xref: /xnu-8020.140.41/bsd/kern/kern_symfile.c (revision 27b03b360a988dfd3dfdf34262bb0042026747cc)
1 /*
2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /* Copyright (c) 1998 Apple Computer, Inc.  All rights reserved.
29  *
30  *	File:	bsd/kern/kern_symfile.c
31  *
32  * HISTORY
33  */
34 
35 #include <mach/vm_param.h>
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/signalvar.h>
40 #include <sys/resourcevar.h>
41 #include <sys/namei.h>
42 #include <sys/vnode_internal.h>
43 #include <sys/proc_internal.h>
44 #include <sys/kauth.h>
45 #include <sys/timeb.h>
46 #include <sys/times.h>
47 #include <sys/acct.h>
48 #include <sys/file_internal.h>
49 #include <sys/uio.h>
50 #include <sys/kernel.h>
51 #include <sys/stat.h>
52 #include <sys/disk.h>
53 #include <sys/conf.h>
54 #include <sys/content_protection.h>
55 #include <sys/fsctl.h>
56 
57 #include <mach-o/loader.h>
58 #include <mach-o/nlist.h>
59 
60 #include <kern/kalloc.h>
61 #include <vm/vm_kern.h>
62 #include <pexpert/pexpert.h>
63 #include <IOKit/IOPolledInterface.h>
64 
65 #define HIBERNATE_MIN_PHYSICAL_LBA_512    (34)
66 #define HIBERNATE_MIN_PHYSICAL_LBA_4096   (6)
67 #define HIBERNATE_MIN_FILE_SIZE           (1024*1024)
68 
69 /* This function is called from kern_sysctl in the current process context;
70  * it is exported with the System6.0.exports, but this appears to be a legacy
71  * export, as there are no internal consumers.
72  */
73 int
74 get_kernel_symfile(__unused proc_t p, __unused char const **symfile);
75 int
get_kernel_symfile(__unused proc_t p,__unused char const ** symfile)76 get_kernel_symfile(__unused proc_t p, __unused char const **symfile)
77 {
78 	return KERN_FAILURE;
79 }
80 
81 struct kern_direct_file_io_ref_t {
82 	vfs_context_t      ctx;
83 	struct vnode      * vp;
84 	dev_t               device;
85 	uint32_t            blksize;
86 	off_t               filelength;
87 	char                cf;
88 	char                pinned;
89 	char                frozen;
90 	char                wbcranged;
91 };
92 
93 
94 static int
file_ioctl(void * p1,void * p2,u_long theIoctl,caddr_t result)95 file_ioctl(void * p1, void * p2, u_long theIoctl, caddr_t result)
96 {
97 	dev_t device = *(dev_t*) p1;
98 
99 	return (*bdevsw[major(device)].d_ioctl)
100 	       (device, theIoctl, result, S_IFBLK, p2);
101 }
102 
103 static int
device_ioctl(void * p1,__unused void * p2,u_long theIoctl,caddr_t result)104 device_ioctl(void * p1, __unused void * p2, u_long theIoctl, caddr_t result)
105 {
106 	return VNOP_IOCTL(p1, theIoctl, result, 0, p2);
107 }
108 
109 static int
kern_ioctl_file_extents(struct kern_direct_file_io_ref_t * ref,u_long theIoctl,off_t offset,off_t end)110 kern_ioctl_file_extents(struct kern_direct_file_io_ref_t * ref, u_long theIoctl, off_t offset, off_t end)
111 {
112 	int error = 0;
113 	int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
114 	void * p1;
115 	void * p2;
116 	uint64_t    fileblk = 0;
117 	size_t      filechunk = 0;
118 	dk_extent_t  extent;
119 	dk_unmap_t   unmap;
120 	_dk_cs_pin_t pin;
121 
122 	bzero(&extent, sizeof(dk_extent_t));
123 	bzero(&unmap, sizeof(dk_unmap_t));
124 	bzero(&pin, sizeof(pin));
125 	if (ref->vp->v_type == VREG) {
