]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/log
mirror_ubuntu-artful-kernel.git
11 years agorbd: set initial capacity in rbd_init_disk()
Alex Elder [Wed, 29 Aug 2012 22:11:07 +0000 (17:11 -0500)]
rbd: set initial capacity in rbd_init_disk()

Move the setting of the initial capacity for an rbd image mapping
into rb_init_disk().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: drop dev registration check for new snap
Alex Elder [Fri, 31 Aug 2012 22:29:53 +0000 (17:29 -0500)]
rbd: drop dev registration check for new snap

By the time rbd_dev_snaps_register() gets called during rbd device
initialization, the main device will have already been registered.
Similarly, a header refresh will only occur for an rbd device whose
Linux device is registered.  There is therefore no need to verify
the main device is registered when registering a snapshot device.

For the time being, turn the check into a WARN_ON(), but it can
eventually just go away.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: call rbd_init_disk() sooner
Alex Elder [Wed, 29 Aug 2012 22:11:07 +0000 (17:11 -0500)]
rbd: call rbd_init_disk() sooner

Call rbd_init_disk() from rbd_add() as soon as we have the major
device number for the mapping.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: defer setting device id
Alex Elder [Fri, 27 Jul 2012 04:37:14 +0000 (23:37 -0500)]
rbd: defer setting device id

Hold off setting the device id and formatting the device name
in rbd_add() until just before it's needed.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: read the header before registering device
Alex Elder [Wed, 29 Aug 2012 22:11:07 +0000 (17:11 -0500)]
rbd: read the header before registering device

Read the rbd header information and call rbd_dev_set_mapping()
earlier--before registering the block device or setting up the sysfs
entries for the image.  The sysfs entries provide users access to
some information that's only available after doing the rbd header
initialization, so this will make sure it's valid right away.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: call set_snap() before snap_devs_update()
Alex Elder [Wed, 29 Aug 2012 22:11:07 +0000 (17:11 -0500)]
rbd: call set_snap() before snap_devs_update()

rbd_header_set_snap() is a simple initialization routine for an rbd
device's mapping.  It has to be called after the snapshot context
for the rbd_dev has been updated, but can be done before snapshot
devices have been registered.

Change the name to rbd_dev_set_mapping() to better reflect its
purpose, and call it a little sooner, before registering snapshot
devices.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: defer registering snapshot devices
Alex Elder [Fri, 31 Aug 2012 22:29:52 +0000 (17:29 -0500)]
rbd: defer registering snapshot devices

When a new snapshot is found in an rbd device's updated snapshot
context, __rbd_add_snap_dev() is called to create and insert an
entry in the rbd devices list of snapshots.  In addition, a Linux
device is registered to represent the snapshot.

For version 2 rbd images, it will be undesirable to initialize the
device right away.  So in anticipation of that, this patch separates
the insertion of a snapshot entry in the snaps list from the
creation of devices for those snapshots.

To do this, create a new function rbd_dev_snaps_register() which
traverses the list of snapshots and calls rbd_register_snap_dev()
on any that have not yet been registered.

Rename rbd_dev_snap_devs_update() to be rbd_dev_snaps_update()
to better reflect that only the entry in the snaps list and not
the snapshot's device is affected by the function.

For now, call rbd_dev_snaps_register() immediately after each
call to rbd_dev_snaps_update().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: assign header name later
Alex Elder [Tue, 3 Jul 2012 21:01:19 +0000 (16:01 -0500)]
rbd: assign header name later

Move the assignment of the header name for an rbd image a bit later,
outside rbd_add_parse_args() and into its caller.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: use snaps list in rbd_snap_by_name()
Alex Elder [Wed, 11 Jul 2012 01:30:11 +0000 (20:30 -0500)]
rbd: use snaps list in rbd_snap_by_name()

An rbd_dev structure maintains a list of current snapshots that have
already been fully initialized.  The entries on the list have type
struct rbd_snap, and each entry contains a copy of information
that's found in the rbd_dev's snapshot context and header.

The only caller of snap_by_name() is rbd_header_set_snap().  In that
call site any positive return value (the index in the snapshot
array) is ignored, so there's no need to return the index in
the snapshot context's id array when it's found.

rbd_header_set_snap() also has only one caller--rbd_add()--and that
call is made after a call to rbd_dev_snap_devs_update().  Because
the rbd_snap structures are initialized in that function, the
current snapshot list can be used instead of the snapshot context to
look up a snapshot's information by name.

Change snap_by_name() so it uses the snapshot list rather than the
rbd_dev's snapshot context in looking up snapshot information.
Return 0 if it's found rather than the snapshot id.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: don't register snapshots in bus_add_dev()
Alex Elder [Thu, 30 Aug 2012 05:16:38 +0000 (00:16 -0500)]
rbd: don't register snapshots in bus_add_dev()

When rbd_bus_add_dev() is called (one spot--in rbd_add()), the rbd
image header has not even been read yet.  This means that the list
of snapshots will be empty at the time of the call.  As a result,
there is no need for the code that calls rbd_register_snap_dev()
for each entry in that list--so get rid of it.

Once the header has been read (just after returning), a call will
be made to rbd_dev_snap_devs_update(), which will then find every
snapshot in the context to be new and will therefore call
rbd_register_snap_dev() via __rbd_add_snap_dev() accomplishing
the same thing.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: move locking out of rbd_header_set_snap()
Alex Elder [Fri, 24 Aug 2012 04:48:49 +0000 (23:48 -0500)]
rbd: move locking out of rbd_header_set_snap()

Move the calls to get the header semaphore out of
rbd_header_set_snap() and into its caller.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: simplify rbd_init_disk() a bit
Alex Elder [Wed, 29 Aug 2012 22:11:06 +0000 (17:11 -0500)]
rbd: simplify rbd_init_disk() a bit

This just simplifies a few things in rbd_init_disk(), now that the
previous patch has moved a bunch of initialization code out if it.
Done separately to facilitate review.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: do some header initialization earlier
Alex Elder [Wed, 11 Jul 2012 01:30:10 +0000 (20:30 -0500)]
rbd: do some header initialization earlier

Move some of the code that initializes an rbd header out of
rbd_init_disk() and into its caller.

Move the code at the end of rbd_init_disk() that sets the device
capacity and activates the Linux device out of that function and
into the caller, ensuring we still have the disk size available
where we need it.

Update rbd_free_disk() so it still aligns well as an inverse of
rbd_init_disk(), moving the rbd_header_free() call out to its
caller.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: simplify snap_by_name() interface
Alex Elder [Thu, 30 Aug 2012 19:42:15 +0000 (14:42 -0500)]
rbd: simplify snap_by_name() interface

There is only one caller of snap_by_name(), and it passes two values
to be assigned, both of which are found within an rbd device
structure.

Change the interface so it just passes the address of the rbd_dev,
and make the assignments to its fields directly.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: set mapping name with the rest
Alex Elder [Fri, 31 Aug 2012 22:29:52 +0000 (17:29 -0500)]
rbd: set mapping name with the rest

With the exception of the snapshot name, all of the mapping-specific
fields in an rbd device structure are set in rbd_header_set_snap().

Pass the snapshot name to be assigned into rbd_header_set_snap()
to keep all of the mapping assignments together.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: return snap name from rbd_add_parse_args()
Alex Elder [Fri, 31 Aug 2012 22:29:52 +0000 (17:29 -0500)]
rbd: return snap name from rbd_add_parse_args()

This is the first of two patches aimed at isolating the code that
sets the mapping information into a single spot.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: record mapped size
Alex Elder [Thu, 30 Aug 2012 19:42:15 +0000 (14:42 -0500)]
rbd: record mapped size

Add the size of the mapped image to the set of mapping-specific
fields in an rbd_device, and use it when setting the capacity of the
disk.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: separate mapping info in rbd_dev
Alex Elder [Fri, 31 Aug 2012 22:29:51 +0000 (17:29 -0500)]
rbd: separate mapping info in rbd_dev

Several fields in a struct rbd_dev are related to what is mapped, as
opposed to the actual base rbd image.  If the base image is mapped
these are almost unneeded, but if a snapshot is mapped they describe
information about that snapshot.

In some contexts this can be a little bit confusing.  So group these
mapping-related field into a structure to make it clear what they
are describing.

This also includes a minor change that rearranges the fields in the
in-core image header structure so that invariant fields are at the
top, followed by those that change.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: kill rbd_image_header->total_snaps
Alex Elder [Thu, 30 Aug 2012 19:42:15 +0000 (14:42 -0500)]
rbd: kill rbd_image_header->total_snaps

The "total_snaps" field in an rbd header structure is never any
different from the value of "num_snaps" stored within a snapshot
context.  Avoid any confusion by just using the value held within
the snapshot context, and get rid of the "total_snaps" field.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: kill rbd_dev->q
Alex Elder [Wed, 29 Aug 2012 22:11:06 +0000 (17:11 -0500)]
rbd: kill rbd_dev->q

A copy of rbd_dev->disk->queue is held in rbd_dev->q, but it's
never actually used.  So get just get rid of the field.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: rename __rbd_init_snaps_header()
Alex Elder [Fri, 24 Aug 2012 04:48:49 +0000 (23:48 -0500)]
rbd: rename __rbd_init_snaps_header()

The name __rbd_init_snaps_header() doesn't really convey what that
function does very well.  Its purpose is to scan a new snapshot
context and either create or destroy snapshot device entries so
that local host's view is consistent with the reality maintained
on the OSDs.  This patch just changes the name of this function,
to be rbd_dev_snap_devs_update().  Still not perfect, but I think
better.

