diff -urN v2.4.19-pre5/AIO-NOTES linux.diff/AIO-NOTES --- v2.4.19-pre5/AIO-NOTES Wed Dec 31 19:00:00 1969 +++ linux.diff/AIO-NOTES Tue Apr 2 19:24:02 2002 @@ -0,0 +1,11 @@ +20020102 + - ABI adjustment to io_getevents: the prototype is now + long io_getevents(aio_context_t ctx_id, + long nr, + struct io_event *events, + const struct timespec *timeout) + which does not affect binary compatibility on x86, only + for 64 bit machines. + + + diff -urN v2.4.19-pre5/Documentation/Configure.help linux.diff/Documentation/Configure.help --- v2.4.19-pre5/Documentation/Configure.help Wed Apr 3 21:04:25 2002 +++ linux.diff/Documentation/Configure.help Tue Apr 30 17:29:31 2002 @@ -17983,6 +17983,26 @@ contains more information and the location of the joystick package that you'll need. +/dev/epoll support +CONFIG_EVENTPOLL + This option will allow for the creation of a '/dev/epoll' character + device, with major number 10 (MISC_MAJOR) and minor number 124 + (EVENTPOLL_MINOR). + + This device can be used to very efficiently handle incoming events on a + socket, much more so than select() or poll(). There is a paper that + describes this device and how to program for it (as well as including + some very impressive benchmarks) at the following URL: + http://www.xmailserver.org/linux-patches/nio-improve.html + + If you are writing very scalable servers and wish to code against + /dev/epoll for enhanced speed, say 'Y' or 'M' here. If you have + software in hand that requires (or can make use of) /dev/epoll, + also say 'Y' or 'M' here. + + The vast majority of the planet can very safely say 'N' here + and breathe easily. + Game port support CONFIG_INPUT_GAMEPORT Gameport support is for the standard 15-pin PC gameport. If you diff -urN v2.4.19-pre5/MAINTAINERS linux.diff/MAINTAINERS --- v2.4.19-pre5/MAINTAINERS Wed Apr 3 21:04:25 2002 +++ linux.diff/MAINTAINERS Tue Apr 2 18:56:58 2002 @@ -228,6 +228,12 @@ L: linux-net@vger.kernel.org S: Maintained +ASYNC IO +P: Benjamin LaHaise +M: bcrl@redhat.com +L: linux-aio@kvack.org +S: Maintained + AX.25 NETWORK LAYER P: Matthias Welwarsky M: dg2fef@afthd.tu-darmstadt.de diff -urN v2.4.19-pre5/Makefile linux.diff/Makefile --- v2.4.19-pre5/Makefile Wed Apr 3 21:04:25 2002 +++ linux.diff/Makefile Fri Apr 19 20:57:16 2002 @@ -226,7 +226,7 @@ drivers/sound/pndsperm.c \ drivers/sound/pndspini.c \ drivers/atm/fore200e_*_fw.c drivers/atm/.fore200e_*.fw \ - .version .config* config.in config.old \ + .uniquebytes .version .config* config.in config.old \ scripts/tkparse scripts/kconfig.tk scripts/kconfig.tmp \ scripts/lxdialog/*.o scripts/lxdialog/lxdialog \ .menuconfig.log \ @@ -268,6 +268,7 @@ --end-group \ -o vmlinux $(NM) vmlinux | grep -v '\(compiled\)\|\(\.o$$\)\|\( [aUw] \)\|\(\.\.ng$$\)\|\(LASH[RL]DI\)' | sort > System.map + @$(MAKE) -C ulib symlinks: rm -f include/asm @@ -296,7 +297,7 @@ linuxsubdirs: $(patsubst %, _dir_%, $(SUBDIRS)) -$(patsubst %, _dir_%, $(SUBDIRS)) : dummy include/linux/version.h include/config/MARKER +$(patsubst %, _dir_%, $(SUBDIRS)) : dummy include/linux/compile.h include/config/MARKER $(MAKE) CFLAGS="$(CFLAGS) $(CFLAGS_KERNEL)" -C $(patsubst _dir_%, %, $@) $(TOPDIR)/include/linux/version.h: include/linux/version.h @@ -322,6 +323,11 @@ echo \#define LINUX_COMPILE_DOMAIN ; \ fi >> .ver @echo \#define LINUX_COMPILER \"`$(CC) $(CFLAGS) -v 2>&1 | tail -1`\" >> .ver + @rm -f .uniquebytes + @dd if=/dev/urandom of=.uniquebytes bs=1 count=16 + @echo -n \#"define LINUX_UNIQUE_BYTES " >>.ver + @hexdump -v -e '1/1 "0x%02x, "' .uniquebytes | sed -e 's/, $$//g' >>.ver + @echo "" >>.ver @mv -f .ver $@ include/linux/version.h: ./Makefile @@ -404,6 +410,8 @@ .PHONY: $(patsubst %, _modinst_%, $(SUBDIRS)) $(patsubst %, _modinst_%, $(SUBDIRS)) : $(MAKE) -C $(patsubst _modinst_%, %, $@) modules_install + mkdir -p $(INSTALL_MOD_PATH)/lib/kernel/$(KERNELRELEASE)/ + install -m 755 ulib/libredhat-kernel.so.1.0.1 $(INSTALL_MOD_PATH)/lib/kernel/$(KERNELRELEASE)/ # modules disabled.... @@ -423,6 +431,7 @@ rm -f $(CLEAN_FILES) rm -rf $(CLEAN_DIRS) $(MAKE) -C Documentation/DocBook clean + $(MAKE) -C ulib clean mrproper: clean archmrproper find . \( -size 0 -o -name .depend \) -type f -print | xargs rm -f diff -urN v2.4.19-pre5/arch/i386/Makefile linux.diff/arch/i386/Makefile --- v2.4.19-pre5/arch/i386/Makefile Thu May 3 11:22:07 2001 +++ linux.diff/arch/i386/Makefile Tue Apr 2 18:56:58 2002 @@ -98,7 +98,7 @@ DRIVERS += arch/i386/math-emu/math.o endif -arch/i386/kernel: dummy +arch/i386/kernel: dummy include/linux/compile.h $(MAKE) linuxsubdirs SUBDIRS=arch/i386/kernel arch/i386/mm: dummy diff -urN v2.4.19-pre5/arch/i386/kernel/Makefile linux.diff/arch/i386/kernel/Makefile --- v2.4.19-pre5/arch/i386/kernel/Makefile Mon Nov 26 23:43:07 2001 +++ linux.diff/arch/i386/kernel/Makefile Tue Apr 2 18:56:58 2002 @@ -30,6 +30,10 @@ endif endif +obj-y += vsysdata.o vunique.o dynamic_syscall.o + +vunique.o: $(TOPDIR)/include/linux/compile.h + obj-$(CONFIG_MCA) += mca.o obj-$(CONFIG_MTRR) += mtrr.o obj-$(CONFIG_X86_MSR) += msr.o diff -urN v2.4.19-pre5/arch/i386/kernel/dynamic_syscall.c linux.diff/arch/i386/kernel/dynamic_syscall.c --- v2.4.19-pre5/arch/i386/kernel/dynamic_syscall.c Wed Dec 31 19:00:00 1969 +++ linux.diff/arch/i386/kernel/dynamic_syscall.c Tue Apr 2 18:56:58 2002 @@ -0,0 +1,90 @@ +/* arch/i386/kernel/dynamic_syscall.c + * Entry code for dynamic syscalls on i386. + */ +#include +#include +#include +#include +#include +#include + +struct dummy_args { + long data[8]; +}; + +extern struct vsyscall_entry { + long eip; + long (*call)(struct dummy_args args); +} vsyscall_list_begin, vsyscall_list_end; + +long sys_dynamic_syscall(struct pt_regs regs) __attribute__((regparm(0))); + +long sys_dynamic_syscall(struct pt_regs regs) +{ + struct dummy_args dummy_args; + struct vsyscall_entry *ent = (void *)regs.edx; + void *args = (void *)regs.ecx; + long ret; + + pr_debug("ent = %p args = %p\n", ent, args); + pr_debug("eip = 0x%08lx\n", regs.eip); + + if (unlikely(!current->mm->vsys_mapped)) + goto err; + + /* The pointer must be aligned in the table. */ + if (unlikely((long)ent & (sizeof(*ent) - 1))) { + pr_debug("unaligned\n"); + goto err; + } + + /* Bounds checking... */ + if (unlikely(ent < &vsyscall_list_begin) || + unlikely(ent >= &vsyscall_list_end)) { + pr_debug("out of range %p <= %p < %p\n", + &vsyscall_list_begin, ent, + &vsyscall_list_end); + goto err; + } + /* The entry should be valid now. Verify that the caller's eip + * is correct. + */ + if (unlikely(ent->eip != regs.eip)) { + pr_debug("eip mismatch (0x%lx vs 0x%lx)\n", ent->eip, regs.eip); + goto err; + } + + pr_debug("ent->call = %p\n", ent->call); + + if (unlikely(verify_area(VERIFY_READ, args, sizeof(dummy_args)))) + return -EFAULT; + + __asm__ volatile ( + " cld \n" + " sub $0x20, %%esp \n" + " movl %%esp, %%edi \n" + " movl $0x8, %%ecx \n" + "1: rep movsl \n" + " call %%edx \n" + "2: add $0x20, %%esp \n" + +/* the exception handling: just return -EFAULT */ + ".section .fixup, \"ax\" \n" + "3: movl $0xfffffff2, %%eax \n" /* -EFAULT */ + " jmp 2b \n" + ".previous \n" + ".section __ex_table,\"a\" \n" + " .align 4 \n" + " .long 1b, 3b \n" + ".previous \n" + + : "=a" (ret) + : "S" (args), "d" (ent->call) + : "%edi", "%ecx" ); + + pr_debug("ret = 0x%08lx\n", ret); + + return ret; +err: + return -ENOSYS; +} diff -urN v2.4.19-pre5/arch/i386/kernel/entry.S linux.diff/arch/i386/kernel/entry.S --- v2.4.19-pre5/arch/i386/kernel/entry.S Wed Apr 3 21:04:26 2002 +++ linux.diff/arch/i386/kernel/entry.S Tue Apr 2 18:56:58 2002 @@ -45,6 +45,7 @@ #include #include #include +#include EBX = 0x00 ECX = 0x04 @@ -636,6 +637,11 @@ .long SYMBOL_NAME(sys_ni_syscall) /* reserved for fremovexattr */ .long SYMBOL_NAME(sys_tkill) + .rept __NR_sys_dynamic_syscall-(.-sys_call_table)/4 + .long SYMBOL_NAME(sys_ni_syscall) + .endr + .long SYMBOL_NAME(sys_dynamic_syscall) + .rept NR_syscalls-(.-sys_call_table)/4 .long SYMBOL_NAME(sys_ni_syscall) .endr diff -urN v2.4.19-pre5/arch/i386/kernel/irq.c linux.diff/arch/i386/kernel/irq.c --- v2.4.19-pre5/arch/i386/kernel/irq.c Mon Nov 12 17:49:47 2001 +++ linux.diff/arch/i386/kernel/irq.c Fri Apr 5 18:28:20 2002 @@ -577,7 +577,17 @@ irq_desc_t *desc = irq_desc + irq; struct irqaction * action; unsigned int status; + long esp; + /* Debugging check for stack overflow: is there less than 2KB free? */ + __asm__ __volatile__("andl %%esp,%0" : "=r" (esp) : "0" (8191)); + if (esp < (sizeof(struct task_struct) + 2048)) { + printk("do_IRQ: stack overflow: %ld\n", + esp - sizeof(struct task_struct)); + __asm__ __volatile__("movl %%esp,%0" : "=r" (esp)); + show_stack((void *)esp); + } + kstat.irqs[cpu][irq]++; spin_lock(&desc->lock); desc->handler->ack(irq); diff -urN v2.4.19-pre5/arch/i386/kernel/vsysdata.c linux.diff/arch/i386/kernel/vsysdata.c --- v2.4.19-pre5/arch/i386/kernel/vsysdata.c Wed Dec 31 19:00:00 1969 +++ linux.diff/arch/i386/kernel/vsysdata.c Tue Apr 2 18:56:58 2002 @@ -0,0 +1,11 @@ +/* vsysdata.c - declarations for variables shared with the kernel + * + * Items placed in .data.vsyscall have a kernel virtual address + * and are read/write from kernel space only. The copy placed + * in .vsyscall_data are linked at a userspace address and are + * read only accessible from userland. + */ +#include + +union vsys_union user_vsys_cpudata[256] __attribute__((section(".vsyscall_data"))); +//asm(".globl vsys_cpudata ; bobbob = user_vsys_cpudata - vsyscall_text_begin ; vsys_cpudata = bobbob + VSYSCALL_text"); diff -urN v2.4.19-pre5/arch/i386/kernel/vunique.S linux.diff/arch/i386/kernel/vunique.S --- v2.4.19-pre5/arch/i386/kernel/vunique.S Wed Dec 31 19:00:00 1969 +++ linux.diff/arch/i386/kernel/vunique.S Tue Apr 2 18:56:58 2002 @@ -0,0 +1,7 @@ +#include + + .section .first_vsyscall_text,"xa" + .globl signature +signature: + .byte LINUX_UNIQUE_BYTES + .size signature,.-signature diff -urN v2.4.19-pre5/arch/i386/mm/fault.c linux.diff/arch/i386/mm/fault.c --- v2.4.19-pre5/arch/i386/mm/fault.c Wed Apr 3 21:04:26 2002 +++ linux.diff/arch/i386/mm/fault.c Tue Apr 9 18:03:05 2002 @@ -27,6 +27,8 @@ extern void die(const char *,struct pt_regs *,long); +spinlock_t oops_lock = SPIN_LOCK_UNLOCKED; + /* * Ugly, ugly, but the goto's result in better assembly.. */ @@ -306,7 +308,7 @@ * Oops. The kernel tried to access some bad page. We'll have to * terminate things with extreme prejudice. */ - + spin_lock(&oops_lock); bust_spinlocks(1); if (address < PAGE_SIZE) @@ -327,6 +329,7 @@ } die("Oops", regs, error_code); bust_spinlocks(0); + spin_unlock(&oops_lock); do_exit(SIGKILL); /* diff -urN v2.4.19-pre5/arch/i386/vmlinux.lds linux.diff/arch/i386/vmlinux.lds --- v2.4.19-pre5/arch/i386/vmlinux.lds Thu Mar 7 16:39:56 2002 +++ linux.diff/arch/i386/vmlinux.lds Tue Apr 2 18:56:58 2002 @@ -14,6 +14,27 @@ *(.gnu.warning) } = 0x9090 + /* Note: most of these declarations are in kernel/vsysdata.c,vsyscall.S. + * We use two segments for the data liked at a kernel virtual address + * (.data.vsyscall) and user virtual address (.vsyscall_data). + * .vsyscall_text is linked at a kernel virtual address + */ + . = ALIGN(4096); + VSYSCALL_text = .; + VSYSCALL 0xbfff0000 : AT ( VSYSCALL_text ) { + vsyscall_text_begin = .; + *(.first_vsyscall_text) + *(.vsyscall_text) + . = ALIGN(4096); + vsyscall_text_end = .; + *(.vsyscall_data) + . = ALIGN(4096); + vsyscall_data_end = .; + } + vsys_cpudata = user_vsys_cpudata - vsyscall_text_begin + VSYSCALL_text; + . = VSYSCALL_text + SIZEOF(VSYSCALL); + VSYSCALL_text_end = .; + _etext = .; /* End of text section */ .rodata : { *(.rodata) *(.rodata.*) } @@ -30,6 +51,17 @@ .data : { /* Data */ *(.data) + + . = ALIGN(8); + vsyscall_list_begin = .; + *(.data.vsyscall_list) + vsyscall_list_end = .; + . = ALIGN(4096); + kernel_vsyscall_data_begin = .; + *(.data.vsyscall) + . = ALIGN(4096); + kernel_vsyscall_data_end = .; + CONSTRUCTORS } @@ -79,4 +111,15 @@ .stab.index 0 : { *(.stab.index) } .stab.indexstr 0 : { *(.stab.indexstr) } .comment 0 : { *(.comment) } +/* +# VSYSCALL : { +# / * vsyscall area *i / +# __vsyscall_begin = .; +# *(vsyscall_text) +# . = ALIGN(4096); +# *(.data.vsyscall) +# . = ALIGN(4096); +# __vsyscall_end = .; +# } >vsyscall_area +*/ } diff -urN v2.4.19-pre5/drivers/block/loop.c linux.diff/drivers/block/loop.c --- v2.4.19-pre5/drivers/block/loop.c Wed Apr 3 21:04:30 2002 +++ linux.diff/drivers/block/loop.c Tue Apr 2 18:56:57 2002 @@ -283,7 +283,7 @@ spin_lock_irq(&lo->lo_lock); file = lo->lo_backing_file; spin_unlock_irq(&lo->lo_lock); - do_generic_file_read(file, &pos, &desc, lo_read_actor); + do_generic_file_read(file, &pos, &desc, lo_read_actor, 0); return desc.error; } diff -urN v2.4.19-pre5/drivers/char/Config.in linux.diff/drivers/char/Config.in --- v2.4.19-pre5/drivers/char/Config.in Wed Apr 3 21:04:30 2002 +++ linux.diff/drivers/char/Config.in Tue Apr 30 17:29:31 2002 @@ -220,6 +220,7 @@ dep_tristate 'AMD 768 Random Number Generator support' CONFIG_AMD_RNG $CONFIG_PCI dep_tristate 'Intel i8x0 Random Number Generator support' CONFIG_INTEL_RNG $CONFIG_PCI tristate '/dev/nvram support' CONFIG_NVRAM +tristate '/dev/epoll - Efficent file event polling method' CONFIG_EVENTPOLL tristate 'Enhanced Real Time Clock Support' CONFIG_RTC if [ "$CONFIG_IA64" = "y" ]; then bool 'EFI Real Time Clock Services' CONFIG_EFI_RTC diff -urN v2.4.19-pre5/drivers/char/Makefile linux.diff/drivers/char/Makefile --- v2.4.19-pre5/drivers/char/Makefile Wed Apr 3 21:04:30 2002 +++ linux.diff/drivers/char/Makefile Tue Apr 30 17:29:31 2002 @@ -208,6 +208,7 @@ ifeq ($(CONFIG_PPC),) obj-$(CONFIG_NVRAM) += nvram.o endif +obj-$(CONFIG_EVENTPOLL) += eventpoll.o obj-$(CONFIG_TOSHIBA) += toshiba.o obj-$(CONFIG_I8K) += i8k.o obj-$(CONFIG_DS1620) += ds1620.o diff -urN v2.4.19-pre5/drivers/char/eventpoll.c linux.diff/drivers/char/eventpoll.c --- v2.4.19-pre5/drivers/char/eventpoll.c Wed Dec 31 19:00:00 1969 +++ linux.diff/drivers/char/eventpoll.c Tue Apr 30 17:29:31 2002 @@ -0,0 +1,800 @@ +/* + * drivers/char/eventpoll.c + * + * Copyright (C) 2001, Davide Libenzi + * + * Efficent event polling implementation + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + + + + + +#define DEBUG 0 +#ifdef DEBUG +#define DPRINTK(x) printk x +#define DNPRINTK(n,x) if (n <= DEBUG) printk x +#else +#define DPRINTK(x) +#define DNPRINTK(n,x) +#endif + +#define DEBUG_DPI 0 + +#if DEBUG_DPI +#define DPI_SLAB_DEBUG (SLAB_DEBUG_FREE | SLAB_RED_ZONE /* | SLAB_POISON */) +#else +#define DPI_SLAB_DEBUG 0 +#endif + +#define INITIAL_HASH_BITS 7 +#define MAX_HASH_BITS 18 +#define RESIZE_LENGTH 2 + +#define dpi_mem_alloc() (struct epitem *) kmem_cache_alloc(dpi_cache, SLAB_KERNEL) +#define dpi_mem_free(p) kmem_cache_free(dpi_cache, p) + + + + + +typedef unsigned long long event_version_t; + +struct eventpoll { + rwlock_t lock; + wait_queue_head_t wq; + wait_queue_head_t poll_wait; + struct list_head *hash; + unsigned int hbits; + unsigned int hmask; + atomic_t hents; + atomic_t resize; + int numpages; + char **pages; + char *pages0[MAX_EVENTPOLL_PAGES]; + char *pages1[MAX_EVENTPOLL_PAGES]; + atomic_t mmapped; + int eventcnt; + event_version_t ver; +}; + +struct epitem { + struct list_head llink; + struct eventpoll *ep; + struct file *file; + struct pollfd pfd; + int index; + event_version_t ver; +}; + + + + + + +static int ep_alloc_pages(char **pages, int numpages); +static int ep_free_pages(char **pages, int numpages); +static int ep_init(struct eventpoll *ep); +static void ep_free(struct eventpoll *ep); +static inline struct epitem *ep_find_nl(struct eventpoll *ep, int fd); +static struct epitem *ep_find(struct eventpoll *ep, int fd); +static int ep_hashresize(struct eventpoll *ep, unsigned long *kflags); +static int ep_insert(struct eventpoll *ep, struct pollfd *pfd); +static int ep_remove(struct eventpoll *ep, struct epitem *dpi); +static void notify_proc(struct file *file, void *data, unsigned long *local, long *event); +static int open_eventpoll(struct inode *inode, struct file *file); +static int close_eventpoll(struct inode *inode, struct file *file); +static unsigned int poll_eventpoll(struct file *file, poll_table *wait); +static int write_eventpoll(struct file *file, const char *buffer, size_t count, + loff_t *ppos); +static int ep_poll(struct eventpoll *ep, void *arg); +static int ioctl_eventpoll(struct inode *inode, struct file *file, + unsigned int cmd, unsigned long arg); +static void eventpoll_mm_open(struct vm_area_struct * vma); +static void eventpoll_mm_close(struct vm_area_struct * vma); +static int mmap_eventpoll(struct file *file, struct vm_area_struct *vma); + + + + +static kmem_cache_t *dpi_cache; + +static struct file_operations eventpoll_fops = { + write: write_eventpoll, + ioctl: ioctl_eventpoll, + mmap: mmap_eventpoll, + open: open_eventpoll, + release: close_eventpoll, + poll: poll_eventpoll +}; + +static struct vm_operations_struct eventpoll_mmap_ops = { + open: eventpoll_mm_open, + close: eventpoll_mm_close, +}; + +static struct miscdevice eventpoll = { + EVENTPOLL_MINOR, "eventpoll", &eventpoll_fops +}; + + + + +static int ep_alloc_pages(char **pages, int numpages) +{ + int ii; + + for (ii = 0; ii < numpages; ii++) { + pages[ii] = (char *) __get_free_pages(GFP_KERNEL, 0); + if (!pages[ii]) { + for (--ii; ii >= 0; ii--) { + clear_bit(PG_reserved, &virt_to_page(pages[ii])->flags); + free_pages((unsigned long) pages[ii], 0); + } + return -ENOMEM; + } + set_bit(PG_reserved, &virt_to_page(pages[ii])->flags); + } + return 0; +} + + +static int ep_free_pages(char **pages, int numpages) +{ + int ii; + + for (ii = 0; ii < numpages; ii++) { + clear_bit(PG_reserved, &virt_to_page(pages[ii])->flags); + free_pages((unsigned long) pages[ii], 0); + } + return 0; +} + + +static int ep_init(struct eventpoll *ep) +{ + int ii, hentries; + + rwlock_init(&ep->lock); + init_waitqueue_head(&ep->wq); + init_waitqueue_head(&ep->poll_wait); + ep->hbits = INITIAL_HASH_BITS; + ep->hmask = (1 << ep->hbits) - 1; + atomic_set(&ep->hents, 0); + atomic_set(&ep->resize, 0); + atomic_set(&ep->mmapped, 0); + ep->numpages = 0; + ep->pages = ep->pages0; + ep->eventcnt = 0; + ep->ver = 1; + + hentries = ep->hmask + 1; + if (!(ep->hash = (struct list_head *) vmalloc(hentries * sizeof(struct list_head)))) + return -ENOMEM; + + for (ii = 0; ii < hentries; ii++) + INIT_LIST_HEAD(&ep->hash[ii]); + + return 0; +} + + +static void ep_free(struct eventpoll *ep) +{ + int ii; + struct list_head *lnk; + + lock_kernel(); + for (ii = 0; ii <= ep->hmask; ii++) { + while ((lnk = list_first(&ep->hash[ii]))) { + struct epitem *dpi = list_entry(lnk, struct epitem, llink); + + file_notify_delcb(dpi->file, notify_proc); + list_del(lnk); + dpi_mem_free(dpi); + } + } + vfree(ep->hash); + if (ep->numpages > 0) { + ep_free_pages(ep->pages0, ep->numpages); + ep_free_pages(ep->pages1, ep->numpages); + } + unlock_kernel(); +} + + +static inline struct epitem *ep_find_nl(struct eventpoll *ep, int fd) +{ + struct epitem *dpi = NULL; + struct list_head *lsthead, *lnk; + + lsthead = &ep->hash[fd & ep->hmask]; + list_for_each(lnk, lsthead) { + dpi = list_entry(lnk, struct epitem, llink); + + if (dpi->pfd.fd == fd) break; + dpi = NULL; + } + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ep_find(%d) -> %p\n", current, fd, dpi)); + + return dpi; +} + + +static struct epitem *ep_find(struct eventpoll *ep, int fd) +{ + struct epitem *dpi; + unsigned long flags; + + read_lock_irqsave(&ep->lock, flags); + + dpi = ep_find_nl(ep, fd); + + read_unlock_irqrestore(&ep->lock, flags); + + return dpi; +} + + +static int ep_hashresize(struct eventpoll *ep, unsigned long *kflags) +{ + struct list_head *hash, *oldhash; + unsigned int hbits = ep->hbits + 1; + unsigned int hmask = (1 << hbits) - 1; + int ii, res, hentries = hmask + 1; + unsigned long flags = *kflags; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ep_hashresize(%p) bits=%u\n", current, ep, hbits)); + + write_unlock_irqrestore(&ep->lock, flags); + + res = -ENOMEM; + if (!(hash = (struct list_head *) vmalloc(hentries * sizeof(struct list_head)))) { + write_lock_irqsave(&ep->lock, flags); + goto out; + } + + for (ii = 0; ii < hentries; ii++) + INIT_LIST_HEAD(&hash[ii]); + + write_lock_irqsave(&ep->lock, flags); + + oldhash = ep->hash; + for (ii = 0; ii <= ep->hmask; ii++) { + struct list_head *oldhead = &oldhash[ii], *lnk; + + while ((lnk = list_first(oldhead))) { + struct epitem *dpi = list_entry(lnk, struct epitem, llink); + + list_del(lnk); + list_add(lnk, &hash[dpi->pfd.fd & hmask]); + } + } + + ep->hash = hash; + ep->hbits = hbits; + ep->hmask = hmask; + + write_unlock_irqrestore(&ep->lock, flags); + vfree(oldhash); + write_lock_irqsave(&ep->lock, flags); + + res = 0; +out: + *kflags = flags; + atomic_dec(&ep->resize); + return res; +} + + +static int ep_insert(struct eventpoll *ep, struct pollfd *pfd) +{ + struct epitem *dpi; + struct file *file; + unsigned long flags; + + if (atomic_read(&ep->hents) >= (ep->numpages * POLLFD_X_PAGE)) + return -E2BIG; + + if (!(file = fcheck(pfd->fd))) + return -EINVAL; + + if (!(dpi = dpi_mem_alloc())) + return -ENOMEM; + + INIT_LIST_HEAD(&dpi->llink); + dpi->ep = ep; + dpi->file = file; + dpi->pfd = *pfd; + dpi->index = -1; + dpi->ver = ep->ver - 1; + + write_lock_irqsave(&ep->lock, flags); + + list_add(&dpi->llink, &ep->hash[pfd->fd & ep->hmask]); + atomic_inc(&ep->hents); + + if (!atomic_read(&ep->resize) && + (atomic_read(&ep->hents) >> ep->hbits) > RESIZE_LENGTH && + ep->hbits < MAX_HASH_BITS) { + atomic_inc(&ep->resize); + ep_hashresize(ep, &flags); + } + + write_unlock_irqrestore(&ep->lock, flags); + + file_notify_addcb(file, notify_proc, dpi); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ep_insert(%p, %d)\n", current, ep, pfd->fd)); + + return 0; +} + + +static int ep_remove(struct eventpoll *ep, struct epitem *dpi) +{ + int fd = dpi->pfd.fd; + unsigned long flags; + struct pollfd *pfd, *lpfd; + struct epitem *ldpi; + + file_notify_delcb(dpi->file, notify_proc); + + write_lock_irqsave(&ep->lock, flags); + + list_del(&dpi->llink); + atomic_dec(&ep->hents); + + if (dpi->index >= 0 && dpi->ver == ep->ver && dpi->index < ep->eventcnt) { + pfd = (struct pollfd *) (ep->pages[EVENT_PAGE_INDEX(dpi->index)] + + EVENT_PAGE_OFFSET(dpi->index)); + if (pfd->fd == dpi->pfd.fd && dpi->index < --ep->eventcnt) { + lpfd = (struct pollfd *) (ep->pages[EVENT_PAGE_INDEX(ep->eventcnt)] + + EVENT_PAGE_OFFSET(ep->eventcnt)); + *pfd = *lpfd; + + if ((ldpi = ep_find_nl(ep, pfd->fd))) ldpi->index = dpi->index; + } + } + + write_unlock_irqrestore(&ep->lock, flags); + + dpi_mem_free(dpi); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ep_remove(%p, %d)\n", current, ep, fd)); + + return 0; +} + + +static void notify_proc(struct file *file, void *data, unsigned long *local, long *event) +{ + struct epitem *dpi = (struct epitem *) data; + struct eventpoll *ep = dpi->ep; + struct pollfd *pfd; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: notify(%p, %p, %ld, %ld) ep=%p\n", + current, file, data, event[0], event[1], ep)); + + write_lock(&ep->lock); + if (!(dpi->pfd.events & event[1])) + goto out; + + if (dpi->index < 0 || dpi->ver != ep->ver) { + if (ep->eventcnt >= (ep->numpages * POLLFD_X_PAGE)) + goto out; + dpi->index = ep->eventcnt++; + dpi->ver = ep->ver; + pfd = (struct pollfd *) (ep->pages[EVENT_PAGE_INDEX(dpi->index)] + + EVENT_PAGE_OFFSET(dpi->index)); + *pfd = dpi->pfd; + } else { + pfd = (struct pollfd *) (ep->pages[EVENT_PAGE_INDEX(dpi->index)] + + EVENT_PAGE_OFFSET(dpi->index)); + if (pfd->fd != dpi->pfd.fd) { + if (ep->eventcnt >= (ep->numpages * POLLFD_X_PAGE)) + goto out; + dpi->index = ep->eventcnt++; + pfd = (struct pollfd *) (ep->pages[EVENT_PAGE_INDEX(dpi->index)] + + EVENT_PAGE_OFFSET(dpi->index)); + *pfd = dpi->pfd; + } + } + + pfd->revents |= (pfd->events & event[1]); + + if (waitqueue_active(&ep->wq)) + wake_up(&ep->wq); + if (waitqueue_active(&ep->poll_wait)) + wake_up(&ep->poll_wait); +out: + write_unlock(&ep->lock); +} + + +static int open_eventpoll(struct inode *inode, struct file *file) +{ + int res; + struct eventpoll *ep; + + if (!(ep = kmalloc(sizeof(struct eventpoll), GFP_KERNEL))) + return -ENOMEM; + + memset(ep, 0, sizeof(*ep)); + if ((res = ep_init(ep))) { + kfree(ep); + return res; + } + + file->private_data = ep; + + MOD_INC_USE_COUNT; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: open() ep=%p\n", current, ep)); + return 0; +} + + +static int close_eventpoll(struct inode *inode, struct file *file) +{ + struct eventpoll *ep = file->private_data; + + ep_free(ep); + + kfree(ep); + + MOD_DEC_USE_COUNT; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: close() ep=%p\n", current, ep)); + return 0; +} + + +static unsigned int poll_eventpoll(struct file *file, poll_table *wait) +{ + struct eventpoll *ep = file->private_data; + + poll_wait(file, &ep->poll_wait, wait); + if (ep->eventcnt) + return POLLIN | POLLRDNORM; + + return 0; +} + + +static int write_eventpoll(struct file *file, const char *buffer, size_t count, + loff_t *ppos) +{ + int res, rcount; + struct eventpoll *ep = file->private_data; + struct epitem *dpi; + struct pollfd pfd; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: write(%p, %d)\n", current, ep, count)); + + if (count % sizeof(struct pollfd)) + return -EINVAL; + + if ((res = verify_area(VERIFY_READ, buffer, count))) + return res; + + rcount = 0; + + lock_kernel(); + + while (count > 0) { + __copy_from_user(&pfd, buffer, sizeof(pfd)); + + dpi = ep_find(ep, pfd.fd); + + if (pfd.fd >= current->files->max_fds || !current->files->fd[pfd.fd]) + pfd.events = POLLREMOVE; + if (pfd.events & POLLREMOVE) { + if (dpi) { + ep_remove(ep, dpi); + rcount += sizeof(pfd); + } + } + else if (dpi) { + dpi->pfd.events = pfd.events; + rcount += sizeof(pfd); + } else { + pfd.revents = 0; + if (!ep_insert(ep, &pfd)) + rcount += sizeof(pfd); + } + + buffer += sizeof(pfd); + count -= sizeof(pfd); + } + + unlock_kernel(); + + return rcount; +} + + +static int ep_poll(struct eventpoll *ep, void *arg) +{ + int res = 0; + long timeout; + unsigned long flags; + struct evpoll dvp; + wait_queue_t wait; + + if (copy_from_user(&dvp, arg, sizeof(struct evpoll))) + return -EFAULT; + + if (!atomic_read(&ep->mmapped)) + return -EINVAL; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ioctl(%p, EP_POLL, %d)\n", current, ep, dvp.ep_timeout)); + + write_lock_irqsave(&ep->lock, flags); + + res = 0; + if (!ep->eventcnt) { + init_waitqueue_entry(&wait, current); + add_wait_queue(&ep->wq, &wait); + timeout = dvp.ep_timeout == -1 || dvp.ep_timeout > MAX_SCHEDULE_TIMEOUT/HZ ? + MAX_SCHEDULE_TIMEOUT: (dvp.ep_timeout * HZ) / 1000; + for (;;) { + if (ep->eventcnt || !timeout) + break; + if (signal_pending(current)) { + res = -EINTR; + break; + } + + set_current_state(TASK_INTERRUPTIBLE); + + write_unlock_irqrestore(&ep->lock, flags); + timeout = schedule_timeout(timeout); + write_lock_irqsave(&ep->lock, flags); + } + remove_wait_queue(&ep->wq, &wait); + + set_current_state(TASK_RUNNING); + } + + if (!res && ep->eventcnt) { + res = ep->eventcnt; + ep->eventcnt = 0; + ++ep->ver; + if (ep->pages == ep->pages0) { + ep->pages = ep->pages1; + dvp.ep_resoff = 0; + } else { + ep->pages = ep->pages0; + dvp.ep_resoff = ep->numpages * PAGE_SIZE; + } + } + + write_unlock_irqrestore(&ep->lock, flags); + + if (res > 0) + copy_to_user(arg, &dvp, sizeof(struct evpoll)); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ioctl(%p, EP_POLL, %d) == %d\n", current, ep, dvp.ep_timeout, res)); + return res; +} + + +static int ioctl_eventpoll(struct inode *inode, struct file *file, + unsigned int cmd, unsigned long arg) +{ + int res, numpages; + struct eventpoll *ep = file->private_data; + struct epitem *dpi; + unsigned long flags; + struct pollfd pfd; + + switch (cmd) { + case EP_ALLOC: + if (atomic_read(&ep->mmapped)) + return -EBUSY; + + numpages = EP_FDS_PAGES(arg); + if (numpages > MAX_EVENTPOLL_PAGES) + return -EINVAL; + + res = 0; + write_lock_irqsave(&ep->lock, flags); + if (numpages > ep->numpages) { + if (!