hi Emil, in view of your answer and after rereading my email, I realize that I was confused in my request. here it is, I hope, more clearly reformulated :-) first of all, I use ArchLinux to, from time to time, compile the slightly modified LTS kernel, and this from PKGBUILD provided by ArchLinux at some point. some technologies such as LVM do not take into account the read-only applied on a block device. see the two links provided in the previous exchanges for more details... until now, I recompiled the kernel by applying a slight modification to the bio_check_ro function present in the blk-core.c source file. the last time I made this modification was on the Linux-LTS-5.10.19 kernel : *(https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/...)* $ diff -u 5.10.19.original/blk-core.c 5.10.19.me/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 5.10.19.me/blk-core.c 2023-03-15 13:44:02.353596114 +0100 @@ -706,7 +706,7 @@ "Trying to write to read-only block-device %s (partno %d)\n", bio_devname(bio, b), part->partno); /* Older lvm-tools actually trigger this */ - return false; + return true; } return false; the compilation of the modified LTS 5.10.19 kernel went well and the correction seems to do the job... since this last time (2022/01), the source file blk-core.c has been modified a lot and the bio_check_ro function is part of these modifications : *(https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/...)* $ diff -u 5.10.19.original/blk-core.c 6.1.15.original/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 @@ -14,11 +14,10 @@ */ #include #include -#include #include #include -#include ... @@ -681,40 +483,22 @@ } late_initcall(fail_make_request_debugfs); - -#else /* CONFIG_FAIL_MAKE_REQUEST */ - -static inline bool should_fail_request(struct hd_struct *part, - unsigned int bytes) -{ - return false; -} - #endif /* CONFIG_FAIL_MAKE_REQUEST */ -static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) +static inline void bio_check_ro(struct bio *bio) { - const int op = bio_op(bio); - - if (part->policy && op_is_write(op)) { - char b[BDEVNAME_SIZE]; - + if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) - return false; - - WARN_ONCE(1, - "Trying to write to read-only block-device %s (partno %d)\n", - bio_devname(bio, b), part->partno); + return; + pr_warn("Trying to write to read-only block-device %pg\n", + bio->bi_bdev); /* Older lvm-tools actually trigger this */ - return false; } - - return false; } ... when I introduce my little modification (see diff below) in the code, makepkg fails to compile with the error "return with a value in function returning void" (see makepkg below) : $ diff -u 6.1.15.original/blk-core.c 6.1.15.me/blk-core.c --- 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 +++ 6.1.15.me/blk-core.c 2023-03-15 13:56:15.246945330 +0100 @@ -493,6 +493,7 @@ pr_warn("Trying to write to read-only block-device %pg\n", bio->bi_bdev); /* Older lvm-tools actually trigger this */ + return true; } } $ makepkg ... CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function returning void [-Werror=return-type] 496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2 so, how to modify the current code of the bio_check_ro function to get the desired result (eg. writes KO on RO blockdevice) ? with the changes in the blk-core.c source code since version 5.10.19, is it still necessary to tweak the bio_check_ro function to disallow technologies that ignore the read-only block? regards, lacsaP. Le mer. 15 mars 2023 à 12:37, Emil Velikov a écrit :

Greetings Pascal,

Couple of suggestions from the peanut gallery, Take them with a heavy pinch of salt: - Is the issue happening with upstream code from kernel.org? - Consider mentioning the commit sha (and URL, if it is missing from kernel.org) in the email - Is "intervened" the right word here - the Cambridge dictionary defines it as "to intentionally become involved in a difficult situation in order to improve it or prevent it from getting worse" - Are you contacting a developer only? Have you considered adding the subsystem maintainer and mailing list in the CC list - scripts/get_maintainer.pl will give you those - Have you considered opening a bug report, or better yet sending a patch? Patch does not have to be perfect and if you have doubts you can mention those in the email/cover-letter.

Hope that helps Emil

[1] https://dictionary.cambridge.org/dictionary/english/intervene

On Wed, 15 Mar 2023 at 08:42, Pascal wrote:

...

hi,

I come to you for lack of feedback (I think the Linux kernel developers

...

would one of you have the answer or a track to follow concerning the question below ? the encountered compilation error is behind the forwarded email.

regards, lacsaP.