Also add some dynamic debug statements to this function.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: rename rbd_id_get()
Alex Elder [Wed, 29 Aug 2012 22:11:06 +0000 (17:11 -0500)]
rbd: rename rbd_id_get()

This should have been done as part of this commit:

    commit de71a2970d57463d3d965025e33ec3adcf391248
    Author: Alex Elder <elder@inktank.com>
    Date:   Tue Jul 3 16:01:19 2012 -0500
    rbd: rename rbd_device->id

rbd_id_get() is assigning the rbd_dev->dev_id field.  Change the
name of that function as well as rbd_id_put() and rbd_id_max
to reflect what they are affecting.

Add some dynamic debug statements related to rbd device id activity.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: define rbd_assert()
Alex Elder [Thu, 6 Sep 2012 21:00:54 +0000 (16:00 -0500)]
rbd: define rbd_assert()

Define rbd_assert() and use it in place of various BUG_ON() calls
now present in the code.  By default assertion checking is enabled;
we want to do this differently at some point.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: split up rbd_get_segment()
Alex Elder [Thu, 9 Aug 2012 17:33:26 +0000 (10:33 -0700)]
rbd: split up rbd_get_segment()

There are two places where rbd_get_segment() is called.  One, in
rbd_rq_fn(), only needs to know the length within a segment that an
I/O request should be.  The other, in rbd_do_op(), also needs the
name of the object and the offset within it for the I/O request.

Split out rbd_segment_name() into three dedicated functions:
    - rbd_segment_name() allocates and formats the name of the
      object for a segment containing a given rbd image offset
    - rbd_segment_offset() computes the offset within a segment for
      a given rbd image offset
    - rbd_segment_length() computes the length to use for I/O within
      a segment for a request, not to exceed the end of a segment
      object.

In the new functions be a bit more careful, checking for possible
error conditions:
    - watch for errors or overflows returned by snprintf()
    - catch (using BUG_ON()) potential overflow conditions
      when computing segment length

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: check for overflow in rbd_get_num_segments()
Alex Elder [Thu, 9 Aug 2012 17:33:26 +0000 (10:33 -0700)]
rbd: check for overflow in rbd_get_num_segments()

It is possible in rbd_get_num_segments() for an overflow to occur
when adding the offset and length.  This is easily avoided.

Since the function returns an int and the one caller is already
prepared to handle errors, have it return -ERANGE if overflow would
occur.

The overflow check would not work if a zero-length request was
being tested, so short-circuit that case, returning 0 for the
number of segments required.  (This condition might be avoided
elsewhere already, I don't know.)

Have the caller end the request if either an error or 0 is returned.
The returned value is passed to __blk_end_request_all(), meaning
a 0 length request is not treated an error.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: drop needless test in rbd_rq_fn()
Alex Elder [Thu, 9 Aug 2012 17:33:25 +0000 (10:33 -0700)]
rbd: drop needless test in rbd_rq_fn()

There's a test for null rq pointer inside the while loop in
rbd_rq_fn() that's not needed.  That same test already occurred
in the immediatly preceding loop condition test.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: bio_chain_clone() cleanups
Alex Elder [Thu, 9 Aug 2012 17:33:25 +0000 (10:33 -0700)]
rbd: bio_chain_clone() cleanups

In bio_chain_clone(), at the end of the function the bi_next field
of the tail of the new bio chain is nulled.  This isn't necessary,
because if "tail" is non-null, its value will be the last bio
structure allocated at the top of the while loop in that function.
And before that structure is added to the end of the new chain, its
bi_next pointer is always made null.

While touching that function, clean a few other things:
    - define each local variable on its own line
    - move the definition of "tmp" to an inner scope
    - move the modification of gfpmask closer to where it's used
    - rearrange the logic that sets the chain's tail pointer

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: kill notify_timeout option
Alex Elder [Fri, 10 Aug 2012 20:12:07 +0000 (13:12 -0700)]
rbd: kill notify_timeout option

The "notify_timeout" rbd device option is never used, so get rid of
it.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: add read_only rbd map option
Alex Elder [Fri, 10 Aug 2012 20:12:07 +0000 (13:12 -0700)]
rbd: add read_only rbd map option

Add the ability to map an rbd image read-only, by specifying either
"read_only" or "ro" as an option on the rbd "command line."  Also
allow the inverse to be explicitly specified using "read_write" or
"rw".

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: move rbd_opts to struct rbd_device
Alex Elder [Fri, 10 Aug 2012 20:12:07 +0000 (13:12 -0700)]
rbd: move rbd_opts to struct rbd_device

The rbd options don't really apply to the ceph client.  So don't
store a pointer to it in the ceph_client structure, and put them
(a struct, not a pointer) into the rbd_dev structure proper.

Pass the rbd device structure to rbd_client_create() so it can
assign rbd_dev->rbdc if successful, and have it return an error code
instead of the rbd client pointer.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: more cleanup in rbd_header_from_disk()
Alex Elder [Fri, 24 Aug 2012 04:22:06 +0000 (23:22 -0500)]
rbd: more cleanup in rbd_header_from_disk()

This just rearranges things a bit more in rbd_header_from_disk()
so that the snapshot sizes are initialized right after the buffer
to hold them is allocated and doing a little further consolidation
that follows from that.  Also adds a few simple comments.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: kill incore snap_names_len
Alex Elder [Fri, 24 Aug 2012 04:22:06 +0000 (23:22 -0500)]
rbd: kill incore snap_names_len

The only thing the on-disk snap_names_len field is needed is to
size the buffer allocated to hold a copy of the snapshot names
for an rbd image.

So don't bother saving it in the in-core rbd_image_header structure.
Just use a local variable to hold the required buffer size while
it's needed.

Move the code that actually copies the snapshot names up closer
to where the required length is saved.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: don't over-allocate space for object prefix
Alex Elder [Fri, 24 Aug 2012 04:22:06 +0000 (23:22 -0500)]
rbd: don't over-allocate space for object prefix

In rbd_header_from_disk() the object prefix buffer is sized based on
the maximum size it's block_name equivalent on disk could be.

Instead, only allocate enough to hold null-terminated string from
the on-disk header--or the maximum size of no NUL is found.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agorbd: handle locking inside __rbd_client_find()
Alex Elder [Fri, 10 Aug 2012 20:12:07 +0000 (13:12 -0700)]
rbd: handle locking inside __rbd_client_find()

There is only caller of __rbd_client_find(), and it somewhat
clumsily gets the appropriate lock and gets a reference to the
existing ceph_client structure if it's found.

Instead, have that function handle its own locking, and acquire the
reference if found while it holds the lock.  Drop the underscores
from the name because there's no need to signify anything special
about this function.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
11 years agoceph: use list_move_tail instead of list_del/list_add_tail
Wei Yongjun [Wed, 5 Sep 2012 06:34:32 +0000 (14:34 +0800)]
ceph: use list_move_tail instead of list_del/list_add_tail

Using list_move_tail() instead of list_del() + list_add_tail().

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Sage Weil <sage@inktank.com>
11 years agorbd: add new snapshots at the tail
Alex Elder [Thu, 30 Aug 2012 05:16:37 +0000 (00:16 -0500)]
rbd: add new snapshots at the tail

This fixes a bug that went in with this commit:

    commit f6e0c99092cca7be00fca4080cfc7081739ca544
    Author: Alex Elder <elder@inktank.com>
    Date:   Thu Aug 2 11:29:46 2012 -0500
    rbd: simplify __rbd_init_snaps_header()

The problem is that a new rbd snapshot needs to go either after an
existing snapshot entry, or at the *end* of an rbd device's snapshot
list.  As originally coded, it is placed at the beginning.  This was
based on the assumption the list would be empty (so it wouldn't
matter), but in fact if multiple new snapshots are added to an empty
list in one shot the list will be non-empty after the first one is
added.

This addresses http://tracker.newdream.net/issues/3063

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: rename block_name -> object_prefix
Alex Elder [Fri, 31 Aug 2012 22:29:51 +0000 (17:29 -0500)]
rbd: rename block_name -> object_prefix

In the on-disk image header structure there is a field "block_name"
which represents what we now call the "object prefix" for an rbd
image.  Rename this field "object_prefix" to be consistent with
modern usage.

This appears to be the only remaining vestige of the use of "block"
in symbols that represent objects in the rbd code.

This addresses http://tracker.newdream.net/issues/1761

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
11 years agolibceph: Fix sparse warning
Iulius Curt [Thu, 23 Aug 2012 12:14:29 +0000 (15:14 +0300)]
libceph: Fix sparse warning

Make ceph_monc_do_poolop() static to remove the following sparse warning:
 * net/ceph/mon_client.c:616:5: warning: symbol 'ceph_monc_do_poolop' was not
   declared. Should it be static?
Also drops the 'ceph_monc_' prefix, now being a private function.

Signed-off-by: Iulius Curt <icurt@ixiacom.com>
Signed-off-by: Sage Weil <sage@inktank.com>
11 years agolibceph: remove unused monc->have_fsid
Sage Weil [Fri, 17 Aug 2012 16:47:49 +0000 (09:47 -0700)]
libceph: remove unused monc->have_fsid

This is unused; use monc->client->have_fsid.

Signed-off-by: Sage Weil <sage@inktank.com>
11 years agoceph: let path portion of mount "device" be optional
Alex Elder [Thu, 9 Aug 2012 17:33:26 +0000 (10:33 -0700)]
ceph: let path portion of mount "device" be optional

A recent change to /sbin/mountall causes any trailing '/' character
in the "device" (or fs_spec) field in /etc/fstab to be stripped.  As
a result, an entry for a ceph mount that intends to mount the root
of the name space ends up with now path portion, and the ceph mount
option processing code rejects this.

That is, an entry in /etc/fstab like:
    cephserver:port:/ /mnt ceph defaults 0 0
provides to the ceph code just "cephserver:port:" as the "device,"
and that gets rejected.

