From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-alma10-1.taild15c8.ts.net [100.103.45.18]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id EC8093EAC90; Mon, 29 Jun 2026 14:56:08 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=100.103.45.18 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1782744970; cv=none; b=qXKaroovqPZkVkfj9hWVr4v+WRl7o/Fs0ZrDyRM5wfcysvr9Z9cCbpxzBrBlhr40Hjegf2gzVw7nDA9EwDVU4spSe0MVLlb+iCHhS/IPb7ZtRRLAz1FybK3wiTydZHrZPRg4v8DEKjt4TLQKkiO0uPp1vbIh2BN8yAgb6g19ZEE= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1782744970; c=relaxed/simple; bh=IMDYlYkmjNZwCd98Nzj9Zb+VjSJnlKtJVRbIJ+wWLtw=; h=Message-ID:Subject:From:To:Cc:Date:In-Reply-To:References: Content-Type:MIME-Version; b=WIaPi4unLRRk6kSvYA1yPqHaKksCYaQl4reK4egsdRwOdkNxXdIpbQHpkBEG+YMjf9/JmLHI7HCwjfJvrXiYAswuKJd9q8EmcQTGKg4zh0c522eIAfI0A3WfkLXODf9un1YqZnsmO11/l8lAF0nH/IRm84/PF37MEBUxOPnDTwA= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=KrgWKuuo; arc=none smtp.client-ip=100.103.45.18 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="KrgWKuuo" Received: by smtp.kernel.org (Postfix) with ESMTPSA id D9B691F00A3A; Mon, 29 Jun 2026 14:56:07 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel.org; s=k20260515; t=1782744968; bh=pRNh9BtnZCGX1do1vvMu/LN6jH/9U5QYoFoUa0UEtf4=; h=Subject:From:To:Cc:Date:In-Reply-To:References; b=KrgWKuuojMvGf7r51WDpTkcCVyrcOxmFR/LFWz/vhDp7z5bhd4j6pk/HFAbwmLsPb D2PV860I19iBp/odh9UIqU6fYTG3b4feyh4qiouKOeI5iwZHQfYSAmuJ2hQ5dzfQo6 Mw92BMPzfdTVdpbvCz1OSE4sgTHmJo2pMzpPtqoV8Q1gzq3XwJRfeoHV2Wt/xnP9hB 9yixtVYZnEE6zhTVSkLmSjt6rUXWEyMaMt+EnObxyEPGwq3N5lZaOQ7dE8R0uuV79M tGWWaOB62zKzx/vUIBFCR4xYDSQpaPV3+SrvgkPXJY7DBH7XfhVUQCogIyr0DK1zOU R2iCXa7dBtH6w== Message-ID: <5084d9c6b6a7025be156354194cf7d3fffebb0e2.camel@kernel.org> Subject: Re: [PATCH] btrfs: don't let shrinker touch extent_maps that are being logged From: Jeff Layton To: Filipe Manana Cc: Chris Mason , David Sterba , Filipe Manana , Josef Bacik , linux-btrfs@vger.kernel.org, linux-kernel@vger.kernel.org Date: Mon, 29 Jun 2026 10:56:06 -0400 In-Reply-To: References: <20260629-btrfs-skip-logging-v1-1-4e3a28c1acaf@kernel.org> Autocrypt: addr=jlayton@kernel.org; prefer-encrypt=mutual; keydata=mQINBE6V0TwBEADXhJg7s8wFDwBMEvn0qyhAnzFLTOCHooMZyx7XO7dAiIhDSi7G1NPxw n8jdFUQMCR/GlpozMFlSFiZXiObE7sef9rTtM68ukUyZM4pJ9l0KjQNgDJ6Fr342Htkjxu/kFV1Wv egyjnSsFt7EGoDjdKqr1TS9syJYFjagYtvWk/UfHlW09X+jOh4vYtfX7iYSx/NfqV3W1D7EDi0PqV T2h6v8i8YqsATFPwO4nuiTmL6I40ZofxVd+9wdRI4Db8yUNA4ZSP2nqLcLtFjClYRBoJvRWvsv4lm 0OX6MYPtv76hka8lW4mnRmZqqx3UtfHX/hF/zH24Gj7A6sYKYLCU3YrI2Ogiu7/ksKcl7goQjpvtV YrOOI5VGLHge0awt7bhMCTM9KAfPc+xL/ZxAMVWd3NCk5SamL2cE99UWgtvNOIYU8m6EjTLhsj8sn