(res = ep_alloc_pages(&ep->pages0[ep->numpages], numpages - ep->numpages))) { + if (!(res = ep_alloc_pages(&ep->pages1[ep->numpages], numpages - ep->numpages))) { + ep->numpages = numpages; + } else { + ep_free_pages(&ep->pages0[ep->numpages], numpages - ep->numpages); + } + } + } + write_unlock_irqrestore(&ep->lock, flags); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ioctl(%p, EP_ALLOC, %lu) == %d\n", + current, ep, arg, res)); + return res; + + case EP_FREE: + if (atomic_read(&ep->mmapped)) + return -EBUSY; + + res = -EINVAL; + write_lock_irqsave(&ep->lock, flags); + if (ep->numpages > 0) { + ep_free_pages(ep->pages0, ep->numpages); + ep_free_pages(ep->pages1, ep->numpages); + ep->numpages = 0; + ep->pages = ep->pages0; + res = 0; + } + write_unlock_irqrestore(&ep->lock, flags); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ioctl(%p, EP_FREE) == %d\n", + current, ep, res)); + return res; + + case EP_POLL: + return ep_poll(ep, (void *) arg); + + case EP_ISPOLLED: + if (copy_from_user(&pfd, (void *) arg, sizeof(struct pollfd))) + return 0; + + read_lock_irqsave(&ep->lock, flags); + + res = 0; + if (!(dpi = ep_find_nl(ep, pfd.fd))) + goto out_ispolled; + + pfd = dpi->pfd; + res = 1; + + out_ispolled: + read_unlock_irqrestore(&ep->lock, flags); + + if (res) + copy_to_user((void *) arg, &pfd, sizeof(struct pollfd)); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: ioctl(%p, EP_ISPOLLED, %d) == %d\n", + current, ep, pfd.fd, res)); + return res; + } + + return -EINVAL; +} + + +static void eventpoll_mm_open(struct vm_area_struct * vma) +{ + struct file *file = vma->vm_file; + struct eventpoll *ep = file->private_data; + + if (ep) atomic_inc(&ep->mmapped); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: mm_open(%p)\n", current, ep)); +} + + +static void eventpoll_mm_close(struct vm_area_struct * vma) +{ + struct file *file = vma->vm_file; + struct eventpoll *ep = file->private_data; + + if (ep) atomic_dec(&ep->mmapped); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: mm_close(%p)\n", current, ep)); +} + + +static int mmap_eventpoll(struct file *file, struct vm_area_struct *vma) +{ + struct eventpoll *ep = file->private_data; + unsigned long start, flags; + int ii, res; + int numpages; + size_t mapsize; + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: mmap(%p, %lx, %lx)\n", + current, ep, vma->vm_start, vma->vm_pgoff << PAGE_SHIFT)); + + if ((vma->vm_pgoff << PAGE_SHIFT) != 0) + return -EINVAL; + + mapsize = PAGE_ALIGN(vma->vm_end - vma->vm_start); + numpages = mapsize >> PAGE_SHIFT; + + write_lock_irqsave(&ep->lock, flags); + + res = -EINVAL; + if (numpages != (2 * ep->numpages)) + goto out; + + start = vma->vm_start; + for (ii = 0; ii < ep->numpages; ii++) { + if (remap_page_range(start, __pa(ep->pages0[ii]), + PAGE_SIZE, vma->vm_page_prot)) + goto out; + start += PAGE_SIZE; + } + for (ii = 0; ii < ep->numpages; ii++) { + if (remap_page_range(start, __pa(ep->pages1[ii]), + PAGE_SIZE, vma->vm_page_prot)) + goto out; + start += PAGE_SIZE; + } + vma->vm_ops = &eventpoll_mmap_ops; + atomic_set(&ep->mmapped, 1); + res = 0; +out: + write_unlock_irqrestore(&ep->lock, flags); + + DNPRINTK(3, (KERN_INFO "[%p] /dev/epoll: mmap(%p, %lx, %lx) == %d\n", + current, ep, vma->vm_start, vma->vm_pgoff << PAGE_SHIFT, res)); + return res; +} + + +int __init eventpoll_init(void) +{ + dpi_cache = kmem_cache_create("eventpoll", + sizeof(struct epitem), + __alignof__(struct epitem), + DPI_SLAB_DEBUG, NULL, NULL); + if (!dpi_cache) { + printk(KERN_INFO "[%p] /dev/epoll: driver install failed.\n", current); + return -ENOMEM; + } + + printk(KERN_INFO "[%p] /dev/epoll: driver installed.\n", current); + + misc_register(&eventpoll); + + return 0; +} + + +module_init(eventpoll_init); + +#ifdef MODULE + +void cleanup_module(void) +{ + misc_deregister(&eventpoll); + kmem_cache_destroy(dpi_cache); +} + +#endif + +MODULE_LICENSE("GPL"); + diff -urN v2.4.19-pre5/drivers/char/mem.c linux.diff/drivers/char/mem.c --- v2.4.19-pre5/drivers/char/mem.c Wed Apr 3 21:04:30 2002 +++ linux.diff/drivers/char/mem.c Tue Apr 2 18:56:58 2002 @@ -25,6 +25,7 @@ #include #include #include +#include #ifdef CONFIG_I2C extern int i2c_init_all(void); @@ -572,6 +573,87 @@ write: write_full, }; +void vsys_mmap_close(struct vm_area_struct *area) +{ + area->vm_mm->vsys_mapped = 0; +} + +static struct vm_operations_struct vsys_mmap_ops = { + close: vsys_mmap_close, +}; + +int vsys_mmap(struct file *file, struct vm_area_struct *vma) +{ + extern unsigned char vsyscall_text_begin, vsyscall_text_end, VSYSCALL_text[]; + unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; + unsigned long len = vma->vm_end - vma->vm_start; + unsigned long actual_len = &vsyscall_text_end - &vsyscall_text_begin; + + if ((offset + len) > actual_len) + len = actual_len - offset; + + pr_debug("len = 0x%lx, actual_len = 0x%lx\n", len, actual_len); + + vma->vm_start = (unsigned long)&vsyscall_text_begin + offset; + vma->vm_end = vma->vm_start + len; + vma->vm_flags |= VM_RESERVED; + + pr_debug("vm_start = 0x%lx, vm_end = 0x%lx\n", + vma->vm_start, vma->vm_end); + pr_debug("va=%p pa=0x%lx\n", + VSYSCALL_text + offset, + __pa(VSYSCALL_text) + offset); + + if (vma->vm_start < (unsigned long)&vsyscall_text_begin) { + pr_debug("vsys_mmap: start < begin\n"); + return -EINVAL; + } + + if (vma->vm_end < (unsigned long)&vsyscall_text_begin) { + pr_debug("vsys_mmap: end < begin\n"); + return -EINVAL; + } + + if (vma->vm_end > (unsigned long)&vsyscall_text_end) { + pr_debug("vsys_mmap: end(%lx) > text_end(%p)\n", + vma->vm_end, &vsyscall_text_end); + return -EINVAL; + } + + if (vma->vm_start >= vma->vm_end) { + pr_debug("vsys_mmap: end\n"); + return -EINVAL; + } + + if (find_vma_intersection(current->mm, vma->vm_start, vma->vm_end)) { + pr_debug("vsyscall: mapping collision\n"); + return -EINVAL; + } + + if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == (VM_SHARED | VM_WRITE)) { + pr_debug("vsyscall: attempt to write to mapping\n"); + return -EPERM; + } + + if (remap_page_range(vma->vm_start, + __pa(VSYSCALL_text) + offset, + vma->vm_end-vma->vm_start, + vma->vm_page_prot)) + return -EAGAIN; + + pr_debug("VSYSCALL_text(%p): %02x %02x %02x %02x\n", + VSYSCALL_text, + VSYSCALL_text[0], VSYSCALL_text[1], + VSYSCALL_text[2], VSYSCALL_text[3]); + + current->mm->vsys_mapped = 1; + return 0; +} + +static struct file_operations vsys_fops = { + mmap: vsys_mmap, +}; + static int memory_open(struct inode * inode, struct file * filp) { switch (MINOR(inode->i_rdev)) { @@ -601,6 +683,9 @@ case 9: filp->f_op = &urandom_fops; break; + case 10: + filp->f_op = &vsys_fops; + break; default: return -ENXIO; } @@ -627,7 +712,8 @@ {5, "zero", S_IRUGO | S_IWUGO, &zero_fops}, {7, "full", S_IRUGO | S_IWUGO, &full_fops}, {8, "random", S_IRUGO | S_IWUSR, &random_fops}, - {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops} + {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops}, + {10,"vsys", S_IRUGO, &vsys_fops}, }; int i; diff -urN v2.4.19-pre5/drivers/char/raw.c linux.diff/drivers/char/raw.c --- v2.4.19-pre5/drivers/char/raw.c Mon Sep 24 02:16:03 2001 +++ linux.diff/drivers/char/raw.c Tue Apr 2 18:56:58 2002 @@ -16,6 +16,8 @@ #include #include #include +#include +#include #define dprintk(x...) @@ -34,13 +36,18 @@ int raw_open(struct inode *, struct file *); int raw_release(struct inode *, struct file *); int raw_ctl_ioctl(struct inode *, struct file *, unsigned int, unsigned long); - +int raw_kvec_read(struct file *filp, kvec_cb_t cb, size_t size, loff_t pos); +int raw_kvec_write(struct file *filp, kvec_cb_t cb, size_t size, loff_t pos); static struct file_operations raw_fops = { read: raw_read, write: raw_write, open: raw_open, release: raw_release, + aio_read: generic_file_aio_read, + aio_write: generic_file_aio_write, + kvec_read: raw_kvec_read, + kvec_write: raw_kvec_write, }; static struct file_operations raw_ctl_fops = { @@ -250,7 +257,6 @@ } - ssize_t raw_read(struct file *filp, char * buf, size_t size, loff_t *offp) { @@ -381,3 +387,99 @@ out: return err; } + +static int raw_kvec_rw(struct file *filp, int rw, kvec_cb_t cb, size_t size, loff_t pos); +int raw_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return raw_kvec_rw(file, READ, cb, size, pos); +} + +int raw_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return raw_kvec_rw(file, WRITE, cb, size, pos); +} + +int raw_kvec_rw(struct file *filp, int rw, kvec_cb_t cb, size_t size, loff_t pos) +{ + int err; + unsigned minor; + kdev_t dev; + unsigned long limit, blocknr, blocks; + + unsigned sector_size, sector_bits, sector_mask; + unsigned max_sectors; + unsigned i; + + pr_debug("raw_kvec_rw: %p %d %d %p %d %d %Lu\n", filp, rw, nr, kiovec, flags, size, pos); + /* + * First, a few checks on device size limits + */ + + minor = MINOR(filp->f_dentry->d_inode->i_rdev); + dev = to_kdev_t(raw_devices[minor].binding->bd_dev); + sector_size = raw_devices[minor].sector_size; + sector_bits = raw_devices[minor].sector_bits; + sector_mask = sector_size- 1; + max_sectors = 25000; //KIO_MAX_SECTORS >> (sector_bits - 9); + + if (blk_size[MAJOR(dev)]) + limit = (((loff_t) blk_size[MAJOR(dev)][MINOR(dev)]) << BLOCK_SIZE_BITS) >> sector_bits; + else + limit = INT_MAX; + pr_debug ("raw_kvec_rw: dev %d:%d (+%d)\n", + MAJOR(dev), MINOR(dev), limit); + + /* EOF at the end */ + err = 0; + if (!size || (pos >> sector_bits) == limit) { + pr_debug("raw_kvec_rw: %Lu > %lu, %d\n", pos >> sector_bits, limit, sector_bits); + cb.fn(cb.data, cb.vec, err); + return 0; + } + + /* ENXIO for io beyond the end */ + err = -ENXIO; + if ((pos >> sector_bits) >= limit) { + pr_debug("raw_kvec_rw: %Lu > %lu, %d\n", pos >> sector_bits, limit, sector_bits); + goto out; + } + + err = -EINVAL; + if ((pos < 0) || (pos & sector_mask) || (size & sector_mask)) { + pr_debug("pos(%Ld)/size(%lu) wrong(%d)\n", pos, size, sector_mask); + goto out; + } + + /* Verify that the scatter-gather list is sector aligned. */ + for (i=0; inr; i++) + if ((cb.vec->veclet[i].offset & sector_mask) || + (cb.vec->veclet[i].length & sector_mask)) { + pr_debug("veclet offset/length wrong"); + goto out; + } + + /* + * Split the IO into KIO_MAX_SECTORS chunks, mapping and + * unmapping the single kiobuf as we go to perform each chunk of + * IO. + */ + + blocknr = pos >> sector_bits; + blocks = size >> sector_bits; + if (blocks > max_sectors) + blocks = max_sectors; + if (blocks > limit - blocknr) + blocks = limit - blocknr; + err = -ENXIO; + if (!blocks) { + pr_debug("raw: !blocks %d %ld %ld\n", max_sectors, limit, blocknr); + goto out; + } + + err = brw_kvec_async(rw, cb, dev, blocks, blocknr, sector_bits); +out: + if (err) + printk(KERN_DEBUG "raw_kvec_rw: ret is %d\n", err); + return err; +} + diff -urN v2.4.19-pre5/fs/Makefile linux.diff/fs/Makefile --- v2.4.19-pre5/fs/Makefile Thu Mar 7 16:40:03 2002 +++ linux.diff/fs/Makefile Tue Apr 30 17:29:31 2002 @@ -7,12 +7,12 @@ O_TARGET := fs.o -export-objs := filesystems.o open.o dcache.o buffer.o +export-objs := filesystems.o open.o dcache.o buffer.o fcblist.o mod-subdirs := nls obj-y := open.o read_write.o devices.o file_table.o buffer.o \ super.o block_dev.o char_dev.o stat.o exec.o pipe.o namei.o \ - fcntl.o ioctl.o readdir.o select.o fifo.o locks.o \ + fcntl.o ioctl.o readdir.o select.o fifo.o locks.o fcblist.o \ dcache.o inode.o attr.o bad_inode.o file.o iobuf.o dnotify.o \ filesystems.o namespace.o seq_file.o @@ -22,6 +22,9 @@ obj-y += noquot.o endif +obj-y += aio.o +export-objs += aio.o + subdir-$(CONFIG_PROC_FS) += proc subdir-y += partitions diff -urN v2.4.19-pre5/fs/aio.c linux.diff/fs/aio.c --- v2.4.19-pre5/fs/aio.c Wed Dec 31 19:00:00 1969 +++ linux.diff/fs/aio.c Mon Apr 29 18:04:24 2002 @@ -0,0 +1,1224 @@ +/* fs/aio.c + * An async IO implementation for Linux + * Written by Benjamin LaHaise + * + * Implements an efficient asynchronous io interface. + * + * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + */ +//#define DEBUG 1 + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include + +#if DEBUG > 1 +#define dprintk printk +#else +#define dprintk(x...) do { ; } while (0) +#endif + +/*------ sysctl variables----*/ +unsigned aio_nr; /* current system wide number of aio requests */ +unsigned aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ +unsigned aio_max_size = 0x20000; /* 128KB per chunk */ +unsigned aio_max_pinned; /* set to mem/4 in aio_setup */ +/*----end sysctl variables---*/ + +static kmem_cache_t *kiocb_cachep; +static kmem_cache_t *kioctx_cachep; + +/* tunable. Needs to be added to sysctl. */ +int max_aio_reqs = 0x10000; + +/* Used for rare fput completion. */ +static void aio_fput_routine(void *); +static struct tq_struct fput_tqueue = { + routine: aio_fput_routine, +}; + +static spinlock_t fput_lock = SPIN_LOCK_UNLOCKED; +LIST_HEAD(fput_head); + +/* forward prototypes */ +static void generic_aio_complete_read(void *_iocb, struct kvec *vec, ssize_t res); +static void generic_aio_complete_write(void *_iocb, struct kvec *vec, ssize_t res); + +/* aio_setup + * Creates the slab caches used by the aio routines, panic on + * failure as this is done early during the boot sequence. + */ +static int __init aio_setup(void) +{ + kiocb_cachep = kmem_cache_create("kiocb", sizeof(struct kiocb), + 0, SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!kiocb_cachep) + panic("unable to create kiocb cache\n"); + + kioctx_cachep = kmem_cache_create("kioctx", sizeof(struct kioctx), + 0, SLAB_HWCACHE_ALIGN, NULL, NULL); + if (!kioctx_cachep) + panic("unable to create kioctx cache"); + + aio_max_pinned = num_physpages/4; + + printk(KERN_NOTICE "aio_setup: num_physpages = %u\n", aio_max_pinned); + printk(KERN_NOTICE "aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); + + return 0; +} + +static void ioctx_free_reqs(struct kioctx *ctx) +{ + struct list_head *pos, *next; + list_for_each_safe(pos, next, &ctx->free_reqs) { + struct kiocb *iocb = list_kiocb(pos); + list_del(&iocb->list); + kmem_cache_free(kiocb_cachep, iocb); + } +} + +/* ioctx_alloc + * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. + */ +static struct kioctx *ioctx_alloc(unsigned nr_reqs) +{ + struct kioctx *ctx; + unsigned i; + long size; + + /* Round off to a power of 2. Needed for cheap mask operations */ + for (i=1; i (0x10000000U / sizeof(struct io_event))) || + (nr_reqs > (0x10000000U / sizeof(struct kiocb)))) { + pr_debug("ENOMEM: nr_reqs too high\n"); + return ERR_PTR(-EINVAL); + } + + if (nr_reqs > aio_max_nr) + return ERR_PTR(-EAGAIN); + + ctx = kmem_cache_alloc(kioctx_cachep, GFP_KERNEL); + if (!ctx) + return ERR_PTR(-ENOMEM); + + memset(ctx, 0, sizeof(*ctx)); + ctx->max_reqs = nr_reqs; + ctx->mm = current->mm; + atomic_inc(&ctx->mm->mm_count); + + atomic_set(&ctx->users, 1); + spin_lock_init(&ctx->lock); + spin_lock_init(&ctx->ring_lock); + init_waitqueue_head(&ctx->wait); + + INIT_LIST_HEAD(&ctx->free_reqs); + INIT_LIST_HEAD(&ctx->active_reqs); + + /* Allocate nr_reqs iocbs for io. Free iocbs are on the + * ctx->free_reqs list. When active they migrate to the + * active_reqs list. During completion and cancellation + * the request may temporarily not be on any list. + */ + for (i=0; ikey = i; + iocb->users = 0; + list_add(&iocb->list, &ctx->free_reqs); + } + + /* Compensate for the ring buffer's head/tail overlap entry */ + nr_reqs *= 2; + + size = sizeof(struct aio_ring); + size += sizeof(struct io_event) * nr_reqs; + + /* Try to use alloc_pages first... */ + for (i=0; (PAGE_SIZE << i) < size; i++) + ; + ctx->ring_order = i; + ctx->ring = (void *)__get_free_pages(GFP_KERNEL, i); + + if (!ctx->ring) { + /* vmalloc it is... */ + ctx->ring = vmalloc(size); + if (!ctx->ring) + goto out_freectx; + ctx->ring_was_vmallocd = 1; + } else + ctx->ring_was_vmallocd = 0; + + memset(ctx->ring, 0, size); + ctx->ring_mask = nr_reqs - 1; /* trusted copy */ + ctx->ring->mask = ctx->ring_mask; /* user copy */ + + /* now link into global list. kludge. FIXME */ + br_write_lock(BR_AIO_REQ_LOCK); + if (unlikely(aio_nr + ctx->max_reqs > aio_max_nr)) + goto out_cleanup; + aio_nr += ctx->max_reqs; /* undone by __put_ioctx */ + ctx->ring->id = ctx->user_id = ++current->mm->new_ioctx_id; + ctx->next = current->mm->ioctx_list; + current->mm->ioctx_list = ctx; + br_write_unlock(BR_AIO_REQ_LOCK); + + dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", + ctx, ctx->user_id, current->mm, ctx->ring->mask); + return ctx; + +out_cleanup: + br_write_unlock(BR_AIO_REQ_LOCK); + ctx->max_reqs = 0; /* prevent __put_ioctx from sub'ing aio_nr */ + __put_ioctx(ctx); + return ERR_PTR(-EAGAIN); + +out_freereqs: + ioctx_free_reqs(ctx); +out_freectx: + kmem_cache_free(kioctx_cachep, ctx); + ctx = ERR_PTR(-ENOMEM); + + dprintk("aio: error allocating ioctx %p\n", ctx); + return ctx; +} + +/* aio_cancel_all + * Cancels all outstanding aio requests on an aio context. Used + * when the processes owning a context have all exited to encourage + * the rapid destruction of the kioctx. + */ +static void aio_cancel_all(struct kioctx *ctx) +{ + int (*cancel)(struct kiocb *); + spin_lock_irq(&ctx->lock); + while (!list_empty(&ctx->active_reqs)) { + struct list_head *pos = ctx->active_reqs.next; + struct kiocb *iocb = list_kiocb(pos); + list_del_init(&iocb->list); + cancel = iocb->cancel; + if (cancel) + iocb->users++; + spin_unlock_irq(&ctx->lock); + if (cancel) + cancel(iocb); + spin_lock_irq(&ctx->lock); + } + spin_unlock_irq(&ctx->lock); +} + +/* exit_aio: called when the last user of mm goes away. At this point, + * there is no way for any new requests to be submited or any of the + * io_* syscalls to be called on the context. However, there may be + * outstanding requests which hold references to the context; as they + * go away, they will call put_ioctx and release any pinned memory + * associated with the request (held via struct page * references). + */ +void exit_aio(struct mm_struct *mm) +{ + struct kioctx *ctx = mm->ioctx_list; + mm->ioctx_list = NULL; + while (ctx) { + struct kioctx *next = ctx->next; + ctx->next = NULL; + aio_cancel_all(ctx); + + if (1 != atomic_read(&ctx->users)) + printk(KERN_DEBUG + "exit_aio:ioctx still alive: %d %d %d\n", + atomic_read(&ctx->users), ctx->dead, + ctx->reqs_active); + put_ioctx(ctx); + ctx = next; + } +} + +/* __put_ioctx + * Called when the last user of an aio context has gone away, + * and the struct needs to be freed. + */ +void __put_ioctx(struct kioctx *ctx) +{ + unsigned nr_reqs = ctx->max_reqs; + + mmdrop(ctx->mm); + ctx->mm = NULL; + pr_debug("__put_ioctx: freeing %p\n", ctx); + if (ctx->ring_was_vmallocd) + vfree(ctx->ring); + else + free_pages((unsigned long)ctx->ring, ctx->ring_order); + + ioctx_free_reqs(ctx); + kmem_cache_free(kioctx_cachep, ctx); + + br_write_lock(BR_AIO_REQ_LOCK); + aio_nr -= nr_reqs; + br_write_unlock(BR_AIO_REQ_LOCK); +} + +/* aio_get_req + * Allocate a slot for an aio request. Increments the users count + * of the kioctx so that the kioctx stays around until all requests are + * complete. Returns -EAGAIN if no requests are free. + */ +static inline struct kiocb *__aio_get_req(struct kioctx *ctx) +{ + struct kiocb *req = NULL; + + /* Use cmpxchg instead of spin_lock? */ + spin_lock_irq(&ctx->lock); + if (!list_empty(&ctx->free_reqs) && + (ctx->reqs_active < aio_ring_avail(ctx->ring))) { + req = list_kiocb(ctx->free_reqs.next); + list_del(&req->list); + list_add(&req->list, &ctx->active_reqs); + ctx->reqs_active++; + req->user_obj = NULL; + get_ioctx(ctx); + + if (req->ctx) + BUG(); + req->ctx = ctx; + if (req->users) + BUG(); + req->users = 1; + } + spin_unlock_irq(&ctx->lock); + + return req; +} + +static inline struct kiocb *aio_get_req(struct kioctx *ctx) +{ + struct kiocb *req; + /* Handle a potential starvation case -- should be exceedingly rare as + * requests will be stuck on fput_head only if the aio_fput_routine is + * delayed and the requests were the last user of the struct file. + */ + req = __aio_get_req(ctx); + if (unlikely(NULL == ctx)) { + aio_fput_routine(NULL); + req = __aio_get_req(ctx); + } + return req; +} + +static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) +{ + req->ctx = NULL; + req->filp = NULL; + req->user_obj = NULL; + ctx->reqs_active--; + list_add(&req->list, &ctx->free_reqs); +} + +static void aio_fput_routine(void *data) +{ + spin_lock_irq(&fput_lock); + while (likely(!list_empty(&fput_head))) { + struct kiocb *req = list_kiocb(fput_head.next); + struct kioctx *ctx = req->ctx; + + list_del(&req->list); + spin_unlock_irq(&fput_lock); + + /* Complete the fput */ + __fput(req->filp); + + /* Link the iocb into the context's free list */ + spin_lock_irq(&ctx->lock); + really_put_req(ctx, req); + spin_unlock_irq(&ctx->lock); + + put_ioctx(ctx); + spin_lock_irq(&fput_lock); + } + spin_unlock_irq(&fput_lock); +} + +/* __aio_put_req + * Returns true if this put was the last user of the request. + */ +static int __aio_put_req(struct kioctx *ctx, struct kiocb *req) +{ + dprintk(KERN_DEBUG "aio_put(%p): f_count=%d\n", + req, atomic_read(&req->filp->f_count)); + + req->users --; + if (unlikely(req->users < 0)) + BUG(); + if (req->users) + return 0; + list_del(&req->list); /* remove from active_reqs */ + req->cancel = NULL; + + /* Must be done under the lock to serialise against cancellation. + * Call this aio_fput as it duplicates fput via the fput_tqueue. + */ + if (unlikely(atomic_dec_and_test(&req->filp->f_count))) { + get_ioctx(ctx); + spin_lock(&fput_lock); + list_add(&req->list, &fput_head); + spin_unlock(&fput_lock); + schedule_task(&fput_tqueue); + } else + really_put_req(ctx, req); + return 1; +} + +/* aio_put_req + * Returns true if this put was the last user of the kiocb, + * false if the request is still in use. + */ +int aio_put_req(struct kioctx *ctx, struct kiocb *req) +{ + int ret; + spin_lock_irq(&ctx->lock); + ret = __aio_put_req(ctx, req); + spin_unlock_irq(&ctx->lock); + if (ret) + put_ioctx(ctx); + return ret; +} + +/* Lookup an ioctx id. ioctx_list is lockless for reads. + * FIXME: this is O(n) and is only suitable for development. + */ +static inline struct kioctx *lookup_ioctx(unsigned long ctx_id) +{ + struct kioctx *ioctx; + struct mm_struct *mm; + + br_read_lock(BR_AIO_REQ_LOCK); + mm = current->mm; + for (ioctx = mm->ioctx_list; ioctx; ioctx = ioctx->next) + if (likely(ioctx->user_id == ctx_id && !ioctx->dead)) { + get_ioctx(ioctx); + break; + } + br_read_unlock(BR_AIO_REQ_LOCK); + + return ioctx; +} + +/* aio_complete + * Called when the io request on the given iocb is complete. + * Returns true if this is the last user of the request. The + * only other user of the request can be the cancellation code. + */ +int aio_complete(struct kiocb *iocb, long res, long res2) +{ + struct kioctx *ctx = iocb->ctx; + struct aio_ring *ring = ctx->ring; + struct io_event *event; + unsigned long flags; + unsigned long tail; + int ret; + + /* add a completion event to the ring buffer. + * must be done holding ctx->lock to prevent + * other code from messing with the tail + * pointer since we might be called from irq + * context. + */ + spin_lock_irqsave(&ctx->lock, flags); + + tail = ring->tail; + event = &ring->io_events[tail]; + tail = (tail + 1) & ring->mask; + + event->obj = (u64)(unsigned long)iocb->user_obj; + event->data = iocb->user_data; + event->res = res; + event->res2 = res2; + + dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n", + ctx, tail, iocb, iocb->user_obj, iocb->user_data, res, res2); + + /* after flagging the request as done, we + * must never even look at it again + */ + barrier(); + + ring->tail = tail; + + wmb(); + if (!ring->woke) + ring->woke = 1; + + pr_debug("added to ring %p at [%lu]\n", iocb, tail); + + /* everything turned out well, dispose of the aiocb. */ + ret = __aio_put_req(ctx, iocb); + + spin_unlock_irqrestore(&ctx->lock, flags); + + wake_up(&ctx->wait); + if (ret) + put_ioctx(ctx); + + return ret; +} + +/* aio_read_evt + * Pull an event off of the ioctx's event ring. Returns the number of + * events fetched (0 or 1 ;-) + * FIXME: make this use cmpxchg. + * TODO: make the ringbuffer user mmap()able (requires FIXME). + */ +static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) +{ + struct aio_ring *ring = ioctx->ring; + unsigned long head; + int ret = 0; + + dprintk("in aio_read_evt h%lu t%lu m%lu\n", + (unsigned long)ring->head, (unsigned long)ring->tail, + (unsigned long)ring->mask); + barrier(); + if (ring->head == ring->tail) + goto out; + + spin_lock(&ioctx->ring_lock); + + head = ring->head; + if (head != ring->tail) { + *ent = ring->io_events[head]; + head = (head + 1) & ioctx->ring_mask; + barrier(); + ring->head = head; + ret = 1; + } + spin_unlock(&ioctx->ring_lock); + +out: + dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret, + (unsigned long)ring->head, (unsigned long)ring->tail); + return ret; +} + +struct timeout { + struct timer_list timer; + int timed_out; + wait_queue_head_t wait; +}; + +static void timeout_func(unsigned long data) +{ + struct timeout *to = (struct timeout *)data; + + to->timed_out = 1; + wake_up(&to->wait); +} + +static inline void init_timeout(struct timeout *to) +{ + init_timer(&to->timer); + to->timer.data = (unsigned long)to; + to->timer.function = timeout_func; + to->timed_out = 0; + init_waitqueue_head(&to->wait); +} + +static inline void set_timeout(struct timeout *to, const struct timespec *ts) +{ + unsigned long how_long; + + if (!ts->tv_sec && !ts->tv_nsec) { + to->timed_out = 1; + return; + } + + how_long = ts->tv_sec * HZ; +#define HZ_NS (1000000000 / HZ) + how_long += (ts->tv_nsec + HZ_NS - 1) / HZ_NS; + + to->timer.expires = jiffies + how_long; + add_timer(&to->timer); +} + +static inline void clear_timeout(struct timeout *to) +{ + del_timer_sync(&to->timer); +} + +static int read_events(struct kioctx *ctx, int nr, struct io_event *event, + const struct timespec *timeout) +{ + struct task_struct *tsk = current; + DECLARE_WAITQUEUE(wait, tsk); + DECLARE_WAITQUEUE(to_wait, tsk); + int ret; + int i = 0; + struct io_event ent; + struct timeout to; + + /* needed to zero any padding within an entry (there shouldn't be + * any, but C is fun! + */ + memset(&ent, 0, sizeof(ent)); + ret = 0; + + while (likely(i < nr)) { + ret = aio_read_evt(ctx, &ent); + if (unlikely(ret <= 0)) + break; + + dprintk("read event: %Lx %Lx %Lx %Lx\n", + ent.data, ent.obj, ent.res, ent.res2); + + /* FIXME: split checks in two */ + ret = -EFAULT; + if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { + dprintk("aio: lost an event due to EFAULT.\n"); + break; + } + ret = 0; + + /* Good, event copied to userland, update counts. */ + event ++; + i ++; + } + + if (i) + return i; + if (ret) + return ret; + + /* End fast path */ + + init_timeout(&to); + if (timeout) { + struct timespec ts; + ret = -EFAULT; + if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) + goto out; + + set_timeout(&to, &ts); + if (to.timed_out) + timeout = 0; + } + + while (likely(i < nr)) { + add_wait_queue_exclusive_lifo(&ctx->wait, &wait); + add_wait_queue(&to.wait, &to_wait); + do { + set_task_state(tsk, TASK_INTERRUPTIBLE); + + ret = aio_read_evt(ctx, &ent); + if (ret) + break; + if (i) + break; + ret = 0; + if (to.timed_out) /* Only check after read evt */ + break; + schedule(); + if (signal_pending(tsk)) { + ret = -EINTR; + break; + } + /*ret = aio_read_evt(ctx, &ent);*/ + } while (1) ; + + set_task_state(tsk, TASK_RUNNING); + remove_wait_queue(&ctx->wait, &wait); + remove_wait_queue(&to.wait, &to_wait); + + if (unlikely(ret <= 0)) + break; + + ret = -EFAULT; + if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { + dprintk("aio: lost an event due to EFAULT.