]>
Commit | Line | Data |
---|---|---|
f5e04665 LP |
1 | /*** |
2 | This file is part of systemd. | |
3 | ||
4 | Copyright 2012 Lennart Poettering | |
5 | ||
6 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
7 | under the terms of the GNU Lesser General Public License as published by |
8 | the Free Software Foundation; either version 2.1 of the License, or | |
f5e04665 LP |
9 | (at your option) any later version. |
10 | ||
11 | systemd is distributed in the hope that it will be useful, but | |
12 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 14 | Lesser General Public License for more details. |
f5e04665 | 15 | |
5430f7f2 | 16 | You should have received a copy of the GNU Lesser General Public License |
f5e04665 LP |
17 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
18 | ***/ | |
19 | ||
20 | #include <errno.h> | |
803a3464 LP |
21 | #include <stdio.h> |
22 | #include <sys/prctl.h> | |
cacd6403 | 23 | #include <sys/xattr.h> |
4f5dd394 | 24 | #include <unistd.h> |
f5e04665 | 25 | |
4d229b31 | 26 | #ifdef HAVE_ELFUTILS |
3c171f0b LP |
27 | #include <dwarf.h> |
28 | #include <elfutils/libdwfl.h> | |
4d229b31 UTL |
29 | #endif |
30 | ||
f11943c5 LP |
31 | #include "sd-journal.h" |
32 | #include "sd-login.h" | |
3c171f0b | 33 | #include "sd-daemon.h" |
4f5dd394 LP |
34 | |
35 | #include "acl-util.h" | |
b5efdb8a | 36 | #include "alloc-util.h" |
430f0182 | 37 | #include "capability-util.h" |
ba1261bc | 38 | #include "cgroup-util.h" |
4f5dd394 | 39 | #include "compress.h" |
34c10968 LP |
40 | #include "conf-parser.h" |
41 | #include "copy.h" | |
f11943c5 | 42 | #include "coredump-vacuum.h" |
a0956174 | 43 | #include "dirent-util.h" |
4f5dd394 | 44 | #include "escape.h" |
3ffd4af2 | 45 | #include "fd-util.h" |
4f5dd394 | 46 | #include "fileio.h" |
f4f15635 | 47 | #include "fs-util.h" |
afc5dbf3 | 48 | #include "io-util.h" |
4f5dd394 LP |
49 | #include "journald-native.h" |
50 | #include "log.h" | |
51 | #include "macro.h" | |
0c773903 | 52 | #include "missing.h" |
4f5dd394 | 53 | #include "mkdir.h" |
6bedfcbb | 54 | #include "parse-util.h" |
0b452006 | 55 | #include "process-util.h" |
3c171f0b | 56 | #include "socket-util.h" |
4f5dd394 LP |
57 | #include "special.h" |
58 | #include "stacktrace.h" | |
8b43440b | 59 | #include "string-table.h" |
07630cea | 60 | #include "string-util.h" |
4f5dd394 | 61 | #include "strv.h" |
b1d4f8e1 | 62 | #include "user-util.h" |
4f5dd394 | 63 | #include "util.h" |
34727273 | 64 | |
34c10968 | 65 | /* The maximum size up to which we process coredumps */ |
59f448cf | 66 | #define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU)) |
34c10968 | 67 | |
bdfd7b2c | 68 | /* The maximum size up to which we leave the coredump around on disk */ |
34c10968 LP |
69 | #define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX |
70 | ||
bdfd7b2c | 71 | /* The maximum size up to which we store the coredump in the journal */ |
34c10968 | 72 | #define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU)) |
f5e04665 | 73 | |
c4aa09b0 | 74 | /* Make sure to not make this larger than the maximum journal entry |
874bc134 ZJS |
75 | * size. See DATA_SIZE_MAX in journald-native.c. */ |
76 | assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX); | |
f5e04665 LP |
77 | |
78 | enum { | |
3c171f0b LP |
79 | /* We use this as array indexes for a couple of special fields we use for naming coredumping files, and |
80 | * attaching xattrs */ | |
81 | CONTEXT_PID, | |
82 | CONTEXT_UID, | |
83 | CONTEXT_GID, | |
84 | CONTEXT_SIGNAL, | |
85 | CONTEXT_TIMESTAMP, | |
bdfd7b2c | 86 | CONTEXT_RLIMIT, |
3c171f0b LP |
87 | CONTEXT_COMM, |
88 | CONTEXT_EXE, | |
89 | _CONTEXT_MAX | |
f5e04665 LP |
90 | }; |
91 | ||
34c10968 LP |
92 | typedef enum CoredumpStorage { |
93 | COREDUMP_STORAGE_NONE, | |
94 | COREDUMP_STORAGE_EXTERNAL, | |
95 | COREDUMP_STORAGE_JOURNAL, | |
34c10968 LP |
96 | _COREDUMP_STORAGE_MAX, |
97 | _COREDUMP_STORAGE_INVALID = -1 | |
98 | } CoredumpStorage; | |
99 | ||
34c10968 LP |
100 | static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = { |
101 | [COREDUMP_STORAGE_NONE] = "none", | |
102 | [COREDUMP_STORAGE_EXTERNAL] = "external", | |
103 | [COREDUMP_STORAGE_JOURNAL] = "journal", | |
34c10968 LP |
104 | }; |
105 | ||
106 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage); | |
