Merge branch '3.1' into 3.1

gns3server/utils/images.py

@@ -49,6 +49,9 @@ from . import force_unix_path
 
 log = logging.getLogger(__name__)
 
+# Include the first ISO 9660 volume descriptor in image type detection.
+IMAGE_HEADER_SIZE = 16 * 2048 + 7
+
 
 async def list_images(image_type):
     """
@@ -145,7 +148,7 @@ def inspect_image_file(path, expected_image_type=None, allow_raw_image=False, st
         info = os.fstat(f.fileno())
         if not stat.S_ISREG(info.st_mode) or stat_fingerprint(info) != before:
             raise ImageChangedError(f"Image changed while opening: {path}")
-        header = f.read(7)
+        header = f.read(IMAGE_HEADER_SIZE)
         if len(header) < 7:
             raise InvalidImageError(f"Image '{path}' is too small to be valid")
         image_type = check_valid_image_header(path, header, allow_raw_image)
@@ -359,6 +362,8 @@ def check_valid_image_header(path: str, data: bytes, allow_raw_image: bool = Fal
     elif data[:4] == b"QFI\xfb" or data[:4] == b"KDMV":
         # for Qemy images: file must be QCOW2 or VMDK
         return "qemu"
+    elif data[16 * 2048 + 1 : 16 * 2048 + 6] == b"CD001":
+        return "qemu"
     else:
         if allow_raw_image is True:
             return "qemu"
@@ -380,12 +385,14 @@ async def write_image(
     if subdirectory is not None:
         validate_image_upload_name(image_filename)
         subfolders = validate_image_subdirectory(subdirectory)
-    # HTTP chunk boundaries need not align with the seven-byte image header.
+    # HTTP chunk boundaries need not align with the image header or ISO descriptor.
     iterator = stream.__aiter__()
     prefix = bytearray()
     async for chunk in iterator:
         prefix.extend(chunk)
-        if len(prefix) >= 7:
+        # Known disk formats keep their original early detection behavior.
+        header_size = 7 if prefix.startswith((b"\x7fELF", b"QFI\xfb", b"KDMV")) else IMAGE_HEADER_SIZE
+        if len(prefix) >= header_size:
             break
     if len(prefix) < 7:
         raise InvalidImageError("The image content is empty or too small to be valid")

gns3server/utils/vxlan.py

@@ -0,0 +1,86 @@
+#
+# Copyright (C) 2026 GNS3 Technologies Inc.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+"""
+Identity of a cross-compute VXLAN link: the device name and VNI the
+controller derives and ships in both endpoints' NIOs.
+
+A cross-compute kernel link is carried by one VXLAN device per endpoint
+(uBridge's ``vxlan`` module, rtnetlink), enslaved into the link's per-link
+kernel bridge next to the node's anchor. Each device is local to its own
+host, but the VNI is the kernel's demux key for received frames on the
+deployment-wide shared dstport socket — both ends must agree on it while
+neither can ask the other, so the controller is the single source for both
+values.
+
+* the device name is a pure function of the link id plus the endpoint's
+  node id: unique per (link, endpoint), within IFNAMSIZ, and outside the
+  ``gns3``/anchor name spaces (``gvl`` prefix). Determinism is what makes
+  a crashed run's leftover device recognisable — the compute deletes a
+  device it finds under the name before re-creating it;
+* the VNI is a pure function of the link id and a controller-side salt —
+  a random base per link object, bumped when a retry needs a fresh VNI —
+  drawn from the upper half of the VNI space so it cannot collide with the
+  low well-known VNIs other software uses (Flannel's 1, Docker overlay
+  networks from 256, ...).
+
+The kernel refuses a second device registering a VNI already taken on the
+same (dstport, socket) pair with EEXIST — a check that doubles as the
+collision detector: the controller retries with a bumped salt and falls
+back to the UDP relay when the space stays occupied (the relay needs no
+VNI at all).
+"""
+
+import hashlib
+
+# The VNI range derived links draw from: at or above 0x100000 (1M) to stay
+# clear of the low VNIs other VXLAN software commonly registers, at or below
+# 0xFFFFFE so every derived value is valid for the module's 1..16777215
+# (0xFFFFFF) check with one value of headroom.
+VXLAN_VNI_MIN = 0x100000
+VXLAN_VNI_MAX = 0xFFFFFE
+
+# IFNAMSIZ (16) - 1 for the terminator: "gvl" (3) + 8 (link) + 4 (node) = 15.
+VXLAN_NAME_PREFIX = "gvl"
+
+
+def derive_vni(link_id, salt=0):
+    """
+    The VNI both endpoints of *link_id* tunnel through.
+
+    :param link_id: link UUID (or str)
+    :param salt: controller-side retry counter — bumped when the kernel
+        refuses the previous VNI as already registered on the shared
+        dstport socket; 0 for the first attempt
+    :returns: VNI in [VXLAN_VNI_MIN, VXLAN_VNI_MAX]
+    """
+
+    digest = hashlib.sha256(f"{link_id}:{salt}".encode()).digest()
+    span = VXLAN_VNI_MAX - VXLAN_VNI_MIN + 1
+    return VXLAN_VNI_MIN + int.from_bytes(digest[:8], "big") % span
+
+
+def vxlan_device_name(link_id, node_id):
+    """
+    The VXLAN device name one endpoint of *link_id* creates on its own host,
+    deterministic from the link id plus the endpoint's node id (the pair
+    normally lives on two hosts, but a forced same-host pair must not name
+    both its devices identically).
+    """
+
+    link = str(link_id).replace("-", "")[:8]
+    node = str(node_id).replace("-", "")[:4]
+    return f"{VXLAN_NAME_PREFIX}{link}{node}"