msm@attctc.Dallas.TX.US (Michael Mattone) (09/28/89)
A few weeks ago, Bruce Evans (evans@ditsyda.oz.au), mentioned a
fix I had posted about a year ago that was supposed to fix the
limit of 7K when writing to a pipe. He also mentioned that it
had a bug that caused blocks to be skipped. Well, I took a look
at FS with my patches again, and believe I have fixed this bug.
I believe the bug can be fixed by adding one line to the fs/read.c
file before the "goto bigpipe" statement (from my original posting):
if (nbytes > xbytes) {
f->filp_pos = position; /* BUG fixed - add this line */
goto bigpipe; /* more in a pipe to do */
}
I'm enclosing my patches again in case there are others interested.
I'd really like to hear if anyone tries this and of any other bugs they
might find.
----------------------------------cut here---------------------------------
echo x - fproc.h.cdiff
sed '/^X/s///' > fproc.h.cdiff << '/'
X*** fproc.h Tue Jun 28 22:28:05 1988
X--- new/fproc.h Mon Aug 15 02:12:25 1988
X***************
X*** 22,27 ****
X--- 22,28 ----
X char fp_task; /* which task is proc suspended on */
X int fp_pid; /* process id */
X int fp_pgrp; /* process group */
X+ int fp_tot_io; /* total cum_io for pipes */
X } fproc[NR_PROCS];
X
X /* Field values. */
/
echo x - main.c.cdiff
sed '/^X/s///' > main.c.cdiff << '/'
X*** main.c Tue Jun 28 22:28:13 1988
X--- new/main.c Mon Aug 15 02:08:10 1988
X***************
X*** 110,115 ****
X--- 110,116 ----
X
X who = m.m_source;
X fs_call = m.m_type;
X+ fproc[who].fp_tot_io = 0; /* in case we do a pipe */
X }
X
X
/
echo x - pipe.c.cdiff
sed '/^X/s///' > pipe.c.cdiff << '/'
X*** pipe.c Sat Jul 16 20:34:19 1988
X--- new/pipe.c Tue Nov 1 01:26:47 1988
X***************
X*** 21,26 ****
X--- 21,27 ----
X #include "../h/signal.h"
X #include "const.h"
X #include "type.h"
X+ #include "dev.h"
X #include "file.h"
X #include "fproc.h"
X #include "glo.h"
X***************
X*** 81,91 ****
X /*===========================================================================*
X * pipe_check *
X *===========================================================================*/
X! PUBLIC int pipe_check(rip, rw_flag, bytes, position)
X register struct inode *rip; /* the inode of the pipe */
X int rw_flag; /* READING or WRITING */
X register int bytes; /* bytes to be read or written (all chunks) */
X register file_pos position; /* pointer to current file position */
X {
X /* Pipes are a little different. If a process reads from an empty pipe for
X * which a writer still exists, suspend the reader. If the pipe is empty
X--- 82,93 ----
X /*===========================================================================*
X * pipe_check *
X *===========================================================================*/
X! PUBLIC int pipe_check(rip, rw_flag, bytes, position, xbytes)
X register struct inode *rip; /* the inode of the pipe */
X int rw_flag; /* READING or WRITING */
X register int bytes; /* bytes to be read or written (all chunks) */
X register file_pos position; /* pointer to current file position */
X+ int *xbytes;
X {
X /* Pipes are a little different. If a process reads from an empty pipe for
X * which a writer still exists, suspend the reader. If the pipe is empty
X***************
X*** 110,122 ****
X }
X } else {
X /* Process is writing to a pipe. */
X- if (bytes > PIPE_SIZE) return(EFBIG);
X if (find_filp(rip, R_BIT) == NIL_FILP) {
X /* Tell kernel to generate a SIGPIPE signal. */
X sys_kill((int)(fp - fproc), SIGPIPE);
X return(EPIPE);
X }
X
X if (position + bytes > PIPE_SIZE) {
X suspend(XPIPE); /* stop writer -- pipe full */
X return(0);
X--- 112,128 ----
X }
X } else {
X /* Process is writing to a pipe. */
X if (find_filp(rip, R_BIT) == NIL_FILP) {
X /* Tell kernel to generate a SIGPIPE signal. */
X sys_kill((int)(fp - fproc), SIGPIPE);
X return(EPIPE);
X }
X
X+ if (bytes > PIPE_SIZE) {
X+ if ((*xbytes = PIPE_SIZE - position) > 0)
X+ bytes = *xbytes;
X+ }
X+
X if (position + bytes > PIPE_SIZE) {
X suspend(XPIPE); /* stop writer -- pipe full */
X return(0);
/
echo x - read.c.cdiff
sed '/^X/s///' > read.c.cdiff << '/'
X*** read.c Fri Sep 30 03:15:42 1988
X--- new/read.c Wed Sep 27 19:23:04 1989
X***************
X*** 54,59 ****
X--- 54,60 ----
X register unsigned off, cum_io;
X file_pos position;
X int r, chunk, mode_word, usr, seg, block_spec, char_spec;
X+ int xbytes;
X struct filp *wf;
X extern struct super_block *get_super();
X extern struct filp *find_filp(), *get_filp();
X***************
