SCM_RIGHTS #146
Merged
SCM_RIGHTS #146
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
heatd
force-pushed
the
pedro/scm_rights
branch
from
7b93e77 to
11ce0f6
Compare
heatd
force-pushed
the
pedro/scm_rights
branch
8 times, most recently
from
c539e0e to
3d95c3c
Compare
This is nicer and safer. Signed-off-by: Pedro Falcato <[email protected]>
Certain systems give you LAPIC structures with id 255. This is supposed to be interpreted as "does not exist" and skipped. While we're at it, check if we indeed can online the CPU (the firmware wasn't marked it unonlineable). Signed-off-by: Pedro Falcato <[email protected]>
We were doing map[] when we really can't do so, because of descriptor_size != sizeof(EFI_MEMORY_DESCRIPTOR). Such is life. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Fix setgroups crashes when clearing it, in case c->groups is NULL (we don't have supplemental groups). Signed-off-by: Pedro Falcato <[email protected]>
MSI seems to be broken and untested. Don't enable it, for now. Signed-off-by: Pedro Falcato <[email protected]>
This should catch bad callers. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Still a work in progress, but already much better than the old one. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
This patch adds MSI-X support and revamps IRQ allocation APIs for PCI drivers. New drivers should use pci_alloc_irqs + pci_install_irq. Signed-off-by: Pedro Falcato <[email protected]>
And while we're at it, revamp log messages with dev_*() helpers. Signed-off-by: Pedro Falcato <[email protected]>
This stops us from missing interrupts when multiple IRQ handlers are installed on a single edge-triggered irq line. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Move a bunch of file_ops into inode_operations (new) and vm_obj_ops. This is merely the first step for unscrewing a bunch of problems with file_ops and making writing filesystems/fds easier and more ergonomic. Signed-off-by: Pedro Falcato <[email protected]>
After d_path started requiring a valid mount pointer, un_get_name started crashing the kernel. Add a few wrappers for it, and make AF_UNIX DTRT. Signed-off-by: Pedro Falcato <[email protected]>
Add getaffinity and setaffinity syscalls. They're still stubbed and we
don't actually properly support affinity yet. However, it's good enough
for _SC_NPROCESSORS_{ONLN,CONF}.
Signed-off-by: Pedro Falcato <[email protected]>
Normally, signal pending would mean we have a proper signal waiting for us. However, stuff like group stop and group exit set signal pending while not having a proper signal number or signal struct. This made sigtimedwait crash if the process exited mid-timedwait. Signed-off-by: Pedro Falcato <[email protected]>
Stopping on inode = 0 is wrong, because this is simply an ignored entry, not the end of the list. This repro'd in directories we were simply not able to delete, because they looked empty from getdirent's PoV, but not from elsewhere (open and ext2_dir_empty did the right thing). Fix it. Signed-off-by: Pedro Falcato <[email protected]>
Make them take better types (i.e don't take struct file in stat). Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Also add a d_path_under_root that returns NULL when the path is not under the current root. Unused, For Now(tm). Signed-off-by: Pedro Falcato <[email protected]>
Small helpers for getdirent implementations to use. Signed-off-by: Pedro Falcato <[email protected]>
sched_lock wasn't safe when thread migration was involved, which ended up causing havoc and crashes. Fixes: 66b6df9 ("sched: Add thread migration") Signed-off-by: Pedro Falcato <[email protected]>
We can actually madvise parts of a vma, so detect whether this is the first VMA we're looking at. Fixes tlb_flush1. Signed-off-by: Pedro Falcato <[email protected]>
Use maple trees with RCU support, and improve lots of stuff. TTYs now hold pid references directly. Signed-off-by: Pedro Falcato <[email protected]>
mm_address_space now has two decoupled refcounts, as with Linux. active_mm is used to delay TLB flushing when switching to kernel threads. They keep a ref (mm_count) which holds the top level pgd (any kernel mappings) and the mm_address_space in place. User-level mappings are wiped out when mm_users = 0. Also make vm.cpp -> vm.c Signed-off-by: Pedro Falcato <[email protected]>
We passed in a wrong number and thus were screwing up fd passing. Signed-off-by: Pedro Falcato <[email protected]>
This fixes fstat breakage. Fixes: dc767a8 ("vfs: Change inode_operations->stat and ->link's prototypes") Signed-off-by: Pedro Falcato <[email protected]>
get_user32 needs to be done under the futex hashtable spinlocks. This means we can't actually take a page fault here. Use the new pagefault_disable() and _enable() mechanisms to silently handle -EFAULT there, and retry in that case (with a page fault that may in fact sleep). Signed-off-by: Pedro Falcato <[email protected]>
Fix SACK option parsing (we were rejecting options with 2 length). Also add more loss recovery mechanisms. This greatly increases our bandwidth when loss is present. Signed-off-by: Pedro Falcato <[email protected]>
Add a simple procfs (right now with only static files). Signed-off-by: Pedro Falcato <[email protected]>
Add a linux-style /proc/slabinfo. Signed-off-by: Pedro Falcato <[email protected]>
Fixes a new build failure: grub-mkrescue: error: xorriso not found. Signed-off-by: Pedro Falcato <[email protected]>
This fixes madvise breakage on riscv64. Signed-off-by: Pedro Falcato <[email protected]>
The copy_addr_to_sockaddr function was completely busted, didn't fill in the sockaddr nor use the 'addr' parameter (which we need to use, obviously). Fix it. Signed-off-by: Pedro Falcato <[email protected]>
kmem_pointer_to_slab_maybe uses get_mapping_info, which has weird semantics for huge pages. Add a special case for pointers inside PHYS_BASE. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
This fixes UBSAN reports and bad behavior from zero-sized vectors. Signed-off-by: Pedro Falcato <[email protected]>
Add per-socket send buffer limits. Also significantly improve segment sending and fix some latent TCP/IP bugs while we're at it. This patch doesn't yet introduce this functionality for UDP or UNIX sockets. Signed-off-by: Pedro Falcato <[email protected]>
Add receive buffer accounting and wire it up with the TCP receive window and the individual NIC drivers allocating packetbufs. Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
Signed-off-by: Pedro Falcato <[email protected]>
We were not doing send window accounting correctly, which resulted in us sending more data than the remote host could handle. This is wrong and against RFC9293. Fix it by taking end seqs and snd_una + snd_wnd into consideration. Signed-off-by: Pedro Falcato <[email protected]>
We can't adjust snd_next in tcp_sendmsg, because then ACKs will result in weird segments with sequence numbers that would indicate a gap in data. Instead of that, adjust snd_next in tcp_output. Signed-off-by: Pedro Falcato <[email protected]>
Correct UNIX behavior requires a SIGPIPE when the connection is broken (i.e we can't send or they won't listen). Add a nice common helper while we're at it. Signed-off-by: Pedro Falcato <[email protected]>
heatd
force-pushed
the
pedro/scm_rights
branch
from
be4385d to
6838c3a
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.