---------- Forwarded message --------- De : Pascal Date: mer. 8 mars 2023 à 14:09 Subject: bio_check_ro @ blk-core.c

hi,

I'm addressing you because you intervened (commit) in the function bio_check_ro @ blk-core.c @ Linux-LTS-6.1.15. the last time I intervened on this file (@ Linux-LTS-5.10.19 for

...

with @ Linux-LTS-6.1.15, if I insert "return true;", I now have a compilation error. in your opinion, is there still a need to "fix" blk-core.c to prevent writing to a read-only locked device and if so, can you help me implement

have other cats to whip :-)) personal use), it was to replace "return false;" by "return true;", which theoretically should prevent the possible writing on a device locked in read-only mode (see here or here). this fix?

...

regards, lacsaP. ---------- End forwarded message ---------

SYNC include/config/auto.conf CC arch/x86/kernel/asm-offsets.s CALL scripts/checksyscalls.sh DESCEND objtool DESCEND bpf/resolve_btfids CC block/bdev.o CC block/fops.o CC block/bio.o CC block/elevator.o CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function

returning void [-Werror=return-type]

...

496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2

hi Emil, thank you for this investigation, this history. for the moment, I simply reversed the function bio_check_ro and the compilation is running... if all is ok, I will just have to test to see if the blocking is well done... $ diff -u linux-6.1.15.original/block/blk-core.c linux-6.1.15.me/block/blk-core.c --- linux-6.1.15.original/block/blk-core.c 2023-03-03 11:52:25.000000000 +0100 +++ linux-6.1.15.me/block/blk-core.c 2023-03-15 13:24:24.680237258 +0100 @@ -485,15 +485,17 @@ late_initcall(fail_make_request_debugfs); #endif /* CONFIG_FAIL_MAKE_REQUEST */ -static inline void bio_check_ro(struct bio *bio) +static inline bool bio_check_ro(struct bio *bio) { if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) - return; + return false; pr_warn("Trying to write to read-only block-device %pg\n", bio->bi_bdev); /* Older lvm-tools actually trigger this */ + return true; } + return false; } I made some tests with archlinux-2023.03.01-x86_64.iso to answer my question about the need to continue to make a correction : the answer is *yes* it is necessary to correct because LVM alters the device set in read-only mode. /tmp : file *.disk lvm.disk: QEMU QCOW Image (v3), 268435456 bytes, AES-encrypted (v3), 268435456 bytes /tmp : sha256sum *.disk *55a9af5144edf235cf0ac16fc60a8dab3c88a0d3451eb6e4d61373e0ad80cb82 lvm.disk* /tmp : qemu -cdrom /mnt/archlinux-2023.03.01-x86_64.iso root@archiso ~ # cat /etc/udev/rules.d/09-scalpel.rules KERNEL!="ram*", SUBSYSTEM=="block", RUN+="/usr/local/sbin/rodev /dev/%k" KERNEL=="rtc*", SUBSYSTEM=="rtc", MODE="0400" root@archiso ~ # cat /usr/local/sbin/rodev #!/usr/bin/bash [ ! "${1}" ] && echo "error: device missing" >/dev/stderr && exit 1 [ "${2}" ] && echo "error: too many devices" >/dev/stderr && exit 1 chmod 440 "${1}" blockdev --setro "${1}" root@archiso ~ # chmod +x !$ chmod +x /usr/local/sbin/rodev root@archiso ~ # udevadm control --reload-rules && udevadm trigger (qemu) drive_add 0 if=none,file=/tmp/lvm.disk,id=usbdrv (qemu) device_add usb-storage,drive=usbdrv,id=usbkey * Mar 16 10:07:51 archiso kernel: usb 2-3: new SuperSpeed USB device number 3 using xhci_hcd Mar 16 10:07:51 archiso kernel: usb 2-3: New USB device found, idVendor=46f4, idProduct=0001, bcdDevice= 0.00 Mar 16 10:07:51 archiso kernel: usb 2-3: New USB device strings: Mfr=1, Product=2, SerialNumber=3 Mar 16 10:07:51 archiso kernel: usb 2-3: Product: QEMU USB HARDDRIVE Mar 16 10:07:51 archiso kernel: usb 2-3: Manufacturer: QEMU Mar 16 10:07:51 archiso kernel: usb 2-3: SerialNumber: 1-0000:00:03.0-3 Mar 16 10:07:51 archiso kernel: usb-storage 2-3:1.0: USB Mass Storage device detected Mar 16 10:07:51 archiso kernel: scsi host6: usb-storage 2-3:1.0 Mar 16 10:07:52 archiso kernel: scsi 6:0:0:0: Direct-Access QEMU QEMU HARDDISK 2.5+ PQ: 