]>
Commit | Line | Data |
---|---|---|
1c79356b | 1 | /* |
2d21ac55 | 2 | * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. |
5d5c5d0d | 3 | * |
2d21ac55 | 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
1c79356b | 5 | * |
2d21ac55 A |
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. | |
8f6c56a5 | 14 | * |
2d21ac55 A |
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 | |
8f6c56a5 A |
20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, | |
2d21ac55 A |
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. | |
8f6c56a5 | 25 | * |
2d21ac55 | 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
1c79356b A |
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> | |
91447636 | 45 | #include <sys/user.h> |
1c79356b | 46 | #include <sys/proc.h> |
91447636 | 47 | #include <sys/kauth.h> |
1c79356b A |
48 | #include <sys/buf.h> |
49 | #include <sys/uio.h> | |
91447636 | 50 | #include <sys/vnode_internal.h> |
1c79356b | 51 | #include <sys/namei.h> |
91447636 A |
52 | #include <sys/mount_internal.h> /* needs internal due to fhandle_t */ |
53 | #include <sys/ubc_internal.h> | |
1c79356b | 54 | #include <sys/lock.h> |
6d2010ae | 55 | #include <sys/disk.h> /* For DKIOC calls */ |
1c79356b A |
56 | |
57 | #include <mach/mach_types.h> | |
58 | #include <mach/memory_object_types.h> | |
b0d623f7 A |
59 | #include <mach/memory_object_control.h> |
60 | #include <mach/vm_map.h> | |
61 | #include <mach/mach_vm.h> | |
62 | #include <mach/upl.h> | |
2d21ac55 | 63 | #include <mach/sdt.h> |
1c79356b A |
64 | |
65 | #include <vm/vm_map.h> | |
66 | #include <vm/vm_kern.h> | |
1c79356b A |
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> | |
9bccf70c | 75 | #include <sys/kdebug.h> |
91447636 A |
76 | #include <nfs/rpcv2.h> |
77 | #include <nfs/nfsproto.h> | |
78 | #include <nfs/nfs.h> | |
79 | ||
80 | #include <vm/vm_protos.h> | |
1c79356b | 81 | |
5ba3f43e | 82 | #include <vfs/vfs_disk_conditioner.h> |
b0d623f7 | 83 | |
6d2010ae A |
84 | void |
85 | vnode_pager_throttle() | |
86 | { | |
87 | struct uthread *ut; | |
88 | ||
89 | ut = get_bsdthread_info(current_thread()); | |
90 | ||
91 | if (ut->uu_lowpri_window) | |
39236c6e | 92 | throttle_lowpri_io(1); |
6d2010ae A |
93 | } |
94 | ||
6d2010ae A |
95 | boolean_t |
96 | vnode_pager_isSSD(vnode_t vp) | |
97 | { | |
5ba3f43e | 98 | return disk_conditioner_mount_is_ssd(vp->v_mount); |
6d2010ae A |
99 | } |
100 | ||
fe8ab488 A |
101 | #if CONFIG_IOSCHED |
102 | void | |
103 | vnode_pager_issue_reprioritize_io(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 | |
6d2010ae | 128 | |
b0d623f7 A |
129 | uint32_t |
130 | vnode_pager_isinuse(struct vnode *vp) | |
131 | { | |
132 | if (vp->v_usecount > vp->v_kusecount) | |
133 | return (1); | |
134 | return (0); | |
135 | } | |
136 | ||
137 | uint32_t | |
39236c6e | 138 | vnode_pager_return_throttle_io_limit(struct vnode *vp, uint32_t *limit) |
b0d623f7 | 139 | { |
39236c6e | 140 | return(cluster_throttle_io_limit(vp, limit)); |
b0d623f7 | 141 | } |
1c79356b | 142 | |
0b4e3aa0 A |
143 | vm_object_offset_t |
144 | vnode_pager_get_filesize(struct vnode *vp) | |
145 | { | |
0b4e3aa0 | 146 | return (vm_object_offset_t) ubc_getsize(vp); |
0b4e3aa0 A |
147 | } |
148 | ||
