1 |
|
|
/** |
2 |
|
|
* This file is part of the CernVM File System. |
3 |
|
|
* |
4 |
|
|
* This tool checks a cvmfs cache directory for consistency. |
5 |
|
|
* If necessary, the managed cache db is removed so that |
6 |
|
|
* it will be rebuilt on next mount. |
7 |
|
|
*/ |
8 |
|
|
|
9 |
|
|
#define _FILE_OFFSET_BITS 64 |
10 |
|
|
|
11 |
|
|
#include "cvmfs_config.h" |
12 |
|
|
|
13 |
|
|
#include <dirent.h> |
14 |
|
|
#include <errno.h> |
15 |
|
|
#include <fcntl.h> |
16 |
|
|
#include <pthread.h> |
17 |
|
|
#include <stdint.h> |
18 |
|
|
#include <sys/stat.h> |
19 |
|
|
#include <unistd.h> |
20 |
|
|
|
21 |
|
|
#include <cstdio> |
22 |
|
|
#include <cstdlib> |
23 |
|
|
#include <cstring> |
24 |
|
|
#include <string> |
25 |
|
|
|
26 |
|
|
#include "atomic.h" |
27 |
|
|
#include "compression.h" |
28 |
|
|
#include "hash.h" |
29 |
|
|
#include "logging.h" |
30 |
|
|
#include "platform.h" |
31 |
|
|
#include "smalloc.h" |
32 |
|
|
#include "util/posix.h" |
33 |
|
|
|
34 |
|
|
using namespace std; // NOLINT |
35 |
|
|
|
36 |
|
|
enum Errors { |
37 |
|
|
kErrorOk = 0, |
38 |
|
|
kErrorFixed = 1, |
39 |
|
|
kErrorReboot = 2, |
40 |
|
|
kErrorUnfixed = 4, |
41 |
|
|
kErrorOperational = 8, |
42 |
|
|
kErrorUsage = 16, |
43 |
|
|
}; |
44 |
|
|
|
45 |
|
|
string *g_cache_dir; |
46 |
|
|
atomic_int32 g_num_files; |
47 |
|
|
atomic_int32 g_num_err_fixed; |
48 |
|
|
atomic_int32 g_num_err_unfixed; |
49 |
|
|
atomic_int32 g_num_err_operational; |
50 |
|
|
atomic_int32 g_num_tmp_catalog; |
51 |
|
|
/** |
52 |
|
|
* Traversal of the file system tree is serialized. |
53 |
|
|
*/ |
54 |
|
|
pthread_mutex_t g_lock_traverse = PTHREAD_MUTEX_INITIALIZER; |
55 |
|
|
DIR *g_DIRP_current = NULL; |
56 |
|
|
int g_num_dirs = -1; /**< Number of cache directories already examined. */ |
57 |
|
|
string *g_current_dir; /**< Current cache sub directory */ |
58 |
|
|
|
59 |
|
|
int g_num_threads = 1; |
60 |
|
|
bool g_fix_errors = false; |
61 |
|
|
bool g_verbose = false; |
62 |
|
|
atomic_int32 g_force_rebuild; |
63 |
|
|
atomic_int32 g_modified_cache; |
64 |
|
|
|
65 |
|
|
|
66 |
|
|
static void Usage() { |
67 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
68 |
|
|
"CernVM File System consistency checker, version %s\n\n" |
69 |
|
|
"This tool checks a cvmfs cache directory for consistency.\n" |
70 |
|
|
"If necessary, the managed cache db is removed so that\n" |
71 |
|
|
"it will be rebuilt on next mount.\n\n" |
72 |
|
|
"Usage: cvmfs_fsck [-v] [-p] [-f] [-j #threads] <cache directory>\n" |
73 |
|
|
"Options:\n" |
74 |
|
|
" -v verbose output\n" |
75 |
|
|
" -p try to fix automatically\n" |
76 |
|
|
" -f force rebuild of managed cache db on next mount\n" |
77 |
|
|
" -j number of concurrent integrity check worker threads\n", |
78 |
|
|
VERSION); |
79 |
|
|
} |
80 |
|
|
|
81 |
|
|
|
82 |
|
|
static bool GetNextFile(string *relative_path, string *hash_name) { |
83 |
|
|
platform_dirent64 *d = NULL; |
84 |
|
|
|
85 |
|
|
pthread_mutex_lock(&g_lock_traverse); |
86 |
|
|
get_next_file_again: |
87 |
|
|
while (g_DIRP_current && ((d = platform_readdir(g_DIRP_current)) != NULL)) { |
88 |
|
|
const string name = d->d_name; |
89 |
|
|
if ((name == ".") || (name == "..")) continue; |
90 |
|
|
|
91 |
|
|
platform_stat64 info; |
92 |
|
|
*relative_path = *g_current_dir + "/" + name; |
93 |
|
|
*hash_name = *g_current_dir + name; |
94 |
|
|
const string path = *g_cache_dir + "/" + *relative_path; |
95 |
|
|
if (platform_lstat(relative_path->c_str(), &info) != 0) { |
96 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Warning: failed to stat() %s (%d)", |
97 |
|
|