126 		p1 = &ref->device;
127 		p2 = kernproc;
128 		do_ioctl = &file_ioctl;
129 	} else {
130 		/* Partition. */
131 		p1 = ref->vp;
132 		p2 = ref->ctx;
133 		do_ioctl = &device_ioctl;
134 	}
135 
136 	if (_DKIOCCSPINEXTENT == theIoctl) {
137 		/* Tell CS the image size, so it knows whether to place the subsequent pins SSD/HDD */
138 		pin.cp_extent.length = end;
139 		pin.cp_flags = _DKIOCCSHIBERNATEIMGSIZE;
140 		(void) do_ioctl(p1, p2, _DKIOCCSPINEXTENT, (caddr_t)&pin);
141 	} else if (_DKIOCCSUNPINEXTENT == theIoctl) {
142 		/* Tell CS hibernation is done, so it can stop blocking overlapping writes */
143 		pin.cp_flags = _DKIOCCSPINDISCARDDENYLIST;
144 		(void) do_ioctl(p1, p2, _DKIOCCSUNPINEXTENT, (caddr_t)&pin);
145 	}
146 
147 	for (; offset < end; offset += filechunk) {
148 		if (ref->vp->v_type == VREG) {
149 			daddr64_t blkno;
150 			filechunk = 1 * 1024 * 1024 * 1024;
151 			if (filechunk > (size_t)(end - offset)) {
152 				filechunk = (size_t)(end - offset);
153 			}
154 			error = VNOP_BLOCKMAP(ref->vp, offset, filechunk, &blkno,
155 			    &filechunk, NULL, VNODE_WRITE | VNODE_BLOCKMAP_NO_TRACK, NULL);
156 			if (error) {
157 				break;
158 			}
159 			if (-1LL == blkno) {
160 				continue;
161 			}
162 			fileblk = blkno * ref->blksize;
163 		} else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) {
164 			fileblk = offset;
165 			filechunk = (unsigned long)((ref->filelength > ULONG_MAX) ? ULONG_MAX: ref->filelength);
166 		}
167 
168 		if (DKIOCUNMAP == theIoctl) {
169 			extent.offset = fileblk;
170 			extent.length = filechunk;
171 			unmap.extents = &extent;
172 			unmap.extentsCount = 1;
173 			error = do_ioctl(p1, p2, theIoctl, (caddr_t)&unmap);
174 //          printf("DKIOCUNMAP(%d) 0x%qx, 0x%qx\n", error, extent.offset, extent.length);
175 		} else if (_DKIOCCSPINEXTENT == theIoctl) {
176 			pin.cp_extent.offset = fileblk;
177 			pin.cp_extent.length = filechunk;
178 			pin.cp_flags = _DKIOCCSPINFORHIBERNATION;
179 			error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin);
180 			if (error && (ENOTTY != error)) {
181 				printf("_DKIOCCSPINEXTENT(%d) 0x%qx, 0x%qx\n", error, pin.cp_extent.offset, pin.cp_extent.length);
182 			}
183 		} else if (_DKIOCCSUNPINEXTENT == theIoctl) {
184 			pin.cp_extent.offset = fileblk;
185 			pin.cp_extent.length = filechunk;
186 			pin.cp_flags = _DKIOCCSPINFORHIBERNATION;
187 			error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin);
188 			if (error && (ENOTTY != error)) {
189 				printf("_DKIOCCSUNPINEXTENT(%d) 0x%qx, 0x%qx\n", error, pin.cp_extent.offset, pin.cp_extent.length);
190 			}
191 		} else {
192 			error = EINVAL;
193 		}
194 
195 		if (error) {
196 			break;
197 		}
198 	}
199 	return error;
200 }
201 
202 extern uint32_t freespace_mb(vnode_t vp);
203 
204 struct kern_direct_file_io_ref_t *
kern_open_file_for_direct_io(const char * name,uint32_t iflags,kern_get_file_extents_callback_t callback,void * callback_ref,off_t set_file_size,off_t fs_free_size,off_t write_file_offset,void * write_file_addr,size_t write_file_len,dev_t * partition_device_result,dev_t * image_device_result,uint64_t * partitionbase_result,uint64_t * maxiocount_result,uint32_t * oflags)205 kern_open_file_for_direct_io(const char * name,
