xref: /xnu-8792.61.2/bsd/vm/vnode_pager.c (revision 42e220869062b56f8d7d0726fd4c88954f87902c)
1 /*
2  * Copyright (c) 2000-2020 Apple 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 /*
29  * Mach Operating System
30  * Copyright (c) 1987 Carnegie-Mellon University
31  * All rights reserved.  The CMU software License Agreement specifies
32  * the terms and conditions for use and redistribution.
33  */
34 /*
35  *	File:	vnode_pager.c
36  *
37  *	"Swap" pager that pages to/from vnodes.  Also
38  *	handles demand paging from files.
39  *
40  */
41 
42 #include <mach/boolean.h>
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/user.h>
46 #include <sys/proc.h>
47 #include <sys/kauth.h>
48 #include <sys/buf.h>
49 #include <sys/uio.h>
50 #include <sys/vnode_internal.h>
51 #include <sys/namei.h>
52 #include <sys/mount_internal.h> /* needs internal due to fhandle_t */
53 #include <sys/ubc_internal.h>
54 #include <sys/lock.h>
55 #include <sys/disk.h>           /* For DKIOC calls */
56 
57 #include <mach/mach_types.h>
58 #include <mach/memory_object_types.h>
59 #include <mach/vm_map.h>
60 #include <mach/mach_vm.h>
61 #include <mach/upl.h>
62 #include <mach/sdt.h>
63 
64 #include <vm/vm_map.h>
65 #include <vm/vm_kern.h>
66 #include <kern/zalloc.h>
67 #include <libkern/libkern.h>
68 
69 #include <vm/vnode_pager.h>
70 #include <vm/vm_pageout.h>
71 
72 #include <kern/assert.h>
73 #include <sys/kdebug.h>
74 #include <nfs/nfs.h>
75 
76 #include <vm/vm_protos.h>
77 
78 #include <sys/kdebug_triage.h>
79 #include <vfs/vfs_disk_conditioner.h>
80 
81 void
vnode_pager_throttle(void)82 vnode_pager_throttle(void)
83 {
84 	if (current_uthread()->uu_lowpri_window) {
85 		throttle_lowpri_io(1);
86 	}
87 }
88 
89 boolean_t
vnode_pager_isSSD(vnode_t vp)90 vnode_pager_isSSD(vnode_t vp)
91 {
92 	return disk_conditioner_mount_is_ssd(vp->v_mount);
93 }
94 
95 #if CONFIG_IOSCHED
96 void
vnode_pager_issue_reprioritize_io(struct vnode * devvp,uint64_t blkno,uint32_t len,int priority)97 vnode_pager_issue_reprioritize_io(struct vnode *devvp, uint64_t blkno, uint32_t len, int priority)
98 {
99 	u_int32_t blocksize = 0;
100 	dk_extent_t extent;
101 	dk_set_tier_t set_tier;
102 	int error = 0;
103 
104 	error = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel());
105 	if (error) {
106 		return;
107 	}
108 
109 	memset(&extent, 0, sizeof(dk_extent_t));
110 	memset(&set_tier, 0, sizeof(dk_set_tier_t));
111 
112 	extent.offset = blkno * (u_int64_t) blocksize;
113 	extent.length = len;
114 
115 	set_tier.extents = &extent;
116 	set_tier.extentsCount = 1;
117 	set_tier.tier = (uint8_t)priority;
118 
119 	error = VNOP_IOCTL(devvp, DKIOCSETTIER, (caddr_t)&set_tier, 0, vfs_context_kernel());
120 	return;
121 }
122 #endif
123 
124 void
vnode_pager_was_dirtied(struct vnode * vp,vm_object_offset_t s_offset,vm_object_offset_t e_offset)125 vnode_pager_was_dirtied(
126 	struct vnode            *vp,
127 	vm_object_offset_t      s_offset,
128 	vm_object_offset_t      e_offset)
129 {
130 	cluster_update_state(vp, s_offset, e_offset, TRUE);
131 }
132 
133 uint32_t
vnode_pager_isinuse(struct vnode * vp)134 vnode_pager_isinuse(struct vnode *vp)
135 {
136 	if (vp->v_usecount > vp->v_kusecount) {
137 		return 1;
138 	}
139 	return 0;
140 }
141 
142 uint32_t
vnode_pager_return_throttle_io_limit(struct vnode * vp,uint32_t * limit)143 vnode_pager_return_throttle_io_limit(struct vnode *vp, uint32_t *limit)
144 {
145 	return cluster_throttle_io_limit(vp, limit);
