ntdb: next-generation trivial key-value database
[ccan] / ccan / ntdb / test / run-57-die-during-transaction.c
1 #include "private.h"
2 #include <unistd.h>
3 #include "lock-tracking.h"
4 #include "tap-interface.h"
5 #include <stdlib.h>
6 #include <assert.h>
7 static ssize_t pwrite_check(int fd, const void *buf, size_t count, off_t offset);
8 static ssize_t write_check(int fd, const void *buf, size_t count);
9 static int ftruncate_check(int fd, off_t length);
10
11 #define pwrite pwrite_check
12 #define write write_check
13 #define fcntl fcntl_with_lockcheck
14 #define ftruncate ftruncate_check
15
16 /* There's a malloc inside transaction_setup_recovery, and valgrind complains
17  * when we longjmp and leak it. */
18 #define MAX_ALLOCATIONS 10
19 static void *allocated[MAX_ALLOCATIONS];
20 static unsigned max_alloc = 0;
21
22 static void *malloc_noleak(size_t len)
23 {
24         unsigned int i;
25
26         for (i = 0; i < MAX_ALLOCATIONS; i++)
27                 if (!allocated[i]) {
28                         allocated[i] = malloc(len);
29                         if (i > max_alloc) {
30                                 max_alloc = i;
31                                 diag("max_alloc: %i", max_alloc);
32                         }
33                         return allocated[i];
34                 }
35         diag("Too many allocations!");
36         abort();
37 }
38
39 static void *realloc_noleak(void *p, size_t size)
40 {
41         unsigned int i;
42
43         for (i = 0; i < MAX_ALLOCATIONS; i++) {
44                 if (allocated[i] == p) {
45                         if (i > max_alloc) {
46                                 max_alloc = i;
47                                 diag("max_alloc: %i", max_alloc);
48                         }
49                         return allocated[i] = realloc(p, size);
50                 }
51         }
52         diag("Untracked realloc!");
53         abort();
54 }
55
56 static void free_noleak(void *p)
57 {
58         unsigned int i;
59
60         /* We don't catch asprintf, so don't complain if we miss one. */
61         for (i = 0; i < MAX_ALLOCATIONS; i++) {
62                 if (allocated[i] == p) {
63                         allocated[i] = NULL;
64                         break;
65                 }
66         }
67         free(p);
68 }
69
70 static void free_all(void)
71 {
72         unsigned int i;
73
74         for (i = 0; i < MAX_ALLOCATIONS; i++) {
75                 free(allocated[i]);
76                 allocated[i] = NULL;
77         }
78 }
79
80 #define malloc malloc_noleak
81 #define free(x) free_noleak(x)
82 #define realloc realloc_noleak
83
84 #include "ntdb-source.h"
85
86 #undef malloc
87 #undef free
88 #undef realloc
89 #undef write
90 #undef pwrite
91 #undef fcntl
92 #undef ftruncate
93
94 #include <stdbool.h>
95 #include <stdarg.h>
96 #include <ccan/err/err.h>
97 #include <setjmp.h>
98 #include "external-agent.h"
99 #include "logging.h"
100
101 static bool in_transaction;
102 static int target, current;
103 static jmp_buf jmpbuf;
104 #define TEST_DBNAME "run-57-die-during-transaction.ntdb"
105 #define KEY_STRING "helloworld"
106 #define DATA_STRING "Helloworld"
107
108 static void maybe_die(int fd)
109 {
110         if (in_transaction && current++ == target) {
111                 longjmp(jmpbuf, 1);
112         }
113 }
114
115 static ssize_t pwrite_check(int fd,
116                             const void *buf, size_t count, off_t offset)
117 {
118         ssize_t ret;
119
120         maybe_die(fd);
121
122         ret = pwrite(fd, buf, count, offset);
123         if (ret != count)
124                 return ret;
125
126         maybe_die(fd);
127         return ret;
128 }
129
130 static ssize_t write_check(int fd, const void *buf, size_t count)
131 {
132         ssize_t ret;
133
134         maybe_die(fd);
135
136         ret = write(fd, buf, count);
137         if (ret != count)
138                 return ret;
139
140         maybe_die(fd);
141         return ret;
142 }
143
144 static int ftruncate_check(int fd, off_t length)
145 {
146         int ret;
147
148         maybe_die(fd);
149
150         ret = ftruncate(fd, length);
151
152         maybe_die(fd);
153         return ret;
154 }
155
156 static bool test_death(enum operation op, struct agent *agent,
157                        bool pre_create_recovery)
158 {
159         struct ntdb_context *ntdb = NULL;
160         NTDB_DATA key, data;
161         enum agent_return ret;
162         int needed_recovery = 0;
163
164         current = target = 0;
165         /* Big long data to force a change. */
166         data = ntdb_mkdata(DATA_STRING, strlen(DATA_STRING));
167
168 reset:
169         unlink(TEST_DBNAME);
170         ntdb = ntdb_open(TEST_DBNAME, NTDB_NOMMAP|MAYBE_NOSYNC,
171                          O_CREAT|O_TRUNC|O_RDWR, 0600, &tap_log_attr);
172         if (!ntdb) {
173                 diag("Failed opening NTDB: %s", strerror(errno));
174                 return false;
175         }
176
177         if (setjmp(jmpbuf) != 0) {
178                 /* We're partway through.  Simulate our death. */
179                 close(ntdb->file->fd);
180                 forget_locking();
181                 in_transaction = false;
182
183                 ret = external_agent_operation(agent, NEEDS_RECOVERY, "");
184                 if (ret == SUCCESS)
185                         needed_recovery++;
