Hi everyone,
This patch (for 2.6.4) exposes a new system call (sys_fdmap) that allows
for mapping a virtual memory range to a file descriptor in the context of
a process. This can be seen as the complement to mmap in that it provides
a mechanism to operate on a range of memory as if it were a file. The
supported file operations are: read, write, llseek, and mmap.
The prototype for the system call is:
int fdmap(void *vma, size_t len, int flags);
or
_syscall3(int,fdmap,void *,vma,size_t,len,int,flags);
``flags'' can be one of O_RDONLY, O_WRONLY, or O_RDWR.
I have verified functionality on ia32 and sparc as these are the only
architectures I currently have some type of access to. To test, start the
kernel configuration process and go under File systems/Pseudo filesystems
and select this option:
[*] Virtual memory file descriptor mapping support
Please let me know about any and all suggestions/bugs/flames. I tried to
be as thorough as possible but do suspect that I've missed some things.
I'm looking forward to hearing feedback.
Matt
-------
Patch details:
Target kernel release : 2.6.4
Target subsystem : fs
Number of new files : 1
Number of modified files: 33
Total file changes : 34
Patched file list:
arch/alpha/kernel/systbls.S | 4
arch/arm/kernel/calls.S | 5
arch/i386/kernel/entry.S | 5
arch/ia64/kernel/entry.S | 4
arch/m68k/kernel/entry.S | 5
arch/mips/kernel/scall32-o32.S | 5
arch/mips/kernel/scall64-64.S | 5
arch/mips/kernel/scall64-n32.S | 5
arch/mips/kernel/scall64-o32.S | 5
arch/parisc/kernel/syscall_table.S | 6
arch/ppc/kernel/misc.S | 5
arch/ppc64/kernel/misc.S | 5
arch/s390/kernel/syscalls.S | 5
arch/sh/kernel/entry.S | 5
arch/sparc/kernel/systbls.S | 8
arch/sparc64/kernel/systbls.S | 16 +
arch/um/kernel/sys_call_table.c | 5
arch/v850/kernel/entry.S | 6
fs/Kconfig | 8
fs/Makefile | 2
fs/fdmap.c | 517
+++++++++++++++++++++++++++++++++++++
include/asm-alpha/unistd.h | 2
include/asm-arm/unistd.h | 1
include/asm-i386/unistd.h | 3
include/asm-ia64/unistd.h | 3
include/asm-m68k/unistd.h | 3
include/asm-mips/unistd.h | 15 -
include/asm-parisc/unistd.h | 3
include/asm-ppc/unistd.h | 3
include/asm-ppc64/unistd.h | 3
include/asm-s390/unistd.h | 5
include/asm-sparc/unistd.h | 2
include/asm-sparc64/unistd.h | 2
include/asm-v850/unistd.h | 1
34 files changed, 657 insertions(+), 20 deletions(-)
Patch contents:
-- CUT HERE --
diff -X dontdiff -uprN linux-2.6.4/arch/alpha/kernel/systbls.S linux-2.6.4-fdmap/arch/alpha/kernel/systbls.S
--- linux-2.6.4/arch/alpha/kernel/systbls.S Wed Mar 10 20:55:35 2004
+++ linux-2.6.4-fdmap/arch/alpha/kernel/systbls.S Sun Mar 21 08:57:18 2004
@@ -292,7 +292,11 @@ sys_call_table:
.quad alpha_ni_syscall
.quad alpha_ni_syscall
.quad alpha_ni_syscall
+#ifdef CONFIG_FDMAP
+ .quad sys_fdmap
+#else
.quad alpha_ni_syscall
+#endif
.quad alpha_ni_syscall /* 275 */
.quad alpha_ni_syscall
.quad alpha_ni_syscall
diff -X dontdiff -uprN linux-2.6.4/arch/arm/kernel/calls.S linux-2.6.4-fdmap/arch/arm/kernel/calls.S
--- linux-2.6.4/arch/arm/kernel/calls.S Wed Mar 10 20:55:55 2004
+++ linux-2.6.4-fdmap/arch/arm/kernel/calls.S Sun Mar 21 08:57:40 2004
@@ -288,6 +288,11 @@ __syscall_start:
.long sys_pciconfig_iobase
.long sys_pciconfig_read
.long sys_pciconfig_write
+#ifdef CONFIG_FDMAP
+ .long sys_fdmap
+#else
+ .long sys_ni_syscall
+#endif
__syscall_end:
.rept NR_syscalls - (__syscall_end - __syscall_start) / 4
diff -X dontdiff -uprN linux-2.6.4/arch/i386/kernel/entry.S linux-2.6.4-fdmap/arch/i386/kernel/entry.S
--- linux-2.6.4/arch/i386/kernel/entry.S Wed Mar 10 20:55:24 2004
+++ linux-2.6.4-fdmap/arch/i386/kernel/entry.S Sun Mar 21 08:58:38 2004
@@ -882,5 +882,10 @@ ENTRY(sys_call_table)
.long sys_utimes
.long sys_fadvise64_64
.long sys_ni_syscall /* sys_vserver */
+#ifdef CONFIG_FDMAP
+ .long sys_fdmap /* 274 */
+#else
+ .long sys_ni_syscall
+#endif
syscall_table_size=(.-sys_call_table)
diff -X dontdiff -uprN linux-2.6.4/arch/ia64/kernel/entry.S linux-2.6.4-fdmap/arch/ia64/kernel/entry.S
--- linux-2.6.4/arch/ia64/kernel/entry.S Wed Mar 10 20:55:24 2004
+++ linux-2.6.4-fdmap/arch/ia64/kernel/entry.S Sun Mar 21 08:58:57 2004
@@ -1501,7 +1501,11 @@ sys_call_table:
data8 sys_clock_nanosleep
data8 sys_fstatfs64
data8 sys_statfs64
+#ifdef CONFIG_FDMAP
+ data8 sys_fdmap
+#else
data8 sys_ni_syscall
+#endif
data8 sys_ni_syscall // 1260
data8 sys_ni_syscall
data8 sys_ni_syscall
diff -X dontdiff -uprN linux-2.6.4/arch/m68k/kernel/entry.S linux-2.6.4-fdmap/arch/m68k/kernel/entry.S
--- linux-2.6.4/arch/m68k/kernel/entry.S Wed Mar 10 20:55:21 2004
+++ linux-2.6.4-fdmap/arch/m68k/kernel/entry.S Sun Mar 21 08:59:17 2004
@@ -663,3 +663,8 @@ sys_call_table:
.long sys_lremovexattr
.long sys_fremovexattr
.long sys_futex /* 235 */
+#ifdef CONFIG_FDMAP
+ .long sys_fdmap
+#else
+ .long sys_ni_syscall
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/mips/kernel/scall32-o32.S linux-2.6.4-fdmap/arch/mips/kernel/scall32-o32.S
--- linux-2.6.4/arch/mips/kernel/scall32-o32.S Wed Mar 10 20:55:22 2004
