]> git.saurik.com Git - apple/xnu.git/blob - osfmk/vm/vm_compressor_backing_store.c
projects / apple / xnu.git / blob
? search:


33f4f7cc4f39245f1b60c9d6a48e41c316978dd2
[apple/xnu.git] / osfmk / vm / vm_compressor_backing_store.c
1 /*
2 * Copyright (c) 2000-2013 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 #include "vm_compressor_backing_store.h"
30 #include <vm/vm_protos.h>
31
32 #include <IOKit/IOHibernatePrivate.h>
33
34
35 boolean_t compressor_store_stop_compaction = FALSE;
36 boolean_t vm_swap_up = FALSE;
37 boolean_t vm_swapfile_create_needed = FALSE;
38 boolean_t vm_swapfile_gc_needed = FALSE;
39
40 int swapper_throttle = -1;
41 boolean_t swapper_throttle_inited = FALSE;
42 uint64_t vm_swapout_thread_id;
43
44 uint64_t vm_swap_put_failures = 0;
45 uint64_t vm_swap_get_failures = 0;
46 int vm_num_swap_files = 0;
47 int vm_swapout_thread_processed_segments = 0;
48 int vm_swapout_thread_awakened = 0;
49 int vm_swapfile_create_thread_awakened = 0;
50 int vm_swapfile_create_thread_running = 0;
51 int vm_swapfile_gc_thread_awakened = 0;
52 int vm_swapfile_gc_thread_running = 0;
53
54 int64_t vm_swappin_avail = 0;
55 unsigned int vm_swapfile_total_segs_alloced = 0;
56 unsigned int vm_swapfile_total_segs_used = 0;
57
58
59 #define SWAP_READY 0x1 /* Swap file is ready to be used */
60 #define SWAP_RECLAIM 0x2 /* Swap file is marked to be reclaimed */
61 #define SWAP_WANTED 0x4 /* Swap file has waiters */
62 #define SWAP_REUSE 0x8 /* Swap file is on the Q and has a name. Reuse after init-ing.*/
63 #define SWAP_PINNED 0x10 /* Swap file is pinned (FusionDrive) */
64
65
66 struct swapfile{
67 queue_head_t swp_queue; /* list of swap files */
68 char *swp_path; /* saved pathname of swap file */
69 struct vnode *swp_vp; /* backing vnode */
70 uint64_t swp_size; /* size of this swap file */
71 uint8_t *swp_bitmap; /* bitmap showing the alloced/freed slots in the swap file */
72 unsigned int swp_pathlen; /* length of pathname */
73 unsigned int swp_nsegs; /* #segments we can use */
74 unsigned int swp_nseginuse; /* #segments in use */
75 unsigned int swp_index; /* index of this swap file */
76 unsigned int swp_flags; /* state of swap file */
77 unsigned int swp_free_hint; /* offset of 1st free chunk */
78 unsigned int swp_io_count; /* count of outstanding I/Os */
79 c_segment_t *swp_csegs; /* back pointers to the c_segments. Used during swap reclaim. */
80
81 struct trim_list *swp_delayed_trim_list_head;
82 unsigned int swp_delayed_trim_count;
83 };
84
85 queue_head_t swf_global_queue;
86 boolean_t swp_trim_supported = FALSE;
87
88 extern clock_sec_t dont_trim_until_ts;
89 clock_sec_t vm_swapfile_last_failed_to_create_ts = 0;
90 clock_sec_t vm_swapfile_last_successful_create_ts = 0;
91 int vm_swapfile_can_be_created = FALSE;
92 boolean_t delayed_trim_handling_in_progress = FALSE;
93
94 static void vm_swapout_thread_throttle_adjust(void);
95 static void vm_swap_free_now(struct swapfile *swf, uint64_t f_offset);
96 static void vm_swapout_thread(void);
97 static void vm_swapfile_create_thread(void);
98 static void vm_swapfile_gc_thread(void);
99 static void vm_swap_defragment();
100 static void vm_swap_handle_delayed_trims(boolean_t);
101 static void vm_swap_do_delayed_trim();
102 static void vm_swap_wait_on_trim_handling_in_progress(void);
103
104
105
106 #define VM_MAX_SWAP_FILE_NUM 100
107 #define VM_SWAPFILE_DELAYED_TRIM_MAX 128
108
109 #define VM_SWAP_SHOULD_DEFRAGMENT() (c_swappedout_sparse_count > (vm_swapfile_total_segs_used / 4) ? 1 : 0)
110 #define VM_SWAP_SHOULD_RECLAIM() (((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) >= SWAPFILE_RECLAIM_THRESHOLD_SEGS) ? 1 : 0)
111 #define VM_SWAP_SHOULD_ABORT_RECLAIM() (((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) <= SWAPFILE_RECLAIM_MINIMUM_SEGS) ? 1 : 0)
112 #define VM_SWAP_SHOULD_PIN(_size) (vm_swappin_avail > 0 && vm_swappin_avail >= (int64_t)(_size))
113 #define VM_SWAP_SHOULD_CREATE(cur_ts) ((vm_num_swap_files < VM_MAX_SWAP_FILE_NUM) && ((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) < (unsigned int)VM_SWAPFILE_HIWATER_SEGS) && \
114 ((cur_ts - vm_swapfile_last_failed_to_create_ts) > VM_SWAPFILE_DELAYED_CREATE) ? 1 : 0)
115 #define VM_SWAP_SHOULD_TRIM(swf) ((swf->swp_delayed_trim_count >= VM_SWAPFILE_DELAYED_TRIM_MAX) ? 1 : 0)
116
117
118 #define VM_SWAPFILE_DELAYED_CREATE 15
119
120 #define VM_SWAP_BUSY() ((c_swapout_count && (swapper_throttle == THROTTLE_LEVEL_COMPRESSOR_TIER1 || swapper_throttle == THROTTLE_LEVEL_COMPRESSOR_TIER0)) ? 1 : 0)
121
122
123 #if CHECKSUM_THE_SWAP
124 extern unsigned int hash_string(char *cp, int len);
125 #endif
126
127 #if RECORD_THE_COMPRESSED_DATA
128 boolean_t c_compressed_record_init_done = FALSE;
129 int c_compressed_record_write_error = 0;
130 struct vnode *c_compressed_record_vp = NULL;
131 uint64_t c_compressed_record_file_offset = 0;
132 void c_compressed_record_init(void);
133 void c_compressed_record_write(char *, int);
134 #endif
135
136 #if ENCRYPTED_SWAP
137 extern boolean_t swap_crypt_ctx_initialized;
138 extern void swap_crypt_ctx_initialize(void);
139 extern const unsigned char swap_crypt_null_iv[AES_BLOCK_SIZE];
140 extern aes_ctx swap_crypt_ctx;
141 extern unsigned long vm_page_encrypt_counter;
142 extern unsigned long vm_page_decrypt_counter;
143 #endif /* ENCRYPTED_SWAP */
144
145 extern void vm_pageout_io_throttle(void);
146 extern void vm_pageout_reinit_tuneables(void);
147 extern void vm_swap_file_set_tuneables(void);
148
149 struct swapfile *vm_swapfile_for_handle(uint64_t);
150
151 /*
152 * Called with the vm_swap_data_lock held.