VluJH0/RcxEeFbnSaswVChNSGa7mXJrTR22lRL6ZPjdMgS2Km90haWPRc8Wolcz07Y2se0xpGVLEQ cDEsvv5IMmeMe1/qLZ6NaVkNuL3WOXvxaVT9USW1+/SGipO2IpKJjeDZfehlB/kpfF24+RrK+seQf CBYyUE8QJpvTZyfUHNYldXlrjO6n5MdOempLqWpfOmcGkwnyNRBR46g/jf8KnPRwXs509yAqDB6sE LZH+yWr9LQZEwARAQABtCVKZWZmIExheXRvbiA8amxheXRvbkBwb29jaGllcmVkcy5uZXQ+iQI7BB MBAgAlAhsDBgsJCAcDAgYVCAIJCgsEFgIDAQIeAQIXgAUCTpXWPAIZAQAKCRAADmhBGVaCFc65D/4 gBLNMHopQYgG/9RIM3kgFCCQV0pLv0hcg1cjr+bPI5f1PzJoOVi9s0wBDHwp8+vtHgYhM54yt43uI 7Htij0RHFL5eFqoVT4TSfAg2qlvNemJEOY0e4daljjmZM7UtmpGs9NN0r9r50W82eb5Kw5bc/r0km R/arUS2st+ecRsCnwAOj6HiURwIgfDMHGPtSkoPpu3DDp/cjcYUg3HaOJuTjtGHFH963B+f+hyQ2B rQZBBE76ErgTDJ2Db9Ey0kw7VEZ4I2nnVUY9B5dE2pJFVO5HJBMp30fUGKvwaKqYCU2iAKxdmJXRI ONb7dSde8LqZahuunPDMZyMA5+mkQl7kpIpR6kVDIiqmxzRuPeiMP7O2FCUlS2DnJnRVrHmCljLkZ Wf7ZUA22wJpepBligemtSRSbqCyZ3B48zJ8g5B8xLEntPo/NknSJaYRvfEQqGxgk5kkNWMIMDkfQO lDSXZvoxqU9wFH/9jTv1/6p8dHeGM0BsbBLMqQaqnWiVt5mG92E1zkOW69LnoozE6Le+12DsNW7Rj iR5K+27MObjXEYIW7FIvNN/TQ6U1EOsdxwB8o//Yfc3p2QqPr5uS93SDDan5ehH59BnHpguTc27Xi QQZ9EGiieCUx6Zh2ze3X2UW9YNzE15uKwkkuEIj60NvQRmEDfweYfOfPVOueC+iFifbQgSmVmZiBM YXl0b24gPGpsYXl0b25AcmVkaGF0LmNvbT6JAjgEEwECACIFAk6V0q0CGwMGCwkIBwMCBhUIAgkKC wQWAgMBAh4BAheAAAoJEAAOaEEZVoIViKUQALpvsacTMWWOd7SlPFzIYy2/fjvKlfB/Xs4YdNcf9q LqF+lk2RBUHdR/dGwZpvw/OLmnZ8TryDo2zXVJNWEEUFNc7wQpl3i78r6UU/GUY/RQmOgPhs3epQC 3PMJj4xFx+VuVcf/MXgDDdBUHaCTT793hyBeDbQuciARDJAW24Q1RCmjcwWIV/pgrlFa4lAXsmhoa c8UPc82Ijrs6ivlTweFf16VBc4nSLX5FB3ls7S5noRhm5/Zsd4PGPgIHgCZcPgkAnU1S/A/rSqf3F LpU+CbVBDvlVAnOq9gfNF+QiTlOHdZVIe4gEYAU3CUjbleywQqV02BKxPVM0C5/oVjMVx3bri75n1 TkBYGmqAXy9usCkHIsG5CBHmphv9MHmqMZQVsxvCzfnI5IO1+7MoloeeW/lxuyd0pU88dZsV/riHw 87i2GJUJtVlMl5IGBNFpqoNUoqmvRfEMeXhy/kUX4Xc03I1coZIgmwLmCSXwx9MaCPFzV/dOOrju2 xjO+2sYyB5BNtxRqUEyXglpujFZqJxxau7E0eXoYgoY9gtFGsspzFkVNntamVXEWVVgzJJr/EWW0y +jNd54MfPRqH+eCGuqlnNLktSAVz1MvVRY1dxUltSlDZT7P2bUoMorIPu8p7ZCg9dyX1+9T6Muc5d Hxf/BBP/ir+3e8JTFQBFOiLNdFtB9KZWZmIExheXRvbiA8amxheXRvbkBzYW1iYS5vcmc+iQI4BBM BAgAiBQJOldK9AhsDBgsJCAcDAgYVCAIJCgsEFgIDAQIeAQIXgAAKCRAADmhBGVaCFWgWD/0ZRi4h N9FK2BdQs9RwNnFZUr7JidAWfCrs37XrA/56olQl3ojn0fQtrP4DbTmCuh0SfMijB24psy1GnkPep naQ6VRf7Dxg/Y8muZELSOtsv2CKt3/02J1BBitrkkqmHyni5fLLYYg6fub0T/8Kwo1qGPdu1hx2BQ RERYtQ/S5d/T0cACdlzi6w8rs5f09hU9Tu4qV1JLKmBTgUWKN969HPRkxiojLQziHVyM/weR5Reu6 FZVNuVBGqBD+sfk/c98VJHjsQhYJijcsmgMb1NohAzwrBKcSGKOWJToGEO/1RkIN8tqGnYNp2G+aR 685D0chgTl1WzPRM6mFG1+n2b2RR95DxumKVpwBwdLPoCkI24JkeDJ7lXSe3uFWISstFGt0HL8Eew P8RuGC8s5h7Ct91HMNQTbjgA+Vi1foWUVXpEintAKgoywaIDlJfTZIl6Ew8ETN/7DLy8bXYgq0Xzh aKg3CnOUuGQV5/nl4OAX/3jocT5Cz/OtAiNYj5mLPeL5z2ZszjoCAH6caqsF2oLyAnLqRgDgR+wTQ T6gMhr2IRsl+cp8gPHBwQ4uZMb+X00c/Amm9VfviT+BI7B66cnC7Zv6Gvmtu2rEjWDGWPqUgccB7h