Although this is a bug in /sbin/mountall, we can have the ceph mount
code support an empty/nonexistent path, interpreting it to mean the
root of the name space.

RFC 5952 offers recommendations for how to express IPv6 addresses,
and recommends the usage found in RFC 3986 (which specifies the
format for URI's) for representing both IPv4 and IPv6 addresses that
include port numbers.  (See in particular the definition of
"authority" found in the Appendix of RFC 3986.)

According to those standards, no host specification will ever
contain a '/' character.  As a result, it is sufficient to scan a
provided "device" from an /etc/fstab entry for the first '/'
character, and if it's found, treat that as the beginning of the
path.  If no '/' character is present, we can treat the entire
string as the monitor host specification(s), and assume the path
to be the root of the name space.  We'll still require a ':' to
separate the host portion from the (possibly empty) path portion.

This means that we can more formally define how ceph will interpret
the "device" it's provided when processing a mount request:

    "device" will look like:
        <server_spec>[,<server_spec>...]:[<path>]
    where
        <server_spec> is <ip>[:<port>]
        <path> is optional, but if present must begin with '/'

This addresses http://tracker.newdream.net/issues/2919

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
11 years agorbd: separate reading header from decoding it
Alex Elder [Thu, 2 Aug 2012 16:29:46 +0000 (11:29 -0500)]
rbd: separate reading header from decoding it

Right now rbd_read_header() both reads the header object for an rbd
image and decodes its contents.  It does this repeatedly if needed,
in order to ensure a complete and intact header is obtained.

Separate this process into two steps--reading of the raw header
data (in new function, rbd_dev_v1_header_read()) and separately
decoding its contents (in rbd_header_from_disk()).  As a result,
the latter function no longer requires its allocated_snaps argument.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: expand rbd_dev_ondisk_valid() checks
Alex Elder [Thu, 2 Aug 2012 16:29:45 +0000 (11:29 -0500)]
rbd: expand rbd_dev_ondisk_valid() checks

Add checks on the validity of the snap_count and snap_names_len
field values in rbd_dev_ondisk_valid().  This eliminates the
need to do them in rbd_header_from_disk().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: return earlier in rbd_header_from_disk()
Alex Elder [Fri, 27 Jul 2012 04:37:15 +0000 (23:37 -0500)]
rbd: return earlier in rbd_header_from_disk()

The only caller of rbd_header_from_disk() is rbd_read_header().
It passes as allocated_snaps the number of snapshots it will
have received from the server for the snapshot context that
rbd_header_from_disk() is to interpret.  The first time through
it provides 0--mainly to extract the number of snapshots from
the snapshot context header--so that it can allocate an
appropriately-sized buffer to receive the entire snapshot
context from the server in a second request.

rbd_header_from_disk() will not fill in the array of snapshot ids
unless the number in the snapshot matches the number the caller
had allocated.

This patch adjusts that logic a little further to be more efficient.
rbd_read_header() doesn't even examine the snapshot context unless
the snapshot count (stored in header->total_snaps) matches the
number of snapshots allocated.  So rbd_header_from_disk() doesn't
need to allocate or fill in the snapshot context field at all in
that case.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: rearrange rbd_header_from_disk()
Alex Elder [Thu, 19 Jul 2012 22:12:59 +0000 (17:12 -0500)]
rbd: rearrange rbd_header_from_disk()

This just moves code around for the most part.  It was pulled out as
a separate patch to avoid cluttering up some upcoming patches which
are more substantive.  The point is basically to group everything
related to initializing the snapshot context together.

The only functional change is that rbd_header_from_disk() now
ensures the (in-core) header it is passed is zero-filled.  This
allows a simpler error handling path in rbd_header_from_disk().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: use sizeof (object) instead of sizeof (type)
Alex Elder [Fri, 27 Jul 2012 04:37:14 +0000 (23:37 -0500)]
rbd: use sizeof (object) instead of sizeof (type)

Fix a few spots in rbd_header_from_disk() to use sizeof (object)
rather than sizeof (type).  Use a local variable to record sizes
to shorten some lines and improve readability.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: ensure invalid pointers are made null
Alex Elder [Fri, 27 Jul 2012 04:37:14 +0000 (23:37 -0500)]
rbd: ensure invalid pointers are made null

Fix a number of spots where a pointer value that is known to
have become invalid but was not reset to null.

Also, toss in a change so we use sizeof (object) rather than
sizeof (type).

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: make snap_names_len a u64
Alex Elder [Thu, 2 Aug 2012 16:29:44 +0000 (11:29 -0500)]
rbd: make snap_names_len a u64

The snap_names_len field of an rbd_image_header structure is defined
with type size_t.  That field is used as both the source and target
of 64-bit byte-order swapping operations though, so it's best to
define it with type u64 instead.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agorbd: simplify __rbd_init_snaps_header()
Alex Elder [Thu, 2 Aug 2012 16:29:46 +0000 (11:29 -0500)]
rbd: simplify __rbd_init_snaps_header()

The purpose of __rbd_init_snaps_header() is to compare a new
snapshot context with an rbd device's list of existing snapshots.
It updates the list by adding any new snapshots or removing any
that are not present in the new snapshot context.

The code as written is a little confusing, because it traverses both
the existing snapshot list and the set of snapshots in the snapshot
context in reverse.  This was done based on an assumption about
snapshots that is not true--namely that a duplicate snapshot name
could cause an error in intepreting things if they were not
processed in ascending order.

These precautions are not necessary, because:
    - all snapshots are uniquely identified by their snapshot id
    - a new snapshot cannot be created if the rbd device has another
      snapshot with the same name
(It is furthermore not currently possible to rename a snapshot.)

This patch re-implements __rbd_init_snaps_header() so it passes
through both the existing snapshot list and the entries in the
snapshot context in forward order.  It still does the same thing
as before, but I find the logic considerably easier to understand.

By going forward through the names in the snapshot context, there
is no longer a need for the rbd_prev_snap_name() helper function.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
11 years agoLinux 3.6
Linus Torvalds [Sun, 30 Sep 2012 23:47:46 +0000 (16:47 -0700)]
Linux 3.6

11 years agovfs: dcache: fix deadlock in tree traversal
Miklos Szeredi [Mon, 17 Sep 2012 20:23:30 +0000 (22:23 +0200)]
vfs: dcache: fix deadlock in tree traversal

IBM reported a deadlock in select_parent().  This was found to be caused
by taking rename_lock when already locked when restarting the tree
traversal.

There are two cases when the traversal needs to be restarted:

 1) concurrent d_move(); this can only happen when not already locked,
    since taking rename_lock protects against concurrent d_move().

 2) racing with final d_put() on child just at the moment of ascending
    to parent; rename_lock doesn't protect against this rare race, so it
    can happen when already locked.

Because of case 2, we need to be able to handle restarting the traversal
when rename_lock is already held.  This patch fixes all three callers of
try_to_ascend().

IBM reported that the deadlock is gone with this patch.

[ I rewrote the patch to be smaller and just do the "goto again" if the
  lock was already held, but credit goes to Miklos for the real work.
   - Linus ]

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'iommu-fixes-v3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 29 Sep 2012 17:37:04 +0000 (10:37 -0700)]
Merge tag 'iommu-fixes-v3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu

Pull IOMMU fixes from Joerg Roedel:
 "Two small patches:

* One patch to fix the function declarations for
  !CONFIG_IOMMU_API. This is causing build errors
  in linux-next and should be fixed for v3.6.

* Another patch to fix an IOMMU group related NULL pointer
  dereference."

* tag 'iommu-fixes-v3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
  iommu/amd: Fix wrong assumption in iommu-group specific code
  iommu: static inline iommu group stub functions

11 years agoMerge git://git.infradead.org/users/willy/linux-nvme
Linus Torvalds [Sat, 29 Sep 2012 17:31:52 +0000 (10:31 -0700)]
Merge git://git.infradead.org/users/willy/linux-nvme

Pull NVMe driver fixes from Matthew Wilcox:
 "Now that actual hardware has been released (don't have any yet
  myself), people are starting to want some of these fixes merged."

Willy doesn't have hardware? Guys...

* git://git.infradead.org/users/willy/linux-nvme:
  NVMe: Cancel outstanding IOs on queue deletion
  NVMe: Free admin queue memory on initialisation failure
  NVMe: Use ida for nvme device instance
  NVMe: Fix whitespace damage in nvme_init
  NVMe: handle allocation failure in nvme_map_user_pages()
  NVMe: Fix uninitialized iod compiler warning
  NVMe: Do not set IO queue depth beyond device max
  NVMe: Set block queue max sectors
  NVMe: use namespace id for nvme_get_features
  NVMe: replace nvme_ns with nvme_dev for user admin
  NVMe: Fix nvme module init when nvme_major is set
  NVMe: Set request queue logical block size

11 years agomtdchar: fix offset overflow detection
Linus Torvalds [Sat, 8 Sep 2012 19:57:30 +0000 (12:57 -0700)]
mtdchar: fix offset overflow detection

Sasha Levin has been running trinity in a KVM tools guest, and was able
to trigger the BUG_ON() at arch/x86/mm/pat.c:279 (verifying the range of
the memory type).  The call trace showed that it was mtdchar_mmap() that
created an invalid remap_pfn_range().

The problem is that mtdchar_mmap() does various really odd and subtle
things with the vma page offset etc, and uses the wrong types (and the
wrong overflow) detection for it.

For example, the page offset may well be 32-bit on a 32-bit
architecture, but after shifting it up by PAGE_SHIFT, we need to use a
potentially 64-bit resource_size_t to correctly hold the full value.

Also, we need to check that the vma length plus offset doesn't overflow
before we check that it is smaller than the length of the mtdmap region.

