Fix a small number of invalid unsafe code

Fix #23520

Author: yuyichao

base/array.jl

@@ -171,6 +171,8 @@ that N is inbounds on either array. Incorrect usage may corrupt or segfault your
 the same manner as C.
 """
 function unsafe_copy!(dest::Array{T}, doffs, src::Array{T}, soffs, n) where T
+    t1 = @_gc_preserve_begin dest
+    t2 = @_gc_preserve_begin src
     if isbits(T)
         unsafe_copy!(pointer(dest, doffs), pointer(src, soffs), n)
     elseif isbitsunion(T)
@@ -185,6 +187,8 @@ function unsafe_copy!(dest::Array{T}, doffs, src::Array{T}, soffs, n) where T
         ccall(:jl_array_ptr_copy, Void, (Any, Ptr{Void}, Any, Ptr{Void}, Int),
               dest, pointer(dest, doffs), src, pointer(src, soffs), n)
     end
+    @_gc_preserve_end t2
+    @_gc_preserve_end t1
     return dest
 end
 
@@ -1570,6 +1574,7 @@ function vcat(arrays::Vector{T}...) where T
     else
         elsz = Core.sizeof(Ptr{Void})
     end
+    t = @_gc_preserve_begin arr
     for a in arrays
         na = length(a)
         nba = na * elsz
@@ -1589,6 +1594,7 @@ function vcat(arrays::Vector{T}...) where T
         end
         ptr += nba
     end
+    @_gc_preserve_end t
     return arr
 end
 

base/datafmt.jl

@@ -130,7 +130,7 @@ function readdlm_auto(input::AbstractString, dlm::Char, T::Type, eol::Char, auto
         # TODO: It would be nicer to use String(a) without making a copy,
         # but because the mmap'ed array is not NUL-terminated this causes
         # jl_try_substrtod to segfault below.
-        return readdlm_string(unsafe_string(pointer(a),length(a)), dlm, T, eol, auto, optsd)
+        return readdlm_string(Base.@gc_preserve(a, unsafe_string(pointer(a),length(a))), dlm, T, eol, auto, optsd)
     else
         return readdlm_string(read(input, String), dlm, T, eol, auto, optsd)
     end
@@ -220,7 +220,7 @@ function DLMStore(::Type{T}, dims::NTuple{2,Integer},
 end
 
 _chrinstr(sbuff::String, chr::UInt8, startpos::Int, endpos::Int) =
-    (endpos >= startpos) && (C_NULL != ccall(:memchr, Ptr{UInt8},
+    Base.@gc_preserve sbuff (endpos >= startpos) && (C_NULL != ccall(:memchr, Ptr{UInt8},
     (Ptr{UInt8}, Int32, Csize_t), pointer(sbuff)+startpos-1, chr, endpos-startpos+1))
 
 function store_cell(dlmstore::DLMStore{T}, row::Int, col::Int,

base/deepcopy.jl

@@ -46,7 +46,7 @@ function deepcopy_internal(x::String, stackdict::ObjectIdDict)
     if haskey(stackdict, x)
         return stackdict[x]
     end
-    y = unsafe_string(pointer(x), sizeof(x))
+    y = @gc_preserve x unsafe_string(pointer(x), sizeof(x))
     stackdict[x] = y
     return y
 end

base/env.jl

@@ -99,7 +99,7 @@ if Sys.iswindows()
         blk = block[2]
         len = ccall(:wcslen, UInt, (Ptr{UInt16},), pos)
         buf = Vector{UInt16}(len)
-        unsafe_copy!(pointer(buf), pos, len)
+        @gc_preserve buf unsafe_copy!(pointer(buf), pos, len)
         env = transcode(String, buf)
         m = match(r"^(=?[^=]+)=(.*)$"s, env)
         if m === nothing

base/essentials.jl

@@ -18,6 +18,14 @@ macro _noinline_meta()
     Expr(:meta, :noinline)
 end
 
+macro _gc_preserve_begin(arg1)
+    Expr(:gc_preserve_begin, esc(arg1))
+end
+
+macro _gc_preserve_end(token)
+    Expr(:gc_preserve_end, esc(token))
+end
+
 """
     @nospecialize
 
@@ -513,16 +521,23 @@ end
 # SimpleVector
 
 function getindex(v::SimpleVector, i::Int)
-    if !(1 <= i <= length(v))
+    @boundscheck if !(1 <= i <= length(v))
         throw(BoundsError(v,i))
     end
+    t = @_gc_preserve_begin v
     x = unsafe_load(convert(Ptr{Ptr{Void}},data_pointer_from_objref(v)) + i*sizeof(Ptr))
     x == C_NULL && throw(UndefRefError())
-    return unsafe_pointer_to_objref(x)
+    o = unsafe_pointer_to_objref(x)
+    @_gc_preserve_end t
+    return o
 end
 