153 */
154
155 struct swapfile *
156 vm_swapfile_for_handle(uint64_t f_offset)
157 {
158
159 uint64_t file_offset = 0;
160 unsigned int swapfile_index = 0;
161 struct swapfile* swf = NULL;
162
163 file_offset = (f_offset & SWAP_SLOT_MASK);
164 swapfile_index = (f_offset >> SWAP_DEVICE_SHIFT);
165
166 swf = (struct swapfile*) queue_first(&swf_global_queue);
167
168 while(queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
169
170 if (swapfile_index == swf->swp_index) {
171 break;
172 }
173
174 swf = (struct swapfile*) queue_next(&swf->swp_queue);
175 }
176
177 if (queue_end(&swf_global_queue, (queue_entry_t) swf)) {
178 swf = NULL;
179 }
180
181 return swf;
182 }
183
184 void
185 vm_compressor_swap_init()
186 {
187 thread_t thread = NULL;
188
189 lck_grp_attr_setdefault(&vm_swap_data_lock_grp_attr);
190 lck_grp_init(&vm_swap_data_lock_grp,
191 "vm_swap_data",
192 &vm_swap_data_lock_grp_attr);
193 lck_attr_setdefault(&vm_swap_data_lock_attr);
194 lck_mtx_init_ext(&vm_swap_data_lock,
195 &vm_swap_data_lock_ext,
196 &vm_swap_data_lock_grp,
197 &vm_swap_data_lock_attr);
198
199 queue_init(&swf_global_queue);
200
201
202 if (kernel_thread_start_priority((thread_continue_t)vm_swapout_thread, NULL,
203 BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
204 panic("vm_swapout_thread: create failed");
205 }
206 vm_swapout_thread_id = thread->thread_id;
207
208 thread_deallocate(thread);
209
210 if (kernel_thread_start_priority((thread_continue_t)vm_swapfile_create_thread, NULL,
211 BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
212 panic("vm_swapfile_create_thread: create failed");
213 }
214
215 thread_deallocate(thread);
216
217 if (kernel_thread_start_priority((thread_continue_t)vm_swapfile_gc_thread, NULL,
218 BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
219 panic("vm_swapfile_gc_thread: create failed");
220 }
221 thread_deallocate(thread);
222
223 proc_set_task_policy_thread(kernel_task, thread->thread_id,
224 TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
225 proc_set_task_policy_thread(kernel_task, thread->thread_id,
226 TASK_POLICY_INTERNAL, TASK_POLICY_PASSIVE_IO, TASK_POLICY_ENABLE);
227
228 #if ENCRYPTED_SWAP
229 if (swap_crypt_ctx_initialized == FALSE) {
230 swap_crypt_ctx_initialize();
231 }
232 #endif /* ENCRYPTED_SWAP */
233
234 memset(swapfilename, 0, MAX_SWAPFILENAME_LEN + 1);
235
236 vm_swap_up = TRUE;
237
238 printf("VM Swap Subsystem is %s\n", (vm_swap_up == TRUE) ? "ON" : "OFF");
239 }
240
241
242 #if RECORD_THE_COMPRESSED_DATA
243
244 void
245 c_compressed_record_init()
246 {
247 if (c_compressed_record_init_done == FALSE) {
248 vm_swapfile_open("/tmp/compressed_data", &c_compressed_record_vp);
249 c_compressed_record_init_done = TRUE;
250 }
251 }
252
253 void
254 c_compressed_record_write(char *buf, int size)
255 {
256 if (c_compressed_record_write_error == 0) {
257 c_compressed_record_write_error = vm_record_file_write(c_compressed_record_vp, c_compressed_record_file_offset, buf, size);
258 c_compressed_record_file_offset += size;
259 }
260 }
261 #endif
262
263
264
265 void
266 vm_swap_file_set_tuneables()
267 {
268 struct vnode *vp;
269 char *pathname;
270 int namelen;
271
272 if (strlen(swapfilename) == 0) {
273 /*
274 * If no swapfile name has been set, we'll
275 * use the default name.
276 *
277 * Also, this function is only called from the vm_pageout_scan thread
278 * via vm_consider_waking_compactor_swapper,
279 * so we don't need to worry about a race in checking/setting the name here.
280 */
281 strlcpy(swapfilename, SWAP_FILE_NAME, MAX_SWAPFILENAME_LEN);
282 }
283 namelen = (int)strlen(swapfilename) + SWAPFILENAME_INDEX_LEN + 1;
284 pathname = (char*)kalloc(namelen);
285 memset(pathname, 0, namelen);
286 snprintf(pathname, namelen, "%s%d", swapfilename, 0);
287
288 vm_swapfile_open(pathname, &vp);
289
290 if (vp == NULL)
291 goto done;
292
293 if (vnode_pager_isSSD(vp) == FALSE)
294 vm_pageout_reinit_tuneables();
295 vnode_setswapmount(vp);
296 vm_swappin_avail = vnode_getswappin_avail(vp);
297 vm_swapfile_close((uint64_t)pathname, vp);
298 done:
299 kfree(pathname, namelen);
300 }
301
302
303 #if ENCRYPTED_SWAP
304 void
305 vm_swap_encrypt(c_segment_t c_seg)
306 {
307 vm_offset_t kernel_vaddr = 0;
308 uint64_t size = 0;
309
310 union {
311 unsigned char aes_iv[AES_BLOCK_SIZE];
312 void *c_seg;
313 } encrypt_iv;
314
315 assert(swap_crypt_ctx_initialized);
316
317 bzero(&encrypt_iv.aes_iv[0], sizeof (encrypt_iv.aes_iv));
318
319 encrypt_iv.c_seg = (void*)c_seg;
320
321 /* encrypt the "initial vector" */
322 aes_encrypt_cbc((const unsigned char *) &encrypt_iv.aes_iv[0],
323 swap_crypt_null_iv,
324 1,
325 &encrypt_iv.aes_iv[0],
326 &swap_crypt_ctx.encrypt);
327
328 kernel_vaddr = (vm_offset_t) c_seg->c_store.c_buffer;
329 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
330
331 /*
332 * Encrypt the c_segment.
333 */
334 aes_encrypt_cbc((const unsigned char *) kernel_vaddr,
335 &encrypt_iv.aes_iv[0],
336 (unsigned int)(size / AES_BLOCK_SIZE),
337 (unsigned char *) kernel_vaddr,
338 &swap_crypt_ctx.encrypt);
339
340 vm_page_encrypt_counter += (size/PAGE_SIZE_64);
341 }
342
343 void
344 vm_swap_decrypt(c_segment_t c_seg)
345 {
346
347 vm_offset_t kernel_vaddr = 0;
348 uint64_t size = 0;
349
350 union {
351 unsigned char aes_iv[AES_BLOCK_SIZE];
352 void *c_seg;
353 } decrypt_iv;
354
355
356 assert(swap_crypt_ctx_initialized);
357
358 /*
359 * Prepare an "initial vector" for the decryption.
360 * It has to be the same as the "initial vector" we
361 * used to encrypt that page.
362 */
363 bzero(&decrypt_iv.aes_iv[0], sizeof (decrypt_iv.aes_iv));
364
365 decrypt_iv.c_seg = (void*)c_seg;
366
367 /* encrypt the "initial vector" */
368 aes_encrypt_cbc((const unsigned char *) &decrypt_iv.aes_iv[0],
369 swap_crypt_null_iv,
370 1,
371 &decrypt_iv.aes_iv[0],
372 &swap_crypt_ctx.encrypt);
373
374 kernel_vaddr = (vm_offset_t) c_seg->c_store.c_buffer;
375 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
376
377 /*
378 * Decrypt the c_segment.
