[PATCH v2] scsi: target: tcmu: add compat mode for 32bit userspace on 64bit kernel

[PATCH v2] scsi: target: tcmu: add compat mode for 32bit userspace on 64bit kernel

[Bodo Stroesser]

When using tcmu it might happen, that userspace application cannot
be built as 64 bit program even on a 64 bit host due to existing
32 bit libraries that must be used, e.g. for compression, encryption,
deduplication, ...

Currently this only works with manual changes in userspace include
file target_core_user.h due to a missing padding field in
struct tcmu_cmd_entry. Additionally one has to change uio.h because
32-bit userspace interprets struct iovec to have 8 byte size while
64-bit kernel creates it with 16 byte size.

This patch adds CONFIG_COMPAT support for X86_64 in tcmu to avoid
header file editing.

During mmap() of the command ring and data area of tcmu_dev's
uio device, tcmu checks and saves the value returned from
in_compat_syscall(). That way it allows multiple tasks to mmap()
only if they are of same type (32-/64-bit).

During SCSI command processing tcmu now creates tcmu_cmd_entry
records according to the saved type of the userspace program.

Offset and size of data fields in tcmu_cmd_entry differ between
32-bit and 64-bit only in the req part.
The field cdb_off (__u64) is on a 4-byte boundary in 32-bit, while
in 64-bit it is on the next 8-byte boundary.
The same is true for the start of the "struct iovec iov[]" array.
It also contains 8 vs. 16 byte per array entry in 32-bit vs 64-bit
mode.

Since differences are not too many, I mostly inserted changes into
existing code where necessary instead of writing a full set of
compat_* functions.

This patch is tested on x86_64.

Signed-off-by: Bodo Stroesser <bstroesser@ts.fujitsu.com>
---

This patch is made on top of Martin's 5.10/scsi-queue tree

History:

- First attempt was a PATCH RFC that didn't compile correctly in
  32-bit kernel, but in 64bit kernel only.

- The first PATCH was fixed to compile in 32bit kernel and in
  64bit kernel with and without CONFIG_COMPAT.
  Additionally I changed it to support CONFIG_COMPAT for X86_64
  only, since I wasn't able to find out what field sizes and
  offsets if struct tcmu_cmd_entry userspace programs on other
  archs would expect in compat mode.

- V2:
  Meanwhile I found out, that the compat handling in the patch
  should be fine for all ARCHes:
  * see iovec vs. compat_iovec handling in fs/io_uring.c
  * see usage of "__attribute__ ((__packed__))" in
    struct btrfs_ioctl_send_args_32 in fs/btrfs/ioctl.c
    vs. struct btrfs_ioctl_send_args in include/uapi/linux/btrfs.h
  So I re-allowed COMPAT support for all ARCHes.
---
 drivers/target/target_core_user.c | 170 +++++++++++++++++++++++++++++++++-----
 1 file changed, 149 insertions(+), 21 deletions(-)

diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index 9b7592350502..5fe73e5ab5a7 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -24,6 +24,15 @@
 #include <net/genetlink.h>
 #include <scsi/scsi_common.h>
 #include <scsi/scsi_proto.h>
+
+#if defined(CONFIG_COMPAT) && defined(CONFIG_64BIT)
+#define TCMU_COMPAT
+#endif
+
+#ifdef TCMU_COMPAT
+#include <linux/compat.h>
+#endif
+
 #include <target/target_core_base.h>
 #include <target/target_core_fabric.h>
 #include <target/target_core_backend.h>
@@ -136,6 +145,11 @@ struct tcmu_dev {
 	uint32_t max_blocks;
 	size_t ring_size;
 
