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>
Several spots in the kernel perform a sequence like:
skb_queue_tail(&sk->s_receive_queue, skb);
sk->sk_data_ready(sk, skb->len);
But at the moment we place the SKB onto the socket receive queue it
can be consumed and freed up. So this skb->len access is potentially
to freed up memory.
Furthermore, the skb->len can be modified by the consumer so it is
possible that the value isn't accurate.
And finally, no actual implementation of this callback actually uses
the length argument. And since nobody actually cared about it's
value, lots of call sites pass arbitrary values in such as '0' and
even '1'.
So just remove the length argument from the callback, that way there
is no confusion whatsoever and all of these use-after-free cases get
fixed as a side effect.
Based upon a patch by Eric Dumazet and his suggestion to audit this
issue tree-wide.
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull Ceph updates from Sage Weil:
"The biggest chunk is a series of patches from Ilya that add support
for new Ceph osd and crush map features, including some new tunables,
primary affinity, and the new encoding that is needed for erasure
coding support. This brings things into parity with the server side
and the looming firefly release. There is also support for allocation
hints in RBD that help limit fragmentation on the server side.
There is also a series of patches from Zheng fixing NFS reexport,
directory fragmentation support, flock vs fnctl behavior, and some
issues with clustered MDS.
Finally, there are some miscellaneous fixes from Yunchuan Wen for
fscache, Fabian Frederick for ACLs, and from me for fsync(dirfd)
behavior"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client: (79 commits)
ceph: skip invalid dentry during dcache readdir
libceph: dump pool {read,write}_tier to debugfs
libceph: output primary affinity values on osdmap updates
ceph: flush cap release queue when trimming session caps
ceph: don't grabs open file reference for aborted request
ceph: drop extra open file reference in ceph_atomic_open()
ceph: preallocate buffer for readdir reply
libceph: enable PRIMARY_AFFINITY feature bit
libceph: redo ceph_calc_pg_primary() in terms of ceph_calc_pg_acting()
libceph: add support for osd primary affinity
libceph: add support for primary_temp mappings
libceph: return primary from ceph_calc_pg_acting()
libceph: switch ceph_calc_pg_acting() to new helpers
libceph: introduce apply_temps() helper
libceph: introduce pg_to_raw_osds() and raw_to_up_osds() helpers
libceph: ceph_can_shift_osds(pool) and pool type defines
libceph: ceph_osd_{exists,is_up,is_down}(osd) definitions
libceph: enable OSDMAP_ENC feature bit
libceph: primary_affinity decode bits
libceph: primary_affinity infrastructure
...
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
One of my pet coding style peeves is the practice of
adding extra return; at the end of function.
Kill several instances of this in network code.
I suppose some coccinelle wizardy could do this automatically.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit f38a5181d9 ("ceph: Convert to immutable biovecs") introduced
a NULL pointer dereference, which broke rbd in -rc1. Fix it.
Cc: Kent Overstreet <kmo@daterainc.com>
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Handling redirect replies requires both map_sem and request_mutex.
Taking map_sem unconditionally near the top of handle_reply() avoids
possible race conditions that arise from releasing request_mutex to be
able to acquire map_sem in redirect reply case. (Lock ordering is:
map_sem, request_mutex, crush_mutex.)
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Factor out logic from ceph_osdc_start_request() into a new helper,
__ceph_osdc_start_request(). ceph_osdc_start_request() now amounts to
taking locks and calling __ceph_osdc_start_request().
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
msgpool_op_reply message pool isn't destroyed if workqueue construction
fails. Fix it.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pull core block IO changes from Jens Axboe:
"The major piece in here is the immutable bio_ve series from Kent, the
rest is fairly minor. It was supposed to go in last round, but
various issues pushed it to this release instead. The pull request
contains:
- Various smaller blk-mq fixes from different folks. Nothing major
here, just minor fixes and cleanups.
- Fix for a memory leak in the error path in the block ioctl code
from Christian Engelmayer.
- Header export fix from CaiZhiyong.
- Finally the immutable biovec changes from Kent Overstreet. This
enables some nice future work on making arbitrarily sized bios
possible, and splitting more efficient. Related fixes to immutable
bio_vecs:
- dm-cache immutable fixup from Mike Snitzer.
- btrfs immutable fixup from Muthu Kumar.
- bio-integrity fix from Nic Bellinger, which is also going to stable"
* 'for-3.14/core' of git://git.kernel.dk/linux-block: (44 commits)
xtensa: fixup simdisk driver to work with immutable bio_vecs
block/blk-mq-cpu.c: use hotcpu_notifier()
blk-mq: for_each_* macro correctness
block: Fix memory leak in rw_copy_check_uvector() handling
bio-integrity: Fix bio_integrity_verify segment start bug
block: remove unrelated header files and export symbol
blk-mq: uses page->list incorrectly
blk-mq: use __smp_call_function_single directly
btrfs: fix missing increment of bi_remaining
Revert "block: Warn and free bio if bi_end_io is not set"
block: Warn and free bio if bi_end_io is not set
blk-mq: fix initializing request's start time
block: blk-mq: don't export blk_mq_free_queue()
block: blk-mq: make blk_sync_queue support mq
block: blk-mq: support draining mq queue
dm cache: increment bi_remaining when bi_end_io is restored
block: fixup for generic bio chaining
block: Really silence spurious compiler warnings
block: Silence spurious compiler warnings
block: Kill bio_pair_split()
...
Follow redirect replies from osds, for details see ceph.git commit
fbbe3ad1220799b7bb00ea30fce581c5eadaf034.
v1 (current) version of redirect reply consists of oloc and oid, which
expands to pool, key, nspace, hash and oid. However, server-side code
that would populate anything other than pool doesn't exist yet, and
hence this commit adds support for pool redirects only. To make sure
that future server-side updates don't break us, we decode all fields
and, if any of key, nspace, hash or oid have a non-default value, error
out with "corrupt osd_op_reply ..." message.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Rename ceph_osd_request::r_{oloc,oid} to r_base_{oloc,oid} before
introducing r_target_{oloc,oid} needed for redirects.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Overwrite ceph_osd_request::r_oloc.pool with read_tier for read ops and
write_tier for write and read+write ops (aka basic tiering support).
{read,write}_tier are part of pg_pool_t since v9. This commit bumps
our pg_pool_t decode compat version from v7 to v9, all new fields
except for {read,write}_tier are ignored.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
"Lookup pool info by ID" function is hidden in osdmap.c. Expose it to
the rest of libceph.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Switch ceph_calc_ceph_pg() to new oloc and oid abstractions and rename
it to ceph_oloc_oid_to_pg() to make its purpose more clear.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
In preparation for tiering support, which would require having two
(base and target) object names for each osd request and also copying
those names around, introduce struct ceph_object_id (oid) and a couple
helpers to facilitate those copies and encapsulate the fact that object
name is not necessarily a NUL-terminated string.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
In preparation for adding oid abstraction, rename MAX_OBJ_NAME_SIZE to
CEPH_MAX_OID_NAME_LEN.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Instead of relying on pool fields in ceph_file_layout (for mapping) and
ceph_pg (for enconding), start using ceph_object_locator (oloc)
abstraction. Note that userspace oloc currently consists of pool, key,
nspace and hash fields, while this one contains only a pool. This is
OK, because at this point we only send (i.e. encode) olocs and never
have to receive (i.e. decode) them.
This makes keeping a copy of ceph_file_layout in every osd request
unnecessary, so ceph_osd_request::r_file_layout field is nuked.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Encapsulate kmalloc vs vmalloc memory allocation and freeing logic into
two helpers, ceph_kvmalloc() and ceph_kvfree(), and switch to them.
ceph_kvmalloc() kmalloc()'s a maximum of 8 pages, anything bigger is
vmalloc()'ed with __GFP_HIGHMEM set. This changes the existing
behaviour:
- for buffers (ceph_buffer_new()), from trying to kmalloc() everything
and using vmalloc() just as a fallback
- for messages (ceph_msg_new()), from going to vmalloc() for anything
bigger than a page
- for messages (ceph_msg_new()), from disallowing vmalloc() to use high
memory
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The check that makes sure that we have enough memory allocated to read
in the entire header of the message in question is currently busted.
