- 29 Apr, 2016 26 commits
-
-
Mikhail Pershin authored
Initialize request session early to make it available in high-priority handlers Signed-off-by:
Mikhail Pershin <mike.pershin@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3467 Reviewed-on: http://review.whamcloud.com/7350Reviewed-by:
Alex Zhuravlev <alexey.zhuravlev@intel.com> Reviewed-by:
Fan Yong <fan.yong@intel.com> Reviewed-by:
Andreas Dilger <andreas.dilger@intel.com> Signed-off-by:
James Simmons <jsimmons@infradead.org> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Lai Siyao authored
The patch for LU-3286 removed vfsmount instances used on the server side. Since this is server side only we can remove it from the upstream client. Signed-off-by:
Lai Siyao <lai.siyao@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3286 Reviewed-on: http://review.whamcloud.com/8286Reviewed-by:
Jian Yu <jian.yu@intel.com> Reviewed-by:
Oleg Drokin <oleg.drokin@intel.com> Signed-off-by:
-
Swapnil Pimpale authored
ll_dir_filler() now checks the returned pointer from ll_prep_md_op_data() using IS_ERR() Signed-off-by:
Swapnil Pimpale <spimpale@ddn.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-4082 Reviewed-on: http://review.whamcloud.com/8073Reviewed-by:
Dmitry Eremin <dmitry.eremin@intel.com> Reviewed-by:
John L. Hammond <john.hammond@intel.com> Reviewed-by:
-
Andreas Dilger authored
Add debugging for LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF) in ll_file_open(), since this is a rarely hit failure under racer, and it would be useful to get more information if this is hit again. Print the full intent disposition, as well as the status, in case Oleg's earlier comment about bailing out on any error is actually the case here. Signed-off-by:
Bob Glossman <bob.glossman@intel.com> Reviewed-by:
-
Andriy Skulysh authored
Kernel commit c1c3443c assigns all allowed cpus to emulated node. End cpt initialization loop when all CPUs are assigned. Signed-off-by:
Andriy Skulysh <Andriy_Skulysh@xyratex.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3992 Reviewed-on: http://review.whamcloud.com/7724Reviewed-by:
Liang Zhen <liang.zhen@intel.com> Reviewed-by:
-
Lai Siyao authored
Add OBD_CONNECT_OPEN_BY_FID for open by FID, if MDS supports this, for open by FID, it won't retry with name if object with the FID doesn't exist; while if client supports this, client won't pack name in open request if FID is known. Signed-off-by:
-
Swapnil Pimpale authored
client_obd_setup() allocates an obd_import which should be cleaned up if there is any failure afterwards in callers of client_obd_setup(). This patch fixes the bug in osc_setup(), mgc_setup(), mdc_setup() and lwp_setup(). The fix is to call obd_cleanup_client_import() before calling client_obd_cleanup() in case of an error. Signed-off-by:
-
Dmitry Eremin authored
Fix ignoring errors from ldebugfs_add_vars() function. Signed-off-by:
-
Mikhail Pershin authored
Switch OST/OFD request processing to the unified request handle. Signed-off-by:
-
Amir Shehata authored
The function class_config_parse_rec() parses the llog record and places it into a buffer to be returned. That buffer needs to end with a newline which is currently missing. Signed-off-by:
Amir Shehata <amir.shehata@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2149 Reviewed-on: http://review.whamcloud.com/4254Reviewed-by:
-
Amir Shehata authored
Remove the newline from the LCONSOLE debug macro in the function class_config_dump_handler(). Signed-off-by:
-
Andreas Dilger authored
The Lustre 2.5.4 client will print a warning about connections with 2.1.3 servers, yet they are still supposed to be interoperable. Increase the window of warning to be up to Lustre 2.5.50, since we do not intend to allow interoperability between 2.1 and 2.6 systems. This was from the time when major releases like Lustre 1.4 and 1.8 were many years apart would have no chance for interoperation. Only print this message once per client, to avoid flooding the console for connections to many servers or frequent network reconnections. Server versions should all be nearly the same in any case. Signed-off-by:
-
Prakash Surya authored
This change adds a BRW page flag, OBD_BRW_SOFT_SYNC. This flag is intended to urge a server to commit a client's unstable pages to stable storage. A client will add this flag to any BRW requests while it is in a state where it has "many" unstable pages pinned in its cache. The server, upon receiving a page with this flag set, *should* begin an async commit. The idea being that, with the proactive async commit, the client's pinned unstable pages will transition into a stable state faster than they would have otherwise. Although, the server side portion of this agreement is still unimplemented, so the OBD_BRW_SOFT_SYNC flag will currently fall on deaf ears. Signed-off-by:
Prakash Surya <surya1@llnl.gov> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2139 Reviewed-on: http://review.whamcloud.com/4375Reviewed-by:
Jinshan Xiong <jinshan.xiong@intel.com> Reviewed-by:
-
Prakash Surya authored
This change adds simple accounting hooks for "unstable" pages on a per OSC basis. Now, in addition to the per filesystem tracking, each OSC will maintain a running total of its unstable pages. These counters are exported through the proc interface, and can be read using the lctl command. For example: # Read number of unstable pages contained by each OSC lctl get_param osc.*.unstable_stats The motivation for this change is in anticipation of implementing a "soft sync" functionality, urging servers to commit these unstable pages to stable storage. The per OSC accounting allows a client to limit the soft sync request to only the OSCs which have outstanding unstable pages. Signed-off-by:
-
Prakash Surya authored
This change adds a global counter to track the number of "unstable" pages held by a given client, along with per file system counters. An "unstable" page is defined as a page which has been sent to the server as part of a bulk request, but is uncommitted to stable storage. In addition to simply tracking the unstable pages, they now also count towards the maximum number of "pinned" pages on the system at any given time. Thus, a client will now be bound on the number of dirty and unstable pages it can pin in memory. Previously only dirty pages were accounted for in this limit. In addition to tracking the number of unstable pages in Lustre, the NR_UNSTABLE_NFS memory zone is also incremented and decremented for easy monitoring using the "NFS_Unstable:" field in /proc/meminfo. This field is also used internally by the kernel to limit the total amount of unstable pages on the system. The motivation for this change is twofold. First, the client must not allow itself to disconnect from an OST while still holding unstable pages. Otherwise, these unstable pages can get lost due to an OST failure, and replay is not possible due to the disconnect via unmount. Secondly, the client needs a mechanism to prevent it from allocating too much of its available RAM to unreclaimable pages pinned by the ptlrpc layer. If this case occurs, out of memory events can trigger as a side effect, which we need to avoid. The current number of unstable pages accounted for on a per file system granularity is exported by the unstable_stats proc file, contained under each file system's llite namespace. An example of retrieving this information is below: $ lctl get_param llite.*.unstable_stats Signed-off-by:
-
Sebastien Buisson authored
Fix 'error handling' issues found by Coverity version 6.6.1: Unchecked return value (CHECKED_RETURN) Calling function without checking return value. Argument cannot be negative (NEGATIVE_RETURNS) Negative value used as argument to a function expecting a positive value. Signed-off-by:
Sebastien Buisson <sbuisson@ddn.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-4055 Reviewed-on: http://review.whamcloud.com/7842Reviewed-by:
James Nunez <james.a.nunez@intel.com> Reviewed-by:
-
Niu Yawei authored