146 }
147 
148 vm_object_offset_t
vnode_pager_get_filesize(struct vnode * vp)149 vnode_pager_get_filesize(struct vnode *vp)
150 {
151 	return (vm_object_offset_t) ubc_getsize(vp);
152 }
153 
154 extern int safe_getpath(struct vnode *dvp, char *leafname, char *path, int _len, int *truncated_path);
155 
156 kern_return_t
vnode_pager_get_name(struct vnode * vp,char * pathname,vm_size_t pathname_len,char * filename,vm_size_t filename_len,boolean_t * truncated_path_p)157 vnode_pager_get_name(
158 	struct vnode    *vp,
159 	char            *pathname,
160 	vm_size_t       pathname_len,
161 	char            *filename,
162 	vm_size_t       filename_len,
163 	boolean_t       *truncated_path_p)
164 {
165 	*truncated_path_p = FALSE;
166 	if (pathname != NULL) {
167 		/* get the path name */
168 		safe_getpath(vp, NULL,
169 		    pathname, (int) pathname_len,
170 		    truncated_path_p);
171 	}
172 	if ((pathname == NULL || *truncated_path_p) &&
173 	    filename != NULL) {
174 		/* get the file name */
175 		const char *name;
176 
177 		name = vnode_getname_printable(vp);
178 		strlcpy(filename, name, (size_t) filename_len);
179 		vnode_putname_printable(name);
180 	}
181 	return KERN_SUCCESS;
182 }
183 
184 kern_return_t
vnode_pager_get_mtime(struct vnode * vp,struct timespec * current_mtime,struct timespec * cs_mtime)185 vnode_pager_get_mtime(
186 	struct vnode    *vp,
187 	struct timespec *current_mtime,
188 	struct timespec *cs_mtime)
189 {
190 	vnode_mtime(vp, current_mtime, vfs_context_current());
191 	if (cs_mtime != NULL) {
192 		ubc_get_cs_mtime(vp, cs_mtime);
193 	}
194 	return KERN_SUCCESS;
195 }
196 
197 kern_return_t
vnode_pager_get_cs_blobs(struct vnode * vp,void ** blobs)198 vnode_pager_get_cs_blobs(
199 	struct vnode    *vp,
200 	void            **blobs)
201 {
202 	*blobs = ubc_get_cs_blobs(vp);
203 	return KERN_SUCCESS;
204 }
205 
206 /*
207  * vnode_trim:
208  * Used to call the DKIOCUNMAP ioctl on the underlying disk device for the specified vnode.
209  * Trims the region at offset bytes into the file, for length bytes.
210  *
211  * Care must be taken to ensure that the vnode is sufficiently reference counted at the time this
212  * function is called; no iocounts or usecounts are taken on the vnode.
213  * This function is non-idempotent in error cases;  We cannot un-discard the blocks if only some of them
214  * are successfully discarded.
215  */
216 u_int32_t
vnode_trim(struct vnode * vp,off_t offset,size_t length)217 vnode_trim(
218 	struct vnode *vp,
219 	off_t offset,
220 	size_t length)
221 {
222 	daddr64_t io_blockno;    /* Block number corresponding to the start of the extent */
223 	size_t io_bytecount;    /* Number of bytes in current extent for the specified range */
224 	size_t trimmed = 0;
225 	off_t current_offset = offset;
226 	size_t remaining_length = length;
227 	int error = 0;
228 	u_int32_t blocksize = 0;
229 	struct vnode *devvp;
230 	dk_extent_t extent;
231 	dk_unmap_t unmap;
232 
233 
234 	/* Get the underlying device vnode */
235 	devvp = vp->v_mount->mnt_devvp;
236 
237 	/* Figure out the underlying device block size */
238 	error  = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel());
239 	if (error) {
240 		goto trim_exit;
241 	}
242 
243 	/*
244 	 * We may not get the entire range from offset -> offset+length in a single
245 	 * extent from the blockmap call.  Keep looping/going until we are sure we've hit
246 	 * the whole range or if we encounter an error.