path.c_str(), errno); |
98 |
|
|
continue; |
99 |
|
|
} |
100 |
|
|
|
101 |
|
|
if (!S_ISREG(info.st_mode)) { |
102 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Warning: %s is not a regular file", |
103 |
|
|
path.c_str()); |
104 |
|
|
continue; |
105 |
|
|
} |
106 |
|
|
|
107 |
|
|
break; |
108 |
|
|
} |
109 |
|
|
|
110 |
|
|
if (!d) { |
111 |
|
|
if (g_DIRP_current) { |
112 |
|
|
closedir(g_DIRP_current); |
113 |
|
|
g_DIRP_current = NULL; |
114 |
|
|
} |
115 |
|
|
g_num_dirs++; |
116 |
|
|
if (g_num_dirs < 256) { |
117 |
|
|
char hex[3]; |
118 |
|
|
snprintf(hex, sizeof(hex), "%02x", g_num_dirs); |
119 |
|
|
*g_current_dir = string(hex, 2); |
120 |
|
|
|
121 |
|
|
if (g_verbose) |
122 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Entering %s", g_current_dir->c_str()); |
123 |
|
|
if ((g_DIRP_current = opendir(hex)) == NULL) { |
124 |
|
|
LogCvmfs(kLogCvmfs, kLogStderr, |
125 |
|
|
"Invalid cache directory, %s/%s does not exist", |
126 |
|
|
g_cache_dir->c_str(), g_current_dir->c_str()); |
127 |
|
|
pthread_mutex_unlock(&g_lock_traverse); |
128 |
|
|
exit(kErrorUnfixed); |
129 |
|
|
} |
130 |
|
|
goto get_next_file_again; |
131 |
|
|
} |
132 |
|
|
} |
133 |
|
|
pthread_mutex_unlock(&g_lock_traverse); |
134 |
|
|
|
135 |
|
|
if (d) |
136 |
|
|
return true; |
137 |
|
|
|
138 |
|
|
return false; |
139 |
|
|
} |
140 |
|
|
|
141 |
|
|
|
142 |
|
|
static void *MainCheck(void *data __attribute__((unused))) { |
143 |
|
|
string relative_path; |
144 |
|
|
string hash_name; |
145 |
|
|
|
146 |
|
|
while (GetNextFile(&relative_path, &hash_name)) { |
147 |
|
|
const string path = *g_cache_dir + "/" + relative_path; |
148 |
|
|
|
149 |
|
|
int n = atomic_xadd32(&g_num_files, 1); |
150 |
|
|
if (g_verbose) |
151 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Checking file %s", path.c_str()); |
152 |
|
|
if (!g_verbose && ((n % 1000) == 0)) |
153 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout | kLogNoLinebreak, "."); |
154 |
|
|
|
155 |
|
|
if (relative_path[relative_path.length()-1] == 'T') { |
156 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
157 |
|
|
"Warning: temporary file catalog %s found", path.c_str()); |
158 |
|
|
atomic_inc32(&g_num_tmp_catalog); |
159 |
|
|
if (g_fix_errors) { |
160 |
|
|
if (unlink(relative_path.c_str()) == 0) { |
161 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Fix: %s unlinked", path.c_str()); |
162 |
|
|
atomic_inc32(&g_num_err_fixed); |
163 |
|
|
} else { |
164 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Error: failed to unlink %s", |
165 |
|
|
path.c_str()); |
166 |
|
|
atomic_inc32(&g_num_err_unfixed); |
167 |
|
|
} |
168 |
|
|
} |
169 |
|
|
continue; |
170 |
|
|
} |
171 |
|
|
|
172 |
|
|
int fd_src = open(relative_path.c_str() , O_RDONLY); |
173 |
|
|
if (fd_src < 0) { |
174 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Error: cannot open %s", path.c_str()); |
175 |
|
|
atomic_inc32(&g_num_err_operational); |
176 |
|
|
continue; |
177 |
|
|
} |
178 |
|
|
// Don't thrash kernel buffers |
179 |
|
|
platform_disable_kcache(fd_src); |
180 |
|
|
|
181 |
|
|
// Compress every file and calculate SHA-1 of stream |
182 |
|
|
shash::Any expected_hash = shash::MkFromHexPtr(shash::HexPtr(hash_name)); |
183 |
|
|
shash::Any hash(expected_hash.algorithm); |
184 |
|
|
if (!zlib::CompressFd2Null(fd_src, &hash)) { |
185 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Error: could not compress %s", |
186 |
|
|
path.c_str()); |
187 |
|
|
atomic_inc32(&g_num_err_operational); |
188 |
|
|
} else { |
189 |
|
|
if (hash != expected_hash) { |
190 |
|
|