379 */
380 aes_decrypt_cbc((const unsigned char *) kernel_vaddr,
381 &decrypt_iv.aes_iv[0],
382 (unsigned int) (size / AES_BLOCK_SIZE),
383 (unsigned char *) kernel_vaddr,
384 &swap_crypt_ctx.decrypt);
385
386 vm_page_decrypt_counter += (size/PAGE_SIZE_64);
387 }
388 #endif /* ENCRYPTED_SWAP */
389
390
391 void
392 vm_swap_consider_defragmenting()
393 {
394 if (compressor_store_stop_compaction == FALSE && !VM_SWAP_BUSY() &&
395 (VM_SWAP_SHOULD_DEFRAGMENT() || VM_SWAP_SHOULD_RECLAIM())) {
396
397 if (!vm_swapfile_gc_thread_running) {
398 lck_mtx_lock(&vm_swap_data_lock);
399
400 if (!vm_swapfile_gc_thread_running)
401 thread_wakeup((event_t) &vm_swapfile_gc_needed);
402
403 lck_mtx_unlock(&vm_swap_data_lock);
404 }
405 }
406 }
407
408
409 int vm_swap_defragment_yielded = 0;
410 int vm_swap_defragment_swapin = 0;
411 int vm_swap_defragment_free = 0;
412 int vm_swap_defragment_busy = 0;
413
414
415 static void
416 vm_swap_defragment()
417 {
418 c_segment_t c_seg;
419
420 /*
421 * have to grab the master lock w/o holding
422 * any locks in spin mode
423 */
424 PAGE_REPLACEMENT_DISALLOWED(TRUE);
425
426 lck_mtx_lock_spin_always(c_list_lock);
427
428 while (!queue_empty(&c_swappedout_sparse_list_head)) {
429
430 if (compressor_store_stop_compaction == TRUE || VM_SWAP_BUSY()) {
431 vm_swap_defragment_yielded++;
432 break;
433 }
434 c_seg = (c_segment_t)queue_first(&c_swappedout_sparse_list_head);
435
436 lck_mtx_lock_spin_always(&c_seg->c_lock);
437
438 assert(c_seg->c_state == C_ON_SWAPPEDOUTSPARSE_Q);
439
440 if (c_seg->c_busy) {
441 lck_mtx_unlock_always(c_list_lock);
442
443 PAGE_REPLACEMENT_DISALLOWED(FALSE);
444 /*
445 * c_seg_wait_on_busy consumes c_seg->c_lock
446 */
447 c_seg_wait_on_busy(c_seg);
448
449 PAGE_REPLACEMENT_DISALLOWED(TRUE);
450
451 lck_mtx_lock_spin_always(c_list_lock);
452
453 vm_swap_defragment_busy++;
454 continue;
455 }
456 if (c_seg->c_bytes_used == 0) {
457 /*
458 * c_seg_free_locked consumes the c_list_lock
459 * and c_seg->c_lock
460 */
461 C_SEG_BUSY(c_seg);
462 c_seg_free_locked(c_seg);
463
464 vm_swap_defragment_free++;
465 } else {
466 lck_mtx_unlock_always(c_list_lock);
467
468 c_seg_swapin(c_seg, TRUE);
469 lck_mtx_unlock_always(&c_seg->c_lock);
470
471 vm_swap_defragment_swapin++;
472 }
473 PAGE_REPLACEMENT_DISALLOWED(FALSE);
474
475 vm_pageout_io_throttle();
476
477 /*
478 * because write waiters have privilege over readers,
479 * dropping and immediately retaking the master lock will
480 * still allow any thread waiting to acquire the
481 * master lock exclusively an opportunity to take it
482 */
483 PAGE_REPLACEMENT_DISALLOWED(TRUE);
484
485 lck_mtx_lock_spin_always(c_list_lock);
486 }
487 lck_mtx_unlock_always(c_list_lock);
488
489 PAGE_REPLACEMENT_DISALLOWED(FALSE);
490 }
491
492
493
494 static void
495 vm_swapfile_create_thread(void)
496 {
497 clock_sec_t sec;
498 clock_nsec_t nsec;
499
500 current_thread()->options |= TH_OPT_VMPRIV;
501
502 vm_swapfile_create_thread_awakened++;
503 vm_swapfile_create_thread_running = 1;
504
505 while (TRUE) {
506 /*
507 * walk through the list of swap files
508 * and do the delayed frees/trims for
509 * any swap file whose count of delayed
510 * frees is above the batch limit
511 */
512 vm_swap_handle_delayed_trims(FALSE);
513
514 lck_mtx_lock(&vm_swap_data_lock);
515
516 clock_get_system_nanotime(&sec, &nsec);
517
518 if (VM_SWAP_SHOULD_CREATE(sec) == 0)
519 break;
520
521 lck_mtx_unlock(&vm_swap_data_lock);
522
523 if (vm_swap_create_file() == FALSE) {
524 vm_swapfile_last_failed_to_create_ts = sec;
525 HIBLOG("vm_swap_create_file failed @ %lu secs\n", (unsigned long)sec);
526
527 } else
528 vm_swapfile_last_successful_create_ts = sec;
529 }
530 vm_swapfile_create_thread_running = 0;
531
532 assert_wait((event_t)&vm_swapfile_create_needed, THREAD_UNINT);
533
534 lck_mtx_unlock(&vm_swap_data_lock);
535
536 thread_block((thread_continue_t)vm_swapfile_create_thread);
537
538 /* NOTREACHED */
539 }
540
541
542 static void
543 vm_swapfile_gc_thread(void)
544 {
545 boolean_t need_defragment;
546 boolean_t need_reclaim;
547
548 vm_swapfile_gc_thread_awakened++;
549 vm_swapfile_gc_thread_running = 1;
550
551 while (TRUE) {
552
553 lck_mtx_lock(&vm_swap_data_lock);
554
555 if (VM_SWAP_BUSY() || compressor_store_stop_compaction == TRUE)
556 break;
557
558 need_defragment = FALSE;
559 need_reclaim = FALSE;
560
561 if (VM_SWAP_SHOULD_DEFRAGMENT())
562 need_defragment = TRUE;
563
564 if (VM_SWAP_SHOULD_RECLAIM()) {
565 need_defragment = TRUE;
566 need_reclaim = TRUE;
567 }
568 if (need_defragment == FALSE && need_reclaim == FALSE)
569 break;
570
571 lck_mtx_unlock(&vm_swap_data_lock);
572
573 if (need_defragment == TRUE)
574 vm_swap_defragment();
575 if (need_reclaim == TRUE)
576 vm_swap_reclaim();
577 }
578 vm_swapfile_gc_thread_running = 0;
579
580 assert_wait((event_t)&vm_swapfile_gc_needed, THREAD_UNINT);
581
582 lck_mtx_unlock(&vm_swap_data_lock);
583
584 thread_block((thread_continue_t)vm_swapfile_gc_thread);
585
586 /* NOTREACHED */
587 }
588
589
590
591 int swapper_entered_T0 = 0;
592 int swapper_entered_T1 = 0;
593 int swapper_entered_T2 = 0;
594
595 static void
596 vm_swapout_thread_throttle_adjust(void)
597 {
598 int swapper_throttle_new;
599
600 if (swapper_throttle_inited == FALSE) {
601 /*
602 * force this thread to be set to the correct
603 * throttling tier
604 */
605 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
606 swapper_throttle = THROTTLE_LEVEL_COMPRESSOR_TIER1;
607 swapper_throttle_inited = TRUE;
608 swapper_entered_T2++;
609 goto done;
610 }
611 swapper_throttle_new = swapper_throttle;
612
613
614 switch(swapper_throttle) {
615
616 case THROTTLE_LEVEL_COMPRESSOR_TIER2:
617
