Skip to contentSkip to conversation

fs: jfs: fix shift-out-of-bounds in dbAllocAG

PATCH New Sasha Levin 1 participant 8 messages Dec 17, 2022

Subsystems (6)
Stable branch VFS ACPI Live update +2 more subsystems
lore
Share & tools
lore

Sasha Levin

author maintainer
#0 lore raw

PATCH 1/8 fs: jfs: fix shift-out-of-bounds in dbAllocAG

From: Dongliang Mu <[email protected]>

[ Upstream commit 898f706695682b9954f280d95e49fa86ffa55d08 ]

Syzbot found a crash : UBSAN: shift-out-of-bounds in dbAllocAG. The underlying bug is the missing check of bmp->db_agl2size. The field can be greater than 64 and trigger the shift-out-of-bounds.

Fix this bug by adding a check of bmp->db_agl2size in dbMount since this field is used in many following functions. The upper bound for this field is L2MAXL2SIZE - L2MAXAG, thanks for the help of Dave Kleikamp. Note that, for maintenance, I reorganized error handling code of dbMount.

---
 fs/jfs/jfs_dmap.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)
fs/jfs/jfs_dmap.c +16 -6
index 1014f2a24697..4f72f7dee78b 100644
@@ -168,7 +168,7 @@ int dbMount(struct inode *ipbmap)
 	struct bmap *bmp;
 	struct dbmap_disk *dbmp_le;
 	struct metapage *mp;
-	int i;
+	int i, err;
 
 	/*
 	 * allocate/initialize the in-memory bmap descriptor
@@ -183,8 +183,8 @@ int dbMount(struct inode *ipbmap)
 			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
 			   PSIZE, 0);
 	if (mp == NULL) {
-		kfree(bmp);
-		return -EIO;
+		err = -EIO;
+		goto err_kfree_bmp;
 	}
 
 	/* copy the on-disk bmap descriptor to its in-memory version. */
@@ -194,9 +194,8 @@ int dbMount(struct inode *ipbmap)
 	bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
 	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
 	if (!bmp->db_numag) {
-		release_metapage(mp);
-		kfree(bmp);
-		return -EINVAL;
+		err = -EINVAL;
+		goto err_release_metapage;
 	}
 
 	bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
@@ -207,6 +206,11 @@ int dbMount(struct inode *ipbmap)
 	bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
 	bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
 	bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
+	if (bmp->db_agl2size > L2MAXL2SIZE - L2MAXAG) {
+		err = -EINVAL;
+		goto err_release_metapage;
+	}
+
 	for (i = 0; i < MAXAG; i++)
 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
@@ -227,6 +231,12 @@ int dbMount(struct inode *ipbmap)
 	BMAP_LOCK_INIT(bmp);
 
 	return (0);
+
+err_release_metapage:
+	release_metapage(mp);
+err_kfree_bmp:
+	kfree(bmp);
+	return err;
 }
 
 
Signature · 1 line
2.35.1
author maintainer
#1 lore raw

PATCH 2/8 udf: Avoid double brelse() in udf_rename()

From: Shigeru Yoshida <[email protected]>

[ Upstream commit c791730f2554a9ebb8f18df9368dc27d4ebc38c2 ]

syzbot reported a warning like below [1]:

VFS: brelse: Trying to free free buffer

WARNING: CPU: 2 PID: 7301 at fs/buffer.c:1145 __brelse+0x67/0xa0
...
Call Trace:
 <TASK>
 invalidate_bh_lru+0x99/0x150
 smp_call_function_many_cond+0xe2a/0x10c0
 ? generic_remap_file_range_prep+0x50/0x50
 ? __brelse+0xa0/0xa0
 ? __mutex_lock+0x21c/0x12d0
 ? smp_call_on_cpu+0x250/0x250
 ? rcu_read_lock_sched_held+0xb/0x60
 ? lock_release+0x587/0x810
 ? __brelse+0xa0/0xa0
 ? generic_remap_file_range_prep+0x50/0x50
 on_each_cpu_cond_mask+0x3c/0x80
 blkdev_flush_mapping+0x13a/0x2f0
 blkdev_put_whole+0xd3/0xf0
 blkdev_put+0x222/0x760
 deactivate_locked_super+0x96/0x160
 deactivate_super+0xda/0x100
 cleanup_mnt+0x222/0x3d0
 task_work_run+0x149/0x240
 ? task_work_cancel+0x30/0x30
 do_exit+0xb29/0x2a40
 ? reacquire_held_locks+0x4a0/0x4a0
 ? do_raw_spin_lock+0x12a/0x2b0
 ? mm_update_next_owner+0x7c0/0x7c0
 ? rwlock_bug.part.0+0x90/0x90
 ? zap_other_threads+0x234/0x2d0
 do_group_exit+0xd0/0x2a0
 __x64_sys_exit_group+0x3a/0x50
 do_syscall_64+0x34/0xb0
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

The cause of the issue is that brelse() is called on both ofibh.sbh and ofibh.ebh by udf_find_entry() when it returns NULL. However, brelse() is called by udf_rename(), too. So, b_count on buffer_head becomes unbalanced.

