diff --git a/api/src/file/inotify.rs b/api/src/file/inotify.rs
new file mode 100644
index 00000000..9b16d655
--- /dev/null
+++ b/api/src/file/inotify.rs
@@ -0,0 +1,327 @@
+use alloc::{
+    borrow::Cow,
+    collections::{BTreeMap, VecDeque},
+    sync::{Arc, Weak},
+    vec::Vec,
+};
+use core::{
+    any::Any,
+    mem::size_of,
+    sync::atomic::{AtomicBool, AtomicI32, AtomicU64, Ordering},
+};
+
+use axerrno::{AxError, AxResult};
+use axfs::ROOT_FS_CONTEXT;
+use axfs_ng_vfs::Location;
+use axio::{BufMut, Write};
+use axpoll::{IoEvents, PollSet, Pollable};
+use axsync::Mutex;
+use axtask::future::{block_on, poll_io};
+use bitflags::bitflags;
+use lazy_static::lazy_static;
+
+use crate::file::{FileLike, Kstat, SealedBuf, SealedBufMut};
+
+/// ========== Inotify event flags ==========
+pub const IN_IGNORED: u32 = 0x00008000; // File was ignored
+/// ==========  Flags for inotify_init1()==========
+pub const IN_CLOEXEC: u32 = 0o2000000; // 02000000, Set FD_CLOEXEC
+pub const IN_NONBLOCK: u32 = 0o0004000; // 00004000, Set O_NONBLOCK
+
+// flags for inotify_syscalls
+bitflags! {
+    #[derive(Debug, Clone, Copy, Default)]
+    pub struct InotifyFlags: u32 {
+        /// Create a file descriptor that is closed on `exec`.
+        const CLOEXEC = IN_CLOEXEC;
+        /// Create a non-blocking inotify instance.
+        const NONBLOCK = IN_NONBLOCK;
+    }
+}
+
+/// inotifyEvent(Memory layout fully compatible with Linux)
+#[repr(C)]
+#[derive(Debug, Clone, Copy)]
+pub struct InotifyEvent {
+    pub wd: i32,     // Watch descriptor
+    pub mask: u32,   // Mask describing event
+    pub cookie: u32, // Unique cookie associating related events
+    pub len: u32,    /* Size of name field  (including null terminator)
+                      * note: the name field is a variable-length array and is not contained in this struct */
+}
+
+/// Monitoring data of each inode（stored in Location::user_data()）
+#[derive(Default)]
+struct InodeWatchData {
+    // key: watch descriptor, value: (instance_id, event mask)
+    watches: Mutex<BTreeMap<i32, (u64, u32)>>, // Using Mutex to wrap
+}
+impl InodeWatchData {
+    fn add_watch(&self, wd: i32, instance_id: u64, mask: u32) {
+        self.watches.lock().insert(wd, (instance_id, mask));
+    }
+
+    fn remove_watch(&self, wd: i32) -> bool {
+        self.watches.lock().remove(&wd).is_some()
+    }
+
+    fn is_empty(&self) -> bool {
+        self.watches.lock().is_empty()
+    }
+}
+
+/// inotify instance
+pub struct InotifyInstance {
+    // event_queue:save serialized event data
+    event_queue: Mutex<VecDeque<Vec<u8>>>,
+
+    // Added: weak reference from wd to Location (for quick lookup and path retrieval)
+    wd_to_location: Mutex<BTreeMap<i32, Weak<Location>>>,
+
+    next_wd: AtomicI32,
+
+    // Instance ID (unique identifier)
+    instance_id: u64,
+
+    // blocking/non-blocking mode
+    non_blocking: AtomicBool,
+
+    // poll support
+    poll_set: PollSet,
+}
+
+impl InotifyInstance {
+    /// create new instance
+    pub fn new(flags: i32) -> AxResult<Arc<Self>> {
+        static NEXT_INSTANCE_ID: AtomicU64 = AtomicU64::new(1);
+
+        let flags = flags as u32;
+        // verify flags