206     uint32_t iflags,
207     kern_get_file_extents_callback_t callback,
208     void * callback_ref,
209     off_t set_file_size,
210     off_t fs_free_size,
211     off_t write_file_offset,
212     void * write_file_addr,
213     size_t write_file_len,
214     dev_t * partition_device_result,
215     dev_t * image_device_result,
216     uint64_t * partitionbase_result,
217     uint64_t * maxiocount_result,
218     uint32_t * oflags)
219 {
220 	struct kern_direct_file_io_ref_t * ref;
221 
222 	proc_t            p;
223 	struct vnode_attr va;
224 	dk_apfs_wbc_range_t wbc_range;
225 	int               error;
226 	off_t             f_offset;
227 	uint64_t          fileblk = 0;
228 	size_t            filechunk = 0;
229 	uint64_t          physoffset, minoffset;
230 	dev_t             device;
231 	dev_t             target = 0;
232 	int               isssd = 0;
233 	uint32_t          flags = 0;
234 	uint32_t          blksize;
235 	off_t             maxiocount, count, segcount, wbctotal;
236 	boolean_t         locked = FALSE;
237 	int               fmode;
238 	mode_t                    cmode;
239 	struct            nameidata nd;
240 	u_int32_t         ndflags;
241 	off_t             mpFree;
242 
243 	wbc_range.count = 0;
244 
245 	int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
246 	do_ioctl = NULL;
247 	void * p1 = NULL;
248 	void * p2 = NULL;
249 
250 	error = EFAULT;
251 
252 	ref = kalloc_type(struct kern_direct_file_io_ref_t,
253 	    Z_WAITOK | Z_ZERO | Z_NOFAIL);
254 
255 	p = kernproc;
256 	ref->ctx = vfs_context_kernel();
257 
258 	fmode  = (kIOPolledFileCreate & iflags) ? (O_CREAT | FWRITE) : FWRITE;
259 	cmode =  S_IRUSR | S_IWUSR;
260 	ndflags = NOFOLLOW;
261 	NDINIT(&nd, LOOKUP, OP_OPEN, ndflags, UIO_SYSSPACE, CAST_USER_ADDR_T(name), ref->ctx);
262 	VATTR_INIT(&va);
263 	VATTR_SET(&va, va_mode, cmode);
264 	VATTR_SET(&va, va_dataprotect_flags, VA_DP_RAWENCRYPTED);
265 	VATTR_SET(&va, va_dataprotect_class, PROTECTION_CLASS_D);
266 	if ((error = vn_open_auth(&nd, &fmode, &va))) {
267 		kprintf("vn_open_auth(fmode: %d, cmode: %d) failed with error: %d\n", fmode, cmode, error);
268 		goto out;
269 	}
270 
271 	ref->vp = nd.ni_vp;
272 	if (ref->vp->v_type == VREG) {
273 		vnode_lock_spin(ref->vp);
274 		SET(ref->vp->v_flag, VSWAP);
275 		vnode_unlock(ref->vp);
276 	}
277 
278 	if (write_file_addr && write_file_len) {
279 		if ((error = kern_write_file(ref, write_file_offset, write_file_addr, write_file_len, IO_SKIP_ENCRYPTION))) {
280 			kprintf("kern_write_file() failed with error: %d\n", error);
281 			goto out;
282 		}
283 	}
284 
285 	VATTR_INIT(&va);
286 	VATTR_WANTED(&va, va_rdev);
287 	VATTR_WANTED(&va, va_fsid);
288 	VATTR_WANTED(&va, va_devid);
289 	VATTR_WANTED(&va, va_data_size);
290 	VATTR_WANTED(&va, va_data_alloc);
291 	VATTR_WANTED(&va, va_nlink);
292 	error = EFAULT;
293 	if (vnode_getattr(ref->vp, &va, ref->ctx)) {
294 		goto out;
295 	}
296 
297 	wbctotal = 0;
298 	mpFree = freespace_mb(ref->vp);
299 	mpFree <<= 20;
300 	kprintf("kern_direct_file(%s): vp size %qd, alloc %qd, mp free %qd, keep free %qd\n",
