Hi Joe, On 3/26/22 00:35, Joe Damato wrote: > The description of SPLICE_F_GIFT explains what happens when the flag is > specified, but it was unclear to whether data spliced into the pipe with > vmsplice, but without SPLICE_F_GIFT, is copied immediately or simply mapped > into the kernel and copied by splice(2) SPLICE_F_MOVE later. > > Tests and kernel code reveal that vmsplice(2) maps the memory into the > kernel and it is copied later; this means applications can't e.g. free the > memory after a successful call to vmsplice(2) even when they omit the > SPLICE_F_GIFT flag. Please paste (if short enough) or link (if too long) to those tests or kernel code, to make it easier to review this commit (and especially if someone wants to review it in the future). Thanks, Alex > > Signed-off-by: Joe Damato <jdamato@xxxxxxxxxx> > --- > man2/vmsplice.2 | 36 +++++++++++++++++++++++++++++++----- > 1 file changed, 31 insertions(+), 5 deletions(-) > > diff --git a/man2/vmsplice.2 b/man2/vmsplice.2 > index 9102c4c..b96bcc6 100644 > --- a/man2/vmsplice.2 > +++ b/man2/vmsplice.2 > @@ -116,17 +116,43 @@ otherwise the page cache and on-disk data may differ. > Gifting pages to the kernel means that a subsequent > .BR splice (2) > .B SPLICE_F_MOVE > -can successfully move the pages; > -if this flag is not specified, then a subsequent > -.BR splice (2) > -.B SPLICE_F_MOVE > -must copy the pages. > +can successfully move the pages. > Data must also be properly page aligned, both in memory and length. > .\" FIXME > .\" It looks like the page-alignment requirement went away with > .\" commit bd1a68b59c8e3bce45fb76632c64e1e063c3962d > .\" > .\" .... if we expect to later SPLICE_F_MOVE to the cache. > + > +If > +.B SPLICE_F_GIFT > +is not specified, the kernel will still map > +the memory into > +.I fd. > +No copy is made during the invocation of > +.BR vmsplice (2) > +itself and so the application must take care when modifying the data after > +successful calls to > +.BR vmsplice(2). > +Subsequent > +.BR splice (2) > +.B SPLICE_F_MOVE > +will make a copy of the data that was mapped into > +.I fd > +during > +.BR vmsplice (2). > +Thus, any modifications to the memory after a successful call to > +.BR vmsplice(2) > +will appear when > +.BR splice(2) > +.B SPLICE_F_MOVE > +generates its copy. > +Once usage of > +.BR splice(2) > +is complete and the pipe is closed, the application will be the only user > +of the data and can freely modify the data as needed (e.g. by > +overwriting it, releasing the memory to the allocator, etc). > + > .SH RETURN VALUE > Upon successful completion, > .BR vmsplice () -- Alejandro Colomar Linux man-pages comaintainer; https://www.kernel.org/doc/man-pages/ http://www.alejandro-colomar.es/