8c9571d0 | 107 | static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting"); |
34727273 ZJS |
108 | |
109 | static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL; | |
8c9571d0 | 110 | static bool arg_compress = true; |
59f448cf LP |
111 | static uint64_t arg_process_size_max = PROCESS_SIZE_MAX; |
112 | static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX; | |
34727273 | 113 | static size_t arg_journal_size_max = JOURNAL_SIZE_MAX; |
59f448cf LP |
114 | static uint64_t arg_keep_free = (uint64_t) -1; |
115 | static uint64_t arg_max_use = (uint64_t) -1; | |
34c10968 LP |
116 | |
117 | static int parse_config(void) { | |
34c10968 | 118 | static const ConfigTableItem items[] = { |
8c9571d0 LP |
119 | { "Coredump", "Storage", config_parse_coredump_storage, 0, &arg_storage }, |
120 | { "Coredump", "Compress", config_parse_bool, 0, &arg_compress }, | |
59f448cf LP |
121 | { "Coredump", "ProcessSizeMax", config_parse_iec_uint64, 0, &arg_process_size_max }, |
122 | { "Coredump", "ExternalSizeMax", config_parse_iec_uint64, 0, &arg_external_size_max }, | |
8c9571d0 | 123 | { "Coredump", "JournalSizeMax", config_parse_iec_size, 0, &arg_journal_size_max }, |
59f448cf LP |
124 | { "Coredump", "KeepFree", config_parse_iec_uint64, 0, &arg_keep_free }, |
125 | { "Coredump", "MaxUse", config_parse_iec_uint64, 0, &arg_max_use }, | |
34c10968 LP |
126 | {} |
127 | }; | |
128 | ||
43688c49 | 129 | return config_parse_many_nulstr(PKGSYSCONFDIR "/coredump.conf", |
75eb6154 | 130 | CONF_PATHS_NULSTR("systemd/coredump.conf.d"), |
301af7e4 JT |
131 | "Coredump\0", |
132 | config_item_table_lookup, items, | |
133 | false, NULL); | |
34c10968 LP |
134 | } |
135 | ||
136 | static int fix_acl(int fd, uid_t uid) { | |
137 | ||
138 | #ifdef HAVE_ACL | |
139 | _cleanup_(acl_freep) acl_t acl = NULL; | |
140 | acl_entry_t entry; | |
141 | acl_permset_t permset; | |
709f6e46 | 142 | int r; |
34c10968 | 143 | |
b59233e6 LP |
144 | assert(fd >= 0); |
145 | ||
34c10968 LP |
146 | if (uid <= SYSTEM_UID_MAX) |
147 | return 0; | |
148 | ||
149 | /* Make sure normal users can read (but not write or delete) | |
150 | * their own coredumps */ | |
151 | ||
152 | acl = acl_get_fd(fd); | |
4a62c710 MS |
153 | if (!acl) |
154 | return log_error_errno(errno, "Failed to get ACL: %m"); | |
34c10968 LP |
155 | |
156 | if (acl_create_entry(&acl, &entry) < 0 || | |
157 | acl_set_tag_type(entry, ACL_USER) < 0 || | |
d710aaf7 ZJS |
158 | acl_set_qualifier(entry, &uid) < 0) |
159 | return log_error_errno(errno, "Failed to patch ACL: %m"); | |
34c10968 LP |
160 | |
161 | if (acl_get_permset(entry, &permset) < 0 || | |
709f6e46 MS |
162 | acl_add_perm(permset, ACL_READ) < 0) |
163 | return log_warning_errno(errno, "Failed to patch ACL: %m"); | |
164 | ||
165 | r = calc_acl_mask_if_needed(&acl); | |
166 | if (r < 0) | |
167 | return log_warning_errno(r, "Failed to patch ACL: %m"); | |
34c10968 | 168 | |
4a62c710 MS |
169 | if (acl_set_fd(fd, acl) < 0) |
170 | return log_error_errno(errno, "Failed to apply ACL: %m"); | |
34c10968 LP |
171 | #endif |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
3c171f0b | 176 | static int fix_xattr(int fd, const char *context[_CONTEXT_MAX]) { |
0cd77f97 | 177 | |
3c171f0b LP |
178 | static const char * const xattrs[_CONTEXT_MAX] = { |
179 | [CONTEXT_PID] = "user.coredump.pid", | |
180 | [CONTEXT_UID] = "user.coredump.uid", | |
181 | [CONTEXT_GID] = "user.coredump.gid", | |
182 | [CONTEXT_SIGNAL] = "user.coredump.signal", | |
183 | [CONTEXT_TIMESTAMP] = "user.coredump.timestamp", | |
184 | [CONTEXT_COMM] = "user.coredump.comm", | |
185 | [CONTEXT_EXE] = "user.coredump.exe", | |
0cd77f97 LP |
186 | }; |
187 | ||
34c10968 | 188 | int r = 0; |
0cd77f97 | 189 | unsigned i; |
34c10968 | 190 | |
b59233e6 LP |
191 | assert(fd >= 0); |
192 | ||
1eef15b1 | 193 | /* Attach some metadata to coredumps via extended |
34c10968 LP |
194 | * attributes. Just because we can. */ |
195 | ||
3c171f0b | 196 | for (i = 0; i < _CONTEXT_MAX; i++) { |
1eef15b1 ZJS |
197 | int k; |
198 | ||
3c171f0b | 199 | if (isempty(context[i]) || !xattrs[i]) |
0cd77f97 | 200 | continue; |
34c10968 | 201 | |
3c171f0b | 202 | k = fsetxattr(fd, xattrs[i], context[i], strlen(context[i]), XATTR_CREATE); |
1eef15b1 | 203 | if (k < 0 && r == 0) |
34c10968 | 204 | r = -errno; |
0cd77f97 | 205 | } |
34c10968 LP |
206 | |
207 | return r; | |