It compares front_len of the incoming message with iov_len field of
ceph_msg::front structure, which is used primarily to indicate the
amount of data already read in, and not the size of the allocated
buffer. Under certain conditions (e.g. a short read from a socket
followed by that socket's shutdown and owning ceph_connection reset)
this results in a warning similar to
[85688.975866] libceph: get_reply front 198 > preallocated 122 (4#0)
and, through another bug, leads to forever hung tasks and forced
reboots. Fix this by comparing front_len with front_alloc_len field of
struct ceph_msg, which stores the actual size of the buffer.
Fixes: http://tracker.ceph.com/issues/5425
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Rename front local variable to front_len in get_reply() to make its
purpose more clear.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Rename front_max field of struct ceph_msg to front_alloc_len to make
its purpose more clear.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Similar to userspace, don't bail with "parse_ips bad ip ..." if the
specified port is port 0, instead use port CEPH_MON_PORT (6789, the
default monitor port).
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This allows all of the tunables to be overridden by a specific rule.
Reflects ceph.git commits d129e09e57fbc61cfd4f492e3ee77d0750c9d292,
0497db49e5973b50df26251ed0e3f4ac7578e66e.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The legacy behavior is to make the normal number of tries for the
recursive chooseleaf call. The descend_once tunable changed this to
making a single try and bail if we get a reject (note that it is
impossible to collide in the recursive case).
The new set_chooseleaf_tries lets you select the number of recursive
chooseleaf attempts for indep mode, or default to 1. Use the same
behavior for firstn, except default to total_tries when the legacy
tunables are set (for compatibility). This makes the rule step
override the (new) default of 1 recursive attempt, keeping behavior
consistent with indep mode.
Reflects ceph.git commit 685c6950ef3df325ef04ce7c986e36ca2514c5f1.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This aligns the internal identifier names with the user-visible names in
the decompiled crush map language.
Reflects ceph.git commit caa0e22e15e4226c3671318ba1f61314bf6da2a6.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Since we can specify the recursive retries in a rule, we may as well also
specify the non-recursive tries too for completeness.
Reflects ceph.git commit d1b97462cffccc871914859eaee562f2786abfd1.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Parameterize the attempts for the _firstn choose method, and apply the
rule-specified tries count to firstn mode as well. Note that we have
slightly different behavior here than with indep:
If the firstn value is not specified for firstn, we pass through the
normal attempt count. This maintains compatibility with legacy behavior.
Note that this is usually *not* actually N^2 work, though, because of the
descend_once tunable. However, descend_once is unfortunately *not* the
same thing as 1 chooseleaf try because it is only checked on a reject but
not on a collision. Sigh.
In contrast, for indep, if tries is not specified we default to 1
recursive attempt, because that is simply more sane, and we have the
option to do so. The descend_once tunable has no effect for indep.
Reflects ceph.git commit 64aeded50d80942d66a5ec7b604ff2fcbf5d7b63.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Explicitly control the number of sample attempts, and allow the number of
tries in the recursive call to be explicitly controlled via the rule. This
is important because the amount of time we want to spend looking for a
solution may be rule dependent (e.g., higher for the wide indep pool than
the rep pools).
(We should do the same for the other tunables, by the way!)
Reflects ceph.git commit c43c893be872f709c787bc57f46c0e97876ff681.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pass down the parent's 'r' value so that we will sample different values in
the recursive call when the parent tries multiple times. This avoids doing
useless work (calling multiple times and trying the same values).
Reflects ceph.git commit 2731d3030d7a3e80922b7f1b7756f9a4a124bac5.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pass numrep (the width of the result) separately from the number of results
we want *this* iteration. This makes things less awkward when we do a
recursive call (for chooseleaf) and want only one item.
Reflects ceph.git commit 1b567ee08972f268c11b43fc881e57b5984dd08b.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Now that indep is handled by crush_choose_indep, rename crush_choose to
crush_choose_firstn and remove all the conditionals. This ends up
stripping out *lots* of code.
Note that it *also* makes it obvious that the shenanigans we were playing
with r' for uniform buckets were broken for firstn mode. This appears to
have happened waaaay back in commit dae8bec9 (or earlier)... 2007.
Reflects ceph.git commit 94350996cb2035850bcbece6a77a9b0394177ec9.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
For firstn mode, if we fail to make a valid placement choice, we just
continue and return a short result to the caller. For indep mode, however,
we need to make the position stable, and return an undefined value on
failed placements to avoid shifting later results to the left.
Reflects ceph.git commit b1d4dd4eb044875874a1d01c01c7d766db5d0a80.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This is only present to size the temporary scratch arrays that we put on
the stack. Let the caller allocate them as they wish and remove the
limitation.
Reflects ceph.git commit 1cfe140bf2dab99517589a82a916f4c75b9492d1.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Pass the size of the weight vector into crush_do_rule() to ensure that we
don't access values past the end. This can happen if the caller misbehaves
and passes a weight vector that is smaller than max_devices.
Currently the monitor tries to prevent that from happening, but this will
gracefully tolerate previous bad osdmaps that got into this state. It's
also a bit more defensive.
Reflects ceph.git commit 5922e2c2b8335b5e46c9504349c3a55b7434c01a.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This updates ceph_features.h so that it has all feature bits defined in
ceph.git. In the interim since the last update, ceph.git crossed the
"32 feature bits" point, and, the addition of the 33rd bit wasn't
handled correctly. The work-around is squashed into this commit and
reflects ceph.git commit 053659d05e0349053ef703b414f44965f368b9f0.
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
In preparation for ceph_features.h update, change all features fields
from unsigned int/u32 to u64. (ceph.git has ~40 feature bits at this
point.)
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
With the current full handling, there is a race between osds and
clients getting the first map marked full. If the osd wins, it will
return -ENOSPC to any writes, but the client may already have writes
in flight. This results in the client getting the error and
propagating it up the stack. For rbd, the block layer turns this into
EIO, which can cause corruption in filesystems above it.
To avoid this race, osds are being changed to drop writes that came
from clients with an osdmap older than the last osdmap marked full.
In order for this to work, clients must resend all writes after they
encounter a full -> not full transition in the osdmap. osds will wait
for an updated map instead of processing a request from a client with
a newer map, so resent writes will not be dropped by the osd unless
there is another not full -> full transition.
This approach requires both osds and clients to be fixed to avoid the
race. Old clients talking to osds with this fix may hang instead of
returning EIO and potentially corrupting an fs. New clients talking to
old osds have the same behavior as before if they encounter this race.
Fixes: http://tracker.ceph.com/issues/6938
Reviewed-by: Sage Weil <sage@inktank.com>
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
The PAUSEWR and PAUSERD flags are meant to stop the cluster from
processing writes and reads, respectively. The FULL flag is set when
the cluster determines that it is out of space, and will no longer
process writes. PAUSEWR and PAUSERD are purely client-side settings
already implemented in userspace clients. The osd does nothing special
with these flags.
When the FULL flag is set, however, the osd responds to all writes
with -ENOSPC. For cephfs, this makes sense, but for rbd the block
layer translates this into EIO. If a cluster goes from full to
non-full quickly, a filesystem on top of rbd will not behave well,
since some writes succeed while others get EIO.
Fix this by blocking any writes when the FULL flag is set in the osd
client. This is the same strategy used by userspace, so apply it by
default. A follow-on patch makes this configurable.
__map_request() is called to re-target osd requests in case the
available osds changed. Add a paused field to a ceph_osd_request, and
set it whenever an appropriate osd map flag is set. Avoid queueing
paused requests in __map_request(), but force them to be resent if
they become unpaused.
Also subscribe to the next osd map from the monitor if any of these
flags are set, so paused requests can be unblocked as soon as
possible.
Fixes: http://tracker.ceph.com/issues/6079
Reviewed-by: Sage Weil <sage@inktank.com>
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
Wake up possible waiters, invoke the call back if any, unregister the request
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Now that we've got a mechanism for immutable biovecs -
bi_iter.bi_bvec_done - we need to convert drivers to use primitives that
respect it instead of using the bvec array directly.
Signed-off-by: Kent Overstreet <kmo@daterainc.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Sage Weil <sage@inktank.com>
Cc: ceph-devel@vger.kernel.org
There are a mix of function prototypes with and without extern
in the kernel sources. Standardize on not using extern for
function prototypes.