618 if (SWAPPER_NEEDS_TO_UNTHROTTLE() || swapout_target_age || hibernate_flushing == TRUE) {
619 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER1;
620 swapper_entered_T1++;
621 break;
622 }
623 break;
624
625 case THROTTLE_LEVEL_COMPRESSOR_TIER1:
626
627 if (VM_PAGEOUT_SCAN_NEEDS_TO_THROTTLE()) {
628 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER0;
629 swapper_entered_T0++;
630 break;
631 }
632 if (COMPRESSOR_NEEDS_TO_SWAP() == 0 && swapout_target_age == 0 && hibernate_flushing == FALSE) {
633 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
634 swapper_entered_T2++;
635 break;
636 }
637 break;
638
639 case THROTTLE_LEVEL_COMPRESSOR_TIER0:
640
641 if (COMPRESSOR_NEEDS_TO_SWAP() == 0) {
642 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
643 swapper_entered_T2++;
644 break;
645 }
646 if (SWAPPER_NEEDS_TO_UNTHROTTLE() == 0) {
647 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER1;
648 swapper_entered_T1++;
649 break;
650 }
651 break;
652 }
653 done:
654 if (swapper_throttle != swapper_throttle_new) {
655 proc_set_task_policy_thread(kernel_task, vm_swapout_thread_id,
656 TASK_POLICY_INTERNAL, TASK_POLICY_IO, swapper_throttle_new);
657 proc_set_task_policy_thread(kernel_task, vm_swapout_thread_id,
658 TASK_POLICY_INTERNAL, TASK_POLICY_PASSIVE_IO, TASK_POLICY_ENABLE);
659
660 swapper_throttle = swapper_throttle_new;
661 }
662 }
663
664
665 int vm_swapout_found_empty = 0;
666
667
668 static void
669 vm_swapout_thread(void)
670 {
671 uint64_t f_offset = 0;
672 uint32_t size = 0;
673 c_segment_t c_seg = NULL;
674 kern_return_t kr = KERN_SUCCESS;
675 vm_offset_t addr = 0;
676
677 current_thread()->options |= TH_OPT_VMPRIV;
678
679 vm_swapout_thread_awakened++;
680
681 lck_mtx_lock_spin_always(c_list_lock);
682
683 while (!queue_empty(&c_swapout_list_head)) {
684
685 c_seg = (c_segment_t)queue_first(&c_swapout_list_head);
686
687 lck_mtx_lock_spin_always(&c_seg->c_lock);
688
689 assert(c_seg->c_state == C_ON_SWAPOUT_Q);
690
691 if (c_seg->c_busy) {
692 lck_mtx_unlock_always(c_list_lock);
693
694 c_seg_wait_on_busy(c_seg);
695
696 lck_mtx_lock_spin_always(c_list_lock);
697
698 continue;
699 }
700 vm_swapout_thread_processed_segments++;
701
702 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
703
704 if (size == 0) {
705 assert(c_seg->c_bytes_used == 0);
706
707 c_seg_switch_state(c_seg, C_IS_EMPTY, FALSE);
708 lck_mtx_unlock_always(&c_seg->c_lock);
709 lck_mtx_unlock_always(c_list_lock);
710
711 vm_swapout_found_empty++;
712 goto c_seg_is_empty;
713 }
714 C_SEG_BUSY(c_seg);
715 c_seg->c_busy_swapping = 1;
716
717 lck_mtx_unlock_always(c_list_lock);
718
719 addr = (vm_offset_t) c_seg->c_store.c_buffer;
720
721 lck_mtx_unlock_always(&c_seg->c_lock);
722
723 #if CHECKSUM_THE_SWAP
724 c_seg->cseg_hash = hash_string((char*)addr, (int)size);
725 c_seg->cseg_swap_size = size;
726 #endif /* CHECKSUM_THE_SWAP */
727
728 #if ENCRYPTED_SWAP
729 vm_swap_encrypt(c_seg);
730 #endif /* ENCRYPTED_SWAP */
731
732 vm_swapout_thread_throttle_adjust();
733
734 kr = vm_swap_put((vm_offset_t) addr, &f_offset, size, c_seg);
735
736 PAGE_REPLACEMENT_DISALLOWED(TRUE);
737
738 if (kr == KERN_SUCCESS) {
739 kernel_memory_depopulate(kernel_map, (vm_offset_t) addr, size, KMA_COMPRESSOR);
740 }
741 lck_mtx_lock_spin_always(c_list_lock);
742 lck_mtx_lock_spin_always(&c_seg->c_lock);
743
744 if (kr == KERN_SUCCESS) {
745 int new_state = C_ON_SWAPPEDOUT_Q;
746 boolean_t insert_head = FALSE;
747
748 if (hibernate_flushing == TRUE) {
749 if (c_seg->c_generation_id >= first_c_segment_to_warm_generation_id &&
750 c_seg->c_generation_id <= last_c_segment_to_warm_generation_id)
751 insert_head = TRUE;
752 } else if (C_SEG_ONDISK_IS_SPARSE(c_seg))
753 new_state = C_ON_SWAPPEDOUTSPARSE_Q;
754
755 c_seg_switch_state(c_seg, new_state, insert_head);
756
757 c_seg->c_store.c_swap_handle = f_offset;
758
759 VM_STAT_INCR_BY(swapouts, size >> PAGE_SHIFT);
760
761 if (c_seg->c_bytes_used)
762 OSAddAtomic64(-c_seg->c_bytes_used, &compressor_bytes_used);
763 } else {
764 #if ENCRYPTED_SWAP
765 vm_swap_decrypt(c_seg);
766 #endif /* ENCRYPTED_SWAP */
767 if (c_seg->c_overage_swap == TRUE) {
768 c_seg->c_overage_swap = FALSE;
769 c_overage_swapped_count--;
770 }
771 c_seg_switch_state(c_seg, C_ON_AGE_Q, FALSE);
772 }
773 lck_mtx_unlock_always(c_list_lock);
774
775 c_seg->c_busy_swapping = 0;
776 C_SEG_WAKEUP_DONE(c_seg);
777 lck_mtx_unlock_always(&c_seg->c_lock);
778
779 PAGE_REPLACEMENT_DISALLOWED(FALSE);
780
781 vm_pageout_io_throttle();
782 c_seg_is_empty:
783 if (c_swapout_count == 0)
784 vm_swap_consider_defragmenting();
785
786 lck_mtx_lock_spin_always(c_list_lock);
787 }
788
789 assert_wait((event_t)&c_swapout_list_head, THREAD_UNINT);
790
791 lck_mtx_unlock_always(c_list_lock);
792
793 thread_block((thread_continue_t)vm_swapout_thread);
794
795 /* NOTREACHED */
796 }
797
798 boolean_t
799 vm_swap_create_file()
800 {
801 uint64_t size = 0;
802 int namelen = 0;
803 boolean_t swap_file_created = FALSE;
804 boolean_t swap_file_reuse = FALSE;
805 boolean_t swap_file_pin = FALSE;
806 struct swapfile *swf = NULL;
807
808 /*
809 * Any swapfile structure ready for re-use?
810 */
811
812 lck_mtx_lock(&vm_swap_data_lock);
813
814 swf = (struct swapfile*) queue_first(&swf_global_queue);
815
816 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
817 if (swf->swp_flags == SWAP_REUSE) {
818 swap_file_reuse = TRUE;
819 break;
820 }
821 swf = (struct swapfile*) queue_next(&swf->swp_queue);
822 }
823
824 lck_mtx_unlock(&vm_swap_data_lock);
825
826 if (swap_file_reuse == FALSE) {
827
828 if (strlen(swapfilename) == 0) {
829 /*
830 * If no swapfile name has been set, we'll
831 * use the default name.
