Files
bun.sh/src/sys_uv.zig
robobun 6140eb5faf fix(windows): chunk large buffers in preadv/pwritev to avoid integer overflow (#26303)
## Summary
- Fix panic "integer does not fit in destination type" when
reading/writing large files on Windows
- Add chunking for iovec arrays that exceed `c_uint` max entries
- Add chunking for individual buffers that exceed 4GB (`u32` max)

The libuv functions `uv_fs_read` and `uv_fs_write` have two size
limitations:
1. `nbufs` parameter is `c_uint` (32-bit), limiting the number of iovec
entries
2. `uv_buf_t.len` is `ULONG` (u32 on Windows), limiting individual
buffer sizes to 4GB

This change processes large operations in chunks, accumulating results
and updating file positions between chunks.

## Test plan
- [x] Verified `bun run zig:check-all` passes on all platforms
- [x] Verified `bun bd` builds successfully
- [x] Verified basic file read/write operations work correctly

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Bot <claude-bot@bun.sh>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
2026-01-20 12:47:07 -08:00

569 lines
18 KiB
Zig

//! bun.sys.sys_uv is a polyfill of bun.sys but with libuv.
//! TODO: Probably should merge this into bun.sys itself with isWindows checks
comptime {
bun.assert(Environment.isWindows);
}
pub const log = bun.sys.syslog;
pub const Error = bun.sys.Error;
pub const PosixStat = bun.sys.PosixStat;
// libuv dont support openat (https://github.com/libuv/libuv/issues/4167)
pub const openat = bun.sys.openat;
pub const getFdPath = bun.sys.getFdPath;
pub const setFileOffset = bun.sys.setFileOffset;
pub const openatOSPath = bun.sys.openatOSPath;
pub const mkdirOSPath = bun.sys.mkdirOSPath;
pub const access = bun.sys.access;
// Note: `req = undefined; req.deinit()` has a safety-check in a debug build
pub fn open(file_path: [:0]const u8, c_flags: i32, _perm: bun.Mode) Maybe(bun.FileDescriptor) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const flags = uv.O.fromBunO(c_flags);
var perm = _perm;
if (perm == 0) {
// Set a sensible default, otherwise on windows the file will be unusable
perm = 0o644;
}
const rc = uv.uv_fs_open(uv.Loop.get(), &req, file_path.ptr, flags, perm, null);
log("uv open({s}, {d}, {d}) = {d}", .{ file_path, flags, perm, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .open, .path = file_path } }
else
.{ .result = req.result.toFD() };
}
pub fn mkdir(file_path: [:0]const u8, flags: bun.Mode) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_mkdir(uv.Loop.get(), &req, file_path.ptr, flags, null);
log("uv mkdir({s}, {d}) = {d}", .{ file_path, flags, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .mkdir, .path = file_path } }
else
.success;
}
pub fn chmod(file_path: [:0]const u8, flags: bun.Mode) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_chmod(uv.Loop.get(), &req, file_path.ptr, flags, null);
log("uv chmod({s}, {d}) = {d}", .{ file_path, flags, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .chmod, .path = file_path } }
else
.success;
}
pub fn fchmod(fd: FileDescriptor, flags: bun.Mode) Maybe(void) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_fchmod(uv.Loop.get(), &req, uv_fd, flags, null);
log("uv fchmod({}, {d}) = {d}", .{ uv_fd, flags, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .fchmod, .fd = fd } }
else
.success;
}
pub fn statfs(file_path: [:0]const u8) Maybe(bun.StatFS) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_statfs(uv.Loop.get(), &req, file_path.ptr, null);
log("uv statfs({s}) = {d}", .{ file_path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .statfs, .path = file_path } }
else
.{ .result = bun.StatFS.init(req.ptrAs(*align(1) bun.StatFS)) };
}
pub fn chown(file_path: [:0]const u8, uid: uv.uv_uid_t, gid: uv.uv_uid_t) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_chown(uv.Loop.get(), &req, file_path.ptr, uid, gid, null);
log("uv chown({s}, {d}, {d}) = {d}", .{ file_path, uid, gid, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .chown, .path = file_path } }
else
.success;
}
pub fn fchown(fd: FileDescriptor, uid: uv.uv_uid_t, gid: uv.uv_uid_t) Maybe(void) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_fchown(uv.Loop.get(), &req, uv_fd, uid, gid, null);
log("uv chown({}, {d}, {d}) = {d}", .{ uv_fd, uid, gid, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .fchown, .fd = fd } }
else
.success;
}
pub fn rmdir(file_path: [:0]const u8) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_rmdir(uv.Loop.get(), &req, file_path.ptr, null);
log("uv rmdir({s}) = {d}", .{ file_path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .rmdir, .path = file_path } }
else
.success;
}
pub fn unlink(file_path: [:0]const u8) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_unlink(uv.Loop.get(), &req, file_path.ptr, null);
log("uv unlink({s}) = {d}", .{ file_path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .unlink, .path = file_path } }
else
.success;
}
pub fn readlink(file_path: [:0]const u8, buf: []u8) Maybe([:0]u8) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
// Edge cases: http://docs.libuv.org/en/v1.x/fs.html#c.uv_fs_realpath
const rc = uv.uv_fs_readlink(uv.Loop.get(), &req, file_path.ptr, null);
if (rc.errno()) |errno| {
log("uv readlink({s}) = {d}, [err]", .{ file_path, rc.int() });
return .{ .err = .{ .errno = errno, .syscall = .readlink, .path = file_path } };
} else {
// Seems like `rc` does not contain the size?
