--[ HNS-2024-05 - HN Security Advisory - https://security.humanativaspa.it/

* Title: Multiple vulnerabilities in RT-Thread RTOS
* OS: RT-Thread <= 5.0.2
* Author: Marco Ivaldi <marco.ivaldi@hnsecurity.it>
* Date: 2024-03-05
* CVE IDs and advisory URLs:
  * CVE-2024-24334 - https://github.com/RT-Thread/rt-thread/issues/8282
  * CVE-2024-24335 - https://github.com/RT-Thread/rt-thread/issues/8271
  * CVE-2024-25388 - https://github.com/RT-Thread/rt-thread/issues/8285
  * CVE-2024-25389 - https://github.com/RT-Thread/rt-thread/issues/8283
  * CVE-2024-25390 - https://github.com/RT-Thread/rt-thread/issues/8286
  * CVE-2024-25391 - https://github.com/RT-Thread/rt-thread/issues/8287
  * CVE-2024-25392 - https://github.com/RT-Thread/rt-thread/issues/8290
  * CVE-2024-25393 - https://github.com/RT-Thread/rt-thread/issues/8288
  * CVE-2024-25394 - https://github.com/RT-Thread/rt-thread/issues/8291
  * CVE-2024-25395 - https://github.com/RT-Thread/rt-thread/issues/8289
  * https://github.com/RT-Thread/rt-thread/issues/8292
* Vendor URL: https://www.rt-thread.io/


--[ 0 - Table of contents

1 - Summary
2 - Background
3 - Vulnerabilities
    3.1 - CVE-2024-24335 - Buffer overflow in RT-Thread dfs_v2 romfs filesystem
    3.2 - CVE-2024-24334 - Heap buffer overflows in RT-Thread dfs_v2 dfs_file
    3.3 - CVE-2024-25389 - Weak random source in RT-Thread rt_random driver
    3.4 - CVE-2024-25388 - Heap buffer overflow in RT-Thread wlan driver
    3.5 - CVE-2024-25390 - Heap buffer overflows in RT-Thread finsh
    3.6 - CVE-2024-25391 - Stack buffer overflow in RT-Thread IPC
    3.7 - CVE-2024-25393 - Stack buffer overflow in RT-Thread AT server
    3.8 - CVE-2024-25395 - Static buffer overflow in RT-Thread rt-link utility
    3.9 - CVE-2024-25392 - Out-of-bounds static array access in RT-Thread var_export utility
    3.10 - CVE-2024-25394 - Multiple vulnerabilities in RT-Thread ymodem utility
    3.11 - Use of outdated lwIP and TinyDir dependencies in RT-Thread
4 - Affected products
5 - Remediation
6 - Disclosure timeline
7 - Acknowledgments
8 - References


--[ 1 - Summary

"Security is in the mind of the programmer and in the mind of the designer.
Not so much in the code." 
                                                             -- Alisa Esage

RT-Thread [1] is an open-source, community-based real-time operating system
(RTOS). RT-Thread can be used in sensing nodes, wireless connection chips,
and many other resource-constrained scenarios. It is also widely applied in
gateways, IPC, smart speakers, and other high-performance IoT applications.

We reviewed RT-Thread's source code hosted on GitHub [2] and identified
multiple security vulnerabilities that may cause memory corruption and
security feature bypass. Their impacts range from denial of service to
potential arbitrary code execution.

We also audited the lwIP [3] and TinyDir [4] codebases on which some
RT-Thread functionalities depend, and found some additional vulnerabilities
that were subsequently fixed by the respective maintainers.


--[ 2 - Background

After our recent vulnerability disclosures [5] in the IoT space, we decided
to keep assisting open-source projects in finding and fixing security
vulnerabilities by reviewing their source code. RT-Thread was selected as a
target of interest. Other RTOSes will be featured in future advisories and
writeups.

During this review, we made use of our Semgrep C/C++ ruleset [6] to
identify hotspots in code on which to focus our attention. We also took
advantage of this opportunity to improve and update the ruleset [7].


--[ 3 - Vulnerabilities

The vulnerabilities resulting from our source code review are briefly
described in the following sections.


