Ugly QOM property names: paths within paths

[Markus Armbruster]

I noticed this the other day:

    (qemu) info qom-tree 
    /machine (pc-i440fx-7.2-machine)
      /fw_cfg (fw_cfg_io)
        /\x2from@etc\x2facpi\x2frsdp[0] (memory-region)b
        /\x2from@etc\x2facpi\x2ftables[0] (memory-region)
        /\x2from@etc\x2ftable-loader[0] (memory-region)
        /fwcfg.dma[0] (memory-region)
        /fwcfg[0] (memory-region)
    [...]

It took me a minute to realize that the "\x2" in these property names
are escaped forms of '/'.  I.e. the unescaped path components of the
first property path are

    machine
    fw_cfg
    /from@etc/facpi/frsdp[0]

We're embedding paths within paths.  Ugh!

The escaping happens in memory_region_init():

    static bool memory_region_need_escape(char c)
    {
        return c == '/' || c == '[' || c == '\\' || c == ']';
    }

    static char *memory_region_escape_name(const char *name)
    {
        const char *p;
        char *escaped, *q;
        uint8_t c;
        size_t bytes = 0;

        for (p = name; *p; p++) {
            bytes += memory_region_need_escape(*p) ? 4 : 1;
        }
        if (bytes == p - name) {
           return g_memdup(name, bytes + 1);
        }

        escaped = g_malloc(bytes + 1);
        for (p = name, q = escaped; *p; p++) {
            c = *p;
            if (unlikely(memory_region_need_escape(c))) {
                *q++ = '\\';
                *q++ = 'x';
                *q++ = "0123456789abcdef"[c >> 4];
                c = "0123456789abcdef"[c & 15];
            }
            *q++ = c;
        }
        *q = 0;
        return escaped;
    }

    static void memory_region_do_init(MemoryRegion *mr,
                                      Object *owner,
                                      const char *name,
                                      uint64_t size)
    {
        mr->size = int128_make64(size);
        if (size == UINT64_MAX) {
            mr->size = int128_2_64();
        }
        mr->name = g_strdup(name);
        mr->owner = owner;
        mr->ram_block = NULL;

        if (name) {
            char *escaped_name = memory_region_escape_name(name);
            char *name_array = g_strdup_printf("%s[*]", escaped_name);

            if (!owner) {
                owner = container_get(qdev_get_machine(), "/unattached");
            }

            object_property_add_child(owner, name_array, OBJECT(mr));
            object_unref(OBJECT(mr));
            g_free(name_array);
            g_free(escaped_name);
        }
    }

    void memory_region_init(MemoryRegion *mr,
                            Object *owner,
                            const char *name,
                            uint64_t size)
    {
        object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
        memory_region_do_init(mr, owner, name, size);
    }

Goes back to

    commit b4fefef9d52003b6d09866501275a9a57995c6b0
    Author: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
    Date:   Thu Jun 5 23:15:52 2014 -0700

        memory: MemoryRegion: QOMify

        QOMify memory regions as an Object. The former init() and destroy()
        routines become instance_init() and instance_finalize() resp.

        memory_region_init() is re-implemented to be:
        object_initialize() + set fields

        memory_region_destroy() is re-implemented to call unparent().

        Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
        [Add newly-created MR as child, unparent on destruction. - Paolo]
        Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

No mention of the escapery.

Questions:

1. Do we really want to embed slash-separated paths into slash-separated
   paths?

2. As far as I can tell, object.c does not guard against "funny"
   characters such as '/' in path components.  Should it?  For what it's
   worth, the kernel doesn't permit '/' in filenames.

3. Should the escapery live in object.c instead of memory.c?