This fixes things up and tries to make the code a bit easier to read.

Reported-and-tested-by: Sasha Levin <levinsasha928@gmail.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Artem Bityutskiy <dedekind1@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: linux-mtd@lists.infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 28 Sep 2012 17:09:33 +0000 (10:09 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

Pull networking fixes from David S Miller:

 1) Netfilter xt_limit module can use uninitialized rules, from Jan
    Engelhardt.

 2) Wei Yongjun has found several more spots where error pointers were
    treated as NULL/non-NULL and vice versa.

 3) bnx2x was converted to pci_io{,un}map() but one remaining plain
    iounmap() got missed.  From Neil Horman.

 4) Due to a fence-post type error in initialization of inetpeer entries
    (which is where we store the ICMP rate limiting information), we can
    erroneously drop ICMPs if the inetpeer was created right around when
    jiffies wraps.

    Fix from Nicolas Dichtel.

 5) smsc75xx resume fix from Steve Glendinnig.

 6) LAN87xx smsc chips need an explicit hardware init, from Marek Vasut.

 7) qlcnic uses msleep() with locks held, fix from Narendra K.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  netdev: octeon: fix return value check in octeon_mgmt_init_phy()
  inetpeer: fix token initialization
  qlcnic: Fix scheduling while atomic bug
  bnx2: Clean up remaining iounmap
  net: phy: smsc: Implement PHY config_init for LAN87xx
  smsc75xx: fix resume after device reset
  netdev: pasemi: fix return value check in pasemi_mac_phy_init()
  team: fix return value check
  l2tp: fix return value check
  netfilter: xt_limit: have r->cost != 0 case work

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Fri, 28 Sep 2012 17:02:53 +0000 (10:02 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "A couple of fixes; one for automount/lazy umount race, another a
  classic "we don't protect the refcount transition to zero with the
  lock that protects looking for object in hash" kind of crap in lockd."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  close the race in nlmsvc_free_block()
  do_add_mount()/umount -l races

11 years agoMerge branch 'for-linus-3.6-rc-final' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 28 Sep 2012 17:00:52 +0000 (10:00 -0700)]
Merge branch 'for-linus-3.6-rc-final' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml

Pull UML fixes from Richard Weinberger.

* 'for-linus-3.6-rc-final' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
  um: Preinclude include/linux/kern_levels.h
  um: Fix IPC on um
  um: kill thread->forking
  um: let signal_delivered() do SIGTRAP on singlestepping into handler
  um: don't leak floating point state and segment registers on execve()
  um: take cleaning singlestep to start_thread()

11 years agoMerge tag 'dm-3.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm
Linus Torvalds [Fri, 28 Sep 2012 17:00:01 +0000 (10:00 -0700)]
Merge tag 'dm-3.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm

Pull dm fixes from Alasdair G Kergon:
 "A few fixes for problems discovered during the 3.6 cycle.

  Of particular note, are fixes to the thin target's discard support,
  which I hope is finally working correctly; and fixes for multipath
  ioctls and device limits when there are no paths."

* tag 'dm-3.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm:
  dm verity: fix overflow check
  dm thin: fix discard support for data devices
  dm thin: tidy discard support
  dm: retain table limits when swapping to new table with no devices
  dm table: clear add_random unless all devices have it set
  dm: handle requests beyond end of device instead of using BUG_ON
  dm mpath: only retry ioctl when no paths if queue_if_no_path set
  dm thin: do not set discard_zeroes_data

11 years agothp: avoid VM_BUG_ON page_count(page) false positives in __collapse_huge_page_copy
Andrea Arcangeli [Fri, 28 Sep 2012 12:35:31 +0000 (14:35 +0200)]
thp: avoid VM_BUG_ON page_count(page) false positives in __collapse_huge_page_copy

Speculative cache pagecache lookups can elevate the refcount from
under us, so avoid the false positive. If the refcount is < 2 we'll be
notified by a VM_BUG_ON in put_page_testzero as there are two
put_page(src_page) in a row before returning from this function.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Petr Holasek <pholasek@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoiommu/amd: Fix wrong assumption in iommu-group specific code
Joerg Roedel [Fri, 28 Sep 2012 14:14:44 +0000 (16:14 +0200)]
iommu/amd: Fix wrong assumption in iommu-group specific code

The new IOMMU groups code in the AMD IOMMU driver makes the
assumption that there is a pci_dev struct available for all
device-ids listed in the IVRS ACPI table. Unfortunatly this
assumption is not true and so this code causes a NULL
pointer dereference at boot on some systems.

Fix it by making sure the given pointer is never NULL when
passed to the group specific code. The real fix is larger
and will be queued for v3.7.

Reported-by: Florian Dazinger <florian@dazinger.net>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
11 years agonetdev: octeon: fix return value check in octeon_mgmt_init_phy()
Wei Yongjun [Thu, 27 Sep 2012 19:04:21 +0000 (19:04 +0000)]
netdev: octeon: fix return value check in octeon_mgmt_init_phy()

In case of error, the function of_phy_connect() returns NULL
pointer not ERR_PTR(). The IS_ERR() test in the return value
check should be replaced with NULL test.

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Thu, 27 Sep 2012 23:51:14 +0000 (16:51 -0700)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "The three nouveau fixes quiten unneeded dmesg spam that people are
   seeing and pondering,

  The udl fix stops it from trying to driver monitors that are too big,
  where we get a black screen.

  And a vmware memory alloc problem."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/nvc0/fifo: ignore bits in PFIFO_INTR that aren't set in PFIFO_INTR_EN
  drm/udl: limit modes to the sku pixel limits.
  vmwgfx: corruption in vmw_event_fence_action_create()
  drm/nvc0/ltcg: mask off intr 0x10
  drm/nouveau: silence a debug message triggered by newer userspace

11 years agoMerge tag 'usb-3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Thu, 27 Sep 2012 23:49:15 +0000 (16:49 -0700)]
Merge tag 'usb-3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb

Pull USB fixes from Greg Kroah-Hartman:
 "Here are two USB bugfixes for your 3.6-rc7 tree.

  The OHCI fix has been reported a number of times and is a regression
  from 3.5, and the patch that causes the regression was on the way to
  the -stable trees before I was reminded (again) that this fix needed
  to get to your tree soon.

  The host controller bugfix was reported in older kernels as being
  pretty easy to trigger, and has been tested by Red Hat and their
  customers.

  Both have been in the usb-next branch in the -next tree for a while
  now, I just cherry-picked them out to get to you in time for the 3.6
  release.

Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>"
* tag 'usb-3.6-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  USB: Fix race condition when removing host controllers
  USB: ohci-at91: fix null pointer in ohci_hcd_at91_overcurrent_irq

11 years agoALSA: snd-usb: fix next_packet_size calls for pause case
Daniel Mack [Thu, 27 Sep 2012 08:26:01 +0000 (10:26 +0200)]
ALSA: snd-usb: fix next_packet_size calls for pause case

Also fix the calls to next_packet_size() for the pause case. This was
missed in 245baf983 ("ALSA: snd-usb: fix calls to next_packet_size").

Signed-off-by: Daniel Mack <zonque@gmail.com>
Reviewed-by: Takashi Iwai <tiwai@suse.de>
Reported-and-tested-by: Christian Tefzer <ctrefzer@gmx.de>
Cc: stable@kernel.org
[ Taking directly because Takashi is on vacation  - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'asoc-3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound
Linus Torvalds [Thu, 27 Sep 2012 23:42:35 +0000 (16:42 -0700)]
Merge tag 'asoc-3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound

Pull ASoC update from Mark Brown:
 "One small and obvious driver-specific fix.

  Takashi is on vacation now so he asked me to send directly, it's a
  pretty bad bug with low regression risk."

* tag 'asoc-3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound:
  ASoC: wm2000: Correct register size

11 years agoinetpeer: fix token initialization
Nicolas Dichtel [Thu, 27 Sep 2012 04:11:00 +0000 (04:11 +0000)]
inetpeer: fix token initialization

When jiffies wraps around (for example, 5 minutes after the boot, see
INITIAL_JIFFIES) and peer has just been created, now - peer->rate_last can be
< XRLIM_BURST_FACTOR * timeout, so token is not set to the maximum value, thus
some icmp packets can be unexpectedly dropped.

Fix this case by initializing last_rate to 60 seconds in the past.

Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoqlcnic: Fix scheduling while atomic bug
Narendra K [Tue, 25 Sep 2012 07:53:19 +0000 (07:53 +0000)]
qlcnic: Fix scheduling while atomic bug

In the device close path, 'qlcnic_fw_destroy_ctx' and
'qlcnic_poll_rsp' call msleep. But  'qlcnic_fw_destroy_ctx' and
'qlcnic_poll_rsp' are called with 'adapter->tx_clean_lock' spin lock
held resulting in scheduling while atomic bug causing the following
trace.

I observed that the commit 012dc19a45b2b9cc2ebd14aaa401cf782c2abba4
from John Fastabend addresses a similar issue in ixgbevf driver.
Adopting the same approach used in the commit, this patch uses mdelay
to address the issue.