+#ifdef TCMU_COMPAT
+	bool compat;
+	bool new_open;
+#endif
+
 	struct mutex cmdr_lock;
 	struct list_head qfull_queue;
 	struct list_head tmr_queue;
@@ -193,6 +207,30 @@ struct tcmu_tmr {
 	uint32_t tmr_cmd_cnt;
 	int16_t tmr_cmd_ids[0];
 };
+#ifdef TCMU_COMPAT
+struct tcmu_compat_cmd_entry {
+	struct tcmu_cmd_entry_hdr hdr;
+
+	union {
+		struct {
+			__u32 iov_cnt;
+			__u32 iov_bidi_cnt;
+			__u32 iov_dif_cnt;
+			__u64 cdb_off;
+			__u64 __pad1;
+			__u64 __pad2;
+			struct compat_iovec iov[0];
+		} __packed req;
+		struct {
+			__u8 scsi_status;
+			__u8 __pad1;
+			__u16 __pad2;
+			__u32 read_len;
+			char sense_buffer[TCMU_SENSE_BUFFERSIZE];
+		} rsp;
+	};
+} __packed;
+#endif
 
 /*
  * To avoid dead lock the mutex lock order should always be:
@@ -671,6 +709,26 @@ static inline size_t iov_tail(struct iovec *iov)
 	return (size_t)iov->iov_base + iov->iov_len;
 }
 
+#ifdef TCMU_COMPAT
+static inline void compat_new_iov(struct iovec **iov, int *iov_cnt)
+{
+	struct compat_iovec **c_iov = (struct compat_iovec **)iov;
+
+	if (*iov_cnt != 0)
+		(*c_iov)++;
+	(*iov_cnt)++;
+
+	memset(*c_iov, 0, sizeof(struct compat_iovec));
+}
+
+static inline size_t compat_iov_tail(struct iovec *iov)
+{
+	struct compat_iovec *c_iov = (struct compat_iovec *)iov;
+
+	return (size_t)c_iov->iov_base + c_iov->iov_len;
+}
+#endif
+
 static void scatter_data_area(struct tcmu_dev *udev,
 	struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
 	unsigned int data_nents, struct iovec **iov,
@@ -705,13 +763,41 @@ static void scatter_data_area(struct tcmu_dev *udev,
 			to_offset = get_block_offset_user(udev, dbi,
 					block_remaining);
 
+			copy_bytes = min_t(size_t, sg_remaining,
+					block_remaining);
+			if (copy_data) {
+				offset = DATA_BLOCK_SIZE - block_remaining;
+				memcpy(to + offset,
+				       from + sg->length - sg_remaining,
+				       copy_bytes);
+			}
+			sg_remaining -= copy_bytes;
+			block_remaining -= copy_bytes;
+
 			/*
 			 * The following code will gather and map the blocks
 			 * to the same iovec when the blocks are all next to
 			 * each other.
 			 */
-			copy_bytes = min_t(size_t, sg_remaining,
-					block_remaining);
+#ifdef TCMU_COMPAT
+			if (udev->compat) {
+				struct compat_iovec *c_iov;
+
+				if (*iov_cnt != 0 &&
+				    to_offset = compat_iov_tail(*iov)) {
+					c_iov = (struct compat_iovec *)*iov;
+					c_iov->iov_len += copy_bytes;
+				} else {
+					compat_new_iov(iov, iov_cnt);
+					c_iov = (struct compat_iovec *)*iov;
+					c_iov->iov_base +						(compat_uptr_t)to_offset;
+					c_iov->iov_len = copy_bytes;
+				}
+				continue;
+			}
+#endif
+
 			if (*iov_cnt != 0 &&
 			    to_offset = iov_tail(*iov)) {
 				/*
@@ -730,16 +816,6 @@ static void scatter_data_area(struct tcmu_dev *udev,
 				(*iov)->iov_base = (void __user *)to_offset;
 				(*iov)->iov_len = copy_bytes;
 			}
-
-			if (copy_data) {
-				offset = DATA_BLOCK_SIZE - block_remaining;
-				memcpy(to + offset,
-				       from + sg->length - sg_remaining,
-				       copy_bytes);
-			}
-
-			sg_remaining -= copy_bytes;
-			block_remaining -= copy_bytes;
 		}
 		kunmap_atomic(from - sg->offset);
 	}
@@ -879,8 +955,15 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
 	return tcmu_get_empty_blocks(udev, cmd);
 }
 