Function prototypes don't need to be written with extern.
extern is assumed by the compiler. Its use is as unnecessary as
using auto to declare automatic/local variables in a block.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull ceph fixes from Sage Weil:
"These fix several bugs with RBD from 3.11 that didn't get tested in
time for the merge window: some error handling, a use-after-free, and
a sequencing issue when unmapping and image races with a notify
operation.
There is also a patch fixing a problem with the new ceph + fscache
code that just went in"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
fscache: check consistency does not decrement refcount
rbd: fix error handling from rbd_snap_name()
rbd: ignore unmapped snapshots that no longer exist
rbd: fix use-after free of rbd_dev->disk
rbd: make rbd_obj_notify_ack() synchronous
rbd: complete notifies before cleaning up osd_client and rbd_dev
libceph: add function to ensure notifies are complete
Without a way to flush the osd client's notify workqueue, a watch
event that is unregistered could continue receiving callbacks
indefinitely.
Unregistering the event simply means no new notifies are added to the
queue, but there may still be events in the queue that will call the
watch callback for the event. If the queue is flushed after the event
is unregistered, the caller can be sure no more watch callbacks will
occur for the canceled watch.
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@linaro.org>
Pull ceph updates from Sage Weil:
"This includes both the first pile of Ceph patches (which I sent to
torvalds@vger, sigh) and a few new patches that add support for
fscache for Ceph. That includes a few fscache core fixes that David
Howells asked go through the Ceph tree. (Thanks go to Milosz Tanski
for putting this feature together)
This first batch of patches (included here) had (has) several
important RBD bug fixes, hole punch support, several different
cleanups in the page cache interactions, improvements in the truncate
code (new truncate mutex to avoid shenanigans with i_mutex), and a
series of fixes in the synchronous striping read/write code.
On top of that is a random collection of small fixes all across the
tree (error code checks and error path cleanup, obsolete wq flags,
etc)"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client: (43 commits)
ceph: use d_invalidate() to invalidate aliases
ceph: remove ceph_lookup_inode()
ceph: trivial buildbot warnings fix
ceph: Do not do invalidate if the filesystem is mounted nofsc
ceph: page still marked private_2
ceph: ceph_readpage_to_fscache didn't check if marked
ceph: clean PgPrivate2 on returning from readpages
ceph: use fscache as a local presisent cache
fscache: Netfs function for cleanup post readpages
FS-Cache: Fix heading in documentation
CacheFiles: Implement interface to check cache consistency
FS-Cache: Add interface to check consistency of a cached object
rbd: fix null dereference in dout
rbd: fix buffer size for writes to images with snapshots
libceph: use pg_num_mask instead of pgp_num_mask for pg.seed calc
rbd: fix I/O error propagation for reads
ceph: use vfs __set_page_dirty_nobuffers interface instead of doing it inside filesystem
ceph: allow sync_read/write return partial successed size of read/write.
ceph: fix bugs about handling short-read for sync read mode.
ceph: remove useless variable revoked_rdcache
...
Fix a typo that used the wrong bitmask for the pg.seed calculation. This
is normally unnoticed because in most cases pg_num == pgp_num. It is, however,
a bug that is easily corrected.
CC: stable@vger.kernel.org
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <alex.elder@linary.org>
create_singlethread_workqueue() returns NULL on error, and it doesn't
return ERR_PTRs.
I tweaked the error handling a little to be consistent with earlier in
the function.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Sage Weil <sage@inktank.com>
There are two places where we read "nr_maps" if both of them are set to
zero then we would hit a NULL dereference here.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Sage Weil <sage@inktank.com>
We've tried to fix the error paths in this function before, but there
is still a hidden goto in the ceph_decode_need() macro which goes to the
wrong place. We need to release the "req" and unlock a mutex before
returning.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Sage Weil <sage@inktank.com>
This patch implements fallocate and punch hole support for Ceph kernel client.
Signed-off-by: Li Wang <liwang@ubuntukylin.com>
Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
dbf2576e37 ("workqueue: make all workqueues non-reentrant") made
WQ_NON_REENTRANT no-op and the flag is going away. Remove its usages.
This patch doesn't introduce any behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Sage Weil <sage@inktank.com>
Cc: ceph-devel@vger.kernel.org
For nofail == false request, if __map_request failed, the caller does
cleanup work, like releasing the relative pages. It doesn't make any sense
to retry this request.
CC: stable@vger.kernel.org
Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Several call sites use the hardcoded following condition :
sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)
Lets use a helper because TCP_NOTSENT_LOWAT support will change this
condition for TCP sockets.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can't use !req->r_sent to check if OSD request is sent for the
first time, this is because __cancel_request() zeros req->r_sent
when OSD map changes. Rather than adding a new variable to struct
ceph_osd_request to indicate if it's sent for the first time, We
can call the unsafe callback only when unsafe OSD reply is received.
If OSD's first reply is safe, just skip calling the unsafe callback.
The purpose of unsafe callback is adding unsafe request to a list,
so that fsync(2) can wait for the safe reply. fsync(2) doesn't need
to wait for a write(2) that hasn't returned yet. So it's OK to add
request to the unsafe list when the first OSD reply is received.
(ceph_sync_write() returns after receiving the first OSD reply)
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
A malicious monitor can craft an auth reply message that could cause a
NULL function pointer dereference in the client's kernel.
To prevent this, the auth_none protocol handler needs an empty
ceph_auth_client_ops->build_request() function.
CVE-2013-1059
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Chanam Park <chanam.park@hkpco.kr>
Reviewed-by: Seth Arnold <seth.arnold@canonical.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Cc: stable@vger.kernel.org
handle_reply() calls complete_request() only if the first OSD reply
has ONDISK flag.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
If an osd client response message arrives that has a front section
that's too big for the buffer set aside to receive it, a warning
gets reported and a new buffer is allocated.
The warning says nothing about which connection had the problem.
Add the peer type and number to what gets reported, to be a bit more
informative.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
When an osd request is set to linger, the osd client holds onto the
request so it can be re-submitted following certain osd map changes.
The osd client holds a reference to the request until it is
unregistered. This is used by rbd for watch requests.
Currently, the reference is taken when the request is marked with
the linger flag. This means that if an error occurs after that
time but before the the request completes successfully, that
reference is leaked.
There's really no reason to take the reference until the request is
registered in the the osd client's list of lingering requests, and
that only happens when the lingering (watch) request completes
successfully.
So take that reference only when it gets registered following
succesful completion, and drop it (as before) when the request
gets unregistered. This avoids the reference problem on error
in rbd.
Rearrange ceph_osdc_unregister_linger_request() to avoid using
the request pointer after it may have been freed.
And hold an extra reference in kick_requests() while handling
a linger request that has not yet been registered, to ensure
it doesn't go away.
This resolves:
http://tracker.ceph.com/issues/3859
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An osd client has a red-black tree describing its osds, and
occasionally we would get crashes due to one of these trees tree
becoming corrupt somehow.
The problem turned out to be that reset_changed_osds() was being
called without protection of the osd client request mutex. That
function would call __reset_osd() for any osd that had changed, and
__reset_osd() would call __remove_osd() for any osd with no
outstanding requests, and finally __remove_osd() would remove the
corresponding entry from the red-black tree. Thus, the tree was
getting modified without having any lock protection, and was
vulnerable to problems due to concurrent updates.
This appears to be the only osd tree updating path that has this
problem. It can be fairly easily fixed by moving the call up
a few lines, to just before the request mutex gets dropped
in kick_requests().
This resolves:
http://tracker.ceph.com/issues/5043
Cc: stable@vger.kernel.org # 3.4+
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The rbd code has a need to be able to restart an osd request that
has already been started and completed once before. This currently
wouldn't work right because the osd client code assumes an osd
request will be started exactly once Certain fields in a request
are never cleared and this leads to trouble if you try to reuse it.
Specifically, the r_sent, r_got_reply, and r_completed fields are
never cleared. The r_sent field records the osd incarnation at the
time the request was sent to that osd. If that's non-zero, the
message won't get re-mapped to a target osd properly, and won't be
put on the unsafe requests list the first time it's sent as it
should. The r_got_reply field is used in handle_reply() to ensure
the reply to a request is processed only once. And the r_completed
field is used for lingering requests to avoid calling the callback
function every time the osd client re-sends the request on behalf of
its initiator.
Each osd request passes through ceph_osdc_start_request() when
responsibility for the request is handed over to the osd client for
completion. We can safely zero these three fields there each time a
request gets started.
