<div dir="ltr"><div dir="ltr">Hi Sean </div> <div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Sep 9, 2022 at 10:12 PM Sean Anderson <<a href="mailto:sean.anderson@seco.com">sean.anderson@seco.com</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">I ran into this issue today. This patch fixes things, so Tested-by: Sean Anderson <<a href="mailto:sean.anderson@seco.com" target="_blank">sean.anderson@seco.com</a>> However, I think the implementation leaves a bit to be desired... On 4/6/22 11:46 AM, Fabio Aiuto wrote: > From: Michael Trimarchi <<a href="mailto:michael@amarulasolutions.com" target="_blank">michael@amarulasolutions.com</a>> > > The device driver can be deferrable and can be a race during > the dm-init early. We need to wait all the probe are really finished > in a loop as is done in do_mounts. This is was tested on kernel 5.4 > but code seems was not changed since that time > </blockquote><div> </div><div>Ok, I know, I was waiting for some feedback. I will try to rework</div><div> </div><div>Michael</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> > 003: imx8mq-usb-phy 381f0040.usb-phy: 381f0040.usb-phy supply vbus not found, using dummy regulator > 003: imx8mq-usb-phy 382f0040.usb-phy: 382f0040.usb-phy supply vbus not found, using dummy regulator > 003: imx-cpufreq-dt imx-cpufreq-dt: cpu speed grade 5 mkt segment 0 supported-hw 0x20 0x1 > 003: caam-dma caam-dma: caam dma support with 2 job rings > 000: hctosys: unable to open rtc device (rtc0) > 000: device-mapper: init: waiting for all devices to be available before creating mapped devices > 002: device-mapper: table: 254:0: verity: Data device lookup failed > 002: device-mapper: ioctl: error adding target to table > 002: crng init done > 003: of_cfs_init > 003: of_cfs_init: OK > 003: Waiting for root device /dev/dm-0... > 001: mmc2: new HS400 Enhanced strobe MMC card at address 0001 > 001: mmcblk2: mmc2:0001 IB2916 14.6 GiB > 001: mmcblk2boot0: mmc2:0001 IB2916 partition 1 4.00 MiB > 001: mmcblk2boot1: mmc2:0001 IB2916 partition 2 4.00 MiB > 001: mmcblk2rpmb: mmc2:0001 IB2916 partition 3 4.00 MiB, chardev (249:0) > 001: mmcblk2: p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 > 001: VSD_3V3: disabling > > with the patch > > 003: device-mapper: init: waiting for all devices to be available before creating mapped devices > > 000: device-mapper: table: 254:0: verity: Data device lookup failed > 000: device-mapper: ioctl: error adding target to table > 002: crng init done > 003: device-mapper: init: waiting for all devices to be available before creating mapped devices > 003: device-mapper: table: 254:0: verity: Data device lookup failed > 003: device-mapper: ioctl: error adding target to table > 003: device-mapper: init: waiting for all devices to be available before creating mapped devices > 000: device-mapper: table: 254:0: verity: Data device lookup failed > 000: device-mapper: ioctl: error adding target to table > 002: device-mapper: init: waiting for all devices to be available before creating mapped devices > 002: device-mapper: table: 254:0: verity: Data device lookup failed > 002: device-mapper: ioctl: error adding target to table > 000: device-mapper: init: waiting for all devices to be available before creating mapped devices > 000: device-mapper: table: 254:0: verity: Data device lookup failed > 000: device-mapper: ioctl: error adding target to table For example, we repeatedly print these errors, even though there is not really an error condition. Additionally, we still loop even if (e.g.) there is a syntax error in the dm-mod.create parameter. It also introduces similar problems to rootdelay, where you have to choose a suitable maximum value. I would much rather see something closer to rootwait (as implemented in prepare_namespace). To do that, I think it is better to modify dm_get_device or dm_get_dev_t, which are closer to the source of the error. One issue is that these functions are not just called from dm_early_create, but also at runtime. So we can't just blindly start waiting for devices to show up if they don't exist and dm-mod.wait (or whatever) is missing. We could solve this by introducing a variable which is cleared at the end of dm_early_create. I think that solution is OK. We could also modify dm_early_create to pass a parameter to dm_table_add_target which indicates that the target is being created "early." However, I think this would be fairly disruptive to existing code. --Sean > 003: mmc2: new HS400 Enhanced strobe MMC card at address 0001 > 003: mmcblk2: mmc2:0001 DG4016 14.7 GiB > 003: mmcblk2boot0: mmc2:0001 DG4016 partition 1 4.00 MiB > 003: mmcblk2boot1: mmc2:0001 DG4016 partition 2 4.00 MiB > 003: mmcblk2rpmb: mmc2:0001 DG4016 partition 3 4.00 