208 | } | |
209 | ||
b0b21dce | 210 | #define filename_escape(s) xescape((s), "./ ") |
34c10968 | 211 | |
0c773903 EV |
212 | static inline const char *coredump_tmpfile_name(const char *s) { |
213 | return s ? s : "(unnamed temporary file)"; | |
214 | } | |
215 | ||
b59233e6 LP |
216 | static int fix_permissions( |
217 | int fd, | |
218 | const char *filename, | |
219 | const char *target, | |
3c171f0b | 220 | const char *context[_CONTEXT_MAX], |
b59233e6 LP |
221 | uid_t uid) { |
222 | ||
03532f0a LP |
223 | int r; |
224 | ||
b59233e6 | 225 | assert(fd >= 0); |
b59233e6 | 226 | assert(target); |
3c171f0b | 227 | assert(context); |
cfd652ed ZJS |
228 | |
229 | /* Ignore errors on these */ | |
3c171f0b LP |
230 | (void) fchmod(fd, 0640); |
231 | (void) fix_acl(fd, uid); | |
232 | (void) fix_xattr(fd, context); | |
cfd652ed | 233 | |
4a62c710 | 234 | if (fsync(fd) < 0) |
0c773903 | 235 | return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename)); |
cfd652ed | 236 | |
03532f0a LP |
237 | r = link_tmpfile(fd, filename, target); |
238 | if (r < 0) | |
239 | return log_error_errno(r, "Failed to move coredump %s into place: %m", target); | |
cfd652ed ZJS |
240 | |
241 | return 0; | |
242 | } | |
243 | ||
59f448cf | 244 | static int maybe_remove_external_coredump(const char *filename, uint64_t size) { |
cfd652ed | 245 | |
b59233e6 | 246 | /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */ |
cfd652ed | 247 | |
fc6cec86 | 248 | if (arg_storage == COREDUMP_STORAGE_EXTERNAL && |
cfd652ed ZJS |
249 | size <= arg_external_size_max) |
250 | return 0; | |
251 | ||
252 | if (!filename) | |
253 | return 1; | |
254 | ||
4a62c710 MS |
255 | if (unlink(filename) < 0 && errno != ENOENT) |
256 | return log_error_errno(errno, "Failed to unlink %s: %m", filename); | |
cfd652ed ZJS |
257 | |
258 | return 1; | |
259 | } | |
260 | ||
3c171f0b | 261 | static int make_filename(const char *context[_CONTEXT_MAX], char **ret) { |
b59233e6 | 262 | _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL; |
a7f7d1bd | 263 | sd_id128_t boot = {}; |
34c10968 LP |
264 | int r; |
265 | ||
3c171f0b | 266 | assert(context); |
34c10968 | 267 | |
3c171f0b | 268 | c = filename_escape(context[CONTEXT_COMM]); |
34c10968 | 269 | if (!c) |
b59233e6 | 270 | return -ENOMEM; |
34c10968 | 271 | |
3c171f0b | 272 | u = filename_escape(context[CONTEXT_UID]); |
0dc5d23c | 273 | if (!u) |
b59233e6 | 274 | return -ENOMEM; |
34c10968 LP |
275 | |
276 | r = sd_id128_get_boot(&boot); | |
b59233e6 | 277 | if (r < 0) |
34c10968 | 278 | return r; |
34c10968 | 279 | |
3c171f0b | 280 | p = filename_escape(context[CONTEXT_PID]); |
b59233e6 LP |
281 | if (!p) |
282 | return -ENOMEM; | |
283 | ||
3c171f0b | 284 | t = filename_escape(context[CONTEXT_TIMESTAMP]); |
b59233e6 LP |
285 | if (!t) |
286 | return -ENOMEM; | |
287 | ||
288 | if (asprintf(ret, | |
0dc5d23c | 289 | "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s000000", |
34c10968 | 290 | c, |
0dc5d23c | 291 | u, |
34c10968 LP |
292 | SD_ID128_FORMAT_VAL(boot), |
293 | p, | |
b59233e6 LP |
294 | t) < 0) |
295 | return -ENOMEM; | |
296 | ||
297 | return 0; | |
298 | } | |
299 | ||
300 | static int save_external_coredump( | |
3c171f0b LP |
301 | const char *context[_CONTEXT_MAX], |
302 | int input_fd, | |
b59233e6 | 303 | char **ret_filename, |
5f3e0a74 HW |
304 | int *ret_node_fd, |
305 | int *ret_data_fd, | |
59f448cf | 306 | uint64_t *ret_size) { |
b59233e6 LP |
307 | |
308 | _cleanup_free_ char *fn = NULL, *tmp = NULL; | |
309 | _cleanup_close_ int fd = -1; | |
bdfd7b2c | 310 | uint64_t rlimit, max_size; |
b59233e6 | 311 | struct stat st; |
3c171f0b | 312 | uid_t uid; |
b59233e6 LP |
313 | int r; |
314 | ||
3c171f0b | 315 | assert(context); |
b59233e6 | 316 | assert(ret_filename); |
5f3e0a74 HW |
317 | assert(ret_node_fd); |
318 | assert(ret_data_fd); | |
b59233e6 LP |
319 | assert(ret_size); |
320 | ||
3c171f0b LP |
321 | r = parse_uid(context[CONTEXT_UID], &uid); |
322 | if (r < 0) | |
323 | return log_error_errno(r, "Failed to parse UID: %m"); | |
324 | ||
bdfd7b2c LP |
325 | r = safe_atou64(context[CONTEXT_RLIMIT], &rlimit); |
326 | if (r < 0) | |
327 | return log_error_errno(r, "Failed to parse resource limit: %s", context[CONTEXT_RLIMIT]); | |
6998b540 ZJS |
328 | if (rlimit < page_size()) { |