\n"); + break; + } + + /* Good, event copied to userland, update counts. */ + event ++; + i ++; + } + + if (timeout) + clear_timeout(&to); +out: + return i ? i : ret; +} + +/* Take an ioctx and remove it from the list of ioctx's. Protects + * against races with itself via ->dead. + */ +static void io_destroy(struct kioctx *ioctx) +{ + struct kioctx **tmp; + int was_dead; + + /* delete the entry from the list is someone else hasn't already */ + br_write_lock(BR_AIO_REQ_LOCK); + was_dead = ioctx->dead; + ioctx->dead = 1; + for (tmp = ¤t->mm->ioctx_list; *tmp && *tmp != ioctx; + tmp = &(*tmp)->next) + ; + if (*tmp) + *tmp = ioctx->next; + br_write_unlock(BR_AIO_REQ_LOCK); + + dprintk("aio_release(%p)\n", ioctx); + put_ioctx(ioctx); /* once for the lookup */ + if (likely(!was_dead)) + put_ioctx(ioctx); /* twice for the list */ +} + +asmlinkage long sys_io_setup(unsigned nr_reqs, aio_context_t *ctxp) +{ + struct kioctx *ioctx = NULL; + unsigned long ctx; + long ret; + + ret = get_user(ctx, ctxp); + if (unlikely(ret)) + goto out; + + ret = -EINVAL; + if (unlikely(ctx || !nr_reqs || (int)nr_reqs < 0)) { + pr_debug("EINVAL: io_setup: ctx or nr_reqs > max\n"); + goto out; + } + + ret = -EAGAIN; + if (unlikely(nr_reqs > max_aio_reqs)) + goto out; + + ioctx = ioctx_alloc(nr_reqs); + ret = PTR_ERR(ioctx); + if (!IS_ERR(ioctx)) { + ret = put_user(ioctx->user_id, ctxp); + if (!ret) + return 0; + io_destroy(ioctx); + } + +out: + return ret; +} + +/* aio_release + * Release the kioctx associated with the userspace handle. + */ +asmlinkage long sys_io_destroy(aio_context_t ctx) +{ + struct kioctx *ioctx = lookup_ioctx(ctx); + if (likely(NULL != ioctx)) { + io_destroy(ioctx); + return 0; + } + pr_debug("EINVAL: io_destroy: invalid context id\n"); + return -EINVAL; +} + +int generic_aio_poll(struct file *file, struct kiocb *req, struct iocb iocb) +{ + unsigned events = iocb.aio_buf; + + /* Did the user set any bits they weren't supposed to? (The + * above is actually a cast. + */ + if (unlikely(events != iocb.aio_buf)) + return -EINVAL; + + return async_poll(req, events); +} + +/* sys_io_submit + * Copy an aiocb from userspace into kernel space, then convert it to + * a kiocb, submit and repeat until done. Error codes on copy/submit + * only get returned for the first aiocb copied as otherwise the size + * of aiocbs copied is returned (standard write sematics). + */ +asmlinkage long sys_io_submit(aio_context_t ctx_id, long nr, struct iocb **iocbpp) +{ + struct kioctx *ctx; + long ret = 0; + int i; + + if (unlikely(nr < 0)) + return -EINVAL; + + ctx = lookup_ioctx(ctx_id); + if (unlikely(!ctx)) { + pr_debug("EINVAL: io_submit: invalid context id\n"); + return -EINVAL; + } + + for (i=0; ifilp = file; + tmp.aio_key = req->key; + ret = put_user(tmp.aio_key, &iocbp->aio_key); + if (unlikely(ret)) { + dprintk("EFAULT: aio_key\n"); + goto out_put_req; + } + + req->user_obj = iocbp; + req->user_data = tmp.aio_data; + req->buf = tmp.aio_buf; + req->pos = tmp.aio_offset; + req->size = tmp.aio_nbytes; + + switch (tmp.aio_lio_opcode) { + case IOCB_CMD_PREAD: + op = file->f_op->aio_read; + ret = -EBADF; + if (!(file->f_mode & FMODE_READ)) + goto out_put_req; + break; + case IOCB_CMD_PREADX: + op = file->f_op->aio_readx; + ret = -EBADF; + if (!(file->f_mode & FMODE_READ)) + goto out_put_req; + break; + case IOCB_CMD_PWRITE: + op = file->f_op->aio_write; + ret = -EBADF; + if (!(file->f_mode & FMODE_WRITE)) + goto out_put_req; + break; + case IOCB_CMD_FSYNC: + op = file->f_op->aio_fsync; + break; + //case IOCB_CMD_POLL: + // op = generic_aio_poll; + // break; + default: + op = NULL; + break; + } + + ret = -EINVAL; + if (unlikely(!op)) { + pr_debug("EINVAL: io_submit: no operation provided\n"); + goto out_put_req; + } + + ret = op(file, req, tmp); + if (likely(!ret)) + continue; + + pr_debug("io_submit: op returned %ld\n", ret); + aio_complete(req, ret, 0); + ret = 0; /* A completion event was sent, so + * submit is a success. */ + continue; + + out_put_req: + aio_put_req(ctx, req); + break; + } + + put_ioctx(ctx); + run_task_queue(&tq_disk); + return i ? i : ret; +} + +static void generic_aio_next_chunk(void *_iocb) +{ + int (*kvec_op)(struct file *, kvec_cb_t, size_t, loff_t); + struct kiocb *iocb = _iocb; + int rw = iocb->this_size; + unsigned long buf = iocb->buf; + kvec_cb_t cb; + ssize_t res; + + iocb->this_size = iocb->size - iocb->nr_transferred; + if (iocb->this_size > aio_max_size) + iocb->this_size = aio_max_size; + + buf += iocb->nr_transferred; + cb.vec = mm_map_user_kvec(iocb->ctx->mm, rw, buf, iocb->this_size); + cb.fn = (rw == READ) ? generic_aio_complete_read + : generic_aio_complete_write; + cb.data = iocb; + + dprintk("generic_aio_rw: cb.vec=%p\n", cb.vec); + if (unlikely(IS_ERR(cb.vec))) + goto done; + + kvec_op = (rw == READ) ? iocb->filp->f_op->kvec_read + : iocb->filp->f_op->kvec_write; + dprintk("submit: %d %d %d\n", iocb->this_size, iocb->nr_transferred, iocb->size); + res = kvec_op(iocb->filp, cb, iocb->this_size, + iocb->pos + iocb->nr_transferred); + if (!res) { + dprintk("submit okay\n"); + return; + } + dprintk("submit failed: %d\n", res); + + cb.fn(cb.data, cb.vec, res); + return; + +done: + if (!iocb->nr_transferred) + BUG(); + aio_complete(iocb, iocb->nr_transferred, 0); +} + +static void generic_aio_complete_rw(int rw, void *_iocb, struct kvec *vec, ssize_t res) +{ + struct kiocb *iocb = _iocb; + + unmap_kvec(vec, rw == READ); + free_kvec(vec); + + if (res > 0) + iocb->nr_transferred += res; + + /* Was this chunk successful? Is there more left to transfer? */ + if (res == iocb->this_size && iocb->nr_transferred < iocb->size) { + /* We may be in irq context, so queue processing in + * process context. + */ + iocb->this_size = rw; + INIT_TQUEUE(&iocb->tq, generic_aio_next_chunk, iocb); + schedule_task(&iocb->tq); + return; + } + + aio_complete(iocb, iocb->nr_transferred ? iocb->nr_transferred : res, + 0); +} + +static void generic_aio_complete_read(void *_iocb, struct kvec *vec, ssize_t res) +{ + generic_aio_complete_rw(READ, _iocb, vec, res); +} + +static void generic_aio_complete_write(void *_iocb, struct kvec *vec, ssize_t res) +{ + generic_aio_complete_rw(WRITE, _iocb, vec, res); +} + +ssize_t generic_aio_rw(int rw, struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size) +{ + int (*kvec_op)(struct file *, kvec_cb_t, size_t, loff_t); + unsigned long buf = iocb.aio_buf; + size_t size = iocb.aio_nbytes; + size_t nr_read = 0; + loff_t pos = iocb.aio_offset; + kvec_cb_t cb; + ssize_t res; + +#if 0 + if (likely(NULL != file->f_op->new_read)) { + nr_read = file->f_op->new_read(file, (void *)buf, size, + &pos, F_ATOMIC); + dprintk("from new_read: nr_read: %ld\n", (long)nr_read); + if ((-EAGAIN == nr_read) || (-EWOULDBLOCKIO == nr_read)) + nr_read = 0; + else if ((nr_read >= min_size) || (nr_read < 0)) { + dprintk("returning nr_read: %ld\n", (long)nr_read); + return nr_read; + } + } + dprintk("nr_read: %ld\n", (long)nr_read); +#endif + + req->nr_transferred = nr_read; + size -= nr_read; + if (size > aio_max_size) + /* We have to split up the request. Pin the mm + * struct for further use with map_user_kvec later. + */ + size = aio_max_size; + else + req->buf = 0; + + req->this_size = size; + + buf += nr_read; + cb.vec = map_user_kvec(rw, buf, size); + cb.fn = (rw == READ) ? generic_aio_complete_read + : generic_aio_complete_write; + cb.data = req; + + dprintk("generic_aio_rw: cb.vec=%p\n", cb.vec); + if (IS_ERR(cb.vec)) + return nr_read ? nr_read : PTR_ERR(cb.vec); + + kvec_op = (rw == READ) ? file->f_op->kvec_read : file->f_op->kvec_write; + + res = kvec_op(file, cb, size, pos); + if (unlikely(res != 0)) { + /* If the first chunk was successful, we have to run + * the callback to attempt the rest of the io. + */ + if (res == size && req->buf) { + cb.fn(cb.data, cb.vec, res); + return 0; + } + + unmap_kvec(cb.vec, rw == READ); + free_kvec(cb.vec); + if (nr_read) { + if (res < 0) + res = 0; + res += nr_read; + } + } + return res; +} + +ssize_t generic_file_aio_read(struct file *file, struct kiocb *req, struct iocb iocb) +{ + return generic_aio_rw(READ, file, req, iocb, iocb.aio_nbytes); +} + +ssize_t generic_sock_aio_read(struct file *file, struct kiocb *req, struct iocb iocb) +{ + return generic_aio_rw(READ, file, req, iocb, 1); +} + +ssize_t generic_aio_write(struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size) +{ + return generic_aio_rw(WRITE, file, req, iocb, 1); +#if 0 + unsigned long buf = iocb.aio_buf; + size_t size = iocb.aio_nbytes; + loff_t pos = iocb.aio_offset; + ssize_t nr_written = 0; + kvec_cb_t cb; + long res; +#if 0 + if (likely(NULL != file->f_op->new_write)) { + nr_written = file->f_op->new_write(file, (void *)buf, size, + &pos, F_ATOMIC); + pr_debug("generic_aio_write: new_write: %ld\n", (long)nr_written); + if (-EAGAIN == nr_written) + nr_written = 0; + if ((nr_written >= min_size) || (nr_written < 0)) + return nr_written; + } +#endif + + req->nr_transferred = nr_written; + size -= nr_written; + if (size > aio_max_size) + size = aio_max_size; + req->this_size = size; + buf += nr_written; + cb.vec = map_user_kvec(WRITE, buf, size); + cb.fn = generic_aio_complete_write; + cb.data = req; + + if (IS_ERR(cb.vec)) { + pr_debug("generic_aio_write: map_user_kvec: %ld\n", PTR_ERR(cb.vec)); + return nr_written ? nr_written : PTR_ERR(cb.vec); + } + + res = file->f_op->kvec_write(file, cb, size, iocb.aio_offset); + pr_debug("generic_aio_write: kvec_write: %ld\n", res); + if (unlikely(res != 0)) { + unmap_kvec(cb.vec, 0); + free_kvec(cb.vec); + if (nr_written) { + if (res < 0) + res = 0; + res += nr_written; + } + } + return res; +#endif +} + +ssize_t generic_file_aio_write(struct file *file, struct kiocb *req, struct iocb iocb) +{ + return generic_aio_write(file, req, iocb, iocb.aio_nbytes); +} + +/* lookup_kiocb + * Finds a given iocb for cancellation. + * MUST be called with ctx->lock held. + */ +struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb *iocb, u32 key) +{ + struct list_head *pos; + /* TODO: use a hash or array, this sucks. */ + list_for_each(pos, &ctx->free_reqs) { + struct kiocb *kiocb = list_kiocb(pos); + if (kiocb->user_obj == iocb && kiocb->key == key) + return kiocb; + } + return NULL; +} + +asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb *iocb) +{ + int (*cancel)(struct kiocb *iocb); + struct kioctx *ctx; + struct kiocb *kiocb; + u32 key; + int ret; + + ret = get_user(key, &iocb->aio_key); + if (unlikely(ret)) + return ret; + + ctx = lookup_ioctx(ctx_id); + if (unlikely(!ctx)) + return -EINVAL; + + spin_lock_irq(&ctx->lock); + ret = -EAGAIN; + kiocb = lookup_kiocb(ctx, iocb, key); + if (kiocb && kiocb->cancel) { + cancel = kiocb->cancel; + kiocb->users ++; + } else + cancel = NULL; + spin_unlock_irq(&ctx->lock); + + if (NULL != cancel) + ret = cancel(kiocb); + + put_ioctx(ctx); + + return ret; +} + +asmlinkage long sys_io_wait(aio_context_t ctx_id, struct iocb *iocb, const struct timespec *timeout) +{ +#if 0 /* FIXME. later. */ + struct kioctx *ioctx; + long ret = -EINVAL; + unsigned key; + long obj = (long)iocb; + + ioctx = lookup_ioctx(ctx_id); + if (!ioctx) + goto out; + + ret = get_user(key, &iocb->aio_key); + if (ret) + goto out; + + ret = __aio_complete(ioctx, key, obj, !!timeout); + put_ioctx(ioctx); + +out: + return ret; +#endif + return -ENOSYS; +} + +asmlinkage long sys_io_getevents(aio_context_t ctx_id, + long nr, + struct io_event *events, + const struct timespec *timeout) +{ + struct kioctx *ioctx = lookup_ioctx(ctx_id); + long ret = -EINVAL; + + if (likely(NULL != ioctx)) { + ret = read_events(ioctx, nr, events, timeout); + put_ioctx(ioctx); + } + + return ret; +} + +__initcall(aio_setup); + +add_dynamic_syscall(sys_io_setup); +add_dynamic_syscall(sys_io_destroy); +add_dynamic_syscall(sys_io_submit); +add_dynamic_syscall(sys_io_cancel); +add_dynamic_syscall(sys_io_wait); +add_dynamic_syscall(sys_io_getevents); +EXPORT_SYMBOL_GPL(generic_file_kvec_read); +EXPORT_SYMBOL_GPL(generic_file_aio_read); +EXPORT_SYMBOL_GPL(generic_file_kvec_write); +EXPORT_SYMBOL_GPL(generic_file_aio_write); +EXPORT_SYMBOL_GPL(generic_file_new_read); diff -urN v2.4.19-pre5/fs/buffer.c linux.diff/fs/buffer.c --- v2.4.19-pre5/fs/buffer.c Wed Apr 3 21:04:36 2002 +++ linux.diff/fs/buffer.c Tue Apr 2 18:56:57 2002 @@ -3084,3 +3084,220 @@ module_init(bdflush_init) +/* async kio interface */ +struct brw_cb { + kvec_cb_t cb; + atomic_t io_count; + int nr; + struct buffer_head *bh[1]; +}; + +static inline void brw_cb_put(struct brw_cb *brw_cb) +{ + if (atomic_dec_and_test(&brw_cb->io_count)) { + ssize_t res = 0, err = 0; + int nr; + + /* Walk the buffer heads associated with this kiobuf + * checking for errors and freeing them as we go. + */ + for (nr=0; nr < brw_cb->nr; nr++) { + struct buffer_head *bh = brw_cb->bh[nr]; + if (!err && buffer_uptodate(bh)) + res += bh->b_size; + else + err = -EIO; + kmem_cache_free(bh_cachep, bh); + } + + if (!res) + res = err; + + brw_cb->cb.fn(brw_cb->cb.data, brw_cb->cb.vec, res); + + kfree(brw_cb); + } +} + +/* + * IO completion routine for a buffer_head being used for kiobuf IO: we + * can't dispatch the kiobuf callback until io_count reaches 0. + */ + +static void end_buffer_io_kiobuf_async(struct buffer_head *bh, int uptodate) +{ + struct brw_cb *brw_cb; + + mark_buffer_uptodate(bh, uptodate); + + brw_cb = bh->b_private; + unlock_buffer(bh); + + brw_cb_put(brw_cb); +} + + +/* + * Start I/O on a physical range of kernel memory, defined by a vector + * of kiobuf structs (much like a user-space iovec list). + * + * The kiobuf must already be locked for IO. IO is submitted + * asynchronously: you need to check page->locked, page->uptodate, and + * maybe wait on page->wait. + * + * It is up to the caller to make sure that there are enough blocks + * passed in to completely map the iobufs to disk. + */ + +int brw_kvec_async(int rw, kvec_cb_t cb, kdev_t dev, unsigned blocks, unsigned long blknr, int sector_shift) +{ + struct kvec *vec = cb.vec; + struct kveclet *veclet; + int err; + int length; + unsigned sector_size = 1 << sector_shift; + int i; + + struct brw_cb *brw_cb; + + if (!vec->nr) + BUG(); + + /* + * First, do some alignment and validity checks + */ + length = 0; + for (veclet=vec->veclet, i=0; i < vec->nr; i++,veclet++) { + length += veclet->length; + if ((veclet->offset & (sector_size-1)) || + (veclet->length & (sector_size-1))) { + printk("brw_kiovec_async: tuple[%d]->offset=0x%x length=0x%x sector_size: 0x%x\n", i, veclet->offset, veclet->length, sector_size); + return -EINVAL; + } + } + + if (length < (blocks << sector_shift)) + BUG(); + + /* + * OK to walk down the iovec doing page IO on each page we find. + */ + err = 0; + + if (!blocks) { + printk("brw_kiovec_async: !i\n"); + return -EINVAL; + } + + /* FIXME: tie into userbeans here */ + brw_cb = kmalloc(sizeof(*brw_cb) + (blocks * sizeof(struct buffer_head *)), GFP_KERNEL); + if (!brw_cb) + return -ENOMEM; + + brw_cb->cb = cb; + brw_cb->nr = 0; + + /* This is ugly. FIXME. */ + for (i=0, veclet=vec->veclet; inr; i++,veclet++) { + struct page *page = veclet->page; + unsigned offset = veclet->offset; + unsigned length = veclet->length; + + if (!page) + BUG(); + + while (length > 0) { + struct buffer_head *tmp; + tmp = kmem_cache_alloc(bh_cachep, GFP_NOIO); + err = -ENOMEM; + if (!tmp) + goto error; + + tmp->b_dev = B_FREE; + tmp->b_size = sector_size; + set_bh_page(tmp, page, offset); + tmp->b_this_page = tmp; + + init_buffer(tmp, end_buffer_io_kiobuf_async, NULL); + tmp->b_dev = dev; + tmp->b_blocknr = blknr++; + tmp->b_state = (1 << BH_Mapped) | (1 << BH_Lock) + | (1 << BH_Req); + tmp->b_private = brw_cb; + + if (rw == WRITE) { + set_bit(BH_Uptodate, &tmp->b_state); + clear_bit(BH_Dirty, &tmp->b_state); + } + + brw_cb->bh[brw_cb->nr++] = tmp; + length -= sector_size; + offset += sector_size; + + if (offset >= PAGE_SIZE) { + offset = 0; + break; + } + + if (brw_cb->nr >= blocks) + goto submit; + } /* End of block loop */ + } /* End of page loop */ + +submit: + atomic_set(&brw_cb->io_count, brw_cb->nr+1); + /* okay, we've setup all our io requests, now fire them off! */ + for (i=0; inr; i++) + submit_bh(rw, brw_cb->bh[i]); + brw_cb_put(brw_cb); + + return 0; + +error: + /* Walk brw_cb_table freeing all the goop associated with each kiobuf */ + if (brw_cb) { + /* We got an error allocating the bh'es. Just free the current + buffer_heads and exit. */ + for (i=0; inr; i++) + kmem_cache_free(bh_cachep, brw_cb->bh[i]); + kfree(brw_cb); + } + + return err; +} +#if 0 +int brw_kiovec(int rw, int nr, struct kiobuf *iovec[], + kdev_t dev, int nr_blocks, unsigned long b[], int sector_size) +{ + int i; + int transferred = 0; + int err = 0; + + if (!nr) + return 0; + + /* queue up and trigger the io */ + err = brw_kiovec_async(rw, nr, iovec, dev, nr_blocks, b, sector_size); + if (err) + goto out; + + /* wait on the last iovec first -- it's more likely to finish last */ + for (i=nr; --i >= 0; ) + kiobuf_wait_for_io(iovec[i]); + + run_task_queue(&tq_disk); + + /* okay, how much data actually got through? */ + for (i=0; ierrno) { + if (!err) + err = iovec[i]->errno; + break; + } + transferred += iovec[i]->length; + } + +out: + return transferred ? transferred : err; +} +#endif diff -urN v2.4.19-pre5/fs/exec.c linux.diff/fs/exec.c --- v2.4.19-pre5/fs/exec.c Wed Apr 3 21:04:36 2002 +++ linux.diff/fs/exec.c Mon Apr 29 15:54:22 2002 @@ -397,6 +397,7 @@ old_mm = current->mm; if (old_mm && atomic_read(&old_mm->mm_users) == 1) { mm_release(); + exit_aio(old_mm); exit_mmap(old_mm); return 0; } diff -urN v2.4.19-pre5/fs/ext2/file.c linux.diff/fs/ext2/file.c --- v2.4.19-pre5/fs/ext2/file.c Thu Nov 1 16:40:02 2001 +++ linux.diff/fs/ext2/file.c Sun Apr 7 18:47:48 2002 @@ -40,6 +40,8 @@ */ struct file_operations ext2_file_operations = { llseek: generic_file_llseek, + kvec_read: generic_file_kvec_read, + kvec_write: generic_file_kvec_write, read: generic_file_read, write: generic_file_write, ioctl: ext2_ioctl, @@ -47,6 +49,8 @@ open: generic_file_open, release: ext2_release_file, fsync: ext2_sync_file, + aio_read: generic_file_aio_read, + aio_write: generic_file_aio_write, }; struct inode_operations ext2_file_inode_operations = { diff -urN v2.4.19-pre5/fs/ext3/file.c linux.diff/fs/ext3/file.c --- v2.4.19-pre5/fs/ext3/file.c Mon Nov 26 23:43:08 2001 +++ linux.diff/fs/ext3/file.c Sun Apr 7 18:47:59 2002 @@ -78,6 +78,8 @@ struct file_operations ext3_file_operations = { llseek: generic_file_llseek, /* BKL held */ + kvec_read: generic_file_kvec_read, + kvec_write: generic_file_kvec_write, /* FIXME: attributes */ read: generic_file_read, /* BKL not held. Don't need */ write: ext3_file_write, /* BKL not held. Don't need */ ioctl: ext3_ioctl, /* BKL held */ @@ -85,6 +87,8 @@ open: ext3_open_file, /* BKL not held. Don't need */ release: ext3_release_file, /* BKL not held. Don't need */ fsync: ext3_sync_file, /* BKL held */ + aio_read: generic_file_aio_read, + aio_write: generic_file_aio_write, }; struct inode_operations ext3_file_inode_operations = { diff -urN v2.4.19-pre5/fs/fcblist.c linux.diff/fs/fcblist.c --- v2.4.19-pre5/fs/fcblist.c Wed Dec 31 19:00:00 1969 +++ linux.diff/fs/fcblist.c Tue Apr 30 17:29:31 2002 @@ -0,0 +1,130 @@ +/* + * linux/fs/fcblist.c + * + * Copyright (C) 2001, Davide Libenzi + * + * Handle file callbacks + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + + +long ion_band_table[NSIGPOLL] = { + ION_IN, /* POLL_IN */ + ION_OUT, /* POLL_OUT */ + ION_IN, /* POLL_MSG */ + ION_ERR, /* POLL_ERR */ + 0, /* POLL_PRI */ + ION_HUP /* POLL_HUP */ +}; +EXPORT_SYMBOL(ion_band_table); + +long poll_band_table[NSIGPOLL] = { + POLLIN | POLLRDNORM, /* POLL_IN */ + POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */ + POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */ + POLLERR, /* POLL_ERR */ + POLLPRI | POLLRDBAND, /* POLL_PRI */ + POLLHUP | POLLERR /* POLL_HUP */ +}; +EXPORT_SYMBOL(poll_band_table); + + +void file_notify_event(struct file *filep, long *event) +{ + unsigned long flags; + struct list_head *lnk, *lsthead; + + fcblist_read_lock(filep, flags); + + lsthead = &filep->f_cblist; + list_for_each(lnk, lsthead) { + struct fcb_struct *fcbp = list_entry(lnk, struct fcb_struct, llink); + + fcbp->cbproc(filep, fcbp->data, fcbp->local, event); + } + + fcblist_read_unlock(filep, flags); +} +EXPORT_SYMBOL(file_notify_event); + + +int file_notify_addcb(struct file *filep, + void (*cbproc)(struct file *, void *, unsigned long *, long *), void *data) +{ + unsigned long flags; + struct fcb_struct *fcbp; + + if (!(fcbp = (struct fcb_struct *) kmalloc(sizeof(struct fcb_struct), GFP_KERNEL))) + return -ENOMEM; + + memset(fcbp, 0, sizeof(struct fcb_struct)); + fcbp->cbproc = cbproc; + fcbp->data = data; + + fcblist_write_lock(filep, flags); + list_add_tail(&fcbp->llink, &filep->f_cblist); + fcblist_write_unlock(filep, flags); + + return 0; +} +EXPORT_SYMBOL(file_notify_addcb); + + +int file_notify_delcb(struct file *filep, + void (*cbproc)(struct file *, void *, unsigned long *, long *)) +{ + unsigned long flags; + struct list_head *lnk, *lsthead; + + fcblist_write_lock(filep, flags); + + lsthead = &filep->f_cblist; + list_for_each(lnk, lsthead) { + struct fcb_struct *fcbp = list_entry(lnk, struct fcb_struct, llink); + + if (fcbp->cbproc == cbproc) { + list_del(lnk); + fcblist_write_unlock(filep, flags); + kfree(fcbp); + return 0; + } + } + + fcblist_write_unlock(filep, flags); + + return -ENOENT; +} +EXPORT_SYMBOL(file_notify_delcb); + + +void file_notify_cleanup(struct file *filep) +{ + unsigned long flags; + struct list_head *lnk, *lsthead; + + fcblist_write_lock(filep, flags); + + lsthead = &filep->f_cblist; + while ((lnk = list_first(lsthead))) { + struct fcb_struct *fcbp = list_entry(lnk, struct fcb_struct, llink); + + list_del(lnk); + fcblist_write_unlock(filep, flags); + kfree(fcbp); + fcblist_write_lock(filep, flags); + } + + fcblist_write_unlock(filep, flags); +} +EXPORT_SYMBOL(file_notify_cleanup); + diff -urN v2.4.19-pre5/fs/file_table.c linux.diff/fs/file_table.c --- v2.4.19-pre5/fs/file_table.c Mon Sep 24 02:16:04 2001 +++ linux.diff/fs/file_table.c Tue Apr 30 17:29:31 2002 @@ -8,6 +8,7 @@ #include #include #include +#include #include #include #include @@ -47,6 +48,7 @@ f->f_uid = current->fsuid; f->f_gid = current->fsgid; list_add(&f->f_list, &anon_list); + file_notify_init(f); file_list_unlock(); return f; } @@ -91,6 +93,7 @@ filp->f_uid = current->fsuid; filp->f_gid = current->fsgid; filp->f_op = dentry->d_inode->i_fop; + file_notify_init(filp); if (filp->f_op->open) return filp->f_op->open(dentry->d_inode, filp); else @@ -99,31 +102,36 @@ void fput(struct file * file) { + if (atomic_dec_and_test(&file->f_count)) + __fput(file); +} + +void __fput(struct file * file) +{ struct dentry * dentry = file->f_dentry; struct vfsmount * mnt = file->f_vfsmnt; struct inode * inode = dentry->d_inode; - if (atomic_dec_and_test(&file->f_count)) { - locks_remove_flock(file); + file_notify_cleanup(file); + locks_remove_flock(file); - if (file->f_iobuf) - free_kiovec(1, &file->f_iobuf); + if (file->f_iobuf) + free_kiovec(1, &file->f_iobuf); - if (file->f_op && file->f_op->release) - file->f_op->release(inode, file); - fops_put(file->f_op); - if (file->f_mode & FMODE_WRITE) - put_write_access(inode); - file_list_lock(); - file->f_dentry = NULL; - file->f_vfsmnt = NULL; - list_del(&file->f_list); - list_add(&file->f_list, &free_list); - files_stat.nr_free_files++; - file_list_unlock(); - dput(dentry); - mntput(mnt); - } + if (file->f_op && file->f_op->release) + file->f_op->release(inode, file); + fops_put(file->f_op); + if (file->f_mode & FMODE_WRITE) + put_write_access(inode); + file_list_lock(); + file->f_dentry = NULL; + file->f_vfsmnt = NULL; + list_del(&file->f_list); + list_add(&file->f_list, &free_list); + files_stat.nr_free_files++; + file_list_unlock(); + dput(dentry); + mntput(mnt); } struct file * fget(unsigned int fd) diff -urN v2.4.19-pre5/fs/locks.c linux.diff/fs/locks.c --- v2.4.19-pre5/fs/locks.c Thu Nov 1 16:40:02 2001 +++ linux.diff/fs/locks.c Mon Apr 8 16:46:00 2002 @@ -440,7 +440,7 @@ while (!list_empty(&blocker->fl_block)) { struct file_lock *waiter = list_entry(blocker->fl_block.next, struct file_lock, fl_block); - if (wait) { + if (0) { locks_notify_blocked(waiter); /* Let the blocked process remove waiter from the * block list when it gets scheduled. diff -urN v2.4.19-pre5/fs/nfs/file.c linux.diff/fs/nfs/file.c --- v2.4.19-pre5/fs/nfs/file.c Thu Mar 7 16:40:04 2002 +++ linux.diff/fs/nfs/file.c Tue Apr 2 18:56:58 2002 @@ -39,9 +39,13 @@ static ssize_t nfs_file_write(struct file *, const char *, size_t, loff_t *); static int nfs_file_flush(struct file *); static int nfs_fsync(struct file *, struct dentry *dentry, int datasync); +static int nfs_kvec_write(struct file *file, kvec_cb_t cb, size_t count, loff_t pos); +static int nfs_kvec_read(struct file *file, kvec_cb_t cb, size_t count, loff_t pos); struct file_operations nfs_file_operations = { llseek: generic_file_llseek, + kvec_read: nfs_kvec_read, + kvec_write: nfs_kvec_write, read: nfs_file_read, write: nfs_file_write, mmap: nfs_file_mmap, @@ -50,6 +54,8 @@ release: nfs_release, fsync: nfs_fsync, lock: nfs_lock, + aio_read: generic_file_aio_read, + aio_write: generic_file_aio_write, }; struct inode_operations nfs_file_inode_operations = { @@ -88,6 +94,28 @@ return status; } +static int nfs_kvec_write(struct file *file, kvec_cb_t cb, size_t count, loff_t pos) +{ + struct dentry * dentry = file->f_dentry; + struct inode * inode = dentry->d_inode; + int ret; + ret = nfs_revalidate_inode(NFS_SERVER(inode), inode); + if (!ret) + return generic_file_kvec_write(file, cb, count, pos); + return ret; +} + +static int nfs_kvec_read(struct file *file, kvec_cb_t cb, size_t count, loff_t pos) +{ + struct dentry * dentry = file->f_dentry; + struct inode * inode = dentry->d_inode; + int ret; + ret = nfs_revalidate_inode(NFS_SERVER(inode), inode); + if (!ret) + return generic_file_kvec_read(file, cb, count, pos); + return ret; +} + static ssize_t nfs_file_read(struct file * file, char * buf, size_t count, loff_t *ppos) { diff -urN v2.4.19-pre5/fs/pipe.c linux.diff/fs/pipe.c --- v2.4.19-pre5/fs/pipe.c Wed Apr 3 21:04:37 2002 +++ linux.diff/fs/pipe.c Tue Apr 30 17:29:31 2002 @@ -10,6 +10,7 @@ #include #include #include +#include #include #include @@ -40,6 +41,7 @@ pipe_read(struct file *filp, char *buf, size_t count, loff_t *ppos) { struct inode *inode = filp->f_dentry->d_inode; + int pfull; ssize_t size, read, ret; /* Seeks are not allowed on pipes. */ @@ -72,6 +74,7 @@ PIPE_WAITING_READERS(*inode)++; pipe_wait(inode); PIPE_WAITING_READERS(*inode)--; + pfull = PIPE_FULL(*inode); ret = -ERESTARTSYS; if (signal_pending(current)) goto out; @@ -82,6 +85,8 @@ goto out; } } + else + pfull = PIPE_FULL(*inode); /* Read what data is available. */ ret = -EFAULT; @@ -104,6 +109,9 @@ count -= chars; buf += chars; } + /* Send notification message */ + if (pfull && !PIPE_FULL(*inode) && PIPE_WRITEFILE(*inode)) + file_send_notify(PIPE_WRITEFILE(*inode), ION_OUT, POLLOUT | POLLWRNORM | POLLWRBAND); /* Cache behaviour optimization */ if (!PIPE_LEN(*inode)) @@ -138,6 +146,7 @@ pipe_write(struct file *filp, const char *buf, size_t count, loff_t *ppos) { struct inode *inode = filp->f_dentry->d_inode; + int pempty; ssize_t free, written, ret; /* Seeks are not allowed on pipes. */ @@ -182,6 +191,7 @@ } /* Copy into available space. */ + pempty = PIPE_EMPTY(*inode); ret = -EFAULT; while (count > 0) { int space; @@ -210,6 +220,9 @@ break; do { + /* Send notification message */ + if (pempty && !PIPE_EMPTY(*inode) && PIPE_READFILE(*inode)) + file_send_notify(PIPE_READFILE(*inode), ION_IN, POLLIN | POLLRDNORM); /* * Synchronous wake-up: it knows that this process * is going to give up this CPU, so it doesnt have @@ -219,6 +232,7 @@ PIPE_WAITING_WRITERS(*inode)++; pipe_wait(inode); PIPE_WAITING_WRITERS(*inode)--; + pempty = PIPE_EMPTY(*inode); if (signal_pending(current)) goto out; if (!PIPE_READERS(*inode)) @@ -227,6 +241,9 @@ ret = -EFAULT; } + /* Send notification message */ + if (pempty && !PIPE_EMPTY(*inode) && PIPE_READFILE(*inode)) + file_send_notify(PIPE_READFILE(*inode), ION_IN, POLLIN | POLLRDNORM); /* Signal readers asynchronously that there is more data. */ wake_up_interruptible(PIPE_WAIT(*inode)); @@ -299,9 +316,22 @@ static int pipe_release(struct inode *inode, int decr, int decw) { + struct file *rdfile, *wrfile; down(PIPE_SEM(*inode)); PIPE_READERS(*inode) -= decr; PIPE_WRITERS(*inode) -= decw; + rdfile = PIPE_READFILE(*inode); + wrfile = PIPE_WRITEFILE(*inode); + if (decr && !PIPE_READERS(*inode)) { + PIPE_READFILE(*inode) = NULL; + if (wrfile) + file_send_notify(wrfile, ION_HUP, POLLHUP); + } + if (decw && !PIPE_WRITERS(*inode)) { + PIPE_WRITEFILE(*inode) = NULL; + if (rdfile) + file_send_notify(rdfile, ION_HUP, POLLHUP); + } if (!PIPE_READERS(*inode) && !PIPE_WRITERS(*inode)) { struct pipe_inode_info *info = inode->i_pipe; inode->i_pipe = NULL; @@ -454,6 +484,7 @@ PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 0; PIPE_WAITING_READERS(*inode) = PIPE_WAITING_WRITERS(*inode) = 0; PIPE_RCOUNTER(*inode) = PIPE_WCOUNTER(*inode) = 1; + PIPE_READFILE(*inode) = PIPE_WRITEFILE(*inode) = NULL; return inode; fail_page: @@ -561,6 +592,9 @@ f2->f_mode = 2; f2->f_version = 0; + PIPE_READFILE(*inode) = f1; + PIPE_WRITEFILE(*inode) = f2; + fd_install(i, f1); fd_install(j, f2); fd[0] = i; diff -urN v2.4.19-pre5/fs/select.c linux.diff/fs/select.c --- v2.4.19-pre5/fs/select.c Mon Sep 24 02:16:05 2001 +++ linux.diff/fs/select.c Thu Apr 11 16:17:49 2002 @@ -12,6 +12,12 @@ * 24 January 2000 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). + * June 2001 + * Added async_poll implementation. -bcrl + * Nov 2001 + * Async poll improvments from Suparna Bhattacharya + * April 2002 + * smp safe async poll plus cancellation. -bcrl */ #include @@ -19,6 +25,8 @@ #include #include /* for STICKY_TIMEOUTS */ #include +#include +#include #include @@ -26,19 +34,36 @@ #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM) struct poll_table_entry { - struct file * filp; wait_queue_t wait; wait_queue_head_t * wait_address; + struct file * filp; + poll_table * p; }; struct poll_table_page { + unsigned long size; struct poll_table_page * next; struct poll_table_entry * entry; struct poll_table_entry entries[0]; }; #define POLL_TABLE_FULL(table) \ - ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) + ((unsigned long)((table)->entry+1) > \ + (table)->size + (unsigned long)(table)) + +/* async poll uses only one entry per poll table as it is linked to an iocb */ +typedef struct async_poll_table_struct { + poll_table pt; + struct worktodo wtd; + int events; /* event mask for async poll */ + int wake; + long sync; + struct poll_table_page pt_page; /* one poll table page hdr */ + struct poll_table_entry entries[1]; /* space for a single entry */ +} async_poll_table; + + +static kmem_cache_t *async_poll_table_cache; /* * Ok, Peter made a complicated, but straightforward multiple_wait() function. @@ -61,6 +86,8 @@ struct poll_table_page *old; entry = p->entry; + if (entry == p->entries) /* may happen with async poll */ + break; do { entry--; remove_wait_queue(entry->wait_address,&entry->wait); @@ -68,8 +95,98 @@ } while (entry > p->entries); old = p; p = p->next; - free_page((unsigned long) old); + if (old->size == PAGE_SIZE) + free_page((unsigned long) old); } + if (pt->iocb) + kmem_cache_free(async_poll_table_cache, pt); +} + +void async_poll_complete(void *data) +{ + async_poll_table *pasync = data; + poll_table *p = data; + struct kiocb *iocb = p->iocb; + unsigned int mask; + + pasync->wake = 0; + wmb(); + do { + mask = iocb->filp->f_op->poll(iocb->filp, p); + mask &= pasync->events | POLLERR | POLLHUP; + if (mask) { + poll_freewait(p); + aio_complete(iocb, mask, 0); + return; + } + pasync->sync = 0; + wmb(); + } while (pasync->wake); +} + +static void async_poll_waiter(wait_queue_t *wait) +{ + struct poll_table_entry *entry = (struct poll_table_entry *)wait; + async_poll_table *pasync = (async_poll_table *)(entry->p); + + /* avoid writes to the cacheline if possible for SMP */ + if (!pasync->wake) { + pasync->wake = 1; + /* ensure only one wake up queues the wtd */ + if (!pasync->sync && !test_and_set_bit(0, &pasync->sync)) + wtd_queue(&pasync->wtd); + } +} + +int async_poll_cancel(struct kiocb *iocb) +{ + poll_table *p = iocb->data; + iocb->cancel = NULL; + //wtd_queue(); + return -EAGAIN; +} + +int async_poll(struct kiocb *iocb, int events) +{ + unsigned int mask; + async_poll_table *pasync; + poll_table *p; + + pasync = kmem_cache_alloc(async_poll_table_cache, SLAB_KERNEL); + if (!pasync) + return -ENOMEM; + + p = (poll_table *)pasync; + poll_initwait(p); + wtd_set_action(&pasync->wtd, async_poll_complete, pasync); + p->iocb = iocb; + pasync->wake = 0; + pasync->sync = 0; + pasync->events = events; + pasync->pt_page.entry = pasync->pt_page.entries; + pasync->pt_page.size = sizeof(pasync->pt_page); + p->table = &pasync->pt_page; + + iocb->data = p; + wmb(); + iocb->cancel = async_poll_cancel; + + mask = DEFAULT_POLLMASK; +#warning broken + iocb->users ++; + if (iocb->filp->f_op && iocb->filp->f_op->poll) + mask = iocb->filp->f_op->poll(iocb->filp, p); + mask &= events | POLLERR | POLLHUP; + if (mask && !test_and_set_bit(0, &pasync->sync)) + aio_complete(iocb, mask, 0); + + if (aio_put_req(iocb->ctx, iocb)) + /* Must be freed after aio_complete to synchronise with + * cancellation of the request. + */ + poll_freewait(p); + + return 0; } void __pollwait(struct file * filp, wait_queue_head_t * wait_address, poll_table *p) @@ -85,6 +202,7 @@ __set_current_state(TASK_RUNNING); return; } + new_table->size = PAGE_SIZE; new_table->entry = new_table->entries; new_table->next = table; p->table = new_table; @@ -98,7 +216,11 @@ get_file(filp); entry->filp = filp; entry->wait_address = wait_address; - init_waitqueue_entry(&entry->wait, current); + entry->p = p; + if (p->iocb) /* async poll */ + init_waitqueue_func_entry(&entry->wait, async_poll_waiter); + else + init_waitqueue_entry(&entry->wait, current); add_wait_queue(wait_address,&entry->wait); } } @@ -494,3 +616,14 @@ poll_freewait(&table); return err; } + +static int __init async_poll_init(void) +{ + async_poll_table_cache = kmem_cache_create("async poll table", + sizeof(async_poll_table), 0, 0, NULL, NULL); + if (!async_poll_table_cache) + panic("unable to alloc poll_table_cache"); + return 0; +} + +module_init(async_poll_init); diff -urN v2.4.19-pre5/include/asm-i386/a.out.h linux.diff/include/asm-i386/a.out.h --- v2.4.19-pre5/include/asm-i386/a.out.h Fri Jun 16 14:33:06 1995 +++ linux.diff/include/asm-i386/a.out.h Tue Apr 2 18:56:58 2002 @@ -19,7 +19,9 @@ #ifdef __KERNEL__ -#define STACK_TOP TASK_SIZE +#define VSYSCALL_SIZE 0x10000 /* 64KB for vsyscalls */ +#define STACK_GUARD_SIZE 0x02000 /* 8KB guard area */ +#define STACK_TOP (TASK_SIZE - VSYSCALL_SIZE - STACK_GUARD_SIZE) #endif diff -urN v2.4.19-pre5/include/asm-i386/poll.h linux.diff/include/asm-i386/poll.h --- v2.4.19-pre5/include/asm-i386/poll.h Thu Jan 23 14:01:28 1997 +++ linux.diff/include/asm-i386/poll.h Tue Apr 30 17:29:31 2002 @@ -15,6 +15,7 @@ #define POLLWRNORM 0x0100 #define POLLWRBAND 0x0200 #define POLLMSG 0x0400 +#define POLLREMOVE 0x1000 struct pollfd { int fd; diff -urN v2.4.19-pre5/include/asm-i386/unistd.h linux.diff/include/asm-i386/unistd.h --- v2.4.19-pre5/include/asm-i386/unistd.h Wed Apr 3 21:04:38 2002 +++ linux.diff/include/asm-i386/unistd.h Tue Apr 2 18:56:58 2002 @@ -245,6 +245,8 @@ #define __NR_tkill 238 +#define __NR_sys_dynamic_syscall 250 + /* user-visible error numbers are in the range -1 - -124: see */ #define __syscall_return(type, res) \ diff -urN v2.4.19-pre5/include/asm-i386/vsyscall.h linux.diff/include/asm-i386/vsyscall.h --- v2.4.19-pre5/include/asm-i386/vsyscall.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/asm-i386/vsyscall.h Mon Apr 22 11:49:21 2002 @@ -0,0 +1,42 @@ +#ifndef __ASM__VSYSCALL_H +#define __ASM__VSYSCALL_H +/* include/asm-i386/vsyscall.h + * Copyright 2002 Red Hat, Inc. + */ +#include +#include + +/* We call sys_dynamic_syscall(long nr, void *args) using regparm(2) + * convention. The .text.vsyscall section is mapped into userspace, + * whereas .data.vsyscall_list is a kernel-only array of the vsyscalls + * and the valid userspace address to call them from. All vsyscalls + * are called with C calling convention (ie args on the stack for x86). + * + * Note: the layout of .data.vsyscall_list must match the entries in + * dynamic_syscall.c. + */ +#define STRINGIFYa(x) #x +#define STRINGIFY(x) STRINGIFYa(x) +#define NR_dyn_sys STRINGIFY(__NR_sys_dynamic_syscall) +#define add_dynamic_syscall(name) \ + __asm__(" \n\ + .section .vsyscall_text, \"xa\" \n\ + .globl v" #name " \n\ + v" #name ": \n\ + push %ecx \n\ + push %edx \n\ + movl $" NR_dyn_sys ",%eax \n\ + movl $2f,%edx \n\ + leal 12(%esp),%ecx \n\ + int $0x80 \n\ + 1: \n\ + popl %edx \n\ + popl %ecx \n\ + ret \n\ + .size v" #name ",.-v" #name " \n\ + .previous \n\ + .section .data.vsyscall_list,\"a\" \n\ + 2: .long 1b," #name " \n\ + .previous") + +#endif diff -urN v2.4.19-pre5/include/linux/aio.h linux.diff/include/linux/aio.h --- v2.4.19-pre5/include/linux/aio.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/aio.h Tue Apr 30 17:56:55 2002 @@ -0,0 +1,97 @@ +#ifndef __LINUX__AIO_H +#define __LINUX__AIO_H + +#include +#include +#include +#include + +#define AIO_MAXSEGS 4 +#define AIO_KIOGRP_NR_ATOMIC 8 + +struct kioctx; + +/* Notes on cancelling a kiocb: + * If a kiocb is cancelled, aio_complete may return 0 to indicate + * that cancel has not yet disposed of the kiocb. All cancel + * operations *must* call aio_put_req to dispose of the kiocb + * to guard against races with the completion code. + */ +#define KIOCB_C_CANCELLED 0x01 +#define KIOCB_C_COMPLETE 0x02 + +struct kiocb { + struct list_head list; + struct file *filp; + struct kioctx *ctx; + void *user_obj; + __u64 user_data; + loff_t pos; + unsigned long buf; + size_t nr_transferred; /* used for chunking */ + size_t size; + size_t this_size; + unsigned key; /* id of this request */ + int (*cancel)(struct kiocb *kiocb); + void *data; /* for use by the the async op */ + int users; + struct tq_struct tq; /* argh. */ +}; + +#define AIO_RING_PAGES 8 +#define AIO_RING_PAGES 8 +struct kioctx { + atomic_t users; + int dead; + struct mm_struct *mm; + + /* This needs improving */ + unsigned long user_id; + struct kioctx *next; + + wait_queue_head_t wait; + + spinlock_t lock; + + int reqs_active; + struct list_head free_reqs; + struct list_head active_reqs; /* used for cancellation */ + + unsigned max_reqs; + unsigned ring_mask; + struct aio_ring *ring; + spinlock_t ring_lock; + int ring_was_vmallocd; + int ring_order; +}; + +extern struct file_operations aio_fops; + +extern int FASTCALL(aio_put_req(struct kioctx *ctx, struct kiocb *iocb)); +extern int FASTCALL(aio_complete(struct kiocb *iocb, long res, long res2)); +extern void __put_ioctx(struct kioctx *ctx); +struct mm_struct; +extern void exit_aio(struct mm_struct *mm); + +#define get_ioctx(kioctx) do { if (unlikely(atomic_read(&(kioctx)->users) <= 0)) BUG(); atomic_inc(&(kioctx)->users); } while (0) +#define put_ioctx(kioctx) do { if (unlikely(atomic_dec_and_test(&(kioctx)->users))) __put_ioctx(kioctx); else if (unlikely(atomic_read(&(kioctx)->users) < 0)) BUG(); } while (0) + +#include + +static inline struct kiocb *list_kiocb(struct list_head *h) +{ + return list_entry(h, struct kiocb, list); +} + +struct file; +extern int generic_aio_poll(struct file *file, struct kiocb *req, struct iocb iocb); +extern ssize_t generic_aio_read(struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size); +extern ssize_t generic_aio_write(struct file *file, struct kiocb *req, struct iocb iocb, size_t min_size); +extern ssize_t generic_file_aio_read(struct file *file, struct kiocb *req, struct iocb iocb); +extern ssize_t generic_file_aio_write(struct file *file, struct kiocb *req, struct iocb iocb); +extern ssize_t generic_sock_aio_read(struct file *file, struct kiocb *req, struct iocb iocb); + +/* for sysctl: */ +extern unsigned aio_nr, aio_max_nr, aio_max_size, aio_max_pinned; + +#endif /* __LINUX__AIO_H */ diff -urN v2.4.19-pre5/include/linux/aio_abi.h linux.diff/include/linux/aio_abi.h --- v2.4.19-pre5/include/linux/aio_abi.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/aio_abi.h Thu Apr 11 19:12:13 2002 @@ -0,0 +1,102 @@ +/* linux/aio_abi.h + * + * Copyright 2000,2001,2002 Red Hat. + * + * Written by Benjamin LaHaise + * + * Permission to use, copy, modify, and distribute this software and its + * documentation is hereby granted, provided that the above copyright + * notice appears in all copies. This software is provided without any + * warranty, express or implied. Red Hat makes no representations about + * the suitability of this software for any purpose. + * + * IN NO EVENT SHALL RED HAT BE LIABLE TO ANY PARTY FOR DIRECT, INDIRECT, + * SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OF + * THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF RED HAT HAS BEEN ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + * + * RED HAT DISCLAIMS ANY WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS ON AN "AS IS" BASIS, AND + * RED HAT HAS NO OBLIGATION TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, + * ENHANCEMENTS, OR MODIFICATIONS. + */ +#ifndef __LINUX__AIO_ABI_H +#define __LINUX__AIO_ABI_H + +#include + +typedef unsigned long aio_context_t; + +enum { + IOCB_CMD_PREAD = 0, + IOCB_CMD_PWRITE = 1, + IOCB_CMD_FSYNC = 2, + IOCB_CMD_FDSYNC = 3, + IOCB_CMD_PREADX = 4, + IOCB_CMD_POLL = 5, + IOCB_CMD_NOOP = 6, +}; + +/* read() from /dev/aio returns these structures. */ +struct io_event { + __u64 data; /* the data field from the iocb */ + __u64 obj; /* what iocb this event came from */ + __s64 res; /* result code for this event */ + __s64 res2; /* secondary result */ +}; + +#if defined(__LITTLE_ENDIAN) +#define PADDED(x,y) x, y +#elif defined(__BIG_ENDIAN) +#define PADDED(x,y) y, x +#else +#error edit for your odd byteorder. +#endif + +struct aio_ring { + __u32 PADDED(id, pad1); /* kernel internal index number */ + __u32 PADDED(mask, pad2); /* number of io_events - 1 */ + __u32 PADDED(head, pad3); + __u32 PADDED(tail, pad4); + + __u32 PADDED(woke, pad5); /* set when a wakeup was sent */ + + __u32 pad6[22]; /* pad out to 128 bytes */ + + struct io_event io_events[0]; +}; /* 128 bytes + ring size */ + +#define aio_ring_avail(ring) (((ring)->head + (ring)->mask - (ring)->tail) & (ring)->mask) + +/* + * we always use a 64bit off_t when communicating + * with userland. its up to libraries to do the + * proper padding and aio_error abstraction + */ + +struct iocb { + /* these are internal to the kernel/libc. */ + __u64 aio_data; /* data to be returned in event's data */ + __u32 PADDED(aio_key, aio_reserved1); + /* the kernel sets aio_key to the req # */ + + /* common fields */ + __u16 aio_lio_opcode; /* see IOCB_CMD_ above */ + __s16 aio_reqprio; + __u32 aio_fildes; + + __u64 aio_buf; + __u64 aio_nbytes; + __s64 aio_offset; + + /* extra parameters */ + __u64 aio_reserved2; + __u64 aio_reserved3; +}; /* 64 bytes */ + +#undef IFBIG +#undef IFLITTLE + +#endif /* __LINUX__AIO_ABI_H */ + diff -urN v2.4.19-pre5/include/linux/brlock.h linux.diff/include/linux/brlock.h --- v2.4.19-pre5/include/linux/brlock.h Wed Apr 3 21:10:30 2002 +++ linux.diff/include/linux/brlock.h Thu Apr 11 17:35:53 2002 @@ -34,6 +34,7 @@ enum brlock_indices { BR_GLOBALIRQ_LOCK, BR_NETPROTO_LOCK, + BR_AIO_REQ_LOCK, __BR_END }; diff -urN v2.4.19-pre5/include/linux/errno.h linux.diff/include/linux/errno.h --- v2.4.19-pre5/include/linux/errno.h Tue Nov 6 20:40:27 2001 +++ linux.diff/include/linux/errno.h Tue Apr 2 18:56:57 2002 @@ -21,6 +21,9 @@ #define EBADTYPE 527 /* Type not supported by server */ #define EJUKEBOX 528 /* Request initiated, but will not complete before timeout */ +/* Defined for TUX async IO */ +#define EWOULDBLOCKIO 530 /* Would block due to block-IO */ + #endif #endif diff -urN v2.4.19-pre5/include/linux/eventpoll.h linux.diff/include/linux/eventpoll.h --- v2.4.19-pre5/include/linux/eventpoll.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/eventpoll.h Tue Apr 30 17:29:31 2002 @@ -0,0 +1,43 @@ +/* + * include/linux/eventpoll.h + * + * Copyright (C) 2001, Davide Libenzi + * + * Efficent event polling implementation + */ + + +#ifndef _LINUX_EVENTPOLL_H +#define _LINUX_EVENTPOLL_H + + + + +#define EVENTPOLL_MINOR 124 +#define POLLFD_X_PAGE (PAGE_SIZE / sizeof(struct pollfd)) +#define MAX_FDS_IN_EVENTPOLL (1024 * 128) +#define MAX_EVENTPOLL_PAGES (MAX_FDS_IN_EVENTPOLL / POLLFD_X_PAGE) +#define EVENT_PAGE_INDEX(n) ((n) / POLLFD_X_PAGE) +#define EVENT_PAGE_REM(n) ((n) % POLLFD_X_PAGE) +#define EVENT_PAGE_OFFSET(n) (((n) % POLLFD_X_PAGE) * sizeof(struct pollfd)) +#define EP_FDS_PAGES(n) (((n) + POLLFD_X_PAGE - 1) / POLLFD_X_PAGE) +#define EP_MAP_SIZE(n) (EP_FDS_PAGES(n) * PAGE_SIZE * 2) + + + + + +struct evpoll { + int ep_timeout; + unsigned long ep_resoff; +}; + +#define EP_ALLOC _IOR('P', 1, int) +#define EP_POLL _IOWR('P', 2, struct evpoll) +#define EP_FREE _IO('P', 3) +#define EP_ISPOLLED _IOWR('P', 4, struct pollfd) + + + +#endif + diff -urN v2.4.19-pre5/include/linux/fcblist.h linux.diff/include/linux/fcblist.h --- v2.4.19-pre5/include/linux/fcblist.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/fcblist.h Tue Apr 30 17:29:31 2002 @@ -0,0 +1,67 @@ +/* + * include/linux/fcblist.h + * + * Copyright (C) 2001, Davide Libenzi + * + * Handle file callbacks + */ + +#ifndef __LINUX_FCBLIST_H +#define __LINUX_FCBLIST_H + +#include +#include +#include +#include + + + +/* file callback notification events */ +#define ION_IN 1 +#define ION_OUT 2 +#define ION_HUP 3 +#define ION_ERR 4 + +#define FCB_LOCAL_SIZE 4 + +#define fcblist_read_lock(fp, fl) read_lock_irqsave(&(fp)->f_cblock, fl) +#define fcblist_read_unlock(fp, fl) read_unlock_irqrestore(&(fp)->f_cblock, fl) +#define fcblist_write_lock(fp, fl) write_lock_irqsave(&(fp)->f_cblock, fl) +#define fcblist_write_unlock(fp, fl) write_unlock_irqrestore(&(fp)->f_cblock, fl) + +struct fcb_struct { + struct list_head llink; + void (*cbproc)(struct file *, void *, unsigned long *, long *); + void *data; + unsigned long local[FCB_LOCAL_SIZE]; +}; + + +extern long ion_band_table[]; +extern long poll_band_table[]; + + +void file_notify_event(struct file *filep, long *event); + +int file_notify_addcb(struct file *filep, + void (*cbproc)(struct file *, void *, unsigned long *, long *), void *data); + +int file_notify_delcb(struct file *filep, + void (*cbproc)(struct file *, void *, unsigned long *, long *)); + +void file_notify_cleanup(struct file *filep); + + +static inline void file_notify_init(struct file *filep) +{ + rwlock_init(&filep->f_cblock); + INIT_LIST_HEAD(&filep->f_cblist); +} + +static inline void file_send_notify(struct file *filep, long ioevt, long plevt) { + long event[] = { ioevt, plevt, -1 }; + + file_notify_event(filep, event); +} + +#endif diff -urN v2.4.19-pre5/include/linux/file.h linux.diff/include/linux/file.h --- v2.4.19-pre5/include/linux/file.h Wed Apr 3 21:04:40 2002 +++ linux.diff/include/linux/file.h Tue Apr 2 18:56:57 2002 @@ -5,6 +5,7 @@ #ifndef __LINUX_FILE_H #define __LINUX_FILE_H +extern void FASTCALL(__fput(struct file *)); extern void FASTCALL(fput(struct file *)); extern struct file * FASTCALL(fget(unsigned int fd)); diff -urN v2.4.19-pre5/include/linux/fs.h linux.diff/include/linux/fs.h --- v2.4.19-pre5/include/linux/fs.h Wed Apr 3 21:12:53 2002 +++ linux.diff/include/linux/fs.h Tue Apr 30 17:56:55 2002 @@ -196,6 +196,8 @@ #define FIGETBSZ _IO(0x00,2) /* get the block size used for bmap */ #ifdef __KERNEL__ +#include +#include #include #include @@ -536,6 +538,10 @@ /* needed for tty driver, and maybe others */ void *private_data; + /* file callback list */ + rwlock_t f_cblock; + struct list_head f_cblist; + /* preallocated helper kiobuf to speedup O_DIRECT */ struct kiobuf *f_iobuf; long f_iobuf_lock; @@ -823,6 +829,10 @@ * read, write, poll, fsync, readv, writev can be called * without the big kernel lock held in all filesystems. */ + +#define F_ATOMIC 0x0001 +#define F_OFFSETOK 0x0002 + struct file_operations { struct module *owner; loff_t (*llseek) (struct file *, loff_t, int); @@ -842,6 +852,16 @@ ssize_t (*writev) (struct file *, const struct iovec *, unsigned long, loff_t *); ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int); unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); + + /* in-kernel fully async api */ + int (*kvec_read)(struct file *, kvec_cb_t, size_t, loff_t); + int (*kvec_write)(struct file *, kvec_cb_t, size_t, loff_t); + + /* userland aio ops */ + ssize_t (*aio_read)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_readx)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_write)(struct file *, struct kiocb *, struct iocb); + ssize_t (*aio_fsync)(struct file *, struct kiocb *, struct iocb); }; struct inode_operations { @@ -1420,12 +1440,16 @@ extern int generic_file_mmap(struct file *, struct vm_area_struct *); extern int file_read_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size); extern ssize_t generic_file_read(struct file *, char *, size_t, loff_t *); +extern ssize_t generic_file_new_read(struct file *, char *, size_t, loff_t *, int); extern ssize_t generic_file_write(struct file *, const char *, size_t, loff_t *); -extern void do_generic_file_read(struct file *, loff_t *, read_descriptor_t *, read_actor_t); -extern loff_t no_llseek(struct file *file, loff_t offset, int origin); -extern loff_t generic_file_llseek(struct file *file, loff_t offset, int origin); +extern void do_generic_file_read(struct file *, loff_t *, read_descriptor_t *, read_actor_t, int); +extern int generic_file_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos); +extern int generic_file_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos); + extern ssize_t generic_read_dir(struct file *, char *, size_t, loff_t *); extern int generic_file_open(struct inode * inode, struct file * filp); +extern loff_t no_llseek(struct file *file, loff_t offset, int origin); +extern loff_t generic_file_llseek(struct file *file, loff_t offset, int origin); extern struct file_operations generic_ro_fops; diff -urN v2.4.19-pre5/include/linux/iobuf.h linux.diff/include/linux/iobuf.h --- v2.4.19-pre5/include/linux/iobuf.h Wed Apr 3 21:12:55 2002 +++ linux.diff/include/linux/iobuf.h Tue Apr 30 17:56:55 2002 @@ -53,8 +53,10 @@ /* Dynamic state for IO completion: */ atomic_t io_count; /* IOs still in progress */ + int transferred; /* Number of bytes of completed IO at the beginning of the buffer */ int errno; /* Status of completed IO */ void (*end_io) (struct kiobuf *); /* Completion callback */ + void *end_io_data; wait_queue_head_t wait_queue; }; @@ -80,6 +82,8 @@ /* fs/buffer.c */ +int brw_kiovec_async(int rw, int nr, struct kiobuf *iovec[], + kdev_t dev, int nr_blocks, unsigned long b[], int size); int brw_kiovec(int rw, int nr, struct kiobuf *iovec[], kdev_t dev, unsigned long b[], int size); diff -urN v2.4.19-pre5/include/linux/kiovec.h linux.diff/include/linux/kiovec.h --- v2.4.19-pre5/include/linux/kiovec.