All of lore.kernel.org
 help / color / mirror / Atom feed
From: Greg KH <gregkh@linuxfoundation.org>
To: Hui Peng <benquike@gmail.com>
Cc: linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] usb: gadget: f_tcm: detach the gadget in usbg_drop_tpg() to fix a use-after-free
Date: Fri, 20 Nov 2026 11:53:50 +0100	[thread overview]
Message-ID: <2026112038-sliceable-nail-bc82@gregkh> (raw)
In-Reply-To: <20260919112820.3885842-1-benquike@gmail.com>

On Sat, Sep 19, 2026 at 11:28:20AM +0000, Hui Peng wrote:
> A target portal group can be removed with rmdir() on its configfs
> directory while it is still enabled and still attached to a USB gadget
> function. usbg_drop_tpg() frees the struct usbg_tpg but never calls
> usbg_detach(), and it leaves opts->can_attach set, so the function
> instance keeps a dangling tpg pointer and a subsequent bind to a UDC
> happily attaches to freed memory.
> 
> Enumerating the gadget then dereferences the freed tpg from the UAS/BOT
> completion path, in softirq context:
> 
>  ==================================================================
>  BUG: KASAN: slab-use-after-free in usbg_submit_command.isra.0+0xbe7/0xd50
>  Read of size 8 at addr ffff888107493478 by task swapper/3/0
> 
>  CPU: 3 UID: 0 PID: 0 Comm: swapper/3 Tainted: G    B D    7.3.0-rc3-g5dd1818b15d9 #1 PREEMPT(lazy)
>  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
>  Call Trace:
>   <IRQ>
>   dump_stack_lvl+0x70/0xa0
>   print_report+0x153/0x4c6
>   kasan_report+0xf1/0x120
>   usbg_submit_command.isra.0+0xbe7/0xd50
>   uasp_cmd_complete+0x83/0xf0
>   dummy_timer+0x1337/0x2d60
>   __hrtimer_run_queues+0x2a3/0x640
>   hrtimer_run_softirq+0x1b1/0x3d0
>   handle_softirqs+0x188/0x4e0
>   __irq_exit_rcu+0x62/0x150
>   sysvec_apic_timer_interrupt+0x6b/0x80
>   </IRQ>
> 
>  Allocated by task 1:
>   __kmalloc_cache_noprof+0x16a/0x380
>   usbg_make_tpg+0x243/0x580
>   target_fabric_make_tpg+0xb1/0x7f0
>   configfs_mkdir+0x4e9/0xe10
>   vfs_mkdir+0x2ed/0x790
>   __x64_sys_mkdir+0x6f/0xa0
> 
>  Freed by task 1:
>   kfree+0x159/0x420
>   config_item_cleanup+0x148/0x1e0
>   config_item_put+0x90/0xb0
>   configfs_rmdir+0x816/0xa50
>  ==================================================================
> 
> Detach the gadget in usbg_drop_tpg() if the tpg is still connected, and
> clear opts->can_attach for every function instance bound to this tpg so
> that a later usbg_attach() cannot pick up the freed object.
> 
> The forward declarations of usbg_attach() and usbg_detach() are moved
> above usbg_drop_tpg() since it now calls usbg_detach().
> 
> Fixes: dc8c46a5ae77 ("usb: gadget: f_tcm: convert to new function interface with backward compatibility")
> Assisted-by: LLM
> Signed-off-by: Hui Peng <benquike@gmail.com>
> ---
> v2: Add the Fixes: tag that v1 said it could not determine.
> 
> dc8c46a5ae77 introduced opts->can_attach, the !opts->can_attach gate in
> tcm_bind(), usbg_attach()/usbg_detach(), the tpg_instances[] table and
> fu->tpg, and the version of usbg_drop_tpg() that tears all of that down
> without ever detaching - i.e. every line this patch touches. Note
> git blame points at 08a1cb0f65fd ("usb: gadget: tcm: factor out
> f_tcm") for the top of usbg_drop_tpg(), but that commit is a pure file
> split out of legacy/tcm_usb_gadget.c, so it is not the tag.
> 
> There is a defensible alternative: 4bb8548df632 ("usb: gadget: f_tcm:
> add configfs support") added the callbacks that set can_attach for a
> pure-configfs gadget, which is what my reproducer uses. I went with
> dc8c46a5ae77 because the legacy gadget already set can_attach
> unconditionally, so the hole was reachable without 4bb8548df632 - and
> in any case both landed in v4.5-rc1, five patches apart in the same
> series, so the backport target is identical. Happy to switch if you
> prefer the other.
> 
> For completeness, the underlying "free the tpg without detaching"
> shape is as old as c52661d60f63 ("usb-gadget: Initial merge of target
> module for UASP + BOT", v3.5-rc1), but f_tcm.c does not exist before
> v4.5-rc1 and opts->can_attach - which this patch has to clear - did
> not either, so dc8c46a5ae77 is the oldest commit this patch is
> actually applicable to.
> 
> Two things worth a reviewer's attention: tpg->gadget_connect is read
> here without the target-core configfs locking that usbg_enable_tpg()
> writes it under, and the opts->can_attach = false lands inside the
> if (i < TPG_INSTANCES) arm added by e877b729c649 ("usb: gadget: f_tcm:
> out of bound access in usbg_drop_tpg"). If you would rather refuse the
> rmdir than auto-detach, note that fabric_drop_tpg returns void, so it
> cannot be failed from here.
> 
> Reproduced on Linux 7.3.0-rc3 (5dd1818b15d9) with KASAN under QEMU using
> dummy_hcd: create a tcm function and a target tpgt_1, write 1 to its
> "enable" attribute, link the function into a gadget config, rmdir tpgt_1
> without disabling it first, then bind the gadget to dummy_udc.0. With
> this patch applied the same sequence produces no KASAN splat.

I'm not reading all of that, and obviously no one else did either :(

      reply	other threads:[~2026-11-20 10:53 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-19  9:07 [PATCH] usb: gadget: f_tcm: detach the gadget in usbg_drop_tpg() to fix a use-after-free Hui Peng
2026-09-19 11:28 ` [PATCH v2] " Hui Peng
2026-11-20 10:53   ` Greg KH [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=2026112038-sliceable-nail-bc82@gregkh \
    --to=gregkh@linuxfoundation.org \
    --cc=benquike@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-usb@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.