summaryrefslogtreecommitdiff
path: root/src/util/disk_cache.c
blob: 5f68fee8d53a3ac7a6c678356463290c4187c1d7 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
/*
 * Copyright © 2014 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#ifdef ENABLE_SHADER_CACHE

#include <ctype.h>
#include <ftw.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <sys/file.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/statvfs.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
#include <pwd.h>
#include <errno.h>
#include <dirent.h>
#include "zlib.h"

#include "util/crc32.h"
#include "util/rand_xor.h"
#include "util/u_atomic.h"
#include "util/u_queue.h"
#include "util/mesa-sha1.h"
#include "util/ralloc.h"
#include "main/errors.h"
#include "util/macros.h"

#include "disk_cache.h"

/* Number of bits to mask off from a cache key to get an index. */
#define CACHE_INDEX_KEY_BITS 16

/* Mask for computing an index from a key. */
#define CACHE_INDEX_KEY_MASK ((1 << CACHE_INDEX_KEY_BITS) - 1)

/* The number of keys that can be stored in the index. */
#define CACHE_INDEX_MAX_KEYS (1 << CACHE_INDEX_KEY_BITS)

struct disk_cache {
   /* The path to the cache directory. */
   char *path;

   /* Thread queue for compressing and writing cache entries to disk */
   struct util_queue cache_queue;

   /* Seed for rand, which is used to pick a random directory */
   uint64_t seed_xorshift128plus[2];

   /* A pointer to the mmapped index file within the cache directory. */
   uint8_t *index_mmap;
   size_t index_mmap_size;

   /* Pointer to total size of all objects in cache (within index_mmap) */
   uint64_t *size;

   /* Pointer to stored keys, (within index_mmap). */
   uint8_t *stored_keys;

   /* Maximum size of all cached objects (in bytes). */
   uint64_t max_size;

   /* Driver cache keys. */
   uint8_t *driver_keys_blob;
   size_t driver_keys_blob_size;
};

struct disk_cache_put_job {
   struct util_queue_fence fence;

   struct disk_cache *cache;

   cache_key key;

   /* Copy of cache data to be compressed and written. */
   void *data;

   /* Size of data to be compressed and written. */
   size_t size;
};

/* Create a directory named 'path' if it does not already exist.
 *
 * Returns: 0 if path already exists as a directory or if created.
 *         -1 in all other cases.
 */
static int
mkdir_if_needed(const char *path)
{
   struct stat sb;

   /* If the path exists already, then our work is done if it's a
    * directory, but it's an error if it is not.
    */
   if (stat(path, &sb) == 0) {
      if (S_ISDIR(sb.st_mode)) {
         return 0;
      } else {
         fprintf(stderr, "Cannot use %s for shader cache (not a directory)"
                         "---disabling.\n", path);
         return -1;
      }
   }

   int ret = mkdir(path, 0755);
   if (ret == 0 || (ret == -1 && errno == EEXIST))
     return 0;

   fprintf(stderr, "Failed to create %s for shader cache (%s)---disabling.\n",
           path, strerror(errno));

   return -1;
}

/* Concatenate an existing path and a new name to form a new path.  If the new
 * path does not exist as a directory, create it then return the resulting
 * name of the new path (ralloc'ed off of 'ctx').
 *
 * Returns NULL on any error, such as:
 *
 *      <path> does not exist or is not a directory
 *      <path>/<name> exists but is not a directory
 *      <path>/<name> cannot be created as a directory
 */
static char *
concatenate_and_mkdir(void *ctx, const char *path, const char *name)
{
   char *new_path;
   struct stat sb;

   if (stat(path, &sb) != 0 || ! S_ISDIR(sb.st_mode))
      return NULL;

   new_path = ralloc_asprintf(ctx, "%s/%s", path, name);

   if (mkdir_if_needed(new_path) == 0)
      return new_path;
   else
      return NULL;
}

struct disk_cache *
disk_cache_create(const char *gpu_name, const char *timestamp)
{
   void *local;
   struct disk_cache *cache = NULL;
   char *path, *max_size_str;
   uint64_t max_size;
   int fd = -1;
   struct stat sb;
   struct statvfs vfs = { 0 };
   size_t size;

