> and that it is terminated properly. If snprintf() (without
> CONFIG_SYS_VSNPRINTF defined) doesn't provide "any overflow checking",
> then why would I use it?

That's why I suggested to enable CONFIG_SYS_VSNPRINTF unconditionally. Then your 
code would not need to duplicate all the overflow checks, would it ?

> >>>> and the fact that CONFIG_SYS_VSNPRINTF is not defined for armv7
> >>>> builds, I am
> >>> 
> >>> not going to use it....
> >>> 
> >>> Is it a problem to define it? Also, even without CONFIG_SYS_VSNPRINTF ,
> >>> the
> >>> 
> >>> functions are still available, see the README:
> >>>    857                 If this option is not given then these functions
> >>>    will 858                 silently discard their buffer size argument
> >>>    - this means 859                 you are not getting any overflow
> >>>    checking in this case.
> >>> 
> >>> I have yet to see some hard-evidence against using safe printing
> >>> functions here.
> >> 
> >> I don't want to be the first to defined it for all of armv7....
> > 
> > Honestly, we should just enable this CONFIG_SYS_VSNPRINTF by default for
> > the good of humanity and all the things, since this unbounded string
> > handling is just evil (see how OpenSSL ended up, partly because of that
> > ... and I am just starting to see the pattern in all the security code).
> > I don't want to go down that road with U-Boot.
> > 
> > So, would you please cook a separate patch to enable this by default, so
> > it would spur the right kind of discussion on this matter ?
> 
> I will apologize in advance, but I just don't know anything about SPL or
> TPL or any other boards (outside of my very limited armv7 and armv8
> scope)....

That's OK.

> I would be happy to review and test this suggested patch (on our
> boards), but would be uncomfortable with proposing this patch.
> Please go ahead and submit a patch, and I'll check it!

The patch would go something like:

#if !defined(CONFIG_SPL_BUILD) && !defined(CONFIG_TPL_BUILD)
#define CONFIG_SYS_VSNPRINTF
#endif

and this would go into include/config_cmd_default.h . Unless I'm wrong.

Patch

diff --git a/common/Makefile b/common/Makefile
index de5cce8..daebe39 100644
--- a/common/Makefile
+++ b/common/Makefile
@@ -266,4 +266,9 @@  obj-$(CONFIG_IO_TRACE) += iotrace.o
 obj-y += memsize.o
 obj-y += stdio.o
 
