GitHub/exynos8895/android_kernel_samsung_universal8895.git
10 years agoceph: reserve caps for file layout/lock MDS requests
Yan, Zheng [Sun, 27 Apr 2014 01:17:45 +0000 (09:17 +0800)]
ceph: reserve caps for file layout/lock MDS requests

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: avoid releasing caps that are being used
Yan, Zheng [Thu, 17 Apr 2014 00:02:02 +0000 (08:02 +0800)]
ceph: avoid releasing caps that are being used

To avoid releasing caps that are being used, encode_inode_release()
should send implemented caps to MDS.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: clear directory's completeness when creating file
Yan, Zheng [Mon, 14 Apr 2014 05:13:02 +0000 (13:13 +0800)]
ceph: clear directory's completeness when creating file

When creating a file, ceph_set_dentry_offset() puts the new dentry
at the end of directory's d_subdirs, then set the dentry's offset
based on directory's max offset. The offset does not reflect the
real postion of the dentry in directory. Later readdir reply from
MDS may change the dentry's position/offset. This inconsistency
can cause missing/duplicate entries in readdir result if readdir
is partly satisfied by dcache_readdir().

The fix is clear directory's completeness after creating/renaming
file. It prevents later readdir from using dcache_readdir().

Fixes: http://tracker.ceph.com/issues/8025
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agolibceph: fix non-default values check in apply_primary_affinity()
Ilya Dryomov [Thu, 10 Apr 2014 14:09:41 +0000 (18:09 +0400)]
libceph: fix non-default values check in apply_primary_affinity()

osd_primary_affinity array is indexed into incorrectly when checking
for non-default primary-affinity values.  This nullifies the impact of
the rest of the apply_primary_affinity() and results in misdirected
requests.

                if (osds[i] != CRUSH_ITEM_NONE &&
                    osdmap->osd_primary_affinity[i] !=
                                                ^^^
                                        CEPH_OSD_DEFAULT_PRIMARY_AFFINITY) {

For a pool with size 2, this always ends up checking osd0 and osd1
primary_affinity values, instead of the values that correspond to the
osds in question.  E.g., given a [2,3] up set and a [max,max,0,max]
primary affinity vector, requests are still sent to osd2, because both
osd0 and osd1 happen to have max primary_affinity values and therefore
we return from apply_primary_affinity() early on the premise that all
osds in the given set have max (default) values.  Fix it.

Fixes: http://tracker.ceph.com/issues/7954

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: use fpos_cmp() to compare dentry positions
Yan, Zheng [Wed, 9 Apr 2014 01:35:19 +0000 (09:35 +0800)]
ceph: use fpos_cmp() to compare dentry positions

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: check directory's completeness before emitting directory entry
Yan, Zheng [Tue, 8 Apr 2014 13:42:59 +0000 (21:42 +0800)]
ceph: check directory's completeness before emitting directory entry

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: skip invalid dentry during dcache readdir
Yan, Zheng [Sun, 6 Apr 2014 06:10:04 +0000 (14:10 +0800)]
ceph: skip invalid dentry during dcache readdir

skip dentries that were added before MDS issued FILE_SHARED to
client.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agolibceph: dump pool {read,write}_tier to debugfs
Ilya Dryomov [Fri, 4 Apr 2014 14:21:32 +0000 (18:21 +0400)]
libceph: dump pool {read,write}_tier to debugfs

Dump pool {read,write}_tier to debugfs.  While at it, fixup printk type
specifiers and remove the unnecessary cast to unsigned long long.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
10 years agolibceph: output primary affinity values on osdmap updates
Ilya Dryomov [Wed, 2 Apr 2014 16:34:04 +0000 (20:34 +0400)]
libceph: output primary affinity values on osdmap updates

Similar to osd weights, output primary affinity values on incremental
osdmap updates.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
10 years agoceph: flush cap release queue when trimming session caps
Yan, Zheng [Tue, 1 Apr 2014 12:34:56 +0000 (20:34 +0800)]
ceph: flush cap release queue when trimming session caps

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: don't grabs open file reference for aborted request
Yan, Zheng [Tue, 1 Apr 2014 12:34:18 +0000 (20:34 +0800)]
ceph: don't grabs open file reference for aborted request

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: drop extra open file reference in ceph_atomic_open()
Yan, Zheng [Tue, 1 Apr 2014 12:20:17 +0000 (20:20 +0800)]
ceph: drop extra open file reference in ceph_atomic_open()

ceph_atomic_open() calls ceph_open() after receiving the MDS reply.
ceph_open() grabs an extra open file reference. (The open request
already holds an open file reference)

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: preallocate buffer for readdir reply
Yan, Zheng [Sat, 29 Mar 2014 05:41:15 +0000 (13:41 +0800)]
ceph: preallocate buffer for readdir reply

Preallocate buffer for readdir reply. Limit number of entries in
readdir reply according to the buffer size.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agolibceph: enable PRIMARY_AFFINITY feature bit
Ilya Dryomov [Mon, 24 Mar 2014 15:12:50 +0000 (17:12 +0200)]
libceph: enable PRIMARY_AFFINITY feature bit

Announce our support for osdmaps with non-default primary affinity
values.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: redo ceph_calc_pg_primary() in terms of ceph_calc_pg_acting()
Ilya Dryomov [Mon, 24 Mar 2014 15:12:49 +0000 (17:12 +0200)]
libceph: redo ceph_calc_pg_primary() in terms of ceph_calc_pg_acting()

Reimplement ceph_calc_pg_primary() in terms of ceph_calc_pg_acting()
and get rid of the now unused calc_pg_raw().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: add support for osd primary affinity
Ilya Dryomov [Mon, 24 Mar 2014 15:12:49 +0000 (17:12 +0200)]
libceph: add support for osd primary affinity

Respond to non-default primary_affinity values accordingly.  (Primary
affinity allows the admin to shift 'primary responsibility' away from
specific osds, effectively shifting around the read side of the
workload and whatever overhead is incurred by peering and writes by
virtue of being the primary).