0 ANSI: 5 Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: Power-on or device reset occurred Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: [sda] 524288 512-byte logical blocks: (268 MB/256 MiB) Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: [sda] Write Protect is off Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: [sda] Mode Sense: 63 00 00 08 Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA Mar 16 10:07:52 archiso kernel: sda: sda1 sda2 sda3 Mar 16 10:07:52 archiso kernel: sd 6:0:0:0: [sda] Attached SCSI disk Mar 16 10:07:52 archiso lvm[1047]: PV /dev/sda3 online, VG test.lvm incomplete (need 2). Mar 16 10:07:52 archiso lvm[1049]: PV /dev/sda1 online, VG test.lvm incomplete (need 1). Mar 16 10:07:52 archiso lvm[1048]: PV /dev/sda2 online, VG test.lvm is complete. Mar 16 10:07:52 archiso kernel: raid6: skipped pq benchmark and selected sse2x4 Mar 16 10:07:52 archiso kernel: raid6: using intx1 recovery algorithm Mar 16 10:07:52 archiso kernel: xor: measuring software checksum speed Mar 16 10:07:52 archiso kernel: prefetch64-sse : 34921 MB/sec Mar 16 10:07:52 archiso kernel: generic_sse : 31738 MB/sec Mar 16 10:07:52 archiso kernel: xor: using function: prefetch64-sse (34921 MB/sec) Mar 16 10:07:52 archiso kernel: async_tx: api initialized (async) Mar 16 10:07:52 archiso kernel: device-mapper: raid: Loading target version 1.15.1 Mar 16 10:07:52 archiso kernel: md/raid1:mdX: active with 3 out of 3 mirrors Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-0 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-2 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-4 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-0 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device sda1 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-2 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device sda2 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device dm-4 Mar 16 10:07:52 archiso kernel: Trying to write to read-only block-device sda3 Mar 16 10:07:52 archiso systemd[1]: Started Device-mapper event daemon. Mar 16 10:07:53 archiso dmeventd[1114]: dmeventd ready for processing. Mar 16 10:07:53 archiso dmeventd[1114]: Monitoring RAID device test.lvm-lvm.test for events. Mar 16 10:07:53 archiso lvm[1060]: 1 logical volume(s) in volume group "test.lvm" now active Mar 16 10:07:53 archiso systemd[1]: lvm-activate-test.lvm.service: Deactivated successfully. Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-0 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-2 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-4 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-0 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device sda1 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-2 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device sda2 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device dm-4 Mar 16 10:08:08 archiso kernel: Trying to write to read-only block-device sda3* root@archiso ~ # lsblk NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINTS loop0 7:0 0 688.5M 1 loop /run/archiso/airootfs sda 8:0 0 256M 1 disk ├─sda1 8:1 0 83M 1 part │ ├─test.lvm-lvm.test_rmeta_0 254:0 0 4M 1 lvm │ │ └─test.lvm-lvm.test 254:6 0 64M 1 lvm │ └─test.lvm-lvm.test_rimage_0 254:1 0 64M 1 lvm │ └─test.lvm-lvm.test 254:6 0 64M 1 lvm ├─sda2 8:2 0 83M 1 part │ ├─test.lvm-lvm.test_rmeta_1 254:2 0 4M 1 lvm │ │ └─test.lvm-lvm.test 254:6 0 64M 1 lvm │ └─test.lvm-lvm.test_rimage_1 254:3 0 64M 1 lvm │ └─test.lvm-lvm.test 254:6 0 64M 1 lvm └─sda3 8:3 0 83M 1 part ├─test.lvm-lvm.test_rmeta_2 254:4 0 4M 1 lvm │ └─test.lvm-lvm.test 254:6 0 64M 1 lvm └─test.lvm-lvm.test_rimage_2 254:5 0 64M 1 lvm └─test.lvm-lvm.test 254:6 0 64M 1 lvm sr0 11:0 1 810.3M 1 rom /run/archiso/bootmnt root@archiso ~ # dd if=/dev/zero of=/dev/sda bs=1M count=16 dd: error writing '/dev/sda': Operation not permitted 1+0 records in 0+0 records out 0 bytes copied, 0.00814249 s, 0.0 kB/s /tmp : sha256sum *.disk *1a11bb40ee4887d0dabd9fca2224cf7219bb37d7efe2a8fbd73a4b656213c228 lvm.disk* root@archiso ~ # pvdisplay --- Physical volume --- PV Name /dev/sda1 VG Name test.lvm PV Size <83.01 MiB / not usable <3.01 MiB Allocatable yes PE Size 4.00 MiB Total PE 20 Free PE 3 Allocated PE 17 PV UUID mQR1eS-K5Zf-5lfl-Dupc-lRhN-Yfpf-7Vl36V --- Physical volume --- PV Name /dev/sda2 VG Name test.lvm PV Size <83.01 MiB / not usable <3.01 MiB Allocatable yes PE Size 4.00 MiB Total PE 20 Free PE 3 Allocated PE 17 PV UUID As2RrD-Cev0-hhB7-yohn-r5ym-P34m-MFV0M9 --- Physical volume --- PV Name /dev/sda3 VG Name test.lvm PV Size <83.01 MiB / not usable <3.01 MiB Allocatable yes PE Size 4.00 MiB Total PE 20 Free PE 3 Allocated PE 17 PV UUID khaOAc-YEX4-bi0u-fdtq-doG9-Wuok-B2kCS8 root@archiso ~ # vgdisplay --- Volume group --- VG Name test.lvm System ID Format lvm2 Metadata Areas 3 Metadata Sequence No 3 VG Access read/write VG Status resizable MAX LV 0 Cur LV 1 Open LV 0 Max PV 0 Cur PV 3 Act PV 3 VG Size 240.00 MiB PE Size 4.00 MiB Total PE 60 Alloc PE / Size 51 / 204.00 MiB Free PE / Size 9 / 36.00 MiB VG UUID sttYAJ-VmEf-biZh-C8Uh-jUCK-orFO-WVNX6t root@archiso ~ # vgchange -ay test.lvm 1 logical volume(s) in volume group "test.lvm" now active /tmp : sha256sum *.disk *c39d62a5e9f87dc33339098c64903b56b2d3ac5248d2f34c10e63241a10a35b3 lvm.disk* root@archiso ~ # lvdisplay --- Logical volume --- LV Path /dev/test.lvm/lvm.test LV Name lvm.test VG Name test.lvm LV UUID G1y4A0-LbbZ-na9w-AJrQ-jVtc-pKDo-RKgc0S LV Write Access read/write (activated read only) LV Creation host, time scalpel, 2019-03-12 14:59:02 +0000 LV Status available # open 0 LV Size 64.00 MiB Current LE 16 Mirrored volumes 3 Segments 1 Allocation inherit Read ahead sectors auto - currently set to 256 Block device 254:6 root@archiso ~ # mount /dev/test.lvm/lvm.test /mnt mount: /mnt: WARNING: source write-protected, mounted read-only. root@archiso ~ # ll /mnt total 16 drwx------ 2 root root 12288 Mar 12 2019 lost+found -rwxr-xr-x 1 root root 3155 Mar 12 2019 lvm.md root@archiso ~ # umount /mnt (qemu) device_del usbkey * Mar 16 10:21:07 archiso kernel: usb 2-3: USB disconnect, device number 3 Mar 16 10:21:07 archiso kernel: sd 6:0:0:0: [sda] Synchronizing SCSI cache Mar 16 10:21:07 archiso kernel: sd 6:0:0:0: [sda] Synchronize Cache(10) failed: Result: hostbyte=DID_NO_CONNECT driverbyte=DRIVER_OK Mar 16 10:21:07 archiso (udev-worker)[1276]: sda2: Process '/usr/local/sbin/rodev /dev/sda2' failed with exit code 1. Mar 16 10:21:07 archiso systemd-homed[268]: block device /sys/devices/pci0000:00/0000:00:03.0/usb2/2-3/2-3:1.0/host6/target6:0:0/6:0:0:0/block/sda/sda2 has been removed. Mar 16 10:21:07 archiso (udev-worker)[1275]: sda1: Process '/usr/local/sbin/rodev /dev/sda1' failed with exit code 1. Mar 16 10:21:07 archiso systemd-homed[268]: block device /sys/devices/pci0000:00/0000:00:03.0/usb2/2-3/2-3:1.0/host6/target6:0:0/6:0:0:0/block/sda/sda1 has been removed. Mar 16 10:21:07 archiso (udev-worker)[1277]: sda3: Process '/usr/local/sbin/rodev /dev/sda3' failed with exit code 1. Mar 16 10:21:07 archiso systemd-homed[268]: block device /sys/devices/pci0000:00/0000:00:03.0/usb2/2-3/2-3:1.0/host6/target6:0:0/6:0:0:0/block/sda/sda3 has been removed. Mar 16 10:21:07 archiso (udev-worker)[1276]: sda: Process '/usr/local/sbin/rodev /dev/sda' failed with exit code 1. Mar 16 10:21:07 archiso systemd-homed[268]: block device /sys/devices/pci0000:00/0000:00:03.0/usb2/2-3/2-3:1.0/host6/target6:0:0/6:0:0:0/block/sda has been removed.* /tmp : sha256sum *.disk c39d62a5e9f87dc33339098c64903b56b2d3ac5248d2f34c10e63241a10a35b3 lvm.disk the same test was performed with for example Linux Software RAID (mdadm) and the device is not modified (as expected from being locked in read-only mode). in conclusion, it seems that LVM, even in recent version, does not take into account the read-only mode... regards, lacsaP. Le mer. 15 mars 2023 à 16:36, Emil Velikov a écrit :