247 	 */
248 	while (trimmed < length) {
249 		/*
250 		 * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
251 		 * specified offset.  It returns blocks in contiguous chunks, so if the logical range is
252 		 * broken into multiple extents, it must be called multiple times, increasing the offset
253 		 * in each call to ensure that the entire range is covered.
254 		 */
255 		error = VNOP_BLOCKMAP(vp, current_offset, remaining_length,
256 		    &io_blockno, &io_bytecount, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL);
257 
258 		if (error) {
259 			goto trim_exit;
260 		}
261 		/*
262 		 * We have a contiguous run.  Prepare & issue the ioctl for the device.
263 		 * the DKIOCUNMAP ioctl takes offset in bytes from the start of the device.
264 		 */
265 		memset(&extent, 0, sizeof(dk_extent_t));
266 		memset(&unmap, 0, sizeof(dk_unmap_t));
267 		extent.offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
268 		extent.length = io_bytecount;
269 		unmap.extents = &extent;
270 		unmap.extentsCount = 1;
271 		error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
272 
273 		if (error) {
274 			goto trim_exit;
275 		}
276 		remaining_length = remaining_length - io_bytecount;
277 		trimmed = trimmed + io_bytecount;
278 		current_offset = current_offset + io_bytecount;
279 	}
280 trim_exit:
281 
282 	return error;
283 }
284 
285 pager_return_t
vnode_pageout(struct vnode * vp,upl_t upl,upl_offset_t upl_offset,vm_object_offset_t f_offset,upl_size_t size,int flags,int * errorp)286 vnode_pageout(struct vnode *vp,
287     upl_t                   upl,
288     upl_offset_t            upl_offset,
289     vm_object_offset_t      f_offset,
290     upl_size_t              size,
291     int                     flags,
292     int                     *errorp)
293 {
294 	int             result = PAGER_SUCCESS;
295 	int             error = 0;
296 	int             error_ret = 0;
297 	daddr64_t blkno;
298 	int isize;
299 	int pg_index;
300 	int base_index;
301 	upl_offset_t offset;
302 	upl_page_info_t *pl;
303 	vfs_context_t ctx = vfs_context_current();      /* pager context */
304 
305 	isize = (int)size;
306 
307 	/*
308 	 * This call is non-blocking and does not ever fail but it can
309 	 * only be made when there is other explicit synchronization
310 	 * with reclaiming of the vnode which, in this path, is provided
311 	 * by the paging in progress counter.
312 	 *
313 	 * In addition, this may also be entered via explicit ubc_msync
314 	 * calls or vm_swapfile_io where the existing iocount provides
315 	 * the necessary synchronization. Ideally we would not take an
316 	 * additional iocount here in the cases where an explcit iocount
317 	 * has already been taken but this call doesn't cause a deadlock
318 	 * as other forms of vnode_get* might if this thread has already
319 	 * taken an iocount.
320 	 */
321 	error = vnode_getalways_from_pager(vp);
322 	if (error != 0) {
323 		/* This can't happen */
324 		panic("vnode_getalways returned %d for vp %p", error, vp);
325 	}
326 
327 	if (isize <= 0) {
328 		result    = PAGER_ERROR;
329 		error_ret = EINVAL;
330 		goto out;
331 	}
332 
333 	if (UBCINFOEXISTS(vp) == 0) {
334 		result    = PAGER_ERROR;
335 		error_ret = EINVAL;
336 
337 		if (upl && !(flags & UPL_NOCOMMIT)) {
338 			ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
339 		}
340 		goto out;
341 	}
342 	if (!(flags & UPL_VNODE_PAGER)) {
343 		/*
344 		 * This is a pageout from the default pager,
345 		 * just go ahead and call vnop_pageout since
346 		 * it has already sorted out the dirty ranges
347 		 */
348 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
349 		    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
350 		    size, 1, 0, 0, 0);
351 
352 		if ((error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset,
353 		    (size_t)size, flags, ctx))) {
354 			result = PAGER_ERROR;
355 		}
356 
357 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
358 		    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
359 		    size, 1, 0, 0, 0);
360 
361 		goto out;
362 	}
363 	if (upl == NULL) {
364 		int                     request_flags;
365 
366 		if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEOUTV2) {
367 			/*
368 			 * filesystem has requested the new form of VNOP_PAGEOUT for file
369 			 * backed objects... we will not grab the UPL befofe calling VNOP_PAGEOUT...