--[ 3.1 - CVE-2024-24335 - Buffer overflow in RT-Thread dfs_v2 romfs filesystem

We spotted a buffer overflow vulnerability at the following location in the
RT-Thread dfs_v2 romfs filesystem source code:

* /components/dfs/dfs_v2/filesystems/romfs/dfs_romfs.c

Lack of length check in the `dfs_romfs_getdents()` function could lead to a
buffer overflow at the marked line:
```c
static int dfs_romfs_getdents(struct dfs_file *file, struct dirent *dirp, uint32_t count)
{
    rt_size_t index;
    const char *name;
    struct dirent *d;
    struct romfs_dirent *dirent, *sub_dirent;

    dirent = (struct romfs_dirent *)file->vnode->data;
    if (check_dirent(dirent) != 0)
    {
        return -EIO;
    }
    RT_ASSERT(dirent->type == ROMFS_DIRENT_DIR);

    /* enter directory */
    dirent = (struct romfs_dirent *)dirent->data;

    /* make integer count */
    count = (count / sizeof(struct dirent));
    if (count == 0)
    {
        return -EINVAL;
    }

    index = 0;
    for (index = 0; index < count && file->fpos < file->vnode->size; index++)
    {
        d = dirp + index;

        sub_dirent = &dirent[file->fpos];
        name = sub_dirent->name;

        /* fill dirent */
        if (sub_dirent->type == ROMFS_DIRENT_DIR)
            d->d_type = DT_DIR;
        else
            d->d_type = DT_REG;

        d->d_namlen = rt_strlen(name);
        d->d_reclen = (rt_uint16_t)sizeof(struct dirent);
        rt_strncpy(d->d_name, name, rt_strlen(name) + 1); /* VULN: buffer overflow if rt_strlen(name) is larger than sizeof(d->d_name) due to missing length check */

        /* move to next position */
        ++ file->fpos;
    }

    return index * sizeof(struct dirent);
}
```

Note: dfs_v1 romfs in /components/dfs/dfs_v1/filesystems/romfs/dfs_romfs.c
is not affected, because the string copy operation is implemented
differently:
```c
        len = rt_strlen(name);
        RT_ASSERT(len <= RT_UINT8_MAX);
        d->d_namlen = (rt_uint8_t)len;
        d->d_reclen = (rt_uint16_t)sizeof(struct dirent);
        rt_strncpy(d->d_name, name, DFS_PATH_MAX);
```

Even if the assertion is compiled-out in production code, `len` is not used
for the copy operation anyway. Therefore, unless `DFS_PATH_MAX` is larger
than `sizeof(d->d_name)`, this code should be safe.

Fixes:
https://github.com/RT-Thread/rt-thread/pull/8278

See also:
https://github.com/RT-Thread/rt-thread/issues/8271


--[ 3.2 - CVE-2024-24334 - Heap buffer overflows in RT-Thread dfs_v2 dfs_file

We spotted some heap buffer overflow vulnerabilities at the following
location in the RT-Thread dfs_v2 dfs_file source code:

* /components/dfs/dfs_v2/src/dfs_file.c

Lack of length check in the the `dfs_nolink_path()` function could lead to
heap buffer overflows at the marked lines:
```c
static char *dfs_nolink_path(struct dfs_mnt **mnt, char *fullpath, int mode)
{
    int index = 0;
    char *path = RT_NULL;
    char link_fn[DFS_PATH_MAX] = {0};
    struct dfs_dentry *dentry = RT_NULL;

    path = (char *)rt_malloc((DFS_PATH_MAX * 2) + 1); // path + syslink + \0
    if (!path)
    {
        return path;
    }

    if (*mnt && fullpath)
    {
        int i = 0;
        char *fp = fullpath;

        while (*fp != '\0')
        {
            fp++;
            i++;
            if (*fp == '/')
            {
                rt_memcpy(path + index, fp - i, i); /* VULN: if fullpath has components large enough so that i+index becomes larger than DFS_PATH_MAX*2+1, we could overflow past the path buffer */
                path[index + i] = '\0';

                dentry = dfs_dentry_lookup(*mnt, path, 0);
                if (dentry && dentry->vnode->type == FT_SYMLINK)
                {
                    int ret = -1;

                    if ((*mnt)->fs_ops->readlink)
                    {
                        if (dfs_is_mounted((*mnt)) == 0)
                        {
                            ret = (*mnt)->fs_ops->readlink(dentry, link_fn, DFS_PATH_MAX);
                        }
                    }

                    if (ret > 0)
                    {
                        int len = rt_strlen(link_fn);
                        if (link_fn[0] == '/')
                        {
                            rt_memcpy(path, link_fn, len);
                            index = len;
                        }
                        else
                        {
                            path[index] = '/';
                            index++;
                            rt_memcpy(path + index, link_fn, len); /* VULN: len can be DFS_PATH_MAX; if index can become larger than DFS_PATH_MAX+1, we can overflow past the path buffer */
                            index += len;
                        }
                        path[index] = '\0';
                        *mnt = dfs_mnt_lookup(path);
                    }
                    else
                    {
                        rt_kprintf("link error: %s\n", path);
                    }
                }
                else
                {
                    index += i;
                }
                dfs_dentry_unref(dentry);
                i = 0;
            }
        }

        if (i)
        {
            rt_memcpy(path + index, fp - i, i); /* VULN: if fullpath has components large enough so that i+index becomes larger than DFS_PATH_MAX*2+1, we could overflow past the path buffer */
            path[index + i] = '\0';

            if (mode)
            {
                dentry = dfs_dentry_lookup(*mnt, path, 0);
                if (dentry && dentry->vnode->type == FT_SYMLINK)
                {
                    int ret = -1;

                    if ((*mnt)->fs_ops->readlink)
                    {
                        if (dfs_is_mounted((*mnt)) == 0)
                        {
                            ret = (*mnt)->fs_ops->readlink(dentry, link_fn, DFS_PATH_MAX);
                        }
                    }

                    if (ret > 0)
                    {
                        int len = rt_strlen(link_fn);
                        if (link_fn[0] == '/')
                        {
                            rt_memcpy(path, link_fn, len);
                            index = len;
                        }
                        else
                        {
                            path[index] = '/';
                            index++;
                            rt_memcpy(path + index, link_fn, len); /* VULN: len can be DFS_PATH_MAX; if index can become larger than DFS_PATH_MAX+1, we can overflow past the path buffer */
                            index += len;
                        }
                        path[index] = '\0';
                        *mnt = dfs_mnt_lookup(path);
                    }
                    else
                    {
                        rt_kprintf("link error: %s\n", path);
                    }

                    char *_fullpath = dfs_normalize_path(RT_NULL, path);
                    if (_fullpath)
                    {
                        strncpy(path, _fullpath, DFS_PATH_MAX);
                        rt_free(_fullpath);
                    }
                }
                dfs_dentry_unref(dentry);
            }
        }
    }
    else
    {
        rt_free(path);
        path = RT_NULL;
    }

    //rt_kprintf("%s: %s => %s\n", __FUNCTION__, fullpath, path);

    return path;
}
```

