On Mon, 14 Apr 2025, Damien Le Moal wrote: > On 4/11/25 8:41 PM, Mikulas Patocka wrote: > > > > > > On Fri, 11 Apr 2025, Damien Le Moal wrote: > > > >> On 4/11/25 04:49, Benjamin Marzinski wrote: > >>> There were multiple places in dm's __bind() function where it could fail > >>> and not completely roll back, leaving the device using the the old > >>> table, but with device limits and resources from the new table. > >>> Additionally, unused mempools for request-based devices were not always > >>> freed immediately. > >>> > >>> Also, there were a number of issues with switching zoned tables that > >>> emulate zone append (in other words, dm-crypt on top of zoned devices). > >>> dm_blk_report_zones() could be called while the device was initially > >>> setting up a zoned table and creating zoned resources or could possibly > >>> fail to end a srcu read section. More importantly, > >>> blk_revalidate_disk_zones() would never get called when updating a zoned > >>> table. This could cause the dm device to see the wrong zone write > >>> offsets, not have a large enough zwplugs reserved in its mempool, or > >>> read invalid memory when checking the conventional zones bitmap. > >>> > >>> Finally, any DM device created on top of a device emulating zone appeads > >>> will automatically have zone write plug resources created for it, since > >>> max_hw_zone_append_sectors will always be 0 for a device stacked on top > >>> of a device max_hw_zone_append_sectors = 0 > >>> > >>> This patchset fixes these issues. It deals with the problems around > >>> blk_revalidate_disk_zones() by only calling it for devices that have no > >>> zone write plug resources. This will always correctly update the zoned > >>> settings. If a device has zone write plug resources, calling > >>> blk_revalidate_disk_zones() will not correctly update them in many > >>> cases, so DM simply doesn't call it for devices with zone write plug > >>> resources. Instead of allowing people to load tables that can break the > >>> device, like currently happens, DM disallosw any table reloads that > >>> change the zoned setting for devices that already have zone write plug > >>> resources. Finally, it deals with the max_hw_zone_append_sectors issue > >>> by making sure that it is non-zero for zoned DM devices that do not need > >>> zone write append emulation. > >>> > >>> Changes in V4: > >>> - Fixed typos in 2 commit messages (patches 4 and 5) > >>> - rephrased comment in dm_revalidate_zones() (patch 5) > >> > >> I gave this a quick spin and did not detect any issues. > >> So feel free to add: > >> > >> Tested-by: Damien Le Moal <dlemoal@xxxxxxxxxx> > >> > >> -- > >> Damien Le Moal > >> Western Digital Research > > > > Thanks. > > > > I staged the patches for the next merge window. > > Hmmm... These are all fixes, so shouldn't they go into rc3 ? I think the fixed bugs are not serious (we haven't received any user's bug report that is fixed by these patches). And the patches are big and there's a risk that they break something. So, I wouldn't hurry with them. Mikulas