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/kiovec.h Tue Apr 30 17:56:55 2002 @@ -0,0 +1,123 @@ +#ifndef __LINUX__KIOVEC_H +#define __LINUX__KIOVEC_H + +struct page; +#include + +struct kveclet { + struct page *page; + unsigned offset; + unsigned length; +}; + +struct kvec { + unsigned max_nr; + unsigned nr; + struct kveclet veclet[0]; +}; + +struct kvec_cb { + struct kvec *vec; + void (*fn)(void *data, struct kvec *vec, ssize_t res); + void *data; +}; + +struct kvec_cb_list { + struct list_head list; + struct kvec_cb cb; +}; + +#ifndef _LINUX_TYPES_H +#include +#endif +#ifndef _LINUX_KDEV_T_H +#include +#endif +#ifndef _ASM_KMAP_TYPES_H +#include +#endif + +extern struct kvec *map_user_kvec(int rw, unsigned long va, size_t len); +extern struct kvec *mm_map_user_kvec(struct mm_struct *, int rw, + unsigned long va, size_t len); +extern void unmap_kvec(struct kvec *, int dirtied); +extern void free_kvec(struct kvec *); + +/* brw_kvec_async: + * Performs direct io to/from disk into cb.vec. Count is the number + * of sectors to read, sector_shift is the blocksize (which must be + * compatible with the kernel's current idea of the device's sector + * size) in log2. blknr is the starting sector offset on dev. + * + */ +extern int brw_kvec_async(int rw, kvec_cb_t cb, kdev_t dev, unsigned count, + unsigned long blknr, int sector_shift); + +/* Memory copy helpers usage: + * void foo(... struct kveclet *veclet...) + * + * struct kvec_dst dst; + * + * kvec_dst_init(&dst, KM_USER0); -- resets type + * kvec_dst_set(&dst, veclet); -- set target & clear offset + * kvec_dst_map(&dst); -- activates kmap + * for (...) + * memcpy_to_kvec_dst(&dst, data, size); -- each copy appends + * kvec_dst_unmap(&dst); -- releases kmap + * + * Note that scheduling is not permitted between kvec_dst_map() and + * kvec_dst_unmap(). This is because internally the routines make use + * of an atomic kmap. + */ +struct kvec_dst { + char *addr; + char *dst; + struct kveclet *let; + int space; + int offset; + enum km_type type; +}; + + +#define kvec_dst_set(Xdst, Xlet) \ + do { \ + struct kvec_dst *_dst = (Xdst); \ + struct kveclet *_let = (Xlet); \ + _dst->let = _let; \ + _dst->space = _let->length; \ + _dst->offset = 0; \ + } while(0) + +#define kvec_dst_map(Xdst) \ + do { \ + struct kvec_dst *_dst = (Xdst); \ + struct kveclet *_let = _dst->let; \ + _dst->dst = _dst->addr = kmap_atomic(_let->page, _dst->type);\ + _dst->dst += _let->offset + _dst->offset; \ + _dst->space = _let->length - _dst->offset; \ + _dst->offset = 0; \ + } while(0) + +#define kvec_dst_init(Xdst, Xtype) \ + do { \ + (Xdst)->space = 0; \ + (Xdst)->addr = 0; \ + (Xdst)->offset = 0; \ + (Xdst)->type = Xtype; \ + } while(0) + +#define kvec_dst_unmap(Xdst) \ + do { \ + struct kvec_dst *_dst = (Xdst); \ + kunmap_atomic(_dst->addr, _dst->type); \ + _dst->offset = _dst->dst - _dst->addr; \ + _dst->offset -= _dst->let->offset; \ + _dst->addr = NULL; \ + } while(0) + +extern void FASTCALL(memcpy_to_kvec_dst(struct kvec_dst *dst, + const char *from, long len)); +extern void FASTCALL(memcpy_from_kvec_dst(char *to, + struct kvec_dst *from, long len)); + +#endif /* __LINUX__KIOVEC_H */ diff -urN v2.4.19-pre5/include/linux/list.h linux.diff/include/linux/list.h --- v2.4.19-pre5/include/linux/list.h Wed Apr 3 21:12:49 2002 +++ linux.diff/include/linux/list.h Tue Apr 30 17:29:31 2002 @@ -172,6 +172,11 @@ pos = pos->prev, prefetch(pos->prev)) +#define list_first(head) (((head)->next != (head)) ? (head)->next: (struct list_head *) 0) +#define list_last(head) (((head)->prev != (head)) ? (head)->prev: (struct list_head *) 0) +#define list_next(pos, head) (((pos)->next != (head)) ? (pos)->next: (struct list_head *) 0) +#define list_prev(pos, head) (((pos)->prev != (head)) ? (pos)->prev: (struct list_head *) 0) + #endif /* __KERNEL__ || _LVM_H_INCLUDE */ #endif diff -urN v2.4.19-pre5/include/linux/net.h linux.diff/include/linux/net.h --- v2.4.19-pre5/include/linux/net.h Wed Apr 3 21:12:49 2002 +++ linux.diff/include/linux/net.h Tue Apr 30 17:56:55 2002 @@ -83,6 +83,9 @@ struct scm_cookie; struct vm_area_struct; struct page; +struct iocb; +struct kioctx; +#include /* shut gcc up */ struct proto_ops { int family; @@ -110,6 +113,8 @@ int (*recvmsg) (struct socket *sock, struct msghdr *m, int total_len, int flags, struct scm_cookie *scm); int (*mmap) (struct file *file, struct socket *sock, struct vm_area_struct * vma); ssize_t (*sendpage) (struct socket *sock, struct page *page, int offset, size_t size, int flags); + int (*kvec_read) (struct socket *sock, kvec_cb_t cb, size_t size); + int (*kvec_write) (struct socket *sock, kvec_cb_t cb, size_t size); }; struct net_proto_family diff -urN v2.4.19-pre5/include/linux/pagemap.h linux.diff/include/linux/pagemap.h --- v2.4.19-pre5/include/linux/pagemap.h Wed Apr 3 21:12:55 2002 +++ linux.diff/include/linux/pagemap.h Tue Apr 30 17:56:55 2002 @@ -88,6 +88,7 @@ extern void add_to_page_cache(struct page * page, struct address_space *mapping, unsigned long index); extern void add_to_page_cache_locked(struct page * page, struct address_space *mapping, unsigned long index); extern int add_to_page_cache_unique(struct page * page, struct address_space *mapping, unsigned long index, struct page **hash); +extern wait_queue_head_t *FASTCALL(page_waitqueue(struct page *page)); extern void ___wait_on_page(struct page *); diff -urN v2.4.19-pre5/include/linux/pipe_fs_i.h linux.diff/include/linux/pipe_fs_i.h --- v2.4.19-pre5/include/linux/pipe_fs_i.h Thu May 3 11:22:20 2001 +++ linux.diff/include/linux/pipe_fs_i.h Tue Apr 30 17:29:31 2002 @@ -13,6 +13,8 @@ unsigned int waiting_writers; unsigned int r_counter; unsigned int w_counter; + struct file *rdfile; + struct file *wrfile; }; /* Differs from PIPE_BUF in that PIPE_SIZE is the length of the actual @@ -30,6 +32,8 @@ #define PIPE_WAITING_WRITERS(inode) ((inode).i_pipe->waiting_writers) #define PIPE_RCOUNTER(inode) ((inode).i_pipe->r_counter) #define PIPE_WCOUNTER(inode) ((inode).i_pipe->w_counter) +#define PIPE_READFILE(inode) ((inode).i_pipe->rdfile) +#define PIPE_WRITEFILE(inode) ((inode).i_pipe->wrfile) #define PIPE_EMPTY(inode) (PIPE_LEN(inode) == 0) #define PIPE_FULL(inode) (PIPE_LEN(inode) == PIPE_SIZE) diff -urN v2.4.19-pre5/include/linux/poll.h linux.diff/include/linux/poll.h --- v2.4.19-pre5/include/linux/poll.h Wed Apr 3 21:12:55 2002 +++ linux.diff/include/linux/poll.h Tue Apr 30 17:56:56 2002 @@ -9,12 +9,15 @@ #include #include #include +#include struct poll_table_page; +struct kiocb; typedef struct poll_table_struct { - int error; - struct poll_table_page * table; + int error; + struct poll_table_page *table; + struct kiocb *iocb; /* iocb for async poll */ } poll_table; extern void __pollwait(struct file * filp, wait_queue_head_t * wait_address, poll_table *p); @@ -29,8 +32,11 @@ { pt->error = 0; pt->table = NULL; + pt->iocb = NULL; } + extern void poll_freewait(poll_table* pt); +extern int async_poll(struct kiocb *iocb, int events); /* diff -urN v2.4.19-pre5/include/linux/sched.h linux.diff/include/linux/sched.h --- v2.4.19-pre5/include/linux/sched.h Wed Apr 3 21:12:55 2002 +++ linux.diff/include/linux/sched.h Tue Apr 30 17:56:55 2002 @@ -203,6 +203,7 @@ extern int max_map_count; +struct kioctx; struct mm_struct { struct vm_area_struct * mmap; /* list of VMAs */ rb_root_t mm_rb; @@ -231,6 +232,10 @@ /* Architecture-specific MM context */ mm_context_t context; + + struct kioctx *ioctx_list; + unsigned long new_ioctx_id; + int vsys_mapped; }; extern int mmlist_nr; @@ -243,6 +248,7 @@ mm_count: ATOMIC_INIT(1), \ mmap_sem: __RWSEM_INITIALIZER(name.mmap_sem), \ page_table_lock: SPIN_LOCK_UNLOCKED, \ + vsys_mapped: 0, \ mmlist: LIST_HEAD_INIT(name.mmlist), \ } @@ -794,6 +800,7 @@ extern void FASTCALL(add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)); extern void FASTCALL(add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)); +extern void FASTCALL(add_wait_queue_exclusive_lifo(wait_queue_head_t *q, wait_queue_t * wait)); extern void FASTCALL(remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)); #define __wait_event(wq, condition) \ diff -urN v2.4.19-pre5/include/linux/skbuff.h linux.diff/include/linux/skbuff.h --- v2.4.19-pre5/include/linux/skbuff.h Wed Apr 3 21:12:55 2002 +++ linux.diff/include/linux/skbuff.h Tue Apr 30 17:56:56 2002 @@ -1128,6 +1128,15 @@ extern unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, int len, unsigned int csum); extern void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); +/* skb <-> kvec helpers */ +extern void skb_copy_datagram_kvec(const struct sk_buff *skb, int offset, + struct kvec *vec, int len); +extern int skb_copy_and_csum_datagram_kvec(const struct sk_buff *skb, + int offset, struct kvec *vec, int len); +extern int skb_kvec_recv_datagram(struct sock * sk, kvec_cb_t cb, int len, + void (*finish)(struct sock *sk, kvec_cb_t cb, int len, struct sk_buff *skb)); + + extern void skb_init(void); extern void skb_add_mtu(int mtu); diff -urN v2.4.19-pre5/include/linux/sysctl.h linux.diff/include/linux/sysctl.h --- v2.4.19-pre5/include/linux/sysctl.h Wed Apr 3 21:12:49 2002 +++ linux.diff/include/linux/sysctl.h Tue Apr 30 17:56:56 2002 @@ -546,6 +546,13 @@ FS_LEASES=13, /* int: leases enabled */ FS_DIR_NOTIFY=14, /* int: directory notification enabled */ FS_LEASE_TIME=15, /* int: maximum time to wait for a lease break */ + /* 16 == jbd-debug */ + /* 17 == jbd-oom-retry */ + + FS_AIO_NR=18, /* int: current number of aio requests */ + FS_AIO_MAX_NR=19, /* int: max system wide aio requests */ + FS_AIO_MAX_SIZE=20, /* int: max size of read/write chunks */ + FS_AIO_MAX_PINNED=21, /* long: max memory pinned (in pages) */ }; /* CTL_DEBUG names: */ diff -urN v2.4.19-pre5/include/linux/tasklet.h linux.diff/include/linux/tasklet.h --- v2.4.19-pre5/include/linux/tasklet.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/tasklet.h Wed Apr 10 17:06:48 2002 @@ -0,0 +1,154 @@ +#ifndef __LINUX__TASKLET_H +#define __LINUX__TASKLET_H + +#include +#include +#include +#include /* for smp_mb */ + +/* Tasklets --- multithreaded analogue of BHs. + + Main feature differing them of generic softirqs: tasklet + is running only on one CPU simultaneously. + + Main feature differing them of BHs: different tasklets + may be run simultaneously on different CPUs. + + Properties: + * If tasklet_schedule() is called, then tasklet is guaranteed + to be executed on some cpu at least once after this. + * If the tasklet is already scheduled, but its excecution is still not + started, it will be executed only once. + * If this tasklet is already running on another CPU (or schedule is called + from tasklet itself), it is rescheduled for later. + * Tasklet is strictly serialized wrt itself, but not + wrt another tasklets. If client needs some intertask synchronization, + he makes it with spinlocks. + */ + +struct tasklet_struct +{ + struct tasklet_struct *next; + unsigned long state; + atomic_t count; + void (*func)(unsigned long); + unsigned long data; + int *unlocked; +}; + +#define DECLARE_TASKLET(name, func, data) \ +struct tasklet_struct name = { NULL, 0, ATOMIC_INIT(0), func, data, NULL } + +#define DECLARE_TASKLET_DISABLED(name, func, data) \ +struct tasklet_struct name = { NULL, 0, ATOMIC_INIT(1), func, data, NULL } + + +enum +{ + TASKLET_STATE_SCHED, /* Tasklet is scheduled for execution */ + TASKLET_STATE_RUN /* Tasklet is running (SMP only) */ +}; + +struct tasklet_head +{ + struct tasklet_struct *list; +} __attribute__ ((__aligned__(SMP_CACHE_BYTES))); + +extern struct tasklet_head tasklet_vec[NR_CPUS]; +extern struct tasklet_head tasklet_hi_vec[NR_CPUS]; + +#ifdef CONFIG_SMP +static inline int tasklet_trylock(struct tasklet_struct *t) +{ + return !test_and_set_bit(TASKLET_STATE_RUN, &(t)->state); +} + +static inline void tasklet_unlock(struct tasklet_struct *t) +{ + smp_mb__before_clear_bit(); + clear_bit(TASKLET_STATE_RUN, &(t)->state); +} + +static inline void tasklet_unlock_self(struct tasklet_struct *t) +{ + *t->unlocked = 1; + t->unlocked = NULL; + tasklet_unlock(t); +} + +static inline void tasklet_unlock_wait(struct tasklet_struct *t) +{ + while (test_bit(TASKLET_STATE_RUN, &(t)->state)) { barrier(); } +} +#else +#define tasklet_trylock(t) 1 +#define tasklet_unlock_wait(t) do { } while (0) +#define tasklet_unlock(t) do { } while (0) +#endif + +extern void FASTCALL(__tasklet_schedule(struct tasklet_struct *t)); + +static inline void tasklet_schedule(struct tasklet_struct *t) +{ + if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) + __tasklet_schedule(t); +} + +extern void FASTCALL(__tasklet_hi_schedule(struct tasklet_struct *t)); + +static inline void tasklet_hi_schedule(struct tasklet_struct *t) +{ + if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) + __tasklet_hi_schedule(t); +} + + +static inline void tasklet_disable_nosync(struct tasklet_struct *t) +{ + atomic_inc(&t->count); + smp_mb__after_atomic_inc(); +} + +static inline void tasklet_disable(struct tasklet_struct *t) +{ + tasklet_disable_nosync(t); + tasklet_unlock_wait(t); + smp_mb(); +} + +static inline void tasklet_enable(struct tasklet_struct *t) +{ + smp_mb__before_atomic_dec(); + atomic_dec(&t->count); +} + +static inline void tasklet_hi_enable(struct tasklet_struct *t) +{ + smp_mb__before_atomic_dec(); + atomic_dec(&t->count); +} + +extern void tasklet_kill(struct tasklet_struct *t); +extern void tasklet_init(struct tasklet_struct *t, + void (*func)(unsigned long), unsigned long data); + +#ifdef CONFIG_SMP + +#define SMP_TIMER_NAME(name) name##__thr + +#define SMP_TIMER_DEFINE(name, task) \ +DECLARE_TASKLET(task, name##__thr, 0); \ +static void name (unsigned long dummy) \ +{ \ + tasklet_schedule(&(task)); \ +} + +#else /* CONFIG_SMP */ + +#define SMP_TIMER_NAME(name) name +#define SMP_TIMER_DEFINE(name, task) + +#endif /* CONFIG_SMP */ + + +#endif /* __LINUX__TASKLET_H */ diff -urN v2.4.19-pre5/include/linux/tqueue.h linux.diff/include/linux/tqueue.h --- v2.4.19-pre5/include/linux/tqueue.h Wed Apr 3 21:12:49 2002 +++ linux.diff/include/linux/tqueue.h Tue Apr 30 17:56:55 2002 @@ -67,6 +67,7 @@ #define TQ_ACTIVE(q) (!list_empty(&q)) extern task_queue tq_timer, tq_immediate, tq_disk; +extern struct tq_struct run_disk_tq; /* * To implement your own list of active bottom halfs, use the following diff -urN v2.4.19-pre5/include/linux/types.h linux.diff/include/linux/types.h --- v2.4.19-pre5/include/linux/types.h Wed Apr 3 21:10:29 2002 +++ linux.diff/include/linux/types.h Tue Apr 2 19:14:27 2002 @@ -127,4 +127,9 @@ char f_fpack[6]; }; +/* kernel typedefs -- they belong here. */ +#ifdef __KERNEL__ +typedef struct kvec_cb kvec_cb_t; +#endif /* __KERNEL__ */ + #endif /* _LINUX_TYPES_H */ diff -urN v2.4.19-pre5/include/linux/vsyscall.h linux.diff/include/linux/vsyscall.h --- v2.4.19-pre5/include/linux/vsyscall.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/vsyscall.h Tue Apr 2 18:56:58 2002 @@ -0,0 +1,20 @@ +#ifndef _LINUX__VSYSCALL_H +#define _LINUX__VSYSCALL_H + +struct vsys_cpudata { + unsigned long context_switches; + unsigned long tv_sec; + unsigned long tsc_low, tsc_high; + unsigned long cycles_per_sec; +}; + +union vsys_union { + struct vsys_cpudata data; + char pad[128]; +}; + +extern union vsys_union vsys_cpudata[256] __attribute__((section(".data.vsyscall"))); + +#define vsys_data(cpu) (&vsys_cpudata[cpu].data) + +#endif /*ndef _LINUX__VSYSCALL_H*/ diff -urN v2.4.19-pre5/include/linux/wait.h linux.diff/include/linux/wait.h --- v2.4.19-pre5/include/linux/wait.h Wed Apr 3 21:12:49 2002 +++ linux.diff/include/linux/wait.h Tue Apr 30 17:56:55 2002 @@ -28,17 +28,20 @@ #define WAITQUEUE_DEBUG 0 #endif +typedef struct __wait_queue wait_queue_t; +typedef void (*wait_queue_func_t)(wait_queue_t *wait); + struct __wait_queue { unsigned int flags; #define WQ_FLAG_EXCLUSIVE 0x01 struct task_struct * task; struct list_head task_list; + wait_queue_func_t func; #if WAITQUEUE_DEBUG long __magic; long __waker; #endif }; -typedef struct __wait_queue wait_queue_t; /* * 'dual' spinlock architecture. Can be switched between spinlock_t and @@ -137,6 +140,7 @@ #endif #define __WAITQUEUE_INITIALIZER(name, tsk) { \ + func: NULL, \ task: tsk, \ task_list: { NULL, NULL }, \ __WAITQUEUE_DEBUG_INIT(name)} @@ -174,6 +178,22 @@ #endif q->flags = 0; q->task = p; + q->func = NULL; +#if WAITQUEUE_DEBUG + q->__magic = (long)&q->__magic; +#endif +} + +static inline void init_waitqueue_func_entry(wait_queue_t *q, + wait_queue_func_t func) +{ +#if WAITQUEUE_DEBUG + if (!q || !p) + WQ_BUG(); +#endif + q->flags = 0; + q->task = NULL; + q->func = func; #if WAITQUEUE_DEBUG q->__magic = (long)&q->__magic; #endif @@ -231,6 +251,22 @@ list_del(&old->task_list); } +#define add_wait_queue_cond(q, wait, cond) \ + ({ \ + unsigned long flags; \ + int _raced = 0; \ + wq_write_lock_irqsave(&(q)->lock, flags); \ + (wait)->flags = 0; \ + __add_wait_queue((q), (wait)); \ + rmb(); \ + if (!(cond)) { \ + _raced = 1; \ + __remove_wait_queue((q), (wait)); \ + } \ + wq_write_unlock_irqrestore(&(q)->lock, flags); \ + _raced; \ + }) + #endif /* __KERNEL__ */ #endif diff -urN v2.4.19-pre5/include/linux/worktodo.h linux.diff/include/linux/worktodo.h --- v2.4.19-pre5/include/linux/worktodo.h Wed Dec 31 19:00:00 1969 +++ linux.diff/include/linux/worktodo.h Tue Apr 30 17:56:56 2002 @@ -0,0 +1,75 @@ +/* + * Written by Benjamin LaHaise. + * + * Copyright 2000-2001 Red Hat, Inc. + * + * #include "gpl.h" + * + * Basic design idea from Jeff Merkey. + * Stack based on ideas from Ingo Molnar. + */ +#ifndef __LINUX__WORKTODO_H +#define __LINUX__WORKTODO_H + +#ifndef _LINUX_WAIT_H +#include +#endif +#ifndef _LINUX_TQUEUE_H +#include +#endif + +struct wtd_stack { + void (*fn)(void *data); + void *data; +}; + +struct worktodo { + wait_queue_t wait; + struct tq_struct tq; + + void *data; /* for use by the wtd_ primatives */ + + int sp; + struct wtd_stack stack[3]; +}; + +/* FIXME NOTE: factor from kernel/context.c */ +#define wtd_init(wtd, routine) do { \ + INIT_TQUEUE(&(wtd)->tq, (routine), (wtd)); \ + (wtd)->data = 0; \ + (wtd)->sp = 0; \ +} while (0) + +#define wtd_queue(wtd) schedule_task(&(wtd)->tq) + +#define wtd_push(wtd, action, wtddata) \ +do { \ + (wtd)->stack[(wtd)->sp].fn = (wtd)->tq.routine; \ + (wtd)->stack[(wtd)->sp++].data = (wtd)->tq.data;\ + (wtd)->tq.routine = action; \ + (wtd)->tq.data = wtddata; \ +} while (0) + +static inline void wtd_pop(struct worktodo *wtd) +{ + if (wtd->sp) { + wtd->sp--; + wtd->tq.routine = wtd->stack[wtd->sp].fn; + wtd->tq.data = wtd->stack[wtd->sp].data; + } +} + +#define wtd_set_action(wtd, action, wtddata) INIT_TQUEUE(&(wtd)->tq, action, wtddata) + +struct page; +struct buffer_head; +extern int wtd_lock_page(struct worktodo *wtd, struct page *page); +extern int wtd_wait_on_buffer(struct worktodo *wtd, struct buffer_head *bh); + +#if 0 /* not implemented yet */ +extern void wtd_down(struct worktodo *wtd, struct semaphore *sem); +extern void wtd_down_write(struct worktodo *wtd, struct rw_semaphore *sem); +extern void wtd_down_read(struct worktodo *wtd, struct rw_semaphore *sem); +#endif + +#endif /* __LINUX__WORKTODO_H */ diff -urN v2.4.19-pre5/include/net/sock.h linux.diff/include/net/sock.h --- v2.4.19-pre5/include/net/sock.h Wed Apr 3 21:12:56 2002 +++ linux.diff/include/net/sock.h Tue Apr 30 17:57:03 2002 @@ -105,7 +105,18 @@ #include #include +#include +#include +#include +#include + + +struct sock_iocb { + struct list_head list; + kvec_cb_t cb; + struct kvec_dst dst; +}; /* The AF_UNIX specific socket options */ struct unix_opt { @@ -283,9 +294,9 @@ /* Data for direct copy to user */ struct { struct sk_buff_head prequeue; - int memory; struct task_struct *task; struct iovec *iov; + int memory; int len; } ucopy; @@ -560,6 +571,9 @@ struct sk_buff *tail; } backlog; + struct list_head kvec_read_list; + struct list_head kvec_write_list; + rwlock_t callback_lock; /* Error queue, rarely used. */ @@ -721,6 +735,8 @@ int (*recvmsg)(struct sock *sk, struct msghdr *msg, int len, int noblock, int flags, int *addr_len); + int (*kvec_read)(struct sock *, kvec_cb_t cb, int len); + int (*kvec_write)(struct sock *, kvec_cb_t cb, int len); int (*bind)(struct sock *sk, struct sockaddr *uaddr, int addr_len); @@ -795,7 +811,7 @@ if ((__sk)->backlog.tail != NULL) \ __release_sock(__sk); \ (__sk)->lock.users = 0; \ - if (waitqueue_active(&((__sk)->lock.wq))) wake_up(&((__sk)->lock.wq)); \ + wake_up(&((__sk)->lock.wq)); \ spin_unlock_bh(&((__sk)->lock.slock)); \ } while(0) @@ -1215,8 +1231,13 @@ static inline void sk_wake_async(struct sock *sk, int how, int band) { - if (sk->socket && sk->socket->fasync_list) + if (sk->socket) { + if (sk->socket->file) + file_send_notify(sk->socket->file, ion_band_table[band - POLL_IN], + poll_band_table[band - POLL_IN]); + if (sk->socket->fasync_list) sock_wake_async(sk->socket, how, band); + } } #define SOCK_MIN_SNDBUF 2048 diff -urN v2.4.19-pre5/include/net/tcp.h linux.diff/include/net/tcp.h --- v2.4.19-pre5/include/net/tcp.h Wed Apr 3 21:12:57 2002 +++ linux.diff/include/net/tcp.h Tue Apr 30 17:59:28 2002 @@ -732,6 +732,8 @@ struct msghdr *msg, int len, int nonblock, int flags, int *addr_len); +extern int tcp_kvec_read(struct sock *sk, kvec_cb_t cb, int len); +extern int tcp_kvec_write(struct sock *sk, kvec_cb_t cb, int len); extern int tcp_listen_start(struct sock *sk); diff -urN v2.4.19-pre5/kernel/fork.c linux.diff/kernel/fork.c --- v2.4.19-pre5/kernel/fork.c Wed Apr 3 21:04:41 2002 +++ linux.diff/kernel/fork.c Mon Apr 29 15:54:22 2002 @@ -46,6 +46,16 @@ wq_write_unlock_irqrestore(&q->lock, flags); } +void add_wait_queue_exclusive_lifo(wait_queue_head_t *q, wait_queue_t * wait) +{ + unsigned long flags; + + wq_write_lock_irqsave(&q->lock, flags); + wait->flags = WQ_FLAG_EXCLUSIVE; + __add_wait_queue(q, wait); + wq_write_unlock_irqrestore(&q->lock, flags); +} + void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait) { unsigned long flags; @@ -216,6 +226,7 @@ static struct mm_struct * mm_init(struct mm_struct * mm) { + mm->ioctx_list = NULL; atomic_set(&mm->mm_users, 1); atomic_set(&mm->mm_count, 1); init_rwsem(&mm->mmap_sem); @@ -251,6 +262,9 @@ */ inline void __mmdrop(struct mm_struct *mm) { + if (mm->ioctx_list) + BUG(); + if (mm == &init_mm) BUG(); pgd_free(mm->pgd); destroy_context(mm); @@ -269,6 +283,7 @@ list_del(&mm->mmlist); mmlist_nr--; spin_unlock(&mmlist_lock); + exit_aio(mm); exit_mmap(mm); mmdrop(mm); } diff -urN v2.4.19-pre5/kernel/sched.c linux.diff/kernel/sched.c --- v2.4.19-pre5/kernel/sched.c Tue Jan 1 14:09:35 2002 +++ linux.diff/kernel/sched.c Tue Apr 2 18:56:57 2002 @@ -705,33 +705,44 @@ } /* - * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just wake everything - * up. If it's an exclusive wakeup (nr_exclusive == small +ve number) then we wake all the - * non-exclusive tasks and one exclusive task. + * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just + * wake everything up. If it's an exclusive wakeup (nr_exclusive == small + * +ve number) then we wake all the non-exclusive tasks and one exclusive task. * * There are circumstances in which we can try to wake a task which has already - * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns zero - * in this (rare) case, and we handle it by contonuing to scan the queue. + * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns + * zero in this (rare) case, and we handle it by contonuing to scan the queue. */ static inline void __wake_up_common (wait_queue_head_t *q, unsigned int mode, int nr_exclusive, const int sync) { - struct list_head *tmp; + struct list_head *tmp, *next; struct task_struct *p; CHECK_MAGIC_WQHEAD(q); WQ_CHECK_LIST_HEAD(&q->task_list); - list_for_each(tmp,&q->task_list) { + list_for_each_safe(tmp, next, &q->task_list) { unsigned int state; - wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); + wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); + wait_queue_func_t func; CHECK_MAGIC(curr->__magic); + func = curr->func; + if (func) { + unsigned flags = curr->flags; + func(curr); + if ((flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) + break; + continue; + } p = curr->task; state = p->state; if (state & mode) { WQ_NOTE_WAKER(curr); - if (try_to_wake_up(p, sync) && (curr->flags&WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) + if (try_to_wake_up(p, sync) && + (curr->flags & WQ_FLAG_EXCLUSIVE) && + !--nr_exclusive) break; } } diff -urN v2.4.19-pre5/kernel/sysctl.c linux.diff/kernel/sysctl.c --- v2.4.19-pre5/kernel/sysctl.c Wed Apr 3 21:04:41 2002 +++ linux.diff/kernel/sysctl.c Mon Apr 29 17:58:53 2002 @@ -30,6 +30,7 @@ #include #include #include +#include #include @@ -284,6 +285,8 @@ {0} }; +extern int user_pinned_pages; + static ctl_table fs_table[] = { {FS_NRINODE, "inode-nr", &inodes_stat, 2*sizeof(int), 0444, NULL, &proc_dointvec}, @@ -309,6 +312,16 @@ sizeof(int), 0644, NULL, &proc_dointvec}, {FS_LEASE_TIME, "lease-break-time", &lease_break_time, sizeof(int), 0644, NULL, &proc_dointvec}, + {FS_AIO_NR, "aio-nr", &aio_nr, sizeof(aio_nr), + 0444, NULL, &proc_dointvec}, + {FS_AIO_MAX_NR, "aio-max-nr", &aio_max_nr, sizeof(aio_max_nr), + 0644, NULL, &proc_dointvec}, + {FS_AIO_MAX_SIZE, "aio-max-size", &aio_max_size, sizeof(aio_max_size), + 0644, NULL, &proc_dointvec}, + {FS_AIO_MAX_PINNED, "aio-max-pinned", &aio_max_pinned, sizeof(aio_max_pinned), + 0644, NULL, &proc_dointvec}, + {FS_AIO_MAX_PINNED+1, "aio-pinned", &user_pinned_pages, 4, + 0644, NULL, &proc_dointvec}, {0} }; diff -urN v2.4.19-pre5/mm/Makefile linux.diff/mm/Makefile --- v2.4.19-pre5/mm/Makefile Wed Apr 3 21:04:41 2002 +++ linux.diff/mm/Makefile Tue Apr 2 18:56:57 2002 @@ -17,5 +17,6 @@ shmem.o obj-$(CONFIG_HIGHMEM) += highmem.o +obj-y += wtd.o include $(TOPDIR)/Rules.make diff -urN v2.4.19-pre5/mm/filemap.c linux.diff/mm/filemap.c --- v2.4.19-pre5/mm/filemap.c Wed Apr 3 21:04:41 2002 +++ linux.diff/mm/filemap.c Mon Apr 29 18:30:59 2002 @@ -29,6 +29,8 @@ #include #include +#include +#include /* * Shared mappings implemented 30.11.1994. It's not fully working yet, @@ -774,7 +776,7 @@ * at a cost of "thundering herd" phenomena during rare hash * collisions. */ -static inline wait_queue_head_t *page_waitqueue(struct page *page) +static inline wait_queue_head_t *__page_waitqueue(struct page *page) { const zone_t *zone = page_zone(page); wait_queue_head_t *wait = zone->wait_table; @@ -805,6 +807,13 @@ return &wait[hash]; } +wait_queue_head_t *page_waitqueue(struct page *page) +{ + return __page_waitqueue(page); +} + +#define page_waitqueue(page) __page_waitqueue(page) + /* * Wait for a page to get unlocked. * @@ -1185,7 +1194,7 @@ static void generic_file_readahead(int reada_ok, struct file * filp, struct inode * inode, - struct page * page) + struct page * page, int flags) { unsigned long end_index; unsigned long index = page->index; @@ -1315,7 +1324,7 @@ * This is really ugly. But the goto's actually try to clarify some * of the logic when it comes to error handling etc. */ -void do_generic_file_read(struct file * filp, loff_t *ppos, read_descriptor_t * desc, read_actor_t actor) +void do_generic_file_read(struct file * filp, loff_t *ppos, read_descriptor_t * desc, read_actor_t actor, int flags) { struct address_space *mapping = filp->f_dentry->d_inode->i_mapping; struct inode *inode = mapping->host; @@ -1324,10 +1333,17 @@ int reada_ok; int error; int max_readahead = get_max_readahead(inode); + loff_t pos; + + pos = *ppos; + if (unlikely(pos < 0)) { + desc->error = -EINVAL; + return; + } cached_page = NULL; - index = *ppos >> PAGE_CACHE_SHIFT; - offset = *ppos & ~PAGE_CACHE_MASK; + index = pos >> PAGE_CACHE_SHIFT; + offset = pos & ~PAGE_CACHE_MASK; /* * If the current position is outside the previous read-ahead window, @@ -1374,13 +1390,17 @@ end_index = inode->i_size >> PAGE_CACHE_SHIFT; - if (index > end_index) + if (index > end_index) { + desc->error = 0; break; + } nr = PAGE_CACHE_SIZE; if (index == end_index) { nr = inode->i_size & ~PAGE_CACHE_MASK; - if (nr <= offset) + if (nr <= offset) { + desc->error = 0; break; + } } nr = nr - offset; @@ -1400,7 +1420,7 @@ if (!Page_Uptodate(page)) goto page_not_up_to_date; - generic_file_readahead(reada_ok, filp, inode, page); + generic_file_readahead(reada_ok, filp, inode, page, flags); page_ok: /* If users can be writing to this page using arbitrary * virtual addresses, take care about potential aliasing @@ -1440,13 +1460,23 @@ * Ok, the page was not immediately readable, so let's try to read ahead while we're at it.. */ page_not_up_to_date: - generic_file_readahead(reada_ok, filp, inode, page); + generic_file_readahead(reada_ok, filp, inode, page, flags); if (Page_Uptodate(page)) goto page_ok; /* Get exclusive access to the page ... */ - lock_page(page); + if (flags & F_ATOMIC) { + if (TryLockPage(page)) { + if (Page_Uptodate(page)) + goto page_ok; + desc->error = -EWOULDBLOCKIO; + page_cache_release(page); + break; + } + printk("page_not_up_to_date: atomic trylock succeeded\n"); + } else + lock_page(page); /* Did it get unhashed before we got the lock? */ if (!page->mapping) { @@ -1470,11 +1500,12 @@ goto page_ok; /* Again, try some read-ahead while waiting for the page to finish.. */ - generic_file_readahead(reada_ok, filp, inode, page); - wait_on_page(page); + generic_file_readahead(reada_ok, filp, inode, page, flags); + if (!