301 	    name, va.va_data_size, va.va_data_alloc, mpFree, fs_free_size);
302 
303 	if (ref->vp->v_type == VREG) {
304 		/* Don't dump files with links. */
305 		if (va.va_nlink != 1) {
306 			goto out;
307 		}
308 
309 		device = (VATTR_IS_SUPPORTED(&va, va_devid)) ? va.va_devid : va.va_fsid;
310 		ref->filelength = va.va_data_size;
311 
312 		p1 = &device;
313 		p2 = p;
314 		do_ioctl = &file_ioctl;
315 
316 		if (kIOPolledFileHibernate & iflags) {
317 			error = do_ioctl(p1, p2, DKIOCAPFSGETWBCRANGE, (caddr_t) &wbc_range);
318 			ref->wbcranged = (error == 0);
319 		}
320 		if (ref->wbcranged) {
321 			uint32_t idx;
322 			assert(wbc_range.count <= (sizeof(wbc_range.extents) / sizeof(wbc_range.extents[0])));
323 			for (idx = 0; idx < wbc_range.count; idx++) {
324 				wbctotal += wbc_range.extents[idx].length;
325 			}
326 			kprintf("kern_direct_file(%s): wbc %qd\n", name, wbctotal);
327 			if (wbctotal) {
328 				target = wbc_range.dev;
329 			}
330 		}
331 
332 		if (set_file_size) {
333 			if (wbctotal) {
334 				if (wbctotal >= set_file_size) {
335 					set_file_size = HIBERNATE_MIN_FILE_SIZE;
336 				} else {
337 					set_file_size -= wbctotal;
338 					if (set_file_size < HIBERNATE_MIN_FILE_SIZE) {
339 						set_file_size = HIBERNATE_MIN_FILE_SIZE;
340 					}
341 				}
342 			}
343 			if (fs_free_size) {
344 				mpFree += va.va_data_alloc;
345 				if ((mpFree < set_file_size) || ((mpFree - set_file_size) < fs_free_size)) {
346 					error = ENOSPC;
347 					goto out;
348 				}
349 			}
350 			error = vnode_setsize(ref->vp, set_file_size, IO_NOZEROFILL | IO_NOAUTH, ref->ctx);
351 			if (error) {
352 				goto out;
353 			}
354 			ref->filelength = set_file_size;
355 		}
356 	} else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) {
357 		/* Partition. */
358 		device = va.va_rdev;
359 
360 		p1 = ref->vp;
361 		p2 = ref->ctx;
362 		do_ioctl = &device_ioctl;
363 	} else {
364 		/* Don't dump to non-regular files. */
365 		error = EFAULT;
366 		goto out;
367 	}
368 	ref->device = device;
369 
370 	// probe for CF
371 	dk_corestorage_info_t cs_info;
372 	memset(&cs_info, 0, sizeof(dk_corestorage_info_t));
373 	error = do_ioctl(p1, p2, DKIOCCORESTORAGE, (caddr_t)&cs_info);
374 	ref->cf = (error == 0) && (cs_info.flags & DK_CORESTORAGE_ENABLE_HOTFILES);
375 
376 	// get block size
377 
378 	error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize);
379 	if (error) {
380 		goto out;
381 	}
382 
383 	if (ref->blksize == 4096) {
384 		minoffset = HIBERNATE_MIN_PHYSICAL_LBA_4096 * ref->blksize;
385 	} else {
386 		minoffset = HIBERNATE_MIN_PHYSICAL_LBA_512 * ref->blksize;
387 	}
388 
389 	if (ref->vp->v_type != VREG) {
390 		error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk);
391 		if (error) {
392 			goto out;
393 		}
394 		ref->filelength = fileblk * ref->blksize;
395 	}
396 
397 	// pin logical extents, CS version
398 
399 	error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength);
400 	if (error && (ENOTTY != error)) {
401 		goto out;
402 	}
403 	ref->pinned = (error == 0);
404 
405 	// pin logical extents, apfs version
406 
407 	error = VNOP_IOCTL(ref->vp, FSCTL_FREEZE_EXTENTS, NULL, 0, ref->ctx);
