Core Changes Versus UZI180 WRS

Changes made (base UZI-WRS) before this became git tracked:

• Correct order of cr/lf in putchar()/printk etc
• O_APPEND mode
• Added saving of O_flags beyond r/w/rw
• O_CREAT, O_TRUNC, O_EXCL
• creat now obsolete (frobs and runs open)
• mknod only allows non root user to create pipes
• readi/writei are passed file flag and pass it on
• readi/writei implement non blocking on pipes (minimally)
• rwsetup retrieves access bits for readi/writei to use
• flags are propogated to char devices (but not yet to block devices)
• added uarea.u_root and ref/derefs for it on fork/exit/startup
• n_open starts from udata.u_root and also checks for /.. through root
• modified execve() to honour O_CLOEXEC (sort of) (CLOEXEC should be per fd not per file handle!: FIXME)
• support for O_NDELAY on tty_read (open doesn't do blocking for carrier at all yet, and write is complicated)
• Added fcntl method
• Added u_cloexec to uarea (16bit mask), and made CLOEXEC handlers use it including fcntl, moved CLOEXEC into open only flags.
• Added fchown, fchmod, fchdir
• check if fchown/fchmod are smaller by using shared helpers (yes)
• use a helper for fchdir/chdir/chroot
• add a helper for suser that also sets EPERM
• switch from lib printf to a simple kprintf and kputs etc
• panic tidied to use kputs (also PANIC is now panic in the text)
• added mkdir/rmdir
• forbid unlink of a dir, fix perms checking
• add uf_alloc_base/uf_alloc split, implement F_DUPFD
• added an analyser for size reporting and memory hogs
• return resulting not old position on seek()
• keep a pointer to the offsets we play with in seek (saves 139 bytes)
• add CONFIG_ options for idump and monitor trap (saves over 1K by default and fixes a bug)
• Move long boot up texts into buf[0-2] (cheap trick) (saves 302 bytes)
• All swap* names are now switch* ready for adding a real swapper
• Fix bug in new open() that gave unknown error for opening an unknown file
• Add setpgrp(), and change kill() to honour 0, and -ve for process groups (still need to fix tty ^C to match)
• Fixed "kill 0" semantics
• Added O_NOCTTY to devtty.c and switched to pgrps by changing siggrp to do pgrp not tty matching
• Added a uname that passes 0 seperated strings, return is len so caller can tell if new ones added to ABI
• Added waitpid, and WNOHANG, no support for stopped etc waits
• save p->p_pid before marking slot empty
• dieted sd_init
• added profil syscall pieces, and configurable. Not yet added int handler bits (needs care on mmu mappings)
• Set MAXTICKS to 10 (no reason for switching every tick!)
• implemented the nice syscall
• minimal implementation of uadmin for shutdown/reboot
• extend uname to cover sysconf data and "free" type stuff
• fixed bug introduced in the new tiny kprintf
• fixed memory size in socz80.s
• added rename()
• fix "link" to check only dir permissions and also disallow link to dir
• uninline long shifts in sd_init which bloat the code horribly
• split sd_init parition printing code
• remove fname checks by pushing the "creat /" handlin into ch_link
• clear u_cloexec on exec
• proc interface return 0 for EOF not 1 !!
• remove proc and zero private no-write interfaces and make no_read no_rdwr
• switch to new ugetsys/uputsys for driver buffer copies
• fix error returns on no_ioctl, no_rdwr
• remove progtop uput hacks on minor code
• Make valaddr() handle wraps correctly
• Initial split towards platform/ and generic code
• Fixed optimisation bug introduced in devps
• Allow uname customisation via makefile
• Tweak makeversion to set up uname_len, and make it const
• Fixed bug in devsd SPI init that broke MMC and SDSC cards
• Fix handling of smaller than 32MB final device
• Correct broken permission checks in mount (introdced in esuser mods)
• Correct esuper cases - fixes mkdir with no . and .. and umount bugs
• Clean up fname checking leftovers
• Split the platform dir properly with its own Makefile and moved includes
• Move kprint and friends out of devtty into devio
• Add a sleep/O_NDELAY/signal helper
• NMOUNT so mount table can be smaller than number of devices (can't yet set it differently) [put mount ptr in device table ???]
• fs_tab is device sized pointer table, small mount table of actual filesys objects
• further clean up of shared v private asm code
• get to the point a kernel can be built for either socz80 or z80pack (untested on z80pack!)
• put into git
• UZIX calls driver poll in idle loop rather than halting, implement this choice
• Mini UZI for small boxes with no hd, always runs just the "top" process of the stack (ie fork is fork/sleep). Possibly plus tty irq only mode for some other crap boxes if needed (bg tasks only when fg is in tty waits)
• Clean init died handler
• Add brk() to expand data space within chmem space only
• Adjust allocator so program only get the number of banks they need
• Support better use of memory. New binary type to allow chmem
• Z80pack port needs the bank copier and user copiers done properly
• create a "library" of standard mmu types etc
• split includes further
• clean up assembler common/private bits
• Platform specific mapping
• Introduce 4 bytes of pagemap, pass ptr to pagemap not pagemap data in HL
• Fixed banks, no change Single user - allocate only that space also
• single user swap out in strict parental order and swap just the range specifie d.
• Map small binaries in less pages when non relocatable, map the image at bottom and top repeating. UAREA remains as expected at top, vectors bottom but uses less real pages
• Single user mode with crap swap, instead of relocatables we can ldir binaries up in memory and stack ram from the PROGTOP
  • means PROGTOP becomes a variable
  • means SWAP mapping is not fixed (good thing to sort anyway)
  • stack into upper ram, swap to disk if option, each new app gets
  • 0x0000->the base of the stacked apps
• 30 char names
• Proper major/minor numbers
• uput and efault API
• mapping table for swaps (hand a swap out to users with page1 = swapmap page0=0 meaning swapped). Means we can have a swap where we openswap (check for sig), read the size and set up that many maps. Means swap out can fail but that is fine a s we already cover the out of memory paths it causes. Means we need less swaps as w hile we eat 48K or 64K per task we don't waste any - big deal with large proc tab
• remove doclose hacks on device close occuring each close not final ?
• Add getpagesize() [can be part of a single getsysinfo call with uname etc]
• fix off_t/time_t to be proper values in libc API
• make kernel provide lseek and proper off_t
• make kernel provide proper time_t
• 32 byte directory entries (SYS5 style)
• steal the 'reuse the last buffer exec and file optimisation trick' from OMU