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 | /* |
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/memory_object_control.h> |
60 | #include <mach/vm_map.h> |
61 | #include <mach/mach_vm.h> |
62 | #include <mach/upl.h> |
63 | #include <mach/sdt.h> |
64 | |
65 | #include <vm/vm_map.h> |
66 | #include <vm/vm_kern.h> |
67 | #include <kern/zalloc.h> |
68 | #include <kern/kalloc.h> |
69 | #include <libkern/libkern.h> |
70 | |
71 | #include <vm/vnode_pager.h> |
72 | #include <vm/vm_pageout.h> |
73 | |
74 | #include <kern/assert.h> |
75 | #include <sys/kdebug.h> |
76 | #include <nfs/rpcv2.h> |
77 | #include <nfs/nfsproto.h> |
78 | #include <nfs/nfs.h> |
79 | |
80 | #include <vm/vm_protos.h> |
81 | |
82 | #include <vfs/vfs_disk_conditioner.h> |
83 | |
84 | void |
85 | () |
86 | { |
87 | struct uthread *ut; |
88 | |
89 | ut = get_bsdthread_info(current_thread()); |
90 | |
91 | if (ut->uu_lowpri_window) |
92 | throttle_lowpri_io(1); |
93 | } |
94 | |
95 | boolean_t |
96 | (vnode_t vp) |
97 | { |
98 | return disk_conditioner_mount_is_ssd(vp->v_mount); |
99 | } |
100 | |
101 | #if CONFIG_IOSCHED |
102 | void |
103 | (struct vnode *devvp, uint64_t blkno, uint32_t len, int priority) |
104 | { |
105 | u_int32_t blocksize = 0; |
106 | dk_extent_t extent; |
107 | dk_set_tier_t set_tier; |
108 | int error = 0; |
109 | |
110 | error = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel()); |
111 | if (error) |
112 | return; |
113 | |
114 | memset(&extent, 0, sizeof(dk_extent_t)); |
115 | memset(&set_tier, 0, sizeof(dk_set_tier_t)); |
116 | |
117 | extent.offset = blkno * (u_int64_t) blocksize; |
118 | extent.length = len; |
119 | |
120 | set_tier.extents = &extent; |
121 | set_tier.extentsCount = 1; |
122 | set_tier.tier = priority; |
123 | |
124 | error = VNOP_IOCTL(devvp, DKIOCSETTIER, (caddr_t)&set_tier, 0, vfs_context_kernel()); |
125 | return; |
126 | } |
127 | #endif |
128 | |
129 | void |
130 | ( |
131 | struct vnode *vp, |
132 | vm_object_offset_t s_offset, |
133 | vm_object_offset_t e_offset) |
134 | { |
135 | cluster_update_state(vp, s_offset, e_offset, TRUE); |
136 | } |
137 | |
138 | uint32_t |
139 | (struct vnode *vp) |
140 | { |
141 | if (vp->v_usecount > vp->v_kusecount) |
142 | return (1); |
143 | return (0); |
144 | } |
145 | |
146 | uint32_t |
147 | (struct vnode *vp, uint32_t *limit) |
148 | { |
149 | return(cluster_throttle_io_limit(vp, limit)); |
150 | } |
151 | |
152 | vm_object_offset_t |
153 | (struct vnode *vp) |
154 | { |
155 | return (vm_object_offset_t) ubc_getsize(vp); |
156 | } |
157 | |
158 | extern int safe_getpath(struct vnode *dvp, char *leafname, char *path, int _len, int *truncated_path); |
159 | |
160 | kern_return_t |
161 | ( |
162 | struct vnode *vp, |
163 | char *pathname, |
164 | vm_size_t pathname_len, |
165 | char *filename, |
166 | vm_size_t filename_len, |
167 | boolean_t *truncated_path_p) |
168 | { |
169 | *truncated_path_p = FALSE; |
170 | if (pathname != NULL) { |
171 | /* get the path name */ |
172 | safe_getpath(vp, NULL, |
173 | pathname, (int) pathname_len, |
174 | truncated_path_p); |
175 | } |
176 | if ((pathname == NULL || *truncated_path_p) && |
177 | filename != NULL) { |
178 | /* get the file name */ |
179 | const char *name; |
180 | |
181 | name = vnode_getname_printable(vp); |
182 | strlcpy(filename, name, (size_t) filename_len); |
183 | vnode_putname_printable(name); |
184 | } |
185 | return KERN_SUCCESS; |
186 | } |
187 | |
188 | kern_return_t |
189 | ( |
190 | struct vnode *vp, |
191 | struct timespec *current_mtime, |
192 | struct timespec *cs_mtime) |
193 | { |
194 | vnode_mtime(vp, current_mtime, vfs_context_current()); |
