- 16 Oct, 2016 40 commits
-
-
Gregoire Pichon authored
This patch introduces the maximum modify RPCs in flight variable of a mdc client obd device. Its value is set from connection flag and and connection data. It can later be tuned through the max_mod_rpcs_in_flight procfs file. Signed-off-by:
Gregoire Pichon <gregoire.pichon@bull.net> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5319 Reviewed-on: http://review.whamcloud.com/14153Reviewed-by:
Andreas Dilger <andreas.dilger@intel.com> Reviewed-by:
Dmitry Eremin <dmitry.eremin@intel.com> Reviewed-by:
Mike Pershin <mike.pershin@intel.com> Reviewed-by:
Alex Zhuravlev <alexey.zhuravlev@intel.com> Reviewed-by:
Oleg Drokin <oleg.drokin@intel.com> Signed-off-by:
James Simmons <jsimmons@infradead.org> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
James Simmons authored
With certain version of gcc lov_obd.c failes to compile with the following warning. In function copy_to_user, inlined from lov_iocontrol at lustre/lustre/lov/lov_obd.c:1168: ./arch/x86/include/asm/uaccess.h:735: error: call to __copy_to_user_overflow declared with attribute warning: copy_to_user() buffer size is not probably correct In lov_iocontrol the data was being casted to int instead of the required unsigned long. This patch changes the cast to what is needed for copy_to_user. Signed-off-by:
James Simmons <uja.ornl@yahoo.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6302 Reviewed-on: http://review.whamcloud.com/14613Reviewed-by:
frank zago <fzago@cray.com> Reviewed-by:
John L. Hammond <john.hammond@intel.com> Reviewed-by:
-
Vitaly Fertman authored
replace the linear search by interval_tree one for granted list in ldlm_lock_match() Signed-off-by:
Vitaly Fertman <vitaly_fertman@xyratex.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5739 Xyratex-bug-id: MRP-2089 Reviewed-on: http://review.whamcloud.com/12294Reviewed-by:
Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by:
-
Niu Yawei authored
It doesn't make sense to keep the very aged lock even with the LRUR policy. This patch decreased the default ns_max_age from 10 hours to 65 minutes and changed LRUR policy to cancel very aged locks. Signed-off-by:
Niu Yawei <yawei.niu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6529 Reviewed-on: http://review.whamcloud.com/14856Reviewed-by:
-
John L. Hammond authored
Use obd_get_info() to get the default and maximum LOV EA sizes (along with maximum cookiesize) from LOV. Remove the then unused function obd_size_diskmd() and the unused get info key KEY_LOVDESC. When computing the maximum LOV EA size use the active OST count (ld_active_tgt_count) rather than the OST count (ld_tgt_count). Signed-off-by:
-
John L. Hammond authored
Clients no longer need to track the max and default MDS cookiesizes so remove the obsolete obod o_valid flag OBD_MD_FLCOOKIE, the struct client_obd members cl_{default,max}_mds_cookiesize, the struct obd_trans_info and parameters of this type, the cookiesize parameters from md_init_ea_size(), the files llite/*/{default,max}_cookiesize, and any code that needlessly handled these values. Signed-off-by:Lai Siyao <lai.siyao@intel.com> Reviewed-by:
-
Niu Yawei authored
Remove the obsolete quotacheck, quotaon and quotaoff, which were retained for the interoperability with old (< 2.4) client and server. Some other obsolete quota code related to LL_IOC_QUOTACTL_18, Q_INVLIDATE and Q_FINVALIDATE are removed as well. Signed-off-by:
Fan Yong <fan.yong@intel.com> Reviewed-by:
-
Chris Horn authored
NRS specific structures are not needed in the rest of the PtlRPC code. It is more appropriate for these structures to be defined in a separate header. This commit creates a lustre_nrs.h header for the generic NRS structures, and policy-specific headers for the various NRS policies. Signed-off-by:
Chris Horn <hornc@cray.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2667 Reviewed-on: http://review.whamcloud.com/13966Reviewed-by:
Henri Doreau <henri.doreau@cea.fr> Reviewed-by:
-
Mikhail Pershin authored
There are several HSM requests which modify data on server and reply on Lustre recovery, e.g. they should replay changes in case of recovery. Patch allows such requests to be replayed in recovery time and they are issued from client using mdc_rpc_lock to serialize them and avoid concurrent last_rcvd update on server. Signed-off-by:
Faccini Bruno <bruno.faccini@intel.com> Reviewed-by:
-
John L. Hammond authored
Add cl_object_maxbytes() to return the maximum supported size of a cl_object. Remove the lli_maxbytes member from struct ll_inode_info. Change the lsm_maxbytes member of struct lov_stripe_md from __u64 to loff_t. Correct the computation of lsm_maxbytes in the released layout case. Signed-off-by:
