This patchset adds kernel AIO support for CIFS module. Currently the code processes all read and write requests synchronously regardless of existence of iocb->ki_complete() callback. It is not what kernel AIO expects when a user submits i/o calls through io_submit() interface. That's why new aio context was introduced: it allowed to process i/o responses in a separate thread and return to the caller immediately if ki_complete() is specified. For synchronous variants the code simply needs to wait in the current thread until aio context processing is completed. The work scope is divided into two parts for reading and writing. With these patches applied i/o performance benefits from increasing of i/o queue depth size (e.g. in FIO). The checkpatch.pl script shows 2 warnings related to using of ENOSYS error code but fixing this is out of scope of this patchset (we have plenty of places that use ENOSYS for the same logic and they all should be fixed at once). Changes since v1: - move async i/o context introduction to a separate patch - use preallocated page array and iov_iter_get_pages - call kmalloc for allocations less than 1MB and vmalloc for others - minor cleanup Pavel Shilovsky (3): CIFS: Add asynchronous context to support kernel AIO CIFS: Add asynchronous read support through kernel AIO CIFS: Add asynchronous write support through kernel AIO fs/cifs/cifsglob.h | 19 +++ fs/cifs/cifsproto.h | 3 + fs/cifs/file.c | 357 +++++++++++++++++++++++++++++++++++++++------------- fs/cifs/misc.c | 120 ++++++++++++++++++ 4 files changed, 409 insertions(+), 90 deletions(-) -- 2.7.4 -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