[79884.999115] BUG: scheduling while atomic: ip/30846/0x00000002
[79885.005562] INFO: lockdep is turned off.
[79885.009958] Modules linked in: qlcnic fuse nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE bnep bluetooth rfkill ip6table_mangle ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables iptable_nat nf_nat iptable_mangle ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 xt_conntrack nf_conntrack iptable_filter ip_tables dcdbas coretemp kvm_intel kvm iTCO_wdt ixgbe iTCO_vendor_support crc32c_intel ghash_clmulni_intel nfsd microcode sb_edac pcspkr edac_core dca bnx2x shpchp auth_rpcgss nfs_acl lpc_ich mfd_core mdio lockd libcrc32c wmi acpi_pad acpi_power_meter sunrpc uinput sd_mod sr_mod cdrom crc_t10dif ahci libahci libata megaraid_sas usb_storage dm_mirror dm_region_hash dm_log dm_mod [last unloaded: qlcnic]
[79885.083608] Pid: 30846, comm: ip Tainted: G        W  O 3.6.0-rc7+ #1
[79885.090805] Call Trace:
[79885.093569]  [<ffffffff816764d8>] __schedule_bug+0x68/0x76
[79885.099699]  [<ffffffff8168358e>] __schedule+0x99e/0xa00
[79885.105634]  [<ffffffff81683929>] schedule+0x29/0x70
[79885.111186]  [<ffffffff81680def>] schedule_timeout+0x16f/0x350
[79885.117724]  [<ffffffff811afb7a>] ? init_object+0x4a/0x90
[79885.123770]  [<ffffffff8107c190>] ? __internal_add_timer+0x140/0x140
[79885.130873]  [<ffffffff81680fee>] schedule_timeout_uninterruptible+0x1e/0x20
[79885.138773]  [<ffffffff8107e830>] msleep+0x20/0x30
[79885.144159]  [<ffffffffa04c7fbf>] qlcnic_issue_cmd+0xef/0x290 [qlcnic]
[79885.151478]  [<ffffffffa04c8265>] qlcnic_fw_cmd_destroy_rx_ctx+0x55/0x90 [qlcnic]
[79885.159868]  [<ffffffffa04c92fd>] qlcnic_fw_destroy_ctx+0x2d/0xa0 [qlcnic]
[79885.167576]  [<ffffffffa04bf2ed>] __qlcnic_down+0x11d/0x180 [qlcnic]
[79885.174708]  [<ffffffffa04bf6f8>] qlcnic_close+0x18/0x20 [qlcnic]
[79885.181547]  [<ffffffff8153b4c5>] __dev_close_many+0x95/0xe0
[79885.187899]  [<ffffffff8153b548>] __dev_close+0x38/0x50
[79885.193761]  [<ffffffff81545101>] __dev_change_flags+0xa1/0x180
[79885.200419]  [<ffffffff81545298>] dev_change_flags+0x28/0x70
[79885.206779]  [<ffffffff815531b8>] do_setlink+0x378/0xa00
[79885.212731]  [<ffffffff81354fe1>] ? nla_parse+0x31/0xe0
[79885.218612]  [<ffffffff815558ee>] rtnl_newlink+0x37e/0x560
[79885.224768]  [<ffffffff812cfa19>] ? selinux_capable+0x39/0x50
[79885.231217]  [<ffffffff812cbf98>] ? security_capable+0x18/0x20
[79885.237765]  [<ffffffff81555114>] rtnetlink_rcv_msg+0x114/0x2f0
[79885.244412]  [<ffffffff81551f87>] ? rtnl_lock+0x17/0x20
[79885.250280]  [<ffffffff81551f87>] ? rtnl_lock+0x17/0x20
[79885.256148]  [<ffffffff81555000>] ? __rtnl_unlock+0x20/0x20
[79885.262413]  [<ffffffff81570fc1>] netlink_rcv_skb+0xa1/0xb0
[79885.268661]  [<ffffffff81551fb5>] rtnetlink_rcv+0x25/0x40
[79885.274727]  [<ffffffff815708bd>] netlink_unicast+0x19d/0x220
[79885.281146]  [<ffffffff81570c45>] netlink_sendmsg+0x305/0x3f0
[79885.287595]  [<ffffffff8152b188>] ? sock_update_classid+0x148/0x2e0
[79885.294650]  [<ffffffff81525c2c>] sock_sendmsg+0xbc/0xf0
[79885.300600]  [<ffffffff8152600c>] __sys_sendmsg+0x3ac/0x3c0
[79885.306853]  [<ffffffff8109be23>] ? up_read+0x23/0x40
[79885.312510]  [<ffffffff816896cc>] ? do_page_fault+0x2bc/0x570
[79885.318968]  [<ffffffff81191854>] ? sys_brk+0x44/0x150
[79885.324715]  [<ffffffff811c458c>] ? fget_light+0x24c/0x520
[79885.330875]  [<ffffffff815286f9>] sys_sendmsg+0x49/0x90
[79885.336707]  [<ffffffff8168e429>] system_call_fastpath+0x16/0x1b

Signed-off-by: Narendra K <narendra_k@dell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobnx2: Clean up remaining iounmap
Neil Horman [Wed, 26 Sep 2012 07:22:02 +0000 (07:22 +0000)]
bnx2: Clean up remaining iounmap

commit c0357e975afdbbedab5c662d19bef865f02adc17 modified bnx2 to switch from
using ioremap/iounmap to pci_iomap/pci_iounmap.  They missed a spot in the error
path of bnx2_init_one though.  This patch just cleans that up.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Michael Chan <mcan@broadcom.com>
CC: "David S. Miller" <davem@davemloft.net>
Acked-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Thu, 27 Sep 2012 22:47:24 +0000 (15:47 -0700)]
Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc

Pull one more arm-soc bugfix from Olof Johansson:
 "Here's a bugfix for orion5x.  Without this, PCI doesn't initialize
  properly because of too small coherent pool to cover the allocations
  needed.

  A similar fix has already been done on kirkwood."

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: Orion5x: Fix too small coherent pool.

11 years agoMerge branch 'fixes-for-3.6' of git://git.linaro.org/people/mszyprowski/linux-dma...
Linus Torvalds [Thu, 27 Sep 2012 22:46:04 +0000 (15:46 -0700)]
Merge branch 'fixes-for-3.6' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping

Pull ARM dma-mapping fix from Marek Szyprowski:
 "This patch fixes a potential memory leak in the ARM dma-mapping code."

* 'fixes-for-3.6' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping:
  ARM: dma-mapping: Fix potential memory leak in atomic_pool_init()

11 years agoMerge tag 'gpio-fixes-v3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Thu, 27 Sep 2012 22:45:20 +0000 (15:45 -0700)]
Merge tag 'gpio-fixes-v3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio

Pull GPIO fix from Linus Walleij:
 "A late GPIO fix: Roland Stigge found a problem in the LPC32xx driver
  where a callback ignores one of its arguments.  It needs to go into
  stable too so sending this upstream immediately."

* tag 'gpio-fixes-v3.6' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio:
  gpio-lpc32xx: Fix value handling of gpio_direction_output()

11 years agoMerge tag 'md-3.6-fixes' of git://neil.brown.name/md
Linus Torvalds [Thu, 27 Sep 2012 22:44:31 +0000 (15:44 -0700)]
Merge tag 'md-3.6-fixes' of git://neil.brown.name/md

Pull two md bugfixes from NeilBrown:
 "One (missing spinlock init) was only introduced recently.  The other
  has been present as long as raid10 has been supported, so is tagged
  for -stable."

* tag 'md-3.6-fixes' of git://neil.brown.name/md:
  md/raid10: fix "enough" function for detecting if array is failed.
  md/raid5: add missing spin_lock_init.

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac
Linus Torvalds [Thu, 27 Sep 2012 22:43:36 +0000 (15:43 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac

Pull EDAC fixes from Mauro Carvalho Chehab:
 "Three edac fixes at the memory enumeration logic:
        - i3200_edac: Fixes a regression at the memory rank size, when the
                memorias are dual-rank;
        - i5000_edac: Fix a longstanding bug when calculating the memory
                size: before Kernel 3.6, the memory size were right only
                with one specific configuration;
        - sb_edac: Fixes a bug since the initial release of the driver:
                with 16GB DIMMs, there's an overflow at the memory size,
                causing the number of pages per dimm (an unsigned value)
                to have the highest bit equal to 1, effectively mangling
                the memory size.

  The third bug can potentially affect the error decoding logic as well."

* git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac:
  sb_edac: Avoid overflow errors at memory size calculation
  i5000: Fix the memory size calculation with 2R memories
  i3200_edac: Fix memory rank size

11 years agotrivial select_parent documentation fix
J. Bruce Fields [Tue, 18 Sep 2012 20:35:51 +0000 (16:35 -0400)]
trivial select_parent documentation fix

"Search list for X" sounds like you're trying to find X on a list.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agonet: phy: smsc: Implement PHY config_init for LAN87xx
Marek Vasut [Tue, 25 Sep 2012 10:17:42 +0000 (10:17 +0000)]
net: phy: smsc: Implement PHY config_init for LAN87xx

The LAN8710/LAN8720 chips do have broken the "FlexPWR" smart power-saving
capability. Enabling it leads to the PHY not being able to detect Link when
cold-started without cable connected. Thus, make sure this is disabled.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Christian Hohnstaedt <chohnstaedt@innominate.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Otavio Salvador <otavio@ossystems.com.br>
Acked-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agosmsc75xx: fix resume after device reset
Steve Glendinning [Mon, 24 Sep 2012 04:42:59 +0000 (04:42 +0000)]
smsc75xx: fix resume after device reset

On some systems this device fails to properly resume after suspend,
this patch fixes it by running the usbnet_resume handler.

I suspect this also fixes this bug:

http://code.google.com/p/chromium-os/issues/detail?id=31871

Signed-off-by: Steve Glendinning <steve.glendinning@shawell.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoum: Preinclude include/linux/kern_levels.h
Geert Uytterhoeven [Thu, 16 Aug 2012 18:15:05 +0000 (20:15 +0200)]
um: Preinclude include/linux/kern_levels.h

The userspace part of UML uses the asm-offsets.h generator mechanism to
create definitions for UM_KERN_<LEVEL> that match the in-kernel
KERN_<LEVEL> constant definitions.

