Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Qiujun Huang <hqjagain@gmail.com>
Content-Type: text/plain;
        charset=us-ascii
Content-Transfer-Encoding: quoted-printable
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3696.80.82.1.1\))
Subject: [PATCH]pstore/zone: Preallocate zone buffer for 
 psz_kmsg_write_record
Message-Id: <778A2949-C5E7-4C4A-AF30-57CD1B45B3C0@gmail.com>
Date:   Wed, 28 Dec 2022 21:04:32 +0800
Cc:     anton@enomsg.org, ccross@android.com, tony.luck@intel.com,
        LKML <linux-kernel@vger.kernel.org>
To:     keescook@chromium.org
Precedence: bulk

The case found when triggering a panic_on_oom, pstore fails to dump
kmsg.
Partly fixed by: commit 99b3b837855b ("pstore/zone: Use GFP_ATOMIC to
allocate zone buffer"). After the patch, it also fails sometimes. As I =
use
64k as kmsg_size, it's hard to get 4-order free pages sometimes at this
case. So it's better to preallocate the buffer.

Signed-off-by: Qiujun Huang <hqjagain@gmail.com>
---
fs/pstore/zone.c | 26 +++++++++++++++++---------
1 file changed, 17 insertions(+), 9 deletions(-)

diff --git a/fs/pstore/zone.c b/fs/pstore/zone.c
index 2770746bb7aa..86e47dbc9d53 100644
--- a/fs/pstore/zone.c
+++ b/fs/pstore/zone.c
@@ -136,6 +136,7 @@ struct psz_context {
	struct mutex pstore_zone_info_lock;
	struct pstore_zone_info *pstore_zone_info;
	struct pstore_info pstore;
+	struct psz_buffer *kmsg_buf_swap;
};
static struct psz_context pstore_zone_cxt;

@@ -758,14 +759,11 @@ static inline int notrace =
psz_kmsg_write_record(struct psz_context *cxt,
		if (unlikely(!zone))
			return -ENOSPC;

-		/* avoid destroying old data, allocate a new one */
+		/* avoid destroying old data, use the swap buffer */
		len =3D zone->buffer_size + sizeof(*zone->buffer);
		zone->oldbuf =3D zone->buffer;
-		zone->buffer =3D kzalloc(len, GFP_ATOMIC);
-		if (!zone->buffer) {
-			zone->buffer =3D zone->oldbuf;
-			return -ENOMEM;
-		}
+		zone->buffer =3D cxt->kmsg_buf_swap;
+		memset(zone->buffer, 0, len);
		zone->buffer->sig =3D zone->oldbuf->sig;

		pr_debug("write %s to zone id %d\n", zone->name, =
zonenum);
@@ -776,15 +774,14 @@ static inline int notrace =
psz_kmsg_write_record(struct psz_context *cxt,
		if (likely(!ret || ret !=3D -ENOMSG)) {
			cxt->kmsg_write_cnt =3D zonenum + 1;
			cxt->kmsg_write_cnt %=3D cxt->kmsg_max_cnt;
-			/* no need to try next zone, free last zone =
buffer */
-			kfree(zone->oldbuf);
+			/* no need to try next zone, put last one to =
swap buffer */
+			cxt->kmsg_buf_swap =3D zone->oldbuf;
			zone->oldbuf =3D NULL;
			return ret;
		}

		pr_debug("zone %u may be broken, try next dmesg zone\n",
				zonenum);
-		kfree(zone->buffer);
		zone->buffer =3D zone->oldbuf;
		zone->oldbuf =3D NULL;
	}
@@ -1373,6 +1370,12 @@ int register_pstore_zone(struct pstore_zone_info =
*info)
			err =3D -ENOMEM;
			goto fail_free;
		}
+
+		cxt->kmsg_buf_swap =3D kzalloc(info->kmsg_size, =
GFP_KERNEL);
+		if (!cxt->kmsg_buf_swap) {
+			err =3D -ENOMEM;
+			goto fail_free;
+		}
	}
	cxt->pstore.data =3D cxt;

@@ -1411,6 +1414,8 @@ int register_pstore_zone(struct pstore_zone_info =
*info)
	return 0;

fail_free:
+	kfree(cxt->kmsg_buf_swap);
+	cxt->kmsg_buf_swap =3D NULL;
	kfree(cxt->pstore.buf);
	cxt->pstore.buf =3D NULL;
	cxt->pstore.bufsize =3D 0;
@@ -1445,6 +1450,9 @@ void unregister_pstore_zone(struct =
pstore_zone_info *info)
	flush_delayed_work(&psz_cleaner);

	/* Clean up allocations. */
+	kfree(cxt->kmsg_buf_swap);
+	cxt->kmsg_buf_swap =3D NULL;
+
	kfree(cxt->pstore.buf);
	cxt->pstore.buf =3D NULL;
	cxt->pstore.bufsize =3D 0;
--=20
2.30.2=