832 *
833 * Also, this function is only called from the swapfile management thread.
834 * So we don't need to worry about a race in checking/setting the name here.
835 */
836
837 strlcpy(swapfilename, SWAP_FILE_NAME, MAX_SWAPFILENAME_LEN);
838 }
839
840 namelen = (int)strlen(swapfilename) + SWAPFILENAME_INDEX_LEN + 1;
841
842 swf = (struct swapfile*) kalloc(sizeof *swf);
843 memset(swf, 0, sizeof(*swf));
844
845 swf->swp_index = vm_num_swap_files + 1;
846 swf->swp_pathlen = namelen;
847 swf->swp_path = (char*)kalloc(swf->swp_pathlen);
848
849 memset(swf->swp_path, 0, namelen);
850
851 snprintf(swf->swp_path, namelen, "%s%d", swapfilename, vm_num_swap_files);
852 }
853
854 vm_swapfile_open(swf->swp_path, &swf->swp_vp);
855
856 if (swf->swp_vp == NULL) {
857 if (swap_file_reuse == FALSE) {
858 kfree(swf->swp_path, swf->swp_pathlen);
859 kfree(swf, sizeof *swf);
860 }
861 return FALSE;
862 }
863 vm_swapfile_can_be_created = TRUE;
864
865 size = MAX_SWAP_FILE_SIZE;
866
867 while (size >= MIN_SWAP_FILE_SIZE) {
868
869 swap_file_pin = VM_SWAP_SHOULD_PIN(size);
870
871 if (vm_swapfile_preallocate(swf->swp_vp, &size, &swap_file_pin) == 0) {
872
873 int num_bytes_for_bitmap = 0;
874
875 swap_file_created = TRUE;
876
877 swf->swp_size = size;
878 swf->swp_nsegs = (unsigned int) (size / COMPRESSED_SWAP_CHUNK_SIZE);
879 swf->swp_nseginuse = 0;
880 swf->swp_free_hint = 0;
881
882 num_bytes_for_bitmap = MAX((swf->swp_nsegs >> 3) , 1);
883 /*
884 * Allocate a bitmap that describes the
885 * number of segments held by this swapfile.
886 */
887 swf->swp_bitmap = (uint8_t*)kalloc(num_bytes_for_bitmap);
888 memset(swf->swp_bitmap, 0, num_bytes_for_bitmap);
889
890 swf->swp_csegs = (c_segment_t *) kalloc(swf->swp_nsegs * sizeof(c_segment_t));
891 memset(swf->swp_csegs, 0, (swf->swp_nsegs * sizeof(c_segment_t)));
892
893 /*
894 * passing a NULL trim_list into vnode_trim_list
895 * will return ENOTSUP if trim isn't supported
896 * and 0 if it is
897 */
898 if (vnode_trim_list(swf->swp_vp, NULL, FALSE) == 0)
899 swp_trim_supported = TRUE;
900
901 lck_mtx_lock(&vm_swap_data_lock);
902
903 swf->swp_flags = SWAP_READY;
904
905 if (swap_file_reuse == FALSE) {
906 queue_enter(&swf_global_queue, swf, struct swapfile*, swp_queue);
907 }
908
909 vm_num_swap_files++;
910
911 vm_swapfile_total_segs_alloced += swf->swp_nsegs;
912
913 if (swap_file_pin == TRUE) {
914 swf->swp_flags |= SWAP_PINNED;
915 vm_swappin_avail -= swf->swp_size;
916 }
917
918 lck_mtx_unlock(&vm_swap_data_lock);
919
920 thread_wakeup((event_t) &vm_num_swap_files);
921 break;
922 } else {
923
924 size = size / 2;
925 }
926 }
927 if (swap_file_created == FALSE) {
928
929 vm_swapfile_close((uint64_t)(swf->swp_path), swf->swp_vp);
930
931 swf->swp_vp = NULL;
932
933 if (swap_file_reuse == FALSE) {
934 kfree(swf->swp_path, swf->swp_pathlen);
935 kfree(swf, sizeof *swf);
936 }
937 }
938 return swap_file_created;
939 }
940
941
942 kern_return_t
943 vm_swap_get(vm_offset_t addr, uint64_t f_offset, uint64_t size)
944 {
945 struct swapfile *swf = NULL;
946 uint64_t file_offset = 0;
947 int retval = 0;
948
949 if (addr == 0) {
950 return KERN_FAILURE;
951 }
952
953 lck_mtx_lock(&vm_swap_data_lock);
954
955 swf = vm_swapfile_for_handle(f_offset);
956
957 if (swf == NULL || ( !(swf->swp_flags & SWAP_READY) && !(swf->swp_flags & SWAP_RECLAIM))) {
958 retval = 1;
959 goto done;
960 }
961 swf->swp_io_count++;
962
963 lck_mtx_unlock(&vm_swap_data_lock);
964
965 file_offset = (f_offset & SWAP_SLOT_MASK);
966 retval = vm_swapfile_io(swf->swp_vp, file_offset, addr, (int)(size / PAGE_SIZE_64), SWAP_READ);
967
968 if (retval == 0)
969 VM_STAT_INCR_BY(swapins, size >> PAGE_SHIFT);
970 else
971 vm_swap_get_failures++;
972
973 /*
974 * Free this slot in the swap structure.
975 */
976 vm_swap_free(f_offset);
977
978 lck_mtx_lock(&vm_swap_data_lock);
979 swf->swp_io_count--;
980
981 if ((swf->swp_flags & SWAP_WANTED) && swf->swp_io_count == 0) {
982
983 swf->swp_flags &= ~SWAP_WANTED;
984 thread_wakeup((event_t) &swf->swp_flags);
985 }
986 done:
987 lck_mtx_unlock(&vm_swap_data_lock);
988
989 if (retval == 0)
990 return KERN_SUCCESS;
991 else
992 return KERN_FAILURE;
993 }
994
995 kern_return_t
996 vm_swap_put(vm_offset_t addr, uint64_t *f_offset, uint64_t size, c_segment_t c_seg)
997 {
998 unsigned int segidx = 0;
999 struct swapfile *swf = NULL;
1000 uint64_t file_offset = 0;
1001 uint64_t swapfile_index = 0;
1002 unsigned int byte_for_segidx = 0;
1003 unsigned int offset_within_byte = 0;
1004 boolean_t swf_eligible = FALSE;
1005 boolean_t waiting = FALSE;
1006 boolean_t retried = FALSE;
1007 int error = 0;
1008 clock_sec_t sec;
1009 clock_nsec_t nsec;
1010
1011 if (addr == 0 || f_offset == NULL) {
1012 return KERN_FAILURE;
1013 }
1014 retry:
1015 lck_mtx_lock(&vm_swap_data_lock);
1016
1017 swf = (struct swapfile*) queue_first(&swf_global_queue);
1018
1019 while(queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1020
1021 segidx = swf->swp_free_hint;
1022
1023 swf_eligible = (swf->swp_flags & SWAP_READY) && (swf->swp_nseginuse < swf->swp_nsegs);
1024
1025 if (swf_eligible) {
1026
1027 while(segidx < swf->swp_nsegs) {
1028
1029 byte_for_segidx = segidx >> 3;
1030 offset_within_byte = segidx % 8;
1031
1032 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1033 segidx++;
1034 continue;
1035 }
1036
1037 (swf->swp_bitmap)[byte_for_segidx] |= (1 << offset_within_byte);
1038
1039 file_offset = segidx * COMPRESSED_SWAP_CHUNK_SIZE;
1040 swf->swp_nseginuse++;
1041 swf->swp_io_count++;
1042 swapfile_index = swf->swp_index;
1043
1044 vm_swapfile_total_segs_used++;
1045
1046 clock_get_system_nanotime(&sec, &nsec);
1047
1048 if (VM_SWAP_SHOULD_CREATE(sec) && !vm_swapfile_create_thread_running)
1049 thread_wakeup((event_t) &vm_swapfile_create_needed);
1050
1051 lck_mtx_unlock(&vm_swap_data_lock);
1052
1053 goto done;
1054 }
1055 }
1056 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1057 }
1058 assert(queue_end(&swf_global_queue, (queue_entry_t) swf));
1059
1060 /*
1061 * we've run out of swap segments, but may not
1062 * be in a position to immediately create a new swap
1063 * file if we've recently failed to create due to a lack
1064 * of free space in the root filesystem... we'll try
1065 * to kick that create off, but in any event we're going
1066 * to take a breather (up to 1 second) so that we're not caught in a tight
1067 * loop back in "vm_compressor_compact_and_swap" trying to stuff
1068 * segments into swap files only to have them immediately put back
1069 * on the c_age queue due to vm_swap_put failing.
