At the moment UBIFS reserves twice old index size space for the
index. But this is not enough in some cases, because if the indexing
node are very fragmented and there are many small gaps, while the
dirty index has big znodes - in-the-gaps method would fail.
Thus, reserve trise as more, in which case we are guaranteed that
we can commit in any case.
If the the background thread is disabled for debugging purposes,
start it anyway, because otherwise we see tonns of kernel debugging
complaints like this:
UBIFS aligns node lengths to 8, so budgeting has to do the
same. Well, direntry, inode, and page budgets are already
aligned, but not inode data budget (e.g., data in special
devices or symlinks). Do this for inode data as well.
Also, add corresponding debugging checks.
Adrian Hunter [Wed, 23 Jul 2008 13:55:55 +0000 (16:55 +0300)]
UBIFS: correct orphan deletion order
The debug function that checks orphans, does so using the
TNC mutex. That means it will not see a correct picture
if the inode is removed from the orphan tree before it is
removed from TNC.
Signed-off-by: Adrian Hunter <ext-adrian.hunter@nokia.com>
Every time anything is deleted, UBIFS writes the deletion inode
node twice - once in 'ubifs_jnl_update()' and the second time in
'ubifs_jnl_write_inode()'. However, the second write is not needed
if no commit happened after 'ubifs_jnl_update()'. This patch
checke that condition and avoids writing the deletion inode for
the second time.
Simplify 'ubifs_jnl_write_inode()' by removing the 'deletion'
parameter which is not really needed because we may test
inode->i_nlink and check whether this is a deletion or not.
Orphan inodes are deleted inodes which will disappear after FS
re-mount. There is not need to write orphan inodes back, because
they are not needed on the flash media.
So optimize orphans a little by not writing them back. Just mark
them as clean, free the budget, and report success to VFS.
Adrian Hunter [Mon, 21 Jul 2008 12:39:05 +0000 (15:39 +0300)]
UBIFS: make ubifs_ro_mode() not inline
We use ubifs_ro_mode() quite a lot, and not in fast-path, so
there is no reason to blow the code up by having it inlined.
Also, we usually want R/O mode change to be seen to other
CPUs as soon as possible, so when we make this a function
call, we will automatically have a memory barrier.
Signed-off-by: Adrian Hunter <ext-adrian.hunter@nokia.com> Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Adrian Hunter [Fri, 18 Jul 2008 14:56:37 +0000 (17:56 +0300)]
UBIFS: ensure UBIFS switches to read-only on error writing the log
UBI transparently handles write errors by automatically copying
and remapping the affected eraseblock. If UBI is unable to do
that, for example its pool of eraseblocks reserved for bad block
handling is empty, then the error is propagated to UBIFS. UBIFS
must protect the media from falling into an inconsistent state
by immediately switching to read-only mode. In the case of log
updates, this was not being done.
Signed-off-by: Adrian Hunter <ext-adrian.hunter@nokia.com>
Adrian Hunter [Fri, 18 Jul 2008 13:47:41 +0000 (16:47 +0300)]
UBIFS: fix error return in failure mode
UBIFS recovery testing debug facility simulates media failures.
When simulating an IO error, the error code returned must be
-EIO but it was not always if the user switched off the
debug recovery testing option at the same time.
Signed-off-by: Adrian Hunter <ext-adrian.hunter@nokia.com>
Although the inode is marked as clean when it is being deleted,
it might stay and be used as orphan, and be marked as dirty.
So we have to free the budget when we delete it.
The 'ubifs_release_dirty_inode_budget()' was buggy and incorrectly
freed the budget, which led to not freeing all dirty data budget.
This patch fixes that.
Also, this patch fixes ubifs_mkdir() which passed 1 in dirty_ino_d,
which makes no sense. Well, it is harmless though.
Also, add few more useful assertions. And improve few debugging
messages.
If bit-flips happen often, UBI prints to many messages. Lessen
the amount by only printing the messages when the PEB has been
scrubbed. Also, print torturing messages.
This commit adds full UBIFS at the same state as it was pulled
by Linus on "Wed Jul 16 15:02:57 2008 -0700", commit number 9c1be0c4712fe760d8969427ef91107e9c062d91
This patch exports the 'sync_sb_inodes()' which is needed for
UBIFS because it have to force write-back from time to time.
Namely, the UBIFS budgeting subsystem forces write-back when
its pessimistic callculations show that there is not free
space on the media.
Artem Bityutskiy [Mon, 16 Jun 2008 10:35:23 +0000 (13:35 +0300)]
UBI: fix LEB locking
leb_read_unlock() may be called simultaniously by several tasks.
