xref: /dragonfly/sys/kern/kern_xio.c (revision 71126e33)
1 /*
2  * Copyright (c) 2003,2004 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
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
16  *    distribution.
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.
20  *
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
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sys/kern/kern_xio.c,v 1.7 2004/07/16 05:51:10 dillon Exp $
35  */
36 /*
37  * Kernel XIO interface.  An initialized XIO is basically a collection of
38  * appropriately held vm_page_t's.  XIO buffers are vmspace agnostic and
39  * can represent userspace or kernelspace buffers, and can be passed to
40  * foreign threads outside of the originating vmspace.  XIO buffers are
41  * not mapped into KVM and thus can be manipulated and passed around with
42  * very low overheads.
43  *
44  * The intent is for XIO to be used in the I/O path, VFS, CAPS, and other
45  * places that need to pass (possibly userspace) data between threads.
46  *
47  * TODO: check for busy page when modifying, check writeable.
48  */
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/malloc.h>
53 #include <sys/proc.h>
54 #include <sys/vmmeter.h>
55 #include <sys/vnode.h>
56 #include <sys/xio.h>
57 #include <sys/sfbuf.h>
58 
59 #include <vm/vm.h>
60 #include <vm/vm_param.h>
61 #include <sys/lock.h>
62 #include <vm/vm_kern.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pageout.h>
68 #include <vm/vm_pager.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_page2.h>
71 
72 /*
73  * Just do basic initialization of an empty XIO
74  */
75 void
76 xio_init(xio_t xio)
77 {
78     xio->xio_flags = 0;
79     xio->xio_bytes = 0;
80     xio->xio_error = 0;
81     xio->xio_offset = 0;
82     xio->xio_npages = 0;
83     xio->xio_pages = xio->xio_internal_pages;
84 }
85 
86 /*
87  * Initialize an XIO given a userspace buffer.  0 is returned on success,
88  * an error code on failure.  The actual number of bytes that could be
89  * accomodated in the XIO will be stored in xio_bytes.
90  *
91  * Note that you cannot legally accessed a previously cached linmap with
92  * a newly initialized xio until after calling xio_linmap().
93  */
94 int
95 xio_init_ubuf(xio_t xio, void *ubase, size_t ubytes, int flags)
96 {
97     vm_offset_t addr;
98     vm_paddr_t paddr;
99     vm_page_t m;
100     int i;
101     int n;
102     int s;
103     int vmprot;
104 
105     addr = trunc_page((vm_offset_t)ubase);
106     xio->xio_flags = flags;
107     xio->xio_bytes = 0;
108     xio->xio_error = 0;
109     if (ubytes == 0) {
110 	xio->xio_offset = 0;
111 	xio->xio_npages = 0;
112     } else {
113 	vmprot = (flags & XIOF_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
114 	xio->xio_offset = (vm_offset_t)ubase & PAGE_MASK;
115 	xio->xio_pages = xio->xio_internal_pages;
116 	if ((n = PAGE_SIZE - xio->xio_offset) > ubytes)
117 	    n = ubytes;
118 	for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
119 	    if (vm_fault_quick((caddr_t)addr, vmprot) < 0)
120 		break;
121 	    if ((paddr = pmap_kextract(addr)) == 0)
122 		break;
123 	    s = splvm();
124 	    m = PHYS_TO_VM_PAGE(paddr);
125 	    vm_page_hold(m);
126 	    splx(s);
127 	    xio->xio_pages[i] = m;
128 	    ubytes -= n;
129 	    xio->xio_bytes += n;
130 	    if ((n = ubytes) > PAGE_SIZE)
131 		n = PAGE_SIZE;
132 	    addr += PAGE_SIZE;
133 	}
134 	xio->xio_npages = i;
135 
136 	/*
137 	 * If a failure occured clean out what we loaded and return EFAULT.
138 	 * Return 0 on success.
139 	 */
140 	if (i < XIO_INTERNAL_PAGES && n) {
141 	    xio_release(xio);
142 	    xio->xio_error = EFAULT;
143 	}
144     }
145     return(xio->xio_error);
146 }
147 
148 /*
149  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
150  * an error code on failure.  The actual number of bytes that could be
151  * accomodated in the XIO will be stored in xio_bytes.
152  *
153  * vmprot is usually either VM_PROT_READ or VM_PROT_WRITE.
154  *
155  * Note that you cannot legally accessed a previously cached linmap with
156  * a newly initialized xio until after calling xio_linmap().
157  */
158 int
159 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
160 {
161     vm_offset_t addr;
162     vm_paddr_t paddr;
163     vm_page_t m;
164     int i;
165     int n;
166     int s;
167 
168     addr = trunc_page((vm_offset_t)kbase);
169     xio->xio_flags = 0;
170     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
171     xio->xio_bytes = 0;
172     xio->xio_pages = xio->xio_internal_pages;
173     xio->xio_error = 0;
174     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
175 	n = kbytes;
176     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
177 	if ((paddr = pmap_kextract(addr)) == 0)
178 	    break;
179 	s = splvm();
180 	m = PHYS_TO_VM_PAGE(paddr);
181 	vm_page_hold(m);
182 	splx(s);
183 	xio->xio_pages[i] = m;
184 	kbytes -= n;
185 	xio->xio_bytes += n;
186 	if ((n = kbytes) > PAGE_SIZE)
187 	    n = PAGE_SIZE;
188 	addr += PAGE_SIZE;
189     }
190     xio->xio_npages = i;
191 
192     /*
193      * If a failure occured clean out what we loaded and return EFAULT.