   /* If running as a users other than the real user disable cache */
   if (geteuid() != getuid())
      return NULL;

   /* A ralloc context for transient data during this invocation. */
   local = ralloc_context(NULL);
   if (local == NULL)
      goto fail;

   /* At user request, disable shader cache entirely. */
   if (getenv("MESA_GLSL_CACHE_DISABLE"))
      goto fail;

   /* Determine path for cache based on the first defined name as follows:
    *
    *   $MESA_GLSL_CACHE_DIR
    *   $XDG_CACHE_HOME/mesa
    *   <pwd.pw_dir>/.cache/mesa
    */
   path = getenv("MESA_GLSL_CACHE_DIR");
   if (path) {
      if (mkdir_if_needed(path) == -1)
         goto fail;

      path = concatenate_and_mkdir(local, path, "mesa");
      if (path == NULL)
         goto fail;
   }

   if (path == NULL) {
      char *xdg_cache_home = getenv("XDG_CACHE_HOME");

      if (xdg_cache_home) {
         if (mkdir_if_needed(xdg_cache_home) == -1)
            goto fail;

         path = concatenate_and_mkdir(local, xdg_cache_home, "mesa");
         if (path == NULL)
            goto fail;
      }
   }

   if (path == NULL) {
      char *buf;
      size_t buf_size;
      struct passwd pwd, *result;

      buf_size = sysconf(_SC_GETPW_R_SIZE_MAX);
      if (buf_size == -1)
         buf_size = 512;

      /* Loop until buf_size is large enough to query the directory */
      while (1) {
         buf = ralloc_size(local, buf_size);

         getpwuid_r(getuid(), &pwd, buf, buf_size, &result);
         if (result)
            break;

         if (errno == ERANGE) {
            ralloc_free(buf);
            buf = NULL;
            buf_size *= 2;
         } else {
            goto fail;
         }
      }

      path = concatenate_and_mkdir(local, pwd.pw_dir, ".cache");
      if (path == NULL)
         goto fail;

      path = concatenate_and_mkdir(local, path, "mesa");
      if (path == NULL)
         goto fail;
   }

   cache = ralloc(NULL, struct disk_cache);
   if (cache == NULL)
      goto fail;

   cache->path = ralloc_strdup(cache, path);
   if (cache->path == NULL)
      goto fail;

   path = ralloc_asprintf(local, "%s/index", cache->path);
   if (path == NULL)
      goto fail;

   fd = open(path, O_RDWR | O_CREAT | O_CLOEXEC, 0644);
   if (fd == -1)
      goto fail;

   if (fstat(fd, &sb) == -1)
      goto fail;

   /* Force the index file to be the expected size. */
   size = sizeof(*cache->size) + CACHE_INDEX_MAX_KEYS * CACHE_KEY_SIZE;
   if (sb.st_size != size) {
      if (ftruncate(fd, size) == -1)
         goto fail;
   }

   /* We map this shared so that other processes see updates that we
    * make.
    *
    * Note: We do use atomic addition to ensure that multiple
    * processes don't scramble the cache size recorded in the
    * index. But we don't use any locking to prevent multiple
    * processes from updating the same entry simultaneously. The idea
    * is that if either result lands entirely in the index, then
    * that's equivalent to a well-ordered write followed by an
    * eviction and a write. On the other hand, if the simultaneous
    * writes result in a corrupt entry, that's not really any
    * different than both entries being evicted, (since within the
    * guarantees of the cryptographic hash, a corrupt entry is
    * unlikely to ever match a real cache key).
    */
   cache->index_mmap = mmap(NULL, size, PROT_READ | PROT_WRITE,
                            MAP_SHARED, fd, 0);
   if (cache->index_mmap == MAP_FAILED)
      goto fail;
   cache->index_mmap_size = size;

   close(fd);

   cache->size = (uint64_t *) cache->index_mmap;
   cache->stored_keys = cache->index_mmap + sizeof(uint64_t);

   max_size = 0;

   max_size_str = getenv("MESA_GLSL_CACHE_MAX_SIZE");
   if (max_size_str) {
      char *end;
      max_size = strtoul(max_size_str, &end, 10);
      if (end == max_size_str) {
         max_size = 0;
      } else {
         switch (*end) {
         case 'K':
         case 'k':
            max_size *= 1024;
            break;
         case 'M':
         case 'm':
            max_size *= 1024*1024;
            break;
         case '\0':
         case 'G':
         case 'g':
         default:
            max_size *= 1024*1024*1024;
            break;
         }
      }
   }