The would race at the following code:
up_read(&le->mutex);
if (free)
kfree(le);
And it is possible that one task frees 'le' before the other tasks
do 'up_read()'. Fix this by doing up_read and free inside the
'ubi->ltree' lock. Below it the oops we had because of this:
BUG: spinlock bad magic on CPU#0, integck/7504
BUG: unable to handle kernel paging request at 6b6b6c4f
IP: [<c0211221>] spin_bug+0x5c/0xdb
*pde = 00000000 Oops: 0000 [#1] PREEMPT SMP Modules linked in: ubifs ubi nandsim nand nand_ids nand_ecc video output
Normally UBI volumes are freed in the release function of
the struct device object. However, on error path they may
have to be freed before the struct device objects have been
initialized.
ubi_free_volume() function sets ubi->volumes[] to NULL, so
ubi_eba_close() is useless, it does not free what has to be freed.
So zap it and free vol->eba_tbl at the volume release function.
Pointed-out-by: Adrian Hunter <ext-adrian.hunter@nokia.com> Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Print information about logicale eraseblock size, sub-page
size and so on at early stage, befor an attempt to attach
the MTD device was made. This is more convenient to do so
because the attempt to attach may fail, and the information
is never printed then.
(a + b) / (c + d) != a / c + b / d. The old code errornously
assumed this incorrect formuld. Instead, just sum all erase
counters in a 64-bit variable and divide to the number of EBs
at the end.
Artem Bityutskiy [Sat, 16 Feb 2008 13:42:52 +0000 (15:42 +0200)]
UBI: make ubi-header.h local
The new trend in linux is not to store headers which define
on-media format in the include/ directory, but instead, store
them locally. This is because these headers "do not define any
kernel<->userspace interface".
Artem Bityutskiy [Thu, 24 Jan 2008 14:15:14 +0000 (16:15 +0200)]
UBI: introduce atomic LEB change ioctl
We have to be able to change individual LEBs for utilities like
ubifsck, ubifstune. For example, ubifsck has to be able to fix
errors on the media, ubifstune has to be able to change the
the superblock, hence this ioctl.
Artem Bityutskiy [Thu, 24 Jan 2008 09:23:23 +0000 (11:23 +0200)]
UBI: add layout volume information
Add more information about layout volume to make userspace tools
use the macros instead of constants. Also rename UBI_LAYOUT_VOL_ID
to make it consistent with other macros.
Artem Bityutskiy [Thu, 24 Jan 2008 09:19:14 +0000 (11:19 +0200)]
UBI: fix warnings
Old gcc complains:
CC drivers/mtd/ubi/wl.o
drivers/mtd/ubi/wl.c: In function 'wear_leveling_worker':
drivers/mtd/ubi/wl.c:746: warning: 'pe' may be used uninitialized in this function
CC drivers/mtd/ubi/scan.o
drivers/mtd/ubi/scan.c: In function 'ubi_scan':
drivers/mtd/ubi/scan.c:772: warning: 'ec' may be used uninitialized in this function
drivers/mtd/ubi/scan.c:772: note: 'ec' was declared here
Artem Bityutskiy [Wed, 16 Jan 2008 13:44:24 +0000 (15:44 +0200)]
UBI: add auto-resize feature
The problem: NAND flashes have different amount of initial bad physical
eraseblocks (marked as bad by the manufacturer). For example, for 256MiB
Samsung OneNAND flash there might be from 0 to 40 bad initial eraseblocks,
which is about 2%. When UBI is used as the base system, one needs to know
the exact amount of good physical eraseblocks, because this number is
needed to create the UBI image which is put to the devices during
production. But this number is not know, which forces us to use the
minimum number of good physical eraseblocks. And UBI additionally
reserves some percentage of physical eraseblocks for bad block handling
(default is 1%), so we have 1-3% of PEBs reserved at the end, depending
on the amount of initial bad PEBs. But it is desired to always have
1% (or more, depending on the configuration).
Solution: this patch adds an "auto-resize" flag to the volume table.
The volume which has the "auto-resize" flag will automatically be re-sized
(enlarged) on the first UBI initialization. UBI clears the flag when
the volume is re-sized. Only one volume may have the "auto-resize" flag.
So, the production UBI image may have one volume with "auto-resize"
flag set, and its size is automatically adjusted on the first boot
of the device.