bun.assert(rc.int() == 0);
const slice = bun.span(req.ptrAs([*:0]u8));
if (slice.len > buf.len) {
log("uv readlink({s}) = {d}, {s} TRUNCATED", .{ file_path, rc.int(), slice });
return .{ .err = .{ .errno = @intFromEnum(bun.sys.E.NOMEM), .syscall = .readlink, .path = file_path } };
}
log("uv readlink({s}) = {d}, {s}", .{ file_path, rc.int(), slice });
@memcpy(buf[0..slice.len], slice);
buf[slice.len] = 0;
return .{ .result = buf[0..slice.len :0] };
}
}
pub fn rename(from: [:0]const u8, to: [:0]const u8) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_rename(uv.Loop.get(), &req, from.ptr, to.ptr, null);
log("uv rename({s}, {s}) = {d}", .{ from, to, rc.int() });
return if (rc.errno()) |errno|
// which one goes in the .path field?
.{ .err = .{ .errno = errno, .syscall = .rename } }
else
.success;
}
pub fn link(from: [:0]const u8, to: [:0]const u8) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_link(uv.Loop.get(), &req, from.ptr, to.ptr, null);
log("uv link({s}, {s}) = {d}", .{ from, to, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .link, .path = from, .dest = to } }
else
.success;
}
pub fn symlinkUV(target: [:0]const u8, new_path: [:0]const u8, flags: c_int) Maybe(void) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_symlink(uv.Loop.get(), &req, target.ptr, new_path.ptr, flags, null);
log("uv symlink({s}, {s}) = {d}", .{ target, new_path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .symlink } }
else
.success;
}
pub fn ftruncate(fd: FileDescriptor, size: isize) Maybe(void) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_ftruncate(uv.Loop.get(), &req, uv_fd, size, null);
log("uv ftruncate({}, {d}) = {d}", .{ uv_fd, size, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .ftruncate, .fd = fd } }
else
.success;
}
pub fn fstat(fd: FileDescriptor) Maybe(bun.Stat) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_fstat(uv.Loop.get(), &req, uv_fd, null);
log("uv fstat({}) = {d}", .{ uv_fd, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .fstat, .fd = fd } }
else
.{ .result = req.statbuf };
}
pub fn fdatasync(fd: FileDescriptor) Maybe(void) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_fdatasync(uv.Loop.get(), &req, uv_fd, null);
log("uv fdatasync({}) = {d}", .{ uv_fd, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .fdatasync, .fd = fd } }
else
.success;
}
pub fn fsync(fd: FileDescriptor) Maybe(void) {
const uv_fd = fd.uv();
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_fsync(uv.Loop.get(), &req, uv_fd, null);
log("uv fsync({d}) = {d}", .{ uv_fd, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .fsync, .fd = fd } }
else
.success;
}
pub fn stat(path: [:0]const u8) Maybe(bun.Stat) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_stat(uv.Loop.get(), &req, path.ptr, null);
log("uv stat({s}) = {d}", .{ path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .stat, .path = path } }
else
.{ .result = req.statbuf };
}
pub fn lstat(path: [:0]const u8) Maybe(bun.Stat) {
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_lstat(uv.Loop.get(), &req, path.ptr, null);
log("uv lstat({s}) = {d}", .{ path, rc.int() });
return if (rc.errno()) |errno|
.{ .err = .{ .errno = errno, .syscall = .lstat, .path = path } }
else
.{ .result = req.statbuf };
}
pub fn close(fd: FileDescriptor) ?bun.sys.Error {
return fd.closeAllowingBadFileDescriptor(@returnAddress());
}
pub fn closeAllowingStdoutAndStderr(fd: FileDescriptor) ?bun.sys.Error {
return fd.closeAllowingStandardIo(@returnAddress());
}
/// Maximum number of iovec buffers that can be passed to uv_fs_read/uv_fs_write.