   /* Default to 1GB or 10% of filesystem for maximum cache size. */
   if (max_size == 0) {
      statvfs(path, &vfs);
      max_size = MAX2(1024*1024*1024, vfs.f_blocks * vfs.f_bsize / 10);
   }

   cache->max_size = max_size;

   /* A limit of 32 jobs was choosen as observations of Deus Ex start-up times
    * showed that we reached at most 11 jobs on an Intel i5-6400 CPU@2.70GHz
    * (a fairly modest desktop CPU). 1 thread was chosen because we don't
    * really care about getting things to disk quickly just that it's not
    * blocking other tasks.
    */
   util_queue_init(&cache->cache_queue, "disk_cache", 32, 1);

   /* Create driver id keys */
   size_t ts_size = strlen(timestamp) + 1;
   size_t gpu_name_size = strlen(gpu_name) + 1;
   cache->driver_keys_blob_size = ts_size;
   cache->driver_keys_blob_size += gpu_name_size;

   /* We sometimes store entire structs that contains a pointers in the cache,
    * use pointer size as a key to avoid hard to debug issues.
    */
   uint8_t ptr_size = sizeof(void *);
   size_t ptr_size_size = sizeof(ptr_size);
   cache->driver_keys_blob_size += ptr_size_size;

   cache->driver_keys_blob =
      ralloc_size(cache, cache->driver_keys_blob_size);
   if (!cache->driver_keys_blob)
      goto fail;

   memcpy(cache->driver_keys_blob, timestamp, ts_size);
   memcpy(cache->driver_keys_blob + ts_size, gpu_name, gpu_name_size);
   memcpy(cache->driver_keys_blob + ts_size + gpu_name_size, &ptr_size,
          ptr_size_size);

   /* Seed our rand function */
   s_rand_xorshift128plus(cache->seed_xorshift128plus, true);

   ralloc_free(local);

   return cache;

 fail:
   if (fd != -1)
      close(fd);
   if (cache)
      ralloc_free(cache);
   ralloc_free(local);

   return NULL;
}

void
disk_cache_destroy(struct disk_cache *cache)
{
   if (cache) {
      util_queue_destroy(&cache->cache_queue);
      munmap(cache->index_mmap, cache->index_mmap_size);
   }

   ralloc_free(cache);
}

/* Return a filename within the cache's directory corresponding to 'key'. The
 * returned filename is ralloced with 'cache' as the parent context.
 *
 * Returns NULL if out of memory.
 */
static char *
get_cache_file(struct disk_cache *cache, const cache_key key)
{
   char buf[41];
   char *filename;

   _mesa_sha1_format(buf, key);
   if (asprintf(&filename, "%s/%c%c/%s", cache->path, buf[0],
                buf[1], buf + 2) == -1)
      return NULL;

   return filename;
}

/* Create the directory that will be needed for the cache file for \key.
 *
 * Obviously, the implementation here must closely match
 * _get_cache_file above.
*/
static void
make_cache_file_directory(struct disk_cache *cache, const cache_key key)
{
   char *dir;
   char buf[41];