Artem Bityutskiy [Wed, 16 Jan 2008 10:15:47 +0000 (12:15 +0200)]
UBI: fix warnings
drivers/mtd/ubi/cdev.c: In function ‘vol_cdev_read’:
drivers/mtd/ubi/cdev.c:187: warning: unused variable ‘vol_id’
CC [M] drivers/mtd/ubi/kapi.o
drivers/mtd/ubi/kapi.c: In function ‘ubi_leb_erase’:
drivers/mtd/ubi/kapi.c:483: warning: unused variable ‘vol_id’
drivers/mtd/ubi/kapi.c: In function ‘ubi_leb_unmap’:
drivers/mtd/ubi/kapi.c:544: warning: unused variable ‘vol_id’
drivers/mtd/ubi/kapi.c: In function ‘ubi_leb_map’:
drivers/mtd/ubi/kapi.c:582: warning: unused variable ‘vol_id’
Artem Bityutskiy [Wed, 16 Jan 2008 10:11:54 +0000 (12:11 +0200)]
UBI: get rid of ubi_ltree_slab
This slab cache is not really needed since the number of objects
is low and the constructor does not make much sense because we
allocate oblects when doint I/O, which is way slower then allocation.
Artem Bityutskiy [Tue, 25 Dec 2007 17:17:00 +0000 (19:17 +0200)]
UBI: fix mtd device string parsing
UBI allows to specify MTD device name or number when the module is being
loaded. When parsing MTD device identity string, it first tries to treat
it as device NAME, and if that fails, it treats it as device number.
Make it vice-versa as this is more logical and makes less troubles when
you have an MTD device named "1" and try to load mtd1 which has different
name. This is especially easy to hit when gluebi is enabled.
Artem Bityutskiy [Tue, 25 Dec 2007 16:13:33 +0000 (18:13 +0200)]
UBI: use separate mutex for volumes checking
Introduce a separate mutex which serializes volumes checking,
because we cammot really use volumes_mutex - it cases reverse
locking problems with mtd_tbl_mutex when gluebi is used -
thanks to lockdep.
Artem Bityutskiy [Mon, 17 Dec 2007 15:37:26 +0000 (17:37 +0200)]
UBI: add UBI devices reference counting
This is one more step on the way to "removable" UBI devices. It
adds reference counting for UBI devices. Every time a volume on
this device is opened - the device's refcount is increased. It
is also increased if someone is reading any sysfs file of this
UBI device or of one of its volumes.
Artem Bityutskiy [Sun, 16 Dec 2007 14:59:31 +0000 (16:59 +0200)]
UBI: add UBI control device
This patch is a preparation to make UBI devices dynamic. It
adds an UBI control device which has dynamically allocated
major number and registers itself as "ubi_ctrl". It does not
do anything so far. The idea is that this device will allow
to attach/detach MTD devices from userspace.
This is symilar to what the Linux device mapper has.
The next things to do are:
* Fix UBI, because it now assumes UBI devices cannot go away
* Implement control device ioctls which will attach/detach MTD
devices
Artem Bityutskiy [Tue, 18 Dec 2007 13:06:55 +0000 (15:06 +0200)]
UBI: bugfix: protect from volume removal
When the WL worker is moving an LEB, the volume might go away
occasionally. UBI does not handle these situations correctly.
This patch introduces a new mutex which serializes wear-levelling
worker and the the 'ubi_wl_put_peb()' function. Now, if one puts
an LEB, and its PEB is being moved, it will wait on the mutex.
And because we unmap all LEBs when removing volumes, this will make
the volume remove function to wait while the LEB movement
finishes.
Below is an example of an oops which should be fixed by this patch:
Artem Bityutskiy [Mon, 17 Dec 2007 13:42:57 +0000 (15:42 +0200)]
UBI: introduce volume refcounting
Add ref_count field to UBI volumes and remove weired "vol->removed"
field. This way things are better understandable and we do not have
to do whold show_attr operation under spinlock.
Artem Bityutskiy [Mon, 17 Dec 2007 13:48:49 +0000 (15:48 +0200)]
UBI: simplify error handling
If we fail halfway through sysfs file creation, we may just call
sysfs remove function and it will delete all the files we created.
For non-existing files it will also be OK - the remove functions
just return -ENOENT.
Artem Bityutskiy [Mon, 17 Dec 2007 12:22:55 +0000 (14:22 +0200)]
UBI: fix and cleanup volume opening functions
This patch fixes error codes of the functions - if the device number
is out of range, -EINVAL should be returned. It also removes unneeded
try_module_get call from the open by name function.
Artem Bityutskiy [Mon, 17 Dec 2007 12:02:09 +0000 (14:02 +0200)]
UBI: fix error path
Error path in volume creation is bogus. First of, it ovverrides the
'err' variable and returns zero to the caller. Second, ubi_assert()
in the release function is wrong.
Artem Bityutskiy [Mon, 17 Dec 2007 11:09:09 +0000 (13:09 +0200)]
UBI: get device when opening volume
When a volume is opened, get its kref via get_device() call.
And put the reference when closing the volume. With this, we
may have a bit saner volume delete.
projects / pandora-kernel.git / log