-static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
+static inline size_t tcmu_cmd_get_base_cmd_size(struct tcmu_dev *tcmu_dev,
+						size_t iov_cnt)
 {
+#ifdef TCMU_COMPAT
+	if (tcmu_dev->compat) {
+		return max(offsetof(struct tcmu_compat_cmd_entry, req.iov[iov_cnt]),
+				sizeof(struct tcmu_compat_cmd_entry));
+	}
+#endif
 	return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
 			sizeof(struct tcmu_cmd_entry));
 }
@@ -1016,7 +1099,7 @@ static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 	 * The size will be recalculated later as actually needed to save
 	 * cmd area memories.
 	 */
-	base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
+	base_command_size = tcmu_cmd_get_base_cmd_size(udev, tcmu_cmd->dbi_cnt);
 	command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 
 	if (!list_empty(&udev->qfull_queue))
@@ -1049,7 +1132,15 @@ static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 
 	/* Handle allocating space from the data area */
 	tcmu_cmd_reset_dbi_cur(tcmu_cmd);
-	iov = &entry->req.iov[0];
+#ifdef TCMU_COMPAT
+	if (udev->compat) {
+		struct compat_iovec *c_iov;
+
+		c_iov = &((struct tcmu_compat_cmd_entry *)entry)->req.iov[0];
+		iov = (struct iovec *)c_iov;
+	} else
+#endif
+		iov = &entry->req.iov[0];
 	iov_cnt = 0;
 	copy_to_data_area = (se_cmd->data_direction = DMA_TO_DEVICE
 		|| se_cmd->se_cmd_flags & SCF_BIDI);
@@ -1061,7 +1152,12 @@ static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 	/* Handle BIDI commands */
 	iov_cnt = 0;
 	if (se_cmd->se_cmd_flags & SCF_BIDI) {
-		iov++;
+#ifdef TCMU_COMPAT
+		if (udev->compat)
+			iov = (struct iovec *)((struct compat_iovec *)iov + 1);
+		else
+#endif
+			iov++;
 		scatter_data_area(udev, tcmu_cmd, se_cmd->t_bidi_data_sg,
 				  se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
 				  false);
@@ -1089,7 +1185,7 @@ static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 	 * Recalaulate the command's base size and size according
 	 * to the actual needs
 	 */
-	base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
+	base_command_size = tcmu_cmd_get_base_cmd_size(udev, entry->req.iov_cnt +
 						       entry->req.iov_bidi_cnt);
 	command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
 
@@ -1098,7 +1194,14 @@ static int queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, sense_reason_t *scsi_err)
 	/* All offsets relative to mb_addr, not start of entry! */
 	cdb_off = CMDR_OFF + cmd_head + base_command_size;
 	memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
-	entry->req.cdb_off = cdb_off;
+
+#ifdef TCMU_COMPAT
+	if (udev->compat)
+		((struct tcmu_compat_cmd_entry *)entry)->req.cdb_off = cdb_off;
+	else
+#endif
+		entry->req.cdb_off = cdb_off;
+
 	tcmu_flush_dcache_range(entry, command_size);
 
 	UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
@@ -1729,6 +1832,8 @@ static const struct vm_operations_struct tcmu_vm_ops = {
 	.fault = tcmu_vma_fault,
 };
 
+static void __tcmu_reset_ring(struct tcmu_dev *, u8);
+
 static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
 {
 	struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
@@ -1742,6 +1847,23 @@ static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
 	if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
 		return -EINVAL;
 
+#ifdef TCMU_COMPAT
+	mutex_lock(&udev->cmdr_lock);
+	if (udev->new_open) {
+		if (in_compat_syscall() != udev->compat) {
+			udev->compat = !udev->compat;
+			__tcmu_reset_ring(udev, 1);
+		}
+		udev->new_open = false;
+	} else if (in_compat_syscall() != udev->compat) {
+		mutex_unlock(&udev->cmdr_lock);
+		return -EINVAL;
+	}
+	mutex_unlock(&udev->cmdr_lock);
+
+	pr_debug("mmap() successful on %s, compat = %d\n", udev->name, udev->compat);
+#endif
+
 	return 0;
 }
 