1070 *
1071 * if we're doing these puts due to a hibernation flush,
1072 * no need to block... setting hibernate_no_swapspace to TRUE,
1073 * will cause "vm_compressor_compact_and_swap" to immediately abort
1074 */
1075 clock_get_system_nanotime(&sec, &nsec);
1076
1077 if (VM_SWAP_SHOULD_CREATE(sec) && !vm_swapfile_create_thread_running)
1078 thread_wakeup((event_t) &vm_swapfile_create_needed);
1079
1080 if (hibernate_flushing == FALSE || VM_SWAP_SHOULD_CREATE(sec)) {
1081 waiting = TRUE;
1082 assert_wait_timeout((event_t) &vm_num_swap_files, THREAD_INTERRUPTIBLE, 1000, 1000*NSEC_PER_USEC);
1083 } else
1084 hibernate_no_swapspace = TRUE;
1085
1086 lck_mtx_unlock(&vm_swap_data_lock);
1087
1088 if (waiting == TRUE) {
1089 thread_block(THREAD_CONTINUE_NULL);
1090
1091 if (retried == FALSE && hibernate_flushing == TRUE) {
1092 retried = TRUE;
1093 goto retry;
1094 }
1095 }
1096 vm_swap_put_failures++;
1097
1098 return KERN_FAILURE;
1099
1100 done:
1101 error = vm_swapfile_io(swf->swp_vp, file_offset, addr, (int) (size / PAGE_SIZE_64), SWAP_WRITE);
1102
1103 lck_mtx_lock(&vm_swap_data_lock);
1104
1105 swf->swp_csegs[segidx] = c_seg;
1106
1107 swf->swp_io_count--;
1108
1109 *f_offset = (swapfile_index << SWAP_DEVICE_SHIFT) | file_offset;
1110
1111 if ((swf->swp_flags & SWAP_WANTED) && swf->swp_io_count == 0) {
1112
1113 swf->swp_flags &= ~SWAP_WANTED;
1114 thread_wakeup((event_t) &swf->swp_flags);
1115 }
1116
1117 lck_mtx_unlock(&vm_swap_data_lock);
1118
1119 if (error) {
1120 vm_swap_free(*f_offset);
1121
1122 vm_swap_put_failures++;
1123
1124 return KERN_FAILURE;
1125 }
1126 return KERN_SUCCESS;
1127 }
1128
1129
1130
1131 static void
1132 vm_swap_free_now(struct swapfile *swf, uint64_t f_offset)
1133 {
1134 uint64_t file_offset = 0;
1135 unsigned int segidx = 0;
1136
1137
1138 if ((swf->swp_flags & SWAP_READY) || (swf->swp_flags & SWAP_RECLAIM)) {
1139
1140 unsigned int byte_for_segidx = 0;
1141 unsigned int offset_within_byte = 0;
1142
1143 file_offset = (f_offset & SWAP_SLOT_MASK);
1144 segidx = (unsigned int) (file_offset / COMPRESSED_SWAP_CHUNK_SIZE);
1145
1146 byte_for_segidx = segidx >> 3;
1147 offset_within_byte = segidx % 8;
1148
1149 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1150
1151 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1152
1153 swf->swp_csegs[segidx] = NULL;
1154
1155 swf->swp_nseginuse--;
1156 vm_swapfile_total_segs_used--;
1157
1158 if (segidx < swf->swp_free_hint) {
1159 swf->swp_free_hint = segidx;
1160 }
1161 }
1162 if (VM_SWAP_SHOULD_RECLAIM() && !vm_swapfile_gc_thread_running)
1163 thread_wakeup((event_t) &vm_swapfile_gc_needed);
1164 }
1165 }
1166
1167
1168 uint32_t vm_swap_free_now_count = 0;
1169 uint32_t vm_swap_free_delayed_count = 0;
1170
1171
1172 void
1173 vm_swap_free(uint64_t f_offset)
1174 {
1175 struct swapfile *swf = NULL;
1176 struct trim_list *tl = NULL;
1177 clock_sec_t sec;
1178 clock_nsec_t nsec;
1179
1180 if (swp_trim_supported == TRUE)
1181 tl = kalloc(sizeof(struct trim_list));
1182
1183 lck_mtx_lock(&vm_swap_data_lock);
1184
1185 swf = vm_swapfile_for_handle(f_offset);
1186
1187 if (swf && (swf->swp_flags & (SWAP_READY | SWAP_RECLAIM))) {
1188
1189 if (swp_trim_supported == FALSE || (swf->swp_flags & SWAP_RECLAIM)) {
1190 /*
1191 * don't delay the free if the underlying disk doesn't support
1192 * trim, or we're in the midst of reclaiming this swap file since
1193 * we don't want to move segments that are technically free
1194 * but not yet handled by the delayed free mechanism
1195 */
1196 vm_swap_free_now(swf, f_offset);
1197
1198 vm_swap_free_now_count++;
1199 goto done;
1200 }
1201 tl->tl_offset = f_offset & SWAP_SLOT_MASK;
1202 tl->tl_length = COMPRESSED_SWAP_CHUNK_SIZE;
1203
1204 tl->tl_next = swf->swp_delayed_trim_list_head;
1205 swf->swp_delayed_trim_list_head = tl;
1206 swf->swp_delayed_trim_count++;
1207 tl = NULL;
1208
1209 if (VM_SWAP_SHOULD_TRIM(swf) && !vm_swapfile_create_thread_running) {
1210 clock_get_system_nanotime(&sec, &nsec);
1211
1212 if (sec > dont_trim_until_ts)
1213 thread_wakeup((event_t) &vm_swapfile_create_needed);
1214 }
1215 vm_swap_free_delayed_count++;
1216 }
1217 done:
1218 lck_mtx_unlock(&vm_swap_data_lock);
1219
1220 if (tl != NULL)
1221 kfree(tl, sizeof(struct trim_list));
1222 }
1223
1224
1225 static void
1226 vm_swap_wait_on_trim_handling_in_progress()
1227 {
1228 while (delayed_trim_handling_in_progress == TRUE) {
1229
1230 assert_wait((event_t) &delayed_trim_handling_in_progress, THREAD_UNINT);
1231 lck_mtx_unlock(&vm_swap_data_lock);
1232
1233 thread_block(THREAD_CONTINUE_NULL);
1234
1235 lck_mtx_lock(&vm_swap_data_lock);
1236 }
1237 }
1238
1239
1240 static void
1241 vm_swap_handle_delayed_trims(boolean_t force_now)
1242 {
1243 struct swapfile *swf = NULL;
1244
1245 /*
1246 * serialize the race between us and vm_swap_reclaim...