370 			 * it is the fileystem's responsibility to grab the range we're denoting
371 			 * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first
372 			 * take any locks it needs, before effectively locking the pages into a UPL...
373 			 */
374 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
375 			    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
376 			    size, (int)f_offset, 0, 0, 0);
377 
378 			if ((error_ret = VNOP_PAGEOUT(vp, NULL, upl_offset, (off_t)f_offset,
379 			    size, flags, ctx))) {
380 				result = PAGER_ERROR;
381 			}
382 			KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
383 			    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
384 			    size, 0, 0, 0, 0);
385 
386 			goto out;
387 		}
388 		if (flags & UPL_MSYNC) {
389 			request_flags = UPL_UBC_MSYNC | UPL_RET_ONLY_DIRTY;
390 		} else {
391 			request_flags = UPL_UBC_PAGEOUT | UPL_RET_ONLY_DIRTY;
392 		}
393 
394 		if (ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, request_flags, VM_KERN_MEMORY_FILE) != KERN_SUCCESS) {
395 			result    = PAGER_ERROR;
396 			error_ret = EINVAL;
397 			goto out;
398 		}
399 		upl_offset = 0;
400 	} else {
401 		pl = ubc_upl_pageinfo(upl);
402 	}
403 
404 	/*
405 	 * Ignore any non-present pages at the end of the
406 	 * UPL so that we aren't looking at a upl that
407 	 * may already have been freed by the preceeding
408 	 * aborts/completions.
409 	 */
410 	base_index = upl_offset / PAGE_SIZE;
411 
412 	for (pg_index = (upl_offset + isize) / PAGE_SIZE; pg_index > base_index;) {
413 		if (upl_page_present(pl, --pg_index)) {
414 			break;
415 		}
416 		if (pg_index == base_index) {
417 			/*
418 			 * no pages were returned, so release
419 			 * our hold on the upl and leave
420 			 */
421 			if (!(flags & UPL_NOCOMMIT)) {
422 				ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY);
423 			}
424 
425 			goto out;
426 		}
427 	}
428 	isize = ((pg_index + 1) - base_index) * PAGE_SIZE;
429 
430 	/*
431 	 * we come here for pageouts to 'real' files and
432 	 * for msyncs...  the upl may not contain any
433 	 * dirty pages.. it's our responsibility to sort
434 	 * through it and find the 'runs' of dirty pages
435 	 * to call VNOP_PAGEOUT on...
436 	 */
437 
438 	if (ubc_getsize(vp) == 0) {
439 		/*
440 		 * if the file has been effectively deleted, then
441 		 * we need to go through the UPL and invalidate any
442 		 * buffer headers we might have that reference any
443 		 * of it's pages
444 		 */
445 		for (offset = upl_offset; isize; isize -= PAGE_SIZE, offset += PAGE_SIZE) {
446 			if (vp->v_tag == VT_NFS) {
447 				/* check with nfs if page is OK to drop */
448 				error = nfs_buf_page_inval(vp, (off_t)f_offset);
449 			} else {
450 				blkno = ubc_offtoblk(vp, (off_t)f_offset);
451 				error = buf_invalblkno(vp, blkno, 0);
452 			}
453 			if (error) {
454 				if (!(flags & UPL_NOCOMMIT)) {
455 					ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
456 				}
457 				if (error_ret == 0) {
458 					error_ret = error;
459 				}
460 				result = PAGER_ERROR;
461 			} else if (!(flags & UPL_NOCOMMIT)) {
462 				ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
463 			}
464 			f_offset += PAGE_SIZE;
465 		}
466 		goto out;
467 	}
468 
469 	offset = upl_offset;
470 	pg_index = base_index;
471 
472 	while (isize) {
473 		int  xsize;
474 		int  num_of_pages;
475 
476 		if (!upl_page_present(pl, pg_index)) {
477 			/*
478 			 * we asked for RET_ONLY_DIRTY, so it's possible
479 			 * to get back empty slots in the UPL
480 			 * just skip over them
481 			 */
482 			f_offset += PAGE_SIZE;
483 			offset   += PAGE_SIZE;
484 			isize    -= PAGE_SIZE;
485 			pg_index++;
486 
487 			continue;
488 		}
489 		if (!upl_dirty_page(pl, pg_index)) {
490 			/*
491 			 * if the page is not dirty and reached here it is
492 			 * marked precious or it is due to invalidation in
493 			 * memory_object_lock request as part of truncation
494 			 * We also get here from vm_object_terminate()
495 			 * So all you need to do in these
496 			 * cases is to invalidate incore buffer if it is there
497 			 * Note we must not sleep here if the buffer is busy - that is
498 			 * a lock inversion which causes deadlock.