329 | /* Is coredumping disabled? Then don't bother saving/processing the coredump. | |
330 | * Anything below PAGE_SIZE cannot give a readable coredump (the kernel uses | |
331 | * ELF_EXEC_PAGESIZE which is not easily accessible, but is usually the same as PAGE_SIZE. */ | |
332 | log_info("Core dumping has been disabled for process %s (%s).", context[CONTEXT_PID], context[CONTEXT_COMM]); | |
bdfd7b2c LP |
333 | return -EBADSLT; |
334 | } | |
335 | ||
336 | /* Never store more than the process configured, or than we actually shall keep or process */ | |
337 | max_size = MIN(rlimit, MAX(arg_process_size_max, arg_external_size_max)); | |
338 | ||
3c171f0b | 339 | r = make_filename(context, &fn); |
23bbb0de MS |
340 | if (r < 0) |
341 | return log_error_errno(r, "Failed to determine coredump file name: %m"); | |
34c10968 | 342 | |
d2e54fae | 343 | mkdir_p_label("/var/lib/systemd/coredump", 0755); |
803a3464 | 344 | |
03532f0a | 345 | fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp); |
4a62c710 | 346 | if (fd < 0) |
03532f0a | 347 | return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn); |
803a3464 | 348 | |
bdfd7b2c | 349 | r = copy_bytes(input_fd, fd, max_size, false); |
84ee0960 | 350 | if (r == -EFBIG) { |
3c171f0b | 351 | log_error("Coredump of %s (%s) is larger than configured processing limit, refusing.", context[CONTEXT_PID], context[CONTEXT_COMM]); |
93240d3a LP |
352 | goto fail; |
353 | } else if (IN_SET(r, -EDQUOT, -ENOSPC)) { | |
3c171f0b | 354 | log_error("Not enough disk space for coredump of %s (%s), refusing.", context[CONTEXT_PID], context[CONTEXT_COMM]); |
93240d3a LP |
355 | goto fail; |
356 | } else if (r < 0) { | |
da927ba9 | 357 | log_error_errno(r, "Failed to dump coredump to file: %m"); |
34c10968 LP |
358 | goto fail; |
359 | } | |
803a3464 | 360 | |
34c10968 | 361 | if (fstat(fd, &st) < 0) { |
0c773903 | 362 | log_error_errno(errno, "Failed to fstat coredump %s: %m", coredump_tmpfile_name(tmp)); |
34c10968 LP |
363 | goto fail; |
364 | } | |
365 | ||
7849c2ac | 366 | if (lseek(fd, 0, SEEK_SET) == (off_t) -1) { |
0c773903 | 367 | log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp)); |
b59233e6 | 368 | goto fail; |
7849c2ac TA |
369 | } |
370 | ||
d89c8fdf | 371 | #if defined(HAVE_XZ) || defined(HAVE_LZ4) |
cfd652ed | 372 | /* If we will remove the coredump anyway, do not compress. */ |
6e9ef603 | 373 | if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) { |
cfd652ed | 374 | |
b59233e6 LP |
375 | _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL; |
376 | _cleanup_close_ int fd_compressed = -1; | |
cfd652ed | 377 | |
d89c8fdf | 378 | fn_compressed = strappend(fn, COMPRESSED_EXT); |
b59233e6 | 379 | if (!fn_compressed) { |
d89c8fdf | 380 | log_oom(); |
cfd652ed ZJS |
381 | goto uncompressed; |
382 | } | |
383 | ||
03532f0a LP |
384 | fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed); |
385 | if (fd_compressed < 0) { | |
386 | log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed); | |
b59233e6 | 387 | goto uncompressed; |
03532f0a | 388 | } |
cfd652ed | 389 | |
d89c8fdf | 390 | r = compress_stream(fd, fd_compressed, -1); |
b59233e6 | 391 | if (r < 0) { |
0c773903 | 392 | log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed)); |
b59233e6 LP |
393 | goto fail_compressed; |
394 | } | |
395 | ||
3c171f0b | 396 | r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid); |
cfd652ed | 397 | if (r < 0) |
b59233e6 LP |
398 | goto fail_compressed; |
399 | ||
400 | /* OK, this worked, we can get rid of the uncompressed version now */ | |
0c773903 EV |
401 | if (tmp) |
402 | unlink_noerrno(tmp); | |
cfd652ed | 403 | |
59f448cf | 404 | *ret_filename = fn_compressed; /* compressed */ |
5f3e0a74 HW |
405 | *ret_node_fd = fd_compressed; /* compressed */ |
406 | *ret_data_fd = fd; /* uncompressed */ | |
59f448cf | 407 | *ret_size = (uint64_t) st.st_size; /* uncompressed */ |
cfd652ed | 408 | |
b59233e6 | 409 | fn_compressed = NULL; |
5f3e0a74 | 410 | fd = fd_compressed = -1; |
cfd652ed ZJS |
411 | |
412 | return 0; | |
413 | ||
b59233e6 | 414 | fail_compressed: |
0c773903 EV |
415 | if (tmp_compressed) |
416 | (void) unlink(tmp_compressed); | |
34c10968 | 417 | } |
cfd652ed ZJS |
418 | |
419 | uncompressed: | |
3b1a55e1 | 420 | #endif |
5f3e0a74 | 421 | |
3c171f0b | 422 | r = fix_permissions(fd, tmp, fn, context, uid); |
cfd652ed ZJS |