-
Jinshan Xiong authored
Remove the lli_has_smd flag from struct ll_inode_info. The empty layout case will be handled by the LOV layer. Remove the unused function cl_local_size(). Signed-off-by:
-
John L. Hammond authored
Add cl_object_layout_get() to return the layout and generation of an object. Replace some direct accesses to object LSM with calls to this function. In ll_getxattr() factor out the LOV xattr specific handling into a new function ll_getxattr_lov() which calls cl_object_layout_get(). In ll_listxattr() call ll_getxattr_lov() to determine if a lustre.lov xattr should be emitted. Add lov_lsm_pack() to generate LOV xattrs from a LSM. Remove the unused functions ccc_inode_lsm_{get,put}() and lov_lsm_get(). Signed-off-by:Bobi Jam <bobijam@hotmail.com> Reviewed-by:
-
John L. Hammond authored
During development a new api, cl_object_obd_info_get() and cl_object_data_version() which then were later replaced by a better solution CIT_DATA_VERSION. For the case of the upstream client their is no point in introducing a API to only have it removed later. Due to the way the patches landed with their dependencies it is not possible to separate out two patches. These two combined patches do the following: * Add a new cl_io type CIT_DATA_VERSION to get file data version. * Remove the unused IOC_LOV_GETINFO ioctl. * Remove ll_glimpse_ioctl() and ll_lsm_getattr(). * Remove the OBD API method obd_getattr_async(). Signed-off-by:
Bobi Jam <bobijam.xu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5823 Reviewed-on: http://review.whamcloud.com/12748 Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6356 Reviewed-on: http://review.whamcloud.com/14649Reviewed-by:
-
wang di authored
After http://review.whamcloud.com/#/c/9967/ is landed, mgc does not wait the import connected(state = FULL), then enqueue and retrieve config log, which will cause the mount process to fail, especially if the mgc is shared by multiple targets. So once mgc enqueue is failed, it will give another chance to wait the import to recover, if the import comes back in time, it will try to enqueue again. Otherwise it will use local config log. Signed-off-by:
wang di <di.wang@intel.com> Signed-off-by:
-
John L. Hammond authored
Remove unused struct lov_stripe_md * parameters from obd_get_info(), mgc_enqueue(), and osc_brw_prep_request(). Signed-off-by:
-
Henri Doreau authored
Do not bother checking the return value of changelog_kuc_hdr() against NULL since this value was dereferenced earlier. Signed-off-by:
-
John L. Hammond authored
Remove unused definitions related to Size on MDS support from lustre/include/lustre/lustre_idl.h. Remove unused code from several places in lustre/. Signed-off-by:
-
Sebastien Buisson authored
Do not take unwrap time in req_waittime calculation on client part as decryption is not related to request service time. Signed-off-by:
Sebastien Buisson <sebastien.buisson@bull.net> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6356 Reviewed-on: http://review.whamcloud.com/14404Reviewed-by:
Jeremy Filizetti <jeremy.filizetti@gmail.com> Reviewed-by:
Andrew Perepechko <andrew.perepechko@seagate.com> Reviewed-by:
-
Henri Doreau authored
Failures in ll_ioc_copy_{start,end} are reported to coordinator. The return code delivered to the caller should indicate what this error was, not whether coordinator was successfully notified. Signed-off-by: -
Andreas Dilger authored
Some old protocol compatibility workarounds are still present in master that should have been removed when LUSTRE_MSG_MAGIC_V1 was. In particular, the process for upgrading LUSTRE_MSG_MAGIC_V1 to LUSTRE_MSG_MAGIC_V2 had the client to connect to the server with the V1 protocol with op_flag=MSG_CONNECT_NEXT_VER set, and if the server supported the V2 protocol it would reply with LUSTRE_MSG_MAGIC_V2. This ensured that if the new client connected to an old server the connection would be allowed. However, even with V1 protocol support removed, the 2.x clients are still connecting with NEXT_VER set. In b1_8 this flag was contingent on LUSTRE_MSG_MAGIC_V1 being used, which is how it should have been in 2.x as well. A few other cleanups are be done at the same time: - disallow 1.8 clients (or at least those that don't understand OBD_CONNECT_FULL20) so we can remove workarounds for 1.8 clients - remove support for pre-2.1 DLM flock lock handling - don't workaround the lack of MDS_ATTR_xTIME_SET flags in setattr - always set MSGHDR_CKSUM_INCOMPAT18 (it can eventually be removed) Signed-off-by:
Liang Zhen <liang.zhen@intel.com> Reviewed-by:
-
Patrick Farrell authored