This patch fixes the issue by not calling brelse() by udf_rename() when udf_find_entry() returns NULL.

---
 fs/udf/namei.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
fs/udf/namei.c +4 -4
index d13ded8e2c30..ef251622da13 100644
@@ -1106,8 +1106,9 @@ static int udf_rename(struct inode *old_dir, struct dentry *old_dentry,
 		return -EINVAL;
 
 	ofi = udf_find_entry(old_dir, &old_dentry->d_name, &ofibh, &ocfi);
-	if (IS_ERR(ofi)) {
-		retval = PTR_ERR(ofi);
+	if (!ofi || IS_ERR(ofi)) {
+		if (IS_ERR(ofi))
+			retval = PTR_ERR(ofi);
 		goto end_rename;
 	}
 
@@ -1116,8 +1117,7 @@ static int udf_rename(struct inode *old_dir, struct dentry *old_dentry,
 
 	brelse(ofibh.sbh);
 	tloc = lelb_to_cpu(ocfi.icb.extLocation);
-	if (!ofi || udf_get_lb_pblock(old_dir->i_sb, &tloc, 0)
-	    != old_inode->i_ino)
+	if (udf_get_lb_pblock(old_dir->i_sb, &tloc, 0) != old_inode->i_ino)
 		goto end_rename;
 
 	nfi = udf_find_entry(new_dir, &new_dentry->d_name, &nfibh, &ncfi);
Signature · 1 line
2.35.1
author maintainer
#2 lore raw

PATCH 3/8 fs: jfs: fix shift-out-of-bounds in dbDiscardAG

In reply to Sasha Levin

From: Hoi Pok Wu <[email protected]>

[ Upstream commit 25e70c6162f207828dd405b432d8f2a98dbf7082 ]

This should be applied to most URSAN bugs found recently by syzbot, by guarding the dbMount. As syzbot feeding rubbish into the bmap descriptor.

---
 fs/jfs/jfs_dmap.c | 5 +++++
 1 file changed, 5 insertions(+)
fs/jfs/jfs_dmap.c +5 -0
index 4f72f7dee78b..f06796cad9aa 100644
@@ -211,6 +211,11 @@ int dbMount(struct inode *ipbmap)
 		goto err_release_metapage;
 	}
 
+	if (((bmp->db_mapsize - 1) >> bmp->db_agl2size) > MAXAG) {
+		err = -EINVAL;
+		goto err_release_metapage;
+	}
+
 	for (i = 0; i < MAXAG; i++)
 		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
 	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
Signature · 1 line
2.35.1
author maintainer
#3 lore raw

PATCH 4/8 ACPICA: Fix error code path in acpi_ds_call_control_method()

In reply to Sasha Levin

From: "Rafael J. Wysocki" <[email protected]>

[ Upstream commit 404ec60438add1afadaffaed34bb5fe4ddcadd40 ]

A use-after-free in acpi_ps_parse_aml() after a failing invocaion of acpi_ds_call_control_method() is reported by KASAN [1] and code inspection reveals that next_walk_state pushed to the thread by acpi_ds_create_walk_state() is freed on errors, but it is not popped from the thread beforehand. Thus acpi_ds_get_current_walk_state() called by acpi_ps_parse_aml() subsequently returns it as the new walk state which is incorrect.

To address this, make acpi_ds_call_control_method() call acpi_ds_pop_walk_state() to pop next_walk_state from the thread before returning an error.

---
 drivers/acpi/acpica/dsmethod.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/acpi/acpica/dsmethod.c +8 -2
index dd4deb678d13..a00516d9538c 100644
@@ -517,7 +517,7 @@ acpi_ds_call_control_method(struct acpi_thread_state *thread,
 	info = ACPI_ALLOCATE_ZEROED(sizeof(struct acpi_evaluate_info));
 	if (!info) {
 		status = AE_NO_MEMORY;
-		goto cleanup;
+		goto pop_walk_state;
 	}
 
 	info->parameters = &this_walk_state->operands[0];
@@ -529,7 +529,7 @@ acpi_ds_call_control_method(struct acpi_thread_state *thread,
 
 	ACPI_FREE(info);
 	if (ACPI_FAILURE(status)) {
-		goto cleanup;
+		goto pop_walk_state;
 	}
 
 	/*
@@ -561,6 +561,12 @@ acpi_ds_call_control_method(struct acpi_thread_state *thread,
 
 	return_ACPI_STATUS(status);
 
+pop_walk_state:
+
+	/* On error, pop the walk state to be deleted from thread */
+
+	acpi_ds_pop_walk_state(thread);
+
 cleanup:
 
 	/* On error, we must terminate the method properly */
Signature · 1 line
2.35.1
author maintainer
#4 lore raw

PATCH 5/8 proc/vmcore: fix potential memory leak in vmcore_init()

In reply to Sasha Levin

From: Jianglei Nie <[email protected]>

[ Upstream commit 12b9d301ff73122aebd78548fa4c04ca69ed78fe ]

Patch series "Some minor cleanup patches resent".