15129b1c A |
149 | extern int safe_getpath(struct vnode *dvp, char *leafname, char *path, int _len, int *truncated_path); |
150 | ||
0c530ab8 | 151 | kern_return_t |
15129b1c | 152 | vnode_pager_get_name( |
0c530ab8 A |
153 | struct vnode *vp, |
154 | char *pathname, | |
15129b1c A |
155 | vm_size_t pathname_len, |
156 | char *filename, | |
157 | vm_size_t filename_len, | |
158 | boolean_t *truncated_path_p) | |
0c530ab8 | 159 | { |
15129b1c A |
160 | *truncated_path_p = FALSE; |
161 | if (pathname != NULL) { | |
162 | /* get the path name */ | |
163 | safe_getpath(vp, NULL, | |
164 | pathname, (int) pathname_len, | |
165 | truncated_path_p); | |
166 | } | |
167 | if ((pathname == NULL || *truncated_path_p) && | |
168 | filename != NULL) { | |
169 | /* get the file name */ | |
170 | const char *name; | |
171 | ||
172 | name = vnode_getname_printable(vp); | |
173 | strlcpy(filename, name, (size_t) filename_len); | |
174 | vnode_putname_printable(name); | |
0c530ab8 | 175 | } |
0c530ab8 A |
176 | return KERN_SUCCESS; |
177 | } | |
178 | ||
179 | kern_return_t | |
15129b1c | 180 | vnode_pager_get_mtime( |
0c530ab8 | 181 | struct vnode *vp, |
15129b1c A |
182 | struct timespec *current_mtime, |
183 | struct timespec *cs_mtime) | |
0c530ab8 | 184 | { |
15129b1c A |
185 | vnode_mtime(vp, current_mtime, vfs_context_current()); |
186 | if (cs_mtime != NULL) { | |
187 | ubc_get_cs_mtime(vp, cs_mtime); | |
188 | } | |
0c530ab8 A |
189 | return KERN_SUCCESS; |
190 | } | |
191 | ||
2d21ac55 A |
192 | kern_return_t |
193 | vnode_pager_get_cs_blobs( | |
194 | struct vnode *vp, | |
195 | void **blobs) | |
196 | { | |
197 | *blobs = ubc_get_cs_blobs(vp); | |
198 | return KERN_SUCCESS; | |
199 | } | |
200 | ||
6d2010ae A |
201 | /* |
202 | * vnode_trim: | |
203 | * Used to call the DKIOCUNMAP ioctl on the underlying disk device for the specified vnode. | |
204 | * Trims the region at offset bytes into the file, for length bytes. | |
205 | * | |
206 | * Care must be taken to ensure that the vnode is sufficiently reference counted at the time this | |
207 | * function is called; no iocounts or usecounts are taken on the vnode. | |
208 | * This function is non-idempotent in error cases; We cannot un-discard the blocks if only some of them | |
209 | * are successfully discarded. | |
210 | */ | |
211 | u_int32_t vnode_trim ( | |
212 | struct vnode *vp, | |
213 | off_t offset, | |
214 | size_t length) | |
215 | { | |
216 | daddr64_t io_blockno; /* Block number corresponding to the start of the extent */ | |
217 | size_t io_bytecount; /* Number of bytes in current extent for the specified range */ | |
218 | size_t trimmed = 0; | |
219 | off_t current_offset = offset; | |
220 | size_t remaining_length = length; | |
221 | int error = 0; | |
222 | u_int32_t blocksize = 0; | |
223 | struct vnode *devvp; | |
224 | dk_extent_t extent; | |
225 | dk_unmap_t unmap; | |
226 | ||
227 | ||
228 | /* Get the underlying device vnode */ | |
229 | devvp = vp->v_mount->mnt_devvp; | |
230 | ||
231 | /* Figure out the underlying device block size */ | |
232 | error = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel()); | |
233 | if (error) { | |
234 | goto trim_exit; | |
235 | } | |
236 | ||
237 | /* | |