499 			 */
500 			if (vp->v_tag == VT_NFS) {
501 				/* check with nfs if page is OK to drop */
502 				error = nfs_buf_page_inval(vp, (off_t)f_offset);
503 			} else {
504 				blkno = ubc_offtoblk(vp, (off_t)f_offset);
505 				error = buf_invalblkno(vp, blkno, 0);
506 			}
507 			if (error) {
508 				if (!(flags & UPL_NOCOMMIT)) {
509 					ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
510 				}
511 				if (error_ret == 0) {
512 					error_ret = error;
513 				}
514 				result = PAGER_ERROR;
515 			} else if (!(flags & UPL_NOCOMMIT)) {
516 				ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
517 			}
518 			f_offset += PAGE_SIZE;
519 			offset   += PAGE_SIZE;
520 			isize    -= PAGE_SIZE;
521 			pg_index++;
522 
523 			continue;
524 		}
525 		num_of_pages = 1;
526 		xsize = isize - PAGE_SIZE;
527 
528 		while (xsize) {
529 			if (!upl_dirty_page(pl, pg_index + num_of_pages)) {
530 				break;
531 			}
532 			num_of_pages++;
533 			xsize -= PAGE_SIZE;
534 		}
535 		xsize = num_of_pages * PAGE_SIZE;
536 
537 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
538 		    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
539 		    xsize, (int)f_offset, 0, 0, 0);
540 
541 		if ((error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset,
542 		    xsize, flags, ctx))) {
543 			if (error_ret == 0) {
544 				error_ret = error;
545 			}
546 			result = PAGER_ERROR;
547 		}
548 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
549 		    (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
550 		    xsize, 0, 0, 0, 0);
551 
552 		f_offset += xsize;
553 		offset   += xsize;
554 		isize    -= xsize;
555 		pg_index += num_of_pages;
556 	}
557 out:
558 	vnode_put_from_pager(vp);
559 
560 	if (errorp) {
561 		*errorp = error_ret;
562 	}
563 
564 	return result;
565 }
566 
567 
568 pager_return_t
vnode_pagein(struct vnode * vp,upl_t upl,upl_offset_t upl_offset,vm_object_offset_t f_offset,upl_size_t size,int flags,int * errorp)569 vnode_pagein(
570 	struct vnode            *vp,
571 	upl_t                   upl,
572 	upl_offset_t            upl_offset,
573 	vm_object_offset_t      f_offset,
574 	upl_size_t              size,
575 	int                     flags,
576 	int                     *errorp)
577 {
578 	upl_page_info_t *pl;
579 	int             result = PAGER_SUCCESS;
580 	int             error = 0;
581 	int             pages_in_upl;
582 	int             start_pg;
583 	int             last_pg;
584 	int             first_pg;
585 	int             xsize;
586 	int             must_commit = 1;
587 	int             ignore_valid_page_check = 0;
588 
589 	if (flags & UPL_NOCOMMIT) {
590 		must_commit = 0;
591 	}
592 
593 	if (flags & UPL_IGNORE_VALID_PAGE_CHECK) {
594 		ignore_valid_page_check = 1;
595 	}
596 
597 	/*
598 	 * This call is non-blocking and does not ever fail but it can
599 	 * only be made when there is other explicit synchronization
600 	 * with reclaiming of the vnode which, in this path, is provided
601 	 * by the paging in progress counter.
602 	 *
603 	 * In addition, this may also be entered via vm_swapfile_io
604 	 * where the existing iocount provides the necessary synchronization.