The first three patches trivial clean up patches.

And for the patch "kexec: replace crash_mem_range with range", I got a ibm-p9wr ppc64le system to test, it works well.

This patch (of 4):

elfcorehdr_alloc() allocates a memory chunk for elfcorehdr_addr with kzalloc(). If is_vmcore_usable() returns false, elfcorehdr_addr is a predefined value. If parse_crash_elf_headers() gets some error and returns a negetive value, the elfcorehdr_addr should be released with elfcorehdr_free().

Fix it by calling elfcorehdr_free() when parse_crash_elf_headers() fails.

---
 fs/proc/vmcore.c | 1 +
 1 file changed, 1 insertion(+)
fs/proc/vmcore.c +1 -0
index f5dfedc01552..7f53c4d56e23 100644
@@ -1533,6 +1533,7 @@ static int __init vmcore_init(void)
 		return rc;
 	rc = parse_crash_elf_headers();
 	if (rc) {
+		elfcorehdr_free(elfcorehdr_addr);
 		pr_warn("Kdump: vmcore not initialized\n");
 		return rc;
 	}
Signature · 7 lines
2.35.1


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
author maintainer
#5 lore raw

PATCH 6/8 nilfs2: fix shift-out-of-bounds/overflow in nilfs_sb2_bad_offset()

In reply to Sasha Levin

From: Ryusuke Konishi <[email protected]>

[ Upstream commit 610a2a3d7d8be3537458a378ec69396a76c385b6 ]

Patch series "nilfs2: fix UBSAN shift-out-of-bounds warnings on mount time".

The first patch fixes a bug reported by syzbot, and the second one fixes the remaining bug of the same kind. Although they are triggered by the same super block data anomaly, I divided it into the above two because the details of the issues and how to fix it are different.

Both are required to eliminate the shift-out-of-bounds issues at mount time.

This patch (of 2):

If the block size exponent information written in an on-disk superblock is corrupted, nilfs_sb2_bad_offset helper function can trigger shift-out-of-bounds warning followed by a kernel panic (if panic_on_warn is set):

shift exponent 38983 is too large for 64-bit type 'unsigned long long'

 Call Trace:
  <TASK>
  __dump_stack lib/dump_stack.c:88 [inline]
  dump_stack_lvl+0x1b1/0x28e lib/dump_stack.c:106
  ubsan_epilogue lib/ubsan.c:151 [inline]
  __ubsan_handle_shift_out_of_bounds+0x33d/0x3b0 lib/ubsan.c:322
  nilfs_sb2_bad_offset fs/nilfs2/the_nilfs.c:449 [inline]
  nilfs_load_super_block+0xdf5/0xe00 fs/nilfs2/the_nilfs.c:523
  init_nilfs+0xb7/0x7d0 fs/nilfs2/the_nilfs.c:577
  nilfs_fill_super+0xb1/0x5d0 fs/nilfs2/super.c:1047
  nilfs_mount+0x613/0x9b0 fs/nilfs2/super.c:1317
  ...

In addition, since nilfs_sb2_bad_offset() performs multiplication without considering the upper bound, the computation may overflow if the disk layout parameters are not normal.

This fixes these issues by inserting preliminary sanity checks for those parameters and by converting the comparison from one involving multiplication and left bit-shifting to one using division and right bit-shifting.

---
 fs/nilfs2/the_nilfs.c | 31 +++++++++++++++++++++++++++----
 1 file changed, 27 insertions(+), 4 deletions(-)
fs/nilfs2/the_nilfs.c +27 -4
index fb61c33c6004..74ef3d313686 100644
@@ -13,6 +13,7 @@
 #include <linux/blkdev.h>
 #include <linux/backing-dev.h>
 #include <linux/random.h>
+#include <linux/log2.h>
 #include <linux/crc32.h>
 #include "nilfs.h"
 #include "segment.h"
@@ -448,11 +449,33 @@ static int nilfs_valid_sb(struct nilfs_super_block *sbp)
 	return crc == le32_to_cpu(sbp->s_sum);
 }
 