408 	if (error && (ENOTTY != error)) {
409 		goto out;
410 	}
411 	ref->frozen = (error == 0);
412 
413 	// generate the block list
414 
415 	error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL);
416 	if (error) {
417 		goto out;
418 	}
419 	locked = TRUE;
420 
421 	f_offset = 0;
422 	for (; f_offset < ref->filelength; f_offset += filechunk) {
423 		if (ref->vp->v_type == VREG) {
424 			filechunk = 1 * 1024 * 1024 * 1024;
425 			daddr64_t blkno;
426 
427 			error = VNOP_BLOCKMAP(ref->vp, f_offset, filechunk, &blkno,
428 			    &filechunk, NULL, VNODE_WRITE | VNODE_BLOCKMAP_NO_TRACK, NULL);
429 			if (error) {
430 				goto out;
431 			}
432 			if (-1LL == blkno) {
433 				continue;
434 			}
435 			fileblk = blkno * ref->blksize;
436 		} else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) {
437 			fileblk = f_offset;
438 			filechunk = f_offset ? 0 : (unsigned long)ref->filelength;
439 		}
440 
441 		physoffset = 0;
442 		while (physoffset < filechunk) {
443 			dk_physical_extent_t getphysreq;
444 			bzero(&getphysreq, sizeof(getphysreq));
445 
446 			getphysreq.offset = fileblk + physoffset;
447 			getphysreq.length = (filechunk - physoffset);
448 			error = do_ioctl(p1, p2, DKIOCGETPHYSICALEXTENT, (caddr_t) &getphysreq);
449 			if (error) {
450 				goto out;
451 			}
452 			if (!target) {
453 				target = getphysreq.dev;
454 			} else if (target != getphysreq.dev) {
455 				error = ENOTSUP;
456 				goto out;
457 			}
458 
459 			assert(getphysreq.offset >= minoffset);
460 
461 #if HIBFRAGMENT
462 			uint64_t rev;
463 			for (rev = 4096; rev <= getphysreq.length; rev += 4096) {
464 				callback(callback_ref, getphysreq.offset + getphysreq.length - rev, 4096);
465 			}
466 #else
467 			callback(callback_ref, getphysreq.offset, getphysreq.length);
468 #endif
469 			physoffset += getphysreq.length;
470 		}
471 	}
472 	if (ref->wbcranged) {
473 		uint32_t idx;
474 		for (idx = 0; idx < wbc_range.count; idx++) {
475 			assert(wbc_range.extents[idx].offset >= minoffset);
476 			callback(callback_ref, wbc_range.extents[idx].offset, wbc_range.extents[idx].length);
477 		}
478 	}
479 	callback(callback_ref, 0ULL, 0ULL);
480 
481 	if (ref->vp->v_type == VREG) {
482 		p1 = &target;
483 	} else {
484 		p1 = &target;
485 		p2 = p;
486 		do_ioctl = &file_ioctl;
487 	}
488 
489 	// get partition base
490 
491 	if (partitionbase_result) {
492 		error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result);
493 		if (error) {
494 			goto out;
495 		}
496 	}
497 
498 	// get block size & constraints
499 
500 	error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &blksize);
501 	if (error) {
502 		goto out;
503 	}
504 
505 	maxiocount = 1 * 1024 * 1024 * 1024;
506 
507 	error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTREAD, (caddr_t) &count);
508 	if (error) {
509 		count = 0;
510 	}
511 	count *= blksize;
512 	if (count && (count < maxiocount)) {
513 		maxiocount = count;
514 	}
515 
516 	error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTWRITE, (caddr_t) &count);
517 	if (error) {
518 		count = 0;
519 	}
520 	count *= blksize;
521 	if (count && (count < maxiocount)) {
522 		maxiocount = count;
523 	}
524 