   _mesa_sha1_format(buf, key);
   if (asprintf(&dir, "%s/%c%c", cache->path, buf[0], buf[1]) == -1)
      return;

   mkdir_if_needed(dir);
   free(dir);
}

/* Given a directory path and predicate function, find the entry with
 * the oldest access time in that directory for which the predicate
 * returns true.
 *
 * Returns: A malloc'ed string for the path to the chosen file, (or
 * NULL on any error). The caller should free the string when
 * finished.
 */
static char *
choose_lru_file_matching(const char *dir_path,
                         bool (*predicate)(const char *dir_path,
                                           const struct stat *,
                                           const char *, const size_t))
{
   DIR *dir;
   struct dirent *entry;
   char *filename;
   char *lru_name = NULL;
   time_t lru_atime = 0;

   dir = opendir(dir_path);
   if (dir == NULL)
      return NULL;

   while (1) {
      entry = readdir(dir);
      if (entry == NULL)
         break;

      struct stat sb;
      if (fstatat(dirfd(dir), entry->d_name, &sb, 0) == 0) {
         if (!lru_atime || (sb.st_atime < lru_atime)) {
            size_t len = strlen(entry->d_name);

            if (!predicate(dir_path, &sb, entry->d_name, len))
               continue;

            char *tmp = realloc(lru_name, len + 1);
            if (tmp) {
               lru_name = tmp;
               memcpy(lru_name, entry->d_name, len + 1);
               lru_atime = sb.st_atime;
            }
         }
      }
   }

   if (lru_name == NULL) {
      closedir(dir);
      return NULL;
   }

   if (asprintf(&filename, "%s/%s", dir_path, lru_name) < 0)
      filename = NULL;

   free(lru_name);
   closedir(dir);

   return filename;
}

/* Is entry a regular file, and not having a name with a trailing
 * ".tmp"
 */
static bool
is_regular_non_tmp_file(const char *path, const struct stat *sb,
                        const char *d_name, const size_t len)
{
   if (!S_ISREG(sb->st_mode))
      return false;

   if (len >= 4 && strcmp(&d_name[len-4], ".tmp") == 0)
      return false;

   return true;
}

/* Returns the size of the deleted file, (or 0 on any error). */
static size_t
unlink_lru_file_from_directory(const char *path)
{
   struct stat sb;
   char *filename;

   filename = choose_lru_file_matching(path, is_regular_non_tmp_file);
   if (filename == NULL)
      return 0;

   if (stat(filename, &sb) == -1) {
      free (filename);
      return 0;
   }

   unlink(filename);
   free (filename);

   return sb.st_size;
}

/* Is entry a directory with a two-character name, (and not the
 * special name of ".."). We also return false if the dir is empty.
 */
static bool
is_two_character_sub_directory(const char *path, const struct stat *sb,
                               const char *d_name, const size_t len)
{
   if (!S_ISDIR(sb->st_mode))
      return false;

   if (len != 2)
      return false;

   if (strcmp(d_name, "..") == 0)
      return false;

   char *subdir;
   if (asprintf(&subdir, "%s/%s", path, d_name) == -1)
      return false;
   DIR *dir = opendir(subdir);
   free(subdir);

   if (dir == NULL)
     return false;

   unsigned subdir_entries = 0;
   struct dirent *d;
   while ((d = readdir(dir)) != NULL) {
      if(++subdir_entries > 2)
         break;
   }
   closedir(dir);

   /* If dir only contains '.' and '..' it must be empty */
   if (subdir_entries <= 2)
      return false;

   return true;
}

static void
evict_lru_item(struct disk_cache *cache)
{
   char *dir_path;

   /* With a reasonably-sized, full cache, (and with keys generated
    * from a cryptographic hash), we can choose two random hex digits
    * and reasonably expect the directory to exist with a file in it.
    * Provides pseudo-LRU eviction to reduce checking all cache files.
    */
   uint64_t rand64 = rand_xorshift128plus(cache->seed_xorshift128plus);
   if (asprintf(&dir_path, "%s/%02" PRIx64 , cache->path, rand64 & 0xff) < 0)
      return;

   size_t size = unlink_lru_file_from_directory(dir_path);

   free(dir_path);

   if (size) {
      p_atomic_add(cache->size, - (uint64_t)size);
      return;
   }