1247 * if vm_swap_reclaim wins it will turn off SWAP_READY
1248 * on the victim it has chosen... we can just skip over
1249 * that file since vm_swap_reclaim will first process
1250 * all of the delayed trims associated with it
1251 */
1252 lck_mtx_lock(&vm_swap_data_lock);
1253
1254 delayed_trim_handling_in_progress = TRUE;
1255
1256 lck_mtx_unlock(&vm_swap_data_lock);
1257
1258 /*
1259 * no need to hold the lock to walk the swf list since
1260 * vm_swap_create (the only place where we add to this list)
1261 * is run on the same thread as this function
1262 * and vm_swap_reclaim doesn't remove items from this list
1263 * instead marking them with SWAP_REUSE for future re-use
1264 */
1265 swf = (struct swapfile*) queue_first(&swf_global_queue);
1266
1267 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1268
1269 if ((swf->swp_flags & SWAP_READY) && (force_now == TRUE || VM_SWAP_SHOULD_TRIM(swf))) {
1270
1271 assert(!(swf->swp_flags & SWAP_RECLAIM));
1272 vm_swap_do_delayed_trim(swf);
1273 }
1274 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1275 }
1276 lck_mtx_lock(&vm_swap_data_lock);
1277
1278 delayed_trim_handling_in_progress = FALSE;
1279 thread_wakeup((event_t) &delayed_trim_handling_in_progress);
1280
1281 if (VM_SWAP_SHOULD_RECLAIM() && !vm_swapfile_gc_thread_running)
1282 thread_wakeup((event_t) &vm_swapfile_gc_needed);
1283
1284 lck_mtx_unlock(&vm_swap_data_lock);
1285
1286 }
1287
1288 static void
1289 vm_swap_do_delayed_trim(struct swapfile *swf)
1290 {
1291 struct trim_list *tl, *tl_head;
1292
1293 lck_mtx_lock(&vm_swap_data_lock);
1294
1295 tl_head = swf->swp_delayed_trim_list_head;
1296 swf->swp_delayed_trim_list_head = NULL;
1297 swf->swp_delayed_trim_count = 0;
1298
1299 lck_mtx_unlock(&vm_swap_data_lock);
1300
1301 vnode_trim_list(swf->swp_vp, tl_head, TRUE);
1302
1303 while ((tl = tl_head) != NULL) {
1304 unsigned int segidx = 0;
1305 unsigned int byte_for_segidx = 0;
1306 unsigned int offset_within_byte = 0;
1307
1308 lck_mtx_lock(&vm_swap_data_lock);
1309
1310 segidx = (unsigned int) (tl->tl_offset / COMPRESSED_SWAP_CHUNK_SIZE);
1311
1312 byte_for_segidx = segidx >> 3;
1313 offset_within_byte = segidx % 8;
1314
1315 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1316
1317 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1318
1319 swf->swp_csegs[segidx] = NULL;
1320
1321 swf->swp_nseginuse--;
1322 vm_swapfile_total_segs_used--;
1323
1324 if (segidx < swf->swp_free_hint) {
1325 swf->swp_free_hint = segidx;
1326 }
1327 }
1328 lck_mtx_unlock(&vm_swap_data_lock);
1329
1330 tl_head = tl->tl_next;
1331
1332 kfree(tl, sizeof(struct trim_list));
1333 }
1334 }
1335
1336
1337 void
1338 vm_swap_flush()
1339 {
1340 return;
1341 }
1342
1343 int vm_swap_reclaim_yielded = 0;
1344
1345 void
1346 vm_swap_reclaim(void)
1347 {
1348 vm_offset_t addr = 0;
1349 unsigned int segidx = 0;
1350 uint64_t f_offset = 0;
1351 struct swapfile *swf = NULL;
1352 struct swapfile *smallest_swf = NULL;
1353 unsigned int min_nsegs = 0;
1354 unsigned int byte_for_segidx = 0;
1355 unsigned int offset_within_byte = 0;
1356 uint32_t c_size = 0;
1357
1358 c_segment_t c_seg = NULL;
1359
1360 if (kernel_memory_allocate(kernel_map, (vm_offset_t *)(&addr), C_SEG_BUFSIZE, 0, KMA_KOBJECT, VM_KERN_MEMORY_COMPRESSOR) != KERN_SUCCESS) {
1361 panic("vm_swap_reclaim: kernel_memory_allocate failed\n");
1362 }
1363
1364 lck_mtx_lock(&vm_swap_data_lock);
1365
1366 /*
1367 * if we're running the swapfile list looking for
1368 * candidates with delayed trims, we need to
1369 * wait before making our decision concerning
1370 * the swapfile we want to reclaim
1371 */
1372 vm_swap_wait_on_trim_handling_in_progress();
1373
1374 /*
1375 * from here until we knock down the SWAP_READY bit,
1376 * we need to remain behind the vm_swap_data_lock...
1377 * once that bit has been turned off, "vm_swap_handle_delayed_trims"
1378 * will not consider this swapfile for processing
1379 */
1380 swf = (struct swapfile*) queue_first(&swf_global_queue);
1381 min_nsegs = MAX_SWAP_FILE_SIZE / COMPRESSED_SWAP_CHUNK_SIZE;
1382 smallest_swf = NULL;
1383
1384 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1385
1386 if ((swf->swp_flags & SWAP_READY) && (swf->swp_nseginuse <= min_nsegs)) {
1387
1388 smallest_swf = swf;
1389 min_nsegs = swf->swp_nseginuse;
1390 }
1391 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1392 }
1393
1394 if (smallest_swf == NULL)
1395 goto done;
1396
1397 swf = smallest_swf;
1398
1399
1400 swf->swp_flags &= ~SWAP_READY;
1401 swf->swp_flags |= SWAP_RECLAIM;
1402
1403 if (swf->swp_delayed_trim_count) {
1404
1405 lck_mtx_unlock(&vm_swap_data_lock);
1406
1407 vm_swap_do_delayed_trim(swf);
1408
1409 lck_mtx_lock(&vm_swap_data_lock);
1410 }
1411 segidx = 0;
1412
1413 while (segidx < swf->swp_nsegs) {
1414
1415 ReTry_for_cseg:
1416 /*
1417 * Wait for outgoing I/Os.
1418 */
1419 while (swf->swp_io_count) {
1420
1421 swf->swp_flags |= SWAP_WANTED;
1422
1423 assert_wait((event_t) &swf->swp_flags, THREAD_UNINT);
1424 lck_mtx_unlock(&vm_swap_data_lock);
1425
1426 thread_block(THREAD_CONTINUE_NULL);
1427
1428 lck_mtx_lock(&vm_swap_data_lock);
1429 }
1430 if (compressor_store_stop_compaction == TRUE || VM_SWAP_SHOULD_ABORT_RECLAIM() || VM_SWAP_BUSY()) {
1431 vm_swap_reclaim_yielded++;
1432 break;
1433 }
1434
1435 byte_for_segidx = segidx >> 3;
1436 offset_within_byte = segidx % 8;
1437
1438 if (((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) == 0) {
1439
1440 segidx++;
1441 continue;
1442 }
1443
1444 c_seg = swf->swp_csegs[segidx];
1445 assert(c_seg);
1446
1447 lck_mtx_lock_spin_always(&c_seg->c_lock);
1448
1449 if (c_seg->c_busy) {
1450 /*
1451 * a swapped out c_segment in the process of being freed will remain in the
1452 * busy state until after the vm_swap_free is called on it... vm_swap_free
1453 * takes the vm_swap_data_lock, so can't change the swap state until after
1454 * we drop the vm_swap_data_lock... once we do, vm_swap_free will complete
1455 * which will allow c_seg_free_locked to clear busy and wake up this thread...
