If we have multiple simultaneous readers on a single /dev/bpf fd it's possible
for the assertion after the bpf_uiomove() in bpfread() to fail.
Note that the bpf_uiomove() is done outside of the BPFD_LOCK, because uiomove
may sleep. As a result it's possible for another thread to have already
reclaimed toe bd_hbuf, thus causing us to fail the assertion.
Even without INVARIANTS this may provoke panics.
That results (with INVARIANTS) in a panic such as:
login: panic: bpfread: lost bd_hbuf
cpuid = 13
time = 1740567635
KDB: stack backtrace:
db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe003972db10
vpanic() at vpanic+0x136/frame 0xfffffe003972dc40
panic() at panic+0x43/frame 0xfffffe003972dca0
bpfread() at bpfread+0x2e8/frame 0xfffffe003972dce0
devfs_read_f() at devfs_read_f+0xe4/frame 0xfffffe003972dd40
dofileread() at dofileread+0x80/frame 0xfffffe003972dd90
sys_read() at sys_read+0xb7/frame 0xfffffe003972de00
amd64_syscall() at amd64_syscall+0x15a/frame 0xfffffe003972df30
fast_syscall_common() at fast_syscall_common+0xf8/frame 0xfffffe003972df30
- syscall (3, FreeBSD ELF64, read), rip = 0x302787166afa, rsp = 0x302782638a78, rbp = 0x302782638aa0 ---
Also add a test case replicating the known trigger for this panic.
Sponsored by: Rubicon Communications, LLC ("Netgate")