client: crash caused by invalid iterator in _readdir_cache_cb

Capacity of `readdir_cache` may change after `client_lock` is unlocked in iterations of `readdir_cache`,
and it can cause the iterator to be invalid, then using the invalid iterator in the next iteration will cause crash.
Crash may happen at `Dentry *dn = *pd` (pd points to invalid memory),
or at `if (pd >= dir->readdir_cache.end() || *pd != dn)` (pd is smaller than begin() if idx is negative).
Use index instead of iterator to solve this problem.

Fixes: https://tracker.ceph.com/issues/72247
Signed-off-by: Zhansong Gao <[email protected]>

Author: zhsgao

src/client/Client.cc

@@ -9461,25 +9461,22 @@ int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
 						  dirp->offset, dentry_off_lt());
 
   string dn_name;
-  while (true) {
+  for (unsigned idx = pd - dir->readdir_cache.begin();
+       idx < dir->readdir_cache.size();
+       ++idx) {
     int mask = caps;
     if (!dirp->inode->is_complete_and_ordered())
       return -EAGAIN;
-    if (pd == dir->readdir_cache.end())
-      break;
-    Dentry *dn = *pd;
+    Dentry *dn = dir->readdir_cache[idx];
     if (dn->inode == NULL) {
       ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl;
-      ++pd;
       continue;
     }
     if (dn->cap_shared_gen != dir->parent_inode->shared_gen) {
       ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl;
-      ++pd;
       continue;
     }
 
-    int idx = pd - dir->readdir_cache.begin();
     if (dn->inode->is_dir() && cct->_conf->client_dirsize_rbytes) {
       mask |= CEPH_STAT_RSTAT;
     }
@@ -9493,9 +9490,8 @@ int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
       return -EAGAIN;
     }
 
-    // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator    
-    pd = dir->readdir_cache.begin() + idx;
-    if (pd >= dir->readdir_cache.end() || *pd != dn)
+    // the content of readdir_cache may change after _getattr()
+    if (idx >= dir->readdir_cache.size() || dir->readdir_cache[idx] != dn)
       return -EAGAIN;
 
     struct ceph_statx stx;
@@ -9505,8 +9501,7 @@ int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
     uint64_t next_off = dn->offset + 1;
     auto dname = _unwrap_name(*diri, dn->name, dn->alternate_name);
     fill_dirent(&de, dname.c_str(), stx.stx_mode, stx.stx_ino, next_off);
-    ++pd;
-    if (pd == dir->readdir_cache.end())
+    if (idx + 1 == dir->readdir_cache.size())
       next_off = dir_result_t::END;
 
     Inode *in = NULL;
@@ -9517,6 +9512,7 @@ int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p,
 
     dn_name = dn->name; // fill in name while we have lock
 
+    // the content of readdir_cache may change after unlocking
     client_lock.unlock();
     r = cb(p, &de, &stx, next_off, in);  // _next_ offset
     client_lock.lock();