+++ linux-2.6.4-fdmap/arch/mips/kernel/scall32-o32.S Sun Mar 21 08:59:43 2004
@@ -614,6 +614,11 @@ out: jr ra
sys sys_clock_nanosleep 4 /* 4265 */
sys sys_tgkill 3
sys sys_utimes 2
+#ifdef CONFIG_FDMAP
+ sys sys_fdmap 3
+#else
+ sys sys_ni_syscall 0
+#endif
.endm
diff -X dontdiff -uprN linux-2.6.4/arch/mips/kernel/scall64-64.S linux-2.6.4-fdmap/arch/mips/kernel/scall64-64.S
--- linux-2.6.4/arch/mips/kernel/scall64-64.S Wed Mar 10 20:55:36 2004
+++ linux-2.6.4-fdmap/arch/mips/kernel/scall64-64.S Sun Mar 21 09:07:05 2004
@@ -432,3 +432,8 @@ sys_call_table:
PTR sys_clock_nanosleep
PTR sys_tgkill /* 5225 */
PTR sys_utimes
+#ifdef CONFIG_FDMAP
+ PTR sys_fdmap
+#else
+ PTR sys_ni_syscall
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/mips/kernel/scall64-n32.S linux-2.6.4-fdmap/arch/mips/kernel/scall64-n32.S
--- linux-2.6.4/arch/mips/kernel/scall64-n32.S Wed Mar 10 20:55:44 2004
+++ linux-2.6.4-fdmap/arch/mips/kernel/scall64-n32.S Sun Mar 21 09:07:17 2004
@@ -341,3 +341,8 @@ EXPORT(sysn32_call_table)
PTR sys_clock_nanosleep
PTR sys_tgkill
PTR compat_sys_utimes /* 6230 */
+#ifdef CONFIG_FDMAP
+ PTR sys_fdmap
+#else
+ PTR sys_ni_syscall
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/mips/kernel/scall64-o32.S linux-2.6.4-fdmap/arch/mips/kernel/scall64-o32.S
--- linux-2.6.4/arch/mips/kernel/scall64-o32.S Wed Mar 10 20:55:20 2004
+++ linux-2.6.4-fdmap/arch/mips/kernel/scall64-o32.S Sun Mar 21 09:07:29 2004
@@ -523,6 +523,11 @@ out: jr ra
sys sys_clock_nanosleep 4 /* 4265 */
sys sys_tgkill 3
sys compat_sys_utimes 2
+#ifdef CONFIG_FDMAP
+ sys sys_fdmap 3
+#else
+ sys sys_ni_syscall 0
+#endif
.endm
diff -X dontdiff -uprN linux-2.6.4/arch/parisc/kernel/syscall_table.S linux-2.6.4-fdmap/arch/parisc/kernel/syscall_table.S
--- linux-2.6.4/arch/parisc/kernel/syscall_table.S Wed Mar 10 20:55:21 2004
+++ linux-2.6.4-fdmap/arch/parisc/kernel/syscall_table.S Sun Mar 21 09:00:08 2004
@@ -334,3 +334,9 @@
ENTRY_SAME(epoll_ctl) /* 225 */
ENTRY_SAME(epoll_wait)
ENTRY_SAME(remap_file_pages)
+ ENTRY_SMAE(ni_syscall)
+#ifdef CONFIG_FDMAP
+ ENTRY_SAME(fdmap)
+#else
+ ENTRY_SAME(ni_syscall)
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/ppc/kernel/misc.S linux-2.6.4-fdmap/arch/ppc/kernel/misc.S
--- linux-2.6.4/arch/ppc/kernel/misc.S Wed Mar 10 20:55:44 2004
+++ linux-2.6.4-fdmap/arch/ppc/kernel/misc.S Sun Mar 21 09:00:20 2004
@@ -1380,3 +1380,8 @@ _GLOBAL(sys_call_table)
.long sys_fstatfs64
.long ppc_fadvise64_64
.long sys_ni_syscall /* 255 - rtas (used on ppc64) */
+#ifdef CONFIG_FDMAP
+ .long sys_fdmap
+#else
+ .long sys_ni_syscall
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/ppc64/kernel/misc.S linux-2.6.4-fdmap/arch/ppc64/kernel/misc.S
--- linux-2.6.4/arch/ppc64/kernel/misc.S Wed Mar 10 20:55:44 2004
+++ linux-2.6.4-fdmap/arch/ppc64/kernel/misc.S Sun Mar 21 09:00:33 2004
@@ -1116,3 +1116,8 @@ _GLOBAL(sys_call_table)
.llong .sys_fstatfs64
.llong .sys_ni_syscall /* 32bit only fadvise64_64 */
.llong .ppc_rtas /* 255 */
+#ifdef CONFIG_FDMAP
+ .llong .sys_fdmap
+#else
+ .llong .sys_ni_syscall
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/s390/kernel/syscalls.S linux-2.6.4-fdmap/arch/s390/kernel/syscalls.S
--- linux-2.6.4/arch/s390/kernel/syscalls.S Wed Mar 10 20:55:22 2004
+++ linux-2.6.4-fdmap/arch/s390/kernel/syscalls.S Sun Mar 21 09:01:02 2004
@@ -273,3 +273,8 @@ SYSCALL(sys_clock_getres,sys_clock_getre
SYSCALL(sys_clock_nanosleep,sys_clock_nanosleep,sys32_clock_nanosleep_wrapper)
NI_SYSCALL /* reserved for vserver */
SYSCALL(s390_fadvise64_64,sys_ni_syscall,sys32_fadvise64_64_wrapper)
+#ifdef CONFIG_FDMAP
+SYSCALL(sys_fdmap,sys_fdmap,sys_fdmap) /* 265 */
+#else
+NI_SYSCALL
+#endif
diff -X dontdiff -uprN linux-2.6.4/arch/sh/kernel/entry.S linux-2.6.4-fdmap/arch/sh/kernel/entry.S
--- linux-2.6.4/arch/sh/kernel/entry.S Wed Mar 10 20:55:25 2004
+++ linux-2.6.4-fdmap/arch/sh/kernel/entry.S Sun Mar 21 09:01:13 2004
@@ -1129,6 +1129,11 @@ ENTRY(sys_call_table)
.long sys_utimes
.long sys_fadvise64_64_wrapper
.long sys_ni_syscall /* Reserved for vserver */
+#ifdef CONFIG_FDMAP
+ .long sys_fdmap
+#else
+ .long sys_ni_syscall
+#endif
.rept NR_syscalls-(.-sys_call_table)/4
.long sys_ni_syscall
diff -X dontdiff -uprN linux-2.6.4/arch/sparc/kernel/systbls.S linux-2.6.4-fdmap/arch/sparc/kernel/systbls.S
--- linux-2.6.4/arch/sparc/kernel/systbls.S Wed Mar 10 20:55:27 2004
+++ linux-2.6.4-fdmap/arch/sparc/kernel/systbls.S Sun Mar 21 09:01:34 2004
@@ -61,7 +61,13 @@ sys_call_table:
/*200*/ .long sys_ssetmask, sys_sigsuspend, sys_newlstat, sys_uselib, old_readdir
/*205*/ .long sys_readahead, sys_socketcall, sys_syslog, sys_lookup_dcookie, sys_fadvise64
/*210*/ .long sys_fadvise64_64, sys_tgkill, sys_waitpid, sys_swapoff, sys_sysinfo
-/*215*/ .long sys_ipc, sys_sigreturn, sys_clone, sys_nis_syscall, sys_adjtimex
+/*215*/ .long sys_ipc, sys_sigreturn, sys_clone,
+#ifdef CONFIG_MAP
+/*218*/ .long sys_fdmap
+#else
+/*218*/ .long sys_ni_syscall
+#endif
+/*219*/ .long sys_adjtimex