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: add support for primary_temp mappings
Ilya Dryomov [Mon, 24 Mar 2014 15:12:48 +0000 (17:12 +0200)]
libceph: add support for primary_temp mappings

Change apply_temp() to override primary in the same way pg_temp
overrides osd set.  primary_temp overrides pg_temp primary too.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: return primary from ceph_calc_pg_acting()
Ilya Dryomov [Mon, 24 Mar 2014 15:12:48 +0000 (17:12 +0200)]
libceph: return primary from ceph_calc_pg_acting()

In preparation for adding support for primary_temp, stop assuming
primaryness: add a primary out parameter to ceph_calc_pg_acting() and
change call sites accordingly.  Primary is now specified separately
from the order of osds in the set.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: switch ceph_calc_pg_acting() to new helpers
Ilya Dryomov [Mon, 24 Mar 2014 15:12:48 +0000 (17:12 +0200)]
libceph: switch ceph_calc_pg_acting() to new helpers

Switch ceph_calc_pg_acting() to new helpers: pg_to_raw_osds(),
raw_to_up_osds() and apply_temps().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: introduce apply_temps() helper
Ilya Dryomov [Mon, 24 Mar 2014 15:12:47 +0000 (17:12 +0200)]
libceph: introduce apply_temps() helper

apply_temp() helper for applying various temporary mappings (at this
point only pg_temp mappings) to the up set, therefore transforming it
into an acting set.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: introduce pg_to_raw_osds() and raw_to_up_osds() helpers
Ilya Dryomov [Mon, 24 Mar 2014 15:12:47 +0000 (17:12 +0200)]
libceph: introduce pg_to_raw_osds() and raw_to_up_osds() helpers

pg_to_raw_osds() helper for computing a raw (crush) set, which can
contain non-existant and down osds.

raw_to_up_osds() helper for pruning non-existant and down osds from the
raw set, therefore transforming it into an up set, and determining up
primary.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: ceph_can_shift_osds(pool) and pool type defines
Ilya Dryomov [Mon, 24 Mar 2014 15:12:47 +0000 (17:12 +0200)]
libceph: ceph_can_shift_osds(pool) and pool type defines

Bring in pg_pool_t::can_shift_osds() counterpart along with pool type
defines.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: ceph_osd_{exists,is_up,is_down}(osd) definitions
Ilya Dryomov [Mon, 24 Mar 2014 15:12:46 +0000 (17:12 +0200)]
libceph: ceph_osd_{exists,is_up,is_down}(osd) definitions

Sync up with ceph.git definitions.  Bring in ceph_osd_is_down().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: enable OSDMAP_ENC feature bit
Ilya Dryomov [Fri, 21 Mar 2014 17:05:31 +0000 (19:05 +0200)]
libceph: enable OSDMAP_ENC feature bit

Announce our support for "new" (v7 - split and separately versioned
client and osd sections) osdmap enconding.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: primary_affinity decode bits
Ilya Dryomov [Fri, 21 Mar 2014 17:05:30 +0000 (19:05 +0200)]
libceph: primary_affinity decode bits

Add two helpers to decode primary_affinity (full map, vector<u32>) and
new_primary_affinity (inc map, map<u32, u32>) and switch to them.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: primary_affinity infrastructure
Ilya Dryomov [Fri, 21 Mar 2014 17:05:30 +0000 (19:05 +0200)]
libceph: primary_affinity infrastructure

Add primary_affinity infrastructure.  primary_affinity values are
stored in an max_osd-sized array, hanging off ceph_osdmap, similar to
a osd_weight array.

Introduce {get,set}_primary_affinity() helpers, primarily to return
CEPH_OSD_DEFAULT_PRIMARY_AFFINITY when no affinity has been set and to
abstract out osd_primary_affinity array allocation and initialization.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: primary_temp decode bits
Ilya Dryomov [Fri, 21 Mar 2014 17:05:30 +0000 (19:05 +0200)]
libceph: primary_temp decode bits

Add a common helper to decode both primary_temp (full map, map<pg_t,
u32>) and new_primary_temp (inc map, same) and switch to it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: primary_temp infrastructure
Ilya Dryomov [Fri, 21 Mar 2014 17:05:29 +0000 (19:05 +0200)]
libceph: primary_temp infrastructure

Add primary_temp mappings infrastructure.  struct ceph_pg_mapping is
overloaded, primary_temp mappings are stored in an rb-tree, rooted at
ceph_osdmap, in a manner similar to pg_temp mappings.