As of commit 04d2c8c83d0e3ac5f78aeede51babb3236200112 ("printk: convert
the format for KERN_<LEVEL> to a 2 byte pattern"), KERN_<LEVEL> is no
longer expanded to the literal '"<LEVEL>"', but to '"\001" "LEVEL"', i.e.
it contains two parts.

However, the combo of DEFINE_STR() in
arch/x86/um/shared/sysdep/kernel-offsets.h and sed-y in Kbuild doesn't
support string literals consisting of multiple parts. Hence for all
UM_KERN_<LEVEL> definitions, only the SOH character is retained in the actual
definition, while the remainder ends up in the comment. E.g. in
include/generated/asm-offsets.h we get

    #define UM_KERN_INFO "\001" /* "6" KERN_INFO */

instead of

    #define UM_KERN_INFO "\001" "6" /* KERN_INFO */

This causes spurious '^A' output in some kernel messages:

    Calibrating delay loop... 4640.76 BogoMIPS (lpj=23203840)
    pid_max: default: 32768 minimum: 301
    Mount-cache hash table entries: 256
    ^AChecking that host ptys support output SIGIO...Yes
    ^AChecking that host ptys support SIGIO on close...No, enabling workaround
    ^AUsing 2.6 host AIO
    NET: Registered protocol family 16
    bio: create slab <bio-0> at 0
    Switching to clocksource itimer

To fix this:
  - Move the mapping from UM_KERN_<LEVEL> to KERN_<LEVEL> from
    arch/um/include/shared/common-offsets.h to
    arch/um/include/shared/user.h, which is preincluded for all userspace
    parts,
  - Preinclude include/linux/kern_levels.h for all userspace parts, to
    obtain the in-kernel KERN_<LEVEL> constant definitions. This doesn't
    violate the kernel/userspace separation, as include/linux/kern_levels.h
    is self-contained and doesn't expose any other kernel internals.
  - Remove the now unused STR() and DEFINE_STR() macros.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
11 years agoum: Fix IPC on um
Richard Weinberger [Thu, 27 Sep 2012 18:10:57 +0000 (20:10 +0200)]
um: Fix IPC on um

commit c1d7e01d (ipc: use Kconfig options for __ARCH_WANT_[COMPAT_]IPC_PARSE_VERSION)
forgot UML and broke IPC on it.
Also UML has to select ARCH_WANT_IPC_PARSE_VERSION usin Kconfig.

Reported-and-tested-by: <Toralf Förster toralf.foerster@gmx.de>
Signed-off-by: Richard Weinberger <richard@nod.at>
11 years agonetdev: pasemi: fix return value check in pasemi_mac_phy_init()
Wei Yongjun [Wed, 26 Sep 2012 19:51:58 +0000 (19:51 +0000)]
netdev: pasemi: fix return value check in pasemi_mac_phy_init()

In case of error, the function of_phy_connect() returns NULL
pointer not ERR_PTR(). The IS_ERR() test in the return value
check should be replaced with NULL test.

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoteam: fix return value check
Wei Yongjun [Mon, 24 Sep 2012 18:29:35 +0000 (18:29 +0000)]
team: fix return value check

In case of error, the function genlmsg_put() returns NULL pointer
not ERR_PTR(). The IS_ERR() test in the return value check should
be replaced with NULL test.

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agol2tp: fix return value check
Wei Yongjun [Mon, 24 Sep 2012 18:29:01 +0000 (18:29 +0000)]
l2tp: fix return value check

In case of error, the function genlmsg_put() returns NULL pointer
not ERR_PTR(). The IS_ERR() test in the return value check should
be replaced with NULL test.

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'master' of git://1984.lsi.us.es/nf
David S. Miller [Thu, 27 Sep 2012 17:16:14 +0000 (13:16 -0400)]
Merge branch 'master' of git://1984.lsi.us.es/nf

Pablo Neira Ayuso says:

====================
If time allows, I'd appreciate if you can take the following fix
for the xt_limit match.

As Jan indicates, random things may occur while using the xt_limit
match due to use of uninitialized memory.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoUSB: Fix race condition when removing host controllers
Alan Stern [Wed, 26 Sep 2012 17:09:53 +0000 (13:09 -0400)]
USB: Fix race condition when removing host controllers

This patch (as1607) fixes a race that can occur if a USB host
controller is removed while a process is reading the
/sys/kernel/debug/usb/devices file.

The usb_device_read() routine uses the bus->root_hub pointer to
determine whether or not the root hub is registered.  The is not a
valid test, because the pointer is set before the root hub gets
registered and remains set even after the root hub is unregistered and
deallocated.  As a result, usb_device_read() or usb_device_dump() can
access freed memory, causing an oops.

The patch changes the test to use the hcd->rh_registered flag, which
does get set and cleared at the appropriate times.  It also makes sure
to hold the usb_bus_list_lock mutex while setting the flag, so that
usb_device_read() will become aware of new root hubs as soon as they
are registered.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Don Zickus <dzickus@redhat.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
11 years agoUSB: ohci-at91: fix null pointer in ohci_hcd_at91_overcurrent_irq
Joachim Eastwood [Sun, 23 Sep 2012 20:56:00 +0000 (22:56 +0200)]
USB: ohci-at91: fix null pointer in ohci_hcd_at91_overcurrent_irq