-# TODO: add gc use intrinsic call instead of noinline
-length(v::SimpleVector) = (@_noinline_meta; unsafe_load(convert(Ptr{Int},data_pointer_from_objref(v))))
+function length(v::SimpleVector)
+    t = @_gc_preserve_begin v
+    l = unsafe_load(convert(Ptr{Int},data_pointer_from_objref(v)))
+    @_gc_preserve_end t
+    return l
+end
 endof(v::SimpleVector) = length(v)
 start(v::SimpleVector) = 1
 next(v::SimpleVector,i) = (v[i],i+1)
@@ -573,7 +588,9 @@ function isassigned end
 
 function isassigned(v::SimpleVector, i::Int)
     @boundscheck 1 <= i <= length(v) || return false
+    t = @_gc_preserve_begin v
     x = unsafe_load(convert(Ptr{Ptr{Void}},data_pointer_from_objref(v)) + i*sizeof(Ptr))
+    @_gc_preserve_end t
     return x != C_NULL
 end
 

base/gmp.jl

@@ -252,7 +252,7 @@ function tryparse_internal(::Type{BigInt}, s::AbstractString, startpos::Int, end
     if Base.containsnul(bstr)
         err = -1 # embedded NUL char (not handled correctly by GMP)
     else
-        err = MPZ.set_str!(z, pointer(bstr)+(i-start(bstr)), base)
+        err = Base.@gc_preserve bstr MPZ.set_str!(z, pointer(bstr)+(i-start(bstr)), base)
     end
     if err != 0
         raise && throw(ArgumentError("invalid BigInt: $(repr(bstr))"))
@@ -612,7 +612,7 @@ function base(b::Integer, n::BigInt, pad::Integer=1)
     nd1 = ndigits(n, b)
     nd  = max(nd1, pad)
     sv  = Base.StringVector(nd + isneg(n))
-    MPZ.get_str!(pointer(sv) + nd - nd1, b, n)
+    Base.@gc_preserve sv MPZ.get_str!(pointer(sv) + nd - nd1, b, n)
     @inbounds for i = (1:nd-nd1) .+ isneg(n)
         sv[i] = '0' % UInt8
     end

base/io.jl

@@ -365,9 +365,9 @@ function write(s::IO, A::AbstractArray)
     return nb
 end
 
-@noinline function write(s::IO, a::Array)  # mark noinline to ensure the array is gc-rooted somewhere (by the caller)
+function write(s::IO, a::Array)
     if isbits(eltype(a))
-        return unsafe_write(s, pointer(a), sizeof(a))
+        return @gc_preserve a unsafe_write(s, pointer(a), sizeof(a))
     else
         depwarn("Calling `write` on non-isbits arrays is deprecated. Use a loop or `serialize` instead.", :write)
         nb = 0
@@ -384,7 +384,7 @@ function write(s::IO, a::SubArray{T,N,<:Array}) where {T,N}
     end
     elsz = sizeof(T)
     colsz = size(a,1) * elsz
-    if stride(a,1) != 1
+    @gc_preserve a if stride(a,1) != 1
         for idxs in CartesianRange(size(a))
             unsafe_write(s, pointer(a, idxs.I), elsz)
         end
@@ -444,14 +444,14 @@ end
 read(s::IO, ::Type{Bool}) = (read(s, UInt8) != 0)
 read(s::IO, ::Type{Ptr{T}}) where {T} = convert(Ptr{T}, read(s, UInt))
 
-@noinline function read!(s::IO, a::Array{UInt8}) # mark noinline to ensure the array is gc-rooted somewhere (by the caller)
-    unsafe_read(s, pointer(a), sizeof(a))
+function read!(s::IO, a::Array{UInt8})
+    @gc_preserve a unsafe_read(s, pointer(a), sizeof(a))
     return a
 end
 