One last related change--clear the r_linger flag when a request
is no longer registered as a linger request.
This resolves:
http://tracker.ceph.com/issues/5026
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Create a slab cache to manage allocation of ceph_osdc_request
structures.
This resolves:
http://tracker.ceph.com/issues/3926
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Create a slab cache to manage ceph_msg_data structure allocation.
This is part of:
http://tracker.ceph.com/issues/3926
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Create a slab cache to manage ceph_msg structure allocation.
This is part of:
http://tracker.ceph.com/issues/3926
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This creates a new source file "net/ceph/snapshot.c" to contain
utility routines related to ceph snapshot contexts. The main
motivation was to define ceph_create_snap_context() as a common way
to create these structures, but I've moved the definitions of
ceph_get_snap_context() and ceph_put_snap_context() there too.
(The benefit of inlining those is very small, and I'd rather
keep this collection of functions together.)
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
A WATCH op includes an object version. The version that's supplied
is incorrectly byte-swapped osd_req_op_watch_init() where it's first
assigned (it's been this way since that code was first added).
The result is that the version sent to the osd is wrong, because
that value gets byte-swapped again in osd_req_encode_op(). This
is the source of a sparse warning related to improper byte order in
the assignment.
The approach of using the version to avoid a race is deprecated
(see http://tracker.ceph.com/issues/3871), and the watch parameter
is no longer even examined by the osd. So fix the assignment in
osd_req_op_watch_init() so it no longer does the byte swap.
This resolves:
http://tracker.ceph.com/issues/3847
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Add the ability to provide an array of pages as outbound request
data for object class method calls.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This patch makes four small changes in the ceph messenger.
While getting copyup functionality working I found two bugs in the
messenger. Existing paths through the code did not trigger these
problems, but they're fixed here:
- In ceph_msg_data_pagelist_cursor_init(), the cursor's
last_piece field was being checked against the length
supplied. This was OK until this commit: ccba6d98 libceph:
implement multiple data items in a message That commit changed
the cursor init routines to allow lengths to be supplied that
exceeded the size of the current data item. Because of this,
we have to use the assigned cursor resid field rather than the
provided length in determining whether the cursor points to
the last piece of a data item.
- In ceph_msg_data_add_pages(), a BUG_ON() was erroneously
catching attempts to add page data to a message if the message
already had data assigned to it. That was OK until that same
commit, at which point it was fine for messages to have
multiple data items. It slipped through because that BUG_ON()
call was present twice in that function. (You can never be too
careful.)
In addition two other minor things are changed:
- In ceph_msg_data_cursor_init(), the local variable "data" was
getting assigned twice.
- In ceph_msg_data_advance(), it was assumed that the
type-specific advance routine would set new_piece to true
after it advanced past the last piece. That may have been
fine, but since we check for that case we might as well set it
explicitly in ceph_msg_data_advance().
This resolves:
http://tracker.ceph.com/issues/4762
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Allow osd request ops that aren't otherwise structured (not class,
extent, or watch ops) to specify "raw" data to be used to hold
incoming data for the op. Make use of this capability for the osd
STAT op.
Prefix the name of the private function osd_req_op_init() with "_",
and expose a new function by that (earlier) name whose purpose is to
initialize osd ops with (only) implied data.
For now we'll just support the use of a page array for an osd op
with incoming raw data.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
There are a bunch of functions defined to encapsulate getting the
address of a data field for a particular op in an osd request.
They're all defined the same way, so create a macro to take the
place of all of them.
Two of these are used outside the osd client code, so preserve them
(but convert them to use the new macro internally). Stop exporting
the ones that aren't used elsewhere.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
In the incremental move toward supporting distinct data items in an
osd request some of the functions had "write_request" parameters to
indicate, basically, whether the data belonged to in_data or the
out_data. Now that we maintain the data fields in the op structure
there is no need to indicate the direction, so get rid of the
"write_request" parameters.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An osd request currently has two callbacks. They inform the
initiator of the request when we've received confirmation for the
target osd that a request was received, and when the osd indicates
all changes described by the request are durable.
The only time the second callback is used is in the ceph file system
for a synchronous write. There's a race that makes some handling of
this case unsafe. This patch addresses this problem. The error
handling for this callback is also kind of gross, and this patch
changes that as well.
In ceph_sync_write(), if a safe callback is requested we want to add
the request on the ceph inode's unsafe items list. Because items on
this list must have their tid set (by ceph_osd_start_request()), the
request added *after* the call to that function returns. The
problem with this is that there's a race between starting the
request and adding it to the unsafe items list; the request may
already be complete before ceph_sync_write() even begins to put it
on the list.
To address this, we change the way the "safe" callback is used.
Rather than just calling it when the request is "safe", we use it to
notify the initiator the bounds (start and end) of the period during
which the request is *unsafe*. So the initiator gets notified just
before the request gets sent to the osd (when it is "unsafe"), and
again when it's known the results are durable (it's no longer
unsafe). The first call will get made in __send_request(), just
before the request message gets sent to the messenger for the first
time. That function is only called by __send_queued(), which is
always called with the osd client's request mutex held.
We then have this callback function insert the request on the ceph
inode's unsafe list when we're told the request is unsafe. This
will avoid the race because this call will be made under protection
of the osd client's request mutex. It also nicely groups the setup
and cleanup of the state associated with managing unsafe requests.
The name of the "safe" callback field is changed to "unsafe" to
better reflect its new purpose. It has a Boolean "unsafe" parameter
to indicate whether the request is becoming unsafe or is now safe.
Because the "msg" parameter wasn't used, we drop that.
This resolves the original problem reportedin:
http://tracker.ceph.com/issues/4706
Reported-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Right now the data for a method call is specified via a pointer and
length, and it's copied--along with the class and method name--into
a pagelist data item to be sent to the osd. Instead, encode the
data in a data item separate from the class and method names.
This will allow large amounts of data to be supplied to methods
without copying. Only rbd uses the class functionality right now,
and when it really needs this it will probably need to use a page
array rather than a page list. But this simple implementation
demonstrates the functionality on the osd client, and that's enough
for now.
This resolves:
http://tracker.ceph.com/issues/4104
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Change the names of the functions that put data on a pagelist to
reflect that we're adding to whatever's already there rather than
just setting it to the one thing. Currently only one data item is
ever added to a message, but that's about to change.
This resolves:
http://tracker.ceph.com/issues/2770
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This patch adds support to the messenger for more than one data item
in its data list.
A message data cursor has two more fields to support this:
- a count of the number of bytes left to be consumed across
all data items in the list, "total_resid"
- a pointer to the head of the list (for validation only)
The cursor initialization routine has been split into two parts: the
outer one, which initializes the cursor for traversing the entire
list of data items; and the inner one, which initializes the cursor
to start processing a single data item.
When a message cursor is first initialized, the outer initialization
routine sets total_resid to the length provided. The data pointer
is initialized to the first data item on the list. From there, the
inner initialization routine finishes by setting up to process the
data item the cursor points to.
Advancing the cursor consumes bytes in total_resid. If the resid
field reaches zero, it means the current data item is fully
consumed. If total_resid indicates there is more data, the cursor
is advanced to point to the next data item, and then the inner
initialization routine prepares for using that. (A check is made at
this point to make sure we don't wrap around the front of the list.)
The type-specific init routines are modified so they can be given a
length that's larger than what the data item can support. The resid
field is initialized to the smaller of the provided length and the
length of the entire data item.
When total_resid reaches zero, we're done.
This resolves:
http://tracker.ceph.com/issues/3761
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
In place of the message data pointer, use a list head which links
through message data items. For now we only support a single entry
on that list.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Rather than having a ceph message data item point to the cursor it's
associated with, have the cursor point to a data item. This will
allow a message cursor to be used for more than one data item.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
A message will only be processing a single data item at a time, so
there's no need for each data item to have its own cursor.
Move the cursor embedded in the message data structure into the
message itself. To minimize the impact, keep the data->cursor
field, but make it be a pointer to the cursor in the message.
Move the definition of ceph_msg_data above ceph_msg_data_cursor so
the cursor can point to the data without a forward definition rather
than vice-versa.
This and the upcoming patches are part of:
http://tracker.ceph.com/issues/3761
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The bio is the only data item type that doesn't record its full
length. Fix that.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
We know the length of our message buffers. If we get a message
that's too long, just dump it and ignore it. If skip was set
then con->in_msg won't be valid, so be careful not to dereference
a null pointer in the process.