+# This option is not just y/n - it can have a numeric value
+ifdef CONFIG_FASTBOOT_FLASH_MMC_DEV
+obj-y += fb_mmc.o
+endif
+
 CFLAGS_env_embedded.o := -Wa,--no-warn -DENV_CRC=$(shell tools/envcrc 2>/dev/null)
diff --git a/common/fb_mmc.c b/common/fb_mmc.c
new file mode 100644
index 0000000..9163d8c
--- /dev/null
+++ b/common/fb_mmc.c
@@ -0,0 +1,190 @@ 
+/*
+ * Copyright TODO
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#include <common.h>
+#include <fb_mmc.h>
+#include <part.h>
+#include <sparse_format.h>
+
+/* The 64 defined bytes plus \0 */
+#define RESPONSE_LEN	(64 + 1)
+
+static char *response_str;
+
+static void fastboot_resp(const char *s)
+{
+	strncpy(response_str, s, RESPONSE_LEN);
+	response_str[RESPONSE_LEN - 1] = '\0';
+}
+
+static int is_sparse_image(void *buf)
+{
+	sparse_header_t *s_header = (sparse_header_t *)buf;
+
+	if ((le32_to_cpu(s_header->magic) == SPARSE_HEADER_MAGIC) &&
+	    (le16_to_cpu(s_header->major_version) == 1))
+		return 1;
+
+	return 0;
+}
+
+static void write_sparse_image(block_dev_desc_t *dev_desc,
+		disk_partition_t *info, const char *part_name,
+		void *buffer, unsigned int download_bytes)
+{
+	lbaint_t blk;
+	lbaint_t blkcnt;
+	lbaint_t blks;
+	sparse_header_t *s_header = (sparse_header_t *)buffer;
+	chunk_header_t *c_header;
+	void *buf;
+	uint32_t blk_sz;
+	uint32_t remaining_chunks;
+	uint32_t bytes_written = 0;
+
+	blk_sz = le32_to_cpu(s_header->blk_sz);
+
+	/* verify s_header->blk_sz is exact multiple of info->blksz */
+	if (blk_sz != (blk_sz & ~(info->blksz - 1))) {
+		printf("%s: Sparse image block size issue [%u]\n",
+		       __func__, blk_sz);
+		fastboot_resp("FAILsparse image block size issue");
+		return;
+	}
+
+	if ((le32_to_cpu(s_header->total_blks) * blk_sz) >
+	    (info->size * info->blksz)) {
+		printf("%s: Sparse image is too large for the partition\n",
+		       __func__);
+		fastboot_resp("FAILsparse image is too large");
+		return;
+	}
+
+	printf("Flashing Sparse Image\n");
+
+	remaining_chunks = le32_to_cpu(s_header->total_chunks);
+	c_header = (chunk_header_t *)(buffer +
+	    le16_to_cpu(s_header->file_hdr_sz));
+	blk = info->start;
+	while (remaining_chunks) {
+		blkcnt =
+		    (le32_to_cpu(c_header->chunk_sz) * blk_sz) / info->blksz;
+
+		switch (le16_to_cpu(c_header->chunk_type)) {
+		case CHUNK_TYPE_RAW:
+			buf = (void *)c_header +
+			    le16_to_cpu(s_header->chunk_hdr_sz);
+
+			if (blk + blkcnt > info->start + info->size) {
+				printf(
+				    "%s: Request would exceed partition size!\n",
+				    __func__);
+				fastboot_resp(
+				    "FAILRequest would exceed partition size!");
+				return;
+			}
+
+			blks = dev_desc->block_write(dev_desc->dev, blk, blkcnt,
+						    buf);
+			if (blks != blkcnt) {
+				printf("%s: Write failed " LBAFU "\n",
+				       __func__, blks);
+				fastboot_resp("FAILwrite failure");
+				return;
+			}
+
+			bytes_written += blkcnt * info->blksz;
+			break;
+
+		case CHUNK_TYPE_FILL:
+		case CHUNK_TYPE_DONT_CARE:
+		case CHUNK_TYPE_CRC32:
+			/* do nothing */
+			break;
+
+		default:
+			/* error */
+			printf("%s: Unknown chunk type\n", __func__);
+			fastboot_resp("FAILunknown chunk type in sparse image");
+			return;
+		}
+
+		blk += blkcnt;
+		c_header = (chunk_header_t *)((void *)c_header +
+		    le32_to_cpu(c_header->total_sz));
+		remaining_chunks--;
+	}
+
+	printf("........ wrote %u bytes to '%s'\n", bytes_written, part_name);
+	fastboot_resp("OKAY");
+}
+
+static void write_raw_image(block_dev_desc_t *dev_desc, disk_partition_t *info,
+		const char *part_name, void *buffer,
+		unsigned int download_bytes)
+{
+	lbaint_t blkcnt;
+	lbaint_t blks;
+
+	/* determine number of blocks to write */
+	blkcnt = ((download_bytes + (info->blksz - 1)) & ~(info->blksz - 1));
+	blkcnt = blkcnt / info->blksz;
+
+	if (blkcnt > info->size) {
+		printf("%s: too large for partition: '%s'\n", __func__,
+		       part_name);
+		fastboot_resp("FAILtoo large for partition");
+		return;
+	}
+
+	printf("Flashing Raw Image\n");
+
+	blks = dev_desc->block_write(dev_desc->dev, info->start, blkcnt,
+				     buffer);
+	if (blks != blkcnt) {
+		printf("%s: failed writing to device %d\n", __func__,
+		       dev_desc->dev);
+		fastboot_resp("FAILfailed writing to device");
+		return;
+	}
+
+	printf("........ wrote " LBAFU " bytes to '%s'\n", blkcnt * info->blksz,
+	       part_name);
+	fastboot_resp("OKAY");
+}
+
+void fb_mmc_flash_write(const char *cmd, void *download_buffer,
+			unsigned int download_bytes, char *response)
+{
+	int ret;
+	block_dev_desc_t *dev_desc;
+	disk_partition_t info;
+
+	/* initialize the response buffer */
+	response_str = response;
+
+	dev_desc = mmc_get_dev(CONFIG_FASTBOOT_FLASH_MMC_DEV);
+	if (!dev_desc || dev_desc->type == DEV_TYPE_UNKNOWN) {
+		printf("%s: invalid mmc device\n", __func__);
+		fastboot_resp("FAILinvalid mmc device");
+		return;
+	}
+
+	ret = get_partition_info_efi_by_name(dev_desc, cmd, &info);
+	if (ret) {
+		printf("%s: cannot find partition: '%s'\n", __func__, cmd);
+		fastboot_resp("FAILcannot find partition");
+		return;
+	}
+
+	if (is_sparse_image(download_buffer)) {
+		write_sparse_image(dev_desc, &info, cmd, download_buffer,
+				   download_bytes);
+	} else {
+		write_raw_image(dev_desc, &info, cmd, download_buffer,
+				download_bytes);
+	}
+}
diff --git a/include/fb_mmc.h b/include/fb_mmc.h
new file mode 100644
index 0000000..1ad1d13
--- /dev/null
+++ b/include/fb_mmc.h
@@ -0,0 +1,8 @@ 
+/*
+ * Copyright 2014 Broadcom Corporation.
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+void fb_mmc_flash_write(const char *cmd, void *download_buffer,
+			unsigned int download_bytes, char *response);