-@noinline function read!(s::IO, a::Array{T}) where T # mark noinline to ensure the array is gc-rooted somewhere (by the caller)
+function read!(s::IO, a::Array{T}) where T
     if isbits(T)
-        unsafe_read(s, pointer(a), sizeof(a))
+        @gc_preserve a unsafe_read(s, pointer(a), sizeof(a))
     else
         for i in eachindex(a)
             a[i] = read(s, T)

base/iobuffer.jl

@@ -99,7 +99,7 @@ function unsafe_read(from::GenericIOBuffer, p::Ptr{UInt8}, nb::UInt)
     from.readable || throw(ArgumentError("read failed, IOBuffer is not readable"))
     avail = nb_available(from)
     adv = min(avail, nb)
-    unsafe_copy!(p, pointer(from.data, from.ptr), adv)
+    @gc_preserve from unsafe_copy!(p, pointer(from.data, from.ptr), adv)
     from.ptr += adv
     if nb > avail
         throw(EOFError())
@@ -114,7 +114,7 @@ function read_sub(from::GenericIOBuffer, a::AbstractArray{T}, offs, nel) where T
     end
     if isbits(T) && isa(a,Array)
         nb = UInt(nel * sizeof(T))
-        unsafe_read(from, pointer(a, offs), nb)
+        @gc_preserve a unsafe_read(from, pointer(a, offs), nb)
     else
         for i = offs:offs+nel-1
             a[i] = read(to, T)
@@ -334,7 +334,7 @@ function write_sub(to::GenericIOBuffer, a::AbstractArray{UInt8}, offs, nel)
     if offs+nel-1 > length(a) || offs < 1 || nel < 0
         throw(BoundsError())
     end
-    unsafe_write(to, pointer(a, offs), UInt(nel))
+    @gc_preserve a unsafe_write(to, pointer(a, offs), UInt(nel))
 end
 
 @inline function write(to::GenericIOBuffer, a::UInt8)
@@ -367,7 +367,7 @@ read(io::GenericIOBuffer, nb::Integer) = read!(io,StringVector(min(nb, nb_availa
 
 function search(buf::IOBuffer, delim::UInt8)
     p = pointer(buf.data, buf.ptr)
-    q = ccall(:memchr,Ptr{UInt8},(Ptr{UInt8},Int32,Csize_t),p,delim,nb_available(buf))
+    q = @gc_preserve buf ccall(:memchr,Ptr{UInt8},(Ptr{UInt8},Int32,Csize_t),p,delim,nb_available(buf))
     nb::Int = (q == C_NULL ? 0 : q-p+1)
     return nb
 end
@@ -413,7 +413,7 @@ function crc32c(io::IOBuffer, nb::Integer, crc::UInt32=0x00000000)
     io.readable || throw(ArgumentError("read failed, IOBuffer is not readable"))
     n = min(nb, nb_available(io))
     n == 0 && return crc
-    crc = unsafe_crc32c(pointer(io.data, io.ptr), n, crc)
+    crc = @gc_preserve io unsafe_crc32c(pointer(io.data, io.ptr), n, crc)
     io.ptr += n
     return crc
 end

base/iostream.jl

@@ -264,7 +264,7 @@ end
 function readbytes_all!(s::IOStream, b::Array{UInt8}, nb)
     olb = lb = length(b)
     nr = 0
-    while nr < nb
+    @gc_preserve b while nr < nb
         if lb < nr+1
             lb = max(65536, (nr+1) * 2)
             resize!(b, lb)
@@ -284,8 +284,8 @@ function readbytes_some!(s::IOStream, b::Array{UInt8}, nb)
     if nb > lb
         resize!(b, nb)
     end
-    nr = Int(ccall(:ios_read, Csize_t, (Ptr{Void}, Ptr{Void}, Csize_t),
-                   s.ios, pointer(b), nb))
+    nr = @gc_preserve b Int(ccall(:ios_read, Csize_t, (Ptr{Void}, Ptr{Void}, Csize_t),
+                                  s.ios, pointer(b), nb))
     if lb > olb && lb > nr
         resize!(b, nr)
     end

base/libc.jl

@@ -249,7 +249,7 @@ function gethostname()
         ccall(:gethostname, Int32, (Ptr{UInt8}, UInt), hn, length(hn))
     end
     systemerror("gethostname", err != 0)
-    return unsafe_string(pointer(hn))
+    return Base.@gc_preserve hn unsafe_string(pointer(hn))
 end
 
 ## system error handling ##
@@ -305,7 +305,7 @@ if Sys.iswindows()
         p = lpMsgBuf[]
         len == 0 && return ""
         buf = Vector{UInt16}(len)
-        unsafe_copy!(pointer(buf), p, len)
+        Base.@gc_preserve buf unsafe_copy!(pointer(buf), p, len)
         ccall(:LocalFree, stdcall, Ptr{Void}, (Ptr{Void},), p)
         return transcode(String, buf)
     end