1456 * at that point, we re-look up the swap state which will now indicate that
1457 * this c_segment no longer exists.
1458 */
1459 c_seg->c_wanted = 1;
1460
1461 assert_wait((event_t) (c_seg), THREAD_UNINT);
1462 lck_mtx_unlock_always(&c_seg->c_lock);
1463
1464 lck_mtx_unlock(&vm_swap_data_lock);
1465
1466 thread_block(THREAD_CONTINUE_NULL);
1467
1468 lck_mtx_lock(&vm_swap_data_lock);
1469
1470 goto ReTry_for_cseg;
1471 }
1472 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1473
1474 f_offset = segidx * COMPRESSED_SWAP_CHUNK_SIZE;
1475
1476 assert(c_seg == swf->swp_csegs[segidx]);
1477 swf->swp_csegs[segidx] = NULL;
1478 swf->swp_nseginuse--;
1479
1480 vm_swapfile_total_segs_used--;
1481
1482 lck_mtx_unlock(&vm_swap_data_lock);
1483
1484 assert(C_SEG_IS_ONDISK(c_seg));
1485
1486 C_SEG_BUSY(c_seg);
1487 c_seg->c_busy_swapping = 1;
1488 #if !CHECKSUM_THE_SWAP
1489 c_seg_trim_tail(c_seg);
1490 #endif
1491 c_size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
1492
1493 assert(c_size <= C_SEG_BUFSIZE && c_size);
1494
1495 lck_mtx_unlock_always(&c_seg->c_lock);
1496
1497 if (vm_swapfile_io(swf->swp_vp, f_offset, addr, (int)(c_size / PAGE_SIZE_64), SWAP_READ)) {
1498
1499 /*
1500 * reading the data back in failed, so convert c_seg
1501 * to a swapped in c_segment that contains no data
1502 */
1503 c_seg_swapin_requeue(c_seg, FALSE);
1504 /*
1505 * returns with c_busy_swapping cleared
1506 */
1507
1508 vm_swap_get_failures++;
1509 goto swap_io_failed;
1510 }
1511 VM_STAT_INCR_BY(swapins, c_size >> PAGE_SHIFT);
1512
1513 if (vm_swap_put(addr, &f_offset, c_size, c_seg)) {
1514 vm_offset_t c_buffer;
1515
1516 /*
1517 * the put failed, so convert c_seg to a fully swapped in c_segment
1518 * with valid data
1519 */
1520 c_buffer = (vm_offset_t)C_SEG_BUFFER_ADDRESS(c_seg->c_mysegno);
1521
1522 kernel_memory_populate(kernel_map, c_buffer, c_size, KMA_COMPRESSOR, VM_KERN_MEMORY_COMPRESSOR);
1523
1524 memcpy((char *)c_buffer, (char *)addr, c_size);
1525
1526 c_seg->c_store.c_buffer = (int32_t *)c_buffer;
1527 #if ENCRYPTED_SWAP
1528 vm_swap_decrypt(c_seg);
1529 #endif /* ENCRYPTED_SWAP */
1530 c_seg_swapin_requeue(c_seg, TRUE);
1531 /*
1532 * returns with c_busy_swapping cleared
1533 */
1534 OSAddAtomic64(c_seg->c_bytes_used, &compressor_bytes_used);
1535
1536 goto swap_io_failed;
1537 }
1538 VM_STAT_INCR_BY(swapouts, c_size >> PAGE_SHIFT);
1539
1540 lck_mtx_lock_spin_always(&c_seg->c_lock);
1541
1542 assert(C_SEG_IS_ONDISK(c_seg));
1543 /*
1544 * The c_seg will now know about the new location on disk.
1545 */
1546 c_seg->c_store.c_swap_handle = f_offset;
1547 c_seg->c_busy_swapping = 0;
1548 swap_io_failed:
1549 C_SEG_WAKEUP_DONE(c_seg);
1550
1551 lck_mtx_unlock_always(&c_seg->c_lock);
1552 lck_mtx_lock(&vm_swap_data_lock);
1553 }
1554
1555 if (swf->swp_nseginuse) {
1556
1557 swf->swp_flags &= ~SWAP_RECLAIM;
1558 swf->swp_flags |= SWAP_READY;
1559
1560 goto done;
1561 }
1562 /*
1563 * We don't remove this inactive swf from the queue.
1564 * That way, we can re-use it when needed again and
1565 * preserve the namespace. The delayed_trim processing
1566 * is also dependent on us not removing swfs from the queue.
1567 */
1568 //queue_remove(&swf_global_queue, swf, struct swapfile*, swp_queue);
1569
1570 vm_num_swap_files--;
1571
1572 vm_swapfile_total_segs_alloced -= swf->swp_nsegs;
1573
1574 lck_mtx_unlock(&vm_swap_data_lock);
1575
1576 vm_swapfile_close((uint64_t)(swf->swp_path), swf->swp_vp);
1577
1578 kfree(swf->swp_csegs, swf->swp_nsegs * sizeof(c_segment_t));
1579 kfree(swf->swp_bitmap, MAX((swf->swp_nsegs >> 3), 1));
1580
1581 lck_mtx_lock(&vm_swap_data_lock);
1582
1583 if (swf->swp_flags & SWAP_PINNED) {
1584 vm_swappin_avail += swf->swp_size;
1585 }
1586
1587 swf->swp_vp = NULL;
1588 swf->swp_size = 0;
1589 swf->swp_free_hint = 0;
1590 swf->swp_nsegs = 0;
1591 swf->swp_flags = SWAP_REUSE;
1592
1593 done:
1594 thread_wakeup((event_t) &swf->swp_flags);
1595 lck_mtx_unlock(&vm_swap_data_lock);
1596
1597 kmem_free(kernel_map, (vm_offset_t) addr, C_SEG_BUFSIZE);
1598 }
1599
1600
1601 uint64_t
1602 vm_swap_get_total_space(void)
1603 {
1604 uint64_t total_space = 0;
1605
1606 total_space = (uint64_t)vm_swapfile_total_segs_alloced * COMPRESSED_SWAP_CHUNK_SIZE;
1607
1608 return total_space;
1609 }
1610
1611 uint64_t
1612 vm_swap_get_used_space(void)
1613 {
1614 uint64_t used_space = 0;
1615
1616 used_space = (uint64_t)vm_swapfile_total_segs_used * COMPRESSED_SWAP_CHUNK_SIZE;
1617
1618 return used_space;
1619 }
1620
1621 uint64_t
1622 vm_swap_get_free_space(void)
1623 {
1624 return (vm_swap_get_total_space() - vm_swap_get_used_space());
1625 }
1626
1627
1628 int
1629 vm_swap_low_on_space(void)
1630 {
1631
1632 if (vm_num_swap_files == 0 && vm_swapfile_can_be_created == FALSE)
1633 return (0);
1634
1635 if (((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) < ((unsigned int)VM_SWAPFILE_HIWATER_SEGS) / 8)) {
1636
1637 if (vm_num_swap_files == 0 && !SWAPPER_NEEDS_TO_UNTHROTTLE())
1638 return (0);
1639
1640 if (vm_swapfile_last_failed_to_create_ts >= vm_swapfile_last_successful_create_ts)
1641 return (1);
1642 }
1643 return (0);
1644 }
mirror of XNU