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>
49 #include <sys/malloc.h>
50 #include <sys/sysmsg.h>
52 #include <sys/filedesc.h>
53 #include <sys/sysctl.h>
54 #include <sys/vnode.h>
57 #include <sys/nlookup.h>
60 #include <sys/filio.h>
61 #include <sys/fcntl.h>
62 #include <sys/unistd.h>
63 #include <sys/resourcevar.h>
64 #include <sys/event.h>
68 #include <vm/vm_param.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vm_pageout.h>
76 #include <vm/vm_extern.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 * vn_open() asserts that the cred must match the process's cred.
91 * NOTE! when fp_open() is called from a pure thread, root creds are
95 fp_open(const char *path, int flags, int mode, file_t *fpp)
97 struct nlookupdata nd;
102 if ((error = falloc(NULL, fpp, NULL)) != 0)
107 fsetcred(fp, td->td_proc->p_ucred);
108 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_LOCKVP);
109 flags = FFLAGS(flags);
111 error = vn_open(&nd, fp, flags, mode);
122 * fp_vpopen(): convert a vnode to a file pointer, call VOP_OPEN() on the
123 * the vnode. The vnode must be refd and locked.
125 * On success the vnode's ref is inherited by the file pointer and the caller
126 * should not vrele() it, and the vnode is unlocked.
128 * On failure the vnode remains locked and refd and the caller is responsible
132 fp_vpopen(struct vnode *vp, int flags, file_t *fpp)
142 * Vnode checks (from vn_open())
144 if (vp->v_type == VLNK) {
148 if (vp->v_type == VSOCK) {
152 flags = FFLAGS(flags);
154 if (flags & (FWRITE | O_TRUNC)) {
155 if (vp->v_type == VDIR) {
159 error = vn_writechk(vp);
167 error = VOP_ACCESS(vp, vmode, td->td_proc->p_ucred);
175 if ((error = falloc(NULL, fpp, NULL)) != 0)
179 fsetcred(fp, td->td_proc->p_ucred);
181 error = VOP_OPEN(vp, flags, td->td_proc->p_ucred, fp);
188 fp->f_ops = &badfileops; /* open failed, don't close */
191 /* leave the vnode intact, but fall through and unlock it anyway */
198 * fp_*read() is meant to operate like the normal descriptor based syscalls
199 * would. Note that if 'buf' points to user memory a UIO_USERSPACE
200 * transfer will be used.
203 fp_pread(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res,
213 if (nbytes > LONG_MAX)
215 bzero(&auio, sizeof(auio));
216 aiov.iov_base = (caddr_t)buf;
217 aiov.iov_len = nbytes;
218 auio.uio_iov = &aiov;
220 auio.uio_offset = offset;
221 auio.uio_resid = nbytes;
222 auio.uio_rw = UIO_READ;
223 auio.uio_segflg = seg;
224 auio.uio_td = curthread;
227 error = fo_read(fp, &auio, fp->f_cred, O_FOFFSET);
229 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
230 error == EWOULDBLOCK)
235 count -= auio.uio_resid;
242 fp_read(file_t fp, void *buf, size_t nbytes, ssize_t *res, int all,
252 if (nbytes > LONG_MAX)
254 bzero(&auio, sizeof(auio));
255 aiov.iov_base = (caddr_t)buf;
256 aiov.iov_len = nbytes;
257 auio.uio_iov = &aiov;
260 auio.uio_resid = nbytes;
261 auio.uio_rw = UIO_READ;
262 auio.uio_segflg = seg;
263 auio.uio_td = curthread;
266 * If all is false call fo_read() once.
267 * If all is true we attempt to read the entire request. We have to
268 * break out of the loop if an unrecoverable error or EOF occurs.
271 lastresid = auio.uio_resid;
272 error = fo_read(fp, &auio, fp->f_cred, 0);
273 } while (all && auio.uio_resid &&
274 ((error == 0 && auio.uio_resid != lastresid) ||
275 error == ERESTART || error == EINTR));
276 if (all && error == 0 && auio.uio_resid)
280 * If an error occured but some data was read, silently forget the
281 * error. However, if this is a non-blocking descriptor and 'all'
282 * was specified, return an error even if some data was read (this
283 * is considered a bug in the caller for using an illegal combination
284 * of 'all' and a non-blocking descriptor).
