| Message ID | 20231018114221.441526-4-libaokun1@huawei.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | ext4: fix WARN_ON in alloc_flex_gd() | expand |
On Wed 18-10-23 19:42:20, Baokun Li wrote: > When we online resize an ext4 filesystem with a oversized flexbg_size, > > mkfs.ext4 -F -G 67108864 $dev -b 4096 100M > mount $dev $dir > resize2fs $dev 16G > > the following WARN_ON is triggered: > ================================================================== > WARNING: CPU: 0 PID: 427 at mm/page_alloc.c:4402 __alloc_pages+0x411/0x550 > Modules linked in: sg(E) > CPU: 0 PID: 427 Comm: resize2fs Tainted: G E 6.6.0-rc5+ #314 > RIP: 0010:__alloc_pages+0x411/0x550 > Call Trace: > <TASK> > __kmalloc_large_node+0xa2/0x200 > __kmalloc+0x16e/0x290 > ext4_resize_fs+0x481/0xd80 > __ext4_ioctl+0x1616/0x1d90 > ext4_ioctl+0x12/0x20 > __x64_sys_ioctl+0xf0/0x150 > do_syscall_64+0x3b/0x90 > ================================================================== > > This is because flexbg_size is too large and the size of the new_group_data > array to be allocated exceeds MAX_ORDER. Currently, the minimum value of > MAX_ORDER is 8, the minimum value of PAGE_SIZE is 4096, the corresponding > maximum number of groups that can be allocated is: > > (PAGE_SIZE << MAX_ORDER) / sizeof(struct ext4_new_group_data) ≈ 21845 > > And the value that is down-aligned to the power of 2 is 16384. Therefore, > this value is defined as MAX_RESIZE_BG, and the number of groups added > each time does not exceed this value during resizing, and is added multiple > times to complete the online resizing. The difference is that the metadata > in a flex_bg may be more dispersed. > > Signed-off-by: Baokun Li <libaokun1@huawei.com> Looks good. Feel free to add: Reviewed-by: Jan Kara <jack@suse.cz> Honza > --- > fs/ext4/resize.c | 25 +++++++++++++++++-------- > 1 file changed, 17 insertions(+), 8 deletions(-) > > diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c > index 0a57b199883c..e168a9f59600 100644 > --- a/fs/ext4/resize.c > +++ b/fs/ext4/resize.c > @@ -218,10 +218,17 @@ struct ext4_new_flex_group_data { > in the flex group */ > __u16 *bg_flags; /* block group flags of groups > in @groups */ > + ext4_group_t resize_bg; /* number of allocated > + new_group_data */ > ext4_group_t count; /* number of groups in @groups > */ > }; > > +/* > + * Avoiding memory allocation failures due to too many groups added each time. > + */ > +#define MAX_RESIZE_BG 16384 > + > /* > * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of > * @flexbg_size. > @@ -236,14 +243,18 @@ static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size) > if (flex_gd == NULL) > goto out3; > > - flex_gd->count = flexbg_size; > - flex_gd->groups = kmalloc_array(flexbg_size, > + if (unlikely(flexbg_size > MAX_RESIZE_BG)) > + flex_gd->resize_bg = MAX_RESIZE_BG; > + else > + flex_gd->resize_bg = flexbg_size; > + > + flex_gd->groups = kmalloc_array(flex_gd->resize_bg, > sizeof(struct ext4_new_group_data), > GFP_NOFS); > if (flex_gd->groups == NULL) > goto out2; > > - flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16), > + flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16), > GFP_NOFS); > if (flex_gd->bg_flags == NULL) > goto out1; > @@ -1602,8 +1613,7 @@ static int ext4_flex_group_add(struct super_block *sb, > > static int ext4_setup_next_flex_gd(struct super_block *sb, > struct ext4_new_flex_group_data *flex_gd, > - ext4_fsblk_t n_blocks_count, > - unsigned int flexbg_size) > + ext4_fsblk_t n_blocks_count) > { > struct ext4_sb_info *sbi = EXT4_SB(sb); > struct ext4_super_block *es = sbi->s_es; > @@ -1627,7 +1637,7 @@ static int ext4_setup_next_flex_gd(struct super_block *sb, > BUG_ON(last); > ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); > > - last_group = group | (flexbg_size - 1); > + last_group = group | (flex_gd->resize_bg - 1); > if (last_group > n_group) > last_group = n_group; > > @@ -2130,8 +2140,7 @@ int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) > /* Add flex groups. Note that a regular group is a > * flex group with 1 group. > */ > - while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count, > - flexbg_size)) { > + while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) { > if (time_is_before_jiffies(last_update_time + HZ * 10)) { > if (last_update_time) > ext4_msg(sb, KERN_INFO, > -- > 2.31.1 >
