Re: ubiblock: null pointer dereference using scatterlist in work_queue

From: hch
Date: Wed Aug 09 2023 - 09:53:16 EST


Well, a scatterlist culd contain a highmem page, in which case sg_virt
isn't going to cut it and you need to kmap. Is this a 32-bit system
with highmem enabled?

On Tue, Aug 08, 2023 at 12:02:56AM +0200, Richard Weinberger wrote:
> [Replying with full context for block layer folks]
>
> Stephan reported that ubiblock causes a NULL pointer dereference on his system
> as soon a DM is configured above. He reproduced with both dmcrypt and dm-verity.
>
> The NULL pointer dereference happens in ubi_eba_read_leb_sg() because sg_virt()
> returns NULL. I'm a bit puzzled why.
>
> We checked that the sg list is not empty and that blk_rq_map_sg() returns a
> positive number. So, sg_set_page() should have happened.
> >From the logs it seems also that the first few reads from the kernel succeed,
> I guess the partition scan, but as soon cryptsetup triggers a read from userspace
> the crash happens. It happens always, so it does not look like a random memory
> corruption.
> Enabling CONFIG_DEBUG_SG does also not indicate a bug.
>
> Do you have an idea what could case this problem? Is the usage of sg_virt() in UBI
> wrong?
> ubiblock exists now for some time and is heavily used with dmcrypt on similar devices.
> Also backporting 91cc8fbcc8c7 ("ubi: block: set BLK_MQ_F_BLOCKING") to Stephan's 6.1
> kernel didn't change the symptoms.
>
> Thanks,
> //richard
>
> ----- Ursprüngliche Mail -----
> > Von: "Stephan Wurm" <stephan.wurm@xxxxxxxxxxx>
> > An: "richard" <richard@xxxxxx>, "Miquel Raynal" <miquel.raynal@xxxxxxxxxxx>, "Vignesh Raghavendra" <vigneshr@xxxxxx>,
> > "linux-mtd" <linux-mtd@xxxxxxxxxxxxxxxxxxx>, "linux-kernel" <linux-kernel@xxxxxxxxxxxxxxx>
> > CC: "Johannes Eigner" <johannes.eigner@xxxxxxxxxxx>
> > Gesendet: Montag, 17. Juli 2023 10:07:04
> > Betreff: ubiblock: null pointer dereference using scatterlist in work_queue
>
> > Hello!
> >
> > We struggle with a weird problem accessing an ubiblock device for quite some
> > time now. As we are running out of ideas, I reach out for your help.
> >
> > The device we develop has an i.MX6 Solo SoC equipped with gpmi-nand flash
> > (Micron MT29F8G08ABACA, 1GB). The flash is partitioned into five UBI volumes,
> > forming an A/B boot setup, consisting of two pairs with each a UBIFS volume
> > holding a FIT image and a ubiblock volume containing a SQUASHFS root filesystem
> > with dm-verity hashes attached, and the fifth UBIFS volume for common data.
> > The bootloader (barebox 2022.06.0) is stored on an additional spi-nor flash.
> >
> > In order to mount the root filesystem, the FIT image contains an initramfs,
> > calling cryptsetup (tried v2.4.3 and v2.6.1) on the ubiblock device,
> > followed by mount of the root filesystem.
> > MTD attach and creation of the ubiblock device is triggered by kernel
> > arguments:
> >
> > commandline: console=ttymxc1,115200n8 ubi.block=0,root0 root=/dev/ubiblock0_2
> > rootfstype=squashfs ubi.mtd=nand
> >
> > This fails to boot with a null pointer dereference when calling cryptsetup:
> >
> > [ 5.637207] ubi0: default fastmap pool size: 200
> > [ 5.642020] ubi0: default fastmap WL pool size: 100
> > [ 5.647000] ubi0: attaching mtd0
> > [ 6.450307] ubi0: scanning is finished
> > [ 6.472816] ubi0: attached mtd0 (name "nand", size 1024 MiB)
> > [ 6.478561] ubi0: PEB size: 262144 bytes (256 KiB), LEB size: 253952 bytes
> > [ 6.485508] ubi0: min./max. I/O unit sizes: 4096/4096, sub-page size 4096
> > [ 6.492395] ubi0: VID header offset: 4096 (aligned 4096), data offset: 8192
> > [ 6.499378] ubi0: good PEBs: 4092, bad PEBs: 4, corrupted PEBs: 0
> > [ 6.505518] ubi0: user volume: 5, internal volumes: 1, max. volumes count:
> > 128
> > [ 6.512810] ubi0: max/mean erase counter: 14/8, WL threshold: 4096, image
> > sequence number: 228025450
> > [ 6.521965] ubi0: available PEBs: 177, total reserved PEBs: 3915, PEBs
> > reserved for bad PEB handling: 76
> > [ 6.531750] ubi0: background thread "ubi_bgt0d" started, PID 163
> > [ 12.919477] block ubiblock0_2: created from ubi0:2(root0)
> > [ 12.929687] ALSA device list:
> > [ 12.932750] #0: sgtl5000
> > [ 12.940024] Freeing unused kernel image (initmem) memory: 1024K
> > [ 12.946926] Run /init as init process
> > Starting version 250.5+
> > [ 15.601749] mtdblock: MTD device 'nand' is NAND, please consider using UBI
> > block devices instead.