238 | * We may not get the entire range from offset -> offset+length in a single | |
239 | * extent from the blockmap call. Keep looping/going until we are sure we've hit | |
240 | * the whole range or if we encounter an error. | |
241 | */ | |
242 | while (trimmed < length) { | |
243 | /* | |
244 | * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the | |
245 | * specified offset. It returns blocks in contiguous chunks, so if the logical range is | |
246 | * broken into multiple extents, it must be called multiple times, increasing the offset | |
247 | * in each call to ensure that the entire range is covered. | |
248 | */ | |
249 | error = VNOP_BLOCKMAP (vp, current_offset, remaining_length, | |
5ba3f43e | 250 | &io_blockno, &io_bytecount, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL); |
6d2010ae A |
251 | |
252 | if (error) { | |
253 | goto trim_exit; | |
254 | } | |
255 | /* | |
256 | * We have a contiguous run. Prepare & issue the ioctl for the device. | |
257 | * the DKIOCUNMAP ioctl takes offset in bytes from the start of the device. | |
258 | */ | |
259 | memset (&extent, 0, sizeof(dk_extent_t)); | |
260 | memset (&unmap, 0, sizeof(dk_unmap_t)); | |
261 | extent.offset = (uint64_t) io_blockno * (u_int64_t) blocksize; | |
262 | extent.length = io_bytecount; | |
263 | unmap.extents = &extent; | |
264 | unmap.extentsCount = 1; | |
265 | error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel()); | |
266 | ||
267 | if (error) { | |
268 | goto trim_exit; | |
269 | } | |
270 | remaining_length = remaining_length - io_bytecount; | |
271 | trimmed = trimmed + io_bytecount; | |
272 | current_offset = current_offset + io_bytecount; | |
273 | } | |
274 | trim_exit: | |
275 | ||
276 | return error; | |
277 | ||
278 | } | |
279 | ||
1c79356b A |
280 | pager_return_t |
281 | vnode_pageout(struct vnode *vp, | |
282 | upl_t upl, | |
b0d623f7 | 283 | upl_offset_t upl_offset, |
1c79356b | 284 | vm_object_offset_t f_offset, |
b0d623f7 | 285 | upl_size_t size, |
1c79356b A |
286 | int flags, |
287 | int *errorp) | |
288 | { | |
91447636 | 289 | int result = PAGER_SUCCESS; |
1c79356b | 290 | int error = 0; |
91447636 A |
291 | int error_ret = 0; |
292 | daddr64_t blkno; | |
293 | int isize; | |
1c79356b | 294 | int pg_index; |
91447636 | 295 | int base_index; |
b0d623f7 | 296 | upl_offset_t offset; |
1c79356b | 297 | upl_page_info_t *pl; |
2d21ac55 | 298 | vfs_context_t ctx = vfs_context_current(); /* pager context */ |
1c79356b | 299 | |
1c79356b A |
300 | isize = (int)size; |
301 | ||
9bccf70c | 302 | if (isize <= 0) { |
91447636 A |
303 | result = PAGER_ERROR; |
304 | error_ret = EINVAL; | |
9bccf70c A |
305 | goto out; |
306 | } | |
1c79356b | 307 | |
2d21ac55 | 308 | if (UBCINFOEXISTS(vp) == 0) { |
91447636 A |
309 | result = PAGER_ERROR; |
310 | error_ret = EINVAL; | |
9bccf70c | 311 | |
fa4905b1 | 312 | if (upl && !(flags & UPL_NOCOMMIT)) |
9bccf70c | 313 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY); |
1c79356b A |
314 | goto out; |
315 | } | |
91447636 | 316 | if ( !(flags & UPL_VNODE_PAGER)) { |
1c79356b | 317 | /* |
91447636 A |
318 | * This is a pageout from the default pager, |
319 | * just go ahead and call vnop_pageout since | |