/*220*/ .long sys_sigprocmask, sys_ni_syscall, sys_delete_module, sys_ni_syscall, sys_getpgid
/*225*/ .long sys_bdflush, sys_sysfs, sys_nis_syscall, sys_setfsuid16, sys_setfsgid16
/*230*/ .long sys_select, sys_time, sys_nis_syscall, sys_stime, sys_statfs64
diff -X dontdiff -uprN linux-2.6.4/arch/sparc64/kernel/systbls.S linux-2.6.4-fdmap/arch/sparc64/kernel/systbls.S
--- linux-2.6.4/arch/sparc64/kernel/systbls.S Wed Mar 10 20:55:26 2004
+++ linux-2.6.4-fdmap/arch/sparc64/kernel/systbls.S Sun Mar 21 09:01:59 2004
@@ -62,7 +62,13 @@ sys_call_table32:
/*200*/ .word sys_ssetmask, sys_sigsuspend, compat_sys_newlstat, sys_uselib, old32_readdir
.word sys32_readahead, sys32_socketcall, sys_syslog, sys32_lookup_dcookie, sys32_fadvise64
/*210*/ .word sys32_fadvise64_64, sys_tgkill, sys_waitpid, sys_swapoff, sys32_sysinfo
- .word sys32_ipc, sys32_sigreturn, sys_clone, sys_nis_syscall, sys32_adjtimex
+/*215*/ .word sys32_ipc, sys32_sigreturn, sys_clone
+#ifdef CONFIG_FDMAP
+/*218*/ .word sys_fdmap,
+#else
+/*218*/ .word sys_ni_syscall
+#endif
+/*219*/ .word sys32_adjtimex
/*220*/ .word compat_sys_sigprocmask, sys_ni_syscall, sys32_delete_module, sys_ni_syscall, sys_getpgid
.word sys32_bdflush, sys32_sysfs, sys_nis_syscall, sys32_setfsuid16, sys32_setfsgid16
/*230*/ .word sys32_select, sys_time, sys_nis_syscall, sys_stime, compat_statfs64
@@ -124,7 +130,13 @@ sys_call_table:
/*200*/ .word sys_ssetmask, sys_nis_syscall, sys_newlstat, sys_uselib, sys_nis_syscall
.word sys_readahead, sys_socketcall, sys_syslog, sys_lookup_dcookie, sys_fadvise64
/*210*/ .word sys_fadvise64_64, sys_tgkill, sys_waitpid, sys_swapoff, sys_sysinfo
- .word sys_ipc, sys_nis_syscall, sys_clone, sys_nis_syscall, sys_adjtimex
+ .word sys_ipc, sys_nis_syscall, sys_clone
+#ifdef CONFIG_FDMAP
+/*218*/ .word sys_fdmap
+#else
+/*218*/ .word sys_ni_syscall
+#endif
+/*219*/ .word sys_adjtimex
/*220*/ .word sys_nis_syscall, sys_ni_syscall, sys_delete_module, sys_ni_syscall, sys_getpgid
.word sys_bdflush, sys_sysfs, sys_nis_syscall, sys_setfsuid, sys_setfsgid
/*230*/ .word sys_select, sys_nis_syscall, sys_nis_syscall, sys_stime, sys_statfs64
diff -X dontdiff -uprN linux-2.6.4/arch/um/kernel/sys_call_table.c linux-2.6.4-fdmap/arch/um/kernel/sys_call_table.c
--- linux-2.6.4/arch/um/kernel/sys_call_table.c Wed Mar 10 20:55:43 2004
+++ linux-2.6.4-fdmap/arch/um/kernel/sys_call_table.c Sun Mar 21 08:56:45 2004
@@ -489,6 +489,11 @@ syscall_handler_t *sys_call_table[] = {
[ __NR_epoll_wait ] = sys_epoll_wait,
[ __NR_remap_file_pages ] = sys_remap_file_pages,
[ __NR_set_tid_address ] = sys_set_tid_address,
+#ifdef __NR_fdmap
+ [ __NR_fdmap ] = sys_fdmap,
+#else
+ [ __NR_fdmap ] = sys_ni_syscall,
+#endif
ARCH_SYSCALLS
[ LAST_SYSCALL + 1 ... NR_syscalls ] =
diff -X dontdiff -uprN linux-2.6.4/arch/v850/kernel/entry.S linux-2.6.4-fdmap/arch/v850/kernel/entry.S
--- linux-2.6.4/arch/v850/kernel/entry.S Wed Mar 10 20:55:44 2004
+++ linux-2.6.4-fdmap/arch/v850/kernel/entry.S Sun Mar 21 09:02:25 2004
@@ -1117,5 +1117,11 @@ C_DATA(sys_call_table):
.long CSYM(sys_pivot_root) // 200
.long CSYM(sys_gettid)
.long CSYM(sys_tkill)
+#ifdef CONFIG_FDMAP
+ .long CSYM(sys_fdmap)
+#else
+ .long CSYM(sys_ni_syscall)
+#endif
+
sys_call_table_end:
C_END(sys_call_table)
diff -X dontdiff -uprN linux-2.6.4/fs/Kconfig linux-2.6.4-fdmap/fs/Kconfig
--- linux-2.6.4/fs/Kconfig Wed Mar 10 20:55:29 2004
+++ linux-2.6.4-fdmap/fs/Kconfig Sat Mar 20 15:21:41 2004
@@ -864,6 +864,14 @@ config TMPFS
See <file:Documentation/filesystems/tmpfs.txt> for details.
+config FDMAP
+ bool "Virtual memory file descriptor mapping support"
+ help
+ Fdmap allows for accessing a range of memory inside a given process
+ as if it were a file on disk. This can be seen as the counterpart
+ to mmap, simply, fdmap allows for opening a range of memory that
+ can be operated on by way of read, write, lseek, et al.
+
config HUGETLBFS
bool "HugeTLB file system support"
depends X86 || IA64 || PPC64 || SPARC64 || X86_64 || BROKEN
diff -X dontdiff -uprN linux-2.6.4/fs/Makefile linux-2.6.4-fdmap/fs/Makefile
--- linux-2.6.4/fs/Makefile Wed Mar 10 20:55:29 2004
+++ linux-2.6.4-fdmap/fs/Makefile Sat Mar 20 15:22:17 2004
@@ -44,6 +44,8 @@ obj-y += devpts/
obj-$(CONFIG_PROFILING) += dcookies.o
+obj-$(CONFIG_FDMAP) += fdmap.o
+
# Do not add any filesystems before this line
obj-$(CONFIG_REISERFS_FS) += reiserfs/
obj-$(CONFIG_EXT3_FS) += ext3/ # Before ext2 so root fs can be ext3
diff -X dontdiff -uprN linux-2.6.4/fs/fdmap.c linux-2.6.4-fdmap/fs/fdmap.c
--- linux-2.6.4/fs/fdmap.c Wed Dec 31 18:00:00 1969
+++ linux-2.6.4-fdmap/fs/fdmap.c Sun Mar 21 04:08:22 2004
@@ -0,0 +1,517 @@
+/*
+ * purpose
+ *
+ * Map process memory ranges to virtual file descriptors. This can be
+ * thought of as being the opposite of mmap, simply, instead of mapping
+ * a file's contents to a memory range, fdmap maps a memory range to
+ * a virtual file. This allows one to read, write, seek, and even
+ * mmap the virtual file's contents as if it were a real file on disk.
