Swatantra-66 opened a new pull request, #21815:
URL: https://github.com/apache/nuttx/pull/21815

   ## Summary
   POSIX and Linux specifications (`epoll(7)`) explicitly define parameter 
validation for the `epoll` family of system calls:
   
   ### 1. `epoll_wait()` and `epoll_pwait()`
   - `[EINVAL]` `maxevents` is less than or equal to zero.
   - `[EFAULT]` The memory area pointed to by `events` is not accessible.
   
   **Previously:**
   - If `maxevents <= 0`, `epoll_teardown()` returned 0, causing `epoll_wait()` 
and `epoll_pwait()` to enter an infinite busy-wait loop (`goto retry;`), 
pegging the CPU at 100% and starving other tasks.
   - If `maxevents > 0` but `evs == NULL`, `epoll_teardown()` dereferenced NULL 
when storing results (CWE-476).
   
   **Fix:**
   - Validates `maxevents > 0` upfront returning `ERROR` with `errno = EINVAL`.
   - Validates `evs != NULL` upfront returning `ERROR` with `errno = EFAULT`.
   
   ### 2. `epoll_ctl()`
   - `[EINVAL]` `fd` is the same as `epfd`.
   - `[EFAULT]` The memory area pointed to by `event` is not accessible.
   
   **Previously:**
   - `epoll_ctl()` did not check for `epfd == fd`, allowing circular epoll 
self-monitoring dependencies.
   - For `EPOLL_CTL_ADD` and `EPOLL_CTL_MOD`, `ev->events` and `ev->data` were 
dereferenced without checking `ev != NULL`, causing kernel panics on NULL 
pointer dereference (CWE-476).
   
   **Fix:**
   - Rejects `epfd == fd` with `EINVAL`.
   - Validates `ev != NULL` for `EPOLL_CTL_ADD` and `EPOLL_CTL_MOD` returning 
`EFAULT`.
   
   ### 3. `epoll_create()` and `epoll_create1()`
   - `[EINVAL]` `size` is not positive.
   - `[EINVAL]` Invalid value specified in `flags`.
   
   **Previously:**
   - `epoll_create()` silently accepted non-positive sizes (`size <= 0 ? 1 : 
size`).
   - Unbounded `size` could overflow 32-bit integer multiplication in 
`sizeof(epoll_head_t) + sizeof(epoll_node_t) * size`, causing a small heap 
allocation followed by an out-of-bounds heap initialization loop 
(CWE-190/CWE-122).
   - `epoll_create1()` accepted arbitrary invalid flags without checking 
against `EPOLL_CLOEXEC`.
   
   **Fix:**
   - In public `epoll_create()`, rejects `size <= 0` or oversized values 
exceeding `(SIZE_MAX - sizeof(epoll_head_t)) / sizeof(epoll_node_t)` with 
`EINVAL`.
   - In `epoll_do_create()`, preserves normalization for configurations where 
`CONFIG_FS_NEPOLL_DESCRIPTORS=0` is used by `epoll_create1()`.
   - In `epoll_create1()`, validates `(flags & ~EPOLL_CLOEXEC) == 0`.
   - In `epoll_head_from_fd()`, defensively guards against `(*filep)->f_inode 
== NULL`.
   
   ---
   
   ## Impact
   - **Is new feature added?** No.
   - **Is existing feature changed?** Yes, `epoll` system calls now validate 
parameter bounds per POSIX/Linux.
   - **Impact on user:** None for valid applications. Invalid calls safely 
return standard error codes (`EINVAL` or `EFAULT`) instead of hanging or 
crashing.
   - **Impact on build:** None.
   - **Impact on hardware:** None.
   - **Impact on documentation:** None.
   - **Impact on security:** Hardens against CWE-476, CWE-190, CWE-122, and 
infinite loop denial-of-service.
   
   ---
   
   ## Testing
   
   ### Before vs After
   
   | Scenario | Before | After |
   | :--- | :--- | :--- |
   | `epoll_wait(epfd, evs, 0, 100)` | Infinite busy-loop (`goto retry;`), 100% 
CPU hang | Returns `ERROR`, `errno = EINVAL` |
   | `epoll_wait(epfd, NULL, 5, 0)` | Kernel NULL pointer dereference (CWE-476) 
| Returns `ERROR`, `errno = EFAULT` |
   | `epoll_ctl(epfd, EPOLL_CTL_ADD, epfd, &ev)` | Allows circular 
self-monitoring | Returns `ERROR`, `errno = EINVAL` |
   | `epoll_ctl(epfd, EPOLL_CTL_ADD, fd, NULL)` | Kernel NULL pointer 
dereference (CWE-476) | Returns `ERROR`, `errno = EFAULT` |
   | `epoll_create(0)` / `epoll_create(-1)` | Silently accepted, normalized to 
1 | Returns `ERROR`, `errno = EINVAL` |
   | `epoll_create(SIZE_MAX / 2)` | Integer overflow & heap memory corruption 
(CWE-190/CWE-122) | Returns `ERROR`, `errno = EINVAL` |
   | `epoll_create1(0xdeadbeef)` | Silently accepted | Returns `ERROR`, `errno 
= EINVAL` |
   | `epoll_create1(0)` with `CONFIG_FS_NEPOLL_DESCRIPTORS=0` | Normalizes to 1 
| Preserved, normalizes to 1 |
   
   ### Build & Verification Log
   
   - **Host:** Linux 6.6.137+ / Ubuntu 24.04 (Docker builder `nuttx-builder`)
   - **Target:** `sim:nsh` (x86_64)
   
   ```text
   $ /repos/nuttx/tools/nxstyle /repos/nuttx/fs/vfs/fs_epoll.c
   (0 errors, 0 warnings)
   
   $ cd /repos/nuttx/fs && make TOPDIR=/repos/nuttx fs_epoll.o
   CC:  vfs/fs_epoll.c
   
   $ make -j8
   IN:  fs/libfs.a -> staging/libfs.a
   LD:  nuttx
   Pac SIM with dynamic libs..
   SIM elf with dynamic libs archive in nuttx.tgz
   
   $ echo -e 'help\nexit' | ./nuttx
   nsh> help
   help usage:  help [-v] [<cmd>]
       .            break        exec         mkdir        rm           true    
     
       [            cat          exit         mkfifo       rmdir        
truncate     
       ?            cd           expr         mkrd         rmmod        uname   
     
       addroute     cp           false        mount        set          umount  
     
       alias        cmp          help         mv           kill         unset   
     
       unalias      dirname      hexdump      nslookup     sleep        uptime  
     
       arp          date         insmod       poweroff     usleep       watch   
     
       base64dec    delroute     losetup      quit         source       xd      
     
       base64enc    dmesg        ls           printf       test         
       basename     echo         du           pwd          time         
   
   Builtin Apps:
       dd      init    nsh     sh      
   nsh> exit
   ```
   
   Tested: `sim:nsh` (flat build, x86_64).
   Untested: kernel and protected builds, real hardware target boards.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to