[BUG] reproducable ubifs reboot assert and corruption
Richard Weinberger
richard.weinberger at gmail.com
Wed Jan 29 14:56:51 EST 2014
On Wed, Jan 29, 2014 at 8:13 PM, Andrew Ruder <andy at aeruder.net> wrote:
> On Wed, Jan 29, 2014 at 04:56:04PM +0100, Richard Weinberger wrote:
>> Does the issue also happen if you disable preemption?
>> i.e. CONFIG_PREEMPT_NONE=y
>
> CONFIG_PREEMPT_NONE=y still breaks. I suspect that sync_filesystem
> has some blocking behavior that allows other processes to schedule.
Okay, I have to test this on real hardware.
May take a few days.
BTW: s/not.at/nod.at/g ;)
> Another log is attached and the patch I used to create this log is
> below. I think the most interesting part of this patch is the last hunk
> that modifies ubifs_remount_ro. This clearly shows that after the mount
> has been marked as read-only we have dirty inodes waiting for the
> writeback to come in and hit all the asserts.
>
> Here's some of the important parts of the log:
>
> 170 pre sync_filesystem
> # Notice that while we were running sync_filesystem
> # a inode (0xd75ab658) snuck in with a sys_rename
> 204 inode: d75ab658
> 205 ------------[ cut here ]------------
> 206 WARNING: CPU: 0 PID: 625 at fs/fs-writeback.c:1213 __mark_inode_dirty+0x2a4/0x2f4()
> 207 Modules linked in:
> 208 CPU: 0 PID: 625 Comm: fsstress Tainted: G W 3.12.0-00041-g7f12d39-dirty #250
> 209 [<c0014f54>] (unwind_backtrace+0x0/0x128) from [<c001235c>] (show_stack+0x20/0x24)
> 210 [<c001235c>] (show_stack+0x20/0x24) from [<c0344574>] (dump_stack+0x20/0x28)
> 211 [<c0344574>] (dump_stack+0x20/0x28) from [<c00206d4>] (warn_slowpath_common+0x78/0x98)
> 212 [<c00206d4>] (warn_slowpath_common+0x78/0x98) from [<c00207b0>] (warn_slowpath_null+0x2c/0x34)
> 213 [<c00207b0>] (warn_slowpath_null+0x2c/0x34) from [<c00f0e8c>] (__mark_inode_dirty+0x2a4/0x2f4)
> 214 [<c00f0e8c>] (__mark_inode_dirty+0x2a4/0x2f4) from [<c0136428>] (ubifs_rename+0x4a4/0x600)
> 215 [<c0136428>] (ubifs_rename+0x4a4/0x600) from [<c00d9f44>] (vfs_rename+0x280/0x3f4)
> 216 [<c00d9f44>] (vfs_rename+0x280/0x3f4) from [<c00dabe4>] (SyS_renameat+0x18c/0x218)
> 217 [<c00dabe4>] (SyS_renameat+0x18c/0x218) from [<c00dac9c>] (SyS_rename+0x2c/0x30)
> 218 [<c00dac9c>] (SyS_rename+0x2c/0x30) from [<c000e820>] (ret_fast_syscall+0x0/0x2c)
> 219 ---[ end trace 35ebec8569a53526 ]---
> 754 post sync_filesystem
> 755 pre prepare_remount
> 756 post prepare_remount
> 757 prepare_remount success
> 758 UBIFS: background thread "ubifs_bgt0_0" stops
> 759 we are now a read-only mount
> 760 bdi.work_list = d7ac4110, .next = d7ac4110
> # WE HAVE DIRTY I/O (notice the .next != &b_dirty)
> 761 bdi.wb.b_dirty = d7ac40d8, .next = d75accd0
> 762 bdi.wb.b_io = d7ac40e0, .next = d7ac40e0
> 763 bdi.wb.b_more_io = d7ac40e8, .next = d7ac40e8
> 764 do_remount_sb success
> # And now our friend (inode 0xd75ab658) comes in with a writeback after
> # we are read-only triggering the assert
> 779 inode: d75ab658
> 780 UBIFS assert failed in reserve_space at 125 (pid 11)
> 781 CPU: 0 PID: 11 Comm: kworker/u2:1 Tainted: G W 3.12.0-00041-g7f12d39-dirty #250