Dump primary_temp mappings to /sys/kernel/debug/ceph/<client>/osdmap,
one 'primary_temp <pgid> <osd>' per line, e.g:

    primary_temp 2.6 4

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: generalize ceph_pg_mapping
Ilya Dryomov [Fri, 21 Mar 2014 17:05:29 +0000 (19:05 +0200)]
libceph: generalize ceph_pg_mapping

In preparation for adding support for primary_temp mappings, generalize
struct ceph_pg_mapping so it can hold mappings other than pg_temp.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: introduce get_osdmap_client_data_v()
Ilya Dryomov [Fri, 21 Mar 2014 17:05:29 +0000 (19:05 +0200)]
libceph: introduce get_osdmap_client_data_v()

Full and incremental osdmaps are structured identically and have
identical headers.  Add a helper to decode both "old" (16-bit version,
v6) and "new" (8-bit struct_v+struct_compat+struct_len, v7) osdmap
enconding headers and switch to it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: introduce decode{,_new}_pg_temp() and switch to them
Ilya Dryomov [Fri, 21 Mar 2014 17:05:28 +0000 (19:05 +0200)]
libceph: introduce decode{,_new}_pg_temp() and switch to them

Consolidate pg_temp (full map, map<pg_t, vector<u32>>) and new_pg_temp
(inc map, same) decoding logic into a common helper and switch to it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: switch osdmap_set_max_osd() to krealloc()
Ilya Dryomov [Fri, 21 Mar 2014 17:05:28 +0000 (19:05 +0200)]
libceph: switch osdmap_set_max_osd() to krealloc()

Use krealloc() instead of rolling our own.  (krealloc() with a NULL
first argument acts as a kmalloc()).  Properly initalize the new array
elements.  This is needed to make future additions to osdmap easier.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: introduce decode{,_new}_pools() and switch to them
Ilya Dryomov [Fri, 21 Mar 2014 17:05:27 +0000 (19:05 +0200)]
libceph: introduce decode{,_new}_pools() and switch to them

Consolidate pools (full map, map<u64, pg_pool_t>) and new_pools (inc
map, same) decoding logic into a common helper and switch to it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: rename __decode_pool{,_names}() to decode_pool{,_names}()
Ilya Dryomov [Fri, 21 Mar 2014 17:05:27 +0000 (19:05 +0200)]
libceph: rename __decode_pool{,_names}() to decode_pool{,_names}()

To be in line with all the other osdmap decode helpers.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: fix and clarify ceph_decode_need() sizes
Ilya Dryomov [Thu, 13 Mar 2014 14:36:16 +0000 (16:36 +0200)]
libceph: fix and clarify ceph_decode_need() sizes

Sum up sizeof(...) results instead of (incorrectly) hard-coding the
number of bytes, expressed in ints and longs.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: nuke bogus encoding version check in osdmap_apply_incremental()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:16 +0000 (16:36 +0200)]
libceph: nuke bogus encoding version check in osdmap_apply_incremental()

Only version 6 of osdmap encoding is supported, anything other than
version 6 results in an error and halts the decoding process.  Checking
if version is >= 5 is therefore bogus.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: fixup error handling in osdmap_apply_incremental()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:15 +0000 (16:36 +0200)]
libceph: fixup error handling in osdmap_apply_incremental()

The existing error handling scheme requires resetting err to -EINVAL
prior to calling any ceph_decode_* macro.  This is ugly and fragile,
and there already are a few places where we would return 0 on error,
due to a missing reset.  Follow osdmap_decode() and fix this by adding
a special e_inval label to be used by all ceph_decode_* macros.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: fix crush_decode() call site in osdmap_decode()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:15 +0000 (16:36 +0200)]
libceph: fix crush_decode() call site in osdmap_decode()

The size of the memory area feeded to crush_decode() should be limited
not only by osdmap end, but also by the crush map length.  Also, drop
unnecessary dout() (dout() in crush_decode() conveys the same info) and
step past crush map only if it is decoded successfully.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: check length of osdmap osd arrays
Ilya Dryomov [Thu, 13 Mar 2014 14:36:14 +0000 (16:36 +0200)]
libceph: check length of osdmap osd arrays

Check length of osd_state, osd_weight and osd_addr arrays.  They
should all have exactly max_osd elements after the call to
osdmap_set_max_osd().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: safely decode max_osd value in osdmap_decode()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:14 +0000 (16:36 +0200)]
libceph: safely decode max_osd value in osdmap_decode()

max_osd value is not covered by any ceph_decode_need().  Use a safe
version of ceph_decode_* macro to decode it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: fixup error handling in osdmap_decode()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:14 +0000 (16:36 +0200)]
libceph: fixup error handling in osdmap_decode()

The existing error handling scheme requires resetting err to -EINVAL
prior to calling any ceph_decode_* macro.  This is ugly and fragile,
and there already are a few places where we would return 0 on error,
due to a missing reset.  Fix this by adding a special e_inval label to
be used by all ceph_decode_* macros.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: split osdmap allocation and decode steps
Ilya Dryomov [Thu, 13 Mar 2014 14:36:13 +0000 (16:36 +0200)]
libceph: split osdmap allocation and decode steps

Split osdmap allocation and initialization into a separate function,
ceph_osdmap_decode().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: dump osdmap and enhance output on decode errors
Ilya Dryomov [Thu, 13 Mar 2014 14:36:13 +0000 (16:36 +0200)]
libceph: dump osdmap and enhance output on decode errors

Dump osdmap in hex on both full and incremental decode errors, to make
it easier to match the contents with error offset.  dout() map epoch
and max_osd value on success.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: dump pg_temp mappings to debugfs
Ilya Dryomov [Thu, 13 Mar 2014 14:36:13 +0000 (16:36 +0200)]
libceph: dump pg_temp mappings to debugfs

Dump pg_temp mappings to /sys/kernel/debug/ceph/<client>/osdmap,
one 'pg_temp <pgid> [<osd>, ..., <osd>]' per line, e.g:

    pg_temp 2.6 [2,3,4]

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: do not prefix osd lines with \t in debugfs output
Ilya Dryomov [Thu, 13 Mar 2014 14:36:12 +0000 (16:36 +0200)]
libceph: do not prefix osd lines with \t in debugfs output

To save screen space in anticipation of more fields (e.g. primary
affinity).