525 	error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTREAD, (caddr_t) &count);
526 	if (error) {
527 		count = 0;
528 	}
529 	if (count && (count < maxiocount)) {
530 		maxiocount = count;
531 	}
532 
533 	error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTWRITE, (caddr_t) &count);
534 	if (error) {
535 		count = 0;
536 	}
537 	if (count && (count < maxiocount)) {
538 		maxiocount = count;
539 	}
540 
541 	error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTREAD, (caddr_t) &count);
542 	if (!error) {
543 		error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTREAD, (caddr_t) &segcount);
544 	}
545 	if (error) {
546 		count = segcount = 0;
547 	}
548 	count *= segcount;
549 	if (count && (count < maxiocount)) {
550 		maxiocount = count;
551 	}
552 
553 	error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count);
554 	if (!error) {
555 		error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTWRITE, (caddr_t) &segcount);
556 	}
557 	if (error) {
558 		count = segcount = 0;
559 	}
560 	count *= segcount;
561 	if (count && (count < maxiocount)) {
562 		maxiocount = count;
563 	}
564 
565 	kprintf("max io 0x%qx bytes\n", maxiocount);
566 	if (maxiocount_result) {
567 		*maxiocount_result = maxiocount;
568 	}
569 
570 	error = do_ioctl(p1, p2, DKIOCISSOLIDSTATE, (caddr_t)&isssd);
571 	if (!error && isssd) {
572 		flags |= kIOPolledFileSSD;
573 	}
574 
575 	if (partition_device_result) {
576 		*partition_device_result = device;
577 	}
578 	if (image_device_result) {
579 		*image_device_result = target;
580 	}
581 	if (oflags) {
582 		*oflags = flags;
583 	}
584 
585 	if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) {
586 		vnode_close(ref->vp, FWRITE, ref->ctx);
587 		ref->vp = NULLVP;
588 		ref->ctx = NULL;
589 	}
590 
591 out:
592 	printf("kern_open_file_for_direct_io(%p, %d)\n", ref, error);
593 
594 
595 	if (error && locked) {
596 		p1 = &device;
597 		if (do_ioctl) {
598 			(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
599 		}
600 	}
601 
602 	if (error && ref) {
603 		if (ref->vp) {
604 			(void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (ref->pinned && ref->cf) ? ref->filelength : 0);
605 
606 			if (ref->frozen) {
607 				(void) VNOP_IOCTL(ref->vp, FSCTL_THAW_EXTENTS, NULL, 0, ref->ctx);
608 			}
609 			if (ref->wbcranged) {
610 				if (do_ioctl) {
611 					(void) do_ioctl(p1, p2, DKIOCAPFSRELEASEWBCRANGE, (caddr_t) NULL);
612 				}
613 			}
614 			vnode_close(ref->vp, FWRITE, ref->ctx);
615 			ref->vp = NULLVP;
616 		}
617 		ref->ctx = NULL;
618 		kfree_type(struct kern_direct_file_io_ref_t, ref);
619 		ref = NULL;
620 	}
621 
622 	return ref;
623 }
624 
625 int
kern_write_file(struct kern_direct_file_io_ref_t * ref,off_t offset,void * addr,size_t len,int ioflag)626 kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, void * addr, size_t len, int ioflag)
627 {
628 	assert(len <= INT32_MAX);
629 	return vn_rdwr(UIO_WRITE, ref->vp,
630 	           addr, (int)len, offset,
631 	           UIO_SYSSPACE, ioflag | IO_SYNC | IO_NODELOCKED | IO_UNIT,
632 	           vfs_context_ucred(ref->ctx), (int *) 0,
633 	           vfs_context_proc(ref->ctx));
634 }
635 
