commit d610020f030bec819f42de327c2bd5437d2766b3 upstream.
During wear-leveing work, the source PEB will be moved into scrub list
when source LEB cannot be locked in ubi_eba_copy_leb(), which is wrong
for non-scrub type source PEB. The problem could bring extra and
ineffective wear-leveing jobs, which makes more or less negative effects
for the life time of flash. Specifically, the process is divided 2 steps:
1. wear_leveling_worker // generate false scrub type PEB
ubi_eba_copy_leb // MOVE_RETRY is returned
leb_write_trylock // trylock failed
scrubbing = 1;
e1 is put into ubi->scrub
2. wear_leveling_worker // schedule false scrub type PEB for wl
scrubbing = 1
e1 = rb_entry(rb_first(&ubi->scrub))
The problem can be reproduced easily by running fsstress on a small
UBIFS partition(<64M, simulated by nandsim) for 5~10mins
(CONFIG_MTD_UBI_FASTMAP=y,CONFIG_MTD_UBI_WL_THRESHOLD=50). Following
message is shown:
ubi0: scrubbed PEB 66 (LEB 0:10), data moved to PEB 165
Since scrub type source PEB has set variable scrubbing as '1', and
variable scrubbing is checked before variable keep, so the problem can
be fixed by setting keep variable as 1 directly if the source LEB cannot
be locked.
Fixes: e801e128b2 ("UBI: fix missing scrub when there is a bit-flip")
CC: stable@vger.kernel.org
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit f773f0a331d6c41733b17bebbc1b6cae12e016f5 ]
During the processing of the bgt, if the sync_erase() return -EBUSY
or some other error code in __erase_worker(),schedule_erase() called
again lead to the down_read(ubi->work_sem) hold twice and may get
block by down_write(ubi->work_sem) in ubi_update_fastmap(),
which cause deadlock.
ubi bgt other task
do_work
down_read(&ubi->work_sem) ubi_update_fastmap
erase_worker # Blocked by down_read
__erase_worker down_write(&ubi->work_sem)
schedule_erase
schedule_ubi_work
down_read(&ubi->work_sem)
Fix this by changing input parameter @nested of the schedule_erase() to
'true' to avoid recursively acquiring the down_read(&ubi->work_sem).
Also, fix the incorrect comment about @nested parameter of the
schedule_erase() because when down_write(ubi->work_sem) is held, the
@nested is also need be true.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=217093
Fixes: 2e8f08deab ("ubi: Fix races around ubi_refill_pools()")
Signed-off-by: ZhaoLong Wang <wangzhaolong1@huawei.com>
Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a240bc5c43130c6aa50831d7caaa02a1d84e1bce ]
Wear-leveling entry could be freed in error path, which may be accessed
again in eraseblk_count_seq_show(), for example:
__erase_worker eraseblk_count_seq_show
wl = ubi->lookuptbl[*block_number]
if (wl)
wl_entry_destroy
ubi->lookuptbl[e->pnum] = NULL
kmem_cache_free(ubi_wl_entry_slab, e)
erase_count = wl->ec // UAF!
Wear-leveling entry updating/accessing in ubi->lookuptbl should be
protected by ubi->wl_lock, fix it by adding ubi->wl_lock to serialize
wl entry accessing between wl_entry_destroy() and
eraseblk_count_seq_show().
Fetch a reproducer in [Link].
Link: https://bugzilla.kernel.org/show_bug.cgi?id=216305
Fixes: 7bccd12d27 ("ubi: Add debugfs file for tracking PEB state")
Fixes: 801c135ce7 ("UBI: Unsorted Block Images")
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit d005f8c6588efcfbe88099b6edafc6f58c84a9c1 upstream.
A detach hung is possible when a race occurs between the detach process
and the ubi background thread. The following sequences outline the race:
ubi thread: if (list_empty(&ubi->works)...
ubi detach: set_bit(KTHREAD_SHOULD_STOP, &kthread->flags)
=> by kthread_stop()
wake_up_process()
=> ubi thread is still running, so 0 is returned
ubi thread: set_current_state(TASK_INTERRUPTIBLE)
schedule()
=> ubi thread will never be scheduled again
ubi detach: wait_for_completion()
=> hung task!
To fix that, we need to check kthread_should_stop() after we set the
task state, so the ubi thread will either see the stop bit and exit or
the task state is reset to runnable such that it isn't scheduled out
indefinitely.
Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
Cc: <stable@vger.kernel.org>
Fixes: 801c135ce7 ("UBI: Unsorted Block Images")
Reported-by: syzbot+853639d0cb16c31c7a14@syzkaller.appspotmail.com
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit f9c34bb529975fe9f85b870a80c53a83a3c5a182 ]
When a new fastmap is about to be written UBI must make sure it has a
free block for a fastmap anchor available. For this ubi_update_fastmap()
calls ubi_ensure_anchor_pebs(). This stopped working with 2e8f08deab
("ubi: Fix races around ubi_refill_pools()"), with this commit the wear
leveling code is blocked and can no longer produce free PEBs. UBI then
more often than not falls back to write the new fastmap anchor to the
same block it was already on which means the same erase block gets
erased during each fastmap write and wears out quite fast.
As the locking prevents us from producing the anchor PEB when we
actually need it, this patch changes the strategy for creating the
anchor PEB. We no longer create it on demand right before we want to
write a fastmap, but instead we create an anchor PEB right after we have
written a fastmap. This gives us enough time to produce a new anchor PEB
before it is needed. To make sure we have an anchor PEB for the very
first fastmap write we call ubi_ensure_anchor_pebs() during
initialisation as well.
Fixes: 2e8f08deab ("ubi: Fix races around ubi_refill_pools()")
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Ben Hutchings pointed out that 29b7a6fa1e ("ubi: fastmap: Don't flush
fastmap work on detach") does not really fix the problem, it just
reduces the risk to hit the race window where fastmap work races against
free()'ing ubi->volumes[].
The correct approach is making sure that no more fastmap work is in
progress before we free ubi data structures.
So we cancel fastmap work right after the ubi background thread is
stopped.
By setting ubi->thread_enabled to zero we make sure that no further work
tries to wake the thread.
Fixes: 29b7a6fa1e ("ubi: fastmap: Don't flush fastmap work on detach")
Fixes: 74cdaf2400 ("UBI: Fastmap: Fix memory leaks while closing the WL sub-system")
Cc: stable@vger.kernel.org
Cc: Ben Hutchings <ben.hutchings@codethink.co.uk>
Cc: Martin Townsend <mtownsend1973@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Fix to return error code -ENOMEM from the kmem_cache_alloc() error
handling case instead of 0, as done elsewhere in this function.
Fixes: f78e5623f4 ("ubi: fastmap: Erase outdated anchor PEBs during
attach")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Fix misspelling of 'available' in function name.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
The fastmap update code might erase the current fastmap anchor PEB
in case it doesn't find any new free PEB. When a power cut happens
in this situation we must not have any outdated fastmap anchor PEB
on the device, because that would be used to attach during next
boot.
The easiest way to make that sure is to erase all outdated fastmap
anchor PEBs synchronously during attach.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Fixes: dbb7d2a88d ("UBI: Add fastmap core")
Cc: <stable@vger.kernel.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
When writing a new Fastmap the first thing that happens
is refilling the pools in memory.
At this stage it is possible that new PEBs from the new pools
get already claimed and written with data.
If this happens before the new Fastmap data structure hits the
flash and we face power cut the freshly written PEB will not
scanned and unnoticed.
Solve the issue by locking the pools until Fastmap is written.
Cc: <stable@vger.kernel.org>
Fixes: dbb7d2a88d ("UBI: Add fastmap core")
Signed-off-by: Richard Weinberger <richard@nod.at>
When Fastmap is used we can face here an -EBADMSG
since Fastmap cannot know about unmaps.
If the erasure was interrupted the PEB may show ECC
errors and UBI would go to ro-mode as it assumes
that the PEB was check during attach time, which is
not the case with Fastmap.
Cc: <stable@vger.kernel.org>
Fixes: dbb7d2a88d ("UBI: Add fastmap core")
Signed-off-by: Richard Weinberger <richard@nod.at>
Currently, all VID headers are allocated and freed using the
ubi_zalloc_vid_hdr() and ubi_free_vid_hdr() function. These functions
make sure to align allocation on ubi->vid_hdr_alsize and adjust the
vid_hdr pointer to match the ubi->vid_hdr_shift requirements.
This works fine, but is a bit convoluted.
Moreover, the future introduction of LEB consolidation (needed to support
MLC/TLC NANDs) will allows a VID buffer to contain more than one VID
header.
Hence the creation of a ubi_vid_io_buf struct to attach extra information
to the VID header.
We currently only store the actual pointer of the underlying buffer, but
will soon add the number of VID headers contained in the buffer.
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Introduce a new list to the UBI attach information
object to be able to deal better with old and corrupted
Fastmap eraseblocks.
Also move more Fastmap specific code into fastmap.c.
Signed-off-by: Richard Weinberger <richard@nod.at>
Set free_count to zero before walking through ai->erase list
in wl_init().
Found in U-Boot as U-Boot has no workqueue/threads, it immediately
calls erase_worker(), which increase for each erased block
free_count. Without this patch, free_count gets after
this initialized to zero in wl_init(), so the free_count
variable always has the maybe wrong value 0 in U-Boot.
