GFP masks used from FS/IO context¶
| Date: | May, 2018 |
|---|---|
| Author: | Michal Hocko <mhocko@kernel.org> |
Introduction¶
Code paths in the filesystem and IO stacks must be careful whenallocating memory to prevent recursion deadlocks caused by directmemory reclaim calling back into the FS or IO paths and blocking onalready held resources (e.g. locks - most commonly those used for thetransaction context).
The traditional way to avoid this deadlock problem is to clear __GFP_FSrespectively __GFP_IO (note the latter implies clearing the first as well) inthe gfp mask when calling an allocator. GFP_NOFS respectively GFP_NOIO can beused as shortcut. It turned out though that above approach has led toabuses when the restricted gfp mask is used “just in case” without adeeper consideration which leads to problems because an excessive useof GFP_NOFS/GFP_NOIO can lead to memory over-reclaim or other memoryreclaim issues.
New API¶
Since 4.12 we do have a generic scope API for both NOFS and NOIO contextmemalloc_nofs_save,memalloc_nofs_restore respectivelymemalloc_noio_save,memalloc_noio_restore which allow to mark a scope to be a criticalsection from a filesystem or I/O point of view. Any allocation from thatscope will inherently drop __GFP_FS respectively __GFP_IO from the givenmask so no memory allocation can recurse back in the FS/IO.
- unsigned int
memalloc_nofs_save(void)¶ Marks implicit GFP_NOFS allocation scope.
Parameters
void- no arguments
Description
This functions marks the beginning of the GFP_NOFS allocation scope.All further allocations will implicitly drop __GFP_FS flag and sothey are safe for the FS critical section from the allocation recursionpoint of view. Use memalloc_nofs_restore to end the scope with flagsreturned by this function.
This function is safe to be used from any context.
- void
memalloc_nofs_restore(unsigned int flags)¶ Ends the implicit GFP_NOFS scope.
Parameters
unsignedintflags- Flags to restore.
Description
Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.Always make sure that the given flags is the return value from thepairing memalloc_nofs_save call.
- unsigned int
memalloc_noio_save(void)¶ Marks implicit GFP_NOIO allocation scope.
Parameters
void- no arguments
Description
This functions marks the beginning of the GFP_NOIO allocation scope.All further allocations will implicitly drop __GFP_IO flag and sothey are safe for the IO critical section from the allocation recursionpoint of view. Use memalloc_noio_restore to end the scope with flagsreturned by this function.
This function is safe to be used from any context.
- void
memalloc_noio_restore(unsigned int flags)¶ Ends the implicit GFP_NOIO scope.
Parameters
unsignedintflags- Flags to restore.
Description
Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.Always make sure that the given flags is the return value from thepairing memalloc_noio_save call.
FS/IO code then simply calls the appropriate save function beforeany critical section with respect to the reclaim is started - e.g.lock shared with the reclaim context or when a transaction contextnesting would be possible via reclaim. The restore function should becalled when the critical section ends. All that ideally along with anexplanation what is the reclaim context for easier maintenance.
Please note that the proper pairing of save/restore functionsallows nesting so it is safe to callmemalloc_noio_save ormemalloc_noio_restore respectively from an existing NOIO or NOFSscope.
What about __vmalloc(GFP_NOFS)¶
vmalloc doesn’t support GFP_NOFS semantic because there are hardcodedGFP_KERNEL allocations deep inside the allocator which are quite non-trivialto fix up. That means that callingvmalloc with GFP_NOFS/GFP_NOIO isalmost always a bug. The good news is that the NOFS/NOIO semantic can beachieved by the scope API.
In the ideal world, upper layers should already mark dangerous contextsand so no special care is required and vmalloc should be called withoutany problems. Sometimes if the context is not really clear or there arelayering violations then the recommended way around that is to wrapvmallocby the scope API with a comment explaining the problem.