dMKnKDthkA227/82tYoFiFMb/NwtgGrn5n2vwJyKN6SEoygGrNt0SI84y6hEVbQlSmVmZiBMYXl0b 24gPGpsYXl0b25AcHJpbWFyeWRhdGEuY29tPokCOQQTAQIAIwUCU4xmKQIbAwcLCQgHAwIBBhUIAg kKCwQWAgMBAh4BAheAAAoJEAAOaEEZVoIV1H0P/j4OUTwFd7BBbpoSp695qb6HqCzWMuExsp8nZjr uymMaeZbGr3OWMNEXRI1FWNHMtcMHWLP/RaDqCJil28proO+PQ/yPhsr2QqJcW4nr91tBrv/MqItu AXLYlsgXqp4BxLP67bzRJ1Bd2x0bWXurpEXY//VBOLnODqThGEcL7jouwjmnRh9FTKZfBDpFRaEfD FOXIfAkMKBa/c9TQwRpx2DPsl3eFWVCNuNGKeGsirLqCxUg5kWTxEorROppz9oU4HPicL6rRH22Ce 6nOAON2vHvhkUuO3GbffhrcsPD4DaYup4ic+DxWm+DaSSRJ+e1yJvwi6NmQ9P9UAuLG93S2MdNNbo sZ9P8k2mTOVKMc+GooI9Ve/vH8unwitwo7ORMVXhJeU6Q0X7zf3SjwDq2lBhn1DSuTsn2DbsNTiDv qrAaCvbsTsw+SZRwF85eG67eAwouYk+dnKmp1q57LDKMyzysij2oDKbcBlwB/TeX16p8+LxECv51a sjS9TInnipssssUDrHIvoTTXWcz7Y5wIngxDFwT8rPY3EggzLGfK5Zx2Q5S/N0FfmADmKknG/D8qG IcJE574D956tiUDKN4I+/g125ORR1v7bP+OIaayAvq17RP+qcAqkxc0x8iCYVCYDouDyNvWPGRhbL UO7mlBpjW9jK9e2fvZY9iw3QzIPGKtClKZWZmIExheXRvbiA8amVmZi5sYXl0b25AcHJpbWFyeWRh dGEuY29tPokCOQQTAQIAIwUCU4xmUAIbAwcLCQgHAwIBBhUIAgkKCwQWAgMBAh4BAheAAAoJEAAOa EEZVoIVzJoQALFCS6n/FHQS+hIzHIb56JbokhK0AFqoLVzLKzrnaeXhE5isWcVg0eoV2oTScIwUSU apy94if69tnUo4Q7YNt8/6yFM6hwZAxFjOXR0ciGE3Q+Z1zi49Ox51yjGMQGxlakV9ep4sV/d5a50 M+LFTmYSAFp6HY23JN9PkjVJC4PUv5DYRbOZ6Y1+TfXKBAewMVqtwT1Y+LPlfmI8dbbbuUX/kKZ5d dhV2736fgyfpslvJKYl0YifUOVy4D1G/oSycyHkJG78OvX4JKcf2kKzVvg7/Rnv+AueCfFQ6nGwPn 0P91I7TEOC4XfZ6a1K3uTp4fPPs1Wn75X7K8lzJP/p8lme40uqwAyBjk+IA5VGd+CVRiyJTpGZwA0 jwSYLyXboX+Dqm9pSYzmC9+/AE7lIgpWj+3iNisp1SWtHc4pdtQ5EU2SEz8yKvDbD0lNDbv4ljI7e flPsvN6vOrxz24mCliEco5DwhpaaSnzWnbAPXhQDWb/lUgs/JNk8dtwmvWnqCwRqElMLVisAbJmC0 BhZ/Ab4sph3EaiZfdXKhiQqSGdK4La3OTJOJYZphPdGgnkvDV9Pl1QZ0ijXQrVIy3zd6VCNaKYq7B AKidn5g/2Q8oio9Tf4XfdZ9dtwcB+bwDJFgvvDYaZ5bI3ln4V3EyW5i2NfXazz/GA/I/ZtbsigCFc 8ftCBKZWZmIExheXRvbiA8amxheXRvbkBrZXJuZWwub3JnPokCOAQTAQIAIgUCWe8u6AIbAwYLCQg HAwIGFQgCCQoLBBYCAwECHgECF4AACgkQAA5oQRlWghUuCg/+Lb/xGxZD2Q1oJVAE37uW308UpVSD 2tAMJUvFTdDbfe3zKlPDTuVsyNsALBGclPLagJ5ZTP+Vp2irAN9uwBuacBOTtmOdz4ZN2tdvNgozz uxp4CHBDVzAslUi2idy+xpsp47DWPxYFIRP3M8QG/aNW052LaPc0cedYxp8+9eiVUNpxF4SiU4i9J DfX/sn9XcfoVZIxMpCRE750zvJvcCUz9HojsrMQ1NFc7MFT1z3MOW2/RlzPcog7xvR5ENPH19ojRD CHqumUHRry+RF0lH00clzX/W8OrQJZtoBPXv9ahka/Vp7kEulcBJr1cH5Wz/WprhsIM7U9pse1f1g Yy9YbXtWctUz8uvDR7shsQxAhX3qO7DilMtuGo1v97I/Kx4gXQ52syh/w6EBny71CZrOgD6kJwPVV AaM1LRC28muq91WCFhs/nzHozpbzcheyGtMUI2Ao4K6mnY+3zIuXPygZMFr9KXE6fF7HzKxKuZMJO aEZCiDOq0anx6FmOzs5E6Jqdpo/mtI8beK+BE7Va6ni7YrQlnT0i3vaTVMTiCThbqsB20VrbMjlhp f8lfK1XVNbRq/R7GZ9zHESlsa35ha60yd/j3pu5hT2xyy8krV8vGhHvnJ1XRMJBAB/UYb6FyC7S+m QZIQXVeAA+smfTT0tDrisj1U5x6ZB9b3nBg65kc= Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable User-Agent: Evolution 