195 | if (cs_mtime != NULL) { |
196 | ubc_get_cs_mtime(vp, cs_mtime); |
197 | } |
198 | return KERN_SUCCESS; |
199 | } |
200 | |
201 | kern_return_t |
202 | ( |
203 | struct vnode *vp, |
204 | void **blobs) |
205 | { |
206 | *blobs = ubc_get_cs_blobs(vp); |
207 | return KERN_SUCCESS; |
208 | } |
209 | |
210 | /* |
211 | * vnode_trim: |
212 | * Used to call the DKIOCUNMAP ioctl on the underlying disk device for the specified vnode. |
213 | * Trims the region at offset bytes into the file, for length bytes. |
214 | * |
215 | * Care must be taken to ensure that the vnode is sufficiently reference counted at the time this |
216 | * function is called; no iocounts or usecounts are taken on the vnode. |
217 | * This function is non-idempotent in error cases; We cannot un-discard the blocks if only some of them |
218 | * are successfully discarded. |
219 | */ |
220 | u_int32_t vnode_trim ( |
221 | struct vnode *vp, |
222 | off_t offset, |
223 | size_t length) |
224 | { |
225 | daddr64_t io_blockno; /* Block number corresponding to the start of the extent */ |
226 | size_t io_bytecount; /* Number of bytes in current extent for the specified range */ |
227 | size_t trimmed = 0; |
228 | off_t current_offset = offset; |
229 | size_t remaining_length = length; |
230 | int error = 0; |
231 | u_int32_t blocksize = 0; |
232 | struct vnode *devvp; |
233 | dk_extent_t extent; |
234 | dk_unmap_t unmap; |
235 | |
236 | |
237 | /* Get the underlying device vnode */ |
238 | devvp = vp->v_mount->mnt_devvp; |
239 | |
240 | /* Figure out the underlying device block size */ |
241 | error = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel()); |
242 | if (error) { |
243 | goto trim_exit; |
244 | } |
245 | |
246 | /* |
247 | * We may not get the entire range from offset -> offset+length in a single |
248 | * extent from the blockmap call. Keep looping/going until we are sure we've hit |
249 | * the whole range or if we encounter an error. |
250 | */ |
251 | while (trimmed < length) { |
252 | /* |
253 | * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the |
254 | * specified offset. It returns blocks in contiguous chunks, so if the logical range is |
255 | * broken into multiple extents, it must be called multiple times, increasing the offset |
256 | * in each call to ensure that the entire range is covered. |
257 | */ |
258 | error = VNOP_BLOCKMAP (vp, current_offset, remaining_length, |
259 | &io_blockno, &io_bytecount, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL); |
260 | |
261 | if (error) { |
262 | goto trim_exit; |
263 | } |
264 | /* |
265 | * We have a contiguous run. Prepare & issue the ioctl for the device. |
266 | * the DKIOCUNMAP ioctl takes offset in bytes from the start of the device. |
267 | */ |
268 | memset (&extent, 0, sizeof(dk_extent_t)); |
269 | memset (&unmap, 0, sizeof(dk_unmap_t)); |
270 | extent.offset = (uint64_t) io_blockno * (u_int64_t) blocksize; |
271 | extent.length = io_bytecount; |
272 | unmap.extents = &extent; |
273 | unmap.extentsCount = 1; |
274 | error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel()); |
275 | |
276 | if (error) { |
277 | goto trim_exit; |
278 | } |
279 | remaining_length = remaining_length - io_bytecount; |
280 | trimmed = trimmed + io_bytecount; |
281 | current_offset = current_offset + io_bytecount; |
282 | } |
283 | trim_exit: |
284 | |
285 | return error; |
286 | |
287 | } |
288 | |
289 | pager_return_t |
290 | vnode_pageout(struct vnode *vp, |
291 | upl_t upl, |
292 | upl_offset_t upl_offset, |
293 | vm_object_offset_t f_offset, |
294 | upl_size_t size, |
295 | int flags, |
296 | int *errorp) |
297 | { |
298 | int result = PAGER_SUCCESS; |
299 | int error = 0; |
