Lecture 017

System-Level I/O

strace: print system call

Unix I/O

/dev/: device /dev/tty2: terminal /boot/: kernel /proc/: kernel data structure

Unix File:

• sequence of bytes

• has [current file position], indicate offset in file to read and write (lseek() to move such pointer)

• types of file

  • regular file
  • directory
  • socket
  • named pipes (FIFOs)
  • symbolic links
  • character and block devices

• file ending: \n(LF) in unix, \r\n(CR+LF) in windows and http

• . to itself

• .. to parent

• ... invalid because there is no link named ... in directory

• ../.. to parent of parent

• pwd path, tree see directory as tree

File Functions

file descriptor:

• STDIN: 0

• STDOUT: 1

• STDERR: 2

open(): return lowest numbered file descriptor not currently open

close():

• closing already closed file in multi-threaded program is dangerous (always check return value)

• closing stdin/stdout/stderr cannot be re-opened again (maybe unless you save it somewhere)

read(fd, buf, sizeof(buf)):

• read from current file position, and update file position

• return number of bytes read

• short counts: read less than requested (not error)

• error EINTR when interrupted by other signal, in this case, typically retry

write(fd, buf, sizeof(buf)):

• write up to size of buffer

• short counts: write less than requested (not error)

• error EINTR when interrupted by other signal, in this case, typically retry

Short count:

• Encountering (end-of-file) EOF on reads

• Reading text lines from a terminal

• Reading and writing network sockets

No short count:

• Reading from disk files (except for reading EOF at the end of file)

• Writing to disk files

Buffer:

• expect short count and write as many byte as possible at a time because syscall is expensive

Metadata, Sharing, Redirection

File Metadata: maintained by kernel

• Accessed by users with the stat and fstat functions

• 

File Sharing

• 

• Open File Multiple Time:

  

• Forking Process:

  

• dup2():

  

• 

I/O redirection

Standard I/O

printf(): using stream buffer, flush on \n by default fflush(stddout): flush

Robust I/O (RIO) Package

Download

Unbuffered version

ssize_t rio_readn(int fd, void *usrbuf, size_t n)

• Same as read but short count only at EOF

• only use when know how many bytes to read

• 

rio_writen

• Same as write but never returns short count

Calls to rio_readn and rio_writen can be interleaved arbitrarily on the same descriptor

Buffered version

Thread-safe, typically no short count, can be interleaved on the same descriptor with all buffered version.

typedef struct {
  int rio_fd; /* descriptor for this internal buf */
  int rio_cnt; /* unread bytes in internal buf */
  char *rio_bufptr; /* next unread byte in internal buf */
  char rio_buf[RIO_BUFSIZE]; /* internal buffer */
} rio_t;

void rio_readinitb(rio_t *rp, int fd)

• init rp for specific fd

ssize_t rio_readlineb(rio_t *rp, void *usrbuf, size_t maxlen)

• read up to maxlen into usrbuf

• use when reading network sockets

• stop when reach maxlen, EOF, or \n

ssize_t rio_readnb(rio_t *rp, void *usrbuf, size_t n)

• stop when reach maxlen, or EOF

mmap()

mmap() is good for copy file to memory, copy files

Pros and Cons & Working w/ Binary

System I/O Good:

• most general, lowest overhead

• provide functions for accessing file metadata

• async-signal-safe, can be used in signal handlers

System I/O Bad:

• short counts is error prone

• need buffering if want to read efficiently

Buffered I/O Good:

• increase efficiency by decreasing number of syscall

• handled short counts

Buffered I/O Bad:

• no interface for accessing metadata

• Standard I/O are not async-signal-safe

• Standard I/O not good for network sockets

Standard I/O: disk, terminal Unix I/O: signal handlers, highest performance RIO: network sockets

Binary:

• don't use fgets, scanf, rio_readlineb since they interpret EOL

• don't use strlen, strcpy, strcat

• use rio_readn or rio_readnb instead