Fixes:
https://github.com/RT-Thread/rt-thread/pull/8305

See also:
https://github.com/RT-Thread/rt-thread/issues/8282


--[ 3.3 - CVE-2024-25389 - Weak random source in RT-Thread rt_random driver

We noticed a weak random source at the following location in the RT-Thread
rt_random driver source code:

* /components/drivers/misc/rt_random.c

Weak random generator in the the `calc_random()` function:
```c
static rt_uint16_t calc_random(void)
{
    seed = 214013L * seed + 2531011L;
    return (seed >> 16) & 0x7FFF;   /* return bits 16~30 */
}

static rt_ssize_t random_read(rt_device_t dev, rt_off_t pos, void *buffer, rt_size_t size)
{
    rt_uint16_t rand = calc_random(); /* VULN: weak random source */
    ssize_t ret = sizeof(rand);
    rt_memcpy(buffer, &rand, ret);
    return ret;
}
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8283


--[ 3.4 - CVE-2024-25388 - Heap buffer overflow in RT-Thread wlan driver

We spotted a heap buffer overflow vulnerability at the following location
in the RT-Thread wlan driver source code:

* /components/drivers/wlan/wlan_mgnt.c

Since `len` is a signed integer in the `rt_wlan_send_to_thread()` function,
a small negative value could lead to a buffer overflow at the marked lines:
```c
#ifdef RT_WLAN_WORK_THREAD_ENABLE
...
static rt_err_t rt_wlan_send_to_thread(rt_wlan_event_t event, void *buff, int len)
{
    struct rt_wlan_msg *msg;

    RT_WLAN_LOG_D("F:%s is run event:%d", __FUNCTION__, event);

    /* Event packing */
    msg = rt_malloc(sizeof(struct rt_wlan_msg) + len); /* VULN: if len is a small negative number, this would result in an under-allocation */
    if (msg == RT_NULL)
    {
        RT_WLAN_LOG_E("wlan mgnt send msg err! No memory");
        return -RT_ENOMEM;
    }
    rt_memset(msg, 0, sizeof(struct rt_wlan_msg) + len);
    msg->event = event;
    if (len != 0)
    {
        msg->buff = (void *)&msg[1];
        rt_memcpy(msg->buff, buff, len); /* VULN: the small negative number would become a large unsigned size, and we would have a wild memcpy and a heap buffer overflow */
        msg->len = len;
    }

    /* send event to wlan thread */
    if (rt_wlan_workqueue_dowork(rt_wlan_mgnt_work, msg) != RT_EOK)
    {
        rt_free(msg);
        RT_WLAN_LOG_E("wlan mgnt do work fail");
        return -RT_ERROR;
    }
    return RT_EOK;
}
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8285


--[ 3.5 - CVE-2024-25390 - Heap buffer overflows in RT-Thread finsh

We spotted some heap buffer overflow vulnerabilities at the following
locations in the RT-Thread finsh source code:

* /components/finsh/msh_file.c
* /components/finsh/msh.c

Unbounded `rt_sprintf()` in the `directory_delete_for_msh()` function could
lead to a heap buffer overflow at the marked line:
```c
static void directory_delete_for_msh(const char *pathname, char f, char v)
{
    DIR *dir = NULL;
    struct dirent *dirent = NULL;
    char *full_path;

    if (pathname == RT_NULL)
        return;

    full_path = (char *)rt_malloc(DFS_PATH_MAX);
    if (full_path == RT_NULL)
        return;

    dir = opendir(pathname);
    if (dir == RT_NULL)
    {
        if (f == 0)
        {
            rt_kprintf("cannot remove '%s'\n", pathname);
        }
        rt_free(full_path);
        return;
    }

    while (1)
    {
        dirent = readdir(dir);
        if (dirent == RT_NULL)
            break;
        if (rt_strcmp(".", dirent->d_name) != 0 &&
                rt_strcmp("..", dirent->d_name) != 0)
        {
            rt_sprintf(full_path, "%s/%s", pathname, dirent->d_name); /* VULN: full_path is DFS_PATH_MAX bytes (only 256 bytes), potentially not enough to accommodate pathname + dirent->d_name */
            if (dirent->d_type != DT_DIR)
            {
                if (unlink(full_path) != 0)
                {
                    if (f == 0)
                        rt_kprintf("cannot remove '%s'\n", full_path);
                }
                else if (v)
                {
                    rt_kprintf("removed '%s'\n", full_path);
                }
            }
            else
            {
                directory_delete_for_msh(full_path, f, v);
            }
        }
    }
    closedir(dir);
    rt_free(full_path);
    if (rmdir(pathname) != 0)
    {
        if (f == 0)
            rt_kprintf("cannot remove '%s'\n", pathname);
    }
    else if (v)
    {
        rt_kprintf("removed directory '%s'\n", pathname);
    }
}
```

Unbounded `rt_sprintf()` in the `directory_setattr()` function could lead
to a heap buffer overflow at the marked line:
```c
static void directory_setattr(const char *pathname, struct dfs_attr *attr, char f, char v)
{
    DIR *dir = NULL;
    struct dirent *dirent = NULL;
    char *full_path;

    if (pathname == RT_NULL)
        return;

    full_path = (char *)rt_malloc(DFS_PATH_MAX);
    if (full_path == RT_NULL)
        return;

    dir = opendir(pathname);
    if (dir == RT_NULL)
    {
        if (f == 0)
        {
            rt_kprintf("cannot open '%s'\n", pathname);
        }
        rt_free(full_path);
        return;
    }

    while (1)
    {
        dirent = readdir(dir);
        if (dirent == RT_NULL)
            break;
        if (rt_strcmp(".", dirent->d_name) != 0 &&
            rt_strcmp("..", dirent->d_name) != 0)
        {
            rt_sprintf(full_path, "%s/%s", pathname, dirent->d_name); /* VULN: full_path is DFS_PATH_MAX bytes (only 256 bytes), potentially not enough to accommodate pathname + dirent->d_name */
            if (dirent->d_type == DT_REG)
            {
                if (dfs_file_setattr(full_path, attr) != 0)
                {
                    if (f == 0)
                    {
                        rt_kprintf("'%s' setattr failed, no such file or directory\n", full_path);
                    }
                }
                else if (v)
                {
                    rt_kprintf("'%s' setattr 0x%X\n", full_path, attr->st_mode);
                }
            }
            else if (dirent->d_type == DT_DIR)
            {
                directory_setattr(full_path, attr, f, v);
            }
        }
    }
    closedir(dir);
    rt_free(full_path);
    if (dfs_file_setattr(pathname, attr) != 0)
    {
        if (f == 0)
        {
            rt_kprintf("'%s' setattr failed, no such file or directory\n", pathname);
        }
    }
    else if (v)
    {
        rt_kprintf("'%s' setattr 0x%X\n", pathname, attr->st_mode);
    }
}
```