Greetings Pascal,

After following the links I see what's happening. Essentially: - Kernel gained RO/RW correctness check - circa Jan 2018, kernel commit 721c7fc701c71f693307d274d2b346a1ecd4a534 - LVM was initially buggy but later fixed, circa Mar 2018, - Kernel check got partially reverted because broken LVM is still used - circa Aug 2018, kernel commit a32e236eb93e62a0f692e79b7c3c9636689559b9 - People used an out of tree patch, reinstating the correctness check - The function return type was dropped since it is unused - Sep 2022, kernel commit bdb7d420c6f6d2618d4c907cd7742c3195c425e2 - kernel patch no longer applies, correct behaviour cannot be enforced

To unblock yourself, it will be a matter of reverting bdb7d420c6f6d2618d4c907cd7742c3195c425e2 and then a32e236eb93e62a0f692e79b7c3c9636689559b9.

For the mid/long run, one should consider a proper upstream solution:

Assuming I'm in your position, I would dig through the data in the linked commits and estimate which/how many distributions ship with buggy LVM. Then formulate a comprehensive cover letter, proposing a) reverts (if LVM is no longer used in the wild) or b) reverts && a (KCONFIG, sysctl, other) toggle to control the behaviour.

Hope that helps, Emil

On Wed, 15 Mar 2023 at 13:38, Pascal wrote:

...

hi Emil,

in view of your answer and after rereading my email, I realize that I

...

here it is, I hope, more clearly reformulated :-)

first of all, I use ArchLinux to, from time to time, compile the slightly modified LTS kernel, and this from PKGBUILD provided by ArchLinux at some point.

some technologies such as LVM do not take into account the read-only applied on a block device. see the two links provided in the previous exchanges for more details...

until now, I recompiled the kernel by applying a slight modification to

...

the last time I made this modification was on the Linux-LTS-5.10.19 kernel : ( https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... )

$ diff -u 5.10.19.original/blk-core.c 5.10.19.me/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 5.10.19.me/blk-core.c 2023-03-15 13:44:02.353596114 +0100 @@ -706,7 +706,7 @@ "Trying to write to read-only block-device %s (partno %d)\n", bio_devname(bio, b), part->partno); /* Older lvm-tools actually trigger this */ - return false; + return true; }

return false;

the compilation of the modified LTS 5.10.19 kernel went well and the correction seems to do the job...