Fixes the following NULL pointer dereference:
[    7.740000] ohci_hcd: USB 1.1 'Open' Host Controller (OHCI) Driver
[    7.810000] Unable to handle kernel NULL pointer dereference at virtual address 00000028
[    7.810000] pgd = c3a38000
[    7.810000] [00000028] *pgd=23a8c831, *pte=00000000, *ppte=00000000
[    7.810000] Internal error: Oops: 17 [#1] PREEMPT ARM
[    7.810000] Modules linked in: ohci_hcd(+) regmap_i2c snd_pcm usbcore snd_page_alloc at91_cf snd_timer pcmcia_rsrc snd soundcore gpio_keys regmap_spi pcmcia_core usb_common nls_base
[    7.810000] CPU: 0    Not tainted  (3.6.0-rc6-mpa+ #264)
[    7.810000] PC is at __gpio_to_irq+0x18/0x40
[    7.810000] LR is at ohci_hcd_at91_overcurrent_irq+0x24/0xb4 [ohci_hcd]
[    7.810000] pc : [<c01392d4>]    lr : [<bf08f694>]    psr: 40000093
[    7.810000] sp : c3a11c40  ip : c3a11c50  fp : c3a11c4c
[    7.810000] r10: 00000000  r9 : c02dcd6e  r8 : fefff400
[    7.810000] r7 : 00000000  r6 : c02cc928  r5 : 00000030  r4 : c02dd168
[    7.810000] r3 : c02e7350  r2 : ffffffea  r1 : c02cc928  r0 : 00000000
[    7.810000] Flags: nZcv  IRQs off  FIQs on  Mode SVC_32  ISA ARM  Segment user
[    7.810000] Control: c000717f  Table: 23a38000  DAC: 00000015
[    7.810000] Process modprobe (pid: 285, stack limit = 0xc3a10270)
[    7.810000] Stack: (0xc3a11c40 to 0xc3a12000)
[    7.810000] 1c40: c3a11c6c c3a11c50 bf08f694 c01392cc c3a11c84 c2c38b00 c3806900 00000030
[    7.810000] 1c60: c3a11ca4 c3a11c70 c0051264 bf08f680 c3a11cac c3a11c80 c003e764 c3806900
[    7.810000] 1c80: c2c38b00 c02cb05c c02cb000 fefff400 c3806930 c3a11cf4 c3a11cbc c3a11ca8
[    7.810000] 1ca0: c005142c c005123c c3806900 c3805a00 c3a11cd4 c3a11cc0 c0053f24 c00513e4
[    7.810000] 1cc0: c3a11cf4 00000030 c3a11cec c3a11cd8 c005120c c0053e88 00000000 00000000
[    7.810000] 1ce0: c3a11d1c c3a11cf0 c00124d0 c00511e0 01400000 00000001 00000012 00000000
[    7.810000] 1d00: ffffffff c3a11d94 00000030 00000000 c3a11d34 c3a11d20 c005120c c0012438
[    7.810000] 1d20: c001dac4 00000012 c3a11d4c c3a11d38 c0009b08 c00511e0 c00523fc 60000013
[    7.810000] 1d40: c3a11d5c c3a11d50 c0008510 c0009ab4 c3a11ddc c3a11d60 c0008eb4 c00084f0
[    7.810000] 1d60: 00000000 00000030 00000000 00000080 60000013 bf08f670 c3806900 c2c38b00
[    7.810000] 1d80: 00000030 c3806930 00000000 c3a11ddc c3a11d88 c3a11da8 c0054190 c00523fc
[    7.810000] 1da0: 60000013 ffffffff c3a11dec c3a11db8 00000000 c2c38b00 bf08f670 c3806900
[    7.810000] 1dc0: 00000000 00000080 c02cc928 00000030 c3a11e0c c3a11de0 c0052764 c00520d8
[    7.810000] 1de0: c3a11dfc 00000000 00000000 00000002 bf090f61 00000004 c02cc930 c02cc928
[    7.810000] 1e00: c3a11e4c c3a11e10 bf090978 c005269c bf090f61 c02cc928 bf093000 c02dd170
[    7.810000] 1e20: c3a11e3c c02cc930 c02cc930 bf0911d0 bf0911d0 bf093000 c3a10000 00000000
[    7.810000] 1e40: c3a11e5c c3a11e50 c0155b7c bf090808 c3a11e7c c3a11e60 c0154690 c0155b6c
[    7.810000] 1e60: c02cc930 c02cc964 bf0911d0 c3a11ea0 c3a11e9c c3a11e80 c015484c c01545e8
[    7.810000] 1e80: 00000000 00000000 c01547e4 bf0911d0 c3a11ec4 c3a11ea0 c0152e58 c01547f4
[    7.810000] 1ea0: c381b88c c384ab10 c2c10540 bf0911d0 00000000 c02d7518 c3a11ed4 c3a11ec8
[    7.810000] 1ec0: c01544c0 c0152e0c c3a11efc c3a11ed8 c01536cc c01544b0 bf091075 c3a11ee8
[    7.810000] 1ee0: bf049af0 bf09120c bf0911d0 00000000 c3a11f1c c3a11f00 c0154e9c c0153628
[    7.810000] 1f00: bf049af0 bf09120c 000ae190 00000000 c3a11f2c c3a11f20 c0155f58 c0154e04
[    7.810000] 1f20: c3a11f44 c3a11f30 bf093054 c0155f1c 00000000 00006a4f c3a11f7c c3a11f48
[    7.810000] 1f40: c0008638 bf093010 bf09120c 000ae190 00000000 c00093c4 00006a4f bf09120c
[    7.810000] 1f60: 000ae190 00000000 c00093c4 00000000 c3a11fa4 c3a11f80 c004fdc4 c000859c
[    7.810000] 1f80: c3a11fa4 000ae190 00006a4f 00016eb8 000ad018 00000080 00000000 c3a11fa8
[    7.810000] 1fa0: c0009260 c004fd58 00006a4f 00016eb8 000ae190 00006a4f 000ae100 00000000
[    7.810000] 1fc0: 00006a4f 00016eb8 000ad018 00000080 000adba0 000ad208 00000000 000ad3d8
[    7.810000] 1fe0: beaf7ae8 beaf7ad8 000172b8 b6e4e940 20000010 000ae190 00000000 00000000
[    7.810000] Backtrace:
[    7.810000] [<c01392bc>] (__gpio_to_irq+0x0/0x40) from [<bf08f694>] (ohci_hcd_at91_overcurrent_irq+0x24/0xb4 [ohci_hcd])
[    7.810000] [<bf08f670>] (ohci_hcd_at91_overcurrent_irq+0x0/0xb4 [ohci_hcd]) from [<c0051264>] (handle_irq_event_percpu+0x38/0x1a8)
[    7.810000]  r6:00000030 r5:c3806900 r4:c2c38b00
[    7.810000] [<c005122c>] (handle_irq_event_percpu+0x0/0x1a8) from [<c005142c>] (handle_irq_event+0x58/0x7c)
[    7.810000] [<c00513d4>] (handle_irq_event+0x0/0x7c) from [<c0053f24>] (handle_simple_irq+0xac/0xd8)
[    7.810000]  r5:c3805a00 r4:c3806900
[    7.810000] [<c0053e78>] (handle_simple_irq+0x0/0xd8) from [<c005120c>] (generic_handle_irq+0x3c/0x48)
[    7.810000]  r4:00000030
[    7.810000] [<c00511d0>] (generic_handle_irq+0x0/0x48) from [<c00124d0>] (gpio_irq_handler+0xa8/0xfc)
[    7.810000]  r4:00000000
[    7.810000] [<c0012428>] (gpio_irq_handler+0x0/0xfc) from [<c005120c>] (generic_handle_irq+0x3c/0x48)
[    7.810000] [<c00511d0>] (generic_handle_irq+0x0/0x48) from [<c0009b08>] (handle_IRQ+0x64/0x88)
[    7.810000]  r4:00000012
[    7.810000] [<c0009aa4>] (handle_IRQ+0x0/0x88) from [<c0008510>] (at91_aic_handle_irq+0x30/0x38)
[    7.810000]  r5:60000013 r4:c00523fc
[    7.810000] [<c00084e0>] (at91_aic_handle_irq+0x0/0x38) from [<c0008eb4>] (__irq_svc+0x34/0x60)
[    7.810000] Exception stack(0xc3a11d60 to 0xc3a11da8)
[    7.810000] 1d60: 00000000 00000030 00000000 00000080 60000013 bf08f670 c3806900 c2c38b00
[    7.810000] 1d80: 00000030 c3806930 00000000 c3a11ddc c3a11d88 c3a11da8 c0054190 c00523fc
[    7.810000] 1da0: 60000013 ffffffff
[    7.810000] [<c00520c8>] (__setup_irq+0x0/0x458) from [<c0052764>] (request_threaded_irq+0xd8/0x134)
[    7.810000] [<c005268c>] (request_threaded_irq+0x0/0x134) from [<bf090978>] (ohci_hcd_at91_drv_probe+0x180/0x41c [ohci_hcd])
[    7.810000] [<bf0907f8>] (ohci_hcd_at91_drv_probe+0x0/0x41c [ohci_hcd]) from [<c0155b7c>] (platform_drv_probe+0x20/0x24)
[    7.810000] [<c0155b5c>] (platform_drv_probe+0x0/0x24) from [<c0154690>] (driver_probe_device+0xb8/0x20c)
[    7.810000] [<c01545d8>] (driver_probe_device+0x0/0x20c) from [<c015484c>] (__driver_attach+0x68/0x88)
[    7.810000]  r7:c3a11ea0 r6:bf0911d0 r5:c02cc964 r4:c02cc930
[    7.810000] [<c01547e4>] (__driver_attach+0x0/0x88) from [<c0152e58>] (bus_for_each_dev+0x5c/0x9c)
[    7.810000]  r6:bf0911d0 r5:c01547e4 r4:00000000
[    7.810000] [<c0152dfc>] (bus_for_each_dev+0x0/0x9c) from [<c01544c0>] (driver_attach+0x20/0x28)
[    7.810000]  r7:c02d7518 r6:00000000 r5:bf0911d0 r4:c2c10540
[    7.810000] [<c01544a0>] (driver_attach+0x0/0x28) from [<c01536cc>] (bus_add_driver+0xb4/0x22c)
[    7.810000] [<c0153618>] (bus_add_driver+0x0/0x22c) from [<c0154e9c>] (driver_register+0xa8/0x144)
[    7.810000]  r7:00000000 r6:bf0911d0 r5:bf09120c r4:bf049af0
[    7.810000] [<c0154df4>] (driver_register+0x0/0x144) from [<c0155f58>] (platform_driver_register+0x4c/0x60)
[    7.810000]  r7:00000000 r6:000ae190 r5:bf09120c r4:bf049af0
[    7.810000] [<c0155f0c>] (platform_driver_register+0x0/0x60) from [<bf093054>] (ohci_hcd_mod_init+0x54/0x8c [ohci_hcd])
[    7.810000] [<bf093000>] (ohci_hcd_mod_init+0x0/0x8c [ohci_hcd]) from [<c0008638>] (do_one_initcall+0xac/0x174)
[    7.810000]  r4:00006a4f
[    7.810000] [<c000858c>] (do_one_initcall+0x0/0x174) from [<c004fdc4>] (sys_init_module+0x7c/0x1a0)
[    7.810000] [<c004fd48>] (sys_init_module+0x0/0x1a0) from [<c0009260>] (ret_fast_syscall+0x0/0x2c)
[    7.810000]  r7:00000080 r6:000ad018 r5:00016eb8 r4:00006a4f
[    7.810000] Code: e24cb004 e59f3028 e1a02000 e7930180 (e5903028)
[    7.810000] ---[ end trace 85aa37ed128143b5 ]---
[    7.810000] Kernel panic - not syncing: Fatal exception in interrupt

Commit 6fffb77c (USB: ohci-at91: fix PIO handling in relation with number of
ports) started setting unused pins to EINVAL. But this exposed a bug in the
ohci_hcd_at91_overcurrent_irq function where the gpio was used without being
checked to see if it is valid.

This patches fixed the issue by adding the gpio valid check.

Signed-off-by: Joachim Eastwood <joachim.eastwood@jotron.com>
Cc: stable <stable@vger.kernel.org> # [3.4+] whereever 6fffb77c went
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
11 years agoum: kill thread->forking
Al Viro [Thu, 20 Sep 2012 13:28:25 +0000 (09:28 -0400)]
um: kill thread->forking

we only use that to tell copy_thread() done by syscall from that
done by kernel_thread().  However, it's easier to do simply by
checking PF_KTHREAD in thread flags.

Merge sys_clone() guts for 32bit and 64bit, while we are at it...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoum: let signal_delivered() do SIGTRAP on singlestepping into handler
Al Viro [Thu, 6 Sep 2012 17:39:47 +0000 (13:39 -0400)]
um: let signal_delivered() do SIGTRAP on singlestepping into handler

... rather than duplicating that in sigframe setup code (and doing that
inconsistently, at that)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoum: don't leak floating point state and segment registers on execve()
Al Viro [Thu, 6 Sep 2012 03:20:33 +0000 (23:20 -0400)]
um: don't leak floating point state and segment registers on execve()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoum: take cleaning singlestep to start_thread()
Al Viro [Mon, 3 Sep 2012 07:24:18 +0000 (03:24 -0400)]
um: take cleaning singlestep to start_thread()

... assuming it's needed to be done at all

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
11 years agoMerge branch 'drm-nouveau-fixes' of git://anongit.freedesktop.org/git/nouveau/linux...
Dave Airlie [Thu, 27 Sep 2012 07:58:53 +0000 (17:58 +1000)]
Merge branch 'drm-nouveau-fixes' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-fixes

Another spurious dmesg quitening.

* 'drm-nouveau-fixes' of git://anongit.freedesktop.org/git/nouveau/linux-2.6:
  drm/nvc0/fifo: ignore bits in PFIFO_INTR that aren't set in PFIFO_INTR_EN

11 years agomd/raid10: fix "enough" function for detecting if array is failed.
NeilBrown [Thu, 27 Sep 2012 02:35:21 +0000 (12:35 +1000)]
md/raid10: fix "enough" function for detecting if array is failed.