/// libuv uses c_uint for nbufs, so we must not exceed its maximum value.
const max_iovec_count: usize = std.math.maxInt(c_uint);
/// Maximum size of a single buffer in uv_buf_t.
/// libuv uses ULONG (u32) for the buffer length on Windows.
const max_buf_len: usize = std.math.maxInt(u32);
/// Returns the total byte capacity of a slice of iovec buffers.
fn sumBufsLen(bufs: []const bun.PlatformIOVec) usize {
var total: usize = 0;
for (bufs) |buf| {
total += buf.len;
}
return total;
}
pub fn preadv(fd: FileDescriptor, bufs: []const bun.PlatformIOVec, position: i64) Maybe(usize) {
const uv_fd = fd.uv();
comptime bun.assert(bun.PlatformIOVec == uv.uv_buf_t);
const debug_timer = bun.Output.DebugTimer.start();
var total_read: usize = 0;
var remaining_bufs = bufs;
var current_position = position;
while (remaining_bufs.len > 0) {
const chunk_len = @min(remaining_bufs.len, max_iovec_count);
const chunk_bufs = remaining_bufs[0..chunk_len];
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_read(
uv.Loop.get(),
&req,
uv_fd,
chunk_bufs.ptr,
@intCast(chunk_len),
current_position,
null,
);
const chunk_capacity = sumBufsLen(chunk_bufs);
if (Environment.isDebug) {
log("uv read({}, {d} total bytes) = {d} ({f})", .{ uv_fd, chunk_capacity, rc.int(), debug_timer });
}
if (rc.errno()) |errno| {
return .{ .err = .{ .errno = errno, .fd = fd, .syscall = .read } };
}
const bytes_read: usize = @intCast(rc.int());
total_read += bytes_read;
// If we read less than requested, we're done (EOF or partial read)
if (bytes_read == 0 or bytes_read < chunk_capacity) {
break;
}
remaining_bufs = remaining_bufs[chunk_len..];
// Update position for the next chunk (if position tracking is enabled)
if (current_position >= 0) {
current_position += @intCast(bytes_read);
}
}
return .{ .result = total_read };
}
pub fn pwritev(fd: FileDescriptor, bufs: []const bun.PlatformIOVecConst, position: i64) Maybe(usize) {
const uv_fd = fd.uv();
comptime bun.assert(bun.PlatformIOVec == uv.uv_buf_t);
const debug_timer = bun.Output.DebugTimer.start();
var total_written: usize = 0;
var remaining_bufs = bufs;
var current_position = position;
while (remaining_bufs.len > 0) {
const chunk_len = @min(remaining_bufs.len, max_iovec_count);
const chunk_bufs = remaining_bufs[0..chunk_len];
var req: uv.fs_t = uv.fs_t.uninitialized;
defer req.deinit();
const rc = uv.uv_fs_write(
uv.Loop.get(),
&req,
uv_fd,
chunk_bufs.ptr,
@intCast(chunk_len),
current_position,
null,
);
const chunk_capacity = sumBufsLen(chunk_bufs);
if (Environment.isDebug) {
log("uv write({}, {d} total bytes) = {d} ({f})", .{ uv_fd, chunk_capacity, rc.int(), debug_timer });
}
if (rc.errno()) |errno| {
return .{ .err = .{ .errno = errno, .fd = fd, .syscall = .write } };
}
const bytes_written: usize = @intCast(rc.int());
total_written += bytes_written;
// If we wrote less than requested, we're done (partial write)
if (bytes_written == 0 or bytes_written < chunk_capacity) {
break;
}
remaining_bufs = remaining_bufs[chunk_len..];
// Update position for the next chunk (if position tracking is enabled)
if (current_position >= 0) {
current_position += @intCast(bytes_written);
}
}
return .{ .result = total_written };
}
pub inline fn readv(fd: FileDescriptor, bufs: []bun.PlatformIOVec) Maybe(usize) {
return preadv(fd, bufs, -1);
}
pub fn pread(fd: FileDescriptor, buf: []u8, position: i64) Maybe(usize) {
// If buffer fits in a single uv_buf_t, use the simple path
if (buf.len <= max_buf_len) {
var bufs: [1]bun.PlatformIOVec = .{bun.platformIOVecCreate(buf)};
