[fuse-devel] Understanding OSX Fuse driver flow.

[Irad K]

Hi all,

I wish to learn more about the backend of osx fuse.

I've started with some research about the kernel-user communication by
following the data path in read system call.
The path starting from user-space entity that initiate the action, to the
driver that serve as a path through to user-space filesystem deamon that
perform the action.

As a start, I've focused on read sys call and discovered that this action
is forwarded from vfs layer to *fuse_vnop_read*, with the data kept in '
*uio*' arguments which is the command arguments in user-space (including
buffer for data to be read).

based on these params and other params like the context details, the struct
fuse_dispatcher is created, and this is the entity that eventually being
sent to the deamon more or less ..

here's the fuse_dispatcher class:


struct fuse_dispatcher

{

struct fuse_ticket *tick;

struct fuse_in_header *finh;


void *indata;

size_t iosize;

uint64_t nodeid;

int answ_stat;

void *answ;

};


and here's the command that send the dispatcher and wait for response (from
daemon)

err = fdisp_wait_answ(&fdi);

However, there's some other layer which seem to actually send the data to
user-space, which called *fuse_device_cdevsw* which posses the method
*fuse_device_read. *

Perhaps do you know how is this method connected to the read flow ?

basically what it does is dequeuing the ticket (part of the dispatcher data
we collected in vnop_read), and send it to the deamon. is it working from a
different thread and signal back to vnop_read (in particular
fdisp_wait_answ)
once the read has been completed and response has returned ?

thanks,
Irad

[Nikolaus Rath]

Post by Irad K
Hi all,
I wish to learn more about the backend of osx fuse.

[...]

Then the best place to ask is probably the OSXFUSE mailing list. I don't
think we can help you here.

Best,
-Nikolaus