Unbounded `strcpy()` in the `msh_auto_complete_path()` function could lead
to a heap buffer overflow at the marked line:
```c
#ifdef DFS_USING_POSIX
void msh_auto_complete_path(char *path)
{
    DIR *dir = RT_NULL;
    struct dirent *dirent = RT_NULL;
    char *full_path, *ptr, *index;

    if (!path)
        return;

    full_path = (char *)rt_malloc(256);
    if (full_path == RT_NULL) return; /* out of memory */

    if (*path != '/')
    {
        getcwd(full_path, 256);
        if (full_path[rt_strlen(full_path) - 1]  != '/')
            strcat(full_path, "/");
    }
    else *full_path = '\0';

    index = RT_NULL;
    ptr = path;
    for (;;)
    {
        if (*ptr == '/') index = ptr + 1;
        if (!*ptr) break;

        ptr ++;
    }
    if (index == RT_NULL) index = path;

    if (index != RT_NULL)
    {
        char *dest = index;

        /* fill the parent path */
        ptr = full_path;
        while (*ptr) ptr ++;

        for (index = path; index != dest;)
            *ptr++ = *index++;
        *ptr = '\0';

        dir = opendir(full_path);
        if (dir == RT_NULL) /* open directory failed! */
        {
            rt_free(full_path);
            return;
        }

        /* restore the index position */
        index = dest;
    }

    /* auto complete the file or directory name */
    if (*index == '\0') /* display all of files and directories */
    {
        for (;;)
        {
            dirent = readdir(dir);
            if (dirent == RT_NULL) break;

            rt_kprintf("%s\n", dirent->d_name);
        }
    }
    else
    {
        int multi = 0;
        rt_size_t length, min_length;

        min_length = 0;
        for (;;)
        {
            dirent = readdir(dir);
            if (dirent == RT_NULL) break;

            /* matched the prefix string */
            if (strncmp(index, dirent->d_name, rt_strlen(index)) == 0)
            {
                multi ++;
                if (min_length == 0)
                {
                    min_length = rt_strlen(dirent->d_name);
                    /* save dirent name */
                    strcpy(full_path, dirent->d_name); /* VULN: if sizeof(dirent->d_name) > 256, we could overflow past the full_path buffer */
                }

                length = str_common(dirent->d_name, full_path);

                if (length < min_length)
                {
                    min_length = length;
                }
            }
        }

        if (min_length)
        {
            if (multi > 1)
            {
                /* list the candidate */
                rewinddir(dir);

                for (;;)
                {
                    dirent = readdir(dir);
                    if (dirent == RT_NULL) break;

                    if (strncmp(index, dirent->d_name, rt_strlen(index)) == 0)
                        rt_kprintf("%s\n", dirent->d_name);
                }
            }

            length = index - path;
            rt_memcpy(index, full_path, min_length);
            path[length + min_length] = '\0';

            /* try to locate folder */
            if (multi == 1)
            {
                struct stat buffer = {0};
                if ((stat(path, &buffer) == 0))
                {
                    if (S_ISDIR(buffer.st_mode))
                    {
                        strcat(path, "/");
                    }
                    else if (S_ISLNK(buffer.st_mode))
                    {
                        DIR *dir = opendir(path);
                        if (dir)
                        {
                            closedir(dir);
                            strcat(path, "/");
                        }
                    }
                }
            }
        }
    }

    closedir(dir);
    rt_free(full_path);
}
#endif /* DFS_USING_POSIX */
```

Fixes:
https://github.com/RT-Thread/rt-thread/pull/8305

See also:
https://github.com/RT-Thread/rt-thread/issues/8286


--[ 3.6 - CVE-2024-25391 - Stack buffer overflow in RT-Thread IPC

We spotted a stack buffer overflow vulnerability at the following location
in the RT-Thread IPC source code:

* /components/libc/posix/ipc/mqueue.c

Unbounded `rt_sprintf()` in the `mq_unlink()` function could lead to a
stack buffer overflow at the marked line:
```c
int mq_unlink(const char *name)
{
    if(*name == '/')
    {
        name++;
    }
    const char *mq_path = "/dev/mqueue/";
    char mq_name[RT_NAME_MAX + 12] = {0};
    rt_sprintf(mq_name, "%s%s", mq_path, name); /* VULN: stack buffer overflow */
    return unlink(mq_name);
}
```

Note: the mq_open() function in /components/libc/posix/ipc/mqueue.c
implements bound checking:
```c
    int len = rt_strlen(name);
    if (len > RT_NAME_MAX)
    {
        rt_set_errno(ENAMETOOLONG);
        return (mqd_t)(-1);
    }
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8287


--[ 3.7 - CVE-2024-25393 - Stack buffer overflow in RT-Thread AT server

We spotted a stack buffer overflow vulnerability at the following location
in the RT-Thread AT server source code:

* /components/net/at/src/at_server.c

Lack of length check in the `at_cmd_get_name()` function could lead to a
stack buffer overflow at the marked line:
```c
#ifdef AT_USING_SERVER
...
static rt_err_t at_cmd_get_name(const char *cmd_buffer, char *cmd_name)
{
    rt_size_t cmd_name_len = 0, i = 0;