return preadv(fd, &bufs, position);
}
// Buffer is too large, need to chunk it
var total_read: usize = 0;
var remaining = buf;
var current_position = position;
while (remaining.len > 0) {
const chunk_len = @min(remaining.len, max_buf_len);
var bufs: [1]bun.PlatformIOVec = .{bun.platformIOVecCreate(remaining[0..chunk_len])};
switch (preadv(fd, &bufs, current_position)) {
.err => |err| return .{ .err = err },
.result => |bytes_read| {
total_read += bytes_read;
if (bytes_read == 0 or bytes_read < chunk_len) {
break;
}
remaining = remaining[chunk_len..];
if (current_position >= 0) {
current_position += @intCast(bytes_read);
}
},
}
}
return .{ .result = total_read };
}
pub fn read(fd: FileDescriptor, buf: []u8) Maybe(usize) {
// If buffer fits in a single uv_buf_t, use the simple path
if (buf.len <= max_buf_len) {
var bufs: [1]bun.PlatformIOVec = .{bun.platformIOVecCreate(buf)};
return readv(fd, &bufs);
}
// Buffer is too large, need to chunk it
var total_read: usize = 0;
var remaining = buf;
while (remaining.len > 0) {
const chunk_len = @min(remaining.len, max_buf_len);
var bufs: [1]bun.PlatformIOVec = .{bun.platformIOVecCreate(remaining[0..chunk_len])};
switch (readv(fd, &bufs)) {
.err => |err| return .{ .err = err },
.result => |bytes_read| {
total_read += bytes_read;
if (bytes_read == 0 or bytes_read < chunk_len) {
break;
}
remaining = remaining[chunk_len..];
},
}
}
return .{ .result = total_read };
}
pub inline fn writev(fd: FileDescriptor, bufs: []bun.PlatformIOVec) Maybe(usize) {
return pwritev(fd, bufs, -1);
}
pub fn pwrite(fd: FileDescriptor, buf: []const u8, position: i64) Maybe(usize) {
// If buffer fits in a single uv_buf_t, use the simple path
if (buf.len <= max_buf_len) {
var bufs: [1]bun.PlatformIOVecConst = .{bun.platformIOVecConstCreate(buf)};
return pwritev(fd, &bufs, position);
}
// Buffer is too large, need to chunk it
var total_written: usize = 0;
var remaining = buf;
var current_position = position;
while (remaining.len > 0) {
const chunk_len = @min(remaining.len, max_buf_len);
var bufs: [1]bun.PlatformIOVecConst = .{bun.platformIOVecConstCreate(remaining[0..chunk_len])};
switch (pwritev(fd, &bufs, current_position)) {
.err => |err| return .{ .err = err },
.result => |bytes_written| {
total_written += bytes_written;
if (bytes_written == 0 or bytes_written < chunk_len) {
break;
}
remaining = remaining[chunk_len..];
if (current_position >= 0) {
current_position += @intCast(bytes_written);
}
},
}
}
return .{ .result = total_written };
}
pub fn write(fd: FileDescriptor, buf: []const u8) Maybe(usize) {
// If buffer fits in a single uv_buf_t, use the simple path
if (buf.len <= max_buf_len) {
var bufs: [1]bun.PlatformIOVecConst = .{bun.platformIOVecConstCreate(buf)};
return writev(fd, &bufs);
}
// Buffer is too large, need to chunk it
var total_written: usize = 0;
var remaining = buf;
while (remaining.len > 0) {
const chunk_len = @min(remaining.len, max_buf_len);
var bufs: [1]bun.PlatformIOVecConst = .{bun.platformIOVecConstCreate(remaining[0..chunk_len])};
switch (writev(fd, &bufs)) {
.err => |err| return .{ .err = err },
.result => |bytes_written| {
total_written += bytes_written;
if (bytes_written == 0 or bytes_written < chunk_len) {
break;
}
remaining = remaining[chunk_len..];
},
}
}
return .{ .result = total_written };
}
pub const Tag = @import("./sys.zig").Tag;
const std = @import("std");
const bun = @import("bun");
const Environment = bun.Environment;
const FileDescriptor = bun.FileDescriptor;
const Maybe = bun.sys.Maybe;
const uv = bun.windows.libuv;