(flags & F_ATOMIC)) + wait_on_page(page); if (Page_Uptodate(page)) goto page_ok; - error = -EIO; + error = (flags & F_ATOMIC) ? -EWOULDBLOCKIO : -EIO; } /* UHHUH! A synchronous read error occurred. Report it */ @@ -1483,6 +1514,11 @@ break; no_cached_page: + if (flags & F_ATOMIC) { + spin_unlock(&pagecache_lock); + desc->error = -EWOULDBLOCKIO; + break; + } /* * Ok, it wasn't cached, so we need to create a new * page.. @@ -1637,6 +1673,11 @@ */ ssize_t generic_file_read(struct file * filp, char * buf, size_t count, loff_t *ppos) { + return generic_file_new_read(filp, buf, count, ppos, 0); +} + +ssize_t generic_file_new_read(struct file * filp, char * buf, size_t count, loff_t *ppos, int flags) +{ ssize_t retval; if ((ssize_t) count < 0) @@ -1656,7 +1697,7 @@ desc.count = count; desc.buf = buf; desc.error = 0; - do_generic_file_read(filp, ppos, &desc, file_read_actor); + do_generic_file_read(filp, ppos, &desc, file_read_actor, flags); retval = desc.written; if (!retval) @@ -1781,7 +1822,7 @@ desc.count = count; desc.buf = (char *) out_file; desc.error = 0; - do_generic_file_read(in_file, ppos, &desc, file_send_actor); + do_generic_file_read(in_file, ppos, &desc, file_send_actor, 0); retval = desc.written; if (!retval) @@ -3177,3 +3218,654 @@ panic("Failed to allocate page hash table\n"); memset((void *)page_hash_table, 0, PAGE_HASH_SIZE * sizeof(struct page *)); } + +/* address_space_map + * Maps a series of pages from the page cache into the given array. + */ +static int address_space_map(struct address_space *as, unsigned long index, + int nr, struct page **pages, + int *nr_newp, struct page **new_pages) +{ + struct page *cached_page = NULL; + int nr_new = 0; + int ret; + + if (unlikely(nr <= 0)) { + *nr_newp = nr_new; + return 0; + } + + ret = 0; + + spin_lock(&pagecache_lock); + + while (nr > 0) { + struct page **hash = page_hash(as, index); + struct page *page; + + page = __find_page_nolock(as, index, *hash); + if (page) { + page_cache_get(page); +got_page: + pages[ret++] = page; + index++; + nr--; + continue; + } + + if (cached_page) { + __add_to_page_cache(cached_page, as, index, hash); + nr_new++; + *new_pages++ = page = cached_page; + cached_page = NULL; + goto got_page; + } + spin_unlock(&pagecache_lock); + + cached_page = page_cache_alloc(as); + if (!cached_page) + goto out; + + /* Okay, we now have an allocated page. Retry + * the search and add. */ + spin_lock(&pagecache_lock); + } + + spin_unlock(&pagecache_lock); + +out: + if (cached_page) + page_cache_release(cached_page); + + *nr_newp = nr_new; + return ret ? ret : -ENOMEM; +} + +struct iodesc { + struct worktodo wtd; + + struct page *good_page; /* the highest Uptodate page */ + int good_idx; + int err; + int did_read; + int rw; + + struct page **pages; + struct page **new_pages; + struct page **cur_pagep; + int nr_pages; + int nr_new_pages; + + struct address_space *as; + struct file *file; + kvec_cb_t cb; + + size_t size; + unsigned long transferred; + unsigned offset; + struct kveclet *veclet; + + struct kvec_dst src; + + int sync; + +#define READDESC_NR_DEF 3 + struct page *def_pages[READDESC_NR_DEF]; + struct page *def_new_pages[READDESC_NR_DEF]; +}; + +static void __iodesc_free(struct iodesc *io, int unlock) +{ + kvec_cb_t cb; + ssize_t res; + + if (unlock) { + unsigned i; + for (i=0; inr_pages; i++) { + struct page *page = io->pages[i]; + UnlockPage(page); + page_cache_release(page); + } + } else { + unsigned i; + for (i=0; inr_pages; i++) + page_cache_release(io->pages[i]); + } + + if (io->new_pages != io->def_new_pages) + kfree(io->new_pages); + if (io->pages != io->def_pages) + kfree(io->pages); + + cb = io->cb; + res = io->transferred ? io->transferred : io->err; + kfree(io); + + cb.fn(cb.data, cb.vec, res); +} + +/* By the time this function is called, all of the pages prior to + * the current good_idx have been released appropriately. The remaining + * duties are to release any remaining pages and to honour O_SYNC. + */ +static void __iodesc_finish_write(struct iodesc *io) +{ + pr_debug("__iodesc_finish_write(%p)\n", io); + + __iodesc_free(io, WRITE == io->rw); +} + +/* This is mostly ripped from generic_file_write */ +static int __iodesc_write_page(struct iodesc *io, struct page *page) +{ + char *kaddr = kmap(page); + unsigned long bytes; + unsigned long offset; + long status; + int done = 0; + + offset = io->offset; + kaddr += offset; + + bytes = PAGE_CACHE_SIZE - offset; + if (io->size < bytes) + bytes = io->size; + + pr_debug("__iodesc_write_page(%p (%lu), %lu %lu %lu)\n", page, page->index, offset, bytes); + + io->err = io->as->a_ops->prepare_write(io->file, page, + offset, offset + bytes); + if (unlikely(io->err)) { + pr_debug("prepare_write: %d\n", io->err); + kunmap(page); + return 1; + } + + kvec_dst_map(&io->src); + memcpy_from_kvec_dst(kaddr, &io->src, bytes); + kvec_dst_unmap(&io->src); /* commit_write may block */ + + flush_dcache_page(page); + status = io->as->a_ops->commit_write(io->file, page, + offset, offset+bytes); + + /* We don't handle short writes */ + if (status > 0 && status != bytes) + done = 1; + + if (!status) + status = bytes; + + if (likely(status > 0)) { + io->transferred += status; + io->size -= status; + io->offset = (offset + status) & (PAGE_CACHE_SIZE - 1); + + if (io->offset) + done = 1; + } else { + io->err = status; + done = 1; + } + + kunmap(page); + return done; +} + +void __iodesc_sync_wait_page(void *data) +{ + struct iodesc *io = data; + + do { + struct buffer_head *bh, *head = io->pages[io->good_idx]->buffers; + + if (!head) + continue; + + bh = head; + do { + if (buffer_locked(bh)) { + pr_debug("waiting on bh=%pi io=%p\n", bh, io); + if (!wtd_wait_on_buffer(&io->wtd, bh)) + return; + } + if (buffer_req(bh) && !buffer_uptodate(bh)) { + pr_debug("io err bh=%p (%p)\n", bh, io); + io->err = -EIO; + break; + } + } while ((bh = bh->b_this_page) != head); + } while (!io->err && ++io->good_idx < io->nr_pages) ; + + pr_debug("finish_write(%p)\n", io); + __iodesc_finish_write(io); +} + +static void __iodesc_do_write(void *data) +{ + struct iodesc *io = data; + unsigned i; + + for (i=0; inr_pages; i++) { + if (__iodesc_write_page(io, io->pages[i])) + break; + } + + up(&io->file->f_dentry->d_inode->i_sem); + + if (io->sync) { + io->good_idx = 0; + + pr_debug("writing out pages(%p)\n", io); + for (i=0; inr_pages; i++) { + if (io->pages[i]->buffers) + writeout_one_page(io->pages[i]); + } + + pr_debug("calling __iodesc_sync_wait_page(%p)\n", io); + wtd_set_action(&io->wtd, __iodesc_sync_wait_page, io); + __iodesc_sync_wait_page(io); + return; + } + + __iodesc_finish_write(io); +} + +static void __iodesc_write_lock_next_page(void *data) +{ + struct iodesc *io = data; + pr_debug("__iodesc_write_next_page(%p)\n", io); + + while (io->good_idx < io->nr_pages) { + io->good_page = io->pages[io->good_idx++]; + if (io->good_page == *io->cur_pagep) + io->cur_pagep++; + else { + if (!wtd_lock_page(&io->wtd, io->good_page)) + return; + } + } + + //Is this faster? __iodesc_do_write(io); + wtd_set_action(&io->wtd, __iodesc_do_write, io); + wtd_queue(&io->wtd); +} + +static void __generic_file_write_iodesc(struct iodesc *io) +{ + struct inode *inode = io->file->f_dentry->d_inode; + time_t now = CURRENT_TIME; + + remove_suid(inode); + if (inode->i_ctime != now || inode->i_mtime != now) { + inode->i_ctime = inode->i_mtime = now; + mark_inode_dirty_sync(inode); + } + + wtd_set_action(&io->wtd, __iodesc_write_lock_next_page, io); + io->sync = !!(io->file->f_flags & O_SYNC); + io->good_idx = 0; + io->cur_pagep = io->new_pages; + __iodesc_write_lock_next_page(io); +} + +static void __iodesc_read_finish(struct iodesc *io) +{ + struct page **src_pagep; + char *dst_addr, *src_addr; + int src_off; + size_t size; + size_t valid; + + struct kveclet *veclet = io->veclet; + struct page *dst_page = veclet->page; + int dst_len = veclet->length; + int dst_off = veclet->offset; + + + pr_debug("__iodesc_read_finish: good_idx = %d\n", io->good_idx); + if (io->good_idx <= 0) + goto no_data; + + size = io->size; + src_off = io->offset; + src_pagep = io->pages; + src_addr = kmap(*src_pagep); + + valid = (size_t)io->good_idx << PAGE_CACHE_SHIFT; + valid -= src_off; + pr_debug("size=%d valid=%d src_off=%d\n", size, valid, src_off); + + if (valid < size) + size = valid; + + dst_addr = kmap(veclet->page); + + while (size > 0) { + int this = PAGE_CACHE_SIZE - src_off; + if ((PAGE_SIZE - dst_off) < this) + this = PAGE_SIZE - dst_off; + if (size < this) + this = size; + pr_debug("this=%d src_off=%d dst_off=%d dst_len=%d\n", + this, src_off, dst_off, dst_len); + memcpy(dst_addr + dst_off, src_addr + src_off, this); + + src_off += this; + dst_off += this; + dst_len -= this; + size -= this; + io->transferred += this; + pr_debug("read_finish: this=%d transferred=%d\n", + this, io->transferred); + + if (size <= 0) + break; + + if (dst_len <= 0) { + kunmap(dst_page); + veclet++; + dst_page = veclet->page; + dst_off = veclet->offset; + dst_len = veclet->length; + dst_addr = kmap(dst_page); + } + + if (src_off >= PAGE_SIZE) { /* FIXME: PAGE_CACHE_SIZE */ + kunmap(*src_pagep); + pr_debug("page(%lu)->count = %d\n", + (*src_pagep)->index, + atomic_read(&(*src_pagep)->count)); + src_pagep++; + src_addr = kmap(*src_pagep); + src_off = 0; + } + } + kunmap(dst_page); + kunmap(*src_pagep); +no_data: + __iodesc_free(io, 0); +} + +static void __iodesc_make_uptodate(void *data) +{ + struct iodesc *io = data; + struct page *page = io->good_page; + int locked = 1; + + pr_debug("__iodesc_make_uptodate: io=%p index=%lu\n", io, page->index); +again: + while (Page_Uptodate(page)) { + pr_debug("page index %lu uptodate\n", page->index); + if (locked) { + UnlockPage(page); + locked = 0; + } + io->did_read = 0; + io->good_idx++; + if (io->good_idx >= io->nr_pages) { + __iodesc_read_finish(io); + return; + } + page = io->good_page = io->pages[io->good_idx]; + pr_debug("__iodesc_make_uptodate: index=%lu\n", page->index); + } + + if (!locked) { + if (!wtd_lock_page(&io->wtd, page)) + return; + locked = 1; + } + + if (!io->did_read) { + /* We haven't tried reading this page before, give it a go. */ + pr_debug("attempting to read %lu\n", page->index); + io->did_read = 1; + locked = 0; + io->err = page->mapping->a_ops->readpage(io->file, page); + if (!io->err) { + if (Page_Uptodate(page)) + goto again; + if (wtd_lock_page(&io->wtd, page)) { + locked = 1; + goto again; + } + return; + } + } + + if (locked) + UnlockPage(page); + + /* We've already read this page before. Set err to EIO and quite */ + if (!io->err) + io->err = -EIO; + __iodesc_read_finish(io); +} + +static void __wtdgeneric_file_read_iodesc(void *data); + +static void __generic_file_read_iodesc(struct iodesc *io, int mayblock) +{ + int (*readpage)(struct file *, struct page *); + int i; + + wtd_set_action(&io->wtd, __iodesc_make_uptodate, io); + readpage = io->as->a_ops->readpage; + for (i=0; inr_new_pages; i++) { + int ret; + if (!mayblock) { + wtd_set_action(&io->wtd, __wtdgeneric_file_read_iodesc, io); + wtd_queue(&io->wtd); + return; + } + ret = readpage(io->file, io->new_pages[i]); + if (ret) + printk(KERN_DEBUG "__generic_file_read_kiovec: readpage(%lu) = %d\n", io->new_pages[i]->index, ret); + } + + for (i=0; inr_pages; i++) { + struct page *page = io->pages[i]; + if (Page_Uptodate(page)) { + pr_debug("__generic_file_read_iodesc: %lu is uptodate\n", page->index); + continue; + } + + if (!mayblock) { + wtd_set_action(&io->wtd, __wtdgeneric_file_read_iodesc, io); + wtd_queue(&io->wtd); + return; + } + if (!TryLockPage(page)) { + int ret = readpage(io->file, page); + if (ret) + printk(KERN_DEBUG "__generic_file_read_iodesc: readpage(%lu): %d\n", page->index, ret); + } + + if (!Page_Uptodate(page) && io->good_idx == -1) { + pr_debug("first good_idx=%d (%lu)\n", i, page->index); + io->good_idx = i; + io->good_page = page; + } + } + + /* Whee, all the pages are uptodate! */ + if (!io->good_page) { + pr_debug("all pages uptodate!\n"); + io->good_idx = io->nr_pages; + __iodesc_read_finish(io); + return; + } + + pr_debug("locking good_page\n"); + if (wtd_lock_page(&io->wtd, io->good_page)) + __iodesc_make_uptodate(io); + return; +} + +static void __wtdgeneric_file_read_iodesc(void *data) +{ + struct iodesc *io = data; + __generic_file_read_iodesc(io, 1); +} + +static int generic_file_rw_kvec(struct file *file, int rw, kvec_cb_t cb, + size_t size, loff_t pos); + +int generic_file_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return generic_file_rw_kvec(file, READ, cb, size, pos); +} + +int generic_file_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + return generic_file_rw_kvec(file, WRITE, cb, size, pos); +} + +int generic_file_rw_kvec(struct file *file, int rw, kvec_cb_t cb, + size_t size, loff_t pos) +{ + struct inode *inode = file->f_dentry->d_inode; + struct address_space *as = inode->i_mapping; + unsigned long index; + unsigned long eindex; + unsigned long nr_pages; + struct iodesc *io = NULL; + int ret; + int append = 0; + + ret = -EINVAL; + if (unlikely(rw != READ && rw != WRITE)) + goto out; + + append = unlikely(0 != (file->f_flags & O_APPEND)); + + /* Don't check pos when appending, but otherwise do santity + * checks before allocating memory. -'ve offsets are invalid. + */ + if (unlikely(!append && pos < 0)) + goto out; + + ret = -ENOMEM; + io = kmalloc(sizeof(*io), GFP_KERNEL); + if (!io) + goto out; + + memset(io, 0, sizeof(*io)); + io->size = size; + + /* FIXME: make the down a WTD_op */ + if (rw == WRITE) { + unsigned long long tmp; + loff_t limit; + + down(&inode->i_sem); + if (append) + pos = inode->i_size; + + limit = current->rlim[RLIMIT_FSIZE].rlim_cur; + if (likely(RLIM_INFINITY == limit)) + limit = OFFSET_MAX; + + /* Filesystem limits take precedence over user limits */ + if (likely(inode->i_sb->s_maxbytes < limit)) + limit = inode->i_sb->s_maxbytes; + + if (unlikely(pos >= limit)) { + pr_debug("maxbytes: %Ld\n", limit); + ret = 0; + if (size || pos > limit) + ret = -EFBIG; + goto out_io; + } + + /* Clamp writes straddling limit. */ + tmp = pos + size; + if (unlikely(tmp > (unsigned long long)limit)) + size = limit - pos; + } + + if (READ == rw) { + pr_debug("pos=%Ld i_size=%Ld\n", pos, inode->i_size); + + if (pos > inode->i_size) + size = 0; + else if ((pos + size) > inode->i_size) + size = inode->i_size - pos; + + if (io->size < size) + size = io->size; + else if (size < io->size) + io->size = size; + + pr_debug("io->size=%d size=%d\n", io->size, size); + } + + ret = 0; + if (unlikely(!size)) + goto out_io; + + index = pos >> PAGE_CACHE_SHIFT; + eindex = (pos + size - 1) >> PAGE_CACHE_SHIFT; + nr_pages = eindex - index + 1; + + pr_debug("nr_pages: %lu\n", nr_pages); + + io->good_idx = -1; + io->good_page = NULL; + io->did_read = 0; + io->err = 0; + io->rw = rw; + io->as = as; + io->offset = (unsigned long)pos & (PAGE_CACHE_SIZE - 1); + io->file = file; + io->cb = cb; + kvec_dst_init(&io->src, KM_USER0); + kvec_dst_set(&io->src, cb.vec->veclet); + io->veclet = cb.vec->veclet; + if (nr_pages < READDESC_NR_DEF) { + io->pages = io->def_pages; + io->new_pages = io->def_new_pages; + } else { + io->pages = kmalloc(sizeof(*io->pages) * (nr_pages + 1), GFP_KERNEL); + if (!io->pages) + goto out_io; + + io->new_pages = kmalloc(sizeof(*io->new_pages) * (nr_pages + 1), GFP_KERNEL); + if (!io->new_pages) + goto out_pages; + } + + ret = address_space_map(as, index, nr_pages, io->pages, + &io->nr_new_pages, io->new_pages); + pr_debug("as_map: %d (%d new)\n", ret, io->nr_new_pages); + if (ret <= 0) + goto out_new_pages; + + io->nr_pages = ret; + io->pages[io->nr_pages] = NULL; + io->new_pages[io->nr_new_pages] = NULL; + + if (rw == READ) + __generic_file_read_iodesc(io, 0); + else if (rw == WRITE) + __generic_file_write_iodesc(io); + + return 0; + +out_new_pages: + if (io->new_pages != io->def_new_pages) + kfree(io->new_pages); +out_pages: + if (io->pages != io->def_pages) + kfree(io->pages); +out_io: + kfree(io); + + if (rw == WRITE) + up(&inode->i_sem); +out: + if (!ret) + cb.fn(cb.data, cb.vec, ret); + return ret; +} diff -urN v2.4.19-pre5/mm/memory.c linux.diff/mm/memory.c --- v2.4.19-pre5/mm/memory.c Wed Apr 3 21:04:41 2002 +++ linux.diff/mm/memory.c Tue Apr 30 11:25:06 2002 @@ -45,6 +45,8 @@ #include #include #include +#include +#include #include #include @@ -1492,3 +1494,229 @@ } return page; } + +/* + * Force in an entire range of pages from the current process's user VA, + * and pin them in physical memory. + * FIXME: some architectures need to flush the cache based on user addresses + * here. Someone please provide a better macro than flush_cache_page. + */ + +#define dprintk(x...) +atomic_t user_pinned_pages = ATOMIC_INIT(0); + +struct kvec *map_user_kvec(int rw, unsigned long ptr, size_t len) +{ + return mm_map_user_kvec(current->mm, rw, ptr, len); +} + +struct kvec *mm_map_user_kvec(struct mm_struct *mm, int rw, unsigned long ptr, + size_t len) +{ + struct kvec *vec; + struct kveclet *veclet; + unsigned long end; + int err; + struct vm_area_struct * vma = 0; + int i; + int datain = (rw == READ); + unsigned nr_pages; + + end = ptr + len; + if (end < ptr) + return ERR_PTR(-EINVAL); + + nr_pages = (ptr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; + nr_pages -= ptr >> PAGE_SHIFT; + nr_pages ++; + + atomic_add(nr_pages, &user_pinned_pages); + err = -EAGAIN; + if (unlikely(atomic_read(&user_pinned_pages) >= aio_max_pinned)) + goto out_adjust; + + vec = kmalloc(sizeof(struct kvec) + nr_pages * sizeof(struct kveclet), + GFP_KERNEL); + err = -ENOMEM; + if (unlikely(!vec)) + goto out_adjust; + + vec->nr = 0; + vec->max_nr = nr_pages; + veclet = vec->veclet; + + /* Make sure the iobuf is not already mapped somewhere. */ + dprintk ("map_user_kiobuf: begin\n"); + + down_read(&mm->mmap_sem); + + err = -EFAULT; + + i = 0; + + /* + * First of all, try to fault in all of the necessary pages + */ + while (ptr < end) { + struct page *map; + veclet->offset = ptr & ~PAGE_MASK; + veclet->length = PAGE_SIZE - veclet->offset; + if (len < veclet->length) + veclet->length = len; + ptr &= PAGE_MASK; + len -= veclet->length; + + if (!vma || ptr >= vma->vm_end) { + vma = find_vma(mm, ptr); + if (!vma) + goto out_unlock; + if (vma->vm_start > ptr) { + if (!(vma->vm_flags & VM_GROWSDOWN)) + goto out_unlock; + if (expand_stack(vma, ptr)) + goto out_unlock; + } + if (((datain) && (!(vma->vm_flags & VM_WRITE))) || + (!(vma->vm_flags & VM_READ))) { + err = -EFAULT; + goto out_unlock; + } + } + spin_lock(&mm->page_table_lock); + while (!(map = follow_page(mm, ptr, datain))) { + int ret; + + spin_unlock(&mm->page_table_lock); + ret = handle_mm_fault(mm, vma, ptr, datain); + if (ret <= 0) { + if (!ret) + goto out_unlock; + else { + err = -ENOMEM; + goto out_unlock; + } + } + spin_lock(&mm->page_table_lock); + } + map = get_page_map(map); + if (map) { + flush_dcache_page(map); + atomic_inc(&map->count); + } else + printk (KERN_INFO "Mapped page missing [%d]\n", i); + spin_unlock(&mm->page_table_lock); + veclet->page = map; + veclet++; + + ptr += PAGE_SIZE; + vec->nr = ++i; + } + + veclet->page = NULL; /* dummy for the prefetch in free_kvec */ + veclet->length = 0; /* bug checking ;-) */ + + up_read(&mm->mmap_sem); + dprintk ("map_user_kiobuf: end OK\n"); + return vec; + + out_unlock: + up_read(&mm->mmap_sem); + unmap_kvec(vec, 0); + kfree(vec); + printk("map_user_kvec: err(%d) rw=%d\n", err, rw); + return ERR_PTR(err); + + out_adjust: + atomic_sub(nr_pages, &user_pinned_pages); + printk("map_user_kvec: err(%d) rw=%d\n", err, rw); + return ERR_PTR(err); +} + +/* + * Unmap all of the pages referenced by a kiobuf. We release the pages, + * and unlock them if they were locked. + */ + +void unmap_kvec (struct kvec *vec, int dirtied) +{ + struct kveclet *veclet = vec->veclet; + struct kveclet *end = vec->veclet + vec->nr; + struct page *map = veclet->page; + + prefetchw(map); + for (; vecletpage) { + prefetchw(veclet[1].page); + if (likely(map != NULL) && !PageReserved(map)) { + if (dirtied) { + SetPageDirty(map); + flush_dcache_page(map); /* FIXME */ + } + __free_page(map); + } + } + + atomic_sub(vec->max_nr, &user_pinned_pages); + vec->nr = 0; +} + +void free_kvec(struct kvec *vec) +{ + if (unlikely(vec->nr)) + BUG(); + kfree(vec); +} + +/* kvec memory copy helper: appends len bytes in from to dst. + */ +void memcpy_to_kvec_dst(struct kvec_dst *dst, const char *from, long len) +{ + if (unlikely(len < 0)) + BUG(); + do { + int cnt = len; + if (dst->space < cnt) + cnt = dst->space; + + memcpy(dst->dst, from, cnt); + from += cnt; + dst->space -= cnt; + dst->dst += cnt; + len -= cnt; + if (!dst->space && len) { + kvec_dst_unmap(dst); + dst->let++; + dst->offset = 0; + kvec_dst_map(dst); + if (unlikely(!dst->space)) + BUG(); + } + } while (len); +} + +/* kvec memory copy helper: copies and consumes len bytes in from to dst. + */ +void memcpy_from_kvec_dst(char *to, struct kvec_dst *from, long len) +{ + if (unlikely(len < 0)) + BUG(); + do { + int cnt = len; + if (from->space < cnt) + cnt = from->space; + + memcpy(to, from->dst, cnt); + to += cnt; + from->space -= cnt; + from->dst += cnt; + len -= cnt; + if (unlikely(!from->space && len)) { + kvec_dst_unmap(from); + from->let++; + from->offset = 0; + kvec_dst_map(from); + if (unlikely(!from->space)) + BUG(); + } + } while (len); +} + diff -urN v2.4.19-pre5/mm/mmap.c linux.diff/mm/mmap.c --- v2.4.19-pre5/mm/mmap.c Wed Apr 3 21:04:41 2002 +++ linux.diff/mm/mmap.c Mon Apr 29 15:54:22 2002 @@ -14,6 +14,7 @@ #include #include #include +#include #include #include @@ -548,7 +549,17 @@ * Answer: Yes, several device drivers can do it in their * f_op->mmap method. -DaveM */ - addr = vma->vm_start; + if (addr != vma->vm_start) { + struct vm_area_struct *stale_vma; + /* Since addr changed, we rely on the mmap op to prevent + * collisions with existing vmas and just use find_vma_prepare + * to update the tree pointers. + */ + addr = vma->vm_start; + stale_vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); + if (unlikely(stale_vma && stale_vma->vm_start < vma->vm_end)) + BUG(); + } vma_link(mm, vma, prev, rb_link, rb_parent); if (correct_wcount) diff -urN v2.4.19-pre5/mm/slab.c linux.diff/mm/slab.c --- v2.4.19-pre5/mm/slab.c Wed Apr 3 21:04:41 2002 +++ linux.diff/mm/slab.c Thu Apr 18 18:37:10 2002 @@ -447,7 +447,7 @@ * eliminates "false sharing". * Note for systems short on memory removing the alignment will * allow tighter packing of the smaller caches. */ - sprintf(name,"size-%Zd",sizes->cs_size); + snprintf(name, sizeof(name), "size-%Zd",sizes->cs_size); if (!(sizes->cs_cachep = kmem_cache_create(name, sizes->cs_size, 0, SLAB_HWCACHE_ALIGN, NULL, NULL))) { @@ -459,7 +459,7 @@ offslab_limit = sizes->cs_size-sizeof(slab_t); offslab_limit /= 2; } - sprintf(name, "size-%Zd(DMA)",sizes->cs_size); + snprintf(name, sizeof(name), "size-%Zd(DMA)",sizes->cs_size); sizes->cs_dmacachep = kmem_cache_create(name, sizes->cs_size, 0, SLAB_CACHE_DMA|SLAB_HWCACHE_ALIGN, NULL, NULL); if (!sizes->cs_dmacachep) @@ -1841,7 +1841,7 @@ off -= len; \ len = 0; \ } else { \ - if (len-off > count) \ + if (len-off >= count) \ goto t; \ } \ } while (0) @@ -1854,7 +1854,7 @@ /* Output format version, so at least we can change it without _too_ * many complaints. */ - len += sprintf(page+len, "slabinfo - version: 1.1" + len += snprintf(page+len, PAGE_SIZE-len, "slabinfo - version: 1.1" #if STATS " (statistics)" #endif @@ -1902,7 +1902,8 @@ num_slabs+=active_slabs; num_objs = num_slabs*cachep->num; - len += sprintf(page+len, "%-17s %6lu %6lu %6u %4lu %4lu %4u", + len += snprintf(page+len, PAGE_SIZE-len, + "%-17s %6lu %6lu %6u %4lu %4lu %4u", cachep->name, active_objs, num_objs, cachep->objsize, active_slabs, num_slabs, (1<gfporder)); @@ -1914,7 +1915,8 @@ unsigned long reaped = cachep->reaped; unsigned long allocs = cachep->num_allocations; - len += sprintf(page+len, " : %6lu %7lu %5lu %4lu %4lu", + len += snprintf(page+len, PAGE_SIZE-len, + " : %6lu %7lu %5lu %4lu %4lu", high, allocs, grown, reaped, errors); } #endif @@ -1928,7 +1930,7 @@ limit = cc->limit; else limit = 0; - len += sprintf(page+len, " : %4u %4u", + len += snprintf(page+len, PAGE_SIZE-len, " : %4u %4u", limit, batchcount); } #endif @@ -1938,21 +1940,27 @@ unsigned long allocmiss = atomic_read(&cachep->allocmiss); unsigned long freehit = atomic_read(&cachep->freehit); unsigned long freemiss = atomic_read(&cachep->freemiss); - len += sprintf(page+len, " : %6lu %6lu %6lu %6lu", + len += snprintf(page+len, PAGE_SIZE-len, + " : %6lu %6lu %6lu %6lu", allochit, allocmiss, freehit, freemiss); } #endif - len += sprintf(page+len,"\n"); + len += snprintf(page+len, PAGE_SIZE-len, "\n"); spin_unlock_irq(&cachep->spinlock); FIXUP(got_data_up); p = cachep->next.next; + if (len > PAGE_SIZE - 512) + break; } while (p != &cache_cache.next); got_data_up: up(&cache_chain_sem); got_data: - *start = page+off; - return len; + if (off < len) { + *start = page+off; + return len; + } + return 0; } /** diff -urN v2.4.19-pre5/mm/wtd.c linux.diff/mm/wtd.c --- v2.4.19-pre5/mm/wtd.c Wed Dec 31 19:00:00 1969 +++ linux.diff/mm/wtd.c Thu Apr 11 16:17:49 2002 @@ -0,0 +1,73 @@ +#include +#include +#include +#include + +static void __wtd_lock_page_waiter(wait_queue_t *wait) +{ + struct worktodo *wtd = (struct worktodo *)wait; + struct page *page = (struct page *)wtd->data; + + if (!TryLockPage(page)) { + __remove_wait_queue(page_waitqueue(page), &wtd->wait); + wtd_queue(wtd); + } else + schedule_task(&run_disk_tq); +} + +int wtd_lock_page(struct worktodo *wtd, struct page *page) +{ + if (TryLockPage(page)) { + wtd->data = page; + init_waitqueue_func_entry(&wtd->wait, __wtd_lock_page_waiter); + + /* Wakeups may race with TryLockPage, so try again within the wait + * queue spinlock. + */ + if (!add_wait_queue_cond(page_waitqueue(page), &wtd->wait, + TryLockPage(page))) { + /* Page is still locked. Kick the disk queue... */ + run_task_queue(&tq_disk); + return 0; + } + } + + return 1; +} + +static void __wtd_bh_waiter(wait_queue_t *wait) +{ + struct worktodo *wtd = (struct worktodo *)wait; + struct buffer_head *bh = (struct buffer_head *)wtd->data; + + if (!buffer_locked(bh)) { + __remove_wait_queue(&bh->b_wait, &wtd->wait); + wtd_queue(wtd); + } else { + schedule_task(&run_disk_tq); + } +} + +int wtd_wait_on_buffer(struct worktodo *wtd, struct buffer_head *bh) +{ + if (!buffer_locked(bh)) { + return 1; + } + wtd->data = bh; + init_waitqueue_func_entry(&wtd->wait, __wtd_bh_waiter); + if (add_wait_queue_cond(&bh->b_wait, &wtd->wait, buffer_locked(bh))) + return 1; + run_task_queue(&tq_disk); + return 0; +} + +void do_run_tq_disk(void *data) +{ + run_task_queue(&tq_disk); +} + +struct tq_struct run_disk_tq = { + routine: do_run_tq_disk, + data: NULL +}; + diff -urN v2.4.19-pre5/net/core/datagram.c linux.diff/net/core/datagram.c --- v2.4.19-pre5/net/core/datagram.c Tue Jan 1 14:09:35 2002 +++ linux.diff/net/core/datagram.c Mon Apr 8 13:06:39 2002 @@ -8,6 +8,8 @@ * * Authors: Alan Cox . (datagram_poll() from old udp.c code) * + * Portions Copyright 2001 Red Hat, Inc. + * * Fixes: * Alan Cox : NULL return from skb_peek_copy() understood * Alan Cox : Rewrote skb_read_datagram to avoid the skb_peek_copy stuff. @@ -21,6 +23,7 @@ * Darryl Miles : Fixed non-blocking SOCK_STREAM. * Alan Cox : POSIXisms * Pete Wyckoff : Unconnected accept() fix. + * Benjamin LaHaise: added kvec operations * */ @@ -37,6 +40,7 @@ #include #include #include +#include #include #include @@ -446,3 +450,321 @@ return mask; } + +/* + */ +static inline void skb_copy_datagram_kvec_dst(const struct sk_buff *skb, + int offset, struct kvec_dst *dst, int len) +{ + int i, copy; + int start = skb->len - skb->data_len; + + /* Copy header. */ + if ((copy = start-offset) > 0) { + if (copy > len) + copy = len; + memcpy_to_kvec_dst(dst, skb->data + offset, copy); + if ((len -= copy) == 0) + return; + offset += copy; + } + + /* Copy paged appendix. Hmm... why does this look so complicated? */ + for (i=0; inr_frags; i++) { + int end; + + BUG_TRAP(start <= offset+len); + + end = start + skb_shinfo(skb)->frags[i].size; + if ((copy = end-offset) > 0) { + u8 *vaddr; + skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; + struct page *page = frag->page; + + if (copy > len) + copy = len; + vaddr = kmap_atomic(page, KM_USER1); + memcpy_to_kvec_dst(dst, vaddr + frag->page_offset + + offset-start, copy); + kunmap_atomic(vaddr, KM_USER1); + if (!