    RT_ASSERT(cmd_name);
    RT_ASSERT(cmd_buffer);

    for (i = 0; i < strlen(cmd_buffer); i++)
    {
        if (*(cmd_buffer + i) == AT_CMD_QUESTION_MARK || *(cmd_buffer + i) == AT_CMD_EQUAL_MARK
                || *(cmd_buffer + i) == AT_CMD_CR
                || (*(cmd_buffer + i) >= AT_CMD_CHAR_0 && *(cmd_buffer + i) <= AT_CMD_CHAR_9))
        {
            cmd_name_len = i;
            rt_memcpy(cmd_name, cmd_buffer, cmd_name_len); /* VULN: cmd_buffer is AT_SERVER_RECV_BUFF_LEN bytes (256), while cmd_name is only AT_CMD_NAME_LEN bytes (16); therefore, it might be possible to overflow past the cmd_name buffer with a carefully crafted cmd_buffer */
            *(cmd_name + cmd_name_len) = '\0';

            return RT_EOK;
        }
    }

    return -RT_ERROR;
}
```

Fixes:
https://github.com/RT-Thread/rt-thread/commit/a8d5a645f2e26af67c8fe94e2c7f1726ae05cca7

See also:
https://github.com/RT-Thread/rt-thread/issues/8288


--[ 3.8 - CVE-2024-25395 - Static buffer overflow in RT-Thread rt-link utility

We spotted a static buffer overflow vulnerability at the following location
in the RT-Thread rt-link utility source code:

* /components/utilities/rt-link/src/rtlink.c

Lack of length check in the `frame_send()` function could lead to a static
buffer overflow at the marked line:
```c
static rt_ssize_t frame_send(struct rt_link_frame *frame)
{
    rt_size_t length = 0;
    rt_uint8_t *data = RT_NULL;

    rt_memset(rt_link_scb->sendbuffer, 0, sizeof(rt_link_scb->sendbuffer));
    data = rt_link_scb->sendbuffer;
    length = RT_LINK_HEAD_LENGTH;
    if (frame->head.crc)
    {
        length += RT_LINK_CRC_LENGTH;
    }
    if (frame->head.extend)
    {
        length += RT_LINK_EXTEND_LENGTH;
    }

    length += frame->data_len;
    frame->head.length = frame->data_len;
    rt_memcpy(data, &frame->head, RT_LINK_HEAD_LENGTH);
    data = data + RT_LINK_HEAD_LENGTH;
    if (frame->head.extend)
    {
        rt_memcpy(data, &frame->extend, RT_LINK_EXTEND_LENGTH);
        data = data + RT_LINK_EXTEND_LENGTH;
    }
    if (frame->attribute == RT_LINK_SHORT_DATA_FRAME || frame->attribute == RT_LINK_LONG_DATA_FRAME)
    {
        rt_memcpy(data, frame->real_data, frame->data_len); /* VULN: static buffer overflow, if frame->data_len > 1024 - 4 (it's a rt_uint16_t so at least in theory can be up to 65535) */
        data = data + frame->data_len;
    }
    if (frame->head.crc)
    {
        frame->crc = rt_link_scb->calculate_crc(RT_FALSE, rt_link_scb->sendbuffer, length - RT_LINK_CRC_LENGTH);
        rt_memcpy(data, &frame->crc, RT_LINK_CRC_LENGTH);
    }

    LOG_D("frame send seq(%d) len(%d) attr:(%d), crc:(0x%08x).", frame->head.sequence, length, frame->attribute, frame->crc);
    return rt_link_hw_send(rt_link_scb->sendbuffer, length);
}
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8289


--[ 3.9 - CVE-2024-25392 - Out-of-bounds static array access in RT-Thread var_export utility

We spotted a potential out-of-bounds static array access at the following
location in the RT-Thread var_export utility source code:

* /components/utilities/var_export/var_export.c

Improper size check due to the use of `RT_ASSERT()` in the
`var_export_init()` function at the marked line, which if compiled out in
production code could lead to multiple out-of-bounds `ve_exporter_tab`
static array accesses in the next lines:
```c
int var_export_init(void)
{
    /* initialize the var export table.*/
#if defined(__ARMCC_VERSION)                        /* for ARM C Compiler */
    ve_exporter_table = &__ve_table_start + 1;
    ve_exporter_num = &__ve_table_end - &__ve_table_start;
#elif defined (__IAR_SYSTEMS_ICC__)                 /* for IAR Compiler */
    ve_exporter_table = &__ve_table_start + 1;
    ve_exporter_num = &__ve_table_end - &__ve_table_start - 1;
#elif defined (__GNUC__)                            /* for GCC Compiler */
    extern const int __ve_table_start;
    extern const int __ve_table_end;
    ve_exporter_table = (const ve_exporter_t *)&__ve_table_start;
    ve_exporter_num = (const ve_exporter_t *)&__ve_table_end - ve_exporter_table;
#elif defined (_MSC_VER)                            /* for MS VC++ compiler */
    unsigned int *ptr_begin = (unsigned int *)&__ve_table_start;
    unsigned int *ptr_end = (unsigned int *)&__ve_table_end;
    static ve_exporter_t ve_exporter_tab[2048];
    static char __vexp_strbuf1[1024];
    static char __vexp_strbuf2[1024];
    ve_exporter_t ve_exporter_temp;
    rt_size_t index_i, index_j;

    /* past the three members in first ptr_begin */
    ptr_begin += (sizeof(struct ve_exporter) / sizeof(unsigned int));
    while (*ptr_begin == 0) ptr_begin++;
    do ptr_end--; while (*ptr_end == 0);

    /* Find var objects in custom segments to solve the problem of holes in objects in different files */
    ve_exporter_num = ve_init_find_obj(ptr_begin, ptr_end, ve_exporter_tab);

    /* check if the ve_exporter_num is out of bounds */
    RT_ASSERT(ve_exporter_num < (sizeof(ve_exporter_tab) / sizeof(ve_exporter_t))); /* VULN: size check is implemented via assertion */

    /* bubble sort algorithms */
    for (index_i = 0; index_i < (ve_exporter_num - 1); index_i++)
    {
        for (index_j = 0; index_j < ((ve_exporter_num - 1) - index_i); index_j++)
        {
            /* splice ve_exporter's module and ve_exporter's identifier into a complete string */
            rt_snprintf(__vexp_strbuf1,
                        sizeof(__vexp_strbuf1),
                        "%s%s",
                        ve_exporter_tab[index_j].module,
                        ve_exporter_tab[index_j].identifier);
            rt_snprintf(__vexp_strbuf2,
                        sizeof(__vexp_strbuf2),
                        "%s%s",
                        ve_exporter_tab[index_j + 1].module,
                        ve_exporter_tab[index_j + 1].identifier);
            if (rt_strcmp(__vexp_strbuf1, __vexp_strbuf2) > 0)
            {
                ve_exporter_temp = ve_exporter_tab[index_j];
                ve_exporter_tab[index_j] = ve_exporter_tab[index_j + 1];
                ve_exporter_tab[index_j + 1] = ve_exporter_temp;
            }
        }
    }

    ve_exporter_table = ve_exporter_tab;
#endif /* __ARMCC_VERSION */

    return ve_exporter_num;
}
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8290


--[ 3.10 - CVE-2024-25394 - Multiple vulnerabilities in RT-Thread ymodem utility

We spotted some potential vulnerabilities at the following location in the
RT-Thread ymodem utility source code:

* /components/utilities/ymodem/ry_sy.c

Unbounded `rt_sprintf()` in the `_rym_send_begin()` function could lead to
a buffer overflow at the marked line:
```c
static enum rym_code _rym_send_begin(
    struct rym_ctx *ctx,
    rt_uint8_t *buf,
    rt_size_t len)
{
    struct custom_ctx *cctx = (struct custom_ctx *)ctx;
    struct stat file_buf;
    char insert_0 = '\0';
    rt_err_t err;

    cctx->fd = open(cctx->fpath, O_RDONLY);
    if (cctx->fd < 0)
    {
        err = rt_get_errno();
        rt_kprintf("error open file: %d\n", err);
        return RYM_ERR_FILE;
    }
    rt_memset(buf, 0, len);
    err = stat(cctx->fpath, &file_buf);
    if (err != RT_EOK)
    {
        rt_kprintf("error open file.\n");
        return RYM_ERR_FILE;
    }

    const char *fdst = _get_path_lastname(cctx->fpath);
    if(fdst != cctx->fpath)
    {
        fdst = dfs_normalize_path(RT_NULL, fdst);
        if (fdst == RT_NULL)
        {
            return RYM_ERR_FILE;
        }
    }

    rt_sprintf((char *)buf, "%s%c%d", fdst, insert_0, file_buf.st_size); /* VULN: if we can make fdst large enough, there's no bound checking */

    return RYM_CODE_SOH;
}
```

Lack of NUL-termination in the `rym_download_file()` function at the marked
line:
```c
static rt_err_t rym_download_file(rt_device_t idev,const char *file_path)
{
    rt_err_t res;
    struct custom_ctx *ctx = rt_calloc(1, sizeof(*ctx));

    if (!ctx)
    {
        rt_kprintf("rt_malloc failed\n");
        return -RT_ENOMEM;
    }
    ctx->fd = -1;
    rt_strncpy(ctx->fpath, file_path, DFS_PATH_MAX); /* VULN: unterm if file_path is at least DFS_PATH_MAX bytes (comes from argv); it might cause infoleak or memory corruption */
    RT_ASSERT(idev);
    res = rym_recv_on_device(&ctx->parent, idev, RT_DEVICE_OFLAG_RDWR | RT_DEVICE_FLAG_INT_RX,
                             _rym_recv_begin, _rym_recv_data, _rym_recv_end, 1000);
    rt_free(ctx);

    return res;
}
```

Lack of NUL-termination in the `rym_upload_file()` function at the marked line:
```c
static rt_err_t rym_upload_file(rt_device_t idev, const char *file_path)
{
    rt_err_t res = 0;

