On 12/08/21 16:21, Richard W.M. Jones wrote:
On Wed, Dec 08, 2021 at 01:20:49PM +0100, Laszlo Ersek wrote:
Add the "Nbdkit.get_disk_allocated" helper function, which calculates the number of allocated bytes in an output disk image, through a corresponding NBD server connection.
Original "NBD.block_status" summation example code by Rich Jones (thanks!), chunking, metadata context simplification, and virt-v2v integration by myself. Chunking added because "NBD.block_status" is not 64-bit clean just yet (Eric Blake is working on it).
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2027598 Signed-off-by: Laszlo Ersek <lersek@redhat.com> --- lib/nbdkit.mli | 8 ++++++ lib/nbdkit.ml | 30 ++++++++++++++++++++ 2 files changed, 38 insertions(+)
diff --git a/lib/nbdkit.mli b/lib/nbdkit.mli index 825755e61dbe..03107dab109d 100644 --- a/lib/nbdkit.mli +++ b/lib/nbdkit.mli @@ -117,3 +117,11 @@ val with_connect_unix : socket:string -> in [meta_contexts] requested (each of which is not necessarily supported by the server though). The connection is torn down either on normal return or if the function [f] throws an exception. *) + +val get_disk_allocated : dir:string -> disknr:int -> int64 option +(** Callable only in the finalization step. [get_disk_allocated dir disknr] + examines output disk [disknr] through the corresponding NBD server socket + that resides in [dir]. Returns the number of bytes allocated in the disk + image, according to the "base:allocation" metadata context. If the context + is not supported by the NBD server behind the socket, the function returns + None. *)
This really is the right function, wrong module.
Since it only applies to output modules, the logical place is in output/output.ml, where there are already a bunch of random helper functions.
That's precisely where I put it first (given that we have "get_disks" there too!). However, I proved simply unable to *call* the function, from "lib/create_ovf.ml", in the next patch! I tried "Output.get_disk_allocated", and I got a weird compilation failure, even though the function was properly declared in "output.mli". At that point, I couldn't decide if I was missing something about the tricky "first class modules" thing, or if it was really a layering violation to call an Output function from a *lower level* utility function (such as "create_ovf" in "lib/create_ovf.ml"). I briefly considered that, whoever called create_ovf from the top level, could *pass in* Output.get_disk_allocated to it, as a callback function. But that seemed so horrendously ugly (and uncalled-for) that I dropped the idea. Yet another idea was to collect the actual sizes for all the output disks in one go, separately, then pass in that list, to create_ovf. Then in create_ovf, I would replace: (* Iterate over the disks, adding them to the OVF document. *) List.iteri ( fun i (size, image_uuid, vol_uuid) -> ... ) (List.combine3 sizes image_uuids vol_uuids) with "combine4", adding a fourth component (the actual size) to the tuple parameter of the nameless iterator function. But that idea looked too complex as well. Halp! :)
diff --git a/lib/nbdkit.ml b/lib/nbdkit.ml index 43e0fc5c728a..e142c210c66f 100644 --- a/lib/nbdkit.ml +++ b/lib/nbdkit.ml @@ -229,3 +229,33 @@ let with_connect_unix ~socket ~meta_contexts ~f = ~finally:(fun () -> NBD.shutdown nbd) ) ~finally:(fun () -> NBD.close nbd) + +let get_disk_allocated ~dir ~disknr = + let socket = sprintf "%s/out%d" dir disknr + and alloc_ctx = "base:allocation" in + with_connect_unix ~socket ~meta_contexts:[alloc_ctx] + ~f:(fun nbd -> + if NBD.can_meta_context nbd alloc_ctx then ( + (* Get the list of extents, using a 2GiB chunk size as hint. *) + let size = NBD.get_size nbd + and allocated = ref 0_L + and fetch_offset = ref 0_L in + while !fetch_offset < size do + let remaining = size -^ !fetch_offset in + let fetch_size = min 0x8000_0000_L remaining in + NBD.block_status nbd fetch_size !fetch_offset + (fun ctx offset entries err -> + assert (ctx = alloc_ctx); + for i = 0 to Array.length entries / 2 - 1 do + let len = Int64.of_int32 entries.(i * 2) + and typ = entries.(i * 2 + 1) in + if Int32.logand typ 1_l = 0_l then + allocated := !allocated +^ len; + fetch_offset := !fetch_offset +^ len + done; + 0 + ) + done; + Some !allocated + ) else None + )
But the function itself is fine.
Thanks! Laszlo