xref: /dflybsd-src/sys/kern/kern_xio.c (revision 23c32883e759b0ea42fdaff39e661bd1a12e3b9f)
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 1.9 2005/03/02 18:42:08 hmp 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 and the page offset
90  * will be stored in xio_offset.
91  */
92 int
93 xio_init_ubuf(xio_t xio, void *ubase, size_t ubytes, int flags)
94 {
95     vm_offset_t addr;
96     vm_paddr_t paddr;
97     vm_page_t m;
98     int i;
99     int n;
100     int s;
101     int vmprot;
102 
103     addr = trunc_page((vm_offset_t)ubase);
104     xio->xio_flags = flags;
105     xio->xio_bytes = 0;
106     xio->xio_error = 0;
107     if (ubytes == 0) {
108 	xio->xio_offset = 0;
109 	xio->xio_npages = 0;
110     } else {
111 	vmprot = (flags & XIOF_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
112 	xio->xio_offset = (vm_offset_t)ubase & PAGE_MASK;
113 	xio->xio_pages = xio->xio_internal_pages;
114 	if ((n = PAGE_SIZE - xio->xio_offset) > ubytes)
115 	    n = ubytes;
116 	for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
117 	    if (vm_fault_quick((caddr_t)addr, vmprot) < 0)
118 		break;
119 	    if ((paddr = pmap_kextract(addr)) == 0)
120 		break;
121 	    s = splvm();
122 	    m = PHYS_TO_VM_PAGE(paddr);
123 	    vm_page_hold(m);
124 	    splx(s);
125 	    xio->xio_pages[i] = m;
126 	    ubytes -= n;
127 	    xio->xio_bytes += n;
128 	    if ((n = ubytes) > PAGE_SIZE)
129 		n = PAGE_SIZE;
130 	    addr += PAGE_SIZE;
131 	}
132 	xio->xio_npages = i;
133 
134 	/*
135 	 * If a failure occured clean out what we loaded and return EFAULT.
136 	 * Return 0 on success.
137 	 */
138 	if (i < XIO_INTERNAL_PAGES && n) {
139 	    xio_release(xio);
140 	    xio->xio_error = EFAULT;
141 	}
142     }
143     return(xio->xio_error);
144 }
145 
146 /*
147  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
148  * an error code on failure.  The actual number of bytes that could be
149  * accomodated in the XIO will be stored in xio_bytes and the page offset
150  * will be stored in xio_offset.
151  */
152 int
153 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
154 {
155     vm_offset_t addr;
156     vm_paddr_t paddr;
157     vm_page_t m;
158     int i;
159     int n;
160     int s;
161 
162     addr = trunc_page((vm_offset_t)kbase);
163     xio->xio_flags = 0;
164     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
165     xio->xio_bytes = 0;
166     xio->xio_pages = xio->xio_internal_pages;
167     xio->xio_error = 0;
168     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
169 	n = kbytes;
170     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
171 	if ((paddr = pmap_kextract(addr)) == 0)
172 	    break;
173 	s = splvm();
174 	m = PHYS_TO_VM_PAGE(paddr);
175 	vm_page_hold(m);
176 	splx(s);
177 	xio->xio_pages[i] = m;
178 	kbytes -= n;
179 	xio->xio_bytes += n;
180 	if ((n = kbytes) > PAGE_SIZE)
181 	    n = PAGE_SIZE;
182 	addr += PAGE_SIZE;
183     }
184     xio->xio_npages = i;
185 
186     /*
187      * If a failure occured clean out what we loaded and return EFAULT.
188      * Return 0 on success.
189      */
190     if (i < XIO_INTERNAL_PAGES && n) {
191 	xio_release(xio);
192 	xio->xio_error = EFAULT;
193     }
194     return(xio->xio_error);
195 }
196 
197 /*
198  * Cleanup an XIO so it can be destroyed.  The pages associated with the
199  * XIO are released.