    struct custom_ctx *ctx = rt_calloc(1, sizeof(*ctx));
    if (!ctx)
    {
        rt_kprintf("rt_malloc failed\n");
        return -RT_ENOMEM;
    }
    ctx->fd = -1;
    rt_strncpy(ctx->fpath, file_path, DFS_PATH_MAX); /* VULN: unterm if file_path is at least DFS_PATH_MAX bytes (comes from argv); it might cause infoleak or memory corruption */
    RT_ASSERT(idev);
    res = rym_send_on_device(&ctx->parent, idev,
                             RT_DEVICE_OFLAG_RDWR | RT_DEVICE_FLAG_INT_RX,
                             _rym_send_begin, _rym_send_data, _rym_send_end, 1000);
    rt_free(ctx);

    return res;
}
```

See also:
https://github.com/RT-Thread/rt-thread/issues/8291


--[ 3.11 - Use of outdated lwIP and TinyDir dependencies in RT-Thread

Finally, we noticed that the lwIP and TinyDir dependencies in RT-Thread are
outdated. In addition, during the source code review we also audited the
lwIP and TinyDir codebases, and found the following vulnerabilities that
were subsequently fixed by maintainers:

* httpc with LWIP_HTTPC_HAVE_FILE_IO: heap buffer overflow for long local filenames
  * https://git.savannah.nongnu.org/cgit/lwip.git/commit/?id=ee1523630a81fffa6b7d93dd0c7a6191de5856cd
* makefsdata: buffer corruption with very long paths
  * https://git.savannah.nongnu.org/cgit/lwip.git/commit/?id=b413b040936f48d4cd9ed632ac579542c710efae
* CVE-2023-49287 - Buffer overflow vulnerabilities with long path names in TinyDir
  * https://github.com/hnsecurity/vulns/blob/main/HNS-2023-04-tinydir.txt

These are not necessarily security concerns for RT-Thread. However, as a
general rule, it is important to keep dependencies as up to date as
possible, because more relevant bugs might get fixed upstream.

See also:
https://github.com/RT-Thread/rt-thread/issues/8292


--[ 4 - Affected products

RT-Thread 5.0.2, TinyDir 1.2.5, lwIP 2.2.0, and (likely) earlier versions
are affected by the vulnerabilities discussed in this advisory.


--[ 5 - Remediation

RT-Thread developers have fixed some of the vulnerabilities discussed in
this advisory. However, some of the reported vulnerabilities are still
open.

Please check the official RT-Thread channels for further information about
fixes.


--[ 6 - Disclosure timeline

We reported the vulnerabilities discussed in this advisory to RT-Thread in
November 2023, by opening issues [8] on GitHub as directed [9] by the
maintainers.

Following our request, MITRE assigned CVE IDs to all vulnerabilities in
February 2024.

Even though not all vulnerabilities were fixed by the maintainers, we
decided to go ahead with public disclosure after about 90 days since the
initial reports. This decision was taken mainly based on the fact that the
vulnerability reports were already publicly accessible as GitHub issues.

The detailed coordinated disclosure timeline follows:

2023-11-12: Asked on GitHub how to report potential security issues.
2023-11-13: RT-Thread founder replied to open issues directly on GitHub.
2023-11-20: Opened the first issue on GitHub.
2023-11-24: First vulnerability fixed; reported some other vulnerabilities.
2023-11-28: Second (and third) vulnerability fixed.
2023-12-24: Asked maintainers for updates.
2024-01-17: Requested the first batch of CVE IDs from MITRE.
2024-01-26: Informed maintainers of plan to publish advisory in February.
2024-02-02: Requested the second batch of CVE IDs from MITRE.
2024-02-07: CVE IDs for all reported vulnerabilities assigned by MITRE.
2024-02-08: Communicated CVE IDs and new 2024-03-05 publication date.
2024-02-18: Maintainers acknowledge publication date and inform about new fixes.
2024-03-05: Published advisory and writeup.


--[ 7 - Acknowledgments

We would like to thank RT-Thread, lwIP, and TinyDir developers for the time
dedicated to triaging the reported vulnerabilities. We would also like to
thank MITRE for assigning CVE IDs upon our request.


--[ 8 - References

[1] https://www.rt-thread.io/
[2] https://github.com/RT-Thread/rt-thread
[3] https://savannah.nongnu.org/projects/lwip/
[4] https://github.com/cxong/tinydir
[5] https://security.humanativaspa.it/ost2-zephyr-rtos-and-a-bunch-of-cves/
[6] https://github.com/0xdea/semgrep-rules
[7] https://security.humanativaspa.it/big-update-to-my-semgrep-c-cpp-ruleset/
[8] https://github.com/search?o=desc&q=author%3A0xdea+rt-thread&s=created&type=issues
[9] https://github.com/RT-Thread/rt-thread/issues/8250


Copyright (c) 2024 Marco Ivaldi and Humanativa Group. All rights reserved.