On 6/20/2025 9:35 PM, Michael Kelley wrote:

From: Naman Jain namjain@linux.microsoft.com Sent: Friday, June 20, 2025 12:06 AM

...

Size of ring buffer, as defined in uio_hv_generic driver, is no longer fixed to 16 KB. This creates a problem in fcopy, since this size was hardcoded. With the change in place to make ring sysfs node actually reflect the size of underlying ring buffer, it is safe to get the size of ring sysfs file and use it for ring buffer size in fcopy daemon. Fix the issue of disparity in ring buffer size, by making it dynamic in fcopy uio daemon.

Cc: stable@vger.kernel.org Fixes: 0315fef2aff9 ("uio_hv_generic: Align ring size to system page") Signed-off-by: Naman Jain namjain@linux.microsoft.com

---

tools/hv/hv_fcopy_uio_daemon.c | 65 ++++++++++++++++++++++++++++++---- 1 file changed, 58 insertions(+), 7 deletions(-)

diff --git a/tools/hv/hv_fcopy_uio_daemon.c b/tools/hv/hv_fcopy_uio_daemon.c index 0198321d14a2..da2b27d6af0e 100644 --- a/tools/hv/hv_fcopy_uio_daemon.c +++ b/tools/hv/hv_fcopy_uio_daemon.c @@ -36,6 +36,7 @@ #define WIN8_SRV_VERSION (WIN8_SRV_MAJOR << 16 | WIN8_SRV_MINOR)

#define FCOPY_UIO "/sys/bus/vmbus/devices/eb765408-105f-49b6-b4aa-c123b64d17d4/uio" +#define FCOPY_CHANNELS_PATH "/sys/bus/vmbus/devices/eb765408-105f-49b6-b4aa-c123b64d17d4/channels"

#define FCOPY_VER_COUNT 1 static const int fcopy_versions[] = { @@ -47,9 +48,51 @@ static const int fw_versions[] = { UTIL_FW_VERSION };

-#define HV_RING_SIZE 0x4000 /* 16KB ring buffer size */ +#define HV_RING_SIZE_DEFAULT 0x4000 /* 16KB ring buffer size default */

-static unsigned char desc[HV_RING_SIZE]; +static uint32_t get_ring_buffer_size(void) +{

• char ring_path[PATH_MAX];
• DIR *dir;
• struct dirent *entry;
• struct stat st;
• uint32_t ring_size = 0;
• /* Find the channel directory */
• dir = opendir(FCOPY_CHANNELS_PATH);
• if (!dir) {

• syslog(LOG_ERR, "Failed to open channels directory, using default ring size");
  

This is where the previous long discussion about racing with user space comes into play. While highly unlikely, it's possible that the "opendir" could fail because of racing with the kernel thread that creates the "channels" directory. The right thing to do would be to sleep for some period of time, then try again. Sleeping for 1 second would be a very generous -- could also go with something like 100 milliseconds.

Makes sense, will add that logic.

...

• return HV_RING_SIZE_DEFAULT;
  

• }
• while ((entry = readdir(dir)) != NULL) {

• if (entry->d_type == DT_DIR && strcmp(entry->d_name, ".") != 0 &&
  

•     strcmp(entry->d_name, "..") != 0) {
  

• 	snprintf(ring_path, sizeof(ring_path), "%s/%s/ring",
  

• 		 FCOPY_CHANNELS_PATH, entry->d_name);
  

• 	if (stat(ring_path, &st) == 0) {
  

• 		/* stat returns size of Tx, Rx rings combined, so take half of it */
  

• 		ring_size = (uint32_t)st.st_size / 2;
  

• 		syslog(LOG_INFO, "Ring buffer size from %s: %u bytes",
  

• 		       ring_path, ring_size);
  

• 		break;
  

• 	}
  

• }
  

• }

The same race problem could happen with this loop. The "channels" directory might have been created, but the entry for the numbered channel might not. The loop could exit having found only "." and "..". Again, if no numbered channel is found, sleep for a short period of time and try again.

Will cover this too.

...

• closedir(dir);
• if (!ring_size) {

• ring_size = HV_RING_SIZE_DEFAULT;
  

• syslog(LOG_ERR, "Could not determine ring size, using default: %u bytes",
  

•        HV_RING_SIZE_DEFAULT);
  

• }
• return ring_size;

+}

+static unsigned char *desc;

static int target_fd; static char target_fname[PATH_MAX]; @@ -406,7 +449,8 @@ int main(int argc, char *argv[]) int daemonize = 1, long_index = 0, opt, ret = -EINVAL; struct vmbus_br txbr, rxbr; void *ring;

• uint32_t len = HV_RING_SIZE;

• uint32_t ring_size = get_ring_buffer_size();

Getting the ring buffer size before even the command line options are parsed could produce unexpected results. For example, if someone just wanted to see the usage (the -h option), they might get an error about not being able to get the ring size. I'd suggest doing this later, after the /dev/uio<N> entry is successfully opened.

Thanks for pointing this out, I'll take care of it in next version.

Regards, Naman

...

• uint32_t len = ring_size; char uio_name[NAME_MAX] = {0}; char uio_dev_path[PATH_MAX] = {0};

@@ -416,6 +460,13 @@ int main(int argc, char *argv[]) {0, 0, 0, 0 } };

• desc = (unsigned char *)malloc(ring_size * sizeof(unsigned char));
• if (!desc) {

• syslog(LOG_ERR, "malloc failed for desc buffer");
  

• ret = -ENOMEM;
  

• goto exit;
  

• }
• while ((opt = getopt_long(argc, argv, "hn", long_options, &long_index)) != -1) { switch (opt) {

@@ -448,14 +499,14 @@ int main(int argc, char *argv[]) goto exit; }

• ring = vmbus_uio_map(&fcopy_fd, HV_RING_SIZE);

• ring = vmbus_uio_map(&fcopy_fd, ring_size); if (!ring) { ret = errno; syslog(LOG_ERR, "mmap ringbuffer failed; error: %d %s", ret, strerror(ret)); goto close; }

• vmbus_br_setup(&txbr, ring, HV_RING_SIZE);
• vmbus_br_setup(&rxbr, (char *)ring + HV_RING_SIZE, HV_RING_SIZE);

• vmbus_br_setup(&txbr, ring, ring_size);

• vmbus_br_setup(&rxbr, (char *)ring + ring_size, ring_size);

  rxbr.vbr->imask = 0;

@@ -472,7 +523,7 @@ int main(int argc, char *argv[]) goto close; }

• len = HV_RING_SIZE;
  

• len = ring_size;
  

  ret = rte_vmbus_chan_recv_raw(&rxbr, desc, &len); if (unlikely(ret <= 0)) { /* This indicates a failure to communicate (or worse) */

base-commit: bc6e0ba6c9bafa6241b05524b9829808056ac4ad

2.34.1