+ *
+ * interface
+ *
+ * fdmap exposes a new system call identifier. The system call takes
+ * arguments as the following prototype conveys:
+ *
+ * int fdmap(void *addr, size_t len, int flags);
+ *
+ * ``flags'' can be one of O_RDONLY, O_WRONLY, or O_RDWR.
+ *
+ * syscall number
+ *
+ * alpha, arm, ia32, sh: 274
+ * sparc, sparc64: 218
+ * ia64: 1259
+ * m68k: 236
+ * mips: 32b=4268 64b=5227 64be32=6231
+ * parisc: 229
+ * ppc, ppc64: 256
+ * s390: 265
+ * v850: 203
+ *
+ * based on
+ *
+ * The underlying virtual filesystem code was adapted from sockfs.
+ *
+ * Matt Miller
+ * [email protected]
+ * 3/20/04
+ */
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/utime.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/cache.h>
+#include <asm/uaccess.h>
+
+typedef struct {
+ unsigned long vma;
+ size_t vma_len;
+} fdmap_context;
+
+typedef struct {
+ fdmap_context fdmap;
+ struct inode vfs_inode;
+} fdmap_alloc;
+
+static struct super_block *fdmap_get_sb(struct file_system_type *fs_type,
+ int flags, const char *dev, void *data);
+
+static void fdmap_init_once(void *ctx, kmem_cache_t *cachep,
+ unsigned long flags);
+static struct inode *fdmap_alloc_inode(struct super_block *sb);
+static void fdmap_destroy_inode(struct inode *inode);
+
+static int fdmap_delete_dentry(struct dentry *dentry);
+
+static struct page* fdmap_nopage(struct vm_area_struct *area,
+ unsigned long address, int *type);
+
+static int fdmap_open_mem(unsigned long addr, size_t len, int flags);
+
+static loff_t fdmap_llseek(struct file *f, loff_t off, int whence);
+static ssize_t fdmap_read(struct file *f, char *buf, size_t size, loff_t *off);
+static ssize_t fdmap_write(struct file *f, const char *buf, size_t size,
+ loff_t *off);
+static int fdmap_mmap(struct file *f, struct vm_area_struct *vma);
+
+static unsigned int fdmap_copy_from_user_to_user(unsigned long base_dst,
+ unsigned long base_src,
+ unsigned int total);
+static fdmap_context *fdmap_get_context(struct inode *inode);
+
+static struct super_operations fdmap_super_ops =
+{
+ .alloc_inode = fdmap_alloc_inode,
+ .destroy_inode = fdmap_destroy_inode,
+};
+
+static struct file_system_type fdmap_fs_type =
+{
+ .name = "fdmapfs",
+ .get_sb = fdmap_get_sb,
+ .kill_sb = kill_anon_super,
+};
+
+static struct dentry_operations fdmap_dentry_ops =
+{
+ .d_delete = fdmap_delete_dentry,
+};
+
+static struct vm_operations_struct fdmap_vm_ops =
+{
+ .nopage = fdmap_nopage,
+};
+
+static struct file_operations fdmap_file_ops =
+{
+ .llseek = fdmap_llseek,
+ .read = fdmap_read,
+ .write = fdmap_write,
+ .mmap = fdmap_mmap,
+};
+
+#define FDMAP_MAGIC 0x46444d41 // FDMA
+
+static kmem_cache_t *fdmap_inode_cachep;
+static struct vfsmount *fdmap_mnt;
+static atomic_t fdmap_initialized = { 1 };
+
+/*
+ * register and mount the fdmapfs
+ */
+static void fdmap_initialize(void)
+{
+ fdmap_inode_cachep = kmem_cache_create("fdmap_inode_cache",
+ sizeof(fdmap_alloc), 0,
+ SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT,
+ fdmap_init_once, NULL);
+
+ if (!fdmap_inode_cachep) {
+ printk("fdmap: failed to allocate cache.\n");
+ goto out;
+ }
+
+ register_filesystem(&fdmap_fs_type);
+
+ if (!(fdmap_mnt = kern_mount(&fdmap_fs_type))) {
+ printk("fdmap: failed to mount fdmapfs.\n");
+ }
+
+out:
+}
+
+static struct super_block *fdmap_get_sb(struct file_system_type *fs_type,
+ int flags, const char *dev, void *data)
+{
+ return get_sb_pseudo(fs_type, "fdmap:", &fdmap_super_ops, FDMAP_MAGIC);
+}
+
+static void fdmap_init_once(void *ctx, kmem_cache_t *cachep,
+ unsigned long flags)
+{
+ fdmap_alloc *fma = (fdmap_alloc *)ctx;
+
+ if ((flags & SLAB_CTOR_CONSTRUCTOR))
+ inode_init_once(&fma->vfs_inode);
+}
+
+static struct inode *fdmap_alloc_inode(struct super_block *sb)
+{
+ fdmap_alloc *fma = (fdmap_alloc *)kmem_cache_alloc(fdmap_inode_cachep,
+ SLAB_KERNEL);
+
+ if (!fma)
+ return NULL;
+
+ return &fma->vfs_inode;
+}
+
+static void fdmap_destroy_inode(struct inode *inode)
+{
+ kmem_cache_free(fdmap_inode_cachep,
+ container_of(inode, fdmap_alloc, vfs_inode));
+}
+
+static int fdmap_delete_dentry(struct dentry *dentry)
+{
+ return 1;
+}
+
+/*
+ * used in conjunction with mmap of the virtual file, fdmap_nopage will
+ * copy the contents of the mapped memory range at the offset
+ * relative to the address being requested
+ */
+static struct page* fdmap_nopage(struct vm_area_struct *area,
+ unsigned long addr, int *type)
+{
+ struct page *page = alloc_page(GFP_USER);
+ struct file *f = area->vm_file;
+ fdmap_context *fmc = (fdmap_context *)f->private_data;
+ unsigned long offset;
+ struct page *fpage;
+ void *saddr, *daddr;
+ int len = PAGE_SIZE;
+
+ addr &= PAGE_MASK;
+ offset = addr - area->vm_start + (area->vm_pgoff << PAGE_SHIFT);
+
+ if (addr + PAGE_SIZE > area->vm_end)
+ len = area->vm_end - addr;
+
+ if (get_user_pages(current, current->mm, fmc->vma + offset, 1, 0, 0,
+ &fpage, NULL) < 0)
+ return NULL;
+
+ daddr = kmap(page);
+ saddr = kmap(fpage);
+
+ memcpy(daddr, saddr, len);
+
+ kunmap(fpage);
+ kunmap(page);
+
+ if (type)
+ *type = VM_FAULT_MINOR;
+
+ set_page_dirty(page);
+
+ return page;
+}
+
+/*
+ * entry point for the fdmap syscall
+ */
+asmlinkage int sys_fdmap(void *vma, size_t len, int flags)
+{
+ unsigned long addr = (unsigned long)vma;
+ int fd = -EINVAL;
+
+ if ((atomic_read(&fdmap_initialized) == 1) &&
+ (atomic_dec_and_test(&fdmap_initialized)))
+ fdmap_initialize();
+
+ if ((!len) ||
+ ((addr + len) > TASK_SIZE) ||
+ ((addr + len) < addr)) {
+ fd = -EINVAL;
+ goto out;
+ }
+
+ fd = fdmap_open_mem(addr, len, flags);
+
+out:
+ return fd;
+}
+
+/*
+ * open a virtual file and corresponding file descriptor that is tied
+ * to the passed in memory range.
