RCU free the struct inode. This will allow:
- Subsequent store-free path walking patch. The inode must be consulted for
permissions when walking, so an RCU inode reference is a must.
- sb_inode_list_lock to be moved inside i_lock because sb list walkers who want
to take i_lock no longer need to take sb_inode_list_lock to walk the list in
the first place. This will simplify and optimize locking.
- Could remove some nested trylock loops in dcache code
- Could potentially simplify things a bit in VM land. Do not need to take the
page lock to follow page->mapping.
The downsides of this is the performance cost of using RCU. In a simple
creat/unlink microbenchmark, performance drops by about 10% due to inability to
reuse cache-hot slab objects. As iterations increase and RCU freeing starts
kicking over, this increases to about 20%.
In cases where inode lifetimes are longer (ie. many inodes may be allocated
during the average life span of a single inode), a lot of this cache reuse is
not applicable, so the regression caused by this patch is smaller.
The cache-hot regression could largely be avoided by using SLAB_DESTROY_BY_RCU,
however this adds some complexity to list walking and store-free path walking,
so I prefer to implement this at a later date, if it is shown to be a win in
real situations. I haven't found a regression in any non-micro benchmark so I
doubt it will be a problem.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
* 'linux-next' of git://git.infradead.org/ubifs-2.6:
UBIFS: do not allocate unneeded scan buffer
UBIFS: do not forget to cancel timers
UBIFS: remove a bit of unneeded code
UBIFS: add a commentary about log recovery
UBIFS: avoid kernel error if ubifs superblock read fails
UBIFS: introduce new flags for RO mounts
UBIFS: introduce new flag for RO due to errors
UBIFS: check return code of pnode_lookup
UBIFS: check return code of ubifs_lpt_lookup
UBIFS: improve error reporting when reading bad node
UBIFS: introduce list sorting debugging checks
UBIFS: fix assertion warnings in comparison function
UBIFS: mark unused key objects as invalid
UBIFS: do not write rubbish into truncation scanning node
UBIFS: improve assertion in node comparison functions
UBIFS: do not use key type in list_sort
UBIFS: do not look up truncation nodes
UBIFS: fix assertion warning
UBIFS: do not treat ENOSPC specially
UBIFS: switch to RO mode after synchronizing
In 'ubifs_replay_journal()' we allocate 'sbuf' for scanning the log.
However, we already have 'c->sbuf' for these purposes, so do not
allocate yet another one. This reduces UBIFS memory consumption while
recovering.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This is a bug-fix: when we unmount, and we are currently in R/O
mode because of an error - we do not sync write-buffers, which
means we also do not cancel write-buffer timers we may possibly
have armed. This patch fixes the issue.
The issue can easily be reproduced by enabling UBIFS failure debug
mode (echo 4 > /sys/module/ubifs/parameters/debug_tsts) and
unmounting as soon as a failure happen. At some point the system
oopses because we have an armed hrtimer but UBIFS is unmounted
already.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This is a clean-up patch which:
1. Removes explicite 'hrtimer_cancel()' after 'ubifs_wbuf_sync()' in
'ubifs_remount_ro()', because the timers will be canceled by
'ubifs_wbuf_sync()', no need to cancel them for the second time.
2. Remove "if (c->jheads)" check from 'ubifs_put_super()', because
at journal heads must always be allocated there, since we checked
earlier that we were mounted R/W, and the olny situation when
journal heads are not allocated is when mounter or re-mounted R/O.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Add a commentary which elaborates that 'ubifs_recover_log_leb()' recovers only
the last log LEB, not any. Also remove some unneeded newlines.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
All file_operations should get a .llseek operation so we can make
nonseekable_open the default for future file operations without a
.llseek pointer.
The three cases that we can automatically detect are no_llseek, seq_lseek
and default_llseek. For cases where we can we can automatically prove that
the file offset is always ignored, we use noop_llseek, which maintains
the current behavior of not returning an error from a seek.