320 | * it has already sorted out the dirty ranges | |
1c79356b | 321 | */ |
316670eb A |
322 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
323 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, | |
324 | size, 1, 0, 0, 0); | |
9bccf70c | 325 | |
91447636 | 326 | if ( (error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset, |
2d21ac55 | 327 | (size_t)size, flags, ctx)) ) |
91447636 | 328 | result = PAGER_ERROR; |
9bccf70c | 329 | |
316670eb A |
330 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
331 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, | |
332 | size, 1, 0, 0, 0); | |
9bccf70c | 333 | |
1c79356b A |
334 | goto out; |
335 | } | |
b0d623f7 A |
336 | if (upl == NULL) { |
337 | int request_flags; | |
338 | ||
339 | if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEOUTV2) { | |
340 | /* | |
341 | * filesystem has requested the new form of VNOP_PAGEOUT for file | |
342 | * backed objects... we will not grab the UPL befofe calling VNOP_PAGEOUT... | |
343 | * it is the fileystem's responsibility to grab the range we're denoting | |
344 | * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first | |
345 | * take any locks it needs, before effectively locking the pages into a UPL... | |
346 | */ | |
316670eb A |
347 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
348 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, | |
349 | size, (int)f_offset, 0, 0, 0); | |
b0d623f7 A |
350 | |
351 | if ( (error_ret = VNOP_PAGEOUT(vp, NULL, upl_offset, (off_t)f_offset, | |
352 | size, flags, ctx)) ) { | |
353 | result = PAGER_ERROR; | |
354 | } | |
316670eb A |
355 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
356 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, | |
357 | size, 0, 0, 0, 0); | |
b0d623f7 A |
358 | |
359 | goto out; | |
360 | } | |
361 | if (flags & UPL_MSYNC) | |
362 | request_flags = UPL_UBC_MSYNC | UPL_RET_ONLY_DIRTY; | |
363 | else | |
364 | request_flags = UPL_UBC_PAGEOUT | UPL_RET_ONLY_DIRTY; | |
365 | ||
5ba3f43e | 366 | if (ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, request_flags, VM_KERN_MEMORY_FILE) != KERN_SUCCESS) { |
b0d623f7 A |
367 | result = PAGER_ERROR; |
368 | error_ret = EINVAL; | |
369 | goto out; | |
370 | } | |
371 | upl_offset = 0; | |
372 | } else | |
373 | pl = ubc_upl_pageinfo(upl); | |
374 | ||
fe8ab488 A |
375 | /* |
376 | * Ignore any non-present pages at the end of the | |
377 | * UPL so that we aren't looking at a upl that | |
378 | * may already have been freed by the preceeding | |
379 | * aborts/completions. | |
380 | */ | |
381 | base_index = upl_offset / PAGE_SIZE; | |
382 | ||
383 | for (pg_index = (upl_offset + isize) / PAGE_SIZE; pg_index > base_index;) { | |
384 | if (upl_page_present(pl, --pg_index)) | |
385 | break; | |
386 | if (pg_index == base_index) { | |
387 | /* | |
388 | * no pages were returned, so release | |
389 | * our hold on the upl and leave | |
390 | */ | |
391 | if ( !(flags & UPL_NOCOMMIT)) | |
392 | ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY); | |
393 | ||
394 | goto out; | |
395 | } | |
396 | } | |
397 | isize = ((pg_index + 1) - base_index) * PAGE_SIZE; | |
398 | ||
9bccf70c | 399 | /* |
91447636 A |
400 | * we come here for pageouts to 'real' files and |