   /* In the case where the random choice of directory didn't find
    * something, we choose the least recently accessed from the
    * existing directories.
    *
    * Really, the only reason this code exists is to allow the unit
    * tests to work, (which use an artificially-small cache to be able
    * to force a single cached item to be evicted).
    */
   dir_path = choose_lru_file_matching(cache->path,
                                       is_two_character_sub_directory);
   if (dir_path == NULL)
      return;

   size = unlink_lru_file_from_directory(dir_path);

   free(dir_path);

   if (size)
      p_atomic_add(cache->size, - (uint64_t)size);
}

void
disk_cache_remove(struct disk_cache *cache, const cache_key key)
{
   struct stat sb;

   char *filename = get_cache_file(cache, key);
   if (filename == NULL) {
      return;
   }

   if (stat(filename, &sb) == -1) {
      free(filename);
      return;
   }

   unlink(filename);
   free(filename);

   if (sb.st_size)
      p_atomic_add(cache->size, - (uint64_t)sb.st_size);
}

static ssize_t
write_all(int fd, const void *buf, size_t count)
{
   const char *out = buf;
   ssize_t written;
   size_t done;

   for (done = 0; done < count; done += written) {
      written = write(fd, out + done, count - done);
      if (written == -1)
         return -1;
   }
   return done;
}

/* From the zlib docs:
 *    "If the memory is available, buffers sizes on the order of 128K or 256K
 *    bytes should be used."
 */
#define BUFSIZE 256 * 1024

/**
 * Compresses cache entry in memory and writes it to disk. Returns the size
 * of the data written to disk.
 */
static size_t
deflate_and_write_to_disk(const void *in_data, size_t in_data_size, int dest,
                          const char *filename)
{
   unsigned char out[BUFSIZE];

   /* allocate deflate state */
   z_stream strm;
   strm.zalloc = Z_NULL;
   strm.zfree = Z_NULL;
   strm.opaque = Z_NULL;
   strm.next_in = (uint8_t *) in_data;
   strm.avail_in = in_data_size;

   int ret = deflateInit(&strm, Z_BEST_COMPRESSION);
   if (ret != Z_OK)
       return 0;

   /* compress until end of in_data */
   size_t compressed_size = 0;
   int flush;
   do {
      int remaining = in_data_size - BUFSIZE;
      flush = remaining > 0 ? Z_NO_FLUSH : Z_FINISH;
      in_data_size -= BUFSIZE;

      /* Run deflate() on input until the output buffer is not full (which
       * means there is no more data to deflate).
       */
      do {
         strm.avail_out = BUFSIZE;
         strm.next_out = out;

         ret = deflate(&strm, flush);    /* no bad return value */
         assert(ret != Z_STREAM_ERROR);  /* state not clobbered */

         size_t have = BUFSIZE - strm.avail_out;
         compressed_size += have;

         ssize_t written = write_all(dest, out, have);
         if (written == -1) {
            (void)deflateEnd(&strm);
            return 0;
         }
      } while (strm.avail_out == 0);

      /* all input should be used */
      assert(strm.avail_in == 0);

   } while (flush != Z_FINISH);

   /* stream should be complete */
   assert(ret == Z_STREAM_END);

   /* clean up and return */
   (void)deflateEnd(&strm);
   return compressed_size;
}

static struct disk_cache_put_job *
create_put_job(struct disk_cache *cache, const cache_key key,
               const void *data, size_t size)
{
   struct disk_cache_put_job *dc_job = (struct disk_cache_put_job *)
      malloc(sizeof(struct disk_cache_put_job) + size);

   if (dc_job) {
      dc_job->cache = cache;
      memcpy(dc_job->key, key, sizeof(cache_key));
      dc_job->data = dc_job + 1;
      memcpy(dc_job->data, data, size);
      dc_job->size = size;
   }

   return dc_job;
}

static void
destroy_put_job(void *job, int thread_index)
{
   if (job) {
      free(job);
   }
}

struct cache_entry_file_data {
   uint32_t crc32;
   uint32_t uncompressed_size;
};

static void
cache_put(void *job, int thread_index)
{
   assert(job);

   int fd = -1, fd_final = -1, err, ret;
   unsigned i = 0;
   char *filename = NULL, *filename_tmp = NULL;
   struct disk_cache_put_job *dc_job = (struct disk_cache_put_job *) job;

   filename = get_cache_file(dc_job->cache, dc_job->key);
   if (filename == NULL)
      goto done;

   /* If the cache is too large, evict something else first. */
   while (*dc_job->cache->size + dc_job->size > dc_job->cache->max_size &&
          i < 8) {
      evict_lru_item(dc_job->cache);
      i++;
   }

