2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * Direct file pointer API functions for in-kernel operations on files. These
37 * functions provide a open/read/write/close like interface within the kernel
38 * for operating on files that are not necessarily associated with processes
39 * and which do not (typically) have descriptors.
41 * FUTURE: file handle conversion routines to support checkpointing,
42 * and additional file operations (ioctl, fcntl).
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
49 #include <sys/sysproto.h>
51 #include <sys/filedesc.h>
52 #include <sys/sysctl.h>
53 #include <sys/vnode.h>
56 #include <sys/nlookup.h>
59 #include <sys/filio.h>
60 #include <sys/fcntl.h>
61 #include <sys/unistd.h>
62 #include <sys/resourcevar.h>
63 #include <sys/event.h>
67 #include <vm/vm_param.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
79 #include <sys/file2.h>
80 #include <machine/limits.h>
82 typedef struct file
*file_t
;
87 * Open a file as specified. Use O_* flags for flags.
89 * NOTE! O_ROOTCRED not quite working yet, vn_open() asserts that the
90 * cred must match the process's cred. XXX
92 * NOTE! when fp_open() is called from a pure thread, root creds are
96 fp_open(const char *path
, int flags
, int mode
, file_t
*fpp
)
98 struct nlookupdata nd
;
103 if ((error
= falloc(NULL
, fpp
, NULL
)) != 0)
108 if ((flags
& O_ROOTCRED
) == 0)
109 fsetcred(fp
, td
->td_proc
->p_ucred
);
111 error
= nlookup_init(&nd
, path
, UIO_SYSSPACE
, NLC_LOCKVP
);
112 flags
= FFLAGS(flags
);
114 error
= vn_open(&nd
, fp
, flags
, mode
);
125 * fp_vpopen(): convert a vnode to a file pointer, call VOP_OPEN() on the
126 * the vnode. The vnode must be refd and locked.
128 * On success the vnode's ref is inherited by the file pointer and the caller
129 * should not vrele() it, and the vnode is unlocked.
131 * On failure the vnode remains locked and refd and the caller is responsible
135 fp_vpopen(struct vnode
*vp
, int flags
, file_t
*fpp
)
145 * Vnode checks (from vn_open())
147 if (vp
->v_type
== VLNK
) {
151 if (vp
->v_type
== VSOCK
) {
155 flags
= FFLAGS(flags
);
157 if (flags
& (FWRITE
| O_TRUNC
)) {
158 if (vp
->v_type
== VDIR
) {
162 error
= vn_writechk(vp
, NULL
);
170 error
= VOP_ACCESS(vp
, vmode
, td
->td_proc
->p_ucred
);
178 if ((error
= falloc(NULL
, fpp
, NULL
)) != 0)
181 if ((flags
& O_ROOTCRED
) == 0 && td
->td_proc
)
182 fsetcred(fp
, td
->td_proc
->p_ucred
);
184 error
= VOP_OPEN(vp
, flags
, td
->td_proc
->p_ucred
, fp
);
191 fp
->f_ops
= &badfileops
; /* open failed, don't close */
194 /* leave the vnode intact, but fall through and unlock it anyway */
201 * fp_*read() is meant to operate like the normal descriptor based syscalls
202 * would. Note that if 'buf' points to user memory a UIO_USERSPACE
203 * transfer will be used.
206 fp_pread(file_t fp
, void *buf
, size_t nbytes
, off_t offset
, ssize_t
*res
,
216 if (nbytes
> LONG_MAX
)
218 bzero(&auio
, sizeof(auio
));
219 aiov
.iov_base
= (caddr_t
)buf
;
220 aiov
.iov_len
= nbytes
;
221 auio
.uio_iov
= &aiov
;
223 auio
.uio_offset
= offset
;
224 auio
.uio_resid
= nbytes
;
225 auio
.uio_rw
= UIO_READ
;
226 auio
.uio_segflg
= seg
;
227 auio
.uio_td
= curthread
;
230 error
= fo_read(fp
, &auio
, fp
->f_cred
, O_FOFFSET
);
232 if (auio
.uio_resid
!= nbytes
&& (error
== ERESTART
|| error
== EINTR
||
233 error
== EWOULDBLOCK
)
238 count
-= auio
.uio_resid
;
245 fp_read(file_t fp
, void *buf
, size_t nbytes
, ssize_t
*res
, int all
,
255 if (nbytes
> LONG_MAX
)
257 bzero(&auio
, sizeof(auio
));
258 aiov
.iov_base
= (caddr_t
)buf
;
259 aiov
.iov_len
= nbytes
;
260 auio
.uio_iov
= &aiov
;
263 auio
.uio_resid
= nbytes
;
264 auio
.uio_rw
= UIO_READ
;
265 auio
.uio_segflg
= seg
;
266 auio
.uio_td
= curthread
;
269 * If all is false call fo_read() once.