@@ -1752,6 +1874,9 @@ static int tcmu_open(struct uio_info *info, struct inode *inode)
 	/* O_EXCL not supported for char devs, so fake it? */
 	if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
 		return -EBUSY;
+#ifdef TCMU_COMPAT
+	udev->new_open = true;
+#endif
 
 	udev->inode = inode;
 	kref_get(&udev->kref);
@@ -2209,14 +2334,12 @@ static void tcmu_block_dev(struct tcmu_dev *udev)
 	mutex_unlock(&udev->cmdr_lock);
 }
 
-static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
+static void __tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
 {
 	struct tcmu_mailbox *mb;
 	struct tcmu_cmd *cmd;
 	int i;
 
-	mutex_lock(&udev->cmdr_lock);
-
 	idr_for_each_entry(&udev->commands, cmd, i) {
 		pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
 			  cmd->cmd_id, udev->name,
@@ -2265,7 +2388,12 @@ static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
 	tcmu_remove_all_queued_tmr(udev);
 
 	run_qfull_queue(udev, false);
+}
 
+static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
+{
+	mutex_lock(&udev->cmdr_lock);
+	__tcmu_reset_ring(udev, err_level);
 	mutex_unlock(&udev->cmdr_lock);
 }
 
-- 
2.12.3

Re: [PATCH v2] scsi: target: tcmu: add compat mode for 32bit userspace on 64bit kernel

[Mike Christie]

On 9/3/20 12:41 PM, Bodo Stroesser wrote:
> +#ifdef TCMU_COMPAT
> +static inline void compat_new_iov(struct iovec **iov, int *iov_cnt)
> +{
> +	struct compat_iovec **c_iov = (struct compat_iovec **)iov;
> +
> +	if (*iov_cnt != 0)
> +		(*c_iov)++;
> +	(*iov_cnt)++;
> +
> +	memset(*c_iov, 0, sizeof(struct compat_iovec));
> +}
> +
> +static inline size_t compat_iov_tail(struct iovec *iov)
> +{
> +	struct compat_iovec *c_iov = (struct compat_iovec *)iov;
> +
> +	return (size_t)c_iov->iov_base + c_iov->iov_len;
> +}
> +#endif
> +
>  static void scatter_data_area(struct tcmu_dev *udev,
>  	struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
>  	unsigned int data_nents, struct iovec **iov,
> @@ -705,13 +763,41 @@ static void scatter_data_area(struct tcmu_dev *udev,
>  			to_offset = get_block_offset_user(udev, dbi,
>  					block_remaining);
>  
> +			copy_bytes = min_t(size_t, sg_remaining,
> +					block_remaining);
> +			if (copy_data) {
> +				offset = DATA_BLOCK_SIZE - block_remaining;
> +				memcpy(to + offset,
> +				       from + sg->length - sg_remaining,
> +				       copy_bytes);
> +			}
> +			sg_remaining -= copy_bytes;
> +			block_remaining -= copy_bytes;
> +
>  			/*
>  			 * The following code will gather and map the blocks
>  			 * to the same iovec when the blocks are all next to
>  			 * each other.
>  			 */
> -			copy_bytes = min_t(size_t, sg_remaining,
> -					block_remaining);
> +#ifdef TCMU_COMPAT
> +			if (udev->compat) {
> +				struct compat_iovec *c_iov;
> +
> +				if (*iov_cnt != 0 &&
> +				    to_offset = compat_iov_tail(*iov)) {
> +					c_iov = (struct compat_iovec *)*iov;
> +					c_iov->iov_len += copy_bytes;
> +				} else {
> +					compat_new_iov(iov, iov_cnt);
> +					c_iov = (struct compat_iovec *)*iov;
> +					c_iov->iov_base > +						(compat_uptr_t)to_offset;
> +					c_iov->iov_len = copy_bytes;
> +				}
> +				continue;
> +			}
> +#endif

I think a couple one or two liner ifdefs in the middle of a function is something people look the other way on. Maybe here we went a little too wild though. If we are not going the callout route, I think it would be nicer here to separate this part into functions. In the ifdef with compat_new_iov/compat_iov_tail above this function, add a tcmu_compat_iov_set and move the code above there. Move the code below it that does the non compat code to a new function tcmu_iov_set. We will still need a ifdef for the tcmu_compat_iov_set somewhere, but it feels more organized.

