Re: [REGRESSION][STABLE] ext4: too many credits wanted / file system issue in v6.16.1

[Greg KH <gregkh@linuxfoundation.org>]

On Tue, Aug 19, 2025 at 11:38:11PM +0100, Luca Boccassi wrote:
> On Tue, 19 Aug 2025 at 21:53, Christian Heusel <christian@heusel.eu> wrote:
> >
> > Hello everyone,
> >
> > the systemd CI has [recently noticed][0] an issue within the ext4 file
> > system after the Arch Linux kernel was upgraded to 6.16.1. The issue is
> > exclusive to the stable tree and does not occur on 6.16 and not on
> > 6.17-rc2. I have also tested 6.16.2-rc1 and it still contains the bug.
> >
> > I was able to bisect the issue between 6.16 and 6.16.1 to the following
> > commit:
> >
> >     b9c561f3f29c2 ("ext4: fix insufficient credits calculation in ext4_meta_trans_blocks()")
> >
> > The issue can be reproduced by running the tests from
> > [TEST-58-REPART.sh][1] by running the [systemd integration tests][2].
> > But if there are any suggestions I can also test myself as the initial
> > setup for the integration tests is a bit involved.
> >
> > It is not yet clear to me whether this has real-world impact besides the
> > test, but the systemd devs said that it's not a particularily demanding
> > workflow, so I guess it is expected to work and could cause issues on
> > other systems too.
> >
> > Also does anybody have an idea which backport could be missing?
> >
> > Cheers,
> > Chris
> >
> > [0]: https://github.com/systemd/systemd/actions/runs/17053272497/job/48345703316#step:14:233
> > [1]: https://github.com/systemd/systemd/blob/main/test/units/TEST-58-REPART.sh
> > [2]: https://github.com/systemd/systemd/blob/main/test/integration-tests/README.md
> >
> > ---
> >
> > #regzbot introduced: b9c561f3f29c2
> > #regzbot title: [STABLE] ext4: too many credits wanted / file system issue in v6.16.1
> > #regzbot link: https://github.com/systemd/systemd/actions/runs/17053272497/job/48345703316#step:14:233
> >
> > ---
> >
> > git bisect start
> > # status: waiting for both good and bad commits
> > # good: [038d61fd642278bab63ee8ef722c50d10ab01e8f] Linux 6.16
> > git bisect good 038d61fd642278bab63ee8ef722c50d10ab01e8f
> > # status: waiting for bad commit, 1 good commit known
> > # bad: [3e0969c9a8c57ff3c6139c084673ebedfc1cf14f] Linux 6.16.1
> > git bisect bad 3e0969c9a8c57ff3c6139c084673ebedfc1cf14f
> > # good: [288f1562e3f6af6d9b461eba49e75c84afa1b92c] media: v4l2-ctrls: Fix H264 SEPARATE_COLOUR_PLANE check
> > git bisect good 288f1562e3f6af6d9b461eba49e75c84afa1b92c
> > # bad: [f427460a1586c2e0865f9326b71ed6e5a0f404f2] f2fs: turn off one_time when forcibly set to foreground GC
> > git bisect bad f427460a1586c2e0865f9326b71ed6e5a0f404f2
> > # bad: [5f57327f41a5bbb85ea382bc389126dd7b8f2d7b] scsi: elx: efct: Fix dma_unmap_sg() nents value
> > git bisect bad 5f57327f41a5bbb85ea382bc389126dd7b8f2d7b
> > # good: [9143c604415328d5dcd4d37b8adab8417afcdd21] leds: pca955x: Avoid potential overflow when filling default_label (take 2)
> > git bisect good 9143c604415328d5dcd4d37b8adab8417afcdd21
> > # good: [9c4f20b7ac700e4b4377f85e36165a4f6ca85995] RDMA/hns: Fix accessing uninitialized resources
> > git bisect good 9c4f20b7ac700e4b4377f85e36165a4f6ca85995
> > # good: [0b21d1962bec2e660c22c4c4231430f97163dcf8] perf tests bp_account: Fix leaked file descriptor
> > git bisect good 0b21d1962bec2e660c22c4c4231430f97163dcf8
> > # good: [3dbe96d5481acd40d6090f174d2be8433d88716d] clk: thead: th1520-ap: Correctly refer the parent of osc_12m
> > git bisect good 3dbe96d5481acd40d6090f174d2be8433d88716d
> > # bad: [c6714f30ef88096a8da9fcafb6034dc4e9aa467d] clk: sunxi-ng: v3s: Fix de clock definition
> > git bisect bad c6714f30ef88096a8da9fcafb6034dc4e9aa467d
> > # bad: [b9c561f3f29c2d6e1c1d3ffc202910bef250b7d8] ext4: fix insufficient credits calculation in ext4_meta_trans_blocks()
> > git bisect bad b9c561f3f29c2d6e1c1d3ffc202910bef250b7d8
> > # first bad commit: [b9c561f3f29c2d6e1c1d3ffc202910bef250b7d8] ext4: fix insufficient credits calculation in ext4_meta_trans_blocks()
> 
> The full kernel warning (immediately after the ext4 fs stops working):

I've pushed out a 6.16.3-rc1 that should hopefully resolve this.

thanks,

greg k-h