-static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
+/**
+ * nilfs_sb2_bad_offset - check the location of the second superblock
+ * @sbp: superblock raw data buffer
+ * @offset: byte offset of second superblock calculated from device size
+ *
+ * nilfs_sb2_bad_offset() checks if the position on the second
+ * superblock is valid or not based on the filesystem parameters
+ * stored in @sbp.  If @offset points to a location within the segment
+ * area, or if the parameters themselves are not normal, it is
+ * determined to be invalid.
+ *
+ * Return Value: true if invalid, false if valid.
+ */
+static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
 {
-	return offset < ((le64_to_cpu(sbp->s_nsegments) *
-			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
-			 (le32_to_cpu(sbp->s_log_block_size) + 10));
+	unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
+	u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
+	u64 nsegments = le64_to_cpu(sbp->s_nsegments);
+	u64 index;
+
+	if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
+	    shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
+		return true;
+
+	index = offset >> (shift_bits + BLOCK_SIZE_BITS);
+	do_div(index, blocks_per_segment);
+	return index < nsegments;
 }
 
 static void nilfs_release_super_block(struct the_nilfs *nilfs)
Signature · 1 line
2.35.1
author maintainer
#6 lore raw

PATCH 7/8 acct: fix potential integer overflow in encode_comp_t()

In reply to Sasha Levin

From: Zheng Yejian <[email protected]>

[ Upstream commit c5f31c655bcc01b6da53b836ac951c1556245305 ]

The integer overflow is descripted with following codes:

Quoted 3 lines

317 static comp_t encode_comp_t(u64 value)
318 {
319 int exp, rnd;

    ......
Quoted 4 lines
341         exp <<= MANTSIZE;
342         exp += value;
343         return exp;

344 }

Currently comp_t is defined as type of '__u16', but the variable 'exp' is type of 'int', so overflow would happen when variable 'exp' in line 343 is greater than 65535.

---
 kernel/acct.c | 2 ++
 1 file changed, 2 insertions(+)
kernel/acct.c +2 -0
index 81f9831a7859..6d98aed403ba 100644
@@ -331,6 +331,8 @@ static comp_t encode_comp_t(unsigned long value)
 		exp++;
 	}
 
+	if (exp > (((comp_t) ~0U) >> MANTSIZE))
+		return (comp_t) ~0U;
 	/*
 	 * Clean it up and polish it off.
 	 */
Signature · 1 line
2.35.1
author maintainer
#7 lore raw

PATCH 8/8 hfs: fix OOB Read in __hfs_brec_find

In reply to Sasha Levin

From: ZhangPeng <[email protected]>

[ Upstream commit 8d824e69d9f3fa3121b2dda25053bae71e2460d2 ]

Syzbot reported a OOB read bug:

==================================================================

BUG: KASAN: slab-out-of-bounds in hfs_strcmp+0x117/0x190
fs/hfs/string.c:84
Read of size 1 at addr ffff88807eb62c4e by task kworker/u4:1/11
CPU: 1 PID: 11 Comm: kworker/u4:1 Not tainted
6.1.0-rc6-syzkaller-00308-g644e9524388a #0
Workqueue: writeback wb_workfn (flush-7:0)
Call Trace:
 <TASK>
 __dump_stack lib/dump_stack.c:88 [inline]
 dump_stack_lvl+0x1b1/0x28e lib/dump_stack.c:106
 print_address_description+0x74/0x340 mm/kasan/report.c:284
 print_report+0x107/0x1f0 mm/kasan/report.c:395
 kasan_report+0xcd/0x100 mm/kasan/report.c:495
 hfs_strcmp+0x117/0x190 fs/hfs/string.c:84
 __hfs_brec_find+0x213/0x5c0 fs/hfs/bfind.c:75
 hfs_brec_find+0x276/0x520 fs/hfs/bfind.c:138
 hfs_write_inode+0x34c/0xb40 fs/hfs/inode.c:462
 write_inode fs/fs-writeback.c:1440 [inline]

If the input inode of hfs_write_inode() is incorrect:
struct inode
struct hfs_inode_info
struct hfs_cat_key
struct hfs_name
u8 len # len is greater than HFS_NAMELEN(31) which is the
maximum length of an HFS filename

OOB read occurred:
hfs_write_inode()
hfs_brec_find()
__hfs_brec_find()
hfs_cat_keycmp()
hfs_strcmp() # OOB read occurred due to len is too large

Fix this by adding a Check on len in hfs_write_inode() before calling hfs_brec_find().

---
 fs/hfs/inode.c | 2 ++
 1 file changed, 2 insertions(+)
fs/hfs/inode.c +2 -0
index da243c84e93b..9c5badc9db00 100644
@@ -453,6 +453,8 @@ int hfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 		/* panic? */
 		return -EIO;
 
+	if (HFS_I(main_inode)->cat_key.CName.len > HFS_NAMELEN)
+		return -EIO;
 	fd.search_key->cat = HFS_I(main_inode)->cat_key;
 	if (hfs_brec_find(&fd))
 		/* panic? */
Signature · 1 line
2.35.1