+ */
+static int fdmap_open_mem(unsigned long addr, size_t len, int flags)
+{
+ struct inode *inode = NULL;
+ fdmap_context *fmc = NULL;
+ struct file *file = NULL;
+ struct qstr this;
+ int fd = 0,
+ res = -ENFILE;
+ char name[32];
+
+ flags &= (O_RDONLY | O_WRONLY | O_RDWR);
+
+ if (flags & O_RDONLY)
+ flags &= ~O_WRONLY;
+ if (flags & O_WRONLY)
+ flags &= ~O_RDONLY;
+
+ if (!(inode = new_inode(fdmap_mnt->mnt_sb)))
+ goto out;
+
+ inode->i_mode = S_IRWXUGO;
+ inode->i_uid = current->fsuid;
+ inode->i_gid = current->fsgid;
+ inode->i_size = (loff_t)len;
+
+ fmc = fdmap_get_context(inode);
+
+ if ((fd = get_unused_fd()) < 0)
+ goto cleanup;
+
+ if (!(file = get_empty_filp()))
+ goto cleanup;
+
+ sprintf(name, "[%lu]", inode->i_ino);
+
+ this.name = name;
+ this.len = strlen(name);
+ this.hash = inode->i_ino;
+
+ file->f_dentry = d_alloc(fdmap_mnt->mnt_sb->s_root, &this);
+
+ if (!file->f_dentry) {
+ res = -ENOMEM;
+ goto cleanup;
+ }
+
+ file->f_dentry->d_op = &fdmap_dentry_ops;
+
+ d_add(file->f_dentry, inode);
+
+ file->f_vfsmnt = mntget(fdmap_mnt);
+ file->f_mapping = file->f_dentry->d_inode->i_mapping;
+ file->f_op = inode->i_fop = &fdmap_file_ops;
+ file->f_mode = 3;
+ file->f_flags = flags;
+ file->f_pos = 0;
+ file->private_data = fmc;
+
+ fd_install(fd, file);
+
+ fmc->vma = addr;
+ fmc->vma_len = len;
+
+ res = fd;
+
+ goto out;
+
+cleanup:
+ if (file)
+ put_filp(file);
+
+ if (fd >= 0)
+ put_unused_fd(fd);
+out:
+ return res;
+}
+
+static loff_t fdmap_llseek(struct file *f, loff_t off, int whence)
+{
+ fdmap_context *fmc = (fdmap_context *)f->private_data;
+ int res = -EINVAL;
+
+ if (!fmc)
+ goto out;
+
+ switch (whence) {
+
+ case 0:
+ if (off < 0 || off > fmc->vma_len)
+ break;
+
+ f->f_pos = off;
+ res = 0;
+ break;
+ case 1:
+ if ((off < 0) && (f->f_pos + off < fmc->vma_len))
+ break;
+ if ((off >= 0) && (f->f_pos + off >= fmc->vma_len))
+ break;
+
+ f->f_pos += off;
+ res = 0;
+ break;
+ case 2:
+ if (off > 0)
+ break;
+ if (fmc->vma_len + off < 0)
+ break;
+
+ f->f_pos = fmc->vma_len + off;
+ res = 0;
+ break;
+ default:
+ break;
+ }
+
+out:
+ return (res < 0) ? res : f->f_pos;
+}
+
+static ssize_t fdmap_read(struct file *f, char *buf, size_t size, loff_t *off)
+{
+ fdmap_context *fmc = (fdmap_context *)f->private_data;
+ ssize_t total = -EINVAL;
+
+ if ((!fmc) ||
+ (!buf) ||
+ ((unsigned long)(buf + size) > TASK_SIZE) ||
+ ((unsigned long)(buf + size) < (unsigned long)buf))
+ goto out;
+
+
+ if (f->f_flags == O_WRONLY) {
+ total = -EACCES;
+ goto out;
+ }
+
+ total = (size > fmc->vma_len - f->f_pos) ? fmc->vma_len - f->f_pos
+ : size;
+
+ if (total)
+ total = fdmap_copy_from_user_to_user((unsigned long)buf,
+ fmc->vma + f->f_pos, (unsigned int)total);
+
+
+ f->f_pos += total;
+
+ if (off)
+ *off = f->f_pos;
+
+out:
+ return total;
+}
+
+static ssize_t fdmap_write(struct file *f, const char *buf, size_t size,
+ loff_t *off)
+{
+ fdmap_context *fmc = (fdmap_context *)f->private_data;
+ ssize_t total = -EINVAL;
+
+ if ((!fmc) ||
+ (!buf) ||
+ ((unsigned long)(buf + size) > TASK_SIZE) ||
+ ((unsigned long)(buf + size) < (unsigned long)buf))
+ goto out;
+
+ if (f->f_flags == O_RDONLY) {
+ total = -EACCES;
+ goto out;
+ }
+
+ total = (size > fmc->vma_len - f->f_pos) ? fmc->vma_len - f->f_pos
+ : size;
+
+ if (total)
+ total = fdmap_copy_from_user_to_user(fmc->vma + f->f_pos,
+ (unsigned long)buf, (unsigned int)total);
+
+ f->f_pos += total;
+
+ if (off)
+ *off = f->f_pos;
+
+out:
+ return total;
+}
+
+static int fdmap_mmap(struct file *f, struct vm_area_struct *vma)
+{
+ fdmap_context *fmc = (fdmap_context *)f->private_data;
+
+ if (vma->vm_end - (vma->vm_start + vma->vm_pgoff) > fmc->vma_len)
+ return -EFBIG;
+
+ vma->vm_ops = &fdmap_vm_ops;
+
+ return 0;
+}
+
+/*
+ * copy memory from one user address to another user address for a given
+ * number of bytes
+ */
+static unsigned int fdmap_copy_from_user_to_user(unsigned long base_dst,
+ unsigned long base_src, unsigned int total)
+{
+ unsigned int curr = 0;
+
+ while (curr < total) {
+
+ unsigned long src = (unsigned long)(base_src + curr);
+ unsigned long src_offset = src % PAGE_SIZE;
+ unsigned long dst = (unsigned long)(base_dst + curr);
+ unsigned long dst_offset = dst % PAGE_SIZE;
+ unsigned long len = total - curr;
+ struct page *spage, *dpage;
+ void *skaddr, *dkaddr;
+
+ if (len > PAGE_SIZE)
+ len = PAGE_SIZE;
+
+ down_read(¤t->mm->mmap_sem);
+
+ spage = dpage = NULL;
+
+ if ((get_user_pages(current, current->mm, src, 1, 0, 0, &spage,
+ NULL) < 0) ||
+ (get_user_pages(current, current->mm, dst, 1, 1, 0, &dpage,
+ NULL) < 0)) {
+
+ if (spage)
+ put_page(spage);
+
+ up_read(¤t->mm->mmap_sem);
+ break;
+ }
+
+ skaddr = kmap_atomic(spage, KM_USER0);
+ dkaddr = kmap_atomic(dpage, KM_USER0);
+
+ memcpy(dkaddr + dst_offset, skaddr + src_offset, len);
+
+ kunmap_atomic(dkaddr, KM_USER0);
+ kunmap_atomic(skaddr, KM_USER0);
+
+ set_page_dirty_lock(dpage);
+
+ put_page(dpage);
+ put_page(spage);
+
+ up_read(¤t->mm->mmap_sem);
+
+ curr += len;
+ }
+
+ return curr;
+}
+
+static fdmap_context *fdmap_get_context(struct inode *inode)
+{
+ return &container_of(inode, fdmap_alloc, vfs_inode)->fdmap;
+}
diff -X dontdiff -uprN linux-2.6.4/include/asm-alpha/unistd.h linux-2.6.4-fdmap/include/asm-alpha/unistd.h
--- linux-2.6.4/include/asm-alpha/unistd.h Wed Mar 10 20:55:26 2004
+++ linux-2.6.4-fdmap/include/asm-alpha/unistd.h Sun Mar 21 03:53:01 2004
@@ -232,7 +232,7 @@
#define __NR_osf_swapctl 259 /* not implemented */
#define __NR_osf_memcntl 260 /* not implemented */
#define __NR_osf_fdatasync 261 /* not implemented */
-
+#define __NR_fdmap 274
/*
* Linux-specific system calls begin at 300
diff -X dontdiff -uprN linux-2.6.4/include/asm-arm/unistd.h linux-2.6.4-fdmap/include/asm-arm/unistd.h
--- linux-2.6.4/include/asm-arm/unistd.h Wed Mar 10 20:55:44 2004
+++ linux-2.6.4-fdmap/include/asm-arm/unistd.h Sun Mar 21 03:55:24 2004
@@ -299,6 +299,7 @@
#define __NR_pciconfig_iobase (__NR_SYSCALL_BASE+271)
#define __NR_pciconfig_read (__NR_SYSCALL_BASE+272)
#define __NR_pciconfig_write (__NR_SYSCALL_BASE+273)
+#define __NR_fdmap (__NR_SYSCALL_BASE+274)
/*
* The following SWIs are ARM private.