300 | int error_ret = 0; |
301 | daddr64_t blkno; |
302 | int isize; |
303 | int pg_index; |
304 | int base_index; |
305 | upl_offset_t offset; |
306 | upl_page_info_t *pl; |
307 | vfs_context_t ctx = vfs_context_current(); /* pager context */ |
308 | |
309 | isize = (int)size; |
310 | |
311 | if (isize <= 0) { |
312 | result = PAGER_ERROR; |
313 | error_ret = EINVAL; |
314 | goto out; |
315 | } |
316 | |
317 | if (UBCINFOEXISTS(vp) == 0) { |
318 | result = PAGER_ERROR; |
319 | error_ret = EINVAL; |
320 | |
321 | if (upl && !(flags & UPL_NOCOMMIT)) |
322 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY); |
323 | goto out; |
324 | } |
325 | if ( !(flags & UPL_VNODE_PAGER)) { |
326 | /* |
327 | * This is a pageout from the default pager, |
328 | * just go ahead and call vnop_pageout since |
329 | * it has already sorted out the dirty ranges |
330 | */ |
331 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
332 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, |
333 | size, 1, 0, 0, 0); |
334 | |
335 | if ( (error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset, |
336 | (size_t)size, flags, ctx)) ) |
337 | result = PAGER_ERROR; |
338 | |
339 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
340 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, |
341 | size, 1, 0, 0, 0); |
342 | |
343 | goto out; |
344 | } |
345 | if (upl == NULL) { |
346 | int request_flags; |
347 | |
348 | if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEOUTV2) { |
349 | /* |
350 | * filesystem has requested the new form of VNOP_PAGEOUT for file |
351 | * backed objects... we will not grab the UPL befofe calling VNOP_PAGEOUT... |
352 | * it is the fileystem's responsibility to grab the range we're denoting |
353 | * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first |
354 | * take any locks it needs, before effectively locking the pages into a UPL... |
355 | */ |
356 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
357 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, |
358 | size, (int)f_offset, 0, 0, 0); |
359 | |
360 | if ( (error_ret = VNOP_PAGEOUT(vp, NULL, upl_offset, (off_t)f_offset, |
361 | size, flags, ctx)) ) { |
362 | result = PAGER_ERROR; |
363 | } |
364 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
365 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, |
366 | size, 0, 0, 0, 0); |
367 | |
368 | goto out; |
369 | } |
370 | if (flags & UPL_MSYNC) |
371 | request_flags = UPL_UBC_MSYNC | UPL_RET_ONLY_DIRTY; |
372 | else |
373 | request_flags = UPL_UBC_PAGEOUT | UPL_RET_ONLY_DIRTY; |
374 | |
375 | if (ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, request_flags, VM_KERN_MEMORY_FILE) != KERN_SUCCESS) { |
376 | result = PAGER_ERROR; |
377 | error_ret = EINVAL; |
378 | goto out; |
379 | } |
380 | upl_offset = 0; |
381 | } else |
382 | pl = ubc_upl_pageinfo(upl); |
383 | |
384 | /* |
385 | * Ignore any non-present pages at the end of the |
386 | * UPL so that we aren't looking at a upl that |
387 | * may already have been freed by the preceeding |
388 | * aborts/completions. |
389 | */ |
390 | base_index = upl_offset / PAGE_SIZE; |
391 | |
392 | for (pg_index = (upl_offset + isize) / PAGE_SIZE; pg_index > base_index;) { |
393 | if (upl_page_present(pl, --pg_index)) |
394 | break; |
395 | if (pg_index == base_index) { |
396 | /* |
397 | * no pages were returned, so release |
398 | * our hold on the upl and leave |
399 | */ |
400 | if ( !(flags & UPL_NOCOMMIT)) |
401 | ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY); |
402 | |
403 | goto out; |
404 | } |
405 | } |
406 | isize = ((pg_index + 1) - base_index) * PAGE_SIZE; |
407 | |
408 | /* |