270 * If all is true we attempt to read the entire request. We have to
271 * break out of the loop if an unrecoverable error or EOF occurs.
274 lastresid
= auio
.uio_resid
;
275 error
= fo_read(fp
, &auio
, fp
->f_cred
, 0);
276 } while (all
&& auio
.uio_resid
&&
277 ((error
== 0 && auio
.uio_resid
!= lastresid
) ||
278 error
== ERESTART
|| error
== EINTR
));
279 if (all
&& error
== 0 && auio
.uio_resid
)
283 * If an error occured but some data was read, silently forget the
284 * error. However, if this is a non-blocking descriptor and 'all'
285 * was specified, return an error even if some data was read (this
286 * is considered a bug in the caller for using an illegal combination
287 * of 'all' and a non-blocking descriptor).
290 if (auio
.uio_resid
!= nbytes
) {
291 if (error
== ERESTART
|| error
== EINTR
)
293 if (error
== EWOULDBLOCK
&& all
== 0)
298 *res
= nbytes
- auio
.uio_resid
;
303 fp_pwrite(file_t fp
, void *buf
, size_t nbytes
, off_t offset
, ssize_t
*res
,
313 if (nbytes
> LONG_MAX
)
315 bzero(&auio
, sizeof(auio
));
316 aiov
.iov_base
= (caddr_t
)buf
;
317 aiov
.iov_len
= nbytes
;
318 auio
.uio_iov
= &aiov
;
320 auio
.uio_offset
= offset
;
321 auio
.uio_resid
= nbytes
;
322 auio
.uio_rw
= UIO_WRITE
;
323 auio
.uio_segflg
= seg
;
324 auio
.uio_td
= curthread
;
327 error
= fo_write(fp
, &auio
, fp
->f_cred
, O_FOFFSET
);
329 if (auio
.uio_resid
!= nbytes
&& (error
== ERESTART
|| error
== EINTR
||
330 error
== EWOULDBLOCK
)
335 count
-= auio
.uio_resid
;
343 fp_write(file_t fp
, void *buf
, size_t nbytes
, ssize_t
*res
, enum uio_seg seg
)
352 if (nbytes
> LONG_MAX
)
354 bzero(&auio
, sizeof(auio
));
355 aiov
.iov_base
= (caddr_t
)buf
;
356 aiov
.iov_len
= nbytes
;
357 auio
.uio_iov
= &aiov
;
360 auio
.uio_resid
= nbytes
;
361 auio
.uio_rw
= UIO_WRITE
;
362 auio
.uio_segflg
= seg
;
363 auio
.uio_td
= curthread
;
366 error
= fo_write(fp
, &auio
, fp
->f_cred
, 0);
368 if (auio
.uio_resid
!= nbytes
&& (error
== ERESTART
|| error
== EINTR
||
369 error
== EWOULDBLOCK
)
374 count
-= auio
.uio_resid
;
381 fp_stat(file_t fp
, struct stat
*ub
)
385 error
= fo_stat(fp
, ub
, fp
->f_cred
);
390 * non-anonymous, non-stack descriptor mappings only!
392 * This routine mostly snarfed from vm/vm_mmap.c
395 fp_mmap(void *addr_arg
, size_t size
, int prot
, int flags
, struct file
*fp
,
396 off_t pos
, void **resp
)
398 struct thread
*td
= curthread
;
399 struct proc
*p
= td
->td_proc
;
406 struct vmspace
*vms
= p
->p_vmspace
;
411 if ((ssize_t
)size
< 0 || (flags
& MAP_ANON
))
414 pageoff
= (pos
& PAGE_MASK
);
417 /* Adjust size for rounding (on both ends). */
418 size
+= pageoff
; /* low end... */
419 size
= (vm_size_t
)round_page(size
); /* hi end */
420 addr
= (vm_offset_t
)addr_arg
;
423 * Check for illegal addresses. Watch out for address wrap... Note
424 * that VM_*_ADDRESS are not constants due to casts (argh).
