)]}'
{
  "commit": "089d755192f09bc60eb9e7454bcb92ce3553f250",
  "tree": "73dc72c9ec7c10bbbe73738bf197fa6b37ef16f1",
  "parents": [
    "988078cdee9a70c9e55c6f6a71109bac145806ea"
  ],
  "author": {
    "name": "Shi Jin",
    "email": "sjina@amazon.com",
    "time": "Thu Nov 06 18:27:00 2025 +0000"
  },
  "committer": {
    "name": "Nicolas Morey",
    "email": "nmorey@suse.com",
    "time": "Fri Nov 21 13:24:48 2025 +0100"
  },
  "message": "providers/efa: Fix the size check in efadv_create_cq\n\n[ Upstream commit 696ed925ac4d48adf0cc97f7dd781dec6d762b8a ]\n\nefadv_create_cq incorrectly uses sizeof(efa_attr) instead\nof sizeof(*efa_attr) to calculate the size of the efa_attr struct,\nas a result, the currentt inlen\u003esizeof(efa_attr) will always be\ntrue as the latter will be 8 bytes always.\n\nThis bug breaks the backward compatibility when a library is built\nwith an older rdma-core that has smaller efa_attr struct, but\nrun with newer rdma-core that has larger efa_attr, as the later called\n`is_ext_clear` check is to make sure when application\ninput a inlen larger than size of efa_attr,\nthe extra space are all zero-ed, and it will fail when the inlen\nis smaller than the actual size of efa_attr.\nHowever when inlen is \u003c\u003d than size of the efa_attr struct,\nsuch is_ext_clear check shouldn\u0027t happen.\n\nFixes: a00b600 (\"efa: Introduce EFA DV CQ\")\nSigned-off-by: Shi Jin \u003csjina@amazon.com\u003e\nSigned-off-by: Nicolas Morey \u003cnmorey@suse.com\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "ddb1af989075078a99e9b978c6e0d9db449efa0c",
      "old_mode": 33188,
      "old_path": "providers/efa/verbs.c",
      "new_id": "2c82caf5ef3b98eae1cf70f9f60f20317485298c",
      "new_mode": 33188,
      "new_path": "providers/efa/verbs.c"
    }
  ]
}