401 | * for msyncs... the upl may not contain any | |
402 | * dirty pages.. it's our responsibility to sort | |
403 | * through it and find the 'runs' of dirty pages | |
404 | * to call VNOP_PAGEOUT on... | |
9bccf70c | 405 | */ |
fe8ab488 | 406 | |
fa4905b1 | 407 | if (ubc_getsize(vp) == 0) { |
91447636 A |
408 | /* |
409 | * if the file has been effectively deleted, then | |
410 | * we need to go through the UPL and invalidate any | |
411 | * buffer headers we might have that reference any | |
412 | * of it's pages | |
413 | */ | |
414 | for (offset = upl_offset; isize; isize -= PAGE_SIZE, offset += PAGE_SIZE) { | |
415 | #if NFSCLIENT | |
416 | if (vp->v_tag == VT_NFS) | |
417 | /* check with nfs if page is OK to drop */ | |
418 | error = nfs_buf_page_inval(vp, (off_t)f_offset); | |
419 | else | |
420 | #endif | |
421 | { | |
422 | blkno = ubc_offtoblk(vp, (off_t)f_offset); | |
423 | error = buf_invalblkno(vp, blkno, 0); | |
424 | } | |
425 | if (error) { | |
426 | if ( !(flags & UPL_NOCOMMIT)) | |
427 | ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY); | |
428 | if (error_ret == 0) | |
429 | error_ret = error; | |
430 | result = PAGER_ERROR; | |
431 | ||
432 | } else if ( !(flags & UPL_NOCOMMIT)) { | |
433 | ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY); | |
fa4905b1 | 434 | } |
91447636 | 435 | f_offset += PAGE_SIZE; |
1c79356b | 436 | } |
1c79356b A |
437 | goto out; |
438 | } | |
91447636 A |
439 | |
440 | offset = upl_offset; | |
441 | pg_index = base_index; | |
1c79356b A |
442 | |
443 | while (isize) { | |
444 | int xsize; | |
445 | int num_of_pages; | |
446 | ||
91447636 A |
447 | if ( !upl_page_present(pl, pg_index)) { |
448 | /* | |
449 | * we asked for RET_ONLY_DIRTY, so it's possible | |
450 | * to get back empty slots in the UPL | |
451 | * just skip over them | |
452 | */ | |
2d21ac55 A |
453 | f_offset += PAGE_SIZE; |
454 | offset += PAGE_SIZE; | |
455 | isize -= PAGE_SIZE; | |
1c79356b A |
456 | pg_index++; |
457 | ||
458 | continue; | |
459 | } | |
460 | if ( !upl_dirty_page(pl, pg_index)) { | |
461 | /* | |
462 | * if the page is not dirty and reached here it is | |
463 | * marked precious or it is due to invalidation in | |
464 | * memory_object_lock request as part of truncation | |
465 | * We also get here from vm_object_terminate() | |
466 | * So all you need to do in these | |
467 | * cases is to invalidate incore buffer if it is there | |
91447636 | 468 | * Note we must not sleep here if the buffer is busy - that is |
fa4905b1 | 469 | * a lock inversion which causes deadlock. |
1c79356b | 470 | */ |
91447636 A |
471 | #if NFSCLIENT |
472 | if (vp->v_tag == VT_NFS) | |
55e303ae | 473 | /* check with nfs if page is OK to drop */ |
2d21ac55 | 474 | error = nfs_buf_page_inval(vp, (off_t)f_offset); |
91447636 A |
475 | else |
476 | #endif | |
477 | { | |
2d21ac55 | 478 | blkno = ubc_offtoblk(vp, (off_t)f_offset); |
91447636 A |
479 | error = buf_invalblkno(vp, blkno, 0); |
480 | } | |
481 | if (error) { | |
482 | if ( !(flags & UPL_NOCOMMIT)) | |
483 | ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY); | |
484 | if (error_ret == 0) | |
485 | error_ret = error; | |
486 | result = PAGER_ERROR; | |
487 | ||
