Interprocess%20Communication

1
Interprocess Communication
2

• Resource Sharing
• Kernel Data structures, Buffers
• Processes Shared Memory, Files
• Synchronization Methods
• Kernel Wait Queues, Semaphores
• Processes Semaphores, file locks
• Data Exchange
• Kernel Signals
• Processes Signals, Message Queues, Sockets,
  Pipes

3
Kernel Synchronization

• Struct wait_queue
• Struct task_struct task
• Struct wait_queue next
• Add_wait_queue(struct wait_queue p, struct
  wait_queue wait)
• Unsigned long flags
• Save_flags(flags)
• Cli()
• __add_wait_queue(p,wait) // actual insert
• Restore_flags(flags)
• Remove_wait_queue(.)

4
Semaphores within Kernel

• Struct semaphore
• int count, waiting
• struct wait_queue wait
• down(struct semaphore psem)
• while (--psem-gtcount lt 0)
• psem-gtwaiting
• if (psem-gtcount psem-gtwaiting lt 0)
• do sleep(psem-gtwait) while (psem-gtcount lt 0)
• psem-gtcount psem-gtwaiting
• psem-gtwaiting 0
• More complicated due to race conditions and not
  turning off interrupts

5
File Locking

• Locking Entire Files
• Auxiliary file (lock file) if it exists, lock
  fails. You can then sleep and retry.
• Using fcntl() system call.
• Locking file areas
• Using fcntl()

6

• int sys_fcntl(unsigned int fd, unsigned int cmd,
  unsigned long arg)
• Cmd FGET_LK (returns if locked or not),
  F_SET_LK (tries to lock, else returns), F_SETLKW
  (locks, else blocks)
• Arg pointer to a flock.
• Struct flock
• Short l_type // F_RDLCK, F_WRLCK, F_UNLCK,
  F_SHLCK, F_EXLCK
• Short l_whence // SEEK_SET (file start),
  SEEK_CUR, SEEK_END
• Off_t l_start
• Off_t l_len //0 indicates current and future
  end of file
• Pid_t pid

7

• Doubly linked list file_lock_table is used
• Struct file_lock
• Struct file_lock fl_next // singly linked for
  this inode
• Struct file_lock fl_nextlink, fl_prevlink //
  across all inodes
• Struct task_struct fl_owner
• Struct wait_queue fl_wait
• Struct file fl_file
• Off_t fl_start, fl_end
• Struct file_lock file_lock_table
• Linux tracks down deadlocks and returns EDEADLK
• Locks are not transferred to child by fork but
  are retained in exec

8
Pipes

• mkfifo ltpathnamegt
• Or
• mknod ltpathnamegt p
• ls l gt fifo more lt fifo

9

• Struct pipe_inode_info
• Struct wait_queue wait
• Char base // address of FIFO bounded buffer
• Unsigned int start
• Unsigned int len
• Unsigned int lock
• Unsigned int rd_openers
• Unsigned int wr_openers
• Unsigned int readers // currently reading
• Unsigned int writers // currently writing
• Read/Write will not block if O_NONBLOCK is set in
  descriptor.
• Write operation is atomic (no intermixing of
  writes) as long as data does not exceed buffer
  size.

10
Debugging

• Int sys_ptrace(long request, long pid, long addr,
  long data)
• Request
• PTRACE_TRACEME // trace me
• PTRACE_ATTACH // to trace child
• PTRACE_PEEKTEXT/PEEKDATA // to examine locns.
• PTRACE_POKETEXT/POKEDATA // to write
• PTRACE_CONT // to continue the child
• PTRACE_SYSCALL // continue until next sys call
• PTRACE_SINGLESTEP

11
How does gdb work?

• Forks a child process
• Child calls function with PTRACE_TRACEME
• Child then does an execve() on the program
• The execve call sebds a SIGTRAP to itself.
• In the handler, the process is halted, and parent
  is informed with SIGCHLD signal.
• Parent (debugger) waits for this signal using
  wait()
• It can inspect child memory, modify it, set
  breakpoints (int 3 instruction)
• It requests PTRACE_CONT, child continues till
  trapping on an int 3 instruction.
• strace simply uses PTRACE_SYSCALL

12
System V IPC Support

• IPC permissions
• Struct ipc_perm
• Key_t key // using file name and char
• Ushort uid, gid // owner
• Ushort cuid, cgid // creator
• Ushort mode // access modes
• Semaphores
• Message Queues
• Shared Memory
• Sockets