// If the hashes don't match, try hashing the uncompressed file |
191 |
|
|
if (!shash::HashFile(relative_path, &hash)) { |
192 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Error: could not hash %s", |
193 |
|
|
path.c_str()); |
194 |
|
|
atomic_inc32(&g_num_err_operational); |
195 |
|
|
} |
196 |
|
|
if (hash != expected_hash) { |
197 |
|
|
if (g_fix_errors) { |
198 |
|
|
const string quarantaine_path = "./quarantaine/" + hash_name; |
199 |
|
|
bool fixed = false; |
200 |
|
|
if (rename(relative_path.c_str(), quarantaine_path.c_str()) == 0) { |
201 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
202 |
|
|
"Fix: %s is corrupted, moved to quarantaine folder", |
203 |
|
|
path.c_str()); |
204 |
|
|
fixed = true; |
205 |
|
|
} else { |
206 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
207 |
|
|
"Warning: failed to move %s into quarantaine folder", |
208 |
|
|
path.c_str()); |
209 |
|
|
if (unlink(relative_path.c_str()) == 0) { |
210 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
211 |
|
|
"Fix: %s is corrupted, file unlinked", path.c_str()); |
212 |
|
|
fixed = true; |
213 |
|
|
} else { |
214 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
215 |
|
|
"Error: %s is corrupted, could not unlink", |
216 |
|
|
path.c_str()); |
217 |
|
|
} |
218 |
|
|
} |
219 |
|
|
|
220 |
|
|
if (fixed) { |
221 |
|
|
atomic_inc32(&g_num_err_fixed); |
222 |
|
|
|
223 |
|
|
// Changes made, we have to rebuild the managed cache db |
224 |
|
|
atomic_cas32(&g_force_rebuild, 0, 1); |
225 |
|
|
atomic_cas32(&g_modified_cache, 0, 1); |
226 |
|
|
} else { |
227 |
|
|
atomic_inc32(&g_num_err_unfixed); |
228 |
|
|
} |
229 |
|
|
} else { |
230 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
231 |
|
|
"Error: %s has compressed checksum %s, " |
232 |
|
|
"delete this file from cache directory!", |
233 |
|
|
path.c_str(), hash.ToString().c_str()); |
234 |
|
|
atomic_inc32(&g_num_err_unfixed); |
235 |
|
|
} |
236 |
|
|
} |
237 |
|
|
} |
238 |
|
|
} |
239 |
|
|
close(fd_src); |
240 |
|
|
} |
241 |
|
|
|
242 |
|
|
return NULL; |
243 |
|
|
} |
244 |
|
|
|
245 |
|
|
|
246 |
|
|
int main(int argc, char **argv) { |
247 |
|
|
atomic_init32(&g_force_rebuild); |
248 |
|
|
atomic_init32(&g_modified_cache); |
249 |
|
|
g_current_dir = new string(); |
250 |
|
|
|
251 |
|
|
int c; |
252 |
|
|
while ((c = getopt(argc, argv, "hvpfj:")) != -1) { |
253 |
|
|
switch (c) { |
254 |
|
|
case 'h': |
255 |
|
|
Usage(); |
256 |
|
|
return kErrorOk; |
257 |
|
|
case 'v': |
258 |
|
|
g_verbose = true; |
259 |
|
|
break; |
260 |
|
|
case 'p': |
261 |
|
|
g_fix_errors = true; |
262 |
|
|
break; |
263 |
|
|
case 'f': |
264 |
|
|
atomic_cas32(&g_force_rebuild, 0, 1); |
265 |
|
|
break; |
266 |
|
|
case 'j': |
267 |
|
|
g_num_threads = atoi(optarg); |
268 |
|
|
if (g_num_threads < 1) { |
269 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
270 |
|
|
"There is at least one worker thread required"); |
271 |
|
|
return kErrorUsage; |
272 |
|
|
} |
273 |
|
|
break; |
274 |
|
|
case '?': |
275 |
|
|
default: |
276 |
|
|
Usage(); |
277 |
|
|
return kErrorUsage; |
278 |
|
|
} |
279 |
|
|
} |
280 |
|
|
|
281 |
|
|
// Switch to cache directory |
282 |
|
|
if (optind >= argc) { |
283 |
|
|
Usage(); |
284 |
|
|
return kErrorUsage; |
285 |
|
|
} |
286 |
|
|
g_cache_dir = new string(MakeCanonicalPath(argv[optind])); |
287 |
|
|
if (chdir(g_cache_dir->c_str()) != 0) { |
288 |
|
|
LogCvmfs(kLogCvmfs, kLogStderr, "Could not chdir to %s", |
289 |
|
|
g_cache_dir->c_str()); |
290 |
|
|
return kErrorOperational; |
291 |
|
|
} |
292 |
|
|
|
293 |
|
|
// Check if txn directory is empty |
294 |
|
|
DIR *dirp_txn; |