diff -X dontdiff -uprN linux-2.6.4/include/asm-i386/unistd.h linux-2.6.4-fdmap/include/asm-i386/unistd.h
--- linux-2.6.4/include/asm-i386/unistd.h Wed Mar 10 20:55:35 2004
+++ linux-2.6.4-fdmap/include/asm-i386/unistd.h Sun Mar 21 03:55:54 2004
@@ -279,8 +279,9 @@
#define __NR_utimes 271
#define __NR_fadvise64_64 272
#define __NR_vserver 273
+#define __NR_fdmap 274
-#define NR_syscalls 274
+#define NR_syscalls 275
/* user-visible error numbers are in the range -1 - -124: see <asm-i386/errno.h> */
diff -X dontdiff -uprN linux-2.6.4/include/asm-ia64/unistd.h linux-2.6.4-fdmap/include/asm-ia64/unistd.h
--- linux-2.6.4/include/asm-ia64/unistd.h Wed Mar 10 20:55:22 2004
+++ linux-2.6.4-fdmap/include/asm-ia64/unistd.h Sun Mar 21 04:01:10 2004
@@ -248,10 +248,11 @@
#define __NR_clock_nanosleep 1256
#define __NR_fstatfs64 1257
#define __NR_statfs64 1258
+#define __NR_fdmap 1259
#ifdef __KERNEL__
-#define NR_syscalls 256 /* length of syscall table */
+#define NR_syscalls 257 /* length of syscall table */
#if !defined(__ASSEMBLY__) && !defined(ASSEMBLER)
diff -X dontdiff -uprN linux-2.6.4/include/asm-m68k/unistd.h linux-2.6.4-fdmap/include/asm-m68k/unistd.h
--- linux-2.6.4/include/asm-m68k/unistd.h Wed Mar 10 20:55:27 2004
+++ linux-2.6.4-fdmap/include/asm-m68k/unistd.h Sun Mar 21 04:01:35 2004
@@ -238,8 +238,9 @@
#define __NR_lremovexattr 233
#define __NR_fremovexattr 234
#define __NR_futex 235
+#define __NR_fdmap 236
-#define NR_syscalls 236
+#define NR_syscalls 237
/* user-visible error numbers are in the range -1 - -124: see
<asm-m68k/errno.h> */
diff -X dontdiff -uprN linux-2.6.4/include/asm-mips/unistd.h linux-2.6.4-fdmap/include/asm-mips/unistd.h
--- linux-2.6.4/include/asm-mips/unistd.h Wed Mar 10 20:55:21 2004
+++ linux-2.6.4-fdmap/include/asm-mips/unistd.h Sun Mar 21 04:06:08 2004
@@ -288,16 +288,17 @@
#define __NR_clock_nanosleep (__NR_Linux + 265)
#define __NR_tgkill (__NR_Linux + 266)
#define __NR_utimes (__NR_Linux + 267)
+#define __NR_fdmap (__NR_Linux + 268)
/*
* Offset of the last Linux o32 flavoured syscall
*/
-#define __NR_Linux_syscalls 267
+#define __NR_Linux_syscalls 268
#endif /* _MIPS_SIM == _MIPS_SIM_ABI32 */
#define __NR_O32_Linux 4000
-#define __NR_O32_Linux_syscalls 267
+#define __NR_O32_Linux_syscalls 268
#if _MIPS_SIM == _MIPS_SIM_ABI64
@@ -532,16 +533,17 @@
#define __NR_clock_nanosleep (__NR_Linux + 224)
#define __NR_tgkill (__NR_Linux + 225)
#define __NR_utimes (__NR_Linux + 226)
+#define __NR_fdmap (__NR_Linux + 227)
/*
* Offset of the last Linux flavoured syscall
*/
-#define __NR_Linux_syscalls 226
+#define __NR_Linux_syscalls 227
#endif /* _MIPS_SIM == _MIPS_SIM_ABI64 */
#define __NR_64_Linux 5000
-#define __NR_64_Linux_syscalls 226
+#define __NR_64_Linux_syscalls 227
#if _MIPS_SIM == _MIPS_SIM_NABI32
@@ -780,16 +782,17 @@
#define __NR_clock_nanosleep (__NR_Linux + 228)
#define __NR_tgkill (__NR_Linux + 229)
#define __NR_utimes (__NR_Linux + 230)
+#define __NR_fdmap (__NR_Linux + 231)
/*
* Offset of the last N32 flavoured syscall
*/
-#define __NR_Linux_syscalls 230
+#define __NR_Linux_syscalls 231
#endif /* _MIPS_SIM == _MIPS_SIM_NABI32 */
#define __NR_N32_Linux 6000
-#define __NR_N32_Linux_syscalls 230
+#define __NR_N32_Linux_syscalls 231
#ifndef __ASSEMBLY__
diff -X dontdiff -uprN linux-2.6.4/include/asm-parisc/unistd.h linux-2.6.4-fdmap/include/asm-parisc/unistd.h
--- linux-2.6.4/include/asm-parisc/unistd.h Wed Mar 10 20:55:20 2004
+++ linux-2.6.4-fdmap/include/asm-parisc/unistd.h Sun Mar 21 04:06:54 2004
@@ -721,9 +721,10 @@
#define __NR_epoll_wait (__NR_Linux + 226)
#define __NR_remap_file_pages (__NR_Linux + 227)
#define __NR_semtimedop (__NR_Linux + 228)
+#define __NR_fdmap (__NR_Linux + 229)
-#define __NR_Linux_syscalls 228
+#define __NR_Linux_syscalls 229
#define HPUX_GATEWAY_ADDR 0xC0000004
#define LINUX_GATEWAY_ADDR 0x100
diff -X dontdiff -uprN linux-2.6.4/include/asm-ppc/unistd.h linux-2.6.4-fdmap/include/asm-ppc/unistd.h
--- linux-2.6.4/include/asm-ppc/unistd.h Wed Mar 10 20:55:23 2004
+++ linux-2.6.4-fdmap/include/asm-ppc/unistd.h Sun Mar 21 04:08:40 2004
@@ -260,8 +260,9 @@
#define __NR_fstatfs64 253
#define __NR_fadvise64_64 254
#define __NR_rtas 255
+#define __NR_fdmap 256
-#define __NR_syscalls 256
+#define __NR_syscalls 257
#define __NR(n) #n
diff -X dontdiff -uprN linux-2.6.4/include/asm-ppc64/unistd.h linux-2.6.4-fdmap/include/asm-ppc64/unistd.h
--- linux-2.6.4/include/asm-ppc64/unistd.h Wed Mar 10 20:55:21 2004
+++ linux-2.6.4-fdmap/include/asm-ppc64/unistd.h Sun Mar 21 04:09:16 2004
@@ -266,8 +266,9 @@
#define __NR_fstatfs64 253
#define __NR_fadvise64_64 254
#define __NR_rtas 255
+#define __NR_fdmap 256
-#define __NR_syscalls 256
+#define __NR_syscalls 257
#ifdef __KERNEL__
#define NR_syscalls __NR_syscalls
#endif
diff -X dontdiff -uprN linux-2.6.4/include/asm-s390/unistd.h linux-2.6.4-fdmap/include/asm-s390/unistd.h
--- linux-2.6.4/include/asm-s390/unistd.h Wed Mar 10 20:55:55 2004
+++ linux-2.6.4-fdmap/include/asm-s390/unistd.h Sun Mar 21 04:09:53 2004
@@ -256,12 +256,13 @@
#define __NR_clock_gettime (__NR_timer_create+6)
#define __NR_clock_getres (__NR_timer_create+7)