since this last time (2022/01), the source file blk-core.c has been modified a lot and the bio_check_ro function is part of these modifications : ( https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... )

$ diff -u 5.10.19.original/blk-core.c 6.1.15.original/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 @@ -14,11 +14,10 @@ */ #include #include -#include #include #include -#include ... @@ -681,40 +483,22 @@ }

late_initcall(fail_make_request_debugfs); - -#else /* CONFIG_FAIL_MAKE_REQUEST */ - -static inline bool should_fail_request(struct hd_struct *part, - unsigned int bytes) -{ - return false; -} - #endif /* CONFIG_FAIL_MAKE_REQUEST */

-static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) +static inline void bio_check_ro(struct bio *bio) { - const int op = bio_op(bio); - - if (part->policy && op_is_write(op)) { - char b[BDEVNAME_SIZE]; - + if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) - return false; - - WARN_ONCE(1, - "Trying to write to read-only block-device %s (partno %d)\n", - bio_devname(bio, b), part->partno); + return; + pr_warn("Trying to write to read-only block-device %pg\n", + bio->bi_bdev); /* Older lvm-tools actually trigger this */ - return false; } - - return false; } ...

when I introduce my little modification (see diff below) in the code, makepkg fails to compile with the error "return with a value in function returning void" (see makepkg below) :

$ diff -u 6.1.15.original/blk-core.c 6.1.15.me/blk-core.c --- 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 +++ 6.1.15.me/blk-core.c 2023-03-15 13:56:15.246945330 +0100 @@ -493,6 +493,7 @@ pr_warn("Trying to write to read-only block-device %pg\n", bio->bi_bdev); /* Older lvm-tools actually trigger this */ + return true; } }

$ makepkg ... CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function returning void [-Werror=return-type] 496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2

so, how to modify the current code of the bio_check_ro function to get

...

with the changes in the blk-core.c source code since version 5.10.19, is it still necessary to tweak the bio_check_ro function to disallow technologies that ignore the read-only block?

regards, lacsaP.

Le mer. 15 mars 2023 à 12:37, Emil Velikov a écrit :

...

Greetings Pascal,

Couple of suggestions from the peanut gallery, Take them with a heavy pinch of salt: - Is the issue happening with upstream code from kernel.org? - Consider mentioning the commit sha (and URL, if it is missing from kernel.org) in the email - Is "intervened" the right word here - the Cambridge dictionary defines it as "to intentionally become involved in a difficult situation in order to improve it or prevent it from getting worse" - Are you contacting a developer only? Have you considered adding the subsystem maintainer and mailing list in the CC list - scripts/get_maintainer.pl will give you those - Have you considered opening a bug report, or better yet sending a patch? Patch does not have to be perfect and if you have doubts you can mention those in the email/cover-letter.

Hope that helps Emil

[1] https://dictionary.cambridge.org/dictionary/english/intervene

On Wed, 15 Mar 2023 at 08:42, Pascal wrote:

...

hi,

I come to you for lack of feedback (I think the Linux kernel

developers have other cats to whip :-))

...

...

would one of you have the answer or a track to follow concerning the question below ? the encountered compilation error is behind the forwarded email.

regards, lacsaP.

---------- Forwarded message --------- De : Pascal Date: mer. 8 mars 2023 à 14:09 Subject: bio_check_ro @ blk-core.c

hi,

I'm addressing you because you intervened (commit) in the function bio_check_ro @ blk-core.c @ Linux-LTS-6.1.15. the last time I intervened on this file (@ Linux-LTS-5.10.19 for

...

...

...

with @ Linux-LTS-6.1.15, if I insert "return true;", I now have a compilation error. in your opinion, is there still a need to "fix" blk-core.c to prevent writing to a read-only locked device and if so, can you help me implement

was confused in my request. the bio_check_ro function present in the blk-core.c source file. the desired result (eg. writes KO on RO blockdevice) ? personal use), it was to replace "return false;" by "return true;", which theoretically should prevent the possible writing on a device locked in read-only mode (see here or here). this fix?

...

...

