| Message ID | PA4PR01MB76467B0DF885133EEC19AFD388652@PA4PR01MB7646.eurprd01.prod.exchangelabs.com |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | core: util: add retry for mount to avoid udev timing | expand |
Hi Daniel, On 13.09.24 10:41, 'Daniel Matt' via swupdate wrote: > After partitioning, there are cases in which swupdate is faster than > > udev can create the /dev/disk symbolic links. If these symlinks are used > > in the sw-description, > the devices will not be available at that time, > > causing the update to fail. Implementing a retry mechanism resolves this > > issue. This is a specific use case. > > Signed-off-by: Daniel Matt <daniel.matt@kuka.com> > > --- > > core/util.c | 17 ++++++++++++++++- > > 1 file changed, 16 insertions(+), 1 deletion(-) > > diff --git a/core/util.c b/core/util.c > > index 2316a57..75810a3 100644 > > --- a/core/util.c > > +++ b/core/util.c > > @@ -52,6 +52,8 @@ struct decryption_key { > > unsigned char ivt[AES_BLK_SIZE]; > > }; > > +#define MAX_AMOUNT_MOUNT_TRIES 2 It seems to me a sort of hack / work-around to fix the issue on just your project. The first question is why is 2 a good number (apart it fixes your project), and what happens on different projects where it takes maybe longer. And this is a second work-around to solve this. So, when a new partition table is generated, kernel must reread it. This is already forced by SWUpdate, but of course the new table should be propagated to all processes like udev that are using it. The cause then is in the new partition table, not in the mount command. https://github.com/sbabic/swupdate/blob/master/handlers/diskpart_handler.c#L1467 Your patch is just doing the same, adding an additional second but somewhere else in code (the additional sleep is what fixes your issue). It is then clear that a "default" sleep cannot be set, and I can imagine projects then where even your patch is not enough. Then we can have another solution and we can make the sleep programmable and set explicitely the time into sw-description. Something like this: partitions: ( { type = "diskpart"; device = <....>; properties: { labeltype = "gpt"; partition-1 = ["size=2G", ....]; partition-2 = ["name=OEM",....]; wait = "3"; // Time for system to reload part table } This should solve your issue and can solve any other project that requires more time instead of hard-coding as it is now. What do you think ? Best regards, Stefano Babic > > + > > static struct decryption_key *aes_key = NULL; > > /* > > @@ -757,7 +759,20 @@ bool strtobool(const char *s) > > int swupdate_mount(const char *device, const char *dir, const char *fstype) > > { > > #if defined(__linux__) > > - return mount(device, dir, fstype, 0, NULL); > > + int ret; > > + > > + for (int i = 0; i < MAX_AMOUNT_MOUNT_TRIES; i++) { > > + ret = mount(device, dir, fstype, 0, NULL); > > + > > + if (ret == -1 && errno == ENOENT) { > > + WARN("Device %s not found, > retry after delay", device); > > + sleep(1); > > + } else { > > + break; > > + } > > + } > > + > > + return ret; > > #elif defined(__FreeBSD__) > > int iovlen = 8; > > struct iovec iov[iovlen]; > > -- > > 2.39.2 > > > Internal > > -- > You received this message because you are subscribed to the Google > Groups "swupdate" group. > To unsubscribe from this group and stop receiving emails from it, send > an email to swupdate+unsubscribe@googlegroups.com > <mailto:swupdate+unsubscribe@googlegroups.com>. > To view this discussion on the web visit > https://groups.google.com/d/msgid/swupdate/PA4PR01MB76467B0DF885133EEC19AFD388652%40PA4PR01MB7646.eurprd01.prod.exchangelabs.com <https://groups.google.com/d/msgid/swupdate/PA4PR01MB76467B0DF885133EEC19AFD388652%40PA4PR01MB7646.eurprd01.prod.exchangelabs.com?utm_medium=email&utm_source=footer>.
diff --git a/core/util.c b/core/util.c index 2316a57..75810a3 100644 --- a/core/util.c +++ b/core/util.c @@ -52,6 +52,8 @@ struct decryption_key { unsigned char ivt[AES_BLK_SIZE]; }; +#define MAX_AMOUNT_MOUNT_TRIES 2 + static struct decryption_key *aes_key = NULL; /* @@ -757,7 +759,20 @@ bool strtobool(const char *s) int swupdate_mount(const char *device, const char *dir, const char *fstype)
After partitioning, there are cases in which swupdate is faster than udev can create the /dev/disk symbolic links. If these symlinks are used in the sw-description, the devices will not be available at that time, causing the update to fail. Implementing a retry mechanism resolves this issue. Signed-off-by: Daniel Matt <daniel.matt@kuka.com> --- core/util.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) { #if defined(__linux__) - return mount(device, dir, fstype, 0, NULL); + int ret; + + for (int i = 0; i < MAX_AMOUNT_MOUNT_TRIES; i++) { + ret = mount(device, dir, fstype, 0, NULL); + + if (ret == -1 && errno == ENOENT) { + WARN("Device %s not found, retry after delay", device); + sleep(1); + } else { + break; + } + } + + return ret; #elif defined(__FreeBSD__) int iovlen = 8; struct iovec iov[iovlen]; -- 2.39.2 Internal