This resolves:
http://tracker.ceph.com/issues/4664
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This patch:
15a0d7b libceph: record message data length
did not enclose some bio-specific code inside CONFIG_BLOCK as
it should have. Fix that.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Finally! Convert the osd op data pointers into real structures, and
make the switch over to using them instead of having all ops share
the in and/or out data structures in the osd request.
Set up a new function to traverse the set of ops and release any
data associated with them (pages).
This and the patches leading up to it resolve:
http://tracker.ceph.com/issues/4657
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Still using the osd request r_data_in and r_data_out pointer, but
we're basically only referring to it via the data pointers in the
osd ops. And we're transferring that information to the request
or reply message only when the op indicates it's needed, in
osd_req_encode_op().
To avoid a forward reference, ceph_osdc_msg_data_set() was moved up
in the file.
Don't bother calling ceph_osd_data_init(), in ceph_osd_alloc(),
because the ops array will already be zeroed anyway.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This ends up being a rather large patch but what it's doing is
somewhat straightforward.
Basically, this is replacing two calls with one. The first of the
two calls is initializing a struct ceph_osd_data with data (either a
page array, a page list, or a bio list); the second is setting an
osd request op so it associates that data with one of the op's
parameters. In place of those two will be a single function that
initializes the op directly.
That means we sort of fan out a set of the needed functions:
- extent ops with pages data
- extent ops with pagelist data
- extent ops with bio list data
and
- class ops with page data for receiving a response
We also have define another one, but it's only used internally:
- class ops with pagelist data for request parameters
Note that we *still* haven't gotten rid of the osd request's
r_data_in and r_data_out fields. All the osd ops refer to them for
their data. For now, these data fields are pointers assigned to the
appropriate r_data_* field when these new functions are called.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
All calls of ceph_osdc_start_request() are preceded (in the case of
rbd, almost) immediately by a call to ceph_osdc_build_request().
Move the build calls at the top of ceph_osdc_start_request() out of
there and into the ceph_osdc_build_request(). Nothing prevents
moving these calls to the top of ceph_osdc_build_request(), either
(and we're going to want them there in the next patch) so put them
at the top.
This and the next patch are related to:
http://tracker.ceph.com/issues/4657
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This simply moves ceph_osdc_build_request() later in its source
file without any change. Done as a separate patch to facilitate
review of the change in the next patch.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An object class method is formatted using a pagelist which contains
the class name, the method name, and the data concatenated into an
osd request's outbound data.
Currently when a class op is initialized in osd_req_op_cls_init(),
the lengths of and pointers to these three items are recorded.
Later, when the op is getting formatted into the request message, a
new pagelist is created and that is when these items get copied into
the pagelist.
This patch makes it so the pagelist to hold these items is created
when the op is initialized instead.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An osd request now holds all of its source op structures, and every
place that initializes one of these is in fact initializing one
of the entries in the the osd request's array.
So rather than supplying the address of the op to initialize, have
caller specify the osd request and an indication of which op it
would like to initialize. This better hides the details the
op structure (and faciltates moving the data pointers they use).
Since osd_req_op_init() is a common routine, and it's not used
outside the osd client code, give it static scope. Also make
it return the address of the specified op (so all the other
init routines don't have to repeat that code).
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An extent type osd operation currently implies that there will
be corresponding data supplied in the data portion of the request
(for write) or response (for read) message. Similarly, an osd class
method operation implies a data item will be supplied to receive
the response data from the operation.
Add a ceph_osd_data pointer to each of those structures, and assign
it to point to eithre the incoming or the outgoing data structure in
the osd message. The data is not always available when an op is
initially set up, so add two new functions to allow setting them
after the op has been initialized.
Begin to make use of the data item pointer available in the osd
operation rather than the request data in or out structure in
places where it's convenient. Add some assertions to verify
pointers are always set the way they're expected to be.
This is a sort of stepping stone toward really moving the data
into the osd request ops, to allow for some validation before
making that jump.
This is the first in a series of patches that resolve:
http://tracker.ceph.com/issues/4657
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
There are fields "indata" and "indata_len" defined the ceph osd
request op structure. The "in" part is with from the point of view
of the osd server, but is a little confusing here on the client
side. Change their names to use "request" instead of "in" to
indicate that it defines data provided with the request (as opposed
the data returned in the response).
Rename the local variable in osd_req_encode_op() to match.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An osd request keeps a pointer to the osd operations (ops) array
that it builds in its request message.
In order to allow each op in the array to have its own distinct
data, we will need to keep track of each op's data, and that
information does not go over the wire.
As long as we're tracking the data we might as well just track the
entire (source) op definition for each of the ops. And if we're
doing that, we'll have no more need to keep a pointer to the
wire-encoded version.
This patch makes the array of source ops be kept with the osd
request structure, and uses that instead of the version encoded in
the message in places where that was previously used. The array
will be embedded in the request structure, and the maximum number of
ops we ever actually use is currently 2. So reduce CEPH_OSD_MAX_OP
to 2 to reduce the size of the structure.
The result of doing this sort of ripples back up, and as a result
various function parameters and local variables become unnecessary.
Make r_num_ops be unsigned, and move the definition of struct
ceph_osd_req_op earlier to ensure it's defined where needed.
It does not yet add per-op data, that's coming soon.
This resolves:
http://tracker.ceph.com/issues/4656
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
One more osd data helper, which returns the length of the
data item, regardless of its type.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Define ceph_osd_data_init() and ceph_osd_data_release() to clean up
a little code.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Define and use functions that encapsulate the initializion of a
ceph_osd_data structure.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This is a simple change, extracting the number of incoming data
bytes just once in handle_reply().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
In prepare_message_data(), the length used to initialize the cursor
is taken from the header of the message provided. I'm working
toward not using the header data length field to determine length in
outbound messages, and this is a step in that direction. For
inbound messages this will be set to be the actual number of bytes
that are arriving (which may be less than the total size of the data
buffer available).
This resolves:
http://tracker.ceph.com/issues/4589
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Hold off building the osd request message in ceph_writepages_start()
until just before it will be submitted to the osd client for
execution.
We'll still create the request and allocate the page pointer array
after we learn we have at least one page to write. A local variable
will be used to keep track of the allocated array of pages. Wait
until just before submitting the request for assigning that page
array pointer to the request message.
Create ands use a new function osd_req_op_extent_update() whose
purpose is to serve this one spot where the length value supplied
when an osd request's op was initially formatted might need to get
changed (reduced, never increased) before submitting the request.
Previously, ceph_writepages_start() assigned the message header's
data length because of this update. That's no longer necessary,
because ceph_osdc_build_request() will recalculate the right
value to use based on the content of the ops in the request.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Defer building the osd request until just before submitting it in
all callers except ceph_writepages_start(). (That caller will be
handed in the next patch.)
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This patch moves the call to ceph_osdc_build_request() out of
ceph_osdc_new_request() and into its caller.
This is in order to defer formatting osd operation information into
the request message until just before request is started.
The only unusual (ab)user of ceph_osdc_build_request() is
ceph_writepages_start(), where the final length of write request may
change (downward) based on the current inode size or the oldest
snapshot context with dirty data for the inode.
The remaining callers don't change anything in the request after has
been built.
This means the ops array is now supplied by the caller. It also
means there is no need to pass the mtime to ceph_osdc_new_request()
(it gets provided to ceph_osdc_build_request()). And rather than
passing a do_sync flag, have the number of ops in the ops array
supplied imply adding a second STARTSYNC operation after the READ or
WRITE requested.
This and some of the patches that follow are related to having the
messenger (only) be responsible for filling the content of the
message header, as described here:
http://tracker.ceph.com/issues/4589
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Keep track of the length of the data portion for a message in a
separate field in the ceph_msg structure. This information has
been maintained in wire byte order in the message header, but
that's going to change soon.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
A field in an osd request keeps track of whether a connection is
currently filling the request's reply message. This patch gets rid
of that field.
An osd request includes two messages--a request and a reply--and
they're both associated with the connection that existed to its
the target osd at the time the request was created.
An osd request can be dropped early, even when it's in flight.
And at that time both messages are released. It's possible the
reply message has been supplied to its connection to receive
an incoming response message at the time the osd request gets
dropped. So ceph_osdc_release_request() revokes that message
from the connection before releasing it so things get cleaned up
properly.