295 |
|
|
if ((dirp_txn = opendir("txn")) == NULL) { |
296 |
|
|
LogCvmfs(kLogCvmfs, kLogStderr, |
297 |
|
|
"Invalid cache directory, %s/txn does not exist", |
298 |
|
|
g_cache_dir->c_str()); |
299 |
|
|
return kErrorOperational; |
300 |
|
|
} |
301 |
|
|
platform_dirent64 *d; |
302 |
|
|
while ((d = platform_readdir(dirp_txn)) != NULL) { |
303 |
|
|
const string name = d->d_name; |
304 |
|
|
if ((name == ".") || (name == "..")) continue; |
305 |
|
|
|
306 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
307 |
|
|
"Warning: temporary directory %s/txn is not empty\n" |
308 |
|
|
"If this repository is currently _not_ mounted, " |
309 |
|
|
"you can remove its contents", g_cache_dir->c_str()); |
310 |
|
|
break; |
311 |
|
|
} |
312 |
|
|
closedir(dirp_txn); |
313 |
|
|
|
314 |
|
|
// Run workers to recalculate checksums |
315 |
|
|
atomic_init32(&g_num_files); |
316 |
|
|
atomic_init32(&g_num_err_fixed); |
317 |
|
|
atomic_init32(&g_num_err_unfixed); |
318 |
|
|
atomic_init32(&g_num_err_operational); |
319 |
|
|
atomic_init32(&g_num_tmp_catalog); |
320 |
|
|
pthread_t *workers = reinterpret_cast<pthread_t *>( |
321 |
|
|
smalloc(g_num_threads * sizeof(pthread_t))); |
322 |
|
|
if (!g_verbose) |
323 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout | kLogNoLinebreak, "Verifying: "); |
324 |
|
|
for (int i = 0; i < g_num_threads; ++i) { |
325 |
|
|
if (g_verbose) |
326 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Starting worker %d", i+1); |
327 |
|
|
if (pthread_create(&workers[i], NULL, MainCheck, NULL) != 0) { |
328 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Fatal: could not create worker thread"); |
329 |
|
|
return kErrorOperational; |
330 |
|
|
} |
331 |
|
|
} |
332 |
|
|
for (int i = g_num_threads-1; i >= 0; --i) { |
333 |
|
|
pthread_join(workers[i], NULL); |
334 |
|
|
if (g_verbose) |
335 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Stopping worker %d", i+1); |
336 |
|
|
} |
337 |
|
|
free(workers); |
338 |
|
|
if (!g_verbose) |
339 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, ""); |
340 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Verified %d files", |
341 |
|
|
atomic_read32(&g_num_files)); |
342 |
|
|
|
343 |
|
|
if (atomic_read32(&g_num_tmp_catalog) > 0) |
344 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "Temporary file catalogs were found."); |
345 |
|
|
|
346 |
|
|
if (atomic_read32(&g_force_rebuild)) { |
347 |
|
|
if (unlink("cachedb") == 0) { |
348 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
349 |
|
|
"Fix: managed cache db unlinked, will be rebuilt on next mount"); |
350 |
|
|
atomic_inc32(&g_num_err_fixed); |
351 |
|
|
} else { |
352 |
|
|
if (errno != ENOENT) { |
353 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, |
354 |
|
|
"Error: could not unlink managed cache database (%d)", errno); |
355 |
|
|
atomic_inc32(&g_num_err_unfixed); |
356 |
|
|
} |
357 |
|
|
} |
358 |
|
|
} |
359 |
|
|
|
360 |
|
|
if (atomic_read32(&g_modified_cache)) { |
361 |
|
|
LogCvmfs(kLogCvmfs, kLogStdout, "\n" |
362 |
|
|
"WARNING: There might by corrupted files in the kernel buffers.\n" |
363 |
|
|
"Remount CernVM-FS or run 'echo 3 > /proc/sys/vm/drop_caches'" |
364 |
|
|
"\n\n"); |
365 |
|
|
} |
366 |
|
|
|
367 |
|
|
int retval = 0; |
368 |
|
|
if (atomic_read32(&g_num_err_fixed) > 0) |
369 |
|
|
retval |= kErrorFixed; |
370 |
|
|
if (atomic_read32(&g_num_err_unfixed) > 0) |
371 |
|
|
retval |= kErrorUnfixed; |
372 |
|
|
if (atomic_read32(&g_num_err_operational) > 0) |
373 |
|
|
retval |= kErrorOperational; |
374 |
|
|
|
375 |
|
|
return retval; |
376 |
|
|
} |