488 | } else if ( !(flags & UPL_NOCOMMIT)) { | |
489 | ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY); | |
490 | } | |
2d21ac55 A |
491 | f_offset += PAGE_SIZE; |
492 | offset += PAGE_SIZE; | |
493 | isize -= PAGE_SIZE; | |
1c79356b A |
494 | pg_index++; |
495 | ||
496 | continue; | |
497 | } | |
1c79356b A |
498 | num_of_pages = 1; |
499 | xsize = isize - PAGE_SIZE; | |
500 | ||
501 | while (xsize) { | |
1c79356b A |
502 | if ( !upl_dirty_page(pl, pg_index + num_of_pages)) |
503 | break; | |
504 | num_of_pages++; | |
505 | xsize -= PAGE_SIZE; | |
506 | } | |
507 | xsize = num_of_pages * PAGE_SIZE; | |
508 | ||
316670eb A |
509 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
510 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, | |
511 | xsize, (int)f_offset, 0, 0, 0); | |
9bccf70c | 512 | |
b0d623f7 | 513 | if ( (error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset, |
2d21ac55 | 514 | xsize, flags, ctx)) ) { |
91447636 A |
515 | if (error_ret == 0) |
516 | error_ret = error; | |
517 | result = PAGER_ERROR; | |
518 | } | |
316670eb A |
519 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
520 | (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, | |
521 | xsize, 0, 0, 0, 0); | |
9bccf70c | 522 | |
2d21ac55 A |
523 | f_offset += xsize; |
524 | offset += xsize; | |
525 | isize -= xsize; | |
1c79356b A |
526 | pg_index += num_of_pages; |
527 | } | |
528 | out: | |
529 | if (errorp) | |
91447636 | 530 | *errorp = error_ret; |
1c79356b | 531 | |
91447636 | 532 | return (result); |
1c79356b A |
533 | } |
534 | ||
535 | ||
536 | pager_return_t | |
537 | vnode_pagein( | |
538 | struct vnode *vp, | |
9bccf70c | 539 | upl_t upl, |
b0d623f7 | 540 | upl_offset_t upl_offset, |
1c79356b | 541 | vm_object_offset_t f_offset, |
b0d623f7 | 542 | upl_size_t size, |
1c79356b A |
543 | int flags, |
544 | int *errorp) | |
545 | { | |
9bccf70c A |
546 | upl_page_info_t *pl; |
547 | int result = PAGER_SUCCESS; | |
1c79356b | 548 | int error = 0; |
9bccf70c A |
549 | int pages_in_upl; |
550 | int start_pg; | |
551 | int last_pg; | |
552 | int first_pg; | |
553 | int xsize; | |
2d21ac55 | 554 | int must_commit = 1; |
39236c6e | 555 | int ignore_valid_page_check = 0; |
1c79356b | 556 | |
2d21ac55 A |
557 | if (flags & UPL_NOCOMMIT) |
558 | must_commit = 0; | |
1c79356b | 559 | |
39236c6e A |
560 | if (flags & UPL_IGNORE_VALID_PAGE_CHECK) |
561 | ignore_valid_page_check = 1; | |
562 | ||
2d21ac55 | 563 | if (UBCINFOEXISTS(vp) == 0) { |
1c79356b A |
564 | result = PAGER_ERROR; |
565 | error = PAGER_ERROR; | |
2d21ac55 A |
566 | |
567 | if (upl && must_commit) | |
9bccf70c | 568 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR); |
2d21ac55 | 569 | |
1c79356b A |
570 | goto out; |
571 | } | |
9bccf70c | 572 | if (upl == (upl_t)NULL) { |
b0d623f7 | 573 | flags &= ~UPL_NOCOMMIT; |
2d21ac55 | 574 | |
fe8ab488 | 575 | if (size > MAX_UPL_SIZE_BYTES) { |
9bccf70c A |
576 | result = PAGER_ERROR; |
577 | error = PAGER_ERROR; | |
578 | goto out; | |
579 | } | |
b0d623f7 A |
580 | if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEINV2) { |
581 | /* | |
582 | * filesystem has requested the new form of VNOP_PAGEIN for file | |
583 | * backed objects... we will not grab the UPL befofe calling VNOP_PAGEIN... | |