+        let valid_flags = IN_NONBLOCK | IN_CLOEXEC;
+        if flags & !valid_flags != 0 {
+            return Err(AxError::InvalidInput);
+        }
+
+        let non_blocking = (flags & IN_NONBLOCK) != 0;
+        let instance_id = NEXT_INSTANCE_ID.fetch_add(1, Ordering::Relaxed);
+
+        let instance = Arc::new(Self {
+            event_queue: Mutex::new(VecDeque::new()),
+            wd_to_location: Mutex::new(BTreeMap::new()),
+            next_wd: AtomicI32::new(1),
+            instance_id,
+            non_blocking: AtomicBool::new(non_blocking),
+            poll_set: PollSet::new(),
+        });
+
+        // Registered to global manager
+        INOTIFY_MANAGER.register(instance_id, Arc::clone(&instance));
+
+        Ok(instance)
+    }
+
+    /// Serialized events are in binary format for users to read with char[]
+    fn serialize_event(event: &InotifyEvent, name: Option<&str>) -> Vec<u8> {
+        // +1 for null terminator
+        let name_len = name.map_or(0, |s| s.len() + 1);
+        let total_len = size_of::<InotifyEvent>() + name_len;
+
+        // Linux requires events to be 4-byte aligned
+        let aligned_len = (total_len + 3) & !3;
+
+        let mut buf = Vec::with_capacity(aligned_len);
+
+        // Write event header (native byte order, matching architecture)
+        buf.extend_from_slice(&event.wd.to_ne_bytes());
+        buf.extend_from_slice(&event.mask.to_ne_bytes());
+        buf.extend_from_slice(&event.cookie.to_ne_bytes());
+        buf.extend_from_slice(&(name_len as u32).to_ne_bytes());
+
+        // Write filename (if any)
+        if let Some(name) = name {
+            buf.extend_from_slice(name.as_bytes());
+            buf.push(0); // null terminator
+
+            // Padding for alignment (using null bytes)
+            let padding = aligned_len - total_len;
+            buf.resize(buf.len() + padding, 0);
+        }
+
+        buf
+    }
+
+    /// add watch for a path
+    /// Returns watch descriptor (wd)
+    pub fn add_watch(&self, path: &str, mask: u32) -> AxResult<i32> {
+        // Convert path to Location
+        let location = self.resolve_path(path)?;
+        // Generate a new watch descriptor
+        let wd = self.next_wd.fetch_add(1, Ordering::Relaxed);
+        if wd == i32::MAX {
+            return Err(AxError::StorageFull);
+        }
+
+        // Get user_data of location
+        let mut user_data = location.user_data();
+
+        // Get or create InodeWatchData
+        // Use get_or_insert_with to get Arc<InodeWatchData>
+        let inode_data = user_data.get_or_insert_with(InodeWatchData::default);
+
+        // Store watch info: wd -> (instance_id, mask)
+        inode_data.add_watch(wd, self.instance_id, mask);
+
+        // Store reverse mapping: wd -> location
+        self.wd_to_location
+            .lock()
+            .insert(wd, Arc::downgrade(&location));
+
+        Ok(wd)
+    }
+
+    /// remove watch (generate IN_IGNORED event)
+    pub fn remove_watch(&self, wd: i32) -> AxResult<()> {
+        // Get location from wd_to_location
+        let location_weak = {
+            let mut wd_map = self.wd_to_location.lock();
+            wd_map.remove(&wd).ok_or(AxError::InvalidInput)?