200  */
201 void
202 xio_release(xio_t xio)
203 {
204     int i;
205     int s;
206     vm_page_t m;
207 
208     s = splvm();
209     for (i = 0; i < xio->xio_npages; ++i) {
210 	m = xio->xio_pages[i];
211 	vm_page_unhold(m);
212     }
213     splx(s);
214     xio->xio_offset = 0;
215     xio->xio_npages = 0;
216     xio->xio_bytes = 0;
217     xio->xio_error = ENOBUFS;
218 }
219 
220 /*
221  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
222  * must be relative to the current context.
223  *
224  * uoffset is the abstracted starting offset in the XIO, not the actual
225  * offset, and usually starts at 0.
226  *
227  * The XIO is not modified.  The UIO is updated to reflect the copy.
228  *
229  * UIO_READ	xio -> uio
230  * UIO_WRITE	uio -> xio
231  */
232 int
233 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, int *sizep)
234 {
235     int error;
236     int bytes;
237 
238     bytes = xio->xio_bytes - uoffset;
239     if (bytes > uio->uio_resid)
240 	bytes = uio->uio_resid;
241     KKASSERT(bytes >= 0);
242     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
243 				bytes, uio);
244     if (error == 0)
245 	*sizep = bytes;
246     else
247 	*sizep = 0;
248     return(error);
249 }
250 
251 /*
252  * Copy the specified number of bytes from the xio to a userland
253  * buffer.  Return an error code or 0 on success.
254  *
255  * uoffset is the abstracted starting offset in the XIO, not the actual
256  * offset, and usually starts at 0.
257  *
258  * The XIO is not modified.
259  */
260 int
261 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
262 {
263     int i;
264     int n;
265     int error;
266     int offset;
267     vm_page_t m;
268     struct sf_buf *sf;
269 
270     if (bytes > xio->xio_bytes)
271 	return(EFAULT);
272 
273     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
274     if ((n = PAGE_SIZE - offset) > bytes)
275 	n = bytes;
276 
277     error = 0;
278     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
279 	 i < xio->xio_npages;
280 	 ++i
281     ) {
282 	m = xio->xio_pages[i];
283 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
284 	error = copyout((char *)sf_buf_kva(sf) + offset, uptr, n);
285 	sf_buf_free(sf);
286 	if (error)
287 	    break;
288 	bytes -= n;
289 	uptr = (char *)uptr + n;
290 	if (bytes == 0)
291 	    break;
292 	if ((n = bytes) > PAGE_SIZE)
293 	    n = PAGE_SIZE;
294 	offset = 0;
295     }
296     return(error);
297 }
298 
299 /*
300  * Copy the specified number of bytes from the xio to a kernel
301  * buffer.  Return an error code or 0 on success.
302  *
303  * uoffset is the abstracted starting offset in the XIO, not the actual
304  * offset, and usually starts at 0.
305  *
306  * The XIO is not modified.
307  */
308 int
309 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
310 {
311     int i;
312     int n;
313     int error;
314     int offset;
315     vm_page_t m;
316     struct sf_buf *sf;
317 
318     if (bytes + uoffset > xio->xio_bytes)
319 	return(EFAULT);
320 
321     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
322     if ((n = PAGE_SIZE - offset) > bytes)
323 	n = bytes;
324 
325     error = 0;
326     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
327 	 i < xio->xio_npages;
328 	 ++i
329     ) {
330 	m = xio->xio_pages[i];
331 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
332 	bcopy((char *)sf_buf_kva(sf) + offset, kptr, n);
333 	sf_buf_free(sf);
334 	bytes -= n;
335 	kptr = (char *)kptr + n;
336 	if (bytes == 0)
337 	    break;
338 	if ((n = bytes) > PAGE_SIZE)
339 	    n = PAGE_SIZE;
340 	offset = 0;
341     }
342     return(error);
343 }
344 
345