584 | * it is the fileystem's responsibility to grab the range we're denoting | |
585 | * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first | |
586 | * take any locks it needs, before effectively locking the pages into a UPL... | |
587 | * so we pass a NULL into the filesystem instead of a UPL pointer... the 'upl_offset' | |
588 | * is used to identify the "must have" page in the extent... the filesystem is free | |
589 | * to clip the extent to better fit the underlying FS blocksize if it desires as | |
590 | * long as it continues to include the "must have" page... 'f_offset' + 'upl_offset' | |
591 | * identifies that page | |
592 | */ | |
593 | if ( (error = VNOP_PAGEIN(vp, NULL, upl_offset, (off_t)f_offset, | |
594 | size, flags, vfs_context_current())) ) { | |
595 | result = PAGER_ERROR; | |
596 | error = PAGER_ERROR; | |
597 | } | |
598 | goto out; | |
599 | } | |
5ba3f43e | 600 | ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT, VM_KERN_MEMORY_FILE); |
1c79356b | 601 | |
9bccf70c A |
602 | if (upl == (upl_t)NULL) { |
603 | result = PAGER_ABSENT; | |
604 | error = PAGER_ABSENT; | |
605 | goto out; | |
1c79356b | 606 | } |
316670eb A |
607 | ubc_upl_range_needed(upl, upl_offset / PAGE_SIZE, 1); |
608 | ||
9bccf70c | 609 | upl_offset = 0; |
2d21ac55 A |
610 | first_pg = 0; |
611 | ||
9bccf70c A |
612 | /* |
613 | * if we get here, we've created the upl and | |
614 | * are responsible for commiting/aborting it | |
615 | * regardless of what the caller has passed in | |
616 | */ | |
2d21ac55 | 617 | must_commit = 1; |
1c79356b | 618 | } else { |
9bccf70c | 619 | pl = ubc_upl_pageinfo(upl); |
2d21ac55 | 620 | first_pg = upl_offset / PAGE_SIZE; |
9bccf70c A |
621 | } |
622 | pages_in_upl = size / PAGE_SIZE; | |
2d21ac55 | 623 | DTRACE_VM2(pgpgin, int, pages_in_upl, (uint64_t *), NULL); |
9bccf70c A |
624 | |
625 | /* | |
626 | * before we start marching forward, we must make sure we end on | |
627 | * a present page, otherwise we will be working with a freed | |
628 | * upl | |
629 | */ | |
630 | for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) { | |
631 | if (upl_page_present(pl, last_pg)) | |
632 | break; | |
2d21ac55 A |
633 | if (last_pg == first_pg) { |
634 | /* | |
635 | * empty UPL, no pages are present | |
636 | */ | |
637 | if (must_commit) | |
638 | ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY); | |
639 | goto out; | |
640 | } | |
9bccf70c A |
641 | } |
642 | pages_in_upl = last_pg + 1; | |
2d21ac55 | 643 | last_pg = first_pg; |
9bccf70c | 644 | |
2d21ac55 | 645 | while (last_pg < pages_in_upl) { |
9bccf70c | 646 | /* |
2d21ac55 | 647 | * skip over missing pages... |
9bccf70c | 648 | */ |
2d21ac55 | 649 | for ( ; last_pg < pages_in_upl; last_pg++) { |
9bccf70c A |
650 | if (upl_page_present(pl, last_pg)) |
651 | break; | |
652 | } | |
39236c6e A |
653 | |
654 | if (ignore_valid_page_check == 1) { | |
655 | start_pg = last_pg; | |
656 | } else { | |
657 | /* | |
658 | * skip over 'valid' pages... we don't want to issue I/O for these | |
659 | */ | |
660 | for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) { | |
661 | if (!upl_valid_page(pl, last_pg)) | |
662 | break; | |
663 | } | |
9bccf70c | 664 | } |