> > realpath: /dev/disk/by-partuuid//dev/ubiblock0_2: No such file or directory
> > [ 26.127460] 8<--- cut here ---
> > [ 26.130689] Unable to handle kernel NULL pointer dereference at virtual
> > address 00000000
> > [ 26.138886] [00000000] *pgd=00000000
> > [ 26.142523] Internal error: Oops: 5 [#1] ARM
> > [ 26.146804] Modules linked in:
> > [ 26.149868] CPU: 0 PID: 18 Comm: kworker/0:3 Not tainted 6.1.38 #1
> > [ 26.156060] Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
> > [ 26.162592] Workqueue: ubiblock0_2 ubiblock_do_work
> > [ 26.167498] PC is at ubi_io_read+0x78/0x2fc
> > [ 26.171693] LR is at ubi_eba_read_leb+0xe8/0x4a4
> > [ 26.176320] pc : [<c075f508>] lr : [<c075cd0c>] psr: 60070013
> > [ 26.182591] sp : f0961dc8 ip : 00000000 fp : 00002000
> > [ 26.187819] r10: c81c6000 r9 : 00000000 r8 : c81c7000
> > [ 26.193046] r7 : 00000200 r6 : 000008e0 r5 : 00002000 r4 : 00000000
> > [ 26.199578] r3 : 00000000 r2 : 00000000 r1 : 00000000 r0 : c81c6000
> > [ 26.206108] Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
> > [ 26.213250] Control: 10c53c7d Table: 1824c059 DAC: 00000051
> > [ 26.218998] Register r0 information: slab kmalloc-8k start c81c6000 pointer
> > offset 0 size 8192
> > [ 26.227638] Register r1 information: NULL pointer
> > [ 26.232352] Register r2 information: NULL pointer
> > [ 26.237062] Register r3 information: NULL pointer
> > [ 26.241772] Register r4 information: NULL pointer
> > [ 26.246481] Register r5 information: non-paged memory
> > [ 26.251537] Register r6 information: non-paged memory
> > [ 26.256594] Register r7 information: non-paged memory
> > [ 26.261651] Register r8 information: slab kmalloc-8k start c81c6000 pointer
> > offset 4096 size 8192
> > [ 26.270545] Register r9 information: NULL pointer
> > [ 26.275254] Register r10 information: slab kmalloc-8k start c81c6000 pointer
> > offset 0 size 8192
> > [ 26.283977] Register r11 information: non-paged memory
> > [ 26.289120] Register r12 information: NULL pointer
> > [ 26.293916] Process kworker/0:3 (pid: 18, stack limit = 0x1a047662)
> > [ 26.300194] Stack: (0xf0961dc8 to 0xf0962000)
> > [ 26.304559] 1dc0: c1d50ec0 c075cc64 00000001 00000000
> > 00000000 c1d50ec0
> > [ 26.312744] 1de0: 00000001 24c676fe c075cc64 00000000 000008e0 c81c6000
> > c80c4000 00000000
> > [ 26.320928] 1e00: c1d50ec0 00000002 c81c7000 c075cd0c 00000200 c01d3930
> > 60070013 00000000
> > [ 26.329111] 1e20: 60070013 c1b43208 000008e0 24c676fe 00000000 c830f5cc
> > c830f5cc 00000200
> > [ 26.337294] 1e40: 00000000 00000001 00000200 c80c4000 00000000 c075d124
> > 00000000 00000200
> > [ 26.345477] 1e60: 00000000 c13e0ec0 c81c6000 c1d50ec0 00000000 00000000
> > 00000000 c80c4000
> > [ 26.353661] 1e80: c830f5cc c1e52c00 00000200 c81c6000 00000002 c075b748
> > 00000000 00000200
> > [ 26.361844] 1ea0: 00000000 c1312e58 ef7d3c00 0003e000 00000000 00000200
> > ef7d3c00 c1e52c00
> > [ 26.370027] 1ec0: c830f5a8 c830f5cc 00000000 c076b228 00000200 00000000
> > c830f500 c1d50ec0
> > [ 26.378211] 1ee0: c830f5d4 24c676fe c0c0e608 c830f5a8 c1ed3800 c1312e58
> > ef7d3c00 c1d50ec0
> > [ 26.386394] 1f00: 00000000 c14e59c0 ef7d3c05 c013c06c 00000001 00000000
> > c013bff4 00000000