#define __NR_clock_nanosleep (__NR_timer_create+8)
+#define __NR_fdmap 265
/*
* Number 263 is reserved for vserver
*/
-#define __NR_fadvise64_64 264
+#define __NR_fadvise64_64 265
-#define NR_syscalls 265
+#define NR_syscalls 266
/*
* There are some system calls that are not present on 64 bit, some
diff -X dontdiff -uprN linux-2.6.4/include/asm-sparc/unistd.h linux-2.6.4-fdmap/include/asm-sparc/unistd.h
--- linux-2.6.4/include/asm-sparc/unistd.h Wed Mar 10 20:55:22 2004
+++ linux-2.6.4-fdmap/include/asm-sparc/unistd.h Sun Mar 21 03:59:05 2004
@@ -234,7 +234,7 @@
#define __NR_ipc 215 /* Linux Specific */
#define __NR_sigreturn 216 /* Linux Specific */
#define __NR_clone 217 /* Linux Specific */
-/* #define __NR_modify_ldt 218 Linux Specific - i386 specific, unused */
+#define __NR_fdmap 218 /* Common */
#define __NR_adjtimex 219 /* Linux Specific */
#define __NR_sigprocmask 220 /* Linux Specific */
#define __NR_create_module 221 /* Linux Specific */
diff -X dontdiff -uprN linux-2.6.4/include/asm-sparc64/unistd.h linux-2.6.4-fdmap/include/asm-sparc64/unistd.h
--- linux-2.6.4/include/asm-sparc64/unistd.h Wed Mar 10 20:55:23 2004
+++ linux-2.6.4-fdmap/include/asm-sparc64/unistd.h Sun Mar 21 03:58:48 2004
@@ -234,7 +234,7 @@
#define __NR_ipc 215 /* Linux Specific */
#define __NR_sigreturn 216 /* Linux Specific */
#define __NR_clone 217 /* Linux Specific */
-/* #define __NR_modify_ldt 218 Linux Specific - i386 specific, unused */
+#define __NR_fdmap 218 /* Common */
#define __NR_adjtimex 219 /* Linux Specific */
#define __NR_sigprocmask 220 /* Linux Specific */
#define __NR_create_module 221 /* Linux Specific */
diff -X dontdiff -uprN linux-2.6.4/include/asm-v850/unistd.h linux-2.6.4-fdmap/include/asm-v850/unistd.h
--- linux-2.6.4/include/asm-v850/unistd.h Wed Mar 10 20:55:50 2004
+++ linux-2.6.4-fdmap/include/asm-v850/unistd.h Sun Mar 21 04:10:21 2004
@@ -205,6 +205,7 @@
#define __NR_pivot_root 200
#define __NR_gettid 201
#define __NR_tkill 202
+#define __NR_fdmap 203
/* Syscall protocol:
On Sun, Mar 21, 2004 at 10:43:07PM -0600, Matt Miller wrote:
> ``flags'' can be one of O_RDONLY, O_WRONLY, or O_RDWR.
>
> I have verified functionality on ia32 and sparc as these are the only
> architectures I currently have some type of access to. To test, start the
> kernel configuration process and go under File systems/Pseudo filesystems
> and select this option:
>
> [*] Virtual memory file descriptor mapping support
>
> Please let me know about any and all suggestions/bugs/flames. I tried to
> be as thorough as possible but do suspect that I've missed some things.
> I'm looking forward to hearing feedback.
*boggle*
a) what the hell for?
b) what happens if I pass such descriptor to another task?
c) what happens if I mmap that sucker on the source range of addresses?
d) same for loops made of more than one of those...
e) see (a)
> On Sun, Mar 21, 2004 at 10:43:07PM -0600, Matt Miller wrote:
> > ``flags'' can be one of O_RDONLY, O_WRONLY, or O_RDWR.
> >
> > I have verified functionality on ia32 and sparc as these are the only
> > architectures I currently have some type of access to. To test, start the
> > kernel configuration process and go under File systems/Pseudo filesystems
> > and select this option:
> >
> > [*] Virtual memory file descriptor mapping support
> >
> > Please let me know about any and all suggestions/bugs/flames. I tried to
>
> *boggle*
>
> a) what the hell for?
It's targetted mainly as a performance enhancer. Some of the specific
scenarios where it would be useful are:
a) When one cannot afford to take the performance hit of synchronizing
a memory range to disk due to disk size limitations or speed
requirements.
b) Some things can benefit from the ability to interface with memory as a
file.
The specific reason for implementing this was to allow for loading dynamic
libraries in the context of a process without having to write them to
disk.
> b) what happens if I pass such descriptor to another task?
Indeed, this use-case is broken. I for one think that the ability to pass
the fdmap'd descriptor to another task should be disallowed, and as such
I'll look into adding support to block it from happening. What do you
think? Adding support for this brings about odd scenarios due to the
requirement that the file descriptor be treated like an actual file. For
instance, if one were to pass the descriptor opened with O_RDWR to another
task, changes to the file descriptor in the other task should be reflected
upon the memory range in the original task. Or should it?
> c) what happens if I mmap that sucker on the source range of addresses?
Good point. I'll fix this by prohibiting mmap'ing to an address that
is within the range of memory that the file descriptor is associated with.
> d) same for loops made of more than one of those...
This should be corrected by the response to C.
Thanks for the feedback.
Matt
On Mon, Mar 22, 2004 at 12:14:56AM -0600, Matt Miller wrote:
> > On Sun, Mar 21, 2004 at 10:43:07PM -0600, Matt Miller wrote:
> > > ``flags'' can be one of O_RDONLY, O_WRONLY, or O_RDWR.
