fileio: lower maximum virtual file buffer size by one byte

When reading virtual files (i.e. procfs, sysfs, …) we currently put a
limit of 4M-1 on that. We have to pick something, and we have to read
these files in a single read() (since the kernel generally doesn't
support continuation read()s for them). 4M-1 is actually the maximum
size the kernel allows for reads from files in /proc/sys/, all larger
reads will result in an ENOMEM error (which is really weird, but the
kernel does what the kernel does). Hence 4M-1 sounds like a smart
choice.

However, we made one mistake here: in order to be able to detect EOFs
properly we actually read one byte more than we actually intend to
return: if that extra byte can be read, then we know the file is
actually larger than our limit and we can generate an EFBIG error from
that. However, if it cannot be read then we know EOF was hit, and we are
good. So ultimately after all we issued a single 4M read, which the
kernel then responds with ENOMEM to.  And that means read_virtual_file()
actually doesn't work properly right now on /proc/sys/. Let's fix that.

The fix is simple, lower the limit of the the buffer we intend to return
by one, i.e. 4M-2. That way, the read() we'll issue is exactly as large
as the limit the kernel allows, and we still get safely detect EOF from
it.

src/basic/fileio.c

@@ -30,14 +30,16 @@
 /* The maximum size of the file we'll read in one go in read_full_file() (64M). */
 #define READ_FULL_BYTES_MAX (64U*1024U*1024U - 1U)
 
-/* The maximum size of virtual files we'll read in one go in read_virtual_file() (4M). Note that this limit
- * is different (and much lower) than the READ_FULL_BYTES_MAX limit. This reflects the fact that we use
- * different strategies for reading virtual and regular files: virtual files are generally size constrained:
- * there we allocate the full buffer size in advance. Regular files OTOH can be much larger, and here we grow
- * the allocations exponentially in a loop. In glibc large allocations are immediately backed by mmap()
- * making them relatively slow (measurably so). Thus, when allocating the full buffer in advance the large
- * limit is a problem. When allocating piecemeal it's not. Hence pick two distinct limits. */
-#define READ_VIRTUAL_BYTES_MAX (4U*1024U*1024U - 1U)
+/* The maximum size of virtual files (i.e. procfs, sysfs, and other virtual "API" files) we'll read in one go
+ * in read_virtual_file(). Note that this limit is different (and much lower) than the READ_FULL_BYTES_MAX
+ * limit. This reflects the fact that we use different strategies for reading virtual and regular files:
+ * virtual files we generally have to read in a single read() syscall since the kernel doesn't support
+ * continuation read()s for them. Thankfully they are somewhat size constrained. Thus we can allocate the
+ * full potential buffer in advance. Regular files OTOH can be much larger, and there we grow the allocations
+ * exponentially in a loop. We use a size limit of 4M-2 because 4M-1 is the maximum buffer that /proc/sys/
+ * allows us to read() (larger reads will fail with ENOMEM), and we want to read one extra byte so that we
+ * can detect EOFs. */
+#define READ_VIRTUAL_BYTES_MAX (4U*1024U*1024U - 2U)
 
 int fopen_unlocked(const char *path, const char *options, FILE **ret) {
         assert(ret);