423 | if (r < 0) |
424 | goto fail; | |
34c10968 LP |
425 | |
426 | *ret_filename = fn; | |
5f3e0a74 HW |
427 | *ret_data_fd = fd; |
428 | *ret_node_fd = -1; | |
59f448cf | 429 | *ret_size = (uint64_t) st.st_size; |
34c10968 LP |
430 | |
431 | fn = NULL; | |
432 | fd = -1; | |
433 | ||
434 | return 0; | |
435 | ||
436 | fail: | |
0c773903 EV |
437 | if (tmp) |
438 | (void) unlink(tmp); | |
34c10968 LP |
439 | return r; |
440 | } | |
441 | ||
442 | static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) { | |
443 | _cleanup_free_ char *field = NULL; | |
444 | ssize_t n; | |
445 | ||
8d4e028f | 446 | assert(fd >= 0); |
34c10968 LP |
447 | assert(ret); |
448 | assert(ret_size); | |
449 | ||
4a62c710 MS |
450 | if (lseek(fd, 0, SEEK_SET) == (off_t) -1) |
451 | return log_warning_errno(errno, "Failed to seek: %m"); | |
803a3464 | 452 | |
34c10968 LP |
453 | field = malloc(9 + size); |
454 | if (!field) { | |
cfd652ed | 455 | log_warning("Failed to allocate memory for coredump, coredump will not be stored."); |
34c10968 LP |
456 | return -ENOMEM; |
457 | } | |
458 | ||
459 | memcpy(field, "COREDUMP=", 9); | |
460 | ||
461 | n = read(fd, field + 9, size); | |
23bbb0de MS |
462 | if (n < 0) |
463 | return log_error_errno((int) n, "Failed to read core data: %m"); | |
34c10968 LP |
464 | if ((size_t) n < size) { |
465 | log_error("Core data too short."); | |
466 | return -EIO; | |
467 | } | |
468 | ||
469 | *ret = field; | |
470 | *ret_size = size + 9; | |
471 | ||
472 | field = NULL; | |
473 | ||
474 | return 0; | |
475 | } | |
803a3464 | 476 | |
3f132692 JF |
477 | /* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines: |
478 | * 0:/dev/pts/23 | |
479 | * pos: 0 | |
480 | * flags: 0100002 | |
481 | * | |
482 | * 1:/dev/pts/23 | |
483 | * pos: 0 | |
484 | * flags: 0100002 | |
485 | * | |
486 | * 2:/dev/pts/23 | |
487 | * pos: 0 | |
488 | * flags: 0100002 | |
489 | * EOF | |
490 | */ | |
491 | static int compose_open_fds(pid_t pid, char **open_fds) { | |
4d84bc2f LP |
492 | _cleanup_closedir_ DIR *proc_fd_dir = NULL; |
493 | _cleanup_close_ int proc_fdinfo_fd = -1; | |
494 | _cleanup_free_ char *buffer = NULL; | |
3f132692 | 495 | _cleanup_fclose_ FILE *stream = NULL; |
59059b4a | 496 | const char *fddelim = "", *path; |
3f132692 | 497 | struct dirent *dent = NULL; |
4d84bc2f | 498 | size_t size = 0; |
3f132692 JF |
499 | int r = 0; |
500 | ||
501 | assert(pid >= 0); | |
502 | assert(open_fds != NULL); | |
503 | ||
59059b4a | 504 | path = procfs_file_alloca(pid, "fd"); |
3f132692 | 505 | proc_fd_dir = opendir(path); |
59059b4a ZJS |
506 | if (!proc_fd_dir) |
507 | return -errno; | |
3f132692 | 508 | |
4d84bc2f | 509 | proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH); |
59059b4a ZJS |
510 | if (proc_fdinfo_fd < 0) |
511 | return -errno; | |
3f132692 | 512 | |
4d84bc2f | 513 | stream = open_memstream(&buffer, &size); |
3f132692 JF |
514 | if (!stream) |
515 | return -ENOMEM; | |
516 | ||
4d84bc2f | 517 | FOREACH_DIRENT(dent, proc_fd_dir, return -errno) { |
3f132692 | 518 | _cleanup_fclose_ FILE *fdinfo = NULL; |
4d84bc2f | 519 | _cleanup_free_ char *fdname = NULL; |
59059b4a | 520 | char line[LINE_MAX]; |
4d84bc2f | 521 | int fd; |
3f132692 | 522 | |
59059b4a | 523 | r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname); |
3f132692 JF |
524 | if (r < 0) |
525 | return r; | |
526 | ||
527 | fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname); | |
528 | fddelim = "\n"; | |
529 | ||
530 | /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */ | |
59059b4a ZJS |
531 | fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY); |
532 | if (fd < 0) | |
3f132692 JF |
533 | continue; |
534 | ||
59059b4a ZJS |
535 | fdinfo = fdopen(fd, "re"); |
536 | if (fdinfo == NULL) { | |
537 | close(fd); | |
3f132692 | 538 | continue; |
59059b4a | 539 | } |
3f132692 | 540 | |
4d84bc2f LP |
541 | FOREACH_LINE(line, fdinfo, break) { |
542 | fputs(line, stream); | |
543 | if (!endswith(line, "\n")) | |
544 | fputc('\n', stream); | |
545 | } | |
3f132692 JF |
546 | } |
547 | ||
4d84bc2f | 548 | errno = 0; |
74ca738f | 549 | stream = safe_fclose(stream); |
4d84bc2f | 550 | |
b3267152 | 551 | if (errno > 0) |
4d84bc2f LP |
552 | return -errno; |
553 | ||
554 | *open_fds = buffer; | |
555 | buffer = NULL; | |
556 | ||
3f132692 JF |
557 | return 0; |
558 | } | |