409 | * we come here for pageouts to 'real' files and |
410 | * for msyncs... the upl may not contain any |
411 | * dirty pages.. it's our responsibility to sort |
412 | * through it and find the 'runs' of dirty pages |
413 | * to call VNOP_PAGEOUT on... |
414 | */ |
415 | |
416 | if (ubc_getsize(vp) == 0) { |
417 | /* |
418 | * if the file has been effectively deleted, then |
419 | * we need to go through the UPL and invalidate any |
420 | * buffer headers we might have that reference any |
421 | * of it's pages |
422 | */ |
423 | for (offset = upl_offset; isize; isize -= PAGE_SIZE, offset += PAGE_SIZE) { |
424 | #if NFSCLIENT |
425 | if (vp->v_tag == VT_NFS) |
426 | /* check with nfs if page is OK to drop */ |
427 | error = nfs_buf_page_inval(vp, (off_t)f_offset); |
428 | else |
429 | #endif |
430 | { |
431 | blkno = ubc_offtoblk(vp, (off_t)f_offset); |
432 | error = buf_invalblkno(vp, blkno, 0); |
433 | } |
434 | if (error) { |
435 | if ( !(flags & UPL_NOCOMMIT)) |
436 | ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY); |
437 | if (error_ret == 0) |
438 | error_ret = error; |
439 | result = PAGER_ERROR; |
440 | |
441 | } else if ( !(flags & UPL_NOCOMMIT)) { |
442 | ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY); |
443 | } |
444 | f_offset += PAGE_SIZE; |
445 | } |
446 | goto out; |
447 | } |
448 | |
449 | offset = upl_offset; |
450 | pg_index = base_index; |
451 | |
452 | while (isize) { |
453 | int xsize; |
454 | int num_of_pages; |
455 | |
456 | if ( !upl_page_present(pl, pg_index)) { |
457 | /* |
458 | * we asked for RET_ONLY_DIRTY, so it's possible |
459 | * to get back empty slots in the UPL |
460 | * just skip over them |
461 | */ |
462 | f_offset += PAGE_SIZE; |
463 | offset += PAGE_SIZE; |
464 | isize -= PAGE_SIZE; |
465 | pg_index++; |
466 | |
467 | continue; |
468 | } |
469 | if ( !upl_dirty_page(pl, pg_index)) { |
470 | /* |
471 | * if the page is not dirty and reached here it is |
472 | * marked precious or it is due to invalidation in |
473 | * memory_object_lock request as part of truncation |
474 | * We also get here from vm_object_terminate() |
475 | * So all you need to do in these |
476 | * cases is to invalidate incore buffer if it is there |
477 | * Note we must not sleep here if the buffer is busy - that is |
478 | * a lock inversion which causes deadlock. |
479 | */ |
480 | #if NFSCLIENT |
481 | if (vp->v_tag == VT_NFS) |
482 | /* check with nfs if page is OK to drop */ |
483 | error = nfs_buf_page_inval(vp, (off_t)f_offset); |
484 | else |
485 | #endif |
486 | { |
487 | blkno = ubc_offtoblk(vp, (off_t)f_offset); |
488 | error = buf_invalblkno(vp, blkno, 0); |
489 | } |
490 | if (error) { |
491 | if ( !(flags & UPL_NOCOMMIT)) |
492 | ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY); |
493 | if (error_ret == 0) |
494 | error_ret = error; |
495 | result = PAGER_ERROR; |
496 | |
497 | } else if ( !(flags & UPL_NOCOMMIT)) { |
498 | ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY); |
499 | } |
500 | f_offset += PAGE_SIZE; |
501 | offset += PAGE_SIZE; |
502 | isize -= PAGE_SIZE; |
503 | pg_index++; |
504 | |
505 | continue; |
506 | } |
507 | num_of_pages = 1; |
508 | xsize = isize - PAGE_SIZE; |
509 | |
510 | while (xsize) { |
511 | if ( !upl_dirty_page(pl, pg_index + num_of_pages)) |
512 | break; |
513 | num_of_pages++; |
514 | xsize -= PAGE_SIZE; |
515 | } |
516 | xsize = num_of_pages * PAGE_SIZE; |
517 | |
518 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
519 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, |
520 | xsize, (int)f_offset, 0, 0, 0); |
521 | |
522 | if ( (error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset, |