> > [ 26.394577] 1f20: c1312e68 24c676fe c1b5f698 c19642b8 00000000 c101b390
> > 00000000 24c676fe
> > [ 26.402760] 1f40: c1d50ec0 c1ed3800 c1312e58 c1ed3818 c1312e94 c13dfc30
> > c1d50ec0 00000008
> > [ 26.410944] 1f60: c1312e58 c013c3f0 00000000 c1f16880 c1d50ec0 c013c3b0
> > c1ed3800 c1ed3880
> > [ 26.419126] 1f80: f095de9c 00000000 00000000 c0143654 c1f16880 c0143588
> > 00000000 00000000
> > [ 26.427309] 1fa0: 00000000 00000000 00000000 c0100128 00000000 00000000
> > 00000000 00000000
> > [ 26.435491] 1fc0: 00000000 00000000 00000000 00000000 00000000 00000000
> > 00000000 00000000
> > [ 26.443673] 1fe0: 00000000 00000000 00000000 00000000 00000013 00000000
> > 00000000 00000000
> > [ 26.451855] ubi_io_read from ubi_eba_read_leb+0xe8/0x4a4
> > [ 26.457277] ubi_eba_read_leb from ubi_eba_read_leb_sg+0x5c/0x154
> > [ 26.463390] ubi_eba_read_leb_sg from ubi_leb_read_sg+0x70/0xb0
> > [ 26.469325] ubi_leb_read_sg from ubiblock_do_work+0x104/0x238
> > [ 26.475180] ubiblock_do_work from process_one_work+0x238/0x57c
> > [ 26.481130] process_one_work from worker_thread+0x40/0x4f8
> > [ 26.486724] worker_thread from kthread+0xcc/0xf0
> > [ 26.491449] kthread from ret_from_fork+0x14/0x2c
> > [ 26.496168] Exception stack(0xf0961fb0 to 0xf0961ff8)
> > [ 26.501225] 1fa0: 00000000 00000000
> > 00000000 00000000
> > [ 26.509410] 1fc0: 00000000 00000000 00000000 00000000 00000000 00000000
> > 00000000 00000000
> > [ 26.517592] 1fe0: 00000000 00000000 00000000 00000000 00000013 00000000
> > [ 26.524215] Code: 1a000041 e59d2010 e1a09fc5 e1a0b005 (e5d23000)
> > [ 26.530367] ---[ end trace 0000000000000000 ]---
> >
> > This kernel Oops happens both on patched or vanilla kernels, of which we tried
> > several releases in 5.15.y and 6.1.y branches. We also tried with several of
> > our devices.
> > Upgrading to the latest mainline kernel did not work out-of-the-box, hence we
> > did not follow that track (yet).
> >
> > We tried with full debugging output for the whole ubi driver and used several
> > additional printks. That way we were able to see, that the first scatterlist
> > entry already pointed to virtual address zero when the first read request
> > caused by cryptsetup was added to ubiblock's work_queue.
> >
> > sg_virt(pdu->usgl.sg[0]) => page_address(sg) => 0x0000000
> >
> > We also tried to use hw_breakpoints to gather more information on (maybe)
> > another module interfering, but did not succeed (yet).
> >
> > But we were not able to narrow down the root cause until now.
> >
> >
> > As an additional twist, the system is able to boot when we mount the ubiblock
> > root filesystem without calling cryptsetup, hence skipping the dm-verity hash
> > verification. And we can verify the root filesystem with cryptsetup once the
> > system boot is finished.
> >
> > It is also possible to boot the same system image, including dm-verity, when
> > using a sdcard instead of the nand flash. Loading the FIT from sdcard but
> > calling cryptsetup on the ubiblock device again leads to the described oops.
> >
> >
> > Is there something we have overlooked?
> > Do you have further ideas to get behind this issue?
> >
> >
> > Thanks!
> > --
> > Stephan Wurm <stephan.wurm@xxxxxxxxxxx>
---end quoted text---