3.60.2 (3.60.2-1.fc44) Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 On Mon, 2026-06-29 at 15:19 +0100, Filipe Manana wrote: > On Mon, Jun 29, 2026 at 2:41=E2=80=AFPM Jeff Layton = wrote: > >=20 > > The extent map shrinker can free an extent map that is still owned by a= n > > in-flight fsync and still linked on the inode's modified_extents list, > > corrupting that list and eventually causing an RCU stall. > >=20 > > btrfs_scan_inode() currently skips EXTENT_FLAG_PINNED maps, then calls > > btrfs_remove_extent_mapping() followed by btrfs_free_extent_map(): > >=20 > > if (em->flags & EXTENT_FLAG_PINNED) > > goto next; > > ... > > btrfs_remove_extent_mapping(inode, em); > > btrfs_free_extent_map(em); > >=20 > > But btrfs_remove_extent_mapping() deliberately does NOT unlink a map th= at > > has EXTENT_FLAG_LOGGING set: > >=20 > > if (!(em->flags & EXTENT_FLAG_LOGGING)) > > list_del_init(&em->list); > > remove_em(inode, em); > >=20 > > This sets up a UAF situation where a later fsync() can trip over the > > now-freed extent_map still on the modified_extents() list. >=20 > I don't see how that can happen, because fsync always takes the > inode's i_mmap_lock (in write/exclusive mode), and btrfs_scan_inode() > takes the same lock too (in shared/read mode). > EXTENT_FLAG_LOGGING is only set and cleared during fsync, so how can > the shrinker race with fsync? >=20 ...and yet, the vmcore shows that the shrinker hit an extent_map that had LOGGING set while holding the i_mmap_lock. I had hoped to add an assertion here that LOGGING couldn't be set on the em's after taking the i_mmap_lock, but that's not viable: The problem is that the i_mmap_lock coverage isn't complete. I had Claude assist me with some analysis so take it with a hefty grain of salt but the problem is that the em tree can contain em's from other inodes where the i_mmap_lock is not held: -----------------------8<--------------------- ...that inode is in progress, therefore no em in its tree should have LOGGI= NG set. Your invariant is the right one. But it's violated, because the i_mmap_lock interlock is incomplete. Two fa= cts: = =20 = = =20 1. tree-log.c never takes i_mmap_lock =E2=80=94 grep returns 0 occurrence= s. Only btrfs_sync_file takes it (file.c:1608/1610), and only = = =20 on the one inode being fsync'd. = = =20 2. A single fsync logs many other inodes, and each can set LOGGING on its= own ems via btrfs_log_changed_extents(). = =20 btrfs_log_inode() is called recursively on conflicting inodes, parent dir= ectories, and referenced inodes: = =20 tree-log.c:5982 btrfs_log_inode(di_inode, log_mode) # dir dentrie= s = =20 