Previously this may have caused a problem, because the connection
that a message was associated with might have gone away before the
revoke request. And to avoid any problems using that connection,
the osd client held a reference to it when it supplies its response
message.
However since this commit:
38941f80 libceph: have messages point to their connection
all messages hold a reference to the connection they are associated
with whenever the connection is actively operating on the message
(i.e. while the message is queued to send or sending, and when it
data is being received into it). And if a message has no connection
associated with it, ceph_msg_revoke_incoming() won't do anything
when asked to revoke it.
As a result, there is no need to keep an additional reference to the
connection associated with a message when we hand the message to the
messenger when it calls our alloc_msg() method to receive something.
If the connection *were* operating on it, it would have its own
reference, and if not, there's no work to be done when we need to
revoke it.
So get rid of the osd request's r_con_filling_msg field.
This resolves:
http://tracker.ceph.com/issues/4647
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
There are two basically identical definitions of __decode_pgid()
in libceph, one in "net/ceph/osdmap.c" and the other in
"net/ceph/osd_client.c". Get rid of both, and instead define
a single inline version in "include/linux/ceph/osdmap.h".
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The osd client mutex is acquired just before getting a reference to
a request in handle_reply(). However the error paths after that
don't drop the mutex before returning as they should.
Drop the mutex after dropping the request reference. Also add a
bad_mutex label at that point and use it so the failed request
lookup case can be handled with the rest.
This resolves:
http://tracker.ceph.com/issues/4615
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Use osd_req_op_extent_init() in ceph_osdc_new_request() to
initialize the one or two ops built in that function.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
All callers of ceph_osd_new_request() pass either CEPH_OSD_OP_READ
or CEPH_OSD_OP_WRITE as the opcode value. The function assumes it
by filling in the extent fields in the ops array it builds. So just
assert that is the case, and don't bother calling op_has_extent()
before filling in the first osd operation in the array.
Define some local variables to gather the information to fill into
the first op, and then fill in the op array all in one place.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The ceph_osdc_new_request() an array of osd operations is built up
and filled in partially within that function and partially in the
called function calc_layout(). Move the latter part back out to
ceph_osdc_new_request() so it's all done in one place. This makes
it unnecessary to pass the op pointer to calc_layout(), so get rid
of that parameter.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The purpose of calc_layout() is to determine, given a file offset
and length and a layout describing the placement of file data across
objects, where in "object space" that data resides.
Specifically, it determines which object should hold the first part
of the specified range of file data, and the offset and length of
data within that object. The length will not exceed the bounds
of the object, and the caller is informed of that maximum length.
Add two parameters to calc_layout() to allow the object-relative
offset and length to be passed back to the caller.
This is the first steps toward having ceph_osdc_new_request() build
its osd op structure using osd_req_op_extent_init().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The rbd code has a function that allocates and populates a
ceph_osd_req_op structure (the in-core version of an osd request
operation). When reviewed, Josh suggested two things: that the
big varargs function might be better split into type-specific
functions; and that this functionality really belongs in the osd
client rather than rbd.
This patch implements both of Josh's suggestions. It breaks
up the rbd function into separate functions and defines them
in the osd client module as exported interfaces. Unlike the
rbd version, however, the functions don't allocate an osd_req_op
structure; they are provided the address of one and that is
initialized instead.
The rbd function has been eliminated and calls to it have been
replaced by calls to the new routines. The rbd code now now use a
stack (struct) variable to hold the op rather than allocating and
freeing it each time.
For now only the capabilities used by rbd are implemented.
Implementing all the other osd op types, and making the rest of the
code use it will be done separately, in the next few patches.
Note that only the extent, cls, and watch portions of the
ceph_osd_req_op structure are currently used. Delete the others
(xattr, pgls, and snap) from its definition so nobody thinks it's
actually implemented or needed. We can add it back again later
if needed, when we know it's been tested.
This (and a few follow-on patches) resolves:
http://tracker.ceph.com/issues/3861
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Define a separate function to determine the validity of an opcode,
and use it inside osd_req_encode_op() in order to unclutter that
function.
Don't update the destination op at all--and return zero--if an
unsupported or unrecognized opcode is seen in osd_req_encode_op().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
In ceph_osdc_build_request() there is a call to cpu_to_le16() which
provides a 64-bit value as its argument. Because of the implied
byte swapping going on it looked pretty suspect to me.
At the moment it turns out the behavior is well defined, but masking
off those bottom bits explicitly eliminates this distraction, and is
in fact more directly related to the purpose of the message header's
data_off field.
This resolves:
http://tracker.ceph.com/issues/4125
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
When a cursor for a page array data message is initialized it needs
to determine the initial value for cursor->last_piece. Currently it
just checks if length is less than a page, but that's not correct.
The data in the first page in the array will be offset by a page
offset based on the alignment recorded for the data. (All pages
thereafter will be aligned at the base of the page, so there's
no need to account for this except for the first page.)
Because this was wrong, there was a case where the length of a piece
would be calculated as all of the residual bytes in the message and
that plus the page offset could exceed the length of a page.
So fix this case. Make sure the sum won't wrap.
This resolves a third issue described in:
http://tracker.ceph.com/issues/4598
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Currently ceph_msg_data_pages_advance() allows the page offset value
to be PAGE_SIZE, apparently assuming ceph_msg_data_pages_next() will
treat it as 0. But that doesn't happen, and the result led to a
helpful assertion failure.
Change ceph_msg_data_pages_advance() to truncate the offset to 0
before returning if it reaches PAGE_SIZE.
Make a few other minor adjustments in this area (comments and a
better assertion) while modifying it.
This resolves a second issue described in:
http://tracker.ceph.com/issues/4598
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
It's OK for the result of a read to come back with fewer bytes than
were requested. So don't trigger a BUG() in that case when
initializing the data cursor.
This resolves the first problem described in:
http://tracker.ceph.com/issues/4598
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Begin the transition from a single message data item to a list of
them by replacing the "data" structure in a message with a pointer
to a ceph_msg_data structure.
A null pointer will indicate the message has no data; replace the
use of ceph_msg_has_data() with a simple check for a null pointer.
Create functions ceph_msg_data_create() and ceph_msg_data_destroy()
to dynamically allocate and free a data item structure of a given type.
When a message has its data item "set," allocate one of these to
hold the data description, and free it when the last reference to
the message is dropped.
This partially resolves:
http://tracker.ceph.com/issues/4429
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The *_msg_pos_next() functions do little more than call
ceph_msg_data_advance(). Replace those wrapper functions with
a simple call to ceph_msg_data_advance().
This cleanup is related to:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
In write_partial_message_data() we aggregate the crc for the data
portion of the message as each new piece of the data item is
encountered. Because it was computed *before* sending the data, if
an attempt to send a new piece resulted in 0 bytes being sent, the
crc crc across that piece would erroneously get computed again and
added to the aggregate result. This would occasionally happen in
the evnet of a connection failure.
The crc value isn't really needed until the complete value is known
after sending all data, so there's no need to compute it before
sending.
So don't calculate the crc for a piece until *after* we know at
least one byte of it has been sent. That will avoid this problem.
This resolves:
http://tracker.ceph.com/issues/4450
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
The only remaining field in the ceph_msg_pos structure is
did_page_crc. In the new cursor model of things that flag (or
something like it) belongs in the cursor.
Define a new field "need_crc" in the cursor (which applies to all
types of data) and initialize it to true whenever a cursor is
initialized.
In write_partial_message_data(), the data CRC still will be computed
as before, but it will check the cursor->need_crc field to determine
whether it's needed. Any time the cursor is advanced to a new piece
of a data item, need_crc will be set, and this will cause the crc
for that entire piece to be accumulated into the data crc.
In write_partial_message_data() the intermediate crc value is now
held in a local variable so it doesn't have to be byte-swapped so
many times. In read_partial_msg_data() we do something similar
(but mainly for consistency there).
With that, the ceph_msg_pos structure can go away, and it no longer
needs to be passed as an argument to prepare_message_data().
This cleanup is related to:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
All but one of the fields in the ceph_msg_pos structure are now
never used (only assigned), so get rid of them. This allows
several small blocks of code to go away.
This is cleanup of old code related to:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Use the "resid" field of a cursor rather than finding when the
message data position has moved up to meet the data length to
determine when all data has been sent or received in
write_partial_message_data() and read_partial_msg_data().
