1ece42d0 | 02-May-2014 |
Simon Klinkert <klinkert@webgods.de> |
4827 nfs4: slow file locking 4837 NFSv4 client lock retry delay upper limit should be shorter
(cherry picked from commit 99ef4b75bc247bdd229cd0820fccabdb0602cf52) |
93297ba3 | 13-May-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: include fs root inode info in mount events
(cherry picked from commit 5beaf3ad100304461dfab72835f386d63abe8568) |
6a6a51ee | 08-May-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: changes for #1891 review (multiplayer)
Review feedback addressed in this commit:
- remove avl, replace with array - read() can block (but can be interrupted by signals) - style changes - q
zev: changes for #1891 review (multiplayer)
Review feedback addressed in this commit:
- remove avl, replace with array - read() can block (but can be interrupted by signals) - style changes - queue threshold can be set to 0 bytes (no delay) - FIND_QUEUE macro replaced by function - new mutex to avoid races when queues are full - zevadm can be told to poll events from a specified queue - zevadm can create blocking queues
additional changes:
- mark events are delivered without delay
(cherry picked from commit e4ea145c5d454bda934f8af75eec9ea17f2489c0)
show more ...
|
add9520f | 19-Mar-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: multi-queue support
(cherry picked from commit bc16f3206d5f60ea7a18a972c3a52afb85102b71) |
919e9f2a | 24-Feb-2014 |
Simon Klinkert <klinkert@webgods.de> |
nfs4_do_delegreturn: fix for endless recovery after nfs server reboot.
Issue: https://redmine.ob1.de/issues/799 (cherry picked from commit fed705ad873bc176fde0522f8b470b3a98e35017) |
b94833c4 | 06-Feb-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
getgen driver added for znode info ioctl()s
The getgen driver provides an ioctl() that can give information about an inode to userspace that is normally unavailable, such as the generation number an
getgen driver added for znode info ioctl()s
The getgen driver provides an ioctl() that can give information about an inode to userspace that is normally unavailable, such as the generation number and the crtime.
The inode has to be on a ZFS filesystem, and an open fd to any file on the same filesystem needs to be provided.
(cherry picked from commit c8fa4655f8d4fdaba0469296ede911a6bb63f76b)
show more ...
|
97dcf88d | 05-Feb-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: pass removed file's inode info into userland
(cherry picked from commit b23df4380880bf151b06786c591579e8a7abafc7) |
a01b300a | 03-Feb-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: remove unnecessary zap_lookup, display ctime
(cherry picked from commit a27b019ed8c12d2702bb90b14500b8932ac1ecd3) |
35d4e8dd | 27-Jan-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: new ioctl (get_gen) and various fixes
(cherry picked from commit 7a84701fb0e8e59f59d30523c82b171d7e510273) |
f985bd4a | 15-Jan-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: small fix for zev_detach() return value
(cherry picked from commit ff33ba5a948e98d96388e544a519285260587106) |
19b08257 | 14-Jan-2014 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: pass xattr flag into userland, fix umount bug
zev events now have a flag in inode structures indicating whether the inode is part of some file's extended attributes. We use this to ignore such
zev: pass xattr flag into userland, fix umount bug
zev events now have a flag in inode structures indicating whether the inode is part of some file's extended attributes. We use this to ignore such events.
Also, there was a bug in umount: the dsl dataset was freed before it was used by the zev callback. This wasn't noted before because it was usually still okay. On debug kernels, the freed space is cleared, and so it was immediatly visible.
(cherry picked from commit 5be8959620e9c9a4179a80521f9be872ab79b065)
show more ...
|
9a535601 | 09-Dec-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: bugfix - zev_detach return value when busy
zev_detach() returned EBUSY when the module was in use, but should return DDI_FAILURE.
EBUSY leads to the titanic_7 kernel trying to unload the modul
zev: bugfix - zev_detach return value when busy
zev_detach() returned EBUSY when the module was in use, but should return DDI_FAILURE.
EBUSY leads to the titanic_7 kernel trying to unload the module, which will fail in _fini(), and also break functionality because of the partial execution of _fini().
(cherry picked from commit b36b61981825c480819dc7087eeb3aad5be53d6d)
show more ...
|
548c8b6e | 19-Nov-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: fix lint and build warnings
(cherry picked from commit 74cdfce6dd066682a1724b52826d50255f2af3dd) |
74ec25ca | 19-Nov-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: fixed warning - removed unused variable
(cherry picked from commit 7aed7d4f6fa42851ce9dcaeecaea5c36ebb2a924) |
8f38b24d | 18-Nov-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: skip mount/umount events for snapshots.
(cherry picked from commit 35f5557688c45a343edf7bae121e177a472ca22b) |
e2e7ab14 | 25-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
zev: bugfix - error events had wrong record size.
(cherry picked from commit 00787cc8a2d0d4dcb6e69efab7719d6c85b6d6cd) |
ca10c38b | 24-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
add close_after_update event for any close()
We tried to be clever and get close_after_update events only for files that have no further open fds after the close() in question. However, the NFSv4 s
add close_after_update event for any close()
We tried to be clever and get close_after_update events only for files that have no further open fds after the close() in question. However, the NFSv4 server seems to keep files open for much longer than anticipated, so we'd get the events much too late.
Now we get an event for any close() call on a file, as long as there's been new data added to the file since the last time we added the event to the queue.
(cherry picked from commit c27e8300d7cdccf9d8df6abcec04e15846bf7862)
show more ...
|
808a670a | 11-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
allow 32-bit ioctl() calls in 64-bit kernels: all structs are now padded and aligned
(cherry picked from commit aadc9bd992b088de347df644453365dff6a4b4d1) |
888fea18 | 10-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
marks can be queued in-between events.
(cherry picked from commit d8d328ed3087e5c59b7f42951ab9dff657ba162a) |
149d0aff | 10-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
struct padding fixed to work with 32 and 64 bit userspace receivers, plus two other small fixes.
(cherry picked from commit 3c715018da3061d1e7dd083ab1699d835073d6ce) |
d979f56c | 02-Oct-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
pass multiple events in one read() call; new convenience typedefs in header
(cherry picked from commit 4a804719ddd0b229ab53217aa4c7a44f7466165b) |
f2dd45e5 | 30-Sep-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
switch protocol from packed nvlists to structs, for expected performance reasons.
(cherry picked from commit e7801d00ec0706117ba51159f125a1ac6695cb59) |
aafc540f | 24-Sep-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
pass additional information for each inode, changed protocoll to native encoded nvlist records.
(cherry picked from commit d41256dd2f25a2cf8d8d1a4dbb550c0d47f9858a) |
dde42681 | 13-Aug-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
switched fs identification from fsid to zfs guid, because fsids are not persistent and can not be obtained from userspace
(cherry picked from commit 2ae38835b515b7ae1e7d893d499036d6dd80307d) |
1f47b6c0 | 01-Aug-2013 |
Andreas Jaekel <jaekel@strato-rz.de> |
setting callbacks directly instead of using an ioctl() call
(cherry picked from commit 144f824a918b74a1927b1945c1841ae49621b137) |