tree-log.c:6339 btrfs_log_inode(inode, LOG_INODE_ALL) # conflicting= inodes = =20 tree-log.c:6900 btrfs_log_inode(di_inode, log_mode) # new dir den= tries = =20 tree-log.c:7389 btrfs_log_inode(dir_inode, LOG_INODE_ALL) # log_all_par= ents = =20 tree-log.c:7488 btrfs_log_inode(inode, ...) # referenced = inodes = =20 2. For all of these, btrfs_sync_file holds i_mmap_lock on the original fs= ync'd inode =E2=80=94 not on these side inodes. So a side = = =20 inode X gets LOGGING set on its ems while X's own i_mmap_lock is free. = = =20 = = =20 So the shrinker can down_read_trylock(X->i_mmap_lock) successfully (nobod= y holds it) and walk X's tree while X's ems are = =20 LOGGING-flagged =E2=86=92 it removes+frees one =E2=86=92 corruption. The = interlock only protects the directly-fsync'd inode, not the inodes = = =20 logged as its dependencies. = = =20 > Also, there should be no () after modified_extents (it's not a function n= ame). >=20 >=20 Sorry, early morning typo. Fixed in my local repo. >=20 > >=20 > > Fix it by having the shrinker skip maps that are being logged, the same > > way it skips pinned maps. Such a map is owned by the in-flight fsync an= d > > will become reclaimable again once logging clears the flag. > >=20 > > Fixes: 956a17d9d050 ("btrfs: add a shrinker for extent maps") > > Signed-off-by: Jeff Layton > > --- > > We've started hitting a number of these problems in our fleet. It > > seems to mostly happen on ARM64 architecture, but there have been some > > WARN_ONs that popped on x86_64 too. > > --- > > fs/btrfs/extent_map.c | 8 +++++++- > > 1 file changed, 7 insertions(+), 1 deletion(-) > >=20 > > diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c > > index fce9c5cc0122..128f7800e101 100644 > > --- a/fs/btrfs/extent_map.c > > +++ b/fs/btrfs/extent_map.c > > @@ -1166,7 +1166,13 @@ static long btrfs_scan_inode(struct btrfs_inode = *inode, struct btrfs_em_shrink_c > > em =3D rb_entry(node, struct extent_map, rb_node); > > ctx->scanned++; > >=20 > > - if (em->flags & EXTENT_FLAG_PINNED) > > + /* > > + * Skip extent maps that are pinned or are being logged= . The > > + * i_mmap_lock should prevent this from seeing LOGGING = on extent_maps > > + * directly associated with inode, but em may be associ= ated with > > + * other, dependent inodes and their locks are not held= . > > + */ > > + if (em->flags & (EXTENT_FLAG_PINNED | EXTENT_FLAG_LOGGI= NG)) > > goto next; > >=20 > > /* > >=20 > > --- > > base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482 > > change-id: 20260629-btrfs-skip-logging-3e31701d9647 > >=20 > > Best regards, > > -- > > Jeff Layton > >=20 > >=20 --=20 Jeff Layton