xref: /dflybsd-src/sys/kern/kern_xio.c (revision e7b4468ce80913950cd099c393f3ce6ece6fcb2c)
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 1.15 2007/08/13 17:20:04 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 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_page_t m;
97     vm_page_t m0;
98     int error;
99     int i;
100     int n;
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 	m0 = NULL;
117 	for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
118 	    m = vm_fault_page_quick(addr, vmprot, &error);
119 	    if (m == NULL)
120 		break;
121 	    xio->xio_pages[i] = m;
122 	    ubytes -= n;
123 	    xio->xio_bytes += n;
124 	    if ((n = ubytes) > PAGE_SIZE)
125 		n = PAGE_SIZE;
126 	    addr += PAGE_SIZE;
127 
128 	    /*
129 	     * Check linearity, used by syslink to memory map DMA buffers.
130 	     */
131 	    if (flags & XIOF_VMLINEAR) {
132 		if (i == 0) {
133 		    m0 = m;
134 		} else
135 		if (m->object != m0->object || m->pindex != m0->pindex + i) {
136 		    error = EINVAL;
137 		    break;
138 		}
139 	    }
140 	}
141 	xio->xio_npages = i;
142 
143 	/*
144 	 * If a failure occured clean out what we loaded and return EFAULT.
145 	 * Return 0 on success.
146 	 */
147 	if (i < XIO_INTERNAL_PAGES && n) {
148 	    xio_release(xio);
149 	    xio->xio_error = EFAULT;
150 	}
151     }
152     return(xio->xio_error);
153 }
154 
155 /*
156  * Initialize an XIO given a kernelspace buffer.  0 is returned on success,
157  * an error code on failure.  The actual number of bytes that could be
158  * accomodated in the XIO will be stored in xio_bytes and the page offset
159  * will be stored in xio_offset.
160  */
161 int
162 xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
163 {
164     vm_offset_t addr;
165     vm_paddr_t paddr;
166     vm_page_t m;
167     int i;
168     int n;
169 
170     addr = trunc_page((vm_offset_t)kbase);
171     xio->xio_flags = 0;
172     xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
173     xio->xio_bytes = 0;
174     xio->xio_pages = xio->xio_internal_pages;
175     xio->xio_error = 0;
176     if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
177 	n = kbytes;
178     for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
179 	if ((paddr = pmap_kextract(addr)) == 0)
180 	    break;
181 	crit_enter();
182 	m = PHYS_TO_VM_PAGE(paddr);
183 	vm_page_hold(m);
184 	crit_exit();
185 	xio->xio_pages[i] = m;
186 	kbytes -= n;
187 	xio->xio_bytes += n;
188 	if ((n = kbytes) > PAGE_SIZE)
189 	    n = PAGE_SIZE;
190 	addr += PAGE_SIZE;
191     }
192     xio->xio_npages = i;
193 
194     /*
195      * If a failure occured clean out what we loaded and return EFAULT.
196      * Return 0 on success.
197      */
198     if (i < XIO_INTERNAL_PAGES && n) {
199 	xio_release(xio);
200 	xio->xio_error = EFAULT;
201     }
202     return(xio->xio_error);
203 }
204 
205 /*
206  * Initialize an XIO given an array of vm_page pointers.
207  */
208 int
209 xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
210 {
211     int i;
212 
213     KKASSERT(npages <= XIO_INTERNAL_PAGES);
214 
215     xio->xio_flags = xflags;
216     xio->xio_offset = 0;
217     xio->xio_bytes = 0;
218     xio->xio_pages = xio->xio_internal_pages;
219     xio->xio_npages = npages;
220     xio->xio_error = 0;
221     crit_enter();
222     for (i = 0; i < npages; ++i) {
223 	vm_page_hold(mbase[i]);
224 	xio->xio_pages[i] = mbase[i];
225     }
226     crit_exit();
227     return(0);
228 }
229 
230 /*
231  * Cleanup an XIO so it can be destroyed.  The pages associated with the
232  * XIO are released.
233  */
234 void
235 xio_release(xio_t xio)
236 {
237     int i;
238     vm_page_t m;
239 
240     crit_enter();
241     for (i = 0; i < xio->xio_npages; ++i) {
242 	m = xio->xio_pages[i];
243 	vm_page_unhold(m);
244     }
245     crit_exit();
246     xio->xio_offset = 0;
247     xio->xio_npages = 0;
248     xio->xio_bytes = 0;
249     xio->xio_error = ENOBUFS;
250 }
251 
252 /*
253  * Copy data between an XIO and a UIO.  If the UIO represents userspace it
254  * must be relative to the current context.
255  *
256  * uoffset is the abstracted starting offset in the XIO, not the actual
257  * offset, and usually starts at 0.
258  *
259  * The XIO is not modified.  The UIO is updated to reflect the copy.
260  *
261  * UIO_READ	xio -> uio
262  * UIO_WRITE	uio -> xio
263  */
264 int
265 xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, int *sizep)
266 {
267     int error;
268     int bytes;
269 
270     bytes = xio->xio_bytes - uoffset;
271     if (bytes > uio->uio_resid)
272 	bytes = uio->uio_resid;
273     KKASSERT(bytes >= 0);
274     error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
275 				bytes, uio);
276     if (error == 0)
277 	*sizep = bytes;
278     else
279 	*sizep = 0;
280     return(error);
281 }
282 
283 /*
284  * Copy the specified number of bytes from the xio to a userland
285  * buffer.  Return an error code or 0 on success.