New drivers should normally not use noop_llseek but instead use no_llseek
and call nonseekable_open at open time. Existing drivers can be converted
to do the same when the maintainer knows for certain that no user code
relies on calling seek on the device file.
The generated code is often incorrectly indented and right now contains
comments that clarify for each added line why a specific variant was
chosen. In the version that gets submitted upstream, the comments will
be gone and I will manually fix the indentation, because there does not
seem to be a way to do that using coccinelle.
Some amount of new code is currently sitting in linux-next that should get
the same modifications, which I will do at the end of the merge window.
Many thanks to Julia Lawall for helping me learn to write a semantic
patch that does all this.
===== begin semantic patch =====
// This adds an llseek= method to all file operations,
// as a preparation for making no_llseek the default.
//
// The rules are
// - use no_llseek explicitly if we do nonseekable_open
// - use seq_lseek for sequential files
// - use default_llseek if we know we access f_pos
// - use noop_llseek if we know we don't access f_pos,
// but we still want to allow users to call lseek
//
@ open1 exists @
identifier nested_open;
@@
nested_open(...)
{
<+...
nonseekable_open(...)
...+>
}
@ open exists@
identifier open_f;
identifier i, f;
identifier open1.nested_open;
@@
int open_f(struct inode *i, struct file *f)
{
<+...
(
nonseekable_open(...)
|
nested_open(...)
)
...+>
}
@ read disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ read_no_fpos disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
... when != off
}
@ write @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ write_no_fpos @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
... when != off
}
@ fops0 @
identifier fops;
@@
struct file_operations fops = {
...
};
@ has_llseek depends on fops0 @
identifier fops0.fops;
identifier llseek_f;
@@
struct file_operations fops = {
...
.llseek = llseek_f,
...
};
@ has_read depends on fops0 @
identifier fops0.fops;
identifier read_f;
@@
struct file_operations fops = {
...
.read = read_f,
...
};
@ has_write depends on fops0 @
identifier fops0.fops;
identifier write_f;
@@
struct file_operations fops = {
...
.write = write_f,
...
};
@ has_open depends on fops0 @
identifier fops0.fops;
identifier open_f;
@@
struct file_operations fops = {
...
.open = open_f,
...
};
// use no_llseek if we call nonseekable_open
////////////////////////////////////////////
@ nonseekable1 depends on !has_llseek && has_open @
identifier fops0.fops;
identifier nso ~= "nonseekable_open";
@@
struct file_operations fops = {
... .open = nso, ...
+.llseek = no_llseek, /* nonseekable */
};
@ nonseekable2 depends on !has_llseek @
identifier fops0.fops;
identifier open.open_f;
@@
struct file_operations fops = {
... .open = open_f, ...
+.llseek = no_llseek, /* open uses nonseekable */
};
// use seq_lseek for sequential files
/////////////////////////////////////
@ seq depends on !has_llseek @
identifier fops0.fops;
identifier sr ~= "seq_read";
@@
struct file_operations fops = {
... .read = sr, ...
+.llseek = seq_lseek, /* we have seq_read */
};
// use default_llseek if there is a readdir
///////////////////////////////////////////
@ fops1 depends on !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier readdir_e;
@@
// any other fop is used that changes pos
struct file_operations fops = {
... .readdir = readdir_e, ...
+.llseek = default_llseek, /* readdir is present */
};
// use default_llseek if at least one of read/write touches f_pos
/////////////////////////////////////////////////////////////////
@ fops2 depends on !fops1 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read.read_f;
@@
// read fops use offset
struct file_operations fops = {
... .read = read_f, ...
+.llseek = default_llseek, /* read accesses f_pos */
};
@ fops3 depends on !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write.write_f;
@@
// write fops use offset
struct file_operations fops = {
... .write = write_f, ...
+ .llseek = default_llseek, /* write accesses f_pos */
};
// Use noop_llseek if neither read nor write accesses f_pos
///////////////////////////////////////////////////////////
@ fops4 depends on !fops1 && !fops2 && !fops3 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
identifier write_no_fpos.write_f;
@@
// write fops use offset
struct file_operations fops = {
...