+        };
+
+        // Generate IN_IGNORED event
+        let event = InotifyEvent {
+            wd,
+            mask: IN_IGNORED,
+            cookie: 0,
+            len: 0,
+        };
+        let event_data = Self::serialize_event(&event, None);
+        self.push_event(event_data);
+
+        // If location exists, remove watch from its user_data
+        if let Some(location) = location_weak.upgrade() {
+            let user_data = location.user_data();
+
+            if let Some(inode_data) = user_data.get::<InodeWatchData>() {
+                // Remove watch
+                inode_data.remove_watch(wd);
+                // If no more watches, remove InodeWatchData
+                if inode_data.is_empty() {
+                    // Actually TypeMap has no remove, can only leave empty
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Push event to queue
+    fn push_event(&self, event_data: Vec<u8>) {
+        let mut queue = self.event_queue.lock();
+        queue.push_back(event_data);
+        self.poll_set.wake();
+    }
+
+    fn resolve_path(&self, path: &str) -> AxResult<Arc<Location>> {
+        let fs_ctx = ROOT_FS_CONTEXT.get().ok_or(AxError::NotFound)?;
+        let loc = fs_ctx.resolve(path).map_err(|_| AxError::NotFound)?;
+        Ok(Arc::new(loc))
+    }
+}
+
+impl FileLike for InotifyInstance {
+    fn read(&self, dst: &mut SealedBufMut) -> axio::Result<usize> {
+        block_on(poll_io(self, IoEvents::IN, self.nonblocking(), || {
+            let mut queue = self.event_queue.lock();
+
+            if queue.is_empty() {
+                return Err(AxError::WouldBlock);
+            }
+
+            let mut bytes_written = 0;
+
+            // Write as many events as possible without exceeding the buffer
+            while let Some(event_data) = queue.front() {
+                if dst.remaining_mut() < event_data.len() {
+                    break;
+                }
+
+                dst.write(event_data)?;
+                bytes_written += event_data.len();
+                queue.pop_front();
+            }
+
+            if bytes_written > 0 {
+                Ok(bytes_written)
+            } else {
+                // Buffer too small to write a complete event
+                Err(AxError::InvalidInput)
+            }
+        }))
+    }
+
+    fn write(&self, _src: &mut SealedBuf) -> axio::Result<usize> {
+        Err(AxError::BadFileDescriptor)
+    }
+
+    fn stat(&self) -> axio::Result<Kstat> {
+        Ok(Kstat::default())
+    }
+
+    fn nonblocking(&self) -> bool {
+        self.non_blocking.load(Ordering::Acquire)
+    }
+
+    fn set_nonblocking(&self, non_blocking: bool) -> axio::Result {
+        self.non_blocking.store(non_blocking, Ordering::Release);
+        Ok(())
+    }
+
+    fn path(&self) -> Cow<str> {
+        "anon_inode:[inotify]".into()
+    }
+
+    fn into_any(self: Arc<Self>) -> Arc<dyn Any + Send + Sync> {
+        self
+    }
+}
+
+impl Pollable for InotifyInstance {
+    fn poll(&self) -> IoEvents {
+        let mut events = IoEvents::empty();
+        let queue = self.event_queue.lock();
+
+        // Events available to read
+        events.set(IoEvents::IN, !queue.is_empty());
+        // inotify file is not writable
+        events.set(IoEvents::OUT, false);
+
+        events
+    }
+
+    fn register(&self, context: &mut core::task::Context<'_>, events: IoEvents) {
+        if events.contains(IoEvents::IN) {
+            self.poll_set.register(context.waker());
+        }
+    }
+}
+
+// Global manager (singleton)
+struct InotifyManager {
+    instances: Mutex<BTreeMap<u64, Arc<InotifyInstance>>>,
+}
+
+impl InotifyManager {
+    fn new() -> Self {
+        Self {
+            instances: Mutex::new(BTreeMap::new()),
+        }
+    }
+
+    fn register(&self, instance_id: u64, instance: Arc<InotifyInstance>) {
+        self.instances.lock().insert(instance_id, instance);
+    }
+}
+
+lazy_static! {
+    static ref INOTIFY_MANAGER: InotifyManager = InotifyManager::new();
+}
diff --git a/api/src/file/mod.rs b/api/src/file/mod.rs
index f6d6e4e5..48d4e4c2 100644
--- a/api/src/file/mod.rs
+++ b/api/src/file/mod.rs
@@ -1,6 +1,7 @@
 pub mod epoll;
 pub mod event;