X*** 111,124 ****
X if (position > f_size) clear_zone(rip, f_size, 0);
X }
X
X! /* Pipes are a little different. Check. */
X! if (rip->i_pipe &&
X! (r = pipe_check(rip, rw_flag, nbytes, position)) <= 0)
X! return r;
X! /* Split the transfer into chunks that don't span two blocks. */
X! while (nbytes != 0) {
X! off = position % BLOCK_SIZE; /* offset within a block */
X! chunk = MIN(nbytes, BLOCK_SIZE - off);
X if (chunk < 0) chunk = BLOCK_SIZE - off;
X
X if (rw_flag == READING || (block_spec && rw_flag == WRITING)) {
X--- 112,127 ----
X if (position > f_size) clear_zone(rip, f_size, 0);
X }
X
X! bigpipe:
X! xbytes = nbytes;
X! /* Pipes are a little different. Check. */
X! if (rip->i_pipe &&
X! (r = pipe_check(rip, rw_flag, nbytes, position, &xbytes)) <= 0)
X! return r;
X! /* Split the transfer into chunks that don't span two blocks. */
X! while (xbytes != 0) {
X! off = position % BLOCK_SIZE; /* offset within a block */
X! chunk = MIN(xbytes, BLOCK_SIZE - off);
X if (chunk < 0) chunk = BLOCK_SIZE - off;
X
X if (rw_flag == READING || (block_spec && rw_flag == WRITING)) {
X***************
X*** 136,141 ****
X--- 139,145 ----
X /* Update counters and pointers. */
X buffer += chunk; /* user buffer address */
X nbytes -= chunk; /* bytes yet to be read */
X+ xbytes -= chunk; /* for pipes */
X cum_io += chunk; /* bytes read so far */
X position += chunk; /* position within the file */
X }
X***************
X*** 148,153 ****
X--- 152,165 ----
X rip->i_modtime = clock_time();
X rip->i_dirt = DIRTY;
X }
X+ if (r == OK && rip->i_pipe) {
X+ fp->fp_tot_io += cum_io;
X+ if (nbytes > xbytes) {
X+ f->filp_pos = position; /* just in case */
X+ goto bigpipe; /* more in a pipe to do */
X+ }
X+ cum_io = fp->fp_tot_io; /* for accurate count */
X+ }
X } else {
X if (rip->i_pipe && position >= rip->i_size) {
X /* Reset pipe pointers. */
/ast@cs.vu.nl (Andy Tanenbaum) (09/28/89)
In article <9507@attctc.Dallas.TX.US> msm@attctc.Dallas.TX.US (Michael Mattone) writes: > goto bigpipe; /* more in a pipe to do */ This is definitely a nono. No gotos please! Andy Tanenbaum (ast@cs.vu.nl)
mccrea@ecf.ncsl.nist.gov (MCCREA, SAM) (09/28/89)
>> goto bigpipe; /* more in a pipe to do */
That is definitely a YES, YES!!
Where did you get your programming instruction? From
a Cereal box?tweten@gilmore.nas.nasa.gov (Dave Tweten) (09/29/89)
From: Andy Tanenbaum <ast%CS.VU.NL@VM1.NoDak.EDU> This is definitely a nono. No gotos please! I hope you are merely overstating you case, Dr. Tanenbaum. I had thought the "No gotos" religion had died, mercifully for the rest of us. I continue to find that something on the order of one "goto" per 5000 lines of C code or so is required to maximize simplicity, clarity and therefore maintainability. Applications include breaking out of two levels of loop at once, breaking out of a loop, but not executing the loop exhaustion code, switch cases which should decide to break out of the surrounding loop, etc., etc. I didn't see enough context to tell if the specific "goto" which triggered your message was a good one, but surely an absolute "No gotos" is not a defensable position. Please say you were kidding.
ast@cs.vu.nl (Andy Tanenbaum) (09/30/89)
In article <24962@louie.udel.EDU> tweten@gilmore.nas.nasa.gov (Dave Tweten) writes: >From: Andy Tanenbaum <ast%CS.VU.NL@VM1.NoDak.EDU> >surely an absolute "No gotos" is not a defensable position. I'll bet you like i = 0; <<VAX some VAX assembly statements XAV>> j=0; more C Andy Tanenbaum (ast@cs.vu.nl)
cs304pal@rata.vuw.ac.nz (Lloyd Parkes) (10/02/89)
In article <24962@louie.udel.EDU> tweten@gilmore.nas.nasa.gov (Dave Tweten) writes: >I hope you are merely overstating you case, Dr. Tanenbaum. I had >thought the "No gotos" religion had died, mercifully for the rest of >us. I have only been programming in C for a couple of years now, but I have been programming for many years before that. I have written most sorts of programs, including an assembler, a heap management program and a multitude of file handling programs. And I can quite truthfully say that since I left BASIC, I have *NEVER* needed a goto. There is always (I hope) a way to write a piece of code without resorting to gotos. If you can't get a readable program without using a goto, then you need to rethink, it might be a major rethink, but there is always a way. Lloyd Quick, send your money to cs304pal@rata.vuw.ac.nz now! If you think anyone believes what I have just said, then you must be daft in the head!