605 	 * Ideally we would not take an additional iocount here in the cases
606 	 * where an explcit iocount has already been taken but this call
607 	 * doesn't cause a deadlock as other forms of vnode_get* might if
608 	 * this thread has already taken an iocount.
609 	 */
610 	error = vnode_getalways_from_pager(vp);
611 	if (error != 0) {
612 		/* This can't happen */
613 		panic("vnode_getalways returned %d for vp %p", error, vp);
614 	}
615 
616 	if (UBCINFOEXISTS(vp) == 0) {
617 		result = PAGER_ERROR;
618 		error  = PAGER_ERROR;
619 
620 		if (upl && must_commit) {
621 			ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
622 		}
623 
624 		ktriage_record(thread_tid(current_thread()), KDBG_TRIAGE_EVENTID(KDBG_TRIAGE_SUBSYS_VM, KDBG_TRIAGE_RESERVED, KDBG_TRIAGE_VM_VNODEPAGEIN_NO_UBCINFO), 0 /* arg */);
625 		goto out;
626 	}
627 	if (upl == (upl_t)NULL) {
628 		flags &= ~UPL_NOCOMMIT;
629 
630 		if (size > MAX_UPL_SIZE_BYTES) {
631 			result = PAGER_ERROR;
632 			error  = PAGER_ERROR;
633 			goto out;
634 		}
635 		if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEINV2) {
636 			/*
637 			 * filesystem has requested the new form of VNOP_PAGEIN for file
638 			 * backed objects... we will not grab the UPL befofe calling VNOP_PAGEIN...
639 			 * it is the fileystem's responsibility to grab the range we're denoting
640 			 * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first
641 			 * take any locks it needs, before effectively locking the pages into a UPL...
642 			 * so we pass a NULL into the filesystem instead of a UPL pointer... the 'upl_offset'
643 			 * is used to identify the "must have" page in the extent... the filesystem is free
644 			 * to clip the extent to better fit the underlying FS blocksize if it desires as
645 			 * long as it continues to include the "must have" page... 'f_offset' + 'upl_offset'
646 			 * identifies that page
647 			 */
648 			if ((error = VNOP_PAGEIN(vp, NULL, upl_offset, (off_t)f_offset,
649 			    size, flags, vfs_context_current()))) {
650 				set_thread_pagein_error(current_thread(), error);
651 				result = PAGER_ERROR;
652 				error  = PAGER_ERROR;
653 				ktriage_record(thread_tid(current_thread()), KDBG_TRIAGE_EVENTID(KDBG_TRIAGE_SUBSYS_VM, KDBG_TRIAGE_RESERVED, KDBG_TRIAGE_VM_VNODEPAGEIN_FSPAGEIN_FAIL), 0 /* arg */);
654 			}
655 			goto out;
656 		}
657 		ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT, VM_KERN_MEMORY_FILE);
658 
659 		if (upl == (upl_t)NULL) {
660 			result =  PAGER_ABSENT;
661 			error = PAGER_ABSENT;
662 			ktriage_record(thread_tid(current_thread()), KDBG_TRIAGE_EVENTID(KDBG_TRIAGE_SUBSYS_VM, KDBG_TRIAGE_RESERVED, KDBG_TRIAGE_VM_VNODEPAGEIN_NO_UPL), 0 /* arg */);
663 			goto out;
664 		}
665 		ubc_upl_range_needed(upl, upl_offset / PAGE_SIZE, 1);
666 
667 		upl_offset = 0;
668 		first_pg = 0;
669 
670 		/*
671 		 * if we get here, we've created the upl and
672 		 * are responsible for commiting/aborting it
673 		 * regardless of what the caller has passed in
674 		 */
675 		must_commit = 1;
676 	} else {
677 		pl = ubc_upl_pageinfo(upl);
678 		first_pg = upl_offset / PAGE_SIZE;
679 	}
680 	pages_in_upl = size / PAGE_SIZE;
681 	DTRACE_VM2(pgpgin, int, pages_in_upl, (uint64_t *), NULL);
682 
683 	/*
684 	 * before we start marching forward, we must make sure we end on
685 	 * a present page, otherwise we will be working with a freed
686 	 * upl
687 	 */
688 	for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) {
689 		if (upl_page_present(pl, last_pg)) {
690 			break;
691 		}
692 		if (last_pg == first_pg) {
693 			/*
694 			 * empty UPL, no pages are present
695 			 */
696 			if (must_commit) {
697 				ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
698 			}
699 			goto out;
700 		}
701 	}
702 	pages_in_upl = last_pg + 1;
703 	last_pg = first_pg;
704 
705 	while (last_pg < pages_in_upl) {
706 		/*
707 		 * skip over missing pages...
