mm: implement mremap(2)

[gburd]

What

Implements mremap(2), which was missing entirely on OSv: it was not in the
syscall table and had no libc entry point, so applications and allocators that
resize an mmap region failed (glibc/musl realloc of an mmap-backed chunk calls
the internal __mremap).

How

mmu::mremap() (core/mmu.cc) is built on the existing VMA machinery and the
public, self-locking map_anon()/map_file()/munmap() primitives. It handles
a single, wholly mapped source vma (Linux likewise requires the old range to
fall within one mapping; a range spanning several vmas or a hole yields
EFAULT):

• shrink / no-op: drop the tail in place, address unchanged;
• grow in place: when the region immediately after the source is free;
• grow with MREMAP_MAYMOVE: allocate a new region of the new size, migrate
  the contents (anonymous: copy; file-backed: remap the file at the same offset
  so the grown tail is served by the fault path with no copy), then unmap the old
  range.

The source vma is inspected once under vma_list_mutex to snapshot its
perm/flags/type/file/offset and whether its tail is free, so the rest of the
function can call the self-locking primitives without nesting the non-recursive
vma write lock.

MREMAP_FIXED is not supported yet and returns EINVAL rather than doing the
wrong thing; the anonymous and file-backed paths cover the callers we have.

Wired as syscall SYS_mremap (via long_mremap, mirroring long_mmap since the
return value is a pointer) with a matching tracepoint, plus the libc
mremap()/__mremap() entry points, and both symbols exported from
libc.so.6 and ld-musl.so.1.

Testing

tests/tst-mremap.cc covers grow, shrink, in-place-blocked (ENOMEM without
MREMAP_MAYMOVE), file-backed move, and the EINVAL/EFAULT error paths. It
passes on OSv under KVM and on native Linux (semantics match). tst-mmap
continues to pass, so there is no regression in the VMA path.

(Recreated from #1412, which GitHub auto-closed when its branch was rebased onto current master. Same change, rebased and build/boot-verified on master 3aba46c.)

[Copilot]

Pull request overview

Adds initial support for Linux mremap(2) in OSv, wiring it end-to-end (MMU implementation → syscall plumbing → libc entry points) and introducing a dedicated test to validate expected grow/shrink/move behaviors.

Changes:

• Implement mmu::mremap() using existing VMA primitives, supporting shrink, in-place grow, and MAYMOVE-based relocation (with MREMAP_FIXED rejected).
• Wire SYS_mremap via long_mremap, add a syscall tracepoint, and add libc mremap() / __mremap() entry points.
• Add tst-mremap and integrate it into the tests module build.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
tests/tst-mremap.cc	New functional test coverage for anon and file-backed mremap() behaviors and error paths.
syscalls/syscalls.cc.in	Adds long_mremap syscall declaration to the generated syscall table.
syscalls/syscall_tracepoints.cc.in	Adds a syscall tracepoint for long_mremap.
modules/tests/Makefile	Registers the new tst-mremap.so test module.
linux.cc	Adds a Linux-hosted long_mremap() shim for syscall emulation/testing.
libc/mman.cc	Adds libc mremap() wrapper and internal __mremap() symbol.
include/osv/mmu.hh	Declares the new mmu::mremap() API.
exported_symbols/osv_libc.so.6.symbols	Exports mremap from OSv’s glibc-compat libc.
exported_symbols/osv_ld-musl.so.1.symbols	Exports mremap from OSv’s musl loader.
core/mmu.cc	Implements mmu::mremap() and supporting helpers using VMA machinery.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[gburd]

Thanks for the review (and Copilot). Addressed in the current tip (3f8152e, rebased onto master and squashed into one coherent commit):

• "ponytail" word (comment): removed; the doc comment now reads as a plain TODO:.
• Missing <errno.h> in the test: added.
• old_size == 0: now rejected with EINVAL (it is only meaningful on Linux for mirroring a shareable mapping, which we do not implement; a zero old_size would otherwise "grow" from nothing or munmap a zero-length tail).
• Shrink path ignored munmap's return: now checked. The source range was verified wholly mapped, so this cannot fail with EINVAL, but any error is now surfaced (via to_libc() + MAP_FAILED) rather than reporting a success that left the mapping unchanged.
• File offset when old_addr starts mid-VMA: inspect_mremap_src() now records fvma->offset() + (start - vma_start), so a sub-range remap re-derives the correct file bytes instead of the VMA's base offset.
• MAP_PRIVATE file move dropping COW-dirtied pages: rather than silently returning success with lost data, a writable MAP_PRIVATE file mapping move is now refused with EINVAL. A correct private-dirty-page copy is left as a TODO in the comment.
• __mremap not exported: added to both osv_ld-musl.so.1.symbols and osv_libc.so.6.symbols (the __mremap entry point already existed in libc/mman.cc). Note OSv uses its own core/mempool allocator rather than musl's mallocng/oldmalloc, so nothing internal depends on __mremap, but exporting it is harmless and covers any caller that resolves the internal name.

Needs a build to confirm it compiles and the test passes on the OSv image; I have queued that.

[gburd]

Pushed a couple of build fixes on top (now c12b8121):

• The shrink-path munmap error check I'd added used a struct error in a boolean context (no implicit conversion) and then an if-init-statement, which the kernel's gnu++14 -Werror rejects. Now a plain two-statement auto e = munmap(...); if (e.bad()) { ... }.
• Added an old_size == 0 -> EINVAL assert to test_errors() so that guard is exercised, not just source-verified.

Rebuilt image=tests: core/mmu.cc compiles, __mremap links (present in both osv_ld-musl.so.1.symbols and osv_libc.so.6.symbols, linker-script sort intact), and tst-mremap passes all asserts (grow/shrink/no-maymove/file-move/errors).