The 'enough' function is written to work with 'near' arrays only
in that is implicitly assumes that the offset from one 'group' of
devices to the next is the same as the number of copies.
In reality it is the number of 'near' copies.

So change it to make this number explicit.

This bug makes it possible to run arrays without enough drives
present, which is dangerous.
It is appropriate for an -stable kernel, but will almost certainly
need to be modified for some of them.

Cc: stable@vger.kernel.org
Reported-by: Jakub Husák <jakub@gooseman.cz>
Signed-off-by: NeilBrown <neilb@suse.de>
11 years agoARM: Orion5x: Fix too small coherent pool.
Andrew Lunn [Mon, 24 Sep 2012 05:54:33 +0000 (07:54 +0200)]
ARM: Orion5x: Fix too small coherent pool.

Some Orion5x devices allocate their coherent buffers from atomic
context. Increase size of atomic coherent pool to make sure such the
allocations won't fail during boot.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Jason Cooper <jason@lakedaemon.net>
Signed-off-by: Olof Johansson <olof@lixom.net>
11 years agodrm/nvc0/fifo: ignore bits in PFIFO_INTR that aren't set in PFIFO_INTR_EN
Ben Skeggs [Wed, 26 Sep 2012 23:13:43 +0000 (09:13 +1000)]
drm/nvc0/fifo: ignore bits in PFIFO_INTR that aren't set in PFIFO_INTR_EN

PFIFO_INTR = 0x40000000 appears to be a normal case on nvc0/nve0 PFIFO,
the binary driver appears to completely ignore it in its PFIFO interrupt
handler and even masks off the bit (as we do) in PFIFO_INTR_EN at init
time.

The bits still light up in the hardware sometimes though, so lets just
ignore any bits we haven't explicitely requested.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodm verity: fix overflow check
Mikulas Patocka [Wed, 26 Sep 2012 22:45:48 +0000 (23:45 +0100)]
dm verity: fix overflow check

This patch fixes sector_t overflow checking in dm-verity.

Without this patch, the code checks for overflow only if sector_t is
smaller than long long, not if sector_t and long long have the same size.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm thin: fix discard support for data devices
Mike Snitzer [Wed, 26 Sep 2012 22:45:47 +0000 (23:45 +0100)]
dm thin: fix discard support for data devices

The discard limits that get established for a thin-pool or thin device
may be incompatible with the pool's data device.  Avoid this by checking
the discard limits of the pool's data device.  If an incompatibility is
found then the pool's 'discard passdown' feature is disabled.

Change thin_io_hints to ensure that a thin device always uses the same
queue limits as its pool device.

Introduce requested_pf to track whether or not the table line originally
contained the no_discard_passdown flag and use this directly for table
output.  We prepare the correct setting for discard_passdown directly in
bind_control_target (called from pool_io_hints) and store it in
adjusted_pf rather than waiting until we have access to pool->pf in
pool_preresume.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm thin: tidy discard support
Mike Snitzer [Wed, 26 Sep 2012 22:45:46 +0000 (23:45 +0100)]
dm thin: tidy discard support

A little thin discard code refactoring to make the next patch (dm thin:
fix discard support for data devices) more readable.
Pull out a couple of functions (and uses bools instead of unsigned for
features).

No functional changes.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm: retain table limits when swapping to new table with no devices
Mike Snitzer [Wed, 26 Sep 2012 22:45:45 +0000 (23:45 +0100)]
dm: retain table limits when swapping to new table with no devices

Add a safety net that will re-use the DM device's existing limits in the
event that DM device has a temporary table that doesn't have any
component devices.  This is to reduce the chance that requests not
respecting the hardware limits will reach the device.

DM recalculates queue limits based only on devices which currently exist
in the table.  This creates a problem in the event all devices are
temporarily removed such as all paths being lost in multipath.  DM will
reset the limits to the maximum permissible, which can then assemble
requests which exceed the limits of the paths when the paths are
restored.  The request will fail the blk_rq_check_limits() test when
sent to a path with lower limits, and will be retried without end by
multipath.  This became a much bigger issue after v3.6 commit fe86cdcef
("block: do not artificially constrain max_sectors for stacking
drivers").

Reported-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm table: clear add_random unless all devices have it set
Milan Broz [Wed, 26 Sep 2012 22:45:43 +0000 (23:45 +0100)]
dm table: clear add_random unless all devices have it set

Always clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
have it set. Otherwise devices with predictable characteristics may
contribute entropy.

QUEUE_FLAG_ADD_RANDOM specifies whether or not queue IO timings
contribute to the random pool.

For bio-based targets this flag is always 0 because such devices have no
real queue.

For request-based devices this flag was always set to 1 by default.

Now set it according to the flags on underlying devices. If there is at
least one device which should not contribute, set the flag to zero: If a
device, such as fast SSD storage, is not suitable for supplying entropy,
a request-based queue stacked over it will not be either.

Because the checking logic is exactly same as for the rotational flag,
share the iteration function with device_is_nonrot().

Signed-off-by: Milan Broz <mbroz@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm: handle requests beyond end of device instead of using BUG_ON
Mike Snitzer [Wed, 26 Sep 2012 22:45:42 +0000 (23:45 +0100)]
dm: handle requests beyond end of device instead of using BUG_ON

The access beyond the end of device BUG_ON that was introduced to
dm_request_fn via commit 29e4013de7ad950280e4b2208 ("dm: implement
REQ_FLUSH/FUA support for request-based dm") was an overly
drastic (but simple) response to this situation.

I have received a report that this BUG_ON was hit and now think
it would be better to use dm_kill_unmapped_request() to fail the clone
and original request with -EIO.

map_request() will assign the valid target returned by
dm_table_find_target to tio->ti.  But when the target
isn't valid tio->ti is never assigned (because map_request isn't
called); so add a check for tio->ti != NULL to dm_done().

Reported-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Cc: stable@vger.kernel.org # v2.6.37+
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm mpath: only retry ioctl when no paths if queue_if_no_path set
Mike Snitzer [Wed, 26 Sep 2012 22:45:41 +0000 (23:45 +0100)]
dm mpath: only retry ioctl when no paths if queue_if_no_path set

When there are no paths and multipath receives an ioctl, it waits until
a path becomes available.  This behaviour is incorrect if the
"queue_if_no_path" setting was not specified, as then the ioctl should
be rejected immediately, which this patch now does.

commit 35991652b ("dm mpath: allow ioctls to trigger pg init") should
have checked if queue_if_no_path was configured before queueing IO.

Checking for the queue_if_no_path feature, like is done in map_io(),
allows the following table load to work without blocking in the
multipath_ioctl retry loop:

  echo "0 1024 multipath 0 0 0 0" | dmsetup create mpath_nodevs

Without this fix the multipath_ioctl will block with the following stack
trace:

  blkid           D 0000000000000002     0 23936      1 0x00000000
   ffff8802b89e5cd8 0000000000000082 ffff8802b89e5fd8 0000000000012440
   ffff8802b89e4010 0000000000012440 0000000000012440 0000000000012440
   ffff8802b89e5fd8 0000000000012440 ffff88030c2aab30 ffff880325794040
  Call Trace:
   [<ffffffff814ce099>] schedule+0x29/0x70
   [<ffffffff814cc312>] schedule_timeout+0x182/0x2e0
   [<ffffffff8104dee0>] ? lock_timer_base+0x70/0x70
   [<ffffffff814cc48e>] schedule_timeout_uninterruptible+0x1e/0x20
   [<ffffffff8104f840>] msleep+0x20/0x30
   [<ffffffffa0000839>] multipath_ioctl+0x109/0x170 [dm_multipath]
   [<ffffffffa06bfb9c>] dm_blk_ioctl+0xbc/0xd0 [dm_mod]
   [<ffffffff8122a408>] __blkdev_driver_ioctl+0x28/0x30
   [<ffffffff8122a79e>] blkdev_ioctl+0xce/0x730
   [<ffffffff811970ac>] block_ioctl+0x3c/0x40
   [<ffffffff8117321c>] do_vfs_ioctl+0x8c/0x340
   [<ffffffff81166293>] ? sys_newfstat+0x33/0x40
   [<ffffffff81173571>] sys_ioctl+0xa1/0xb0
   [<ffffffff814d70a9>] system_call_fastpath+0x16/0x1b

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 3.5+
Acked-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agodm thin: do not set discard_zeroes_data
Mike Snitzer [Wed, 26 Sep 2012 22:45:39 +0000 (23:45 +0100)]
dm thin: do not set discard_zeroes_data

The dm thin pool target claims to support the zeroing of discarded
data areas.  This turns out to be incorrect when processing discards
that do not exactly cover a complete number of blocks, so the target
must always set discard_zeroes_data_unsupported.

The thin pool target will zero blocks when they are allocated if the
skip_block_zeroing feature is not specified.  The block layer
may send a discard that only partly covers a block.  If a thin pool
block is partially discarded then there is no guarantee that the
discarded data will get zeroed before it is accessed again.
Due to this, thin devices cannot claim discards will always zero data.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Cc: stable@vger.kernel.org # 3.4+
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
11 years agoMerge tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming
Linus Torvalds [Wed, 26 Sep 2012 21:28:17 +0000 (14:28 -0700)]
Merge tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming

Pull c6x arch fixes from Mark Salter:
  - Add __NR_kcmp to generic syscall list
  - C6X: Use generic asm/barrier.h

* tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming:
  syscalls: add __NR_kcmp syscall to generic unistd.h
  c6x: use asm-generic/barrier.h