- 23 Apr, 2018 40 commits
-
-
NeilBrown authored
As this is indexed by an integer, an extensible array or extensible bitmap would be better. If/when xarray lands, we should change to use that. For now, just a simple conversion to rhashtable. When removing an entry, we need to hold rcu_read_lock() across the lookup and remove in case we race with another thread performing a removal. This means we need to use call_rcu() to free the quota info so we need an rcu_head in there, which unfortunately doubles the size of the structure. Signed-off-by:
NeilBrown <neilb@suse.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
NeilBrown authored
The rhashtable data type is a perfect fit for the export uuid hash table, so use that instead of cfs_hash (which will eventually be removed). As rhashtable supports lookups and insertions in atomic context, there is no need to drop a spinlock while inserting a new entry, which simplifies code quite a bit. As there are no simple lookups on this hash table (only insertions which might fail and take a spinlock), there is no need to use rcu to free the exports. Signed-off-by:
-
NeilBrown authored
The pools hashtable can be implemented using the rhashtable implementation in lib. This has the benefit that lookups are lock-free. We need to use kfree_rcu() to free a pool so that a lookup racing with a deletion will not access freed memory. rhashtable has no combined lookup-and-delete interface, but as the lookup is lockless and the chains are short, this brings little cost. Even if a lookup finds a pool, we must be prepared for the delete to fail to find it, as we might race with another thread doing a delete. We use atomic_inc_not_zero() after finding a pool in the hash table and if that fails, we must have raced with a deletion, so we treat the lookup as a failure. Use hashlen_string() rather than a hand-crafted hash function. Note that the pool_name, and the search key, are guaranteed to be nul terminated. Signed-off-by:
-
NeilBrown authored
Linux has a resizeable hashtable implementation in lib, so we should use that instead of having one in libcfs. This patch converts the ptlrpc conn_hash to use rhashtable. In the process we gain lockless lookup. As connections are never deleted until the hash table is destroyed, there is no need to count the reference in the hash table. There is also no need to enable automatic_shrinking. Signed-off-by:
-
NeilBrown authored
This include file is only included in one place, and only contains a list of other include directives. So just move all those to the place where this file is included, and discard the file. One include directive uses a local name ("linux-cpu.h"), so that needs to be given a proper path. Probably many of these should be remove from here, and moved to just the files that need them. Signed-off-by: -
NeilBrown authored
CDEBUG_STACK() and CHECK_STACK() are macros to help with debugging, so move them from drivers/staging/lustre/include/linux/libcfs/linux/libcfs.h to drivers/staging/lustre/include/linux/libcfs/libcfs_debug.h This seems a more fitting location, and is a step towards removing linux/libcfs.h and simplifying the include file structure. Signed-off-by:
-
James Simmons authored
In ll_xattr_set_common() detect the removexattr() case correctly by testing for a NULL value as well as XATTR_REPLACE. Signed-off-by:
John L. Hammond <john.hammond@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-10787 Reviewed-on: https://review.whamcloud.com/Reviewed-by:
Dmitry Eremin <dmitry.eremin@intel.com> Reviewed-by:
James Simmons <uja.ornl@yahoo.com> Signed-off-by:
James Simmons <jsimmons@infradead.org> Signed-off-by:
-
John L. Hammond authored
If struct xattr_handler has a name member then use it (rather than prefix) for the ACL xattrs. This avoids a bug where ACL operations failed for some kernels. Signed-off-by:
-
James Simmons authored
Small style changes to match more the kernel code standard and it make it more readable. Signed-off-by:
Bob Glossman <bob.glossman@intel.com> Reviewed-by:
Sebastien Buisson <sbuisson@ddn.com> Reviewed-by:
Oleg Drokin <oleg.drokin@intel.com> Signed-off-by:
-
James Simmons authored
Add proper punctuation to the comments. Change buf_size to size for comment in ll_listxattr() since buf_size doesn't exit which will confuse someone reading the code. Signed-off-by:
-
James Simmons authored
Convert __uXX types to uXX types since this is kernel code. The function ll_lov_user_md_size() returns ssize_t so change lum_size from int to ssize_t. Signed-off-by:
-
James Simmons authored
Having an extra ifdef makes the code harder to read. For the case of ll_xattr_get_common() we have a variable initialized at the start of the function but it is only used in XATTR_ACL_ACCESS_T code block. Lets move that variable to that location since its only used there and make the code look cleaner. Signed-off-by:
-
James Simmons authored
Keep track of attempted deletions as well as changing of the lma/link xattrs. Signed-off-by:
-
James Simmons authored
In creating the full name of a xattr a new line was added that was seen by the remote MDS server which confused it. Remove the newline. Signed-off-by:
-
James Simmons authored
The size check at the start of ll_setstripe_ea() is only valid for a directory. Move that check to the section of code handling the S_ISDIR case. Signed-off-by:
-
Bobi Jam authored
Tools like rsync, tar, cp may copy and restore the xattrs on a file. The client previously ignored the setting of trusted.lov/lustre.lov if the layout had already been specified, to avoid causing these tools to fail for no reason. For PFL files we still need to silently eat -EEXIST on setting these attributes to avoid problems. Signed-off-by:
Bobi Jam <bobijam.xu@intel.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-9484 Reviewed-on: https://review.whamcloud.com/27126Reviewed-by:
Niu Yawei <yawei.niu@intel.com> Reviewed-by:
Andreas Dilger <andreas.dilger@intel.com> Signed-off-by:
-
Bobi Jam authored
No need to check several times if lump is NULL. Just test once and return 0 if NULL. Signed-off-by:
-
Bobi Jam authored
Place all the handling of information of trusted.lov that is not stripe related into the new function ll_adjust_lum(). Now ll_setstripe_ea() only handles striping information. Signed-off-by:
-
Niu Yawei authored
Simple comment added to ll_xattr_set. Signed-off-by:
Lai Siyao <lai.siyao@intel.com> Reviewed-by:
Jinshan Xiong <jinshan.xiong@gmail.com> Signed-off-by:
-
Niu Yawei authored
The function ll_xattr_set() contains special code to handle the lustre specific xattr lustre.lov. Move all this code to a new function ll_setstripe_ea(). Signed-off-by:
-
Robin Humble authored
The security.capability xattr is used to implement File Capabilities in recent Linux versions. Capabilities are a fine grained approach to granting executables elevated privileges. eg. /bin/ping can have capabilities cap_net_admin, cap_net_raw+ep instead of being setuid root. This xattr has long been filtered out by llite, initially for stability reasons (b15587), and later over performance concerns as this xattr is read for every file with eg. 'ls --color'. Since LU-2869 xattr's are cached on clients, alleviating most performance concerns. Removing llite's filtering of the security.capability xattr enables using Lustre as a root filesystem, which is used on some large clusters. Signed-off-by:
Robin Humble <plaguedbypenguins@gmail.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-9562 Reviewed-on: https://review.whamcloud.com/27292Reviewed-by:
-
John L. Hammond authored
In ll_xattr_cache_refill() if the xattr cache was invalid (and no request was sent) then return -EAGAIN so that ll_getxattr_common() caller will fetch the xattr from the MDT. Signed-off-by:
-
Vitaly Fertman authored
On the server side mdt_intent_getxattr() can return EFAULT if a buffer cannot be found, it is returned after lock_replace, where a new lock is installed into lockp. An error forces ldlm_lock_enqueue() to destroy the original lock, but ldlm_handle_enqueue0() drops the reference on the new lock. The xattr client code implied intent error is returned under a lock, which is immediately cancelled. Check if a lock obtained and cancel it properly for error cases. Note: we should support both cases for interop needs, an intent error under a lock and with a lock abort. Keep returning a lock with an intent error for interop purposes for now, to be dropped later when client will get old enough. make all intent ops to work through md_intent_lock: getxattr and layout, which should extract the intent error. Signed-off-by:
Vitaly Fertman <vitaly.fertman@seagate.com> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7433 Seagate-bug-id: MRP-3072 MRP-3137 Reviewed-on: http://review.whamcloud.com/17220Reviewed-by:
Andrew Perepechko <andrew.perepechko@seagate.com> Reviewed-by:
Andriy Skulysh <andriy.skulysh@seagate.com> Tested-by:
Elena V. Gryaznova <elena.gryaznova@seagate.com> Reviewed-by:
-
Vitaly Fertman authored
Remove LL_IT2STR(it) from debug macros in lmv_enqueue(). The removal makes it possible to simplify the md_enqueue() functions. Signed-off-by:
-
Vitaly Fertman authored
The lookup_intent it_op fields in many cases will be compared to the settings of IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR. Create a simple inline function for this common case. Signed-off-by:
-
James Simmons authored
When the allocation of xattr->xe_name was moved to kstrdup() setting xattr->xe_namelen was dropped. This field is used in several parts of the xattr cache code so it broke xattr handling. Initialize xattr->xe_namelen when allocating xattr->xe_name succeeds. Also change the debugging statement to really report the xattr name instead of its length which wasn't event being set. Fixes: b3dd8957 ("staging: lustre: lustre: llite: Use kstrdup" Signed-off-by:
-
John L. Hammond authored
Request dynamic minor allocation when registering /dev/lnet and /dev/obd. Signed-off-by:
Jian Yu <jian.yu@intel.com> Reviewed-by:
-
NeilBrown authored
These defines are unused or nearly unused, and do not help at all. Signed-off-by:
-
NeilBrown authored
Very little is left in linux-time.h. Move CFS_TICK to libcfs.h discard the test. Signed-off-by:
-
NeilBrown authored
cfs_time_before_64 is the same as time_before64() similarly cfs_time_beforeq_64() matsches time_before_eq64() So just use the standard interfaces. Signed-off-by:
-
NeilBrown authored
cfs_duration_sec() simply divides by HZ. It is mostly used to report durations in debug messages. Remove and just use X/HZ. Signed-off-by:
-
NeilBrown authored
This wrapper is only used once, so open-code it as max(). This allows us to remove the libcfs_time.h include file. Signed-off-by:
-
NeilBrown authored
cfs_time_after() behaves exactly like time_after() similarly cfs_time_aftereq() matches time_after_eq() so discard the cfs versions. Signed-off-by:
-
NeilBrown authored
This function simply multiplies by HZ and adds jiffies. This is simple enough to be opencoded, and doing so makes the code easier to read. Same for cfs_time_shift_64() Signed-off-by:
-
NeilBrown authored
cfs_time_add adds its arguments. cfs_time_sub subtracts finds the difference. Discard these and use '+' and '-' directly. This change highlighted a type error. The structure field cr_queued_time was used to store jiffies, but was declared as time_t (meant for seconds). So the time is changed to "unsigned long". Signed-off-by:
-
NeilBrown authored
Discard cfs_time_current() and cfs_time_current64() and use jiffies and get_jiffies_64() like the rest of the kernel. Signed-off-by:
-
Greg Hartman authored
The cuttlefish system is a virtual SoC architecture based on QEMU. It uses the QEMU ivshmem feature to share memory regions between guest and host with a custom protocol. Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Arve Hjønnevåg <arve@android.com> Cc: Todd Kjos <tkjos@android.com> Cc: Martijn Coenen <maco@android.com> Cc: Dan Carpenter <dan.carpenter@oracle.com> Cc: devel@driverdev.osuosl.org Cc: kernel-team@android.com Signed-off-by:
Greg Hartman <ghartman@google.com> [astrachan: rebased against 4.16, added TODO, fixed checkpatch issues] Signed-off-by:
Alistair Strachan <astrachan@google.com> Signed-off-by:
-
Todd Poynor authored
As a result of various previous patches, ion_buffer_create() now has two sets of identical statements for initializing two fields of the buffer struct, next to each other. Remove one set. Move the initialization of these two fields together with the statements that initialize the other two fields from the function parameters, prior to the heap allocate() call, for consistency. Signed-off-by:
Todd Poynor <toddpoynor@google.com> Acked-by:
Sumit Semwal <sumit.semwal@linaro.org> Signed-off-by:
-
NeilBrown authored
ldlm_lock_create() gets a resource, but don't put it on all failure paths. It should. Signed-off-by:
-
NeilBrown authored
1/ the return value of ldlm_resource_putref() is never used, so change it to return 'void'. 2/ Move all of the code to run on the last putref to __ldlm_resource_putref_final(). This means a lock is taken in one function and dropped in another, but that isn't too uncommon, and will disappear in a future patch. Now that the code it together, it becomes apparent that we are dropping a ref on the namespace *before* the last use. So keep the ref until after. Signed-off-by:
-