> 782 Workqueue: writeback bdi_writeback_workfn (flush-ubifs_0_0)
> 783 [<c0014f54>] (unwind_backtrace+0x0/0x128) from [<c001235c>] (show_stack+0x20/0x24)
> 784 [<c001235c>] (show_stack+0x20/0x24) from [<c0344574>] (dump_stack+0x20/0x28)
> 785 [<c0344574>] (dump_stack+0x20/0x28) from [<c012f90c>] (make_reservation+0x8c/0x560)
> 786 [<c012f90c>] (make_reservation+0x8c/0x560) from [<c0130c60>] (ubifs_jnl_write_inode+0xbc/0x214)
> 787 [<c0130c60>] (ubifs_jnl_write_inode+0xbc/0x214) from [<c0137ce0>] (ubifs_write_inode+0xec/0x17c)
> 788 [<c0137ce0>] (ubifs_write_inode+0xec/0x17c) from [<c00f0a00>] (__writeback_single_inode+0x1fc/0x308)
> 789 [<c00f0a00>] (__writeback_single_inode+0x1fc/0x308) from [<c00f1780>] (writeback_sb_inodes+0x1f8/0x3c4)
> 790 [<c00f1780>] (writeback_sb_inodes+0x1f8/0x3c4) from [<c00f19cc>] (__writeback_inodes_wb+0x80/0xc0)
> 791 [<c00f19cc>] (__writeback_inodes_wb+0x80/0xc0) from [<c00f1ba4>] (wb_writeback+0x198/0x310)
> 792 [<c00f1ba4>] (wb_writeback+0x198/0x310) from [<c00f2328>] (bdi_writeback_workfn+0x15c/0x454)
> 793 [<c00f2328>] (bdi_writeback_workfn+0x15c/0x454) from [<c0038348>] (process_one_work+0x280/0x420)
> 794 [<c0038348>] (process_one_work+0x280/0x420) from [<c00389e4>] (worker_thread+0x240/0x380)
> 795 [<c00389e4>] (worker_thread+0x240/0x380) from [<c003dff8>] (kthread+0xbc/0xc8)
> 796 [<c003dff8>] (kthread+0xbc/0xc8) from [<c000e8b0>] (ret_from_fork+0x14/0x20)
>
> - Andy
>
>
>
>
> --- patch ---
> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
> index 9f4935b..48e4272 100644
> --- a/fs/fs-writeback.c
> +++ b/fs/fs-writeback.c
> @@ -93,6 +93,10 @@ static void bdi_queue_work(struct backing_dev_info *bdi,
> {
> trace_writeback_queue(bdi, work);
>
> + do {
> + extern bool ubifs_enable_debug;
> + WARN_ON(ubifs_enable_debug);
> + } while (0);
> spin_lock_bh(&bdi->wb_lock);
> list_add_tail(&work->list, &bdi->work_list);
> spin_unlock_bh(&bdi->wb_lock);
> @@ -194,6 +198,11 @@ static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
> if (time_before(inode->dirtied_when, tail->dirtied_when))
> inode->dirtied_when = jiffies;
> }
> + do {
> + extern bool ubifs_enable_debug;
> + if (ubifs_enable_debug) pr_info("inode: %p\n", inode);
> + WARN_ON(ubifs_enable_debug);
> + } while (0);
> list_move(&inode->i_wb_list, &wb->b_dirty);
> }
>
> @@ -204,6 +213,11 @@ static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
> {
> assert_spin_locked(&wb->list_lock);
> list_move(&inode->i_wb_list, &wb->b_more_io);
> + do {
> + extern bool ubifs_enable_debug;
> + if (ubifs_enable_debug) pr_info("inode: %p\n", inode);
> + WARN_ON(ubifs_enable_debug);
> + } while (0);
> }
>
> static void inode_sync_complete(struct inode *inode)
> @@ -437,6 +451,8 @@ __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
> unsigned dirty;
> int ret;
>
> + extern bool ubifs_enable_debug;
> + if (ubifs_enable_debug) pr_info("inode: %p\n", inode);
> WARN_ON(!(inode->i_state & I_SYNC));
>
> trace_writeback_single_inode_start(inode, wbc, nr_to_write);
> @@ -1191,6 +1207,11 @@ void __mark_inode_dirty(struct inode *inode, int flags)
>
> inode->dirtied_when = jiffies;