286  *
287  * uoffset is the abstracted starting offset in the XIO, not the actual
288  * offset, and usually starts at 0.
289  *
290  * The XIO is not modified.
291  */
292 int
293 xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
294 {
295     int i;
296     int n;
297     int error;
298     int offset;
299     vm_page_t m;
300     struct sf_buf *sf;
301 
302     if (uoffset + bytes > xio->xio_bytes)
303 	return(EFAULT);
304 
305     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
306     if ((n = PAGE_SIZE - offset) > bytes)
307 	n = bytes;
308 
309     error = 0;
310     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
311 	 i < xio->xio_npages;
312 	 ++i
313     ) {
314 	m = xio->xio_pages[i];
315 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
316 	error = copyout((char *)sf_buf_kva(sf) + offset, uptr, n);
317 	sf_buf_free(sf);
318 	if (error)
319 	    break;
320 	bytes -= n;
321 	uptr = (char *)uptr + n;
322 	if (bytes == 0)
323 	    break;
324 	if ((n = bytes) > PAGE_SIZE)
325 	    n = PAGE_SIZE;
326 	offset = 0;
327     }
328     return(error);
329 }
330 
331 /*
332  * Copy the specified number of bytes from the xio to a kernel
333  * buffer.  Return an error code or 0 on success.
334  *
335  * uoffset is the abstracted starting offset in the XIO, not the actual
336  * offset, and usually starts at 0.
337  *
338  * The XIO is not modified.
339  */
340 int
341 xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
342 {
343     int i;
344     int n;
345     int error;
346     int offset;
347     vm_page_t m;
348     struct sf_buf *sf;
349 
350     if (bytes + uoffset > xio->xio_bytes)
351 	return(EFAULT);
352 
353     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
354     if ((n = PAGE_SIZE - offset) > bytes)
355 	n = bytes;
356 
357     error = 0;
358     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
359 	 i < xio->xio_npages;
360 	 ++i
361     ) {
362 	m = xio->xio_pages[i];
363 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
364 	bcopy((char *)sf_buf_kva(sf) + offset, kptr, n);
365 	sf_buf_free(sf);
366 	bytes -= n;
367 	kptr = (char *)kptr + n;
368 	if (bytes == 0)
369 	    break;
370 	if ((n = bytes) > PAGE_SIZE)
371 	    n = PAGE_SIZE;
372 	offset = 0;
373     }
374     return(error);
375 }
376 
377 /*
378  * Copy the specified number of bytes from userland to the xio.
379  * Return an error code or 0 on success.
380  *
381  * uoffset is the abstracted starting offset in the XIO, not the actual
382  * offset, and usually starts at 0.
383  *
384  * Data in pages backing the XIO will be modified.
385  */
386 int
387 xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
388 {
389     int i;
390     int n;
391     int error;
392     int offset;
393     vm_page_t m;
394     struct sf_buf *sf;
395 
396     if (uoffset + bytes > xio->xio_bytes)
397 	return(EFAULT);
398 
399     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
400     if ((n = PAGE_SIZE - offset) > bytes)
401 	n = bytes;
402 
403     error = 0;
404     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
405 	 i < xio->xio_npages;
406 	 ++i
407     ) {
408 	m = xio->xio_pages[i];
409 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
410 	error = copyin(uptr, (char *)sf_buf_kva(sf) + offset, n);
411 	sf_buf_free(sf);
412 	if (error)
413 	    break;
414 	bytes -= n;
415 	uptr = (const char *)uptr + n;
416 	if (bytes == 0)
417 	    break;
418 	if ((n = bytes) > PAGE_SIZE)
419 	    n = PAGE_SIZE;
420 	offset = 0;
421     }
422     return(error);
423 }
424 
425 /*
426  * Copy the specified number of bytes from the kernel to the xio.
427  * Return an error code or 0 on success.
428  *
429  * uoffset is the abstracted starting offset in the XIO, not the actual
430  * offset, and usually starts at 0.
431  *
432  * Data in pages backing the XIO will be modified.
433  */
434 int
435 xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
436 {
437     int i;
438     int n;
439     int error;
440     int offset;
441     vm_page_t m;
442     struct sf_buf *sf;
443 
444     if (uoffset + bytes > xio->xio_bytes)
445 	return(EFAULT);
446 
447     offset = (xio->xio_offset + uoffset) & PAGE_MASK;
448     if ((n = PAGE_SIZE - offset) > bytes)
449 	n = bytes;
450 
451     error = 0;
452     for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
453 	 i < xio->xio_npages;
454 	 ++i
455     ) {
456 	m = xio->xio_pages[i];
457 	sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
458 	bcopy(kptr, (char *)sf_buf_kva(sf) + offset, n);
459 	sf_buf_free(sf);
460 	bytes -= n;
461 	kptr = (const char *)kptr + n;
462 	if (bytes == 0)
463 	    break;
464 	if ((n = bytes) > PAGE_SIZE)
465 	    n = PAGE_SIZE;
466 	offset = 0;
467     }
468     return(error);
469 }
470