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: refer to osdmap directly in osdmap_show()
Ilya Dryomov [Thu, 13 Mar 2014 14:36:12 +0000 (16:36 +0200)]
libceph: refer to osdmap directly in osdmap_show()

To make it more readable and save screen space.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agocrush: support chooseleaf_vary_r tunable (tunables3) by default
Ilya Dryomov [Wed, 19 Mar 2014 14:58:37 +0000 (16:58 +0200)]
crush: support chooseleaf_vary_r tunable (tunables3) by default

Add TUNABLES3 feature (chooseleaf_vary_r tunable) to a set of features
supported by default.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
10 years agocrush: add SET_CHOOSELEAF_VARY_R step
Ilya Dryomov [Wed, 19 Mar 2014 14:58:37 +0000 (16:58 +0200)]
crush: add SET_CHOOSELEAF_VARY_R step

This lets you adjust the vary_r tunable on a per-rule basis.

Reflects ceph.git commit f944ccc20aee60a7d8da7e405ec75ad1cd449fac.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
10 years agocrush: add chooseleaf_vary_r tunable
Ilya Dryomov [Wed, 19 Mar 2014 14:58:37 +0000 (16:58 +0200)]
crush: add chooseleaf_vary_r tunable

The current crush_choose_firstn code will re-use the same 'r' value for
the recursive call.  That means that if we are hitting a collision or
rejection for some reason (say, an OSD that is marked out) and need to
retry, we will keep making the same (bad) choice in that recursive
selection.

Introduce a tunable that fixes that behavior by incorporating the parent
'r' value into the recursive starting point, so that a different path
will be taken in subsequent placement attempts.

Note that this was done from the get-go for the new crush_choose_indep
algorithm.

This was exposed by a user who was seeing PGs stuck in active+remapped
after reweight-by-utilization because the up set mapped to a single OSD.

Reflects ceph.git commit a8e6c9fbf88bad056dd05d3eb790e98a5e43451a.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
10 years agocrush: allow crush rules to set (re)tries counts to 0
Ilya Dryomov [Wed, 19 Mar 2014 14:58:37 +0000 (16:58 +0200)]
crush: allow crush rules to set (re)tries counts to 0

These two fields are misnomers; they are *retry* counts.

Reflects ceph.git commit f17caba8ae0cad7b6f8f35e53e5f73b444696835.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
10 years agocrush: fix off-by-one errors in total_tries refactor
Ilya Dryomov [Wed, 19 Mar 2014 14:58:36 +0000 (16:58 +0200)]
crush: fix off-by-one errors in total_tries refactor

Back in 27f4d1f6bc32c2ed7b2c5080cbd58b14df622607 we refactored the CRUSH
code to allow adjustment of the retry counts on a per-pool basis.  That
commit had an off-by-one bug: the previous "tries" counter was a *retry*
count, not a *try* count, but the new code was passing in 1 meaning
there should be no retries.

Fix the ftotal vs tries comparison to use < instead of <= to fix the
problem.  Note that the original code used <= here, which means the
global "choose_total_tries" tunable is actually counting retries.
Compensate for that by adding 1 in crush_do_rule when we pull the tunable
into the local variable.

This was noticed looking at output from a user provided osdmap.
Unfortunately the map doesn't illustrate the change in mapping behavior
and I haven't managed to construct one yet that does.  Inspection of the
crush debug output now aligns with prior versions, though.

Reflects ceph.git commit 795704fd615f0b008dcc81aa088a859b2d075138.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
10 years agoceph: don't include ceph.{file,dir}.layout vxattr in listxattr()
Yan, Zheng [Mon, 24 Mar 2014 11:15:05 +0000 (19:15 +0800)]
ceph: don't include ceph.{file,dir}.layout vxattr in listxattr()

This avoids 'cp -a' modifying layout of new files/directories.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: check buffer size in ceph_vxattrcb_layout()
Yan, Zheng [Mon, 24 Mar 2014 05:00:54 +0000 (13:00 +0800)]
ceph: check buffer size in ceph_vxattrcb_layout()

If buffer size is zero, return the size of layout vxattr. If buffer
size is not zero, check if it is large enough for layout vxattr.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: fix null pointer dereference in discard_cap_releases()
Yan, Zheng [Mon, 24 Mar 2014 01:56:43 +0000 (09:56 +0800)]
ceph: fix null pointer dereference in discard_cap_releases()

send_mds_reconnect() may call discard_cap_releases() after all
release messages have been dropped by cleanup_cap_releases()

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agolibceph: fix oops in ceph_msg_data_{pages,pagelist}_advance()
Yan, Zheng [Sat, 22 Mar 2014 22:50:39 +0000 (06:50 +0800)]
libceph: fix oops in ceph_msg_data_{pages,pagelist}_advance()

When there is no more data, ceph_msg_data_{pages,pagelist}_advance()
should not move on to the next page.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: Remove get/set acl on symlinks
Fabian Frederick [Fri, 21 Mar 2014 16:52:58 +0000 (09:52 -0700)]
ceph: Remove get/set acl on symlinks

Remove unsupported symlink operations.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
10 years agoceph: set mds_wanted when MDS reply changes a cap to auth cap
Yan, Zheng [Tue, 18 Mar 2014 02:15:29 +0000 (10:15 +0800)]
ceph: set mds_wanted when MDS reply changes a cap to auth cap

When adjusting caps client wants, MDS does not record caps that are
not allowed. For non-auth MDS, it does not record WR caps. So when
a MDS reply changes a non-auth cap to auth cap, client needs to set
cap's mds_wanted according to the reply.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: use fl->fl_file as owner identifier of flock and posix lock
Yan, Zheng [Sun, 9 Mar 2014 15:16:40 +0000 (23:16 +0800)]
ceph: use fl->fl_file as owner identifier of flock and posix lock

flock and posix lock should use fl->fl_file instead of process ID
as owner identifier. (posix lock uses fl->fl_owner. fl->fl_owner
is usually equal to fl->fl_file, but it also can be a customized
value). The process ID of who holds the lock is just for F_GETLK
fcntl(2).