> list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
> + do {
> + extern bool ubifs_enable_debug;
> + if (ubifs_enable_debug) pr_info("inode: %p\n", inode);
> + WARN_ON(ubifs_enable_debug);
> + } while (0);
> spin_unlock(&bdi->wb.list_lock);
>
> if (wakeup_bdi)
> diff --git a/fs/super.c b/fs/super.c
> index 0225c20..29afc68 100644
> --- a/fs/super.c
> +++ b/fs/super.c
> @@ -737,6 +737,12 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
> int retval;
> int remount_ro;
>
> + extern bool ubifs_enable_debug;
> + if (flags & MS_RDONLY) {
> + ubifs_enable_debug = true;
> + }
> + WARN_ON(ubifs_enable_debug);
> +
> if (sb->s_writers.frozen != SB_UNFROZEN)
> return -EBUSY;
>
> @@ -747,8 +753,11 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
>
> if (flags & MS_RDONLY)
> acct_auto_close(sb);
> + pr_info("pre shrink_dcache_sb\n");
> shrink_dcache_sb(sb);
> + pr_info("pre sync_filesystem\n");
> sync_filesystem(sb);
> + pr_info("post sync_filesystem\n");
>
> remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
>
> @@ -758,9 +767,12 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
> if (force) {
> mark_files_ro(sb);
> } else {
> + pr_info("pre prepare_remount\n");
> retval = sb_prepare_remount_readonly(sb);
> + pr_info("post prepare_remount\n");
> if (retval)
> return retval;
> + pr_info("prepare_remount success\n");
> }
> }
>
> @@ -789,6 +801,7 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
> */
> if (remount_ro && sb->s_bdev)
> invalidate_bdev(sb->s_bdev);
> + pr_info("do_remount_sb success\n");
> return 0;
>
> cancel_readonly:
> diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c
> index 123c79b..c9f2d5d 100644
> --- a/fs/ubifs/file.c
> +++ b/fs/ubifs/file.c
> @@ -902,6 +902,8 @@ static int do_writepage(struct page *page, int len)
> struct inode *inode = page->mapping->host;
> struct ubifs_info *c = inode->i_sb->s_fs_info;
>
> + WARN_ON(c->ro_mount);
> +
> #ifdef UBIFS_DEBUG
> spin_lock(&ui->ui_lock);
> ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
> diff --git a/fs/ubifs/super.c b/fs/ubifs/super.c
> index 3e4aa72..8cbb731 100644
> --- a/fs/ubifs/super.c
> +++ b/fs/ubifs/super.c
> @@ -38,6 +38,8 @@
> #include <linux/writeback.h>
> #include "ubifs.h"
>
> +bool ubifs_enable_debug;
> +
> /*
> * Maximum amount of memory we may 'kmalloc()' without worrying that we are
> * allocating too much.
> @@ -1756,6 +1758,11 @@ static void ubifs_remount_ro(struct ubifs_info *c)
> err = dbg_check_space_info(c);
> if (err)
> ubifs_ro_mode(c, err);
> + pr_info("we are now a read-only mount\n");
> + pr_info("bdi.work_list = %p, .next = %p\n", &c->bdi.work_list, c->bdi.work_list.next);
> + pr_info("bdi.wb.b_dirty = %p, .next = %p\n", &c->bdi.wb.b_dirty, c->bdi.wb.b_dirty.next);
> + pr_info("bdi.wb.b_io = %p, .next = %p\n", &c->bdi.wb.b_io, c->bdi.wb.b_io.next);
> + pr_info("bdi.wb.b_more_io = %p, .next = %p\n", &c->bdi.wb.b_more_io, c->bdi.wb.b_more_io.next);
> mutex_unlock(&c->umount_mutex);
> }
>
>
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
>
--
Thanks,
//richard
More information about the linux-mtd
mailing list