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c index 0a57b199883c..e168a9f59600 100644 --- a/fs/ext4/resize.c +++ b/fs/ext4/resize.c @@ -218,10 +218,17 @@ struct ext4_new_flex_group_data { in the flex group */ __u16 *bg_flags; /* block group flags of groups in @groups */ + ext4_group_t resize_bg; /* number of allocated + new_group_data */ ext4_group_t count; /* number of groups in @groups */ }; +/* + * Avoiding memory allocation failures due to too many groups added each time. + */ +#define MAX_RESIZE_BG 16384 + /* * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of * @flexbg_size. @@ -236,14 +243,18 @@ static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size) if (flex_gd == NULL) goto out3; - flex_gd->count = flexbg_size; - flex_gd->groups = kmalloc_array(flexbg_size, + if (unlikely(flexbg_size > MAX_RESIZE_BG)) + flex_gd->resize_bg = MAX_RESIZE_BG; + else + flex_gd->resize_bg = flexbg_size; + + flex_gd->groups = kmalloc_array(flex_gd->resize_bg, sizeof(struct ext4_new_group_data), GFP_NOFS); if (flex_gd->groups == NULL) goto out2; - flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16), + flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16), GFP_NOFS); if (flex_gd->bg_flags == NULL) goto out1; @@ -1602,8 +1613,7 @@ static int ext4_flex_group_add(struct super_block *sb, static int ext4_setup_next_flex_gd(struct super_block *sb, struct ext4_new_flex_group_data *flex_gd, - ext4_fsblk_t n_blocks_count, - unsigned int flexbg_size) + ext4_fsblk_t n_blocks_count) { struct ext4_sb_info *sbi = EXT4_SB(sb); struct ext4_super_block *es = sbi->s_es; @@ -1627,7 +1637,7 @@ static int ext4_setup_next_flex_gd(struct super_block *sb, BUG_ON(last); ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last); - last_group = group | (flexbg_size - 1); + last_group = group | (flex_gd->resize_bg - 1); if (last_group > n_group) last_group = n_group; @@ -2130,8 +2140,7 @@ int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) /* Add flex groups. Note that a regular group is a * flex group with 1 group. */ - while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count, - flexbg_size)) { + while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) { if (time_is_before_jiffies(last_update_time + HZ * 10)) { if (last_update_time) ext4_msg(sb, KERN_INFO,
When we online resize an ext4 filesystem with a oversized flexbg_size, mkfs.ext4 -F -G 67108864 $dev -b 4096 100M mount $dev $dir resize2fs $dev 16G the following WARN_ON is triggered: ================================================================== WARNING: CPU: 0 PID: 427 at mm/page_alloc.c:4402 __alloc_pages+0x411/0x550 Modules linked in: sg(E) CPU: 0 PID: 427 Comm: resize2fs Tainted: G E 6.6.0-rc5+ #314 RIP: 0010:__alloc_pages+0x411/0x550 Call Trace: <TASK> __kmalloc_large_node+0xa2/0x200 __kmalloc+0x16e/0x290 ext4_resize_fs+0x481/0xd80 __ext4_ioctl+0x1616/0x1d90 ext4_ioctl+0x12/0x20 __x64_sys_ioctl+0xf0/0x150 do_syscall_64+0x3b/0x90 ================================================================== This is because flexbg_size is too large and the size of the new_group_data array to be allocated exceeds MAX_ORDER. Currently, the minimum value of MAX_ORDER is 8, the minimum value of PAGE_SIZE is 4096, the corresponding maximum number of groups that can be allocated is: (PAGE_SIZE << MAX_ORDER) / sizeof(struct ext4_new_group_data) ≈ 21845 And the value that is down-aligned to the power of 2 is 16384. Therefore, this value is defined as MAX_RESIZE_BG, and the number of groups added each time does not exceed this value during resizing, and is added multiple times to complete the online resizing. The difference is that the metadata in a flex_bg may be more dispersed. Signed-off-by: Baokun Li <libaokun1@huawei.com> --- fs/ext4/resize.c | 25 +++++++++++++++++-------- 1 file changed, 17 insertions(+), 8 deletions(-)