 mod fs;
+mod inotify;
 mod net;
 mod pidfd;
 mod pipe;
@@ -23,6 +24,7 @@ use starry_core::{resources::AX_FILE_LIMIT, task::AsThread};
 
 pub use self::{
     fs::{Directory, File, ResolveAtResult, metadata_to_kstat, resolve_at, with_fs},
+    inotify::{InotifyFlags, InotifyInstance},
     net::Socket,
     pidfd::PidFd,
     pipe::Pipe,
diff --git a/api/src/syscall/fs/inotify.rs b/api/src/syscall/fs/inotify.rs
new file mode 100644
index 00000000..9f73a18c
--- /dev/null
+++ b/api/src/syscall/fs/inotify.rs
@@ -0,0 +1,62 @@
+use core::ffi::c_char;
+
+use axerrno::{AxError, AxResult};
+
+use crate::{
+    file::{InotifyFlags, InotifyInstance, add_file_like, get_file_like},
+    mm::vm_load_string,
+};
+pub fn sys_inotify_init1(flags: i32) -> AxResult<isize> {
+    let instance = InotifyInstance::new(flags)?;
+
+    // Use add_file_like to add file descriptor
+    let cloexec = (flags as u32) & InotifyFlags::CLOEXEC.bits() != 0;
+    add_file_like(instance, cloexec).map(|fd| {
+        debug!("sys_inotify_init1: allocated fd {fd}");
+        fd as isize
+    })
+}
+
+pub fn sys_inotify_add_watch(fd: i32, pathname: *const c_char, mask: u32) -> AxResult<isize> {
+    debug!("inotify_add_watch called: fd={fd}, mask={mask:#x}");
+
+    // Load pathname (using vm_load_string, same as sys_open)
+    let path = vm_load_string(pathname)?;
+    // Get file corresponding to file descriptor
+    let file = get_file_like(fd)?;
+    // Convert to inotify instance
+    let inotify = match file.clone().into_any().downcast::<InotifyInstance>() {
+        Ok(inst) => inst,
+        Err(_) => {
+            warn!("inotify_add_watch: fd {fd} is not an inotify instance");
+            return Err(AxError::InvalidInput);
+        }
+    };
+
+    // Add watch
+    let wd = inotify.add_watch(&path, mask)?;
+
+    info!("inotify watch added: fd={fd}, path={path}, wd={wd}");
+    Ok(wd as isize)
+}
+
+pub fn sys_inotify_rm_watch(fd: i32, wd: i32) -> AxResult<isize> {
+    debug!("sys_inotify_rm_watch: fd={fd}, wd={wd}");
+
+    // Get file
+    let file = get_file_like(fd)?;
+
+    // Convert to inotify instance
+    let inotify = match file.clone().into_any().downcast::<InotifyInstance>() {
+        Ok(inst) => inst,
+        Err(_) => {
+            warn!("inotify_rm_watch: fd {fd} is not an inotify instance");
+            return Err(AxError::InvalidInput);
+        }
+    };
+
+    // Remove watch
+    inotify.remove_watch(wd)?;
+    info!("inotify_rm_watch: removed watch wd={wd} from fd={fd}");
+    Ok(0)
+}
diff --git a/api/src/syscall/fs/mod.rs b/api/src/syscall/fs/mod.rs
index e54069d4..e63d17b7 100644
--- a/api/src/syscall/fs/mod.rs
+++ b/api/src/syscall/fs/mod.rs
@@ -1,6 +1,7 @@
 mod ctl;
 mod event;
 mod fd_ops;
+mod inotify;
 mod io;
 mod memfd;
 mod mount;
@@ -10,5 +11,6 @@ mod signalfd;
 mod stat;
 
 pub use self::{
-    ctl::*, event::*, fd_ops::*, io::*, memfd::*, mount::*, pidfd::*, pipe::*, signalfd::*, stat::*,
+    ctl::*, event::*, fd_ops::*, inotify::*, io::*, memfd::*, mount::*, pidfd::*, pipe::*,
+    signalfd::*, stat::*,
 };
diff --git a/api/src/syscall/mod.rs b/api/src/syscall/mod.rs
index aea95737..10338ca9 100644
--- a/api/src/syscall/mod.rs
+++ b/api/src/syscall/mod.rs
@@ -284,6 +284,13 @@ pub fn handle_syscall(uctx: &mut UserContext) {
         // event
         Sysno::eventfd2 => sys_eventfd2(uctx.arg0() as _, uctx.arg1() as _),
 
+        // inotify
+        Sysno::inotify_init1 => sys_inotify_init1(uctx.arg0() as _),
+        Sysno::inotify_add_watch => {
+            sys_inotify_add_watch(uctx.arg0() as _, uctx.arg1() as _, uctx.arg2() as _)
+        }
+        Sysno::inotify_rm_watch => sys_inotify_rm_watch(uctx.arg0() as _, uctx.arg1() as _),
+
         // pidfd
         Sysno::pidfd_open => sys_pidfd_open(uctx.arg0() as _, uctx.arg1() as _),
         Sysno::pidfd_getfd => sys_pidfd_getfd(uctx.arg0() as _, uctx.arg1() as _, uctx.arg2() as _),
@@ -593,7 +600,6 @@ pub fn handle_syscall(uctx: &mut UserContext) {
         // dummy fds
         Sysno::timerfd_create
         | Sysno::fanotify_init
-        | Sysno::inotify_init1
         | Sysno::userfaultfd
         | Sysno::perf_event_open
         | Sysno::io_uring_setup