MiB, chardev (249:0) > 003: mmcblk2: p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 > 002: device-mapper: init: waiting for all devices to be available before creating mapped devices > 003: device-mapper: verity: sha256 using implementation "sha256-caam" > 000: device-mapper: ioctl: dm-0 (rootfs) is ready > > Wait loop is limited to 10 at the moment for our use case showed no > more than 4 loops before successfully find data device. > > Signed-off-by: Michael Trimarchi <<a href="mailto:michael@amarulasolutions.com" target="_blank">michael@amarulasolutions.com</a>> > Signed-off-by: Fabio Aiuto <<a href="mailto:fabio.aiuto@amarulasolutions.com" target="_blank">fabio.aiuto@amarulasolutions.com</a>> > --- > Changes from v1: > - limit the loop to 10 iterations > - change variable names > - check only for -ENODEV failures > > Changes from v2: > - use a limit in seconds (not in retry > number) > - add a parameter > - update docs > > .../admin-guide/device-mapper/dm-init.rst | 13 +++++++++++ > drivers/md/dm-init.c | 23 +++++++++++++++++-- > 2 files changed, 34 insertions(+), 2 deletions(-) > > diff --git a/Documentation/admin-guide/device-mapper/dm-init.rst b/Documentation/admin-guide/device-mapper/dm-init.rst > index e5242ff17e9b..5c2f2bf1db03 100644 > --- a/Documentation/admin-guide/device-mapper/dm-init.rst > +++ b/Documentation/admin-guide/device-mapper/dm-init.rst > @@ -123,3 +123,16 @@ Other examples (per target): > 0 1638400 verity 1 8:1 8:2 4096 4096 204800 1 sha256 > fb1a5a0f00deb908d8b53cb270858975e76cf64105d412ce764225d53b8f3cfd > 51934789604d1b92399c52e7cb149d1b3a1b74bbbcb103b2a0aaacbed5c08584 > + > +Delay for waiting deferred probes of block devices > +================================================== > + > +Sometimes the late initcall starting the early creation of mapped > +devices, starts too early. A loop waiting for probing of block > +devices has been added; the default maximum delay is 1 second but > +it can be set through the following kernel command:: > + > + dm-mod.delay=<seconds> > + > +This allows the procedure to retry the creation of a mapped device > +after a short wait (5 msecs). > diff --git a/drivers/md/dm-init.c b/drivers/md/dm-init.c > index b0c45c6ebe0b..f4c5b4a46001 100644 > --- a/drivers/md/dm-init.c > +++ b/drivers/md/dm-init.c > @@ -7,7 +7,9 @@ > * This file is released under the GPLv2. > */ > > +#include <linux/async.h> > #include <linux/ctype.h> > +#include <linux/delay.h> > #include <linux/device.h> > #include <linux/device-mapper.h> > #include <linux/init.h> > @@ -18,8 +20,10 @@ > #define DM_MAX_DEVICES 256 > #define DM_MAX_TARGETS 256 > #define DM_MAX_STR_SIZE 4096 > +#define DM_DEFAULT_MAX_PROBE_DELAY 1 > > static char *create; > +static int delay = DM_DEFAULT_MAX_PROBE_DELAY; > > /* > * Format: dm-mod.create=<name>,<uuid>,<minor>,<flags>,<table>[,<table>+][;<name>,<uuid>,<minor>,<flags>,<table>[,<table>+]+] > @@ -267,6 +271,8 @@ static int __init dm_init_init(void) > LIST_HEAD(devices); > char *str; > int r; > + int loopcnt = delay * 1000 / 5; > + bool devnotfound = false; > > if (!create) > return 0; > @@ -275,6 +281,7 @@ static int __init dm_init_init(void) > DMERR("Argument is too big. Limit is %d", DM_MAX_STR_SIZE); > return -EINVAL; > } > +retry: > str = kstrndup(create, DM_MAX_STR_SIZE, GFP_KERNEL); > if (!str) > return -ENOMEM; > @@ -287,13 +294,23 @@ static int __init dm_init_init(void) > wait_for_device_probe(); > > list_for_each_entry(dev, &devices, list) { > - if (dm_early_create(&dev->dmi, dev->table, > - dev->target_args_array)) > + r = dm_early_create(&dev->dmi, dev->table, dev->target_args_array); > + if (r == -ENODEV) { > + devnotfound = true; > break; > + } > } > + > out: > kfree(str); > dm_setup_cleanup(&devices); > + if (devnotfound && loopcnt) { > + msleep(5); > + devnotfound = false; > + loopcnt--; > + goto retry; > + } > + > return r; > } > > @@ -301,3 +318,5 @@ late_initcall(dm_init_init); > > module_param(create, charp, 0); > MODULE_PARM_DESC(create, "Create a mapped device in early boot"); > +module_param(delay, int, 0); > +MODULE_PARM_DESC(delay, "Max delay to wait for data/hash device probe in seconds"); > </blockquote></div> -- <div dir="ltr" class="gmail_signature"><div dir="ltr">Michael Nazzareno Trimarchi Co-Founder & Chief Executive Officer M. +39 347 913 2170 <a href="mailto:michael@amarulasolutions.com" target="_blank">michael@amarulasolutions.com</a> __________________________________ <div> </div><div>Amarula Solutions BV</div>Joop Geesinkweg 125, 1114 AB, Amsterdam, NL T. +31 (0)85 111 9172 <a href="mailto:info@amarulasolutions.com" target="_blank">info@amarulasolutions.com</a> <a href="http://www.amarulasolutions.com" target="_blank">www.amarulasolutions.com</a> </div></div></div>