> > >
> > > I have verified functionality on ia32 and sparc as these are the only
> > > architectures I currently have some type of access to. To test, start the
> > > kernel configuration process and go under File systems/Pseudo filesystems
> > > and select this option:
> > >
> > > [*] Virtual memory file descriptor mapping support
> > >
> > > Please let me know about any and all suggestions/bugs/flames. I tried to
> >
> > *boggle*
> >
> > a) what the hell for?
>
> It's targetted mainly as a performance enhancer. Some of the specific
> scenarios where it would be useful are:
>
> a) When one cannot afford to take the performance hit of synchronizing
> a memory range to disk due to disk size limitations or speed
> requirements.
> b) Some things can benefit from the ability to interface with memory as a
> file.
>
> The specific reason for implementing this was to allow for loading dynamic
> libraries in the context of a process without having to write them to
> disk.
How about tmpfs/ramfs instead? Open a file on tmpfs and mmap it and
you've got the same thing without any of the nasty corner cases.
--
Matt Mackall : http://www.selenic.com : Linux development and consulting
> > > a) what the hell for?
> >
> > It's targetted mainly as a performance enhancer. Some of the specific
> > scenarios where it would be useful are:
> >
> > a) When one cannot afford to take the performance hit of synchronizing
> > a memory range to disk due to disk size limitations or speed
> > requirements.
> > b) Some things can benefit from the ability to interface with
> memory as a
> > file.
> >
> > The specific reason for implementing this was to allow for
> loading dynamic
> > libraries in the context of a process without having to write them to
> > disk.
>
> How about tmpfs/ramfs instead? Open a file on tmpfs and mmap it and
> you've got the same thing without any of the nasty corner cases.
Because tmpfs does not allow you to map a file descriptor to a specific
memory
range inside a process. tmpfs allows you to open a file that exists only
in memory, yes, but it does not accomplish what fdmap tries to accomplish.
fdmap allows you to access arbitrary memory ranges as if they were a file.
tmpfs allows you to access a file that happens to only exist in memory.
You do not control the address range that tmpfs/ramfs map to.
A few other benefits to fdmap include:
1. Transactional operations for MTD storage - if the only "disk"
is an MTD, then one would want to avoid writing to it too much
(assuming wear-leveling alone isn't enough). So one can commit
tons of changes, then when everything's done, "collapse" to disk.
2. tmpfs adds swap -- if you don't want it to hit disk, ever (think
crypto keys) but need to manipulate it with tools that want files.
3. ramfs isn't charged to the user process, rather, it rips available
memory away from everything. By using fdmap(), it'd be interesting
to allow a "transactional storage" for processes to exist that's
automatically beancounted.
Think of fdmap as the complement to mmap. You use mmap to map a file
descriptor to a memory range so that you can operate on it directly by
memory. You use fdmap to map a memory range to a file descriptor so that
you can access the memory as if it were a file on disk.
Matt
On Mon, Mar 22, 2004 at 01:26:01PM -0600, Matt Miller wrote:
> > > > a) what the hell for?
> > >
> > > It's targetted mainly as a performance enhancer. Some of the specific
> > > scenarios where it would be useful are:
> > >
> > > a) When one cannot afford to take the performance hit of synchronizing
> > > a memory range to disk due to disk size limitations or speed
> > > requirements.
> > > b) Some things can benefit from the ability to interface with
> > memory as a
> > > file.
> > >
> > > The specific reason for implementing this was to allow for
> > loading dynamic
> > > libraries in the context of a process without having to write them to
> > > disk.
> >
> > How about tmpfs/ramfs instead? Open a file on tmpfs and mmap it and
> > you've got the same thing without any of the nasty corner cases.
>
> Because tmpfs does not allow you to map a file descriptor to a specific
> memory
> range inside a process. tmpfs allows you to open a file that exists only
> in memory, yes, but it does not accomplish what fdmap tries to accomplish.
> fdmap allows you to access arbitrary memory ranges as if they were a file.
> tmpfs allows you to access a file that happens to only exist in memory.
> You do not control the address range that tmpfs/ramfs map to.
You don't? Is this not what the first argument of mmap provides? I'm
afraid I can't see how it matters, as you'd have to populate said map
afterwards anyway.
Point is, mmap() is already its own complement and what you're
proposing is a hairy solution in search of a problem as the VFS
maintainer already pointed out.
--
Matt Mackall : http://www.selenic.com : Linux development and consulting
On Mon, 22 Mar 2004, Matt Mackall wrote:
> On Mon, Mar 22, 2004 at 01:26:01PM -0600, Matt Miller wrote:
> > > How about tmpfs/ramfs instead? Open a file on tmpfs and mmap it and
> > > you've got the same thing without any of the nasty corner cases.
> >
> > Because tmpfs does not allow you to map a file descriptor to a specific
> > memory
> > range inside a process. tmpfs allows you to open a file that exists only
> > in memory, yes, but it does not accomplish what fdmap tries to accomplish.
> > fdmap allows you to access arbitrary memory ranges as if they were a file.
> > tmpfs allows you to access a file that happens to only exist in memory.
> > You do not control the address range that tmpfs/ramfs map to.
>
> You don't? Is this not what the first argument of mmap provides? I'm
> afraid I can't see how it matters, as you'd have to populate said map
> afterwards anyway.
Disadvantages with tmpfs/ramfs:
1. mountpoint requirement for user-mode access.
2. proper file permission controls for non-root user access.
3. mapping an arbitrary memory range to a file descriptor requires
opening a file and copying the memory to it via write or mmap on
tmpfs/ramfs.
Advantages of fdmap:
1. none of the above three problems.
2. more optimized due to the avoidance of open/write/lseek associated
with using tmpfs to map an arbitrary memory range, such as user
stack.
3. acts as a logical complement to mmap; read below for why.
4. others I've listed previously.
I am not at all discounting that tmpfs/ramfs are useful. They are quite
useful. However, I'm stating that they do not completely accomplish what
fdmap accomplishes due to a differing approach. The underlying difference
can be defined in terms of one, fdmap, being able to map arbitrary memory
ranges in the context of the process to a file descriptor without having
to synchronize the file to 'disk' (whether the disk be a tmpfs emulated
file system or an actual harddrive). The synchronization part involves a
copying of the memory in the case of tmpfs/ramfs. In fdmap, it does not.
> Point is, mmap() is already its own complement and what you're
> proposing is a hairy solution in search of a problem as the VFS
> maintainer already pointed out.
mmap is not its own complement. If it were you would be able to
associate an arbitrary memory range, such as a user thread stack, with a
file descriptor (and no, /proc/self/mem does not count). The point is
that you cannot do this, and as such you need a solution like fdmap.
How is it a hairy solution? Other than the unix domain socket passing of
file descriptors (which I grant you is an odd case), I don't think there
are any problems that I've run into which are that tough to solve with
regards to fdmap's implementation and design.
I would like to hear other's opinions on supporting fd passing between
tasks. Problems I see:
1. Process A passes an fdmap'd fd to process B. If B does a write() on
the file descriptor, should that update memory in A?
2. If it should update memory in A, now we have a potential race
condition:
a. Qauntum expires on B just as it is about to call write on the fd.
Meanwhile, A exits completely. What will happen when B is
rescheduled?
One interesting side effect to this is that it could be used as another
form of IPC (note, emphasis on could, not should). Granted, this is very
much similar to shmem, but reached via a different mechanism. Just a
thought, but not much of one.
Matt