...

regards, lacsaP. ---------- End forwarded message ---------

SYNC include/config/auto.conf CC arch/x86/kernel/asm-offsets.s CALL scripts/checksyscalls.sh DESCEND objtool DESCEND bpf/resolve_btfids CC block/bdev.o CC block/fops.o CC block/bio.o CC block/elevator.o CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function

returning void [-Werror=return-type]

...

496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2

hi Emil, forgive me if this is the third time I've replied to this message but I received a moderation warning the two previous times that my reply was over 40Kb when my preparation file was only 13Kb... I haven't had any feedback about this moderation since last week, so I don't know if you received my answer... this one concluded that the recent version of LVM integrated in the ISO image of ArchLinux modifies a disk, even if this one and all its partitions are set as read-only (chmod 444 /dev/sdX* && blockdev --setro /dev/sdX* with udev rule). thank you for this investigation, this history. regards; lacsaP. Le mer. 15 mars 2023 à 16:36, Emil Velikov a écrit :

Greetings Pascal,

After following the links I see what's happening. Essentially: - Kernel gained RO/RW correctness check - circa Jan 2018, kernel commit 721c7fc701c71f693307d274d2b346a1ecd4a534 - LVM was initially buggy but later fixed, circa Mar 2018, - Kernel check got partially reverted because broken LVM is still used - circa Aug 2018, kernel commit a32e236eb93e62a0f692e79b7c3c9636689559b9 - People used an out of tree patch, reinstating the correctness check - The function return type was dropped since it is unused - Sep 2022, kernel commit bdb7d420c6f6d2618d4c907cd7742c3195c425e2 - kernel patch no longer applies, correct behaviour cannot be enforced

To unblock yourself, it will be a matter of reverting bdb7d420c6f6d2618d4c907cd7742c3195c425e2 and then a32e236eb93e62a0f692e79b7c3c9636689559b9.

For the mid/long run, one should consider a proper upstream solution:

Assuming I'm in your position, I would dig through the data in the linked commits and estimate which/how many distributions ship with buggy LVM. Then formulate a comprehensive cover letter, proposing a) reverts (if LVM is no longer used in the wild) or b) reverts && a (KCONFIG, sysctl, other) toggle to control the behaviour.

Hope that helps, Emil

On Wed, 15 Mar 2023 at 13:38, Pascal wrote:

...

hi Emil,

in view of your answer and after rereading my email, I realize that I

...

here it is, I hope, more clearly reformulated :-)

first of all, I use ArchLinux to, from time to time, compile the slightly modified LTS kernel, and this from PKGBUILD provided by ArchLinux at some point.

some technologies such as LVM do not take into account the read-only applied on a block device. see the two links provided in the previous exchanges for more details...

until now, I recompiled the kernel by applying a slight modification to

...

the last time I made this modification was on the Linux-LTS-5.10.19 kernel : ( https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... )

$ diff -u 5.10.19.original/blk-core.c 5.10.19.me/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 5.10.19.me/blk-core.c 2023-03-15 13:44:02.353596114 +0100 @@ -706,7 +706,7 @@ "Trying to write to read-only block-device %s (partno %d)\n", bio_devname(bio, b), part->partno); /* Older lvm-tools actually trigger this */ - return false; + return true; }

return false;

the compilation of the modified LTS 5.10.19 kernel went well and the correction seems to do the job...

since this last time (2022/01), the source file blk-core.c has been modified a lot and the bio_check_ro function is part of these modifications : ( https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/block/... )

$ diff -u 5.10.19.original/blk-core.c 6.1.15.original/blk-core.c --- 5.10.19.original/blk-core.c 2023-03-15 13:44:20.176929833 +0100 +++ 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 @@ -14,11 +14,10 @@ */ #include #include -#include #include #include -#include ... @@ -681,40 +483,22 @@ }

late_initcall(fail_make_request_debugfs); - -#else /* CONFIG_FAIL_MAKE_REQUEST */ - -static inline bool should_fail_request(struct hd_struct *part, - unsigned int bytes) -{ - return false; -} - #endif /* CONFIG_FAIL_MAKE_REQUEST */

