[RESEND] "ext4: get rid of ppath in get_ext_path()" 6.6.y backport request

["Yoann Congal" <yoann.congal@smile.fr>]

Hello,

(Resent with developers/maintainers of the patch in CC)

I'd like to request the backport of
6b854d552711 ("ext4: get rid of ppath in get_ext_path()")
on the 6.6.y branch.

Rational:
6.6.130 commit fb138df7d886 ("ext4: get rid of ppath in ext4_ext_insert_extent()")
created a regression in ext4_ext_map_blocks() by changing the path value
under error (NULL -> ERR_PTR). But path is only checked for NULL value
in ext4_free_ext_path (not ERR_PTR).

The check is added in 6b854d552711 ("ext4: get rid of ppath in get_ext_path()"),
hence this backport request.

More details:
This regression was triggered during LTP test on a 6.6.129->6.6.142
upgrade for a Yocto Project stable branch:
https://autobuilder.yoctoproject.org/valkyrie/#/builders/98/builds/3837
-> https://valkyrie.yocto.io/pub/non-release/20260622-121/testresults/qemuarm64-ltp/core-image-sato/qemu_boot_log.20260623002740

[ 6952.500858] Unable to handle kernel paging request at virtual address ffffffffffffffec
[ 6952.503768] Mem abort info:
[ 6952.504431]   ESR = 0x0000000096000005
[ 6952.505333]   EC = 0x25: DABT (current EL), IL = 32 bits
[ 6952.506541]   SET = 0, FnV = 0
[ 6952.507354]   EA = 0, S1PTW = 0
[ 6952.508154]   FSC = 0x05: level 1 translation fault
[ 6952.509208] Data abort info:
[ 6952.509849]   ISV = 0, ISS = 0x00000005, ISS2 = 0x00000000
[ 6952.511175]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 6952.512372]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 6952.513667] swapper pgtable: 4k pages, 39-bit VAs, pgdp=0000000041250000
[ 6952.514909] [ffffffffffffffec] pgd=0000000000000000, p4d=0000000000000000, pud=0000000000000000
[ 6952.516423] Internal error: Oops: 0000000096000005 [#1] PREEMPT SMP
[ 6952.517503] Modules linked in: x_tables tun loop [last unloaded: ip6_tables]
[ 6952.518691] CPU: 1 PID: 1078 Comm: kworker/u12:1 Tainted: G        W          6.6.142-yocto-standard #1
[ 6952.520269] Hardware name: linux,dummy-virt (DT)
[ 6952.521094] Workqueue: writeback wb_workfn (flush-7:0)
[ 6952.521985] pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 6952.523184] pc : ext4_ext_map_blocks+0x260/0x1860
[ 6952.524011] lr : ext4_ext_map_blocks+0xdb8/0x1860
[ 6952.524851] sp : ffffffc086a3b620
[ 6952.525421] x29: ffffffc086a3b740 x28: ffffffffffffffe4 x27: 000000000000808c
[ 6952.526624] x26: ffffff8017dd9000 x25: 000000000000808c x24: 0000000000000002
[ 6952.527849] x23: ffffff8035e766c8 x22: ffffff802e589690 x21: 000000000000042f
[ 6952.529087] x20: ffffffc086a3b948 x19: ffffff8035e767f0 x18: 0000000000000000
[ 6952.530310] x17: ffffffc081691310 x16: fffffffe001ab548 x15: 0000005564d4cb48
[ 6952.531519] x14: 00000000ffffffff x13: 0000000000000000 x12: ffffffffffffffc0
[ 6952.532683] x11: 0000000000000040 x10: ffffff8005d81d80 x9 : ffffffc0803cce14
[ 6952.533886] x8 : 00000000bab647bc x7 : 0000000000000000 x6 : 000000000000d847
[ 6952.535065] x5 : 0000000000000000 x4 : 0000000000316019 x3 : 0000000000000000
[ 6952.536264] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffffff803deec880
[ 6952.537425] Call trace:
[ 6952.537860]  ext4_ext_map_blocks+0x260/0x1860
[ 6952.538589]  ext4_map_blocks+0x19c/0x598
[ 6952.539258]  ext4_do_writepages+0x5a4/0xbe0
[ 6952.539977]  ext4_writepages+0x84/0x110
[ 6952.540624]  do_writepages+0x94/0x1e0
[ 6952.541240]  __writeback_single_inode+0x60/0x4d8
[ 6952.542086]  writeback_sb_inodes+0x208/0x4b0
[ 6952.542812]  __writeback_inodes_wb+0x58/0x118
[ 6952.543578]  wb_writeback+0x274/0x440
[ 6952.544198]  wb_workfn+0x3b0/0x5c8
[ 6952.544788]  process_one_work+0x16c/0x3e0
[ 6952.545434]  worker_thread+0x1b4/0x378
[ 6952.546059]  kthread+0x118/0x128
[ 6952.546599]  ret_from_fork+0x10/0x20
[ 6952.547197] Code: 2a0103f9 b9009fe1 b9000e99 b40055fc (79401398)
[ 6952.548170] ---[ end trace 0000000000000000 ]---
[ 6952.551090] ------------[ cut here ]------------

Reading the resulting code in 6.6.142:
fs/ext4/extents.c:
int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
			struct ext4_map_blocks *map, int flags)
{
	struct ext4_ext_path *path = NULL;
	// ...

got_allocated_blocks:
	path = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
	if (IS_ERR(path)) {
		err = PTR_ERR(path);
		/*
		 * Gracefully handle out of space conditions. If the filesystem
		 * is inconsistent, we'll just leak allocated blocks to avoid
		 * causing even more damage.
		 */
		// ...
		goto out;
	}

	// ...
out:
	ext4_free_ext_path(path);

	trace_ext4_ext_map_blocks_exit(inode, flags, map,
				       err ? err : allocated);
	return err ? err : allocated;
}

=> Under out of space condition (what LTP does a *LOT*): path is given unmodified to
ext4_free_ext_path() that only does a NULL check (no IS_ERR) before
dereferencing it. And that produces the oops and then, the LTP failure.

Notably, master commit 6b854d552711 ("ext4: get rid of ppath in get_ext_path()")
never got backported to 6.6.y. But does add the IS_ERR_OR_NULL() check
to ext4_free_ext_path:
 void ext4_free_ext_path(struct ext4_ext_path *path)
 {
+       if (IS_ERR_OR_NULL(path))
+               return;

Thanks!
-- 
Yoann Congal
Smile ECS