426 if (flags
& MAP_FIXED
) {
428 * The specified address must have the same remainder
429 * as the file offset taken modulo PAGE_SIZE, so it
430 * should be aligned after adjustment by pageoff.
433 if (addr
& PAGE_MASK
)
435 /* Address range must be all in user VM space. */
436 if (VM_MAX_USER_ADDRESS
> 0 && addr
+ size
> VM_MAX_USER_ADDRESS
)
438 if (VM_MIN_USER_ADDRESS
> 0 && addr
< VM_MIN_USER_ADDRESS
)
440 if (addr
+ size
< addr
)
442 } else if (addr
== 0 ||
443 (addr
>= round_page((vm_offset_t
)vms
->vm_taddr
) &&
444 addr
< round_page((vm_offset_t
)vms
->vm_daddr
+ maxdsiz
))
447 * XXX for non-fixed mappings where no hint is provided or
448 * the hint would fall in the potential heap space,
449 * place it after the end of the largest possible heap.
451 * There should really be a pmap call to determine a reasonable
454 addr
= round_page((vm_offset_t
)vms
->vm_daddr
+ maxdsiz
);
458 * Mapping file, get fp for validation. Obtain vnode and make
459 * sure it is of appropriate type.
461 if (fp
->f_type
!= DTYPE_VNODE
)
465 * POSIX shared-memory objects are defined to have
466 * kernel persistence, and are not defined to support
467 * read(2)/write(2) -- or even open(2). Thus, we can
468 * use MAP_ASYNC to trade on-disk coherence for speed.
469 * The shm_open(3) library routine turns on the FPOSIXSHM
470 * flag to request this behavior.
472 if (fp
->f_flag
& FPOSIXSHM
)
474 vp
= (struct vnode
*) fp
->f_data
;
475 if (vp
->v_type
!= VREG
&& vp
->v_type
!= VCHR
)
479 * Get the proper underlying object
481 if (vp
->v_type
== VREG
) {
482 if ((obj
= vp
->v_object
) == NULL
)
484 KKASSERT(vp
== (struct vnode
*)obj
->handle
);
488 * XXX hack to handle use of /dev/zero to map anon memory (ala
491 if (vp
->v_type
== VCHR
&& iszerodev(vp
->v_rdev
)) {
493 maxprot
= VM_PROT_ALL
;
498 * cdevs does not provide private mappings of any kind.
500 if (vp
->v_type
== VCHR
&&
501 (flags
& (MAP_PRIVATE
|MAP_COPY
))) {
506 * Ensure that file and memory protections are
507 * compatible. Note that we only worry about
508 * writability if mapping is shared; in this case,
509 * current and max prot are dictated by the open file.
510 * XXX use the vnode instead? Problem is: what
511 * credentials do we use for determination? What if
512 * proc does a setuid?
514 maxprot
= VM_PROT_EXECUTE
; /* ??? */
515 if (fp
->f_flag
& FREAD
) {
516 maxprot
|= VM_PROT_READ
;
517 } else if (prot
& PROT_READ
) {
522 * If we are sharing potential changes (either via
523 * MAP_SHARED or via the implicit sharing of character
524 * device mappings), and we are trying to get write
525 * permission although we opened it without asking
529 if ((flags
& MAP_SHARED
) != 0 ||
532 if ((fp
->f_flag
& FWRITE
) != 0) {
534 if ((error
= VOP_GETATTR(vp
, &va
))) {
537 if ((va
.va_flags
& (IMMUTABLE
|APPEND
)) == 0) {
538 maxprot
|= VM_PROT_WRITE
;
539 } else if (prot
& PROT_WRITE
) {
543 } else if ((prot
& PROT_WRITE
) != 0) {
548 maxprot
|= VM_PROT_WRITE
;
552 error
= vm_mmap(&vms
->vm_map
, &addr
, size
, prot
,
553 maxprot
, flags
, handle
, pos
);
554 if (error
== 0 && addr_arg
)
555 *resp
= (void *)addr
;
567 fp_shutdown(file_t fp
, int how
)
569 return(fo_shutdown(fp
, how
));