708 		 */
709 		for (; last_pg < pages_in_upl; last_pg++) {
710 			if (upl_page_present(pl, last_pg)) {
711 				break;
712 			}
713 		}
714 
715 		if (ignore_valid_page_check == 1) {
716 			start_pg = last_pg;
717 		} else {
718 			/*
719 			 * skip over 'valid' pages... we don't want to issue I/O for these
720 			 */
721 			for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
722 				if (!upl_valid_page(pl, last_pg)) {
723 					break;
724 				}
725 			}
726 		}
727 
728 		if (last_pg > start_pg) {
729 			/*
730 			 * we've found a range of valid pages
731 			 * if we've got COMMIT responsibility
732 			 * commit this range of pages back to the
733 			 * cache unchanged
734 			 */
735 			xsize = (last_pg - start_pg) * PAGE_SIZE;
736 
737 			if (must_commit) {
738 				ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY);
739 			}
740 		}
741 		if (last_pg == pages_in_upl) {
742 			/*
743 			 * we're done... all pages that were present
744 			 * have either had I/O issued on them or
745 			 * were aborted unchanged...
746 			 */
747 			break;
748 		}
749 
750 		if (!upl_page_present(pl, last_pg)) {
751 			/*
752 			 * we found a range of valid pages
753 			 * terminated by a missing page...
754 			 * bump index to the next page and continue on
755 			 */
756 			last_pg++;
757 			continue;
758 		}
759 		/*
760 		 * scan from the found invalid page looking for a valid
761 		 * or non-present page before the end of the upl is reached, if we
762 		 * find one, then it will be the last page of the request to
763 		 * 'cluster_io'
764 		 */
765 		for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
766 			if ((!ignore_valid_page_check && upl_valid_page(pl, last_pg)) || !upl_page_present(pl, last_pg)) {
767 				break;
768 			}
769 		}
770 		if (last_pg > start_pg) {
771 			int xoff;
772 			xsize = (last_pg - start_pg) * PAGE_SIZE;
773 			xoff  = start_pg * PAGE_SIZE;
774 
775 			if ((error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff,
776 			    (off_t)f_offset + xoff,
777 			    xsize, flags, vfs_context_current()))) {
778 				/*
779 				 * Usually this UPL will be aborted/committed by the lower cluster layer.
780 				 *
781 				 * a)	In the case of decmpfs, however, we may return an error (EAGAIN) to avoid
782 				 *	a deadlock with another thread already inflating the file.
783 				 *
784 				 * b)	In the case of content protection, EPERM is a valid error and we should respect it.
785 				 *
786 				 * In those cases, we must take care of our UPL at this layer itself.
787 				 */
788 				if (must_commit) {
789 					if (error == EAGAIN) {
790 						ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_RESTART);
791 					}
792 					if (error == EPERM) {
793 						ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
794 					}
795 				}
796 				set_thread_pagein_error(current_thread(), error);
797 				result = PAGER_ERROR;
798 				error  = PAGER_ERROR;
799 				ktriage_record(thread_tid(current_thread()), KDBG_TRIAGE_EVENTID(KDBG_TRIAGE_SUBSYS_VM, KDBG_TRIAGE_RESERVED, KDBG_TRIAGE_VM_VNODEPAGEIN_FSPAGEIN_FAIL), 0 /* arg */);
800 			}
801 		}
802 	}
803 out:
804 	vnode_put_from_pager(vp);
805 
806 	if (errorp) {
807 		*errorp = result;
808 	}
809 
810 	return error;
811 }
812 
813 void *
upl_get_internal_page_list(upl_t upl)814 upl_get_internal_page_list(upl_t upl)
815 {
816 	return UPL_GET_INTERNAL_PAGE_LIST(upl);
817 }
818