-static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) +static inline void bio_check_ro(struct bio *bio) { - const int op = bio_op(bio); - - if (part->policy && op_is_write(op)) { - char b[BDEVNAME_SIZE]; - + if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) - return false; - - WARN_ONCE(1, - "Trying to write to read-only block-device %s (partno %d)\n", - bio_devname(bio, b), part->partno); + return; + pr_warn("Trying to write to read-only block-device %pg\n", + bio->bi_bdev); /* Older lvm-tools actually trigger this */ - return false; } - - return false; } ...

when I introduce my little modification (see diff below) in the code, makepkg fails to compile with the error "return with a value in function returning void" (see makepkg below) :

$ diff -u 6.1.15.original/blk-core.c 6.1.15.me/blk-core.c --- 6.1.15.original/blk-core.c 2023-03-15 13:50:36.560271323 +0100 +++ 6.1.15.me/blk-core.c 2023-03-15 13:56:15.246945330 +0100 @@ -493,6 +493,7 @@ pr_warn("Trying to write to read-only block-device %pg\n", bio->bi_bdev); /* Older lvm-tools actually trigger this */ + return true; } }

$ makepkg ... CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function returning void [-Werror=return-type] 496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2

so, how to modify the current code of the bio_check_ro function to get

...

with the changes in the blk-core.c source code since version 5.10.19, is it still necessary to tweak the bio_check_ro function to disallow technologies that ignore the read-only block?

regards, lacsaP.

Le mer. 15 mars 2023 à 12:37, Emil Velikov a écrit :

...

Greetings Pascal,

Couple of suggestions from the peanut gallery, Take them with a heavy pinch of salt: - Is the issue happening with upstream code from kernel.org? - Consider mentioning the commit sha (and URL, if it is missing from kernel.org) in the email - Is "intervened" the right word here - the Cambridge dictionary defines it as "to intentionally become involved in a difficult situation in order to improve it or prevent it from getting worse" - Are you contacting a developer only? Have you considered adding the subsystem maintainer and mailing list in the CC list - scripts/get_maintainer.pl will give you those - Have you considered opening a bug report, or better yet sending a patch? Patch does not have to be perfect and if you have doubts you can mention those in the email/cover-letter.

Hope that helps Emil

[1] https://dictionary.cambridge.org/dictionary/english/intervene

On Wed, 15 Mar 2023 at 08:42, Pascal wrote:

...

hi,

I come to you for lack of feedback (I think the Linux kernel

developers have other cats to whip :-))

...

...

would one of you have the answer or a track to follow concerning the question below ? the encountered compilation error is behind the forwarded email.

regards, lacsaP.

---------- Forwarded message --------- De : Pascal Date: mer. 8 mars 2023 à 14:09 Subject: bio_check_ro @ blk-core.c

hi,

I'm addressing you because you intervened (commit) in the function bio_check_ro @ blk-core.c @ Linux-LTS-6.1.15. the last time I intervened on this file (@ Linux-LTS-5.10.19 for

...

...

...

with @ Linux-LTS-6.1.15, if I insert "return true;", I now have a compilation error. in your opinion, is there still a need to "fix" blk-core.c to prevent writing to a read-only locked device and if so, can you help me implement

was confused in my request. the bio_check_ro function present in the blk-core.c source file. the desired result (eg. writes KO on RO blockdevice) ? personal use), it was to replace "return false;" by "return true;", which theoretically should prevent the possible writing on a device locked in read-only mode (see here or here). this fix?

...

...

...

regards, lacsaP. ---------- End forwarded message ---------

SYNC include/config/auto.conf CC arch/x86/kernel/asm-offsets.s CALL scripts/checksyscalls.sh DESCEND objtool DESCEND bpf/resolve_btfids CC block/bdev.o CC block/fops.o CC block/bio.o CC block/elevator.o CC block/blk-core.o block/blk-core.c: In function 'bio_check_ro': block/blk-core.c:496:24: error: 'return' with a value, in function

returning void [-Werror=return-type]

...

496 | return true; | ^~~~ block/blk-core.c:488:20: note: declared here 488 | static inline void bio_check_ro(struct bio *bio) | ^~~~~~~~~~~~ cc1: some warnings being treated as errors make[2]: *** [scripts/Makefile.build:250: block/blk-core.o] Error 1 make[1]: *** [scripts/Makefile.build:500: block] Error 2 make: *** [Makefile:2005: .] Error 2