523 | xsize, flags, ctx)) ) { |
524 | if (error_ret == 0) |
525 | error_ret = error; |
526 | result = PAGER_ERROR; |
527 | } |
528 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
529 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, |
530 | xsize, 0, 0, 0, 0); |
531 | |
532 | f_offset += xsize; |
533 | offset += xsize; |
534 | isize -= xsize; |
535 | pg_index += num_of_pages; |
536 | } |
537 | out: |
538 | if (errorp) |
539 | *errorp = error_ret; |
540 | |
541 | return (result); |
542 | } |
543 | |
544 | |
545 | pager_return_t |
546 | vnode_pagein( |
547 | struct vnode *vp, |
548 | upl_t upl, |
549 | upl_offset_t upl_offset, |
550 | vm_object_offset_t f_offset, |
551 | upl_size_t size, |
552 | int flags, |
553 | int *errorp) |
554 | { |
555 | upl_page_info_t *pl; |
556 | int result = PAGER_SUCCESS; |
557 | int error = 0; |
558 | int pages_in_upl; |
559 | int start_pg; |
560 | int last_pg; |
561 | int first_pg; |
562 | int xsize; |
563 | int must_commit = 1; |
564 | int ignore_valid_page_check = 0; |
565 | |
566 | if (flags & UPL_NOCOMMIT) |
567 | must_commit = 0; |
568 | |
569 | if (flags & UPL_IGNORE_VALID_PAGE_CHECK) |
570 | ignore_valid_page_check = 1; |
571 | |
572 | if (UBCINFOEXISTS(vp) == 0) { |
573 | result = PAGER_ERROR; |
574 | error = PAGER_ERROR; |
575 | |
576 | if (upl && must_commit) |
577 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR); |
578 | |
579 | goto out; |
580 | } |
581 | if (upl == (upl_t)NULL) { |
582 | flags &= ~UPL_NOCOMMIT; |
583 | |
584 | if (size > MAX_UPL_SIZE_BYTES) { |
585 | result = PAGER_ERROR; |
586 | error = PAGER_ERROR; |
587 | goto out; |
588 | } |
589 | if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEINV2) { |
590 | /* |
591 | * filesystem has requested the new form of VNOP_PAGEIN for file |
592 | * backed objects... we will not grab the UPL befofe calling VNOP_PAGEIN... |
593 | * it is the fileystem's responsibility to grab the range we're denoting |
594 | * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first |
595 | * take any locks it needs, before effectively locking the pages into a UPL... |
596 | * so we pass a NULL into the filesystem instead of a UPL pointer... the 'upl_offset' |
597 | * is used to identify the "must have" page in the extent... the filesystem is free |
598 | * to clip the extent to better fit the underlying FS blocksize if it desires as |
599 | * long as it continues to include the "must have" page... 'f_offset' + 'upl_offset' |
600 | * identifies that page |
601 | */ |
602 | if ( (error = VNOP_PAGEIN(vp, NULL, upl_offset, (off_t)f_offset, |
603 | size, flags, vfs_context_current())) ) { |
604 | result = PAGER_ERROR; |
605 | error = PAGER_ERROR; |
606 | } |
607 | goto out; |
608 | } |
609 | ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT, VM_KERN_MEMORY_FILE); |
610 | |
611 | if (upl == (upl_t)NULL) { |
612 | result = PAGER_ABSENT; |
613 | error = PAGER_ABSENT; |
614 | goto out; |
615 | } |
616 | ubc_upl_range_needed(upl, upl_offset / PAGE_SIZE, 1); |
617 | |
618 | upl_offset = 0; |
619 | first_pg = 0; |
620 | |
621 | /* |
622 | * if we get here, we've created the upl and |
623 | * are responsible for commiting/aborting it |
624 | * regardless of what the caller has passed in |
625 | */ |
626 | must_commit = 1; |
627 | } else { |
628 | pl = ubc_upl_pageinfo(upl); |
629 | first_pg = upl_offset / PAGE_SIZE; |
630 | } |
631 | pages_in_upl = size / PAGE_SIZE; |
632 | DTRACE_VM2(pgpgin, int, pages_in_upl, (uint64_t *), NULL); |
633 | |
634 | /* |
635 | * before we start marching forward, we must make sure we end on |
636 | * a present page, otherwise we will be working with a freed |
637 | * upl |