(len -= copy)) + return; + offset += copy; + } + start = end; + } + + if (skb_shinfo(skb)->frag_list) { + struct sk_buff *list; + + for (list = skb_shinfo(skb)->frag_list; list; list=list->next) { + int end; + + BUG_TRAP(start <= offset+len); + + end = start + list->len; + if ((copy = end-offset) > 0) { + if (copy > len) + copy = len; + skb_copy_datagram_kvec_dst(list, offset-start, dst, copy); + if ((len -= copy) == 0) + return; + offset += copy; + } + start = end; + } + } +} + +void skb_copy_datagram_kvec(const struct sk_buff *skb, int offset, + struct kvec *vec, int len) +{ + struct kvec_dst dst; + kvec_dst_init(&dst, KM_USER0); + kvec_dst_set(&dst, vec->veclet); + kvec_dst_map(&dst); + skb_copy_datagram_kvec_dst(skb, offset, &dst, len); + kvec_dst_unmap(&dst); +} + +/* C++ would be better for this. Please don't torture me with this code + * ever again. + */ +static inline unsigned int csum_and_copy_to_dst(struct kvec_dst *dst, + const char *from, int len, unsigned int csum) +{ + do { + int cnt = len; + if (dst->space < cnt) + cnt = dst->space; + + memcpy(dst->dst, from, cnt); + csum = csum_partial_copy_nocheck(from, dst->dst, cnt, csum); + from += cnt; + dst->space -= cnt; + dst->dst += cnt; + len -= cnt; + if (!dst->space && len) { + kvec_dst_unmap(dst); + dst->let++; + dst->offset = 0; + kvec_dst_map(dst); + if (!dst->space) + BUG(); + } + } while (len); + return csum; +} + +static inline void skb_copy_and_csum_datagram_kvec_dst(const struct sk_buff *skb, int offset, struct kvec_dst *dst, int len, unsigned int *csump) +{ + int i, copy; + int start = skb->len - skb->data_len; + int pos = 0; + + /* Copy header. */ + if ((copy = start-offset) > 0) { + if (copy > len) + copy = len; + *csump = csum_and_copy_to_dst(dst, skb->data+offset, copy, *csump); + if ((len -= copy) == 0) + return; + offset += copy; + pos = copy; + } + + for (i=0; inr_frags; i++) { + int end; + + BUG_TRAP(start <= offset+len); + + end = start + skb_shinfo(skb)->frags[i].size; + if ((copy = end-offset) > 0) { + unsigned int csum2; + u8 *vaddr; + skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; + struct page *page = frag->page; + + if (copy > len) + copy = len; + vaddr = kmap_atomic(page, KM_USER1); + csum2 = csum_and_copy_to_dst(dst, + vaddr + frag->page_offset + offset-start, + copy, 0); + kunmap_atomic(vaddr, KM_USER1); + *csump = csum_block_add(*csump, csum2, pos); + if (!(len -= copy)) + return; + offset += copy; + pos += copy; + } + start = end; + } + + if (skb_shinfo(skb)->frag_list) { + struct sk_buff *list; + + for (list = skb_shinfo(skb)->frag_list; list; list=list->next) { + int end; + + BUG_TRAP(start <= offset+len); + + end = start + list->len; + if ((copy = end-offset) > 0) { + unsigned int csum2 = 0; + if (copy > len) + copy = len; + skb_copy_and_csum_datagram_kvec_dst(list, offset-start, dst, copy, &csum2); + *csump = csum_block_add(*csump, csum2, pos); + if ((len -= copy) == 0) + return; + offset += copy; + pos += copy; + } + start = end; + } + } +} + +int skb_copy_and_csum_datagram_kvec(const struct sk_buff *skb, int offset, + struct kvec *vec, int len) +{ + unsigned int csum; + struct kvec_dst dst; + + csum = csum_partial(skb->data, offset, skb->csum); + + kvec_dst_init(&dst, KM_USER0); + kvec_dst_set(&dst, vec->veclet); + kvec_dst_map(&dst); + skb_copy_and_csum_datagram_kvec_dst(skb, offset, &dst, len, &csum); + kvec_dst_unmap(&dst); + + if ((unsigned short)csum_fold(csum)) + return -EINVAL; + return 0; +} + +struct skb_async_info { + struct worktodo wtd; + struct sock *sk; + int len; + void (*finish)(struct sock *sk, kvec_cb_t cb, int len, struct sk_buff *skb); + kvec_cb_t cb; +}; +static void skb_async_read_worker(void *_data); + +int skb_kvec_recv_datagram(struct sock * sk, kvec_cb_t cb, int len, + void (*finish)(struct sock *sk, kvec_cb_t cb, int len, struct sk_buff *skb)) +{ + struct skb_async_info *info = kmalloc(sizeof(struct skb_async_info), GFP_KERNEL); + if (info) { + wtd_set_action(&info->wtd, skb_async_read_worker, info); + info->sk = sk; + info->len = len; + info->finish = finish; + info->cb = cb; + skb_async_read_worker(info); + return 0; + } + return -EAGAIN; +} + +static void skb_async_read_waiter(wait_queue_t *wait) +{ + struct skb_async_info *info = (void *)wait; + __remove_wait_queue(info->sk->sleep, &info->wtd.wait); + wtd_queue(&info->wtd); +} + +static void skb_async_read_worker(void *_data) +{ + struct skb_async_info *info = _data; + struct sock *sk = info->sk; + struct sk_buff *skb; + int error; + + /* Caller is allowed not to check sk->err before skb_recv_datagram() */ + error = sock_error(sk); + if (error) + goto no_packet; + + + init_waitqueue_func_entry(&info->wtd.wait, skb_async_read_waiter); + + /* Attempted to dequeue and process any skbs that already arrived. + * Note that add_wait_queue_cond is used to check against a race + * where an skb is added to the queue after we checked but before + * the callback is added to the wait queue. + */ + do { + skb = skb_dequeue(&sk->receive_queue); + if (skb) { + info->finish(sk, info->cb, info->len, skb); + kfree(info); + return; + } + } while ( add_wait_queue_cond( sk->sleep, &info->wtd.wait, + (!(error = sock_error(sk)) && + skb_queue_empty(&sk->receive_queue)) ) + && !error); + + if (!error) + return; + +no_packet: + info->cb.fn(info->cb.data, info->cb.vec, error); + kfree(info); + return; +} + +#if 0 +static void skb_async_read_worker(void *_data) +{ + struct skb_async_info *info = _data; + int error; + + /* Socket errors? */ + error = sock_error(sk); + if (error) + goto out_err; + + if (!skb_queue_empty(&sk->receive_queue)) + goto ready; + + /* Socket shut down? */ + if (sk->shutdown & RCV_SHUTDOWN) + goto out_noerr; + + /* Sequenced packets can come disconnected. If so we report the problem */ + error = -ENOTCONN; + if(connection_based(sk) && !(sk->state==TCP_ESTABLISHED || sk->state==TCP_LISTEN)) + goto out_err; + + /* handle signals */ + if (signal_pending(current)) + goto interrupted; + + /* here: queue sleep */ + *timeo_p = schedule_timeout(*timeo_p); + return; + +ready: + current->state = TASK_RUNNING; + remove_wait_queue(sk->sleep, &wait); + return 0; + +interrupted: + error = sock_intr_errno(*timeo_p); +out_err: + *err = error; +out: + current->state = TASK_RUNNING; + remove_wait_queue(sk->sleep, &wait); + return error; +out_noerr: + *err = 0; + error = 1; + goto out; +} +#endif diff -urN v2.4.19-pre5/net/core/sock.c linux.diff/net/core/sock.c --- v2.4.19-pre5/net/core/sock.c Tue Jan 1 14:09:35 2002 +++ linux.diff/net/core/sock.c Mon Apr 8 00:37:02 2002 @@ -586,6 +586,8 @@ if(sk && zero_it) { memset(sk, 0, sizeof(struct sock)); sk->family = family; + INIT_LIST_HEAD(&sk->kvec_read_list); + INIT_LIST_HEAD(&sk->kvec_write_list); sock_lock_init(sk); } @@ -1116,7 +1118,7 @@ void sock_def_wakeup(struct sock *sk) { read_lock(&sk->callback_lock); - if (sk->sleep && waitqueue_active(sk->sleep)) + if (sk->sleep) wake_up_interruptible_all(sk->sleep); read_unlock(&sk->callback_lock); } @@ -1124,7 +1126,7 @@ void sock_def_error_report(struct sock *sk) { read_lock(&sk->callback_lock); - if (sk->sleep && waitqueue_active(sk->sleep)) + if (sk->sleep) wake_up_interruptible(sk->sleep); sk_wake_async(sk,0,POLL_ERR); read_unlock(&sk->callback_lock); @@ -1133,7 +1135,7 @@ void sock_def_readable(struct sock *sk, int len) { read_lock(&sk->callback_lock); - if (sk->sleep && waitqueue_active(sk->sleep)) + if (sk->sleep) wake_up_interruptible(sk->sleep); sk_wake_async(sk,1,POLL_IN); read_unlock(&sk->callback_lock); @@ -1147,7 +1149,7 @@ * progress. --DaveM */ if((atomic_read(&sk->wmem_alloc) << 1) <= sk->sndbuf) { - if (sk->sleep && waitqueue_active(sk->sleep)) + if (sk->sleep) wake_up_interruptible(sk->sleep); /* Should agree with poll, otherwise some programs break */ diff -urN v2.4.19-pre5/net/ipv4/af_inet.c linux.diff/net/ipv4/af_inet.c --- v2.4.19-pre5/net/ipv4/af_inet.c Wed Apr 3 21:04:41 2002 +++ linux.diff/net/ipv4/af_inet.c Mon Apr 8 00:37:02 2002 @@ -729,6 +729,19 @@ } +int inet_kvec_read(struct socket *sock, kvec_cb_t cb, size_t len) +{ + struct sock *sk = sock->sk; + + return sk->prot->kvec_read(sk, cb, len); +} + +int inet_kvec_write(struct socket *sock, kvec_cb_t cb, size_t len) +{ + struct sock *sk = sock->sk; + + return sk->prot->kvec_write(sk, cb, len); +} int inet_recvmsg(struct socket *sock, struct msghdr *msg, int size, int flags, struct scm_cookie *scm) @@ -960,7 +973,9 @@ sendmsg: inet_sendmsg, recvmsg: inet_recvmsg, mmap: sock_no_mmap, - sendpage: tcp_sendpage + sendpage: tcp_sendpage, + kvec_read: inet_kvec_read, + kvec_write: inet_kvec_write, }; struct proto_ops inet_dgram_ops = { @@ -982,6 +997,8 @@ recvmsg: inet_recvmsg, mmap: sock_no_mmap, sendpage: sock_no_sendpage, + kvec_read: inet_kvec_read, + kvec_write: inet_kvec_write, }; struct net_proto_family inet_family_ops = { diff -urN v2.4.19-pre5/net/ipv4/tcp.c linux.diff/net/ipv4/tcp.c --- v2.4.19-pre5/net/ipv4/tcp.c Wed Apr 3 21:04:41 2002 +++ linux.diff/net/ipv4/tcp.c Tue Apr 30 17:29:31 2002 @@ -251,6 +251,7 @@ #include #include #include +#include #include #include @@ -470,8 +471,8 @@ if (sk->sleep && waitqueue_active(sk->sleep)) wake_up_interruptible(sk->sleep); - if (sock->fasync_list && !(sk->shutdown&SEND_SHUTDOWN)) - sock_wake_async(sock, 2, POLL_OUT); + if (!(sk->shutdown&SEND_SHUTDOWN)) + sk_wake_async(sk, 2, POLL_OUT); } } @@ -676,11 +677,266 @@ return 0; } +struct tcp_write_async_info { + struct worktodo wtd; + struct sock *sk; + int len; + int done; + int offset; + struct kveclet *cur_let; + kvec_cb_t cb; + spinlock_t lock; +}; + +static void async_lock_sock_wait(wait_queue_t *wait) +{ + struct tcp_write_async_info *info = (void *)wait; + printk("async_lock_sock_wait(%p)\n", info); + if (!info->sk->lock.users) { + printk("async_lock_sock_wait: queuing\n"); + __remove_wait_queue(info->sk->sleep, &info->wtd.wait); + wtd_queue(&info->wtd); + } +} + +static void async_lock_sock(void *data) +{ + struct tcp_write_async_info *info = data; + struct sock *sk; + printk(KERN_DEBUG "async_lock_sock(%p)\n", info); + sk = info->sk; + spin_lock_bh(&sk->lock.slock); + if (sk->lock.users) { + printk(KERN_DEBUG "async_lock_sock: waiting\n"); + wtd_push(&info->wtd, async_lock_sock, info); + init_waitqueue_func_entry(&info->wtd.wait, async_lock_sock_wait); + if (!add_wait_queue_cond(sk->sleep, &info->wtd.wait, !sk->lock.users)) { + spin_unlock_bh(&sk->lock.slock); + return; + } + wtd_pop(&info->wtd); + } + printk(KERN_DEBUG "async_lock_sock: locking\n"); + sk->lock.users = 1; + spin_unlock_bh(&sk->lock.slock); + wtd_queue(&info->wtd); +} + +static void async_wait_for_tcp_connect(void *data); +int tcp_kvec_write(struct sock *sk, kvec_cb_t cb, int len) +{ + struct tcp_write_async_info *info; + info = kmalloc(sizeof(*info), GFP_KERNEL); + printk(KERN_DEBUG "tcp_kvec_write: %p\n", info); + if (!info) + return -ENOMEM; + wtd_init(&info->wtd, async_wait_for_tcp_connect); + info->sk = sk; + info->len = len; + info->done = 0; + info->offset = 0; + info->cur_let = cb.vec->veclet; + info->cb = cb; + spin_lock_init(&info->lock); + async_lock_sock(info); + return 0; +} + +static void async_cn_wait_task(void *data) +{ + struct tcp_write_async_info *info = (void *)data; + async_lock_sock(info); +} + +static void async_cn_wait(wait_queue_t *wait) +{ + struct tcp_write_async_info *info = (void *)wait; + __remove_wait_queue(info->sk->sleep, &info->wtd.wait); + wtd_set_action(&info->wtd, async_cn_wait_task, info); + wtd_queue(&info->wtd); +} + +/* sock_get_iocb + * Attempts to allocate a local socket iocb, which allows high + * performance for the common cases of a small number of ios + * outstanding per socket. + */ +struct sock_iocb *sock_get_iocb(struct sock *sk) +{ + struct sock_iocb *iocb; + + iocb = kmalloc(sizeof(*iocb), GFP_KERNEL); + return iocb; +} + +void sock_put_iocb(struct sock_iocb *iocb) +{ + kfree(iocb); +} + +/* tcp_kvec_read_kick + * Attempts to process an async read request. Must be called with + * the socket lock held. + */ +void tcp_kvec_read_kick(struct sock *sk, struct sock_iocb *iocb) +{ + TCP_CHECK_TIMER(sk); +#if 0 + if (unlikely(TCP_LISTEN == sk->state)) + goto out; +#endif + return; +} + +/* tcp_kvec_read + * Queues an async read request on a socket. If there were + & no outstanding read requests, kicks the backlog processing. + */ +int tcp_kvec_read(struct sock *sk, kvec_cb_t cb, int size) +{ + struct sock_iocb *iocb; + printk("tcp_kvec_read(%p, %d): blah", sk, size); + + iocb = sock_get_iocb(sk); + if (unlikely(NULL == iocb)) + return -ENOMEM; + + iocb->cb = cb; + kvec_dst_init(&iocb->dst, KM_USER0); + + spin_lock_bh(&sk->lock.slock); + if (sk->lock.users != 0 || !list_empty(&sk->kvec_read_list)) { + list_add_tail(&iocb->list, &sk->kvec_read_list); + spin_unlock_bh(&sk->lock.slock); + return 0; + } + spin_unlock_bh(&sk->lock.slock); + + /* We're the head read request and now own the socket lock; + * attempt to kick off processing. + */ + tcp_kvec_read_kick(sk, iocb); + release_sock(sk); + return 0; +} + +static void tcp_kvec_write_worker(struct tcp_write_async_info *info); +static void async_wait_for_tcp_connect(void *data) +{ + struct tcp_write_async_info *info = data; + struct sock *sk = info->sk; + int err; + /* At this point the socket is locked for us. */ + while((1 << sk->state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) { + if (sk->err) { + err = sock_error(sk); + goto error; + } + if ((1 << sk->state) & + ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) { + err = -EPIPE; + goto error; + } + + sk->tp_pinfo.af_tcp.write_pending++; + init_waitqueue_func_entry(&info->wtd.wait, async_cn_wait); + + /* Add our worker to the socket queue, but make sure the socket + * state isn't changed from when we checked while we do so. + */ + if (!add_wait_queue_cond(sk->sleep, &info->wtd.wait, + ((1 << sk->state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) + )) { + release_sock(sk); + return; + } + } + /* sk is now locked *and* the connection is established, let's + * proceed to the data transfer stage. + */ + tcp_kvec_write_worker(info); + return; + +error: + release_sock(sk); + info->cb.fn(info->cb.data, info->cb.vec, err); + kfree(info); +} + static inline int tcp_memory_free(struct sock *sk) { return sk->wmem_queued < sk->sndbuf; } +static void async_wait_for_tcp_memory(struct tcp_write_async_info *info); +static void async_wait_for_tcp_memory_done(void *data) +{ + struct tcp_write_async_info *info = data; + info->sk->tp_pinfo.af_tcp.write_pending--; + if (tcp_memory_free(info->sk)) + tcp_kvec_write_worker(info); + else + async_wait_for_tcp_memory(info); +} + +static void async_wait_for_tcp_memory_waiting(void *data) +{ + struct tcp_write_async_info *info = data; + wtd_set_action(&info->wtd, async_wait_for_tcp_memory_done, info); + async_lock_sock(info); +} + +static void async_wait_for_tcp_memory_wake(wait_queue_t *wait) +{ + struct tcp_write_async_info *info = (void *)wait; + __remove_wait_queue(info->sk->sleep, &info->wtd.wait); + wtd_set_action(&info->wtd, async_wait_for_tcp_memory_waiting, info); + wtd_queue(&info->wtd); +} + +static void async_wait_for_tcp_memory(struct tcp_write_async_info *info) +{ + struct sock *sk = info->sk; + ssize_t res; + kvec_cb_t cb; + int raced = 0; + + printk("async_wait_for_tcp_memory(%p)\n", info); + res = -EPIPE; + if (sk->err || (sk->shutdown & SEND_SHUTDOWN)) + goto err; + + if (tcp_memory_free(sk)) + printk("async_wait_for_tcp_memory: spinning?\n"); + + init_waitqueue_func_entry(&info->wtd.wait, async_wait_for_tcp_memory_wake); + clear_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags); + set_bit(SOCK_NOSPACE, &sk->socket->flags); + raced = add_wait_queue_cond( sk->sleep, &info->wtd.wait, + !(sk->err || (sk->shutdown & SEND_SHUTDOWN) || tcp_memory_free(sk)) ); + + sk->tp_pinfo.af_tcp.write_pending++; + if (raced) { + /* Requeue to be run here: this allows other tasks to + * get rescheduled in case of bugs + */ + wtd_set_action(&info->wtd, async_wait_for_tcp_memory_done, info); + wtd_queue(&info->wtd); + return; + } + + release_sock(sk); + return; + +err: + printk("async_wait_for_tcp_memory: err %ld\n", (long)res); + if (info->done) + res = info->done; + cb = info->cb; + kfree(info); + cb.fn(cb.data, cb.vec, res); +} + /* * Wait for more memory for a socket */ @@ -691,9 +947,17 @@ long current_timeo = *timeo; DECLARE_WAITQUEUE(wait, current); + if (sk->err || (sk->shutdown & SEND_SHUTDOWN)) + return -EPIPE; + if (tcp_memory_free(sk)) current_timeo = vm_wait = (net_random()%(HZ/5))+2; + if (!*timeo) { + set_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags); + return -EAGAIN; + } + add_wait_queue(sk->sleep, &wait); for (;;) { set_bit(SOCK_ASYNC_NOSPACE, &sk->socket->flags); @@ -744,7 +1008,7 @@ goto out; } -ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset, size_t psize, int flags); +ssize_t do_tcp_sendpages(struct sock *sk, struct kveclet *let, int poffset, size_t psize, int flags); static inline int can_coalesce(struct sk_buff *skb, int i, struct page *page, int off) @@ -823,7 +1087,7 @@ return err; } -ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset, size_t psize, int flags) +ssize_t do_tcp_sendpages(struct sock *sk, struct kveclet *let, int poffset, size_t psize, int flags) { struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp); int mss_now; @@ -845,14 +1109,19 @@ if (sk->err || (sk->shutdown & SEND_SHUTDOWN)) goto do_error; + while (poffset > let->length) { + poffset -= let->length; + let++; + } + while (psize > 0) { struct sk_buff *skb = sk->write_queue.prev; int offset, size, copy, i; struct page *page; - page = pages[poffset/PAGE_SIZE]; - offset = poffset % PAGE_SIZE; - size = min_t(size_t, psize, PAGE_SIZE-offset); + page = let->page; + offset = let->offset; + size = min_t(unsigned int, psize, let->length); if (tp->send_head==NULL || (copy = mss_now - skb->len) <= 0) { new_segment: @@ -892,6 +1161,10 @@ copied += copy; poffset += copy; + if (poffset >= let->length) { + poffset = 0; + let++; + } if (!(psize -= copy)) goto out; @@ -931,6 +1204,7 @@ ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) { + struct kveclet let = { page, 0, PAGE_SIZE }; ssize_t res; struct sock *sk = sock->sk; @@ -940,16 +1214,54 @@ !(sk->route_caps & TCP_ZC_CSUM_FLAGS)) return sock_no_sendpage(sock, page, offset, size, flags); -#undef TCP_ZC_CSUM_FLAGS lock_sock(sk); TCP_CHECK_TIMER(sk); - res = do_tcp_sendpages(sk, &page, offset, size, flags); + res = do_tcp_sendpages(sk, &let, offset, size, flags); TCP_CHECK_TIMER(sk); release_sock(sk); return res; } +static void tcp_kvec_write_worker(struct tcp_write_async_info *info) +{ + struct sock *sk = info->sk; + int res; + if (!(sk->route_caps & NETIF_F_SG) || + !(sk->route_caps & TCP_ZC_CSUM_FLAGS)) + BUG(); + + res = do_tcp_sendpages(sk, info->cur_let, info->offset, info->len - info->done, MSG_DONTWAIT); + if (res > 0) + info->done += res; + + if (res == -EAGAIN) { + printk("tcp_kvec_write_worker: -EAGAIN: queuing\n"); + goto requeue; + } + + while (res > info->cur_let->length) { + res -= info->cur_let->length; + info->cur_let++; + } + + if (res <= 0 || (info->done >= info->len)) { + kvec_cb_t cb = info->cb; + printk("tcp_kvec_write_worker: error(%d)\n", res); + if (info->done) + res = info->done; + release_sock(sk); + kfree(info); + cb.fn(cb.data, cb.vec, res); + return; + } + +requeue: + async_wait_for_tcp_memory(info); +} + +#undef TCP_ZC_CSUM_FLAGS + #define TCP_PAGE(sk) (sk->tp_pinfo.af_tcp.sndmsg_page) #define TCP_OFF(sk) (sk->tp_pinfo.af_tcp.sndmsg_off) diff -urN v2.4.19-pre5/net/ipv4/tcp_ipv4.c linux.diff/net/ipv4/tcp_ipv4.c --- v2.4.19-pre5/net/ipv4/tcp_ipv4.c Wed Apr 3 21:04:41 2002 +++ linux.diff/net/ipv4/tcp_ipv4.c Mon Apr 8 00:37:02 2002 @@ -2299,6 +2299,8 @@ hash: tcp_v4_hash, unhash: tcp_unhash, get_port: tcp_v4_get_port, + kvec_read: tcp_kvec_read, + kvec_write: tcp_kvec_write, }; diff -urN v2.4.19-pre5/net/ipv4/udp.c linux.diff/net/ipv4/udp.c --- v2.4.19-pre5/net/ipv4/udp.c Wed Apr 3 21:04:42 2002 +++ linux.diff/net/ipv4/udp.c Mon Apr 8 00:37:02 2002 @@ -93,6 +93,7 @@ #include #include #include +#include /* * Snmp MIB for the UDP layer @@ -619,6 +620,74 @@ __udp_checksum_complete(skb); } +void udp_kvec_read_finish(struct sock *sk, kvec_cb_t cb, int len, struct sk_buff *skb) +{ + struct sockaddr_in *sin = NULL; + int msg_flags = 0; + int copied, err; + + if (!skb) + BUG(); + + copied = skb->len - sizeof(struct udphdr); + if (copied > len) { + copied = len; + msg_flags |= MSG_TRUNC; + } + + err = 0; + + if (skb->ip_summed==CHECKSUM_UNNECESSARY) { + skb_copy_datagram_kvec(skb, sizeof(struct udphdr), + cb.vec, copied); + } else if (msg_flags&MSG_TRUNC) { + err = -EAGAIN; + if (unlikely(__udp_checksum_complete(skb))) { + UDP_INC_STATS_BH(UdpInErrors); + goto out_free; + } + err = 0; + skb_copy_datagram_kvec(skb, sizeof(struct udphdr), + cb.vec, copied); + } else { + err = skb_copy_and_csum_datagram_kvec(skb, + sizeof(struct udphdr), cb.vec, copied); + } + + if (err) + goto out_free; + + //sock_recv_timestamp(msg, sk, skb); + + /* Copy the address. */ + if (sin) + { + sin->sin_family = AF_INET; + sin->sin_port = skb->h.uh->source; + sin->sin_addr.s_addr = skb->nh.iph->saddr; + memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); + } + //if (sk->protinfo.af_inet.cmsg_flags) + // ip_cmsg_recv(msg, skb); + err = copied; + +out_free: + skb_free_datagram(sk, skb); + cb.fn(cb.data, cb.vec, err); + return; +} + +static int udp_kvec_read(struct sock *sk, kvec_cb_t cb, int len) +{ + return skb_kvec_recv_datagram(sk, cb, len, udp_kvec_read_finish); +} + +static int udp_kvec_write(struct sock *sk, kvec_cb_t cb, int len) +{ + return -EINVAL; /* TODO: someone please write ;-) */ +} + + /* * This should be easy, if there is something there we * return it, otherwise we block. @@ -1037,6 +1106,8 @@ getsockopt: ip_getsockopt, sendmsg: udp_sendmsg, recvmsg: udp_recvmsg, + kvec_read: udp_kvec_read, + kvec_write: udp_kvec_write, backlog_rcv: udp_queue_rcv_skb, hash: udp_v4_hash, unhash: udp_v4_unhash, diff -urN v2.4.19-pre5/net/khttpd/datasending.c linux.diff/net/khttpd/datasending.c --- v2.4.19-pre5/net/khttpd/datasending.c Mon Sep 24 02:16:05 2001 +++ linux.diff/net/khttpd/datasending.c Tue Apr 2 18:56:57 2002 @@ -127,7 +127,7 @@ desc.count = ReadSize; desc.buf = (char *) CurrentRequest->sock; desc.error = 0; - do_generic_file_read(CurrentRequest->filp, ppos, &desc, sock_send_actor); + do_generic_file_read(CurrentRequest->filp, ppos, &desc, sock_send_actor, 0); if (desc.written>0) { CurrentRequest->BytesSent += desc.written; diff -urN v2.4.19-pre5/net/socket.c linux.diff/net/socket.c --- v2.4.19-pre5/net/socket.c Wed Apr 3 21:04:42 2002 +++ linux.diff/net/socket.c Mon Apr 8 13:16:30 2002 @@ -44,6 +44,7 @@ * Tigran Aivazian : sys_send(args) calls sys_sendto(args, NULL, 0) * Tigran Aivazian : Made listen(2) backlog sanity checks * protocol-independent + * Benjamin LaHaise: real aio support. * * * This program is free software; you can redistribute it and/or @@ -104,6 +105,8 @@ unsigned long count, loff_t *ppos); static ssize_t sock_sendpage(struct file *file, struct page *page, int offset, size_t size, loff_t *ppos, int more); +static int sock_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos); +static int sock_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos); /* @@ -123,7 +126,11 @@ fasync: sock_fasync, readv: sock_readv, writev: sock_writev, - sendpage: sock_sendpage + sendpage: sock_sendpage, + aio_read: generic_sock_aio_read, + aio_write: generic_file_aio_write, + kvec_read: sock_kvec_read, + kvec_write: sock_kvec_write, }; /* @@ -533,13 +540,14 @@ static ssize_t sock_read(struct file *file, char *ubuf, size_t size, loff_t *ppos) { + int read_flags = 0; struct socket *sock; struct iovec iov; struct msghdr msg; int flags; - if (ppos != &file->f_pos) - return -ESPIPE; + if (read_flags & ~F_ATOMIC) + return -EINVAL; if (size==0) /* Match SYS5 behaviour */ return 0; @@ -554,6 +562,8 @@ iov.iov_base=ubuf; iov.iov_len=size; flags = !(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT; + if (read_flags & F_ATOMIC) + flags |= MSG_DONTWAIT; return sock_recvmsg(sock, &msg, size, flags); } @@ -567,12 +577,13 @@ static ssize_t sock_write(struct file *file, const char *ubuf, size_t size, loff_t *ppos) { + int flags = 0; struct socket *sock; struct msghdr msg; struct iovec iov; - - if (ppos != &file->f_pos) - return -ESPIPE; + + if (flags & ~F_ATOMIC) + return -EINVAL; if(size==0) /* Match SYS5 behaviour */ return 0; @@ -585,6 +596,8 @@ msg.msg_control=NULL; msg.msg_controllen=0; msg.msg_flags=!(file->f_flags & O_NONBLOCK) ? 