On Mon, May 13, 2024 at 11:30:58PM +0000, Jeremy Bongio wrote:
> From: Li Nan <linan122@xxxxxxxxxx>
>
> commit 6cf350658736681b9d6b0b6e58c5c76b235bb4c4 upstream.
>
> If kobject_add() is fail in bind_rdev_to_array(), 'rdev->serial' will be
> alloc not be freed, and kmemleak occurs.
>
> unreferenced object 0xffff88815a350000 (size 49152):
> comm "mdadm", pid 789, jiffies 4294716910
> hex dump (first 32 bytes):
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> backtrace (crc f773277a):
> [<0000000058b0a453>] kmemleak_alloc+0x61/0xe0
> [<00000000366adf14>] __kmalloc_large_node+0x15e/0x270
> [<000000002e82961b>] __kmalloc_node.cold+0x11/0x7f
> [<00000000f206d60a>] kvmalloc_node+0x74/0x150
> [<0000000034bf3363>] rdev_init_serial+0x67/0x170
> [<0000000010e08fe9>] mddev_create_serial_pool+0x62/0x220
> [<00000000c3837bf0>] bind_rdev_to_array+0x2af/0x630
> [<0000000073c28560>] md_add_new_disk+0x400/0x9f0
> [<00000000770e30ff>] md_ioctl+0x15bf/0x1c10
> [<000000006cfab718>] blkdev_ioctl+0x191/0x3f0
> [<0000000085086a11>] vfs_ioctl+0x22/0x60
> [<0000000018b656fe>] __x64_sys_ioctl+0xba/0xe0
> [<00000000e54e675e>] do_syscall_64+0x71/0x150
> [<000000008b0ad622>] entry_SYSCALL_64_after_hwframe+0x6c/0x74
>
> backport change:
> mddev_destroy_serial_pool third parameter was removed in mainline,
> where there is no need to suspend within this function anymore.
>
> Fixes: 963c555e75b0 ("md: introduce mddev_create/destroy_wb_pool for the change of member device")
> Signed-off-by: Li Nan <linan122@xxxxxxxxxx>
> Signed-off-by: Song Liu <song@xxxxxxxxxx>
> Link: https://lore.kernel.org/r/20240208085556.2412922-1-linan666@xxxxxxxxxxxxxxx
> Signed-off-by: Jeremy Bongio <jbongio@xxxxxxxxxx>
> ---
>
> This backport is tested on LTS 5.10, 6.1, 6.6
So this is not needed in 5.15.y or 5.4.y? Why not?
thanks,
greg k-h