638 | */ |
639 | for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) { |
640 | if (upl_page_present(pl, last_pg)) |
641 | break; |
642 | if (last_pg == first_pg) { |
643 | /* |
644 | * empty UPL, no pages are present |
645 | */ |
646 | if (must_commit) |
647 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY); |
648 | goto out; |
649 | } |
650 | } |
651 | pages_in_upl = last_pg + 1; |
652 | last_pg = first_pg; |
653 | |
654 | while (last_pg < pages_in_upl) { |
655 | /* |
656 | * skip over missing pages... |
657 | */ |
658 | for ( ; last_pg < pages_in_upl; last_pg++) { |
659 | if (upl_page_present(pl, last_pg)) |
660 | break; |
661 | } |
662 | |
663 | if (ignore_valid_page_check == 1) { |
664 | start_pg = last_pg; |
665 | } else { |
666 | /* |
667 | * skip over 'valid' pages... we don't want to issue I/O for these |
668 | */ |
669 | for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) { |
670 | if (!upl_valid_page(pl, last_pg)) |
671 | break; |
672 | } |
673 | } |
674 | |
675 | if (last_pg > start_pg) { |
676 | /* |
677 | * we've found a range of valid pages |
678 | * if we've got COMMIT responsibility |
679 | * commit this range of pages back to the |
680 | * cache unchanged |
681 | */ |
682 | xsize = (last_pg - start_pg) * PAGE_SIZE; |
683 | |
684 | if (must_commit) |
685 | ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY); |
686 | } |
687 | if (last_pg == pages_in_upl) |
688 | /* |
689 | * we're done... all pages that were present |
690 | * have either had I/O issued on them or |
691 | * were aborted unchanged... |
692 | */ |
693 | break; |
694 | |
695 | if (!upl_page_present(pl, last_pg)) { |
696 | /* |
697 | * we found a range of valid pages |
698 | * terminated by a missing page... |
699 | * bump index to the next page and continue on |
700 | */ |
701 | last_pg++; |
702 | continue; |
703 | } |
704 | /* |
705 | * scan from the found invalid page looking for a valid |
706 | * or non-present page before the end of the upl is reached, if we |
707 | * find one, then it will be the last page of the request to |
708 | * 'cluster_io' |
709 | */ |
710 | for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) { |
711 | if (( !ignore_valid_page_check && upl_valid_page(pl, last_pg)) || !upl_page_present(pl, last_pg)) |
712 | break; |
713 | } |
714 | if (last_pg > start_pg) { |
715 | int xoff; |
716 | xsize = (last_pg - start_pg) * PAGE_SIZE; |
717 | xoff = start_pg * PAGE_SIZE; |
718 | |
719 | if ( (error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff, |
720 | (off_t)f_offset + xoff, |
721 | xsize, flags, vfs_context_current())) ) { |
722 | /* |
723 | * Usually this UPL will be aborted/committed by the lower cluster layer. |
724 | * |
725 | * a) In the case of decmpfs, however, we may return an error (EAGAIN) to avoid |
726 | * a deadlock with another thread already inflating the file. |
727 | * |
728 | * b) In the case of content protection, EPERM is a valid error and we should respect it. |
729 | * |
730 | * In those cases, we must take care of our UPL at this layer itself. |
731 | */ |
732 | if (must_commit) { |
733 | if(error == EAGAIN) { |
734 | ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_RESTART); |
735 | } |
736 | if(error == EPERM) { |
737 | ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR); |
738 | } |
739 | } |
740 | result = PAGER_ERROR; |
741 | error = PAGER_ERROR; |
742 | |
743 | } |
744 | } |
745 | } |
746 | out: |
747 | if (errorp) |
748 | *errorp = result; |
749 | |
750 | return (error); |
751 | } |
752 | |
753 | void * |
754 | upl_get_internal_page_list(upl_t upl) |
755 | { |
756 | return(UPL_GET_INTERNAL_PAGE_LIST(upl)); |
757 | |
758 | } |
759 | |