Replace LASSERT with LASSERTF in osc_completion, thus we can get more info when the LASSERT is triggered. Signed-off-by:
Niu Yawei <yawei.niu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3843 Reviewed-on: http://review.whamcloud.com/7494Reviewed-by:
Bobi Jam <bobijam@gmail.com> Reviewed-by:
-
James Nunez authored
The printing of i_ino/i_generation in llite messages is not nearly so useful as printing the full inode FID, since i_ino is a "compressed" version of the FID and there may be duplicate values for i_ino in some cases (especially if running on a 32-bit client). All instances of printing i_ino/i_generation are replaced with the FID using ll_inode2fid(). All instances, except for one, of printing just i_ino was replaced by printing the FID. In all CERROR lines touched by the i_ino replacements, the device name or fsname was added at the beginning of the message if it did not already exist. Signed-off-by:
-
Bruce Korb authored
Add and update documentation about some of the ldlm l_flags. Signed-off-by:
Bruce Korb <bruce.korb@gmail.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2906Reviewed-by:
Keith Mannthey <Keith.Mannthey@intel.com> Reviewed-on: http://review.whamcloud.com/7963Reviewed-by:
Doug Oucharek <doug.s.oucharek@intel.com> Reviewed-by:
-
Bruce Korb authored
Wireshark output moved to userland file "lustre_dlm_flags_wshark.c" and only bits that can actually appear "on the wire" are emitted. The user land "packet-lustre.c" code that references these bits now gets emitted into that file. e.g. the "local_only" bit is never put on the wire, so references to it in wireshark are gone. Signed-off-by:
-
Bruce Korb authored
Remove the now obsolete LDLM_AST_FLAGS and LDLM_INHERIT_FLAGS defines. Remove the obsolete LDLM_FL_HIDE_LOCK_MASK define. Rename "local_only" mask to "off_wire" since it is confusingly similar to a flag that (I think) means, "do not copy this lock over the wire." Signed-off-by:
-
Bruce Korb authored
Convert most of the ldlm lock's l_flags references from direct bit twiddling to using bit specific macros. A few multi-bit operations are left as an exercise for the reader. The changes are mostly in ldlm, but also in llite, osc and quota. Also add a multi-bit (mask) test. Signed-off-by:
-
Dmitry Eremin authored
There are few places with access to lmv->tgts[] without check for NULL. Usually it may happens when MDT configured starting from index 1 instead of 0. For example: mkfs.lustre --reformat --mgs --mdt --index=1 /dev/sdd1 Signed-off-by:
-
Wang Di authored
Add LCT_SERVER_SESSION for server session, and separate the server session flag from LCT_SESSION, so to avoid allocating session info for client stack for each server request, if client and server are on the same node. Signed-off-by:
Wang Di <di.wang@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3806 Reviewed-on: http://review.whamcloud.com/7412Reviewed-by:
-
Dmitry Eremin authored
Suppress erroneous/confusing messages when NFS is out of sync and requests old data. Signed-off-by:
-
Jinshan Xiong authored
Otherwise after one round the writeback index will become beyond the file size and ->writepages() turns into an empty operation. Also, a safety guard is added to limit the wait time for grant to 10 minutes(take recovery into consideration) at maximum in the osc_enter_cache() function. Otherwise, EDQUOT will be returned to the applications to start sync write. Signed-off-by:
-
- 26 Apr, 2016 14 commits
-
-
Iban Rodriguez authored
Variable lustre_profile_list is only used inside obd_config.c, better make it static Signed-off-by:
Iban Rodriguez <iban.rodriguez@ono.com> Acked-by:
-
John L. Hammond authored
In the MDC HSM handlers that do not pack a real suppgid, use -1 rather than 0 for the suppgid in mdt_body. Signed-off-by:
Aurelien Degremont <aurelien.degremont@cea.fr> Reviewed-by:
Faccini Bruno <bruno.faccini@intel.com> Reviewed-by:
-
John L. Hammond authored
In the LL_IOC_HSM_CT_START case of ll_dir_ioctl() require CAP_SYS_ADMIN, since the local handler for this ioctl may modify the global KUC table. Signed-off-by:
-
Andriy Skulysh authored
Patch for LU-631 which was landed before the upstream merge broke OBD_STATFS_NODELAY behaviour. It adds unnecessary delay while running df command with inactive OSTs. We shouldn't try to recover connection to OST in this case. Signed-off-by:
-
Bruno Faccini authored
Since st_blocks = NULL is returned for released files, FIEMAP should at least return a minimal mapping to make users aware that file contains data but it is not immediately available. This will make coreutils and tools such tar happy and have them presume file is sparse. Also, add a new test_228 in sanity-hsm to verify it works for "[cp,tar] --sparse" commands. Also fix a LBUG ("lov_fiemap()) ASSERTION( fm_local ) failed") likely to occur when no-object/ENOMEM conditions and also now when released. Signed-off-by: -
Andrew Perepechko authored
This patch implements log2 using fls. Signed-off-by:
Andrew Perepechko <andrew_perepechko@xyratex.com> Reviewed-by:
Alexander Boyko <Alexander_Boyko@xyratex.com> Reviewed-by: alexander_zarochentsev@xyratex.com Reviewed-by:
Vitaly Fertman <Vitaly_Fertman@xyratex.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3496 Xyratex-bug-id: MRP-999 Reviewed-on: http://review.whamcloud.com/6757Reviewed-by:
-
Jinshan Xiong authored
Otherwise it will hit the assertion at cl_lock.c: cl_lock.c:1967:discard_cb()) ASSERTION( (!(page->cp_type == CPT_CACHEABLE) || (!PageWriteback(cl_page_vmpage(env, page)))) ) failed: This is because in osc_lock_flush() we have to make sure the IO is finished before discarding the pages. Signed-off-by:
-
Jinshan Xiong authored
Truncate up is safe so it won't trigger restore. Copy optimization for truncate down - only copy the part under truncate length. If a file is truncated to zero usually it'll be followed by write so I choose to restore the file and set correct stripe information. Signed-off-by:
jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr> Reviewed-by:
Henri Doreau <henri.doreau@cea.fr> Reviewed-by:
-
Swapnil Pimpale authored
This patch has the following: 1) Fix for the return value from ptlrpcd_init(). It will now return a correct status instead of returning zero always. 2) ptlrpcd_addref() should not increment ptlrpcd_users on error. 3) Added code in a mdc_setup() and mgc_setup() to test the return value of ptlrpcd_addref() and return on error. Signed-off-by:
-
Bobi Jam authored
If we specify multiple --mgsnode for a MDT, when we start MDS upon it while MGS is no the other node, the MGC import connection will always select the local nid (which is one of the candidate mgsnode) since it think its the closest connection. This patch treats further --mgsnode nids as failover nids, so that multiple import connections are added for the MGC import. Signed-off-by:
Bobi Jam <bobijam.xu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3829 Reviewed-on: http://review.whamcloud.com/7509Reviewed-by:
-
Jinshan Xiong authored
The root cause is that when max_cache_mb conf parameter is applied, the client isn't connected to the OST yet so that sbi->ll_dt_exp is NULL. However, it's not necessary to shrink the cache memory in this case so success should be returned. Signed-off-by:
-
Andrew Perepechko authored
Assume dchild argument is fully initialized in ->unlink and ->rmdir callbacks, so additional lookup for ELC is not needed. Signed-off-by:
-
Andrew Perepechko authored
With the patch 6648 a fee variables were renamed. We do these renames in broken out patch to the fix obvious. Signed-off-by:
-
Bruno Faccini authored
Follow-on to patch introduced to fix LU-2665 ticket (Gerrit Change at http://review.whamcloud.com/6415 with Change-Id: I8faa331712abeadee46eabe111ee1c23a05840d5). Original patch introduced regressions against POSIX test suite (fcntl.18/fcntl.35 tests in LSB-VSX POSIX test suite at http://www.opengroup.org/testing/linux-test/lsb-vsx.html), so the idea is to only resend F_UNLCKs to have both LU-2665 bug and POSIX test suite happy. Signed-off-by:
-