The fix is rename the 'pid' fields of struct ceph_mds_request_args
and struct ceph_filelock to 'owner', rename 'pid_namespace' fields
to 'pid'. Assign fl->fl_file to the 'owner' field of lock messages.
We also set the most significant bit of the 'owner' field. MDS can
use that bit to distinguish between old and new clients.

The MDS counterpart of this patch modifies the flock code to not
take the 'pid_namespace' into consideration when checking conflict
locks.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: forbid mandatory file lock
Yan, Zheng [Tue, 4 Mar 2014 07:50:06 +0000 (15:50 +0800)]
ceph: forbid mandatory file lock

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: use fl->fl_type to decide flock operation
Yan, Zheng [Tue, 4 Mar 2014 07:42:24 +0000 (15:42 +0800)]
ceph: use fl->fl_type to decide flock operation

VFS does not directly pass flock's operation code to filesystem's
flock callback. It translates the operation code to the form how
posix lock's parameters are presented.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: update i_max_size even if inode version does not change
Yan, Zheng [Sat, 8 Mar 2014 12:12:23 +0000 (20:12 +0800)]
ceph: update i_max_size even if inode version does not change

handle following sequence of events:
 - client releases a inode with i_max_size > 0. The release message
   is queued. (is not sent to the auth MDS)
 - a 'lookup' request reply from non-auth MDS returns the same inode.
 - client opens the inode in write mode. The version of inode trace
   in 'open' request reply is equal to the cached inode's version.
 - client requests new max size. The MDS ignores the request because
   it does not affect client's write range

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: make sure write caps are registered with auth MDS
Yan, Zheng [Sat, 8 Mar 2014 01:51:45 +0000 (09:51 +0800)]
ceph: make sure write caps are registered with auth MDS

Only auth MDS can issue write caps to clients, so don't consider
write caps registered with non-auth MDS as valid.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: print inode number for LOOKUPINO request
Yan, Zheng [Sat, 1 Mar 2014 14:22:57 +0000 (22:22 +0800)]
ceph: print inode number for LOOKUPINO request

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: add get_name() NFS export callback
Yan, Zheng [Thu, 6 Mar 2014 08:40:32 +0000 (16:40 +0800)]
ceph: add get_name() NFS export callback

Use the newly introduced LOOKUPNAME MDS request to connect child
inode to its parent directory.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: fix ceph_fh_to_parent()
Yan, Zheng [Sat, 1 Mar 2014 15:09:05 +0000 (23:09 +0800)]
ceph: fix ceph_fh_to_parent()

ceph_fh_to_parent() returns dentry that corresponds to the 'ino' field
of struct ceph_nfs_confh. This is wrong, it should return dentry that
corresponds to the 'parent_ino' field.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: add get_parent() NFS export callback
Yan, Zheng [Sat, 1 Mar 2014 14:11:45 +0000 (22:11 +0800)]
ceph: add get_parent() NFS export callback

The callback uses LOOKUPPARENT MDS request to find parent.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: simplify ceph_fh_to_dentry()
Yan, Zheng [Sat, 1 Mar 2014 10:05:41 +0000 (18:05 +0800)]
ceph: simplify ceph_fh_to_dentry()

MDS handles LOOKUPHASH and LOOKUPINO MDS requests in the same way.
So __cfh_to_dentry() is redundant.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoceph: fscache: Wait for completion of object initialization
Yunchuan Wen [Thu, 26 Dec 2013 14:29:28 +0000 (06:29 -0800)]
ceph: fscache: Wait for completion of object initialization

The object store limit needs to be updated after writing,
and this can be done provided the corresponding object has already
been initialized. Current object initialization is done asynchrously,
which introduce a race if a file is opened, then immediately followed
by a writing, the initialization may have not completed, the code will
reach the ASSERT in fscache_submit_exclusive_op() to cause kernel
bug.

Tested-by: Milosz Tanski <milosz@adfin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
Signed-off-by: Min Chen <minchen@ubuntukylin.com>
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
10 years agoceph: fscache: Update object store limit after file writing
Yunchuan Wen [Thu, 26 Dec 2013 14:29:27 +0000 (06:29 -0800)]
ceph: fscache: Update object store limit after file writing

Synchronize object->store_limit[_l] with new inode->i_size after file writing.

Tested-by: Milosz Tanski <milosz@adfin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
Signed-off-by: Min Chen <minchen@ubuntukylin.com>
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
10 years agoceph: fscache: add an interface to synchronize object store limit
Yunchuan Wen [Thu, 26 Dec 2013 14:29:26 +0000 (06:29 -0800)]
ceph: fscache: add an interface to synchronize object store limit

Add an interface to explicitly synchronize object->store_limit[_l]
with inode->i_size

Tested-by: Milosz Tanski <milosz@adfin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
Signed-off-by: Min Chen <minchen@ubuntukylin.com>
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
10 years agoceph: do not set r_old_dentry_dir on link()
Sage Weil [Tue, 5 Feb 2013 21:41:23 +0000 (13:41 -0800)]
ceph: do not set r_old_dentry_dir on link()

This is racy--we do not know whather d_parent has changed out from
underneath us because i_mutex is not held on the source inode's directory.

Also, taking this reference is useless.

Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: do not assume r_old_dentry[_dir] always set together
Sage Weil [Tue, 5 Feb 2013 21:40:09 +0000 (13:40 -0800)]
ceph: do not assume r_old_dentry[_dir] always set together

Do not assume that r_old_dentry implies that r_old_dentry_dir is also
true.  Separate out the ref cleanup and make the debugs dump behave when
it is NULL.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: do not chain inode updates to parent fsync
Sage Weil [Tue, 5 Feb 2013 21:52:29 +0000 (13:52 -0800)]
ceph: do not chain inode updates to parent fsync

The fsync(dirfd) only covers namespace operations, not inode updates.
We do not need to cover setattr variants or O_TRUNC.

Reported-by: Al Viro <viro@xeniv.linux.org.uk>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: avoid useless ceph_get_dentry_parent_inode() in ceph_rename()
Sage Weil [Tue, 5 Feb 2013 21:36:05 +0000 (13:36 -0800)]
ceph: avoid useless ceph_get_dentry_parent_inode() in ceph_rename()

This is just old_dir; no reason to abuse the dcache pointers.

Reported-by: Al Viro <viro.zeniv.linux.org.uk>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: let MDS adjust readdir 'frag'
Yan, Zheng [Mon, 3 Mar 2014 01:20:44 +0000 (09:20 +0800)]
ceph: let MDS adjust readdir 'frag'

If readdir 'frag' is adjusted, readdir 'offset' should be reset.
Otherwise some dentries may be lost when readdir and fragmenting
directory happen at the some.

Another way to fix this issue is let MDS adjust readdir 'frag'.
The code that handles MDS reply reset the readdir 'offset' if
the readdir reply is different than the requested one.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
10 years agoceph: fix reset_readdir()
Yan, Zheng [Fri, 28 Feb 2014 08:36:09 +0000 (16:36 +0800)]
ceph: fix reset_readdir()

When changing readdir postion, fi->next_offset should be set to 0
if the new postion is not in the first dirfrag.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agoceph: fix ceph_dir_llseek()
Yan, Zheng [Thu, 27 Feb 2014 08:26:24 +0000 (16:26 +0800)]
ceph: fix ceph_dir_llseek()

Comparing offset with inode->i_sb->s_maxbytes doesn't make sense for
directory. For a fragmented directory, offset (frag_t, off) can be
larger than inode->i_sb->s_maxbytes.

At the very beginning of ceph_dir_llseek(), local variable old_offset
is initialized to parameter offset. This doesn't make sense neither.
Old_offset should be ceph_make_fpos(fi->frag, fi->next_offset).

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agorbd: prefix rbd writes with CEPH_OSD_OP_SETALLOCHINT osd op
Ilya Dryomov [Tue, 25 Feb 2014 14:22:28 +0000 (16:22 +0200)]
rbd: prefix rbd writes with CEPH_OSD_OP_SETALLOCHINT osd op

In an effort to reduce fragmentation, prefix every rbd write with
a CEPH_OSD_OP_SETALLOCHINT osd op with an expected_write_size value set
to the object size (1 << order).  Backwards compatibility is taken care
of on the libceph/osd side.

"The CEPH_OSD_OP_SETALLOCHINT hint is durable, in that it's enough to
do it once.  The reason every rbd write is prefixed is that rbd doesn't
explicitly create objects and relies on writes creating them
implicitly, so there is no place to stick a single hint op into.  To
get around that we decided to prefix every rbd write with a hint (just
like write and setattr ops, hint op will create an object implicitly if
it doesn't exist)."

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agorbd: num_ops parameter for rbd_osd_req_create()
Ilya Dryomov [Tue, 25 Feb 2014 14:22:27 +0000 (16:22 +0200)]
rbd: num_ops parameter for rbd_osd_req_create()

In preparation for prefixing rbd writes with an allocation hint
introduce a num_ops parameter for rbd_osd_req_create().  The rationale
is that not every write request is a write op that needs to be prefixed
(e.g. watch op), so the num_ops logic needs to be in the callers.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: bump CEPH_OSD_MAX_OP to 3
Ilya Dryomov [Tue, 25 Feb 2014 14:22:27 +0000 (16:22 +0200)]
libceph: bump CEPH_OSD_MAX_OP to 3

Our longest osd request now contains 3 ops: copyup+hint+write.

Also, CEPH_OSD_MAX_OP value in a BUG_ON in rbd_osd_req_callback() was
hard-coded to 2.  Fix it, and switch to rbd_assert while at it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: add support for CEPH_OSD_OP_SETALLOCHINT osd op
Ilya Dryomov [Tue, 25 Feb 2014 14:22:27 +0000 (16:22 +0200)]
libceph: add support for CEPH_OSD_OP_SETALLOCHINT osd op

This is primarily for rbd's benefit and is supposed to combat
fragmentation:

"... knowing that rbd images have a 4m size, librbd can pass a hint
that will let the osd do the xfs allocation size ioctl on new files so
that they are allocated in 1m or 4m chunks.  We've seen cases where
users with rbd workloads have very high levels of fragmentation in xfs
and this would mitigate that and probably have a pretty nice
performance benefit."