This is cleanup of old code related to:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
It turns out that only one of the data item types is ever used at
any one time in a single message (currently).
- A page array is used by the osd client (on behalf of the file
system) and by rbd. Only one osd op (and therefore at most
one data item) is ever used at a time by rbd. And the only
time the file system sends two, the second op contains no
data.
- A bio is only used by the rbd client (and again, only one
data item per message)
- A page list is used by the file system and by rbd for outgoing
data, but only one op (and one data item) at a time.
We can therefore collapse all three of our data item fields into a
single field "data", and depend on the messenger code to properly
handle it based on its type.
This allows us to eliminate quite a bit of duplicated code.
This is related to:
http://tracker.ceph.com/issues/4429
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Now that read_partial_message_pages() and read_partial_message_bio()
are literally identical functions we can factor them out. They're
pretty simple as well, so just move their relevant content into
read_partial_msg_data().
This is and previous patches together resolve:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
There is handling in write_partial_message_data() for the case where
only the length of--and no other information about--the data to be
sent has been specified. It uses the zero page as the source of
data to send in this case.
This case doesn't occur. All message senders set up a page array,
pagelist, or bio describing the data to be sent. So eliminate the
block of code that handles this (but check and issue a warning for
now, just in case it happens for some reason).
This resolves:
http://tracker.ceph.com/issues/4426
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The cursor code for a page array selects the right page, page
offset, and length to use for a ceph_tcp_recvpage() call, so
we can use it to replace a block in read_partial_message_pages().
This partially resolves:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The bio_iter and bio_seg fields in a message are no longer used, we
use the cursor instead. So get rid of them and the functions that
operate on them them.
This is related to:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Replace the use of the information in con->in_msg_pos for incoming
bio data. The old in_msg_pos and the new cursor mechanism do
basically the same thing, just slightly differently.
The main functional difference is that in_msg_pos keeps track of the
length of the complete bio list, and assumed it was fully consumed
when that many bytes had been transferred. The cursor does not assume
a length, it simply consumes all bytes in the bio list. Because the
only user of bio data is the rbd client, and because the length of a
bio list provided by rbd client always matches the number of bytes
in the list, both ways of tracking length are equivalent.
In addition, for in_msg_pos the initial bio vector is selected as
the initial value of the bio->bi_idx, while the cursor assumes this
is zero. Again, the rbd client always passes 0 as the initial index
so the effect is the same.
Other than that, they basically match:
in_msg_pos cursor
---------- ------
bio_iter bio
bio_seg vec_index
page_pos page_offset
The in_msg_pos field is initialized by a call to init_bio_iter().
The bio cursor is initialized by ceph_msg_data_cursor_init().
Both now happen in the same spot, in prepare_message_data().
The in_msg_pos field is advanced by a call to in_msg_pos_next(),
which updates page_pos and calls iter_bio_next() to move to the next
bio vector, or to the next bio in the list. The cursor is advanced
by ceph_msg_data_advance(). That isn't currently happening so
add a call to that in in_msg_pos_next().
Finally, the next piece of data to use for a read is determined
by a bunch of lines in read_partial_message_bio(). Those can be
replaced by an equivalent ceph_msg_data_bio_next() call.
This partially resolves:
http://tracker.ceph.com/issues/4428
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
All of the data types can use this, not just the page array. Until
now, only the bio type doesn't have it available, and only the
initiator of the request (the rbd client) is able to supply the
length of the full request without re-scanning the bio list. Change
the cursor init routines so the length is supplied based on the
message header "data_len" field, and use that length to intiialize
the "resid" field of the cursor.
In addition, change the way "last_piece" is defined so it is based
on the residual number of bytes in the original request. This is
necessary (at least for bio messages) because it is possible for
a read request to succeed without consuming all of the space
available in the data buffer.
This resolves:
http://tracker.ceph.com/issues/4427
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The value passed for "pages" in read_partial_message_pages() is
always the pages pointer from the incoming message, which can be
derived inside that function. So just get rid of the parameter.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
When the last reference to a ceph message is dropped,
ceph_msg_last_put() is called to clean things up.
For "normal" messages (allocated via ceph_msg_new() rather than
being allocated from a memory pool) it's sufficient to just release
resources. But for a mempool-allocated message we actually have to
re-initialize the data fields in the message back to initial state
so they're ready to go in the event the message gets reused.
Some of this was already done; this fleshes it out so it's done
more completely.
This resolves:
http://tracker.ceph.com/issues/4540
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
An osd expects the transaction ids of arriving request messages from
a given client to a given osd to increase monotonically. So the osd
client needs to send its requests in ascending tid order.
The transaction id for a request is set at the time it is
registered, in __register_request(). This is also where the request
gets placed at the end of the osd client's unsent messages list.
At the end of ceph_osdc_start_request(), the request message for a
newly-mapped osd request is supplied to the messenger to be sent
(via __send_request()). If any other messages were present in the
osd client's unsent list at that point they would be sent *after*
this new request message.
Because those unsent messages have already been registered, their
tids would be lower than the newly-mapped request message, and
sending that message first can violate the tid ordering rule.
Rather than sending the new request only, send all queued requests
(including the new one) at that point in ceph_osdc_start_request().
This ensures the tid ordering property is preserved.
With this in place, all messages should now be sent in tid order
regardless of whether they're being sent for the first time or
re-sent as a result of a call to osd_reset().
This resolves:
http://tracker.ceph.com/issues/4392
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-off-by: Sage Weil <sage@inktank.com>
In __map_request(), when adding a request to an osd client's unsent
list, add it to the tail rather than the head. That way the newest
entries (with the highest tid value) will be last.
Maintain an osd's request list in order of increasing tid also.
Finally--to be consistent--maintain an osd client's "notarget" list
in that order as well.
This partially resolves:
http://tracker.ceph.com/issues/4392
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-off-by: Sage Weil <sage@inktank.com>
The osd expects incoming requests for a given object from a given
client to arrive in order, with the tid for each request being
greater than the tid for requests that have already arrived. This
patch fixes two places the osd client might not maintain that
ordering.
For the osd client, the connection fault method is osd_reset().
That function calls __reset_osd() to close and re-open the
connection, then calls __kick_osd_requests() to cause all
outstanding requests for the affected osd to be re-sent after
the connection has been re-established.
When an osd is reset, any in-flight messages will need to be
re-sent. An osd client maintains distinct lists for unsent and
in-flight messages. Meanwhile, an osd maintains a single list of
all its requests (both sent and un-sent). (Each message is linked
into two lists--one for the osd client and one list for the osd.)
To process an osd "kick" operation, the request list for the *osd*
is traversed, and each request is moved off whichever osd *client*
list it was on (unsent or sent) and placed onto the osd client's
unsent list. (It remains where it is on the osd's request list.)
When that is done, osd_reset() calls __send_queued() to cause each
of the osd client's unsent messages to be sent.
OK, with that background...
As the osd request list is traversed each request is prepended to
the osd client's unsent list in the order they're seen. The effect
of this is to reverse the order of these requests as they are put
(back) onto the unsent list.
Instead, build up a list of only the requests for an osd that have
already been sent (by checking their r_sent flag values). Once an
unsent request is found, stop examining requests and prepend the
requests that need re-sending to the osd client's unsent list.
Preserve the original order of requests in the process (previously
re-queued requests were reversed in this process). Because they
have already been sent, they will have lower tids than any request
already present on the unsent list.
Just below that, traverse the linger list in forward order as
before, but add them to the *tail* of the list rather than the head.
These requests get re-registered, and in the process are give a new
(higher) tid, so the should go at the end.
This partially resolves:
http://tracker.ceph.com/issues/4392
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-off-by: Sage Weil <sage@inktank.com>
Since we no longer drop the request mutex between registering and
mapping an osd request in ceph_osdc_start_request(), there is no
chance of a race with kick_requests().
We can now therefore map and send the new request unconditionally
(but we'll issue a warning should it ever occur).
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-off-by: Sage Weil <sage@inktank.com>
One of the first things ceph_osdc_start_request() does is register
the request. It then acquires the osd client's map semaphore and
request mutex and proceeds to map and send the request.
There is no reason the request has to be registered before acquiring
the map semaphore. So hold off doing so until after the map
semaphore is held.
Since register_request() is nothing more than a wrapper around
__register_request(), call the latter function instead, after
acquiring the request mutex.