   /* Write to a temporary file to allow for an atomic rename to the
    * final destination filename, (to prevent any readers from seeing
    * a partially written file).
    */
   if (asprintf(&filename_tmp, "%s.tmp", filename) == -1)
      goto done;

   fd = open(filename_tmp, O_WRONLY | O_CLOEXEC | O_CREAT, 0644);

   /* Make the two-character subdirectory within the cache as needed. */
   if (fd == -1) {
      if (errno != ENOENT)
         goto done;

      make_cache_file_directory(dc_job->cache, dc_job->key);

      fd = open(filename_tmp, O_WRONLY | O_CLOEXEC | O_CREAT, 0644);
      if (fd == -1)
         goto done;
   }

   /* With the temporary file open, we take an exclusive flock on
    * it. If the flock fails, then another process still has the file
    * open with the flock held. So just let that file be responsible
    * for writing the file.
    */
   err = flock(fd, LOCK_EX | LOCK_NB);
   if (err == -1)
      goto done;

   /* Now that we have the lock on the open temporary file, we can
    * check to see if the destination file already exists. If so,
    * another process won the race between when we saw that the file
    * didn't exist and now. In this case, we don't do anything more,
    * (to ensure the size accounting of the cache doesn't get off).
    */
   fd_final = open(filename, O_RDONLY | O_CLOEXEC);
   if (fd_final != -1) {
      unlink(filename_tmp);
      goto done;
   }

   /* OK, we're now on the hook to write out a file that we know is
    * not in the cache, and is also not being written out to the cache
    * by some other process.
    *
    * Create CRC of the data and store at the start of the file. We will
    * read this when restoring the cache and use it to check for corruption.
    */
   struct cache_entry_file_data cf_data;
   cf_data.crc32 = util_hash_crc32(dc_job->data, dc_job->size);
   cf_data.uncompressed_size = dc_job->size;

   size_t cf_data_size = sizeof(cf_data);
   ret = write_all(fd, &cf_data, cf_data_size);
   if (ret == -1) {
      unlink(filename_tmp);
      goto done;
   }

   /* Now, finally, write out the contents to the temporary file, then
    * rename them atomically to the destination filename, and also
    * perform an atomic increment of the total cache size.
    */
   size_t file_size = deflate_and_write_to_disk(dc_job->data, dc_job->size,
                                                fd, filename_tmp);
   if (file_size == 0) {
      unlink(filename_tmp);
      goto done;
   }
   ret = rename(filename_tmp, filename);
   if (ret == -1) {
      unlink(filename_tmp);
      goto done;
   }

   file_size += cf_data_size;
   p_atomic_add(dc_job->cache->size, file_size);

 done:
   if (fd_final != -1)
      close(fd_final);
   /* This close finally releases the flock, (now that the final file
    * has been renamed into place and the size has been added).
    */
   if (fd != -1)
      close(fd);
   if (filename_tmp)
      free(filename_tmp);
   if (filename)
      free(filename);
}

void
disk_cache_put(struct disk_cache *cache, const cache_key key,
               const void *data, size_t size)
{
   struct disk_cache_put_job *dc_job =
      create_put_job(cache, key, data, size);

   if (dc_job) {
      util_queue_fence_init(&dc_job->fence);
      util_queue_add_job(&cache->cache_queue, dc_job, &dc_job->fence,
                         cache_put, destroy_put_job);
   }
}

/**
 * Decompresses cache entry, returns true if successful.
 */
static bool
inflate_cache_data(uint8_t *in_data, size_t in_data_size,
                   uint8_t *out_data, size_t out_data_size)
{
   z_stream strm;

   /* allocate inflate state */
   strm.zalloc = Z_NULL;
   strm.zfree = Z_NULL;
   strm.opaque = Z_NULL;
   strm.next_in = in_data;
   strm.avail_in = in_data_size;
   strm.next_out = out_data;
   strm.avail_out = out_data_size;

   int ret = inflateInit(&strm);
   if (ret != Z_OK)
      return false;

   ret = inflate(&strm, Z_NO_FLUSH);
   assert(ret != Z_STREAM_ERROR);  /* state not clobbered */