Currently, the CBPENDING flag is set on group locks when the osc lock above them is released (osc_cancel_base). This results in a situation where a new group lock request on a resource does not match an existing group lock because LDLM_FL_CBPENDING is set on the existing lock. So two group locks are granted on the same resource, which is not valid, since a given client can only have one group lock on a particular resource. Since group locks are manually released and not called back like other LDLM locks, the CBPENDING flag doesn't make sense. Since they must be manually released, they also cannot go in the LDLM LRU cache and must be fully released immediately once they are no longer in use. This was previously accomplished by setting CBPENDING when the corresponding osc lock is released, but as noted above, this prevents the group lock matching some future lock requests. This patch uses the fact that group locks have an l_writers reference which they keep until they are manually released, so we remove them when they have no more reader or writer references, without checking cbpending. Signed-off-by:
Patrick Farrell <paf@cray.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6368 Reviewed-on: http://review.whamcloud.com/14093Reviewed-by:
-
Wu Libin authored
After setting like "lctl set_param -P osc.*.max_pages_per_rpc", it is possible that the function osc_obd_max_pages_per_rpc_seq_write will be called before ocd_brw_size has been set when mount. ocd_brw_size is meaningless when it is zero. So it should not be the limit at that time. Signed-off-by:
Wu Libin <lwu@ddn.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6421 Reviewed-on: http://review.whamcloud.com/14333Reviewed-by:
Emoly Liu <emoly.liu@intel.com> Reviewed-by:
-
Gregoire Pichon authored
The new tag field is used as a virtual index for multiple modifying RPCs management. It is set by the client and allows the target to release in-memory reply data when the tag is reused by a new RPC. The tag field replaces the unused last_seen field of ptlrpcd_body structure. Additionally, the last_xid field is used to transfer the highest XID for which a reply has been received and does not have an unreplied lower-numbered XID. Signed-off-by:
-
John L. Hammond authored
Add handling of inode flags to the handlers of CIT_SETATTR in lov and osc. In the FSFILT_IOC_SETFLAGS case of ll_iocontrol() use cl_setattr_ost() rather than obd_setattr_rqset() to set inode flags on OST objects. Remove the then unused OBD API methods obd_setattr_rqset() and obd_setattr_async() along with their supporting functions. Signed-off-by:
-
Bobi Jam authored
Get rid of lov_stripe_md reference in setting file's stripe info. Signed-off-by:
-
Gregoire Pichon authored
The new OBD_CONNECT_MULTIMODRPCS connection flag indicates the support of multiple modify RPCs in parallel. It can be specified by the client within the connection request and by the server within the connection reply. The new ocd_maxmodrpcs connection data specifies the maximum modify RPCs in parallel supported by the server. To allow the MDS to send the new ocd_maxmodrpcs field, it has been required to modify RMF_CONNECT_DATA so that its size includes the new field. This change leads to remove the ocd_connect_data_v1 structure. Note that the client has been allocating an extra 16*sizeof(__u64) for the obd_connect_data reply since 2.0 and even in later versions of 1.8) so there is no problem for the MDS to just send the full reply size. This patch fixes a bug in __req_capsule_get() since it wasn't checking RMF_F_NO_SIZE_CHECK when receiving the message. This allows legacy clients (with version lower that this commit) to send connection request with ocd_connect_data structure size smaller (actually size is ocd_connect_data_v1 structure size) than new server ocd_connect_data structure size. This patch also fixes a bug in the routine that displays the import's connect data. Signed-off-by:
-
Liang Zhen authored
If user wants to create CPT table that can match numa topology, she has to query cpu & numa topology, then provide a pattern string to describe the topology, this is inconvenient. To improve it, this patch can support shortcut expression "N" or "n" to create CPT table from NUMA & CPU topology Signed-off-by:
Olaf Weber <olaf@sgi.com> Reviewed-by:
-
wang di authored
Default dir stripe index should only work during mkdir, otherwise it will cause other open/create request being sent to the wrong MDT. Signed-off-by:
-
Sebastien Buisson authored
Return -ECONNREFUSED instead of -ENOMEM in sptlrpc_req_get_ctx() if no context is found in req. It is more graceful. Signed-off-by:
-
Bobi Jam authored