SETALLOCHINT is considered advisory, so our backwards compatibility
mechanism here is to set FAILOK flag for all SETALLOCHINT ops.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: encode CEPH_OSD_OP_FLAG_* op flags
Ilya Dryomov [Tue, 25 Feb 2014 14:22:26 +0000 (16:22 +0200)]
libceph: encode CEPH_OSD_OP_FLAG_* op flags

Encode ceph_osd_op::flags field so that it gets sent over the wire.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agorbd: fix error paths in rbd_img_request_fill()
Ilya Dryomov [Tue, 4 Mar 2014 09:57:17 +0000 (11:57 +0200)]
rbd: fix error paths in rbd_img_request_fill()

Doing rbd_obj_request_put() in rbd_img_request_fill() error paths is
not only insufficient, but also triggers an rbd_assert() in
rbd_obj_request_destroy():

    Assertion failure in rbd_obj_request_destroy() at line 1867:

    rbd_assert(obj_request->img_request == NULL);

rbd_img_obj_request_add() adds obj_requests to the img_request, the
opposite is rbd_img_obj_request_del().  Use it.

Fixes: http://tracker.ceph.com/issues/7327

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agorbd: remove out_partial label in rbd_img_request_fill()
Ilya Dryomov [Tue, 4 Mar 2014 09:57:17 +0000 (11:57 +0200)]
rbd: remove out_partial label in rbd_img_request_fill()

Commit 03507db631c94 ("rbd: fix buffer size for writes to images with
snapshots") moved the call to rbd_img_obj_request_add() up, making the
out_partial label bogus.  Remove it.

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
10 years agolibceph: a per-osdc crush scratch buffer
Ilya Dryomov [Fri, 31 Jan 2014 15:54:26 +0000 (17:54 +0200)]
libceph: a per-osdc crush scratch buffer

With the addition of erasure coding support in the future, scratch
variable-length array in crush_do_rule_ary() is going to grow to at
least 200 bytes on average, on top of another 128 bytes consumed by
rawosd/osd arrays in the call chain.  Replace it with a buffer inside
struct osdmap and a mutex.  This shouldn't result in any contention,
because all osd requests were already serialized by request_mutex at
that point; the only unlocked caller was ceph_ioctl_get_dataloc().

Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
10 years agoLinux 3.14
Linus Torvalds [Mon, 31 Mar 2014 03:40:15 +0000 (20:40 -0700)]
Linux 3.14

10 years agoMerge branch 'for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Mon, 31 Mar 2014 00:26:08 +0000 (17:26 -0700)]
Merge branch 'for-linus-2' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "Switch mnt_hash to hlist, turning the races between __lookup_mnt() and
  hash modifications into false negatives from __lookup_mnt() (instead
  of hangs)"

On the false negatives from __lookup_mnt():
 "The *only* thing we care about is not getting stuck in __lookup_mnt().
  If it misses an entry because something in front of it just got moved
  around, etc, we are fine.  We'll notice that mount_lock mismatch and
  that'll be it"

* 'for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  switch mnt_hash to hlist
  don't bother with propagate_mnt() unless the target is shared
  keep shadowed vfsmounts together
  resizable namespace.c hashes

10 years agoMAINTAINERS: resume as Documentation maintainer
Randy Dunlap [Fri, 28 Mar 2014 16:45:33 +0000 (09:45 -0700)]
MAINTAINERS: resume as Documentation maintainer

I am the new kernel tree Documentation maintainer (except for parts that
are handled by other people, of course).

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Rob Landley <rob@landley.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Mon, 31 Mar 2014 00:20:40 +0000 (17:20 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input updates from Dmitry Torokhov:
 "Some more updates for the input subsystem.

  You will get a fix for race in mousedev that has been causing quite a
  few oopses lately and a small fixup for force feedback support in
  evdev"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: mousedev - fix race when creating mixed device
  Input: don't modify the id of ioctl-provided ff effect on upload failure

10 years agoAUDIT: Allow login in non-init namespaces
Eric Paris [Sun, 30 Mar 2014 23:07:54 +0000 (19:07 -0400)]
AUDIT: Allow login in non-init namespaces

It its possible to configure your PAM stack to refuse login if audit
messages (about the login) were unable to be sent.  This is common in
many distros and thus normal configuration of many containers.  The PAM
modules determine if audit is enabled/disabled in the kernel based on
the return value from sending an audit message on the netlink socket.
If userspace gets back ECONNREFUSED it believes audit is disabled in the
kernel.  If it gets any other error else it refuses to let the login
proceed.

Just about ever since the introduction of namespaces the kernel audit
subsystem has returned EPERM if the task sending a message was not in
the init user or pid namespace.  So many forms of containers have never
worked if audit was enabled in the kernel.

BUT if the container was not in net_init then the kernel network code
would send ECONNREFUSED (instead of the audit code sending EPERM).  Thus
by pure accident/dumb luck/bug if an admin configured the PAM stack to
reject all logins that didn't talk to audit, but then ran the login
untility in the non-init_net namespace, it would work!! Clearly this was
a bug, but it is a bug some people expected.

With the introduction of network namespace support in 3.14-rc1 the two
bugs stopped cancelling each other out.  Now, containers in the
non-init_net namespace refused to let users log in (just like PAM was
configfured!) Obviously some people were not happy that what used to let
users log in, now didn't!

This fix is kinda hacky.  We return ECONNREFUSED for all non-init
relevant namespaces.  That means that not only will the old broken
non-init_net setups continue to work, now the broken non-init_pid or
non-init_user setups will 'work'.  They don't really work, since audit
isn't logging things.  But it's what most users want.

In 3.15 we should have patches to support not only the non-init_net
(3.14) namespace but also the non-init_pid and non-init_user namespace.
So all will be right in the world.  This just opens the doors wide open
on 3.14 and hopefully makes users happy, if not the audit system...