194      * Return 0 on success.
195      */
196     if (i < XIO_INTERNAL_PAGES && n) {
197 	xio_release(xio);
198 	xio->xio_error = EFAULT;
199     }
200     return(xio->xio_error);
201 }
202 
203 /*
204  * Cleanup an XIO so it can be destroyed.  The pages associated with the
205  * XIO are released.  If a linear mapping buffer is active, it will be
206  * unlocked but the mappings will be left intact for optimal reconstitution
207  * in a later xio_linmap() call.
208  *
209  * Note that you cannot legally accessed the linmap on a released XIO.
210  */
211 void
212 xio_release(xio_t xio)
213 {
214     int i;
215     int s;
216     vm_page_t m;
217 
218     s = splvm();
219     for (i = 0; i < xio->xio_npages; ++i) {
220 	m = xio->xio_pages[i];
221 	vm_page_unhold(m);
222     }
223     splx(s);
224     if (xio->xio_flags & XIOF_LINMAP) {
225 	xio->xio_flags &= ~XIOF_LINMAP;
226 	/* XXX */
227     }
228     xio->xio_offset = 0;
229     xio->xio_npages = 0;
230     xio->xio_bytes = 0;
231     xio->xio_error = ENOBUFS;
232 }
233 
234 /*
235  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
236  * must be relative to the current context.  Both the UIO and the XIO are
237  * modified, but the XIO's pages are not released when exhausted.
238  *
239  * UIO_READ	xio -> uio
240  * UIO_WRITE	uio -> xio
241  */
242 int
243 xio_uio_copy(xio_t xio, struct uio *uio, int *sizep)
244 {
245     int error;
246     int bytes;
247 
248     if ((bytes = xio->xio_bytes) > uio->uio_resid)
249 	bytes = uio->uio_resid;
250     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset, bytes, uio);
251     if (error == 0) {
252 	xio->xio_bytes -= bytes;
253 	xio->xio_offset += bytes;
254 	*sizep = bytes;
255     } else {
256 	*sizep = 0;
257     }
258     return(error);
259 }
260 
261 /*
262  * Copy the specified number of bytes from the xio to a userland
263  * buffer.  Return an error code or 0 on success.
264  *
265  * The XIO is modified, but the XIO's pages are not released when exhausted.
266  */
267 int
268 xio_copy_xtou(xio_t xio, void *uptr, int bytes)
269 {
270     int i;
271     int n;
272     int error;
273     int offset;
274     vm_page_t m;
275     struct sf_buf *sf;
276 
277     if (bytes > xio->xio_bytes)
278 	return(EFAULT);
279 
280     offset = xio->xio_offset & PAGE_MASK;
281     if ((n = PAGE_SIZE - offset) > bytes)
282 	n = bytes;
283 
284     error = 0;
285     for (i = xio->xio_offset >> PAGE_SHIFT; i < xio->xio_npages; ++i) {
286 	m = xio->xio_pages[i];
287 	sf = sf_buf_alloc(m, SFBA_QUICK);
288 	error = copyout((char *)sf_buf_kva(sf) + offset, uptr, n);
289 	sf_buf_free(sf);
290 	if (error)
291 	    break;
292 	bytes -= n;
293 	xio->xio_bytes -= n;
294 	xio->xio_offset += n;
295 	uptr = (char *)uptr + n;
296 	if (bytes == 0)
297 	    break;
298 	if ((n = bytes) > PAGE_SIZE)
299 	    n = PAGE_SIZE;
300 	offset = 0;
301     }
302     return(error);
303 }
304 
305 /*
306  * Copy the specified number of bytes from the xio to a kernel
307  * buffer.  Return an error code or 0 on success.
308  *
309  * The XIO is modified, but the XIO's pages are not released when exhausted.
310  */
311 int
312 xio_copy_xtok(xio_t xio, void *kptr, int bytes)
313 {
314     int i;
315     int n;
316     int error;
317     int offset;
318     vm_page_t m;
319     struct sf_buf *sf;
320 
321     if (bytes > xio->xio_bytes)
322 	return(EFAULT);
323 
324     offset = xio->xio_offset & PAGE_MASK;
325     if ((n = PAGE_SIZE - offset) > bytes)
326 	n = bytes;
327 
328     error = 0;
329     for (i = xio->xio_offset >> PAGE_SHIFT; i < xio->xio_npages; ++i) {
330 	m = xio->xio_pages[i];
331 	sf = sf_buf_alloc(m, SFBA_QUICK);
332 	bcopy((char *)sf_buf_kva(sf) + offset, kptr, n);
333 	sf_buf_free(sf);
334 	bytes -= n;
335 	xio->xio_bytes -= n;
336 	xio->xio_offset += n;
337 	kptr = (char *)kptr + n;
338 	if (bytes == 0)
339 	    break;
340 	if ((n = bytes) > PAGE_SIZE)
341 	    n = PAGE_SIZE;
342 	offset = 0;
343     }
344     return(error);
345 }
346 
347