* replace struct ll_user_fiemap with struct fiemap * replace struct ll_fiemap_extent with struct fiemap_extent * remove kernel defined FIEMAP_EXTENT_* constants * remove kernel defined FIEMAP_FLAG_* flags * add member prefix for struct ll_fiemap_info_key * Add cl_object_operations::coo_fiemap(). * Add cl_object_fiemap() to get FIEMAP mappings. Signed-off-by:
Yang Sheng <yang.sheng@intel.com> Reviewed-by:
-
frank zago authored
A lot of symbols don't need to be exported at all because they are only used in the module they belong to. Signed-off-by:
James Simmons <uja.ornl@gmail.com> Reviewed-by:
-
Jinshan Xiong authored
In this implementation, read ahead will hold the underlying DLM lock to add read ahead pages. A new cl_io operation cio_read_ahead() is added for this purpose. It takes parameter cl_read_ahead{} so that each layer can adjust it by their own requirements. For example, at OSC layer, it will make sure the read ahead region is covered by a LDLM lock; at the LOV layer, it will make sure that the region won't cross stripe boundary. Legacy callback cpo_is_under_lock() is removed. Signed-off-by: -
John L. Hammond authored
Remove the unused OBD MD API method md_done_writing(). Remove the unused logcookie and struct md_open_data ** parameters from md_setattr(). Remove the unused functions iattr_from_obdo(), md_from_obdo(), and obdo_refresh_inode(). Signed-off-by:
-
John L. Hammond authored
Size on MDS support have been in preview since at least 2.0.0. Remove support for it from lustre/llite/. Signed-off-by:
-
Patrick Valentin authored
When unloading a module, it may happen that lu_context_key_degister() removes a key while a thread is either registering it in a new context (lu_context_init(), lu_context_refill()), or using it when exiting from a context (lu_context__exit(), lu_context__fini()). In these cases, we reference a key which no longer exists, and the system crashes either because we use a *POISON'ed* pointer in key_fini() -> key->lct_fini(), or because one of the following assertions fails: - lu_context_key_degister(): ASSERTION(cfs_atomic_read(&key->lct_used) == 1) failed: key has instances: 2 - lu_context_exit(): ASSERTION(key != NULL) - key_fini(): ASSERTION(atomic_read(&key->lct_used) > 1) This can also leads to SLAB objects which are not freed: slab error in kmem_cache_destroy(): cache `echo_thread_kmem': Can't free all objects Note: ptlrpc service threads need to call lu_context_init/fini in each loop (for each RPC), and this could be a big performance issue on fat SMP machines if we add serialization by a spinlock and need to lock/unlock it for multiple times for each RPC. So the aim of this patch, which only impacts some low frequently used functions, is: 1) to add a synchronization in lu_context_key_quiesce(), also called by lu_context_key_degister(), to wait until all key::lct_init() methods have completed, by serializing with keys_fill() 2) to add a synchronization in lu_context_key_degister(), to wait until all transient contexts referencing this key have run key::lct_fini() method Signed-off-by:Patrick Valentin <patrick.valentin@bull.net> Signed-off-by:
-
Bruno Faccini authored
Upon umount, presumably of last device using same OSD back-end, to prepare for module unload, lu_context_key_quiesce() is run to remove all module's key reference in any context linked on lu_context_remembered list. Threads must protect against such transversal processing when exiting from its context. Signed-off-by:
-
Bhumika Goyal authored
Remove function remote_event_destroy as it is not used anywhere in the kernel. Function detected using Coccinelle but changes done by hand. Script: @initialize:python@ @@ def display(name,p): print(name,p[0].file) @r1@ identifier func; type T; position p; @@ static T func@p(...) { ... } @r@ identifier r1.func; @@ func @script:python depends on !r@ func << r1.func; p << r1.p; @@ display(func,p) Signed-off-by:Bhumika Goyal <bhumirks@gmail.com> Signed-off-by:
-
Chase Metzger authored
Added braces to else statements where checkpatch complained. Signed-off-by:
Chase Metzger <chasemetzger15@gmail.com> Acked-by:
Viresh Kumar <viresh.kumar@linaro.org> Reviewed-by:
Vaibhav Agarwal <vaibhav.sr@gmail.com> Signed-off-by:
-
Elizabeth Ferdman authored
Add a blank line after a declaration. Error found by checkpatch. Signed-off-by:
Elizabeth Ferdman <gnudevliz@gmail.com> Reviewed-by:
Mark Greer <mgreer@animalcreek.com> Signed-off-by:
-
Hisao Tanabe authored
This patch fix the coding style problem of block comment detected by the checkpatch.pl. WARNING: Block comments use a trailing */ on a separate line Signed-off-by:
Hisao Tanabe <xtanabe@gmail.com> Acked-by:
Johan Hovold <johan@kernel.org> Acked-by:
-