That leaves register_request() unused, so get rid of it.
This partially resolves:
http://tracker.ceph.com/issues/4392
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-off-by: Sage Weil <sage@inktank.com>
The auth code is called from a variety of contexts, include the mon_client
(protected by the monc's mutex) and the messenger callbacks (currently
protected by nothing). Avoid chaos by protecting all auth state with a
mutex. Nothing is blocking, so this should be simple and lightweight.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
Use wrapper functions that check whether the auth op exists so that callers
do not need a bunch of conditional checks. Simplifies the external
interface.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
Currently the messenger calls out to a get_authorizer con op, which will
create a new authorizer if it doesn't yet have one. In the meantime, when
we rotate our service keys, the authorizer doesn't get updated. Eventually
it will be rejected by the server on a new connection attempt and get
invalidated, and we will then rebuild a new authorizer, but this is not
ideal.
Instead, if we do have an authorizer, call a new update_authorizer op that
will verify that the current authorizer is using the latest secret. If it
is not, we will build a new one that does. This avoids the transient
failure.
This fixes one of the sorry sequence of events for bug
http://tracker.ceph.com/issues/4282
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
We were invalidating the authorizer by removing the ticket handler
entirely. This was effective in inducing us to request a new authorizer,
but in the meantime it mean that any authorizer we generated would get a
new and initialized handler with secret_id=0, which would always be
rejected by the server side with a confusing error message:
auth: could not find secret_id=0
cephx: verify_authorizer could not get service secret for service osd secret_id=0
Instead, simply clear the validity field. This will still induce the auth
code to request a new secret, but will let us continue to use the old
ticket in the meantime. The messenger code will probably continue to fail,
but the exponential backoff will kick in, and eventually the we will get a
new (hopefully more valid) ticket from the mon and be able to continue.
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
We maintain a counter of failed auth attempts to allow us to retry once
before failing. However, if the second attempt succeeds, the flag isn't
cleared, which makes us think auth failed again later when the connection
resets for other reasons (like a socket error).
This is one part of the sorry sequence of events in bug
http://tracker.ceph.com/issues/4282
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
This is an old protocol extension that allows the client and server to
avoid resending old messages after a reconnect (following a socket error).
Instead, the exchange their sequence numbers during the handshake. This
avoids sending a bunch of useless data over the socket.
It has been supported in the server code since v0.22 (Sep 2010).
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
Basically all cases in write_partial_msg_pages() use the cursor, and
as a result we can simplify that function quite a bit.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The wart that is the ceph message trail can now be removed, because
its only user was the osd client, and the previous patch made that
no longer the case.
The result allows write_partial_msg_pages() to be simplified
considerably.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The osd trail is a pagelist, used only for a CALL osd operation
to hold the class and method names, along with any input data for
the call.
It is only currently used by the rbd client, and when it's used it
is the only bit of outbound data in the osd request. Since we
already support (non-trail) pagelist data in a message, we can
just save this outbound CALL data in the "normal" pagelist rather
than the trail, and get rid of the trail entirely.
The existing pagelist support depends on the pagelist being
dynamically allocated, and ownership of it is passed to the
messenger once it's been attached to a message. (That is to say,
the messenger releases and frees the pagelist when it's done with
it). That means we need to dynamically allocate the pagelist also.
Note that we simply assert that the allocation of a pagelist
structure succeeds. Appending to a pagelist might require a dynamic
allocation, so we're already assuming we won't run into trouble
doing so (we're just ignore any failures--and that should be fixed
at some point).
This resolves:
http://tracker.ceph.com/issues/4407
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Add support for recording a ceph pagelist as data associated with an
osd request.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
The length of outgoing data in an osd request is dependent on the
osd ops that are embedded in that request. Each op is encoded into
a request message using osd_req_encode_op(), so that should be used
to determine the amount of outgoing data implied by the op as it
is encoded.
Have osd_req_encode_op() return the number of bytes of outgoing data
implied by the op being encoded, and accumulate and use that in
ceph_osdc_build_request().
As a result, ceph_osdc_build_request() no longer requires its "len"
parameter, so get rid of it.
Using the sum of the op lengths rather than the length provided is
a valid change because:
- The only callers of osd ceph_osdc_build_request() are
rbd and the osd client (in ceph_osdc_new_request() on
behalf of the file system).
- When rbd calls it, the length provided is only non-zero for
write requests, and in that case the single op has the
same length value as what was passed here.
- When called from ceph_osdc_new_request(), (it's not all that
easy to see, but) the length passed is also always the same
as the extent length encoded in its (single) write op if
present.
This resolves:
http://tracker.ceph.com/issues/4406
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Implement and use cursor routines for page array message data items
for outbound message data.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Implement and use cursor routines for bio message data items for
outbound message data.
(See the previous commit for reasoning in support of the changes
in out_msg_pos_next().)
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Switch to using the message cursor for the (non-trail) outgoing
pagelist data item in a message if present.
Notes on the logic changes in out_msg_pos_next():
- only the mds client uses a ceph pagelist for message data;
- if the mds client ever uses a pagelist, it never uses a page
array (or anything else, for that matter) for data in the same
message;
- only the osd client uses the trail portion of a message data,
and when it does, it never uses any other data fields for
outgoing data in the same message; and finally
- only the rbd client uses bio message data (never pagelist).
Therefore out_msg_pos_next() can assume:
- if we're in the trail portion of a message, the message data
pagelist, data, and bio can be ignored; and
- if there is a page list, there will never be any a bio or page
array data, and vice-versa.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This just inserts some infrastructure in preparation for handling
other types of ceph message data items. No functional changes,
just trying to simplify review by separating out some noise.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
This patch lays out the foundation for using generic routines to
manage processing items of message data.
For simplicity, we'll start with just the trail portion of a
message, because it stands alone and is only present for outgoing
data.
First some basic concepts. We'll use the term "data item" to
represent one of the ceph_msg_data structures associated with a
message. There are currently four of those, with single-letter
field names p, l, b, and t. A data item is further broken into
"pieces" which always lie in a single page. A data item will
include a "cursor" that will track state as the memory defined by
the item is consumed by sending data from or receiving data into it.
We define three routines to manipulate a data item's cursor: the
"init" routine; the "next" routine; and the "advance" routine. The
"init" routine initializes the cursor so it points at the beginning
of the first piece in the item. The "next" routine returns the
page, page offset, and length (limited by both the page and item
size) of the next unconsumed piece in the item. It also indicates
to the caller whether the piece being returned is the last one in
the data item.
The "advance" routine consumes the requested number of bytes in the
item (advancing the cursor). This is used to record the number of
bytes from the current piece that were actually sent or received by
the network code. It returns an indication of whether the result
means the current piece has been fully consumed. This is used by
the message send code to determine whether it should calculate the
CRC for the next piece processed.
The trail of a message is implemented as a ceph pagelist. The
routines defined for it will be usable for non-trail pagelist data
as well.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Group the types of message data into an abstract structure with a
type indicator and a union containing fields appropriate to the
type of data it represents. Use this to represent the pages,
pagelist, bio, and trail in a ceph message.
Verify message data is of type NONE in ceph_msg_data_set_*()
routines. Since information about message data of type NONE really
should not be interpreted, get rid of the other assertions in those
functions.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
A ceph message has a data payload portion. The memory for that data
(either the source of data to send or the location to place data
that is received) is specified in several ways. The ceph_msg
structure includes fields for all of those ways, but this
mispresents the fact that not all of them are used at a time.
Specifically, the data in a message can be in:
- an array of pages
- a list of pages
- a list of Linux bios
- a second list of pages (the "trail")
(The two page lists are currently only ever used for outgoing data.)
Impose more structure on the ceph message, making the grouping of
some of these fields explicit. Shorten the name of the
"page_alignment" field.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Define and use macros ceph_msg_has_*() to determine whether to
operate on the pages, pagelist, bio, and trail fields of a message.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Factor out a common block of code that updates a CRC calculation
over a range of data in a page.
This and the preceding patches are related to:
http://tracker.ceph.com/issues/4403
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Define a new function ceph_tcp_recvpage() that behaves in a way
comparable to ceph_tcp_sendpage().
Rearrange the code in both read_partial_message_pages() and
read_partial_message_bio() so they have matching structure,
(similar to what's in write_partial_msg_pages()), and use
this new function.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