186                 else if (ret != FAILED) {
187                         diag("Step %u agent NEEDS_RECOVERY = %s", current,
188                              agent_return_name(ret));
189                         return false;
190                 }
191
192                 /* Could be key, or data. */
193                 ret = external_agent_operation(agent, op,
194                                                KEY_STRING "=" KEY_STRING);
195                 if (ret != SUCCESS) {
196                         ret = external_agent_operation(agent, op,
197                                                        KEY_STRING
198                                                        "=" DATA_STRING);
199                 }
200                 if (ret != SUCCESS) {
201                         diag("Step %u op %s failed = %s", current,
202                              operation_name(op),
203                              agent_return_name(ret));
204                         return false;
205                 }
206
207                 ret = external_agent_operation(agent, NEEDS_RECOVERY, "");
208                 if (ret != FAILED) {
209                         diag("Still needs recovery after step %u = %s",
210                              current, agent_return_name(ret));
211                         return false;
212                 }
213
214                 ret = external_agent_operation(agent, CHECK, "");
215                 if (ret != SUCCESS) {
216                         diag("Step %u check failed = %s", current,
217                              agent_return_name(ret));
218                         return false;
219                 }
220
221                 ret = external_agent_operation(agent, CLOSE, "");
222                 if (ret != SUCCESS) {
223                         diag("Step %u close failed = %s", current,
224                              agent_return_name(ret));
225                         return false;
226                 }
227
228                 /* Suppress logging as this tries to use closed fd. */
229                 suppress_logging = true;
230                 suppress_lockcheck = true;
231                 ntdb_close(ntdb);
232                 suppress_logging = false;
233                 suppress_lockcheck = false;
234                 target++;
235                 current = 0;
236                 free_all();
237                 goto reset;
238         }
239
240         /* Put key for agent to fetch. */
241         key = ntdb_mkdata(KEY_STRING, strlen(KEY_STRING));
242
243         if (pre_create_recovery) {
244                 /* Using a transaction now means we allocate the recovery
245                  * area immediately.  That makes the later transaction smaller
246                  * and thus tickles a bug we had. */
247                 if (ntdb_transaction_start(ntdb) != 0)
248                         return false;
249         }
250         if (ntdb_store(ntdb, key, key, NTDB_INSERT) != 0)
251                 return false;
252         if (pre_create_recovery) {
253                 if (ntdb_transaction_commit(ntdb) != 0)
254                         return false;
255         }
256
257         /* This is the key we insert in transaction. */
258         key.dsize--;
259
260         ret = external_agent_operation(agent, OPEN, TEST_DBNAME);
261         if (ret != SUCCESS)
262                 errx(1, "Agent failed to open: %s", agent_return_name(ret));
263
264         ret = external_agent_operation(agent, FETCH, KEY_STRING "=" KEY_STRING);
265         if (ret != SUCCESS)
266                 errx(1, "Agent failed find key: %s", agent_return_name(ret));
267
268         in_transaction = true;
269         if (ntdb_transaction_start(ntdb) != 0)
270                 return false;
271
272         if (ntdb_store(ntdb, key, data, NTDB_INSERT) != 0)
273                 return false;
274
275         if (ntdb_transaction_commit(ntdb) != 0)
276                 return false;
277
278         in_transaction = false;
279
280         /* We made it! */
281         diag("Completed %u runs", current);
282         ntdb_close(ntdb);
283         ret = external_agent_operation(agent, CLOSE, "");
284         if (ret != SUCCESS) {
285                 diag("Step %u close failed = %s", current,
286                      agent_return_name(ret));
287                 return false;
288         }
289
290         ok1(needed_recovery);
291         ok1(locking_errors == 0);
292         ok1(forget_locking() == 0);
293         locking_errors = 0;
294         return true;
295 }
296
297 int main(int argc, char *argv[])
298 {
299         enum operation ops[] = { FETCH, STORE, TRANSACTION_START };
300         struct agent *agent;
301         int i, j;
302
303         plan_tests(24);
304         unlock_callback = maybe_die;
305
306         external_agent_free = free_noleak;
307         agent = prepare_external_agent();
308         if (!agent)
309                 err(1, "preparing agent");
310
311         for (j = 0; j < 2; j++) {
312                 for (i = 0; i < sizeof(ops)/sizeof(ops[0]); i++) {
313                         diag("Testing %s after death (%s recovery area)",
314                              operation_name(ops[i]), j ? "with" : "without");
315                         ok1(test_death(ops[i], agent, j));
316                 }
317         }
318
319         free_external_agent(agent);
320         return exit_status();
321 }