39236c6e | 665 | |
9bccf70c A |
666 | if (last_pg > start_pg) { |
667 | /* | |
668 | * we've found a range of valid pages | |
669 | * if we've got COMMIT responsibility | |
670 | * commit this range of pages back to the | |
671 | * cache unchanged | |
672 | */ | |
673 | xsize = (last_pg - start_pg) * PAGE_SIZE; | |
1c79356b | 674 | |
2d21ac55 | 675 | if (must_commit) |
9bccf70c | 676 | ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY); |
9bccf70c A |
677 | } |
678 | if (last_pg == pages_in_upl) | |
2d21ac55 A |
679 | /* |
680 | * we're done... all pages that were present | |
681 | * have either had I/O issued on them or | |
682 | * were aborted unchanged... | |
683 | */ | |
9bccf70c A |
684 | break; |
685 | ||
2d21ac55 | 686 | if (!upl_page_present(pl, last_pg)) { |
9bccf70c | 687 | /* |
2d21ac55 A |
688 | * we found a range of valid pages |
689 | * terminated by a missing page... | |
690 | * bump index to the next page and continue on | |
9bccf70c | 691 | */ |
2d21ac55 | 692 | last_pg++; |
9bccf70c | 693 | continue; |
2d21ac55 | 694 | } |
9bccf70c A |
695 | /* |
696 | * scan from the found invalid page looking for a valid | |
697 | * or non-present page before the end of the upl is reached, if we | |
698 | * find one, then it will be the last page of the request to | |
699 | * 'cluster_io' | |
700 | */ | |
701 | for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) { | |
39236c6e | 702 | if (( !ignore_valid_page_check && upl_valid_page(pl, last_pg)) || !upl_page_present(pl, last_pg)) |
9bccf70c A |
703 | break; |
704 | } | |
705 | if (last_pg > start_pg) { | |
706 | int xoff; | |
9bccf70c A |
707 | xsize = (last_pg - start_pg) * PAGE_SIZE; |
708 | xoff = start_pg * PAGE_SIZE; | |
709 | ||
b0d623f7 | 710 | if ( (error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff, |
9bccf70c | 711 | (off_t)f_offset + xoff, |
2d21ac55 | 712 | xsize, flags, vfs_context_current())) ) { |
b0d623f7 A |
713 | /* |
714 | * Usually this UPL will be aborted/committed by the lower cluster layer. | |
6d2010ae A |
715 | * |
716 | * a) In the case of decmpfs, however, we may return an error (EAGAIN) to avoid | |
717 | * a deadlock with another thread already inflating the file. | |
718 | * | |
719 | * b) In the case of content protection, EPERM is a valid error and we should respect it. | |
720 | * | |
721 | * In those cases, we must take care of our UPL at this layer itself. | |
b0d623f7 A |
722 | */ |
723 | if (must_commit) { | |
724 | if(error == EAGAIN) { | |
725 | ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_RESTART); | |
726 | } | |
6d2010ae A |
727 | if(error == EPERM) { |
728 | ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR); | |
729 | } | |
b0d623f7 | 730 | } |
0b4e3aa0 A |
731 | result = PAGER_ERROR; |
732 | error = PAGER_ERROR; | |
9bccf70c | 733 | |
0b4e3aa0 | 734 | } |
1c79356b | 735 | } |
9bccf70c | 736 | } |
1c79356b A |
737 | out: |
738 | if (errorp) | |
fa4905b1 | 739 | *errorp = result; |
1c79356b A |
740 | |
741 | return (error); | |
742 | } | |
743 | ||
0b4e3aa0 | 744 | void * |
1c79356b A |
745 | upl_get_internal_page_list(upl_t upl) |
746 | { | |
0b4e3aa0 | 747 | return(UPL_GET_INTERNAL_PAGE_LIST(upl)); |
1c79356b A |
748 | |
749 | } |