636 int
kern_read_file(struct kern_direct_file_io_ref_t * ref,off_t offset,void * addr,size_t len,int ioflag)637 kern_read_file(struct kern_direct_file_io_ref_t * ref, off_t offset, void * addr, size_t len, int ioflag)
638 {
639 	assert(len <= INT32_MAX);
640 	return vn_rdwr(UIO_READ, ref->vp,
641 	           addr, (int)len, offset,
642 	           UIO_SYSSPACE, ioflag | IO_SYNC | IO_NODELOCKED | IO_UNIT,
643 	           vfs_context_ucred(ref->ctx), (int *) 0,
644 	           vfs_context_proc(ref->ctx));
645 }
646 
647 
648 struct mount *
kern_file_mount(struct kern_direct_file_io_ref_t * ref)649 kern_file_mount(struct kern_direct_file_io_ref_t * ref)
650 {
651 	return ref->vp->v_mount;
652 }
653 
654 void
kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,off_t write_offset,void * addr,size_t write_length,off_t discard_offset,off_t discard_end)655 kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,
656     off_t write_offset, void * addr, size_t write_length,
657     off_t discard_offset, off_t discard_end)
658 {
659 	int error;
660 	printf("kern_close_file_for_direct_io(%p)\n", ref);
661 
662 	if (!ref) {
663 		return;
664 	}
665 
666 	if (ref->vp) {
667 		int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
668 		void * p1;
669 		void * p2;
670 
671 		discard_offset = ((discard_offset + ref->blksize - 1) & ~(((off_t) ref->blksize) - 1));
672 		discard_end    = ((discard_end)                       & ~(((off_t) ref->blksize) - 1));
673 
674 		if (ref->vp->v_type == VREG) {
675 			p1 = &ref->device;
676 			p2 = kernproc;
677 			do_ioctl = &file_ioctl;
678 		} else {
679 			/* Partition. */
680 			p1 = ref->vp;
681 			p2 = ref->ctx;
682 			do_ioctl = &device_ioctl;
683 		}
684 		(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
685 
686 		//XXX If unmapping extents then don't also need to unpin; except ...
687 		//XXX if file unaligned (HFS 4k / Fusion 128k) then pin is superset and
688 		//XXX unmap is subset, so save extra walk over file extents (and the risk
689 		//XXX that CF drain starts) vs leaving partial units pinned to SSD
690 		//XXX (until whatever was sharing also unmaps).  Err on cleaning up fully.
691 		boolean_t will_unmap = (!ref->pinned || ref->cf) && (discard_end > discard_offset);
692 		boolean_t will_unpin = (ref->pinned && ref->cf /* && !will_unmap */);
693 
694 		(void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (will_unpin) ? ref->filelength : 0);
695 
696 		if (will_unmap) {
697 			(void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, (ref->cf) ? ref->filelength : discard_end);
698 		}
699 
700 		if (ref->frozen) {
701 			(void) VNOP_IOCTL(ref->vp, FSCTL_THAW_EXTENTS, NULL, 0, ref->ctx);
702 		}
703 		if (ref->wbcranged) {
704 			(void) do_ioctl(p1, p2, DKIOCAPFSRELEASEWBCRANGE, (caddr_t) NULL);
705 		}
706 
707 		if (addr && write_length) {
708 			(void) kern_write_file(ref, write_offset, addr, write_length, IO_SKIP_ENCRYPTION);
709 		}
710 
711 		error = vnode_close(ref->vp, FWRITE, ref->ctx);
712 
713 		ref->vp = NULLVP;
714 		kprintf("vnode_close(%d)\n", error);
715 
716 	}
717 
718 	ref->ctx = NULL;
719 
720 	kfree_type(struct kern_direct_file_io_ref_t, ref);
721 }
722