Re: [PATCH v2] scsi: target: tcmu: add compat mode for 32bit userspace on 64bit kernel

[Bodo Stroesser]

On 2020-09-08 01:14, Mike Christie wrote:
> On 9/3/20 12:41 PM, Bodo Stroesser wrote:
>> +#ifdef TCMU_COMPAT
>> +static inline void compat_new_iov(struct iovec **iov, int *iov_cnt)
>> +{
>> +	struct compat_iovec **c_iov = (struct compat_iovec **)iov;
>> +
>> +	if (*iov_cnt != 0)
>> +		(*c_iov)++;
>> +	(*iov_cnt)++;
>> +
>> +	memset(*c_iov, 0, sizeof(struct compat_iovec));
>> +}
>> +
>> +static inline size_t compat_iov_tail(struct iovec *iov)
>> +{
>> +	struct compat_iovec *c_iov = (struct compat_iovec *)iov;
>> +
>> +	return (size_t)c_iov->iov_base + c_iov->iov_len;
>> +}
>> +#endif
>> +
>>   static void scatter_data_area(struct tcmu_dev *udev,
>>   	struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
>>   	unsigned int data_nents, struct iovec **iov,
>> @@ -705,13 +763,41 @@ static void scatter_data_area(struct tcmu_dev *udev,
>>   			to_offset = get_block_offset_user(udev, dbi,
>>   					block_remaining);
>>   
>> +			copy_bytes = min_t(size_t, sg_remaining,
>> +					block_remaining);
>> +			if (copy_data) {
>> +				offset = DATA_BLOCK_SIZE - block_remaining;
>> +				memcpy(to + offset,
>> +				       from + sg->length - sg_remaining,
>> +				       copy_bytes);
>> +			}
>> +			sg_remaining -= copy_bytes;
>> +			block_remaining -= copy_bytes;
>> +
>>   			/*
>>   			 * The following code will gather and map the blocks
>>   			 * to the same iovec when the blocks are all next to
>>   			 * each other.
>>   			 */
>> -			copy_bytes = min_t(size_t, sg_remaining,
>> -					block_remaining);
>> +#ifdef TCMU_COMPAT
>> +			if (udev->compat) {
>> +				struct compat_iovec *c_iov;
>> +
>> +				if (*iov_cnt != 0 &&
>> +				    to_offset = compat_iov_tail(*iov)) {
>> +					c_iov = (struct compat_iovec *)*iov;
>> +					c_iov->iov_len += copy_bytes;
>> +				} else {
>> +					compat_new_iov(iov, iov_cnt);
>> +					c_iov = (struct compat_iovec *)*iov;
>> +					c_iov->iov_base >> +						(compat_uptr_t)to_offset;
>> +					c_iov->iov_len = copy_bytes;
>> +				}
>> +				continue;
>> +			}
>> +#endif
> 
> I think a couple one or two liner ifdefs in the middle of a function is something people look the other way on. Maybe here we went a little too wild though. If we are not going the callout route, I think it would be nicer here to separate this part into functions. In the ifdef with compat_new_iov/compat_iov_tail above this function, add a tcmu_compat_iov_set and move the code above there. Move the code below it that does the non compat code to a new function tcmu_iov_set. We will still need a ifdef for the tcmu_compat_iov_set somewhere, but it feels more organized.
> 

Ok. I agree.

To be honest, I already prepared a new patch series allowing to
have buffer chunks bigger than one page in data area of tcmu's
uio device. While tcmu still uses single page allocation, the
change allows to have shorter iovec lists, less buffer bits to
maintain, and if necessary allows userspace to choose buffer chunks
as big as max. expected IO data size, which from userspace perspective
makes every IO up to that chosen data size appear in a single
consecutive buffer (1 iovec only)

The first patches of that series already do some re-work and
optimization of queue_cmd_ring and scatter_data_area such, that we
end up with something very similar to what you suggest :)

So instead of this patch I'll soon send a short series consisting
of those 3 patches and a new version of the current patch.
I hope, that series will look more organized.

Thank you,
Bodo