   /* Unless there was an error we should have decompressed everything in one
    * go as we know the uncompressed file size.
    */
   if (ret != Z_STREAM_END) {
      (void)inflateEnd(&strm);
      return false;
   }
   assert(strm.avail_out == 0);

   /* clean up and return */
   (void)inflateEnd(&strm);
   return true;
}

void *
disk_cache_get(struct disk_cache *cache, const cache_key key, size_t *size)
{
   int fd = -1, ret, len;
   struct stat sb;
   char *filename = NULL;
   uint8_t *data = NULL;
   uint8_t *uncompressed_data = NULL;

   if (size)
      *size = 0;

   filename = get_cache_file(cache, key);
   if (filename == NULL)
      goto fail;

   fd = open(filename, O_RDONLY | O_CLOEXEC);
   if (fd == -1)
      goto fail;

   if (fstat(fd, &sb) == -1)
      goto fail;

   data = malloc(sb.st_size);
   if (data == NULL)
      goto fail;

   /* Load the CRC that was created when the file was written. */
   struct cache_entry_file_data cf_data;
   size_t cf_data_size = sizeof(cf_data);
   assert(sb.st_size > cf_data_size);
   for (len = 0; len < cf_data_size; len += ret) {
      ret = read(fd, ((uint8_t *) &cf_data) + len, cf_data_size - len);
      if (ret == -1)
         goto fail;
   }

   /* Load the actual cache data. */
   size_t cache_data_size = sb.st_size - cf_data_size;
   for (len = 0; len < cache_data_size; len += ret) {
      ret = read(fd, data + len, cache_data_size - len);
      if (ret == -1)
         goto fail;
   }

   /* Uncompress the cache data */
   uncompressed_data = malloc(cf_data.uncompressed_size);
   if (!inflate_cache_data(data, cache_data_size, uncompressed_data,
                           cf_data.uncompressed_size))
      goto fail;

   /* Check the data for corruption */
   if (cf_data.crc32 != util_hash_crc32(uncompressed_data,
                                        cf_data.uncompressed_size))
      goto fail;

   free(data);
   free(filename);
   close(fd);

   if (size)
      *size = cf_data.uncompressed_size;

   return uncompressed_data;

 fail:
   if (data)
      free(data);
   if (uncompressed_data)
      free(uncompressed_data);
   if (filename)
      free(filename);
   if (fd != -1)
      close(fd);

   return NULL;
}

void
disk_cache_put_key(struct disk_cache *cache, const cache_key key)
{
   const uint32_t *key_chunk = (const uint32_t *) key;
   int i = *key_chunk & CACHE_INDEX_KEY_MASK;
   unsigned char *entry;

   entry = &cache->stored_keys[i * CACHE_KEY_SIZE];

   memcpy(entry, key, CACHE_KEY_SIZE);
}

/* This function lets us test whether a given key was previously
 * stored in the cache with disk_cache_put_key(). The implement is
 * efficient by not using syscalls or hitting the disk. It's not
 * race-free, but the races are benign. If we race with someone else
 * calling disk_cache_put_key, then that's just an extra cache miss and an
 * extra recompile.
 */
bool
disk_cache_has_key(struct disk_cache *cache, const cache_key key)
{
   const uint32_t *key_chunk = (const uint32_t *) key;
   int i = *key_chunk & CACHE_INDEX_KEY_MASK;
   unsigned char *entry;

   entry = &cache->stored_keys[i * CACHE_KEY_SIZE];

   return memcmp(entry, key, CACHE_KEY_SIZE) == 0;
}

void
disk_cache_compute_key(struct disk_cache *cache, const void *data, size_t size,
                       cache_key key)
{
   struct mesa_sha1 ctx;

   _mesa_sha1_init(&ctx);
   _mesa_sha1_update(&ctx, cache->driver_keys_blob,
                     cache->driver_keys_blob_size);
   _mesa_sha1_update(&ctx, data, size);
   _mesa_sha1_final(&ctx, key);
}

#endif /* ENABLE_SHADER_CACHE */