Signed-off-by: Heiko Schocher <hs@denx.de>
Reviewed-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
wear_leveling_worker() currently unconditionally puts a PEB on erase in
the error case even it just been taken from the free_list and never
used.
In case the PEB was never used it can be put back on the free list
saving a precious erase cycle.
v1…v2:
- to_leb_clean -> dst_leb_clean
- use the nested option for ensure_wear_leveling()
- do_sync_erase() can't go -ENOMEM so we can just go into
RO-mode now.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
If __erase_worker() fails to erase the EB and schedule_erase() fails as
well to do anything about it then we go RO. But that is not a reason to
leak the e argument here. Therefore clean up e.
Cc: <stable@vger.kernel.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
Since fastmap we gained do_sync_erase(). This function can return an error
and its error handling isn't obvious. First the memory allocation for
struct ubi_work can fail and as such struct ubi_wl_entry is leaked.
However if the memory allocation succeeds then the tail function takes
care of the struct ubi_wl_entry. A free here could result in a double
free.
To make the error handling simpler, I split the tail function into one
piece which does the work and another which frees the struct ubi_work
which is passed as argument. As result do_sync_erase() can keep the
struct on stack and we get rid of one error source.
Cc: <stable@vger.kernel.org>
Fixes: 8199b901a ("UBI: Add fastmap support to the WL sub-system")
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
UBI: attaching mtd1 to ubi0
UBI: scanning is finished
UBI error: init_volumes: not enough PEBs, required 706, available 686
UBI error: ubi_wl_init: no enough physical eraseblocks (-20, need 1)
UBI error: ubi_attach_mtd_dev: failed to attach mtd1, error -12 <= NOT ENOMEM
UBI error: ubi_init: cannot attach mtd1
If available PEBs are not enough when initializing volumes, return -ENOSPC
directly. If available PEBs are not enough when initializing WL, return
-ENOSPC instead of -ENOMEM.
Cc: stable@vger.kernel.org
Signed-off-by: Sheng Yong <shengyong1@huawei.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: David Gstir <david@sigma-star.at>
This function was added to fastmap in a very early stage
to have paranoid assertions.
With the current fastmap implementation this assert will never
trigger as fastmap PEBs are not seen by the WL sub-system.
Remove it to save us some CPU cycles.
Signed-off-by: Richard Weinberger <richard@nod.at>
Fastmap is tightly connected to the WL sub-system, many fastmap-specific
functionslive in wl.c.
To get rid of most #ifdefs in wl.c move this functions into a new file
and include it into wl.c
Signed-off-by: Richard Weinberger <richard@nod.at>
In some error paths the WL sub-system gives up on a PEB
and frees it's ubi_wl_entry struct but does not set
the entry in ubi->lookuptbl to NULL.
Fastmap can stumble over such a stale pointer as it uses
ubi->lookuptbl to find all PEBs.
Fix this by introducing a new helper function which free()s
a WL entry and removes the reference from the lookup table.
Signed-off-by: Richard Weinberger <richard@nod.at>
a) Rename ubi->fm_sem to ubi->fm_eba_sem as this semaphore
protects EBA changes.
b) Turn ubi->fm_mutex into a rw semaphore. It will still serialize
fastmap writes but also ensures that ubi_wl_put_peb() is not
interrupted by a fastmap write. We use a rw semaphore to allow
ubi_wl_put_peb() still to be executed in parallel if no fastmap
write is happening.
Signed-off-by: Richard Weinberger <richard@nod.at>
ubi_wl_get_peb() returns a fresh PEB which can be used by
user of UBI. Due to the pool logic fastmap will correctly
map this PEB upon attach time because it will be scanned.
If a new fastmap is written (due to heavy parallel io)
while the before the fresh PEB is assigned to the EBA table
it will not be scanned as it is no longer in the pool.
So, the race window exists between ubi_wl_get_peb()
and the EBA table assignment.
We have to make sure that no new fastmap can be written
while that.
To ensure that ubi_wl_get_peb() will grab ubi->fm_sem in read mode
and the user of ubi_wl_get_peb() has to release it after the PEB
got assigned.
Signed-off-by: Richard Weinberger <richard@nod.at>
If ubi_update_fastmap() fails notify the user.
This is not a hard error as ubi_update_fastmap() makes sure that upon failure
the current on-flash fastmap will no be used upon next UBI attach.
Signed-off-by: Richard Weinberger <richard@nod.at>
There is no need to allocate new ones every time, we can reuse
the existing ones.
This makes the code cleaner and more easy to follow.
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: Tanya Brokhman <tlinder@codeaurora.org>
Reviewed-by: Guido Martínez <guido@vanguardiasur.com.ar>
Currently ubi_refill_pools() first fills the first and then
the second one.