287 if (auio.uio_resid != nbytes) {
288 if (error == ERESTART || error == EINTR)
290 if (error == EWOULDBLOCK && all == 0)
295 *res = nbytes - auio.uio_resid;
300 fp_pwrite(file_t fp, void *buf, size_t nbytes, off_t offset, ssize_t *res,
310 if (nbytes > LONG_MAX)
312 bzero(&auio, sizeof(auio));
313 aiov.iov_base = (caddr_t)buf;
314 aiov.iov_len = nbytes;
315 auio.uio_iov = &aiov;
317 auio.uio_offset = offset;
318 auio.uio_resid = nbytes;
319 auio.uio_rw = UIO_WRITE;
320 auio.uio_segflg = seg;
321 auio.uio_td = curthread;
324 error = fo_write(fp, &auio, fp->f_cred, O_FOFFSET);
326 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
327 error == EWOULDBLOCK)
332 count -= auio.uio_resid;
340 fp_write(file_t fp, void *buf, size_t nbytes, ssize_t *res, enum uio_seg seg)
349 if (nbytes > LONG_MAX)
351 bzero(&auio, sizeof(auio));
352 aiov.iov_base = (caddr_t)buf;
353 aiov.iov_len = nbytes;
354 auio.uio_iov = &aiov;
357 auio.uio_resid = nbytes;
358 auio.uio_rw = UIO_WRITE;
359 auio.uio_segflg = seg;
360 auio.uio_td = curthread;
363 error = fo_write(fp, &auio, fp->f_cred, 0);
365 if (auio.uio_resid != nbytes && (error == ERESTART || error == EINTR ||
366 error == EWOULDBLOCK)
371 count -= auio.uio_resid;
378 fp_stat(file_t fp, struct stat *ub)
382 error = fo_stat(fp, ub, fp->f_cred);
387 * non-anonymous, non-stack descriptor mappings only!
389 * This routine mostly snarfed from vm/vm_mmap.c
392 fp_mmap(void *addr_arg, size_t size, int prot, int flags, struct file *fp,
393 off_t pos, void **resp)
395 struct thread *td = curthread;
396 struct proc *p = td->td_proc;
403 struct vmspace *vms = p->p_vmspace;
408 if ((ssize_t)size < 0 || (flags & MAP_ANON))
411 pageoff = (pos & PAGE_MASK);
414 /* Adjust size for rounding (on both ends). */
415 size += pageoff; /* low end... */
416 size = (vm_size_t)round_page(size); /* hi end */
417 addr = (vm_offset_t)addr_arg;
420 * Check for illegal addresses. Watch out for address wrap... Note
421 * that VM_*_ADDRESS are not constants due to casts (argh).
423 if (flags & MAP_FIXED) {
425 * The specified address must have the same remainder
426 * as the file offset taken modulo PAGE_SIZE, so it
427 * should be aligned after adjustment by pageoff.
430 if (addr & PAGE_MASK)
432 /* Address range must be all in user VM space. */
433 if (VM_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS)
435 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
437 if (addr + size < addr)
439 } else if (addr == 0 ||
440 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
441 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))
444 * XXX for non-fixed mappings where no hint is provided or
445 * the hint would fall in the potential heap space,
446 * place it after the end of the largest possible heap.
448 * There should really be a pmap call to determine a reasonable
451 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
455 * Mapping file, get fp for validation. Obtain vnode and make
456 * sure it is of appropriate type.
458 if (fp->f_type != DTYPE_VNODE)
462 * POSIX shared-memory objects are defined to have
463 * kernel persistence, and are not defined to support
464 * read(2)/write(2) -- or even open(2). Thus, we can
465 * use MAP_ASYNC to trade on-disk coherence for speed.
466 * The shm_open(3) library routine turns on the FPOSIXSHM
467 * flag to request this behavior.
469 if (fp->f_flag & FPOSIXSHM)
471 vp = (struct vnode *) fp->f_data;
472 if (vp->v_type != VREG && vp->v_type != VCHR)
476 * Get the proper underlying object
478 if (vp->v_type == VREG) {
479 if ((obj = vp->v_object) == NULL)
481 KKASSERT(vp == (struct vnode *)obj->handle);
485 * XXX hack to handle use of /dev/zero to map anon memory (ala
488 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
490 maxprot = VM_PROT_ALL;
495 * cdevs does not provide private mappings of any kind.
497 if (vp->v_type == VCHR &&
498 (flags & (MAP_PRIVATE|MAP_COPY))) {
503 * Ensure that file and memory protections are
504 * compatible. Note that we only worry about
505 * writability if mapping is shared; in this case,
506 * current and max prot are dictated by the open file.
507 * XXX use the vnode instead? Problem is: what
508 * credentials do we use for determination? What if
509 * proc does a setuid?
511 maxprot = VM_PROT_EXECUTE; /* ??? */
512 if (fp->f_flag & FREAD) {
513 maxprot |= VM_PROT_READ;
514 } else if (prot & PROT_READ) {
519 * If we are sharing potential changes (either via
520 * MAP_SHARED or via the implicit sharing of character
521 * device mappings), and we are trying to get write
522 * permission although we opened it without asking
526 if ((flags & MAP_SHARED) != 0 ||
529 if ((fp->f_flag & FWRITE) != 0) {
531 if ((error = VOP_GETATTR_FP(vp, &va, fp))) {
534 if ((va.va_flags & (IMMUTABLE|APPEND)) == 0) {
535 maxprot |= VM_PROT_WRITE;
536 } else if (prot & PROT_WRITE) {
540 } else if ((prot & PROT_WRITE) != 0) {
545 maxprot |= VM_PROT_WRITE;
549 error = vm_mmap(&vms->vm_map, &addr, size, prot,
550 maxprot, flags, handle, pos);
551 if (error == 0 && addr_arg)
552 *resp = (void *)addr;
564 fp_shutdown(file_t fp, int how)
566 return(fo_shutdown(fp, how));