.write = write_f,
.read = read_f,
...
+.llseek = noop_llseek, /* read and write both use no f_pos */
};
@ depends on has_write && !has_read && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write_no_fpos.write_f;
@@
struct file_operations fops = {
... .write = write_f, ...
+.llseek = noop_llseek, /* write uses no f_pos */
};
@ depends on has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
@@
struct file_operations fops = {
... .read = read_f, ...
+.llseek = noop_llseek, /* read uses no f_pos */
};
@ depends on !has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
@@
struct file_operations fops = {
...
+.llseek = noop_llseek, /* no read or write fn */
};
===== End semantic patch =====
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Julia Lawall <julia@diku.dk>
Cc: Christoph Hellwig <hch@infradead.org>
.get_sb is called on mounts with automatic fs detection too, so this
function should print an error if it cannot read the superblock in
debug mode only (new behaviour conforms the other fs types)
Signed-off-by: Steffen Sledz <sledz@dresearch.de>
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Commit 2fde99cb55 "UBIFS: mark VFS SB RO too"
introduced regression. This commit made UBIFS set the 'MS_RDONLY' flag in the
VFS superblock when it switches to R/O mode due to an error. This was done
to make VFS show the R/O UBIFS flag in /proc/mounts.
However, several places in UBIFS relied on the 'MS_RDONLY' flag and assume this
flag can only change when we re-mount. For example, 'ubifs_put_super()'.
This patch introduces new UBIFS flag - 'c->ro_mount' which changes only when
we re-mount, and preserves the way UBIFS was originally mounted (R/W or R/O).
This allows us to de-initialize UBIFS cleanly in 'ubifs_put_super()'.
This patch also changes all 'ubifs_assert(!c->ro_media)' assertions to
'ubifs_assert(!c->ro_media && !c->ro_mount)', because we never should write
anything if the FS was mounter R/O.
All the places where we test for 'MS_RDONLY' flag in the VFS SB were changed
and now we test the 'c->ro_mount' flag instead, because it preserves the
original UBIFS mount type, unlike the 'MS_RDONLY' flag.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
The R/O state may have various reasons:
1. The UBI volume is R/O
2. The FS is mounted R/O
3. The FS switched to R/O mode because of an error
However, in UBIFS we have only one variable which represents cases
1 and 3 - 'c->ro_media'. Indeed, we set this to 1 if we switch to
R/O mode due to an error, and then we test it in many places to
make sure that we stop writing as soon as the error happens.
But this is very unclean. One consequence of this, for example, is
that in 'ubifs_remount_fs()' we use 'c->ro_media' to check whether
we are in R/O mode because on an error, and we print a message
in this case. However, if we are in R/O mode because the media
is R/O, our message is bogus.
This patch introduces new flag - 'c->ro_error' which is set when
we switch to R/O mode because of an error. It also changes all
"if (c->ro_media)" checks to "if (c->ro_error)" checks, because
this is what the checks actually mean. We do not need to check
for 'c->ro_media' because if the UBI volume is in R/O mode, we
do not allow R/W mounting, and now writes can happen. This is
guaranteed by VFS. But it is good to double-check this, so this
patch also adds many "ubifs_assert(!c->ro_media)" checks.
In the 'ubifs_remount_fs()' function this patch makes a bit more
changes - it fixes the error messages as well.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Function pnode_lookup may return ERR_PTR(...). Check for it.
Signed-off-by: Vasiliy Kulikov <segooon@gmail.com>
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Function ubifs_lpt_lookup may return ERR_PTR(...). Check for it.
[Tweaked by Artem Bityutskiy]
Signed-off-by: Vasiliy Kulikov <segooon@gmail.com>
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When an error happens during validation of read node, the typical situation is that
the LEB we read is unmapped (due to some bug). It is handy to include the mapping
status into the error message.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
The UBIFS bug in the GC list sorting comparison functions inspired
me to write internal debugging check functions which verify that
the list of nodes is sorted properly.