0 : MSG_DONTWAIT; + if (flags & F_ATOMIC) + msg.msg_flags = MSG_DONTWAIT; if (sock->type == SOCK_SEQPACKET) msg.msg_flags |= MSG_EOR; iov.iov_base=(void *)ubuf; @@ -611,6 +624,29 @@ return sock->ops->sendpage(sock, page, offset, size, flags); } +static int sock_kvec_read(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + struct socket *sock; + sock = socki_lookup(file->f_dentry->d_inode); + if ((int)size < 0 || (size_t)(int)size != size) + return -EINVAL; + if (sock->ops->kvec_read) + return sock->ops->kvec_read(sock, cb, size); + return -EOPNOTSUPP; +} + +static int sock_kvec_write(struct file *file, kvec_cb_t cb, size_t size, loff_t pos) +{ + struct socket *sock; + sock = socki_lookup(file->f_dentry->d_inode); + if ((int)size < 0 || (size_t)(int)size != size) + return -EINVAL; + if (sock->ops->kvec_write) + return sock->ops->kvec_write(sock, cb, size); + return -EOPNOTSUPP; +} + + int sock_readv_writev(int type, struct inode * inode, struct file * file, const struct iovec * iov, long count, long size) { diff -urN v2.4.19-pre5/ulib/COPYING linux.diff/ulib/COPYING --- v2.4.19-pre5/ulib/COPYING Wed Dec 31 19:00:00 1969 +++ linux.diff/ulib/COPYING Fri Apr 19 20:54:05 2002 @@ -0,0 +1,515 @@ + + GNU LESSER GENERAL PUBLIC LICENSE + Version 2.1, February 1999 + + Copyright (C) 1991, 1999 Free Software Foundation, Inc. + 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + +[This is the first released version of the Lesser GPL. It also counts + as the successor of the GNU Library Public License, version 2, hence + the version number 2.1.] + + Preamble + + The licenses for most software are designed to take away your +freedom to share and change it. By contrast, the GNU General Public +Licenses are intended to guarantee your freedom to share and change +free software--to make sure the software is free for all its users. + + This license, the Lesser General Public License, applies to some +specially designated software packages--typically libraries--of the +Free Software Foundation and other authors who decide to use it. You +can use it too, but we suggest you first think carefully about whether +this license or the ordinary General Public License is the better +strategy to use in any particular case, based on the explanations +below. + + When we speak of free software, we are referring to freedom of use, +not price. Our General Public Licenses are designed to make sure that +you have the freedom to distribute copies of free software (and charge +for this service if you wish); that you receive source code or can get +it if you want it; that you can change the software and use pieces of +it in new free programs; and that you are informed that you can do +these things. + + To protect your rights, we need to make restrictions that forbid +distributors to deny you these rights or to ask you to surrender these +rights. These restrictions translate to certain responsibilities for +you if you distribute copies of the library or if you modify it. + + For example, if you distribute copies of the library, whether gratis +or for a fee, you must give the recipients all the rights that we gave +you. You must make sure that they, too, receive or can get the source +code. If you link other code with the library, you must provide +complete object files to the recipients, so that they can relink them +with the library after making changes to the library and recompiling +it. And you must show them these terms so they know their rights. + + We protect your rights with a two-step method: (1) we copyright the +library, and (2) we offer you this license, which gives you legal +permission to copy, distribute and/or modify the library. + + To protect each distributor, we want to make it very clear that +there is no warranty for the free library. Also, if the library is +modified by someone else and passed on, the recipients should know +that what they have is not the original version, so that the original +author's reputation will not be affected by problems that might be +introduced by others. +^L + Finally, software patents pose a constant threat to the existence of +any free program. We wish to make sure that a company cannot +effectively restrict the users of a free program by obtaining a +restrictive license from a patent holder. Therefore, we insist that +any patent license obtained for a version of the library must be +consistent with the full freedom of use specified in this license. + + Most GNU software, including some libraries, is covered by the +ordinary GNU General Public License. This license, the GNU Lesser +General Public License, applies to certain designated libraries, and +is quite different from the ordinary General Public License. We use +this license for certain libraries in order to permit linking those +libraries into non-free programs. + + When a program is linked with a library, whether statically or using +a shared library, the combination of the two is legally speaking a +combined work, a derivative of the original library. The ordinary +General Public License therefore permits such linking only if the +entire combination fits its criteria of freedom. The Lesser General +Public License permits more lax criteria for linking other code with +the library. + + We call this license the "Lesser" General Public License because it +does Less to protect the user's freedom than the ordinary General +Public License. It also provides other free software developers Less +of an advantage over competing non-free programs. These disadvantages +are the reason we use the ordinary General Public License for many +libraries. However, the Lesser license provides advantages in certain +special circumstances. + + For example, on rare occasions, there may be a special need to +encourage the widest possible use of a certain library, so that it +becomes +a de-facto standard. To achieve this, non-free programs must be +allowed to use the library. A more frequent case is that a free +library does the same job as widely used non-free libraries. In this +case, there is little to gain by limiting the free library to free +software only, so we use the Lesser General Public License. + + In other cases, permission to use a particular library in non-free +programs enables a greater number of people to use a large body of +free software. For example, permission to use the GNU C Library in +non-free programs enables many more people to use the whole GNU +operating system, as well as its variant, the GNU/Linux operating +system. + + Although the Lesser General Public License is Less protective of the +users' freedom, it does ensure that the user of a program that is +linked with the Library has the freedom and the wherewithal to run +that program using a modified version of the Library. + + The precise terms and conditions for copying, distribution and +modification follow. Pay close attention to the difference between a +"work based on the library" and a "work that uses the library". The +former contains code derived from the library, whereas the latter must +be combined with the library in order to run. +^L + GNU LESSER GENERAL PUBLIC LICENSE + TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION + + 0. This License Agreement applies to any software library or other +program which contains a notice placed by the copyright holder or +other authorized party saying it may be distributed under the terms of +this Lesser General Public License (also called "this License"). +Each licensee is addressed as "you". + + A "library" means a collection of software functions and/or data +prepared so as to be conveniently linked with application programs +(which use some of those functions and data) to form executables. + + The "Library", below, refers to any such software library or work +which has been distributed under these terms. A "work based on the +Library" means either the Library or any derivative work under +copyright law: that is to say, a work containing the Library or a +portion of it, either verbatim or with modifications and/or translated +straightforwardly into another language. (Hereinafter, translation is +included without limitation in the term "modification".) + + "Source code" for a work means the preferred form of the work for +making modifications to it. For a library, complete source code means +all the source code for all modules it contains, plus any associated +interface definition files, plus the scripts used to control +compilation +and installation of the library. + + Activities other than copying, distribution and modification are not +covered by this License; they are outside its scope. The act of +running a program using the Library is not restricted, and output from +such a program is covered only if its contents constitute a work based +on the Library (independent of the use of the Library in a tool for +writing it). Whether that is true depends on what the Library does +and what the program that uses the Library does. + + 1. You may copy and distribute verbatim copies of the Library's +complete source code as you receive it, in any medium, provided that +you conspicuously and appropriately publish on each copy an +appropriate copyright notice and disclaimer of warranty; keep intact +all the notices that refer to this License and to the absence of any +warranty; and distribute a copy of this License along with the +Library. + + You may charge a fee for the physical act of transferring a copy, +and you may at your option offer warranty protection in exchange for a +fee. + + 2. You may modify your copy or copies of the Library or any portion +of it, thus forming a work based on the Library, and copy and +distribute such modifications or work under the terms of Section 1 +above, provided that you also meet all of these conditions: + + a) The modified work must itself be a software library. + + b) You must cause the files modified to carry prominent notices + stating that you changed the files and the date of any change. + + c) You must cause the whole of the work to be licensed at no + charge to all third parties under the terms of this License. + + d) If a facility in the modified Library refers to a function or a + table of data to be supplied by an application program that uses + the facility, other than as an argument passed when the facility + is invoked, then you must make a good faith effort to ensure that, + in the event an application does not supply such function or + table, the facility still operates, and performs whatever part of + its purpose remains meaningful. + + (For example, a function in a library to compute square roots has + a purpose that is entirely well-defined independent of the + application. Therefore, Subsection 2d requires that any + application-supplied function or table used by this function must + be optional: if the application does not supply it, the square + root function must still compute square roots.) + +These requirements apply to the modified work as a whole. If +identifiable sections of that work are not derived from the Library, +and can be reasonably considered independent and separate works in +themselves, then this License, and its terms, do not apply to those +sections when you distribute them as separate works. But when you +distribute the same sections as part of a whole which is a work based +on the Library, the distribution of the whole must be on the terms of +this License, whose permissions for other licensees extend to the +entire whole, and thus to each and every part regardless of who wrote +it. + +Thus, it is not the intent of this section to claim rights or contest +your rights to work written entirely by you; rather, the intent is to +exercise the right to control the distribution of derivative or +collective works based on the Library. + +In addition, mere aggregation of another work not based on the Library +with the Library (or with a work based on the Library) on a volume of +a storage or distribution medium does not bring the other work under +the scope of this License. + + 3. You may opt to apply the terms of the ordinary GNU General Public +License instead of this License to a given copy of the Library. To do +this, you must alter all the notices that refer to this License, so +that they refer to the ordinary GNU General Public License, version 2, +instead of to this License. (If a newer version than version 2 of the +ordinary GNU General Public License has appeared, then you can specify +that version instead if you wish.) Do not make any other change in +these notices. +^L + Once this change is made in a given copy, it is irreversible for +that copy, so the ordinary GNU General Public License applies to all +subsequent copies and derivative works made from that copy. + + This option is useful when you wish to copy part of the code of +the Library into a program that is not a library. + + 4. You may copy and distribute the Library (or a portion or +derivative of it, under Section 2) in object code or executable form +under the terms of Sections 1 and 2 above provided that you accompany +it with the complete corresponding machine-readable source code, which +must be distributed under the terms of Sections 1 and 2 above on a +medium customarily used for software interchange. + + If distribution of object code is made by offering access to copy +from a designated place, then offering equivalent access to copy the +source code from the same place satisfies the requirement to +distribute the source code, even though third parties are not +compelled to copy the source along with the object code. + + 5. A program that contains no derivative of any portion of the +Library, but is designed to work with the Library by being compiled or +linked with it, is called a "work that uses the Library". Such a +work, in isolation, is not a derivative work of the Library, and +therefore falls outside the scope of this License. + + However, linking a "work that uses the Library" with the Library +creates an executable that is a derivative of the Library (because it +contains portions of the Library), rather than a "work that uses the +library". The executable is therefore covered by this License. +Section 6 states terms for distribution of such executables. + + When a "work that uses the Library" uses material from a header file +that is part of the Library, the object code for the work may be a +derivative work of the Library even though the source code is not. +Whether this is true is especially significant if the work can be +linked without the Library, or if the work is itself a library. The +threshold for this to be true is not precisely defined by law. + + If such an object file uses only numerical parameters, data +structure layouts and accessors, and small macros and small inline +functions (ten lines or less in length), then the use of the object +file is unrestricted, regardless of whether it is legally a derivative +work. (Executables containing this object code plus portions of the +Library will still fall under Section 6.) + + Otherwise, if the work is a derivative of the Library, you may +distribute the object code for the work under the terms of Section 6. +Any executables containing that work also fall under Section 6, +whether or not they are linked directly with the Library itself. +^L + 6. As an exception to the Sections above, you may also combine or +link a "work that uses the Library" with the Library to produce a +work containing portions of the Library, and distribute that work +under terms of your choice, provided that the terms permit +modification of the work for the customer's own use and reverse +engineering for debugging such modifications. + + You must give prominent notice with each copy of the work that the +Library is used in it and that the Library and its use are covered by +this License. You must supply a copy of this License. If the work +during execution displays copyright notices, you must include the +copyright notice for the Library among them, as well as a reference +directing the user to the copy of this License. Also, you must do one +of these things: + + a) Accompany the work with the complete corresponding + machine-readable source code for the Library including whatever + changes were used in the work (which must be distributed under + Sections 1 and 2 above); and, if the work is an executable linked + with the Library, with the complete machine-readable "work that + uses the Library", as object code and/or source code, so that the + user can modify the Library and then relink to produce a modified + executable containing the modified Library. (It is understood + that the user who changes the contents of definitions files in the + Library will not necessarily be able to recompile the application + to use the modified definitions.) + + b) Use a suitable shared library mechanism for linking with the + Library. A suitable mechanism is one that (1) uses at run time a + copy of the library already present on the user's computer system, + rather than copying library functions into the executable, and (2) + will operate properly with a modified version of the library, if + the user installs one, as long as the modified version is + interface-compatible with the version that the work was made with. + + c) Accompany the work with a written offer, valid for at + least three years, to give the same user the materials + specified in Subsection 6a, above, for a charge no more + than the cost of performing this distribution. + + d) If distribution of the work is made by offering access to copy + from a designated place, offer equivalent access to copy the above + specified materials from the same place. + + e) Verify that the user has already received a copy of these + materials or that you have already sent this user a copy. + + For an executable, the required form of the "work that uses the +Library" must include any data and utility programs needed for +reproducing the executable from it. However, as a special exception, +the materials to be distributed need not include anything that is +normally distributed (in either source or binary form) with the major +components (compiler, kernel, and so on) of the operating system on +which the executable runs, unless that component itself accompanies +the executable. + + It may happen that this requirement contradicts the license +restrictions of other proprietary libraries that do not normally +accompany the operating system. Such a contradiction means you cannot +use both them and the Library together in an executable that you +distribute. +^L + 7. You may place library facilities that are a work based on the +Library side-by-side in a single library together with other library +facilities not covered by this License, and distribute such a combined +library, provided that the separate distribution of the work based on +the Library and of the other library facilities is otherwise +permitted, and provided that you do these two things: + + a) Accompany the combined library with a copy of the same work + based on the Library, uncombined with any other library + facilities. This must be distributed under the terms of the + Sections above. + + b) Give prominent notice with the combined library of the fact + that part of it is a work based on the Library, and explaining + where to find the accompanying uncombined form of the same work. + + 8. You may not copy, modify, sublicense, link with, or distribute +the Library except as expressly provided under this License. Any +attempt otherwise to copy, modify, sublicense, link with, or +distribute the Library is void, and will automatically terminate your +rights under this License. However, parties who have received copies, +or rights, from you under this License will not have their licenses +terminated so long as such parties remain in full compliance. + + 9. You are not required to accept this License, since you have not +signed it. However, nothing else grants you permission to modify or +distribute the Library or its derivative works. These actions are +prohibited by law if you do not accept this License. Therefore, by +modifying or distributing the Library (or any work based on the +Library), you indicate your acceptance of this License to do so, and +all its terms and conditions for copying, distributing or modifying +the Library or works based on it. + + 10. Each time you redistribute the Library (or any work based on the +Library), the recipient automatically receives a license from the +original licensor to copy, distribute, link with or modify the Library +subject to these terms and conditions. You may not impose any further +restrictions on the recipients' exercise of the rights granted herein. +You are not responsible for enforcing compliance by third parties with +this License. +^L + 11. If, as a consequence of a court judgment or allegation of patent +infringement or for any other reason (not limited to patent issues), +conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot +distribute so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you +may not distribute the Library at all. For example, if a patent +license would not permit royalty-free redistribution of the Library by +all those who receive copies directly or indirectly through you, then +the only way you could satisfy both it and this License would be to +refrain entirely from distribution of the Library. + +If any portion of this section is held invalid or unenforceable under +any particular circumstance, the balance of the section is intended to +apply, and the section as a whole is intended to apply in other +circumstances. + +It is not the purpose of this section to induce you to infringe any +patents or other property right claims or to contest validity of any +such claims; this section has the sole purpose of protecting the +integrity of the free software distribution system which is +implemented by public license practices. Many people have made +generous contributions to the wide range of software distributed +through that system in reliance on consistent application of that +system; it is up to the author/donor to decide if he or she is willing +to distribute software through any other system and a licensee cannot +impose that choice. + +This section is intended to make thoroughly clear what is believed to +be a consequence of the rest of this License. + + 12. If the distribution and/or use of the Library is restricted in +certain countries either by patents or by copyrighted interfaces, the +original copyright holder who places the Library under this License +may add an explicit geographical distribution limitation excluding those +countries, so that distribution is permitted only in or among +countries not thus excluded. In such case, this License incorporates +the limitation as if written in the body of this License. + + 13. The Free Software Foundation may publish revised and/or new +versions of the Lesser General Public License from time to time. +Such new versions will be similar in spirit to the present version, +but may differ in detail to address new problems or concerns. + +Each version is given a distinguishing version number. If the Library +specifies a version number of this License which applies to it and +"any later version", you have the option of following the terms and +conditions either of that version or of any later version published by +the Free Software Foundation. If the Library does not specify a +license version number, you may choose any version ever published by +the Free Software Foundation. +^L + 14. If you wish to incorporate parts of the Library into other free +programs whose distribution conditions are incompatible with these, +write to the author to ask for permission. For software which is +copyrighted by the Free Software Foundation, write to the Free +Software Foundation; we sometimes make exceptions for this. Our +decision will be guided by the two goals of preserving the free status +of all derivatives of our free software and of promoting the sharing +and reuse of software generally. + + NO WARRANTY + + 15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO +WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW. +EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR +OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY +KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE +LIBRARY IS WITH YOU. SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME +THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN +WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY +AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU +FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR +CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE +LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING +RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A +FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF +SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH +DAMAGES. + + END OF TERMS AND CONDITIONS +^L + How to Apply These Terms to Your New Libraries + + If you develop a new library, and you want it to be of the greatest +possible use to the public, we recommend making it free software that +everyone can redistribute and change. You can do so by permitting +redistribution under these terms (or, alternatively, under the terms +of the ordinary General Public License). + + To apply these terms, attach the following notices to the library. +It is safest to attach them to the start of each source file to most +effectively convey the exclusion of warranty; and each file should +have at least the "copyright" line and a pointer to where the full +notice is found. + + + + Copyright (C) + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. + + This library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this library; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + +Also add information on how to contact you by electronic and paper +mail. + +You should also get your employer (if you work as a programmer) or +your +school, if any, to sign a "copyright disclaimer" for the library, if +necessary. Here is a sample; alter the names: + + Yoyodyne, Inc., hereby disclaims all copyright interest in the + library `Frob' (a library for tweaking knobs) written by James +Random Hacker. + + , 1 April 1990 + Ty Coon, President of Vice + +That's all there is to it! + + diff -urN v2.4.19-pre5/ulib/Makefile linux.diff/ulib/Makefile --- v2.4.19-pre5/ulib/Makefile Wed Dec 31 19:00:00 1969 +++ linux.diff/ulib/Makefile Fri Apr 19 20:58:01 2002 @@ -0,0 +1,50 @@ +# Makefile - libredhat-kernel.so build code. +# +# Copyright 2002 Red Hat, Inc. All Rights Reserved. +# +# This library is free software; you can redistribute it and/or +# modify it under the terms of the GNU Lesser General Public +# License as published by the Free Software Foundation; either +# version 2 of the License, or (at your option) any later version. +# +# This library is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +# Lesser General Public License for more details. +# +# You should have received a copy of the GNU Lesser General Public +# License along with this library; if not, write to the Free Software +# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +# +# +all: libredhat-kernel.so + +ASFLAGS=-D__KERNEL__ -D__ASSEMBLY__ -I../include -nostdlib -nostartfiles +CFLAGS=-D__KERNEL__ -I../include -nostdlib -nostartfiles + +so_objs=vsysaddr.o kso_init.o + +vsysaddr.S: ../System.map stub.S Makefile + rm -f vsysaddr.S + echo '#include "stub.S"' >vsysaddr.S + awk -- "/^00000000bfff.* vsys_/ { print \"dynamic_syscall(\"\$$3 \",0x\" \$$1 \")\"; }" <../System.map >>vsysaddr.S + awk -- "/^bfff.* vsys_/ { print \"dynamic_syscall(\"\$$3 \",0x\" \$$1 \")\"; }" <../System.map >>vsysaddr.S + +vsysaddr.o: vsysaddr.S + +kso_init.o: ../include/linux/compile.h + +libredhat-kernel.so.1.0.1: $(so_objs) libredhat-kernel.map + gcc -nostdlib -nostartfiles -shared -Wl,--version-script=libredhat-kernel.map -Wl,-soname=libredhat-kernel.so.1 -o $@ $(so_objs) + cp $@ $@.save + strip $@ + +libredhat-kernel.so: libredhat-kernel.so.1.0.1 + ln -sf $< $@ + +clean: + rm -f *.o libredhat-kernel.so myln libredhat-kernel.so.1* vsysaddr.S + +# test app +myln: myln.c libredhat-kernel.so Makefile + cc -g -o myln myln.c -L. -lredhat-kernel diff -urN v2.4.19-pre5/ulib/README linux.diff/ulib/README --- v2.4.19-pre5/ulib/README Wed Dec 31 19:00:00 1969 +++ linux.diff/ulib/README Fri Apr 19 20:54:05 2002 @@ -0,0 +1,2 @@ +The libredhat-kernel code is provided under the terms of the LGPL. +See the file COPYING for details. diff -urN v2.4.19-pre5/ulib/kso_init.c linux.diff/ulib/kso_init.c --- v2.4.19-pre5/ulib/kso_init.c Wed Dec 31 19:00:00 1969 +++ linux.diff/ulib/kso_init.c Fri Apr 19 20:54:05 2002 @@ -0,0 +1,67 @@ +/* kso_init.c - libredhat-kernel.so startup code. + + Copyright 2002 Red Hat, Inc. All Rights Reserved. + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. + + This library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this library; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + + */ +#include +#include +#include +#include +#include +#include + +char libredhat_kernel_enosys = 1; /* the asm in stub.S depends on this */ + +long _init(void) +{ + static char unique[] = { LINUX_UNIQUE_BYTES }; + int errno; + long addr; + int fd; + int i; + + _syscall6(int, mmap2, unsigned long, addr, unsigned long, len, + unsigned long, prot, unsigned long, flags, + unsigned long, fd, unsigned long, pgoff) + _syscall2(long, munmap, unsigned long, addr, size_t, len) + _syscall2(int, open, const char *, name, int, flags) + _syscall1(int, close, int, fd) + + if (sizeof(unique) != 16) + return -1; + + fd = open("/dev/vsys", O_RDONLY); + if (-1 == fd) + return -1; + + addr = mmap2(0, VSYSCALL_SIZE, PROT_READ | PROT_EXEC, MAP_SHARED, fd, 0); + if (-1 == addr) + return -1; + + close(fd); + + for (i=0; i +#include +#include +#include +#include + +int main () +{ + long ctx = 0; + extern long vsys_io_setup(long, long *); + unsigned char *bob = (void*)&vsys_io_setup; + long ret; + int i; + printf("%p\n", bob); + //printf("%p\n", mmap(0, 65536, PROT_READ | PROT_EXEC, MAP_SHARED, + // open("/dev/vsys", O_RDONLY), 0)); + //for (i=0; i<16; i++) + // printf(" %02x\n", bob[i]); + //printf("\n"); + + ret = vsys_io_setup(100, &ctx); + + printf("ret=%ld, ctx=0x%lx\n", ret, ctx); + return 0; +} diff -urN v2.4.19-pre5/ulib/stub.S linux.diff/ulib/stub.S --- v2.4.19-pre5/ulib/stub.S Wed Dec 31 19:00:00 1969 +++ linux.diff/ulib/stub.S Fri Apr 19 20:54:05 2002 @@ -0,0 +1,38 @@ +/* stub.S - libredhat-kernel.so jump code. + + Copyright 2002 Red Hat, Inc. All Rights Reserved. + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. + + This library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this library; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + + */ +/* stub.S */ +#include +#include + + .text + +#define dynamic_syscall(x,a) \ + .globl x ;\ + .type x, @function ;\ + .align 16 ;\ + x: ;\ + cmpb $0,libredhat_kernel_enosys ;\ + jne 1f ;\ + ljmp $__USER_CS, $a ;\ + 1: ;\ + movl $-ENOSYS,%eax ;\ + ret ;\ + .size x,.-x +