Reported-by: Andre Tomt <andre@tomt.net>
Reported-by: Adam Richter <adam_richter2004@yahoo.com>
Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoext4: atomically set inode->i_flags in ext4_set_inode_flags()
Theodore Ts'o [Sun, 30 Mar 2014 14:20:01 +0000 (10:20 -0400)]
ext4: atomically set inode->i_flags in ext4_set_inode_flags()

Use cmpxchg() to atomically set i_flags instead of clearing out the
S_IMMUTABLE, S_APPEND, etc. flags and then setting them from the
EXT4_IMMUTABLE_FL, EXT4_APPEND_FL flags, since this opens up a race
where an immutable file has the immutable flag cleared for a brief
window of time.

Reported-by: John Sullivan <jsrhbz@kanargh.force9.co.uk>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoswitch mnt_hash to hlist
Al Viro [Fri, 21 Mar 2014 01:10:51 +0000 (21:10 -0400)]
switch mnt_hash to hlist

fixes RCU bug - walking through hlist is safe in face of element moves,
since it's self-terminating.  Cyclic lists are not - if we end up jumping
to another hash chain, we'll loop infinitely without ever hitting the
original list head.

[fix for dumb braino folded]

Spotted by: Max Kellermann <mk@cm4all.com>
Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agodon't bother with propagate_mnt() unless the target is shared
Al Viro [Fri, 21 Mar 2014 14:14:08 +0000 (10:14 -0400)]
don't bother with propagate_mnt() unless the target is shared

If the dest_mnt is not shared, propagate_mnt() does nothing -
there's no mounts to propagate to and thus no copies to create.
Might as well don't bother calling it in that case.

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agokeep shadowed vfsmounts together
Al Viro [Fri, 21 Mar 2014 00:34:43 +0000 (20:34 -0400)]
keep shadowed vfsmounts together

preparation to switching mnt_hash to hlist

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agoresizable namespace.c hashes
Al Viro [Fri, 28 Feb 2014 18:46:44 +0000 (13:46 -0500)]
resizable namespace.c hashes

* switch allocation to alloc_large_system_hash()
* make sizes overridable by boot parameters (mhash_entries=, mphash_entries=)
* switch mountpoint_hashtable from list_head to hlist_head

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agoMerge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Mar 2014 22:01:09 +0000 (15:01 -0700)]
Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull timer fix from Ingo Molnar:
 "A late breaking fix from John.  (The bug fixed has a hard lockup
  potential, but that was not observed, warnings were)"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  time: Revert to calling clock_was_set_delayed() while in irq context

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Sat, 29 Mar 2014 22:00:27 +0000 (15:00 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

Pull Ceph fix from Sage Weil:
 "This drops a bad assert that a few users have been hitting but we've
  only recently been able to track down"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  rbd: drop an unsafe assertion

10 years agoInput: mousedev - fix race when creating mixed device
Dmitry Torokhov [Thu, 6 Mar 2014 20:57:24 +0000 (12:57 -0800)]
Input: mousedev - fix race when creating mixed device

We should not be using static variable mousedev_mix in methods that can be
called before that singleton gets assigned. While at it let's add open and
close methods to mousedev structure so that we do not need to test if we
are dealing with multiplexor or normal device and simply call appropriate
method directly.

This fixes: https://bugzilla.kernel.org/show_bug.cgi?id=71551

Reported-by: GiulioDP <depasquale.giulio@gmail.com>
Tested-by: GiulioDP <depasquale.giulio@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
10 years agoInput: don't modify the id of ioctl-provided ff effect on upload failure
Elias Vanderstuyft [Sat, 29 Mar 2014 19:08:45 +0000 (12:08 -0700)]
Input: don't modify the id of ioctl-provided ff effect on upload failure

If a new (id == -1) ff effect was uploaded from userspace,
ff-core.c::input_ff_upload() will have assigned a positive number to the
new effect id.  Currently, evdev.c::evdev_do_ioctl() will save this new id
to userspace, regardless of whether the upload succeeded or not.

On upload failure, this can be confusing because the dev->ff->effects[]
array will not contain an element at the index of that new effect id.

This patch fixes this by leaving the id unchanged after upload fails.

Note: Unfortunately applications should still expect changed effect id for
quite some time.

This has been discussed on:
http://www.mail-archive.com/linux-input@vger.kernel.org/msg08513.html
("ff-core effect id handling in case of a failed effect upload")

Suggested-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Elias Vanderstuyft <elias.vds@gmail.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
10 years agorbd: drop an unsafe assertion
Alex Elder [Tue, 25 Mar 2014 13:36:02 +0000 (15:36 +0200)]
rbd: drop an unsafe assertion

Olivier Bonvalet reported having repeated crashes due to a failed
assertion he was hitting in rbd_img_obj_callback():

    Assertion failure in rbd_img_obj_callback() at line 2165:
rbd_assert(which >= img_request->next_completion);

With a lot of help from Olivier with reproducing the problem
we were able to determine the object and image requests had
already been completed (and often freed) at the point the
assertion failed.

There was a great deal of discussion on the ceph-devel mailing list
about this.  The problem only arose when there were two (or more)
object requests in an image request, and the problem was always
seen when the second request was being completed.

The problem is due to a race in the window between setting the
"done" flag on an object request and checking the image request's
next completion value.  When the first object request completes, it
checks to see if its successor request is marked "done", and if
so, that request is also completed.  In the process, the image
request's next_completion value is updated to reflect that both
the first and second requests are completed.  By the time the
second request is able to check the next_completion value, it
has been set to a value *greater* than its own "which" value,
which caused an assertion to fail.

Fix this problem by skipping over any completion processing
unless the completing object request is the next one expected.
Test only for inequality (not >=), and eliminate the bad
assertion.

Tested-by: Olivier Bonvalet <ob@daevel.fr>
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Ilya Dryomov <ilya.dryomov@inktank.com>