If only very few free PEBs are available the second pool can get
zero PEBs.
Change ubi_refill_pools() to distribute free PEBs fair between
all pools.
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: Guido Martínez <guido@vanguardiasur.com.ar>
Make it two functions, wl_get_wle() and wl_get_peb().
wl_get_peb() works exactly like __wl_get_peb() but wl_get_wle()
does not call produce_free_peb().
While refilling the fastmap user pool we cannot release ubi->wl_lock
as produce_free_peb() does.
Hence the fastmap logic uses now wl_get_wle().
Signed-off-by: Richard Weinberger <richard@nod.at>
ubi_wl_get_peb() has two problems, it reads the pool
size and usage counters without any protection.
While reading one value would be perfectly fine it reads multiple
values and compares them. This is racy and can lead to incorrect
pool handling.
Furthermore ubi_update_fastmap() is called without wl_lock held,
before incrementing the used counter it needs to be checked again.
It could happen that another thread consumed all PEBs from the
pool and the counter goes beyond ->size.
Signed-off-by: Richard Weinberger <richard@nod.at>
...otherwise the deferred work might run after datastructures
got freed and corrupt memory.
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: Guido Martínez <guido@vanguardiasur.com.ar>
If the WL pool runs out of PEBs we schedule a fastmap write
to refill it as soon as possible.
Ensure that only one at a time is scheduled otherwise we might end in
a fastmap write storm because writing the fastmap can schedule another
write if bitflips are detected.
Signed-off-by: Richard Weinberger <richard@nod.at>
Reviewed-by: Tanya Brokhman <tlinder@codeaurora.org>
Reviewed-by: Guido Martínez <guido@vanguardiasur.com.ar>
In some of the 'out_not_moved' error paths, lnum may be used
uninitialized. Don't ignore the warning; let's fix it.
This uninitialized variable doesn't have much visible effect in the end,
since we just schedule the PEB for erasure, and its LEB number doesn't
really matter (it just gets printed in debug messages). But let's get it
straight anyway.
Coverity CID #113449
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Brian Norris <computersforpeace@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
If the erase worker is unable to erase a PEB it will
free the ubi_wl_entry itself.
The failing ubi_wl_entry must not free()'d again after
do_sync_erase() returns.
Cc: <stable@vger.kernel.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
If there is more then one UBI device mounted, there is no way to
distinguish between messages from different UBI devices.
Add device number to all ubi layer message types.
The R/O block driver messages were replaced by pr_* since
ubi_device structure is not used by it.
Amended a bit by Artem.
Signed-off-by: Tanya Brokhman <tlinder@codeaurora.org>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
It confused me more than once that the cancel flag of the
work function does not indicate the cancellation of a single work.
In fact it indicates the WL sub-system shutdown and therefore
worker functions have to free their wl_entries too.
That's why you cannot cancel a single work, you can only shutdown
all works.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
The while loop in produce_free_peb() assumes that each work will produce a
free PEB. This is not true.
If ubi->works_count is 1 and the only scheduled work is the
wear_leveling_worker() produce_free_peb() can loop forever in case
nobody schedules an erase work.
Fix this issue by checking in the while loop whether work is scheduled.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Use the _safe variant because we're iterating over a list where items get
deleted and freed.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
In case of an error (if there are not free PEB's for example),
__wl_get_peb will return a negative value. In order to prevent access
violation we need to test the returned value prior to using it later on.
Signed-off-by: Tatyana Brokhman <tlinder@codeaurora.org>
Reviewed-by: Dolev Raviv <draviv@codeaurora.org>
Acked-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
If no free PEBs are available refill_wl_user_pool() must not
return with -ENOSPC immediately.
It has to block till produce_free_peb() produced a free PEB.
Reported-and-Tested-by: Richard Genoud <richard.genoud@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
get_peb_for_wl() removes the PEB from the free list.
If the WL subsystem detects that no wear leveling is needed
it cancels the operation and drops the gained PEB.
In this case we have to put the PEB back into the free list.
This issue was introduced with commit ed4b7021c
(UBI: remove PEB from free tree in get_peb_for_wl()).
Cc: <stable@vger.kernel.org> # 3.7.x
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Pull UBI update from Artem Bityutskiy:
"Nothing exciting, just clean-ups and nicification. Oh, and one small
optimization which makes UBI to use less RAM."
* tag 'upstream-3.8-rc1' of git://git.infradead.org/linux-ubi:
UBI: embed ubi_debug_info field in ubi_device struct
UBI: introduce helpers dbg_chk_{io, gen}
UBI: replace memcpy with struct assignment
UBI: remove spurious comment
UBI: gluebi: rename misleading variables
UBI: do not allocate the memory unnecessarily
UBI: use list_move_tail instead of list_del/list_add_tail
