devel/woodpecker, devel/woodpecker-agent: New ports
Woodpecker is a CI/CD engine driven by YAML pipelines in the repository,
for Forgejo, Gitea, GitHub, GitLab and Bitbucket.
devel/woodpecker installs the server (web UI and API) and the CLI;
devel/woodpecker-agent, a slave port, installs the agent, which runs
pipelines with the local backend. Agents usually run on other hosts than
the server, and only the agent needs git and plugin-git.
The release's source tarball carries vendor/ and the built web UI, so
nothing is fetched at build time.
Add the woodpecker user and group (408).
WWW: https://woodpecker-ci.org/
Reviewed by: dtxdf
Approved by: dtxdf (mentor)
Differential Revision: https://reviews.freebsd.org/D61666
devel/woodpecker-plugin-git: New port
plugin-git clones the repository at the start of a Woodpecker CI
pipeline. With the local backend, woodpecker-agent runs it from PATH;
without it, the agent downloads a binary from GitHub at runtime.
WWW: https://github.com/woodpecker-ci/plugin-git
Reviewed by: dtxdf
Approved by: dtxdf (mentor)
Differential Revision: https://reviews.freebsd.org/D61665
sysutils/cni-epair: Update to 1.2.2
Apply configured MTU (or the bridge's own MTU by default) to both ends
of the epair, keep error messages valid JSON, return the DHCP lease's
DNS server in dns.nameservers, and search CNI_PATH before
${LOCALBASE}/libexec/cni for IPAM plugins.
Changelog: https://github.com/daemonless/cni-epair/releases/tag/v1.2.2
Reviewed by: dtxdf
Approved by: dtxdf (mentor)
Differential Revision: https://reviews.freebsd.org/D61688
devel/py-pyutil: Update to 3.4.1
pyutil still has Python 2 code that fails at run time on every Python
3; upstream master has the same code. Fix it:
- scripts/verinfo imports the Python 2 exceptions module.
- cache, odict, dictutil, memutil and humanreadable use next() only
iterators, has_key(), iteritems()/iterkeys(), dict views as lists, a
lazy map() for eviction, the reduce, unicode and long builtins and
func_name/im_func; SmallOrderedDict's iterators and setdefault() were
wrong as well, and describe_all_objects() did an implicit relative
import.
- benchutil imports threading as thread but calls _thread functions,
and makeg() uses xrange.
- The console scripts: randfile and lines write str to binary files,
lines and passphrase -d use the 'rU' mode removed in Python 3.11,
try_decoding calls .decode() on a str, randcookie passes a float to
os.urandom() and imports zbase32, which is not ported (encode
z-base-32 locally instead).
- platformutil.get_linux_distro() uses 'rU' and platform.dist();
[26 lines not shown]
sysutils/noct: Add new port
Noct is a command-line filesystem inspection and analysis toolkit for
Unix-like systems. It combines fast directory listing with specialized
views for ownership, permissions, sizes, ages, duplicate files, scripts,
extensions, filesystem boundaries, cleanup candidates, and more.
It supports rich terminal colors, icons, themes, Tree output, filtering,
sorting, configurable long listings, and focused inspection modes.
WWW: https://github.com/ferusx/noct
PR: 297935
Approved by: osa, vvd (mentors, implicit)
devel/py-sentinels: Add new port
Sentinels are objects with special meanings. They can be thought of as
singletons, but they serve the need of having special values that have
special meanings. This module provides the Sentinel class along with
useful predefined instances.
WWW: https://github.com/vmalloc/sentinels
PR: 300721
Approved by: osa, vvd (mentors, implicit)
databases/py-mongomock: Add new port
Mongomock is a small library to help testing Python code that interacts
with MongoDB via Pymongo. It implements a fake in-memory MongoDB server
and provides a subset of PyMongo's API, allowing tests to run without a
running MongoDB instance.
WWW: https://github.com/mongomock/mongomock
PR: 300720
Approved by: osa, vvd (mentors, implicit)
vmm: Avoid redundant AMD DEBUGCTL writes
Skip clearing and restoring DEBUGCTL when the saved host value is
zero. The entry path leaves DEBUGCTL zero, and LBR virtualization
restores that host value on VM exit. A rejected guest entry also
returns with the pre-VMRUN DEBUGCTL value. Guest accesses to DEBUGCTL
remain intercepted.
Preserve the existing writes for nonzero host state. Avoiding the
redundant writes can also avoid intercepted MSR accesses when bhyve
runs under another hypervisor.
Reviewed by: markj, bnovkov
Approved by: markj (mentor)
Sponsored by: Alchemilla Ventures Private Limited
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D61218
vmm: Avoid redundant Intel DEBUGCTL writes
Skip clearing and restoring IA32_DEBUGCTL when the saved host value
is zero.
VM exit clears the register. If VM entry fails, the entry path has
already cleared any nonzero host value, so restoring zero is also
unnecessary in that case. Preserve the existing writes for nonzero
host state.
Avoiding these writes can also avoid intercepted MSR accesses when
bhyve runs under another hypervisor.
Reviewed by: markj, bnovkov
Approved by: markj (mentor)
Sponsored by: Alchemilla Ventures Private Limited
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D61219
vfs: fail VOP_DEALLOCATE passes that make no progress
VOP_DEALLOCATE(9) permits a filesystem to return a partial result, and
vn_deallocate_impl() simply loops until the range is consumed. If a
filesystem returns success without shrinking the remaining length, we
spin forever with the rangelock held and no way to interrupt the thread.
Return EIO if a pass doesn't make progress so that a buggy filesystem
doesn't wedge the caller.
Document the progress requirement in VOP_DEALLOCATE(9) and the new EIO
condition in fspacectl(2).
I did wonder if this should be made interruptible for, e.g., slow I/O
over NFS, but I've declined to touch it for now.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D61496
tmpfs: fix a hang reachable through hole-punching
Deallocate requests that clip only the start of a single page need to be
taken into account as well- it misses the first branch because
startofs == 0, but then it also misses the last branch because the start
and end are on the same page.
Handling this case in either branch is equally valid, so I've
arbitrarily chosen the first branch since it already has single-page
logic.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D61490
tmpfs: fix append-only file handling for writes and hole-punching
Write protection assumed that O_APPEND handling at open time is
sufficient, but the flag can be cleared with F_SETFL. That's maybe
worth rejecting, but that's harder to do without some clear layering
violations.
Hole-punching also violates the spirit of append-only, so reject that
on files that are append-only.
Note that we don't reject IMMUTABLE here- UFS seems to historically
not care if a file transitions to IMMUTABLE after it's opened, so follow
suit here instead of trying to tighten it up. ZFS will apparently use
the more constraining logic, but that's OK for now.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D61491
vfs: fail VOP_ALLOCATE passes that make no progress
vn_fallocate() loops over VOP_ALLOCATE until the requested range has
been allocated, permitting partial results. It only asserts that each
pass makes progress, so on a non-INVARIANTS kernel a filesystem that
returns success without shrinking the remaining length leaves the caller
spinning forever.
Drop the assertion and just return EIO in that case, as was just done
for vn_deallocate_impl(). The NFS server already does the same in its
own VOP_ALLOCATE loop.
Document the progress requirement in VOP_ALLOCATE(9) and the new EIO
condition in posix_fallocate(2).
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D61497