so that there is a better chance to recover after down/up.
On top of this implement mwx_activate() so that suspend/resume works.
None of this is perfect but hopefully good enough.
Tested on MT7921
the Linux implementation is now in place. With this update, the
kernel can support ports that talk directly to /dev/fuse0 rather
than relying on libfuse.
sys/fusebuf.h is retained rather than introducing fuse_kernel.h
OK claudio@
panes which want redraw for deferred clients - if they are deferred then
they can just have a full redraw instead. Also return earlier if no
redraw is actually needed, and improve the comments.
This commit starts unifying various CMS signed object parsers in
preparation for a lot more refactoring of this long-accumulated
copy-paste mess.
The signed object is passed as a void * object to various handlers, so
these handlers all have the same signature.
In this step, *_parse() is essentially split into allocation plus:
1. *_cert_info(), which checks some basic things on the EE cert (usually
inheritance, presence or absence of RFC 3779 extensions). For MFTs
it also extracts some info and hangs that off mft.
2. the already existing *_parse_econtent()
3. *_validate() that does some validation steps and sets ->valid.
In most signed object handlers the validate step can't currently
fail. This is one of the many warts we've accumulated and is marked
with an /* XXX */.
This is all straightforward and should not change anything.
The reason cert_info comes after parse_econtent is that the latter
usually has some helpers and it is easier to reason about this if
the future struct members are somewhat close to each other.
ok claudio job
Previously, the ')' command required a sentence to end with two
blanks, but '(' only required a single space. Historic vi (and the
documentation) requires two spaces in both directions.
Based on a diff from Walter Alejandro Iglesias.
When btrace(8) is active, refcounting is traced using an index field
to the dt(4) backend. When two CPU simultaneously decrement the
reference count, one could free the object while the other is still
reading the index. Move the load before the dec separated by a
membar. Crashes seen while testing an experimental diff from dlg@.
OK cludwig@
For a sentence spanning more than one line at the end of the file,
when the cursor is placed at the first character of any line except
for the last one, running '!)<cmd>' won't affect the last line.
From Walter Alejandro Iglesias
Fixes a problem where an extra line is printed at the end of the
output when the "number" command is given the "l" (literal display)
flag.
From Jeremy Mates Walter Alejandro Iglesias and
We check this for extensions we know about in cert_parse_extensions().
We keep doing it there since we need to keep track of the extensions we
encountered anyway.
While cert_parse_extensions() rejects certs with critical extensions we
don't know about, we allow duplicate non-critical ones mainly because
that's annoying to keep track of. LibreSSL's libcrypto checks for this
and rejects the cert when caching the extensions, OpenSSL 4 adds a new
EXFLAG_DUPLICATE flag and accepts the cert, and OpenSSL 3 simply ignores
duplicates. In short: we get to do it ourselves.
This check is basically lifted from libcrypto's x509_purp.c with a few
extra contortions due to const sprinkling and making things opaque. The
warnx is the same as the one already present in cert_parse_extensions().
We do not NULL check X509_EXTENSION_get_object() because the extension
parsed, so an OID is present.
ok job
We clamp the amount of data we are willing to parse to the length of
the UDP packet as indicated by the UDP header length field. While we
made sure that the length field did not point past the received data,
we never checked if the length is smaller than the udp header. Since
we are using BPF, the kernel also does not doe this for us. This might
in turn lead to an underflow and a subsequent crash of the engine
process.
Pointed out and diff provided by Andrew Griffiths, thanks!
Valid Rsync URIs always contain a module component.
To avoid duplication of URI validation code, refactor rsync_base_uri()
to optionally allocate & dup the base URI portion.
Thanks to Ties de Kock for reporting.
OK tb@ claudio@
OpenBIOS reports QEMU IDE disks with an intermediate ide node, while the
sparc64 bootpath code only treats ata as a channel node.
The disk is attached as wd0, but the final disk component is not nailed
as the boot device, so the kernel asks for the root device.
No QEMU level option seems possible to enforce ata disk.
OK: claudio@
claudio