So, this patch implements 2 new debugging functions:
o 'dbg_check_data_nodes_order()' - check order of data nodes list
o 'dbg_check_nondata_nodes_order()' - check order of non-data nodes list
The debugging functions are executed only if general UBIFS debugging checks are
enabled. And they are compiled out if UBIFS debugging is disabled.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When running the integrity test ('integck' from mtd-utils) on current
UBIFS on 2.6.35, I see that assertions in UBIFS 'list_sort()' comparison
functions trigger sometimes, e.g.:
UBIFS assert failed in data_nodes_cmp at 132 (pid 28311)
My investigation showed that this happens when 'list_sort()' calls the 'cmp()'
function with equivalent arguments. In this case, the 'struct list_head'
parameter, passed to 'cmp()' is bogus, and it does not belong to any element in
the original list.
And this issue seems to be introduced by commit:
commit 835cc0c847
Author: Don Mullis <don.mullis@gmail.com>
Date: Fri Mar 5 13:43:15 2010 -0800
It is easy to work around the issue by doing:
if (a == b)
return 0;
in UBIFS. It works, but 'lib_sort()' should nevertheless be fixed. Although it
is harmless to have this piece of code in UBIFS.
This patch adds that code to both UBIFS 'cmp()' functions:
'data_nodes_cmp()' and 'nondata_nodes_cmp()'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When scanning the flash, UBIFS builds a list of flash nodes of type
'struct ubifs_scan_node'. Each scanned node has a 'snod->key' field. This field
is valid for most of the nodes, but invalid for some node type, e.g., truncation
nodes. It is safer to explicitly initialize such keys to something invalid,
rather than leaving them initialized to all zeros, which has key type of
UBIFS_INO_KEY.
This patch introduces new "fake" key type UBIFS_INVALID_KEY and initializes
unused 'snod->key' objects to this type. It also adds debugging assertions in
the TNC code to make sure no one ever tries to look these nodes up in the TNC.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
In the scanning code, in 'ubifs_add_snod()', we write rubbish into
'snod->key', because we assume that on-flash truncation nodes have a key, but
they do not. If the other parts of UBIFS then mistakenly try to look-up
the truncation node key (they should not do this, but may do because of a bug),
we can succeed and corrupt TNC. It looks like we did have such a situation in
'sort_nodes()' in gc.c.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Improve assertions in gc.c in the comparison functions for 'list_sort()': check
key types _and_ node types.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
In comparison function for 'list_sort()' we use key type to distinguish between
node types. However, we have a bit simper way to detect node type -
'snod->type'. This more logical to use, comparing to decoding key types. Also
allows to get rid of 2 local variables.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
When moving nodes in GC, do not try to look up truncation nodes in TNC,
because they do not exist there. This would be harmless, because the TNC
look-up would fail, if we did not have bug 'ubifs_add_snod()' which reads
garbage into 'snod->key'. But in any case, it is less error prone to
explicitly ignore everything but inode, data, dentry and xentry nodes.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
This patch fixes the following false assertion warning:
UBIFS assert failed in data_nodes_cmp at 130 (pid 15107)
The assertion was wrong because it did not take into account that the
node can be an xentry.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
'ubifs_garbage_collect_leb()' should never return '-ENOSPC', and if it
does, this is an error. Thus, do not treat this error code specially.
'-EAGAIN' is a special error code, but not '-ENOSPC'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
In 'ubifs_garbage_collect()' on error path, we first switch to R/O mode, and
then synchronize write-buffers (to make sure no data are lost). But the GC
write-buffer synchronization will fail, because we are already in R/O mode.
This patch re-orders this and makes sure we first synchronize the write-buffer,
and then switch to R/O mode.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Nick Piggin
Al Viro
Linus Torvalds
Artem Bityutskiy
Arnd Bergmann
Steffen Sledz
Vasiliy Kulikov