559 | ||
7ed03ce6 JF |
560 | static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) { |
561 | const char *p; | |
562 | struct stat stbuf; | |
563 | _cleanup_close_ int proc_ns_dir_fd; | |
564 | ||
565 | p = procfs_file_alloca(pid, "ns"); | |
566 | ||
567 | proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY); | |
568 | if (proc_ns_dir_fd < 0) | |
569 | return -errno; | |
570 | ||
571 | if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0) | |
572 | return -errno; | |
573 | ||
574 | *ns = stbuf.st_ino; | |
575 | return 0; | |
576 | } | |
577 | ||
578 | static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) { | |
579 | pid_t cpid = pid, ppid = 0; | |
580 | ino_t proc_mntns; | |
581 | int r = 0; | |
582 | ||
583 | r = get_process_ns(pid, "mnt", &proc_mntns); | |
584 | if (r < 0) | |
585 | return r; | |
586 | ||
587 | while (1) { | |
588 | ino_t parent_mntns; | |
589 | ||
590 | r = get_process_ppid(cpid, &ppid); | |
591 | if (r < 0) | |
592 | return r; | |
593 | ||
594 | r = get_process_ns(ppid, "mnt", &parent_mntns); | |
595 | if (r < 0) | |
596 | return r; | |
597 | ||
598 | if (proc_mntns != parent_mntns) | |
599 | break; | |
600 | ||
601 | if (ppid == 1) | |
602 | return -ENOENT; | |
603 | ||
604 | cpid = ppid; | |
605 | } | |
606 | ||
607 | *container_pid = ppid; | |
608 | return 0; | |
609 | } | |
610 | ||
611 | /* Returns 1 if the parent was found. | |
612 | * Returns 0 if there is not a process we can call the pid's | |
613 | * container parent (the pid's process isn't 'containerized'). | |
614 | * Returns a negative number on errors. | |
615 | */ | |
616 | static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) { | |
617 | int r = 0; | |
618 | pid_t container_pid; | |
619 | const char *proc_root_path; | |
620 | struct stat root_stat, proc_root_stat; | |
621 | ||
622 | /* To compare inodes of / and /proc/[pid]/root */ | |
623 | if (stat("/", &root_stat) < 0) | |
624 | return -errno; | |
625 | ||
626 | proc_root_path = procfs_file_alloca(pid, "root"); | |
627 | if (stat(proc_root_path, &proc_root_stat) < 0) | |
628 | return -errno; | |
629 | ||
630 | /* The process uses system root. */ | |
631 | if (proc_root_stat.st_ino == root_stat.st_ino) { | |
632 | *cmdline = NULL; | |
633 | return 0; | |
634 | } | |
635 | ||
636 | r = get_mount_namespace_leader(pid, &container_pid); | |
637 | if (r < 0) | |
638 | return r; | |
639 | ||
640 | return get_process_cmdline(container_pid, 0, false, cmdline); | |
641 | } | |
642 | ||
3c171f0b LP |
643 | static int change_uid_gid(const char *context[]) { |
644 | uid_t uid; | |
645 | gid_t gid; | |
646 | int r; | |
34c10968 | 647 | |
3c171f0b LP |
648 | r = parse_uid(context[CONTEXT_UID], &uid); |
649 | if (r < 0) | |
650 | return r; | |
8c8549db | 651 | |
888e378d LP |
652 | if (uid <= SYSTEM_UID_MAX) { |
653 | const char *user = "systemd-coredump"; | |
654 | ||
655 | r = get_user_creds(&user, &uid, &gid, NULL, NULL); | |
656 | if (r < 0) { | |
657 | log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user); | |
658 | uid = gid = 0; | |
659 | } | |
660 | } else { | |
661 | r = parse_gid(context[CONTEXT_GID], &gid); | |
662 | if (r < 0) | |
663 | return r; | |
664 | } | |
3c171f0b LP |
665 | |
666 | return drop_privileges(uid, gid, 0); | |
667 | } | |
8c8549db | 668 | |
3c171f0b LP |
669 | static int submit_coredump( |
670 | const char *context[_CONTEXT_MAX], | |
671 | struct iovec *iovec, | |
672 | size_t n_iovec_allocated, | |
673 | size_t n_iovec, | |
674 | int input_fd) { | |
34c10968 | 675 | |
5f3e0a74 | 676 | _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1; |
3c171f0b | 677 | _cleanup_free_ char *core_message = NULL, *filename = NULL, *coredump_data = NULL; |
a5ca3649 | 678 | uint64_t coredump_size = UINT64_MAX; |
3c171f0b | 679 | int r; |
f5e04665 | 680 | |
3c171f0b LP |
681 | assert(context); |
682 | assert(iovec); | |
683 | assert(n_iovec_allocated >= n_iovec + 3); | |
684 | assert(input_fd >= 0); | |
f5e04665 | 685 | |
3c171f0b LP |
686 | /* Vacuum before we write anything again */ |
687 | (void) coredump_vacuum(-1, arg_keep_free, arg_max_use); | |
803a3464 | 688 | |
3c171f0b LP |
689 | /* Always stream the coredump to disk, if that's possible */ |
690 | r = save_external_coredump(context, input_fd, &filename, &coredump_node_fd, &coredump_fd, &coredump_size); | |
691 | if (r < 0) | |
692 | /* Skip whole core dumping part */ | |
693 | goto log; | |
694 | ||
695 | /* If we don't want to keep the coredump on disk, remove it now, as later on we will lack the privileges for | |
696 | * it. However, we keep the fd to it, so that we can still process it and log it. */ | |
697 | r = maybe_remove_external_coredump(filename, coredump_size); | |
698 | if (r < 0) | |
699 | return r; | |
700 | if (r == 0) { | |
701 | const char *coredump_filename; | |
702 | ||
703 | coredump_filename = strjoina("COREDUMP_FILENAME=", filename); | |
704 | IOVEC_SET_STRING(iovec[n_iovec++], coredump_filename); | |
6e9ef603 ZJS |
705 | } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL) |
706 | log_info("The core will not be stored: size %zu is greater than %zu (the configured maximum)", | |
707 | coredump_size, arg_external_size_max); | |
f5e04665 | 708 | |
3c171f0b LP |
709 | /* Vacuum again, but exclude the coredump we just created */ |
710 | (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use); | |
8c9571d0 | 711 | |
3c171f0b LP |
712 | /* Now, let's drop privileges to become the user who owns the segfaulted process and allocate the coredump |
713 | * memory under the user's uid. This also ensures that the credentials journald will see are the ones of the | |
714 | * coredumping user, thus making sure the user gets access to the core dump. Let's also get rid of all | |
715 | * capabilities, if we run as root, we won't need them anymore. */ | |
716 | r = change_uid_gid(context); | |
717 | if (r < 0) | |
718 | return log_error_errno(r, "Failed to drop privileges: %m"); | |
34c10968 | 719 | |
3c171f0b LP |
720 | #ifdef HAVE_ELFUTILS |
721 | /* Try to get a strack trace if we can */ | |
722 | if (coredump_size <= arg_process_size_max) { | |
723 | _cleanup_free_ char *stacktrace = NULL; | |
724 | ||
725 | r = coredump_make_stack_trace(coredump_fd, context[CONTEXT_EXE], &stacktrace); | |
726 | if (r >= 0) | |
727 | core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID], " (", context[CONTEXT_COMM], ") of user ", context[CONTEXT_UID], " dumped core.\n\n", stacktrace, NULL); | |
728 | else if (r == -EINVAL) | |
729 | log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno())); | |
730 | else | |
731 | log_warning_errno(r, "Failed to generate stack trace: %m"); | |
6e9ef603 ZJS |
732 | } else |
733 | log_debug("Not generating stack trace: core size %zu is greater than %zu (the configured maximum)", | |
734 | coredump_size, arg_process_size_max); | |
803a3464 | 735 | |
3c171f0b LP |
736 | if (!core_message) |
737 | #endif | |
738 | log: | |
739 | core_message = strjoin("MESSAGE=Process ", context[CONTEXT_PID], " (", context[CONTEXT_COMM], ") of user ", context[CONTEXT_UID], " dumped core.", NULL); | |
740 | if (core_message) | |
741 | IOVEC_SET_STRING(iovec[n_iovec++], core_message); | |
742 | ||
743 | /* Optionally store the entire coredump in the journal */ | |
6e9ef603 ZJS |
744 | if (arg_storage == COREDUMP_STORAGE_JOURNAL) { |
745 | if (coredump_size <= arg_journal_size_max) { | |
746 | size_t sz = 0; | |
747 | ||
748 | /* Store the coredump itself in the journal */ | |
749 | ||
750 | r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz); | |
751 | if (r >= 0) { | |
752 | iovec[n_iovec].iov_base = coredump_data; | |
753 | iovec[n_iovec].iov_len = sz; | |
754 | n_iovec++; | |
755 | } else | |
756 | log_warning_errno(r, "Failed to attach the core to the journal entry: %m"); | |
757 | } else | |
758 | log_info("The core will not be stored: size %zu is greater than %zu (the configured maximum)", | |
759 | coredump_size, arg_journal_size_max); | |
f5e04665 LP |
760 | } |
761 | ||
3c171f0b LP |
762 | assert(n_iovec <= n_iovec_allocated); |
763 | ||
764 | r = sd_journal_sendv(iovec, n_iovec); | |
765 | if (r < 0) | |
766 | return log_error_errno(r, "Failed to log coredump: %m"); | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
771 | static void map_context_fields(const struct iovec *iovec, const char *context[]) { | |
772 | ||
773 | static const char * const context_field_names[_CONTEXT_MAX] = { | |
774 | [CONTEXT_PID] = "COREDUMP_PID=", | |
775 | [CONTEXT_UID] = "COREDUMP_UID=", | |
776 | [CONTEXT_GID] = "COREDUMP_GID=", | |
777 | [CONTEXT_SIGNAL] = "COREDUMP_SIGNAL=", | |
778 | [CONTEXT_TIMESTAMP] = "COREDUMP_TIMESTAMP=", | |
779 | [CONTEXT_COMM] = "COREDUMP_COMM=", | |
780 | [CONTEXT_EXE] = "COREDUMP_EXE=", | |
bdfd7b2c | 781 | [CONTEXT_RLIMIT] = "COREDUMP_RLIMIT=", |
3c171f0b LP |
782 | }; |
783 | ||
784 | unsigned i; | |
785 | ||
786 | assert(iovec); | |
787 | assert(context); | |
788 | ||
789 | for (i = 0; i < _CONTEXT_MAX; i++) { | |
790 | size_t l; | |
791 | ||
792 | l = strlen(context_field_names[i]); | |
793 | if (iovec->iov_len < l) | |
794 | continue; | |
795 | ||
796 | if (memcmp(iovec->iov_base, context_field_names[i], l) != 0) | |
797 | continue; | |
798 | ||
799 | /* Note that these strings are NUL terminated, because we made sure that a trailing NUL byte is in the | |
800 | * buffer, though not included in the iov_len count. (see below) */ | |
801 | context[i] = (char*) iovec->iov_base + l; | |
802 | break; | |
803 | } | |
804 | } | |
805 | ||
806 | static int process_socket(int fd) { | |
807 | _cleanup_close_ int coredump_fd = -1; | |
808 | struct iovec *iovec = NULL; | |
809 | size_t n_iovec = 0, n_iovec_allocated = 0, i; | |
810 | const char *context[_CONTEXT_MAX] = {}; | |
811 | int r; | |
812 | ||
813 | assert(fd >= 0); | |
814 | ||
815 | log_set_target(LOG_TARGET_AUTO); | |
816 | log_parse_environment(); | |
817 | log_open(); | |
818 | ||
819 | for (;;) { | |
820 | union { | |
821 | struct cmsghdr cmsghdr; | |
822 | uint8_t buf[CMSG_SPACE(sizeof(int))]; | |
823 | } control = {}; | |
824 | struct msghdr mh = { | |
825 | .msg_control = &control, | |
826 | .msg_controllen = sizeof(control), | |
827 | .msg_iovlen = 1, | |
828 | }; | |
829 | ssize_t n; | |
fe1ef0f8 | 830 | ssize_t l; |
3c171f0b LP |
831 | |
832 | if (!GREEDY_REALLOC(iovec, n_iovec_allocated, n_iovec + 3)) { | |
833 | r = log_oom(); | |
834 | goto finish; | |
835 | } | |
836 | ||
fe1ef0f8 EV |
837 | l = next_datagram_size_fd(fd); |
838 | if (l < 0) { | |
839 | r = log_error_errno(l, "Failed to determine datagram size to read: %m"); | |
3c171f0b LP |
840 | goto finish; |
841 | } | |
842 | ||
843 | assert(l >= 0); | |
844 | ||
845 | iovec[n_iovec].iov_len = l; | |
846 | iovec[n_iovec].iov_base = malloc(l + 1); | |
3c171f0b LP |
847 | if (!iovec[n_iovec].iov_base) { |
848 | r = log_oom(); | |
849 | goto finish; | |
850 | } | |
851 | ||
852 | mh.msg_iov = iovec + n_iovec; | |
853 | ||
854 | n = recvmsg(fd, &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC); | |
855 | if (n < 0) { | |
856 | free(iovec[n_iovec].iov_base); | |
857 | r = log_error_errno(errno, "Failed to receive datagram: %m"); | |
858 | goto finish; | |
859 | } | |
860 | ||
861 | if (n == 0) { | |
862 | struct cmsghdr *cmsg, *found = NULL; | |
863 | /* The final zero-length datagram carries the file descriptor and tells us that we're done. */ | |
864 | ||
865 | free(iovec[n_iovec].iov_base); | |
866 | ||
867 | CMSG_FOREACH(cmsg, &mh) { | |
868 | if (cmsg->cmsg_level == SOL_SOCKET && | |
869 | cmsg->cmsg_type == SCM_RIGHTS && | |
870 | cmsg->cmsg_len == CMSG_LEN(sizeof(int))) { | |
871 | assert(!found); | |
872 | found = cmsg; | |
873 | } | |
874 | } | |
875 | ||
876 | if (!found) { | |
877 | log_error("Coredump file descriptor missing."); | |
878 | r = -EBADMSG; | |
879 | goto finish; | |
880 | } | |
881 | ||
882 | assert(coredump_fd < 0); | |
883 | coredump_fd = *(int*) CMSG_DATA(found); | |
884 | break; | |
885 | } | |
886 | ||
887 | /* Add trailing NUL byte, in case these are strings */ | |
888 | ((char*) iovec[n_iovec].iov_base)[n] = 0; | |
889 | iovec[n_iovec].iov_len = (size_t) n; | |
890 | ||
891 | cmsg_close_all(&mh); | |
892 | map_context_fields(iovec + n_iovec, context); | |
893 | n_iovec++; | |
894 | } | |
895 | ||
896 | if (!GREEDY_REALLOC(iovec, n_iovec_allocated, n_iovec + 3)) { | |
897 | r = log_oom(); | |
34c10968 LP |
898 | goto finish; |
899 | } | |
900 | ||
61233823 | 901 | /* Make sure we got all data we really need */ |
3c171f0b LP |
902 | assert(context[CONTEXT_PID]); |
903 | assert(context[CONTEXT_UID]); | |
904 | assert(context[CONTEXT_GID]); | |
905 | assert(context[CONTEXT_SIGNAL]); | |
906 | assert(context[CONTEXT_TIMESTAMP]); | |
bdfd7b2c | 907 | assert(context[CONTEXT_RLIMIT]); |
3c171f0b LP |
908 | assert(context[CONTEXT_COMM]); |
909 | assert(coredump_fd >= 0); | |
910 | ||
911 | r = submit_coredump(context, iovec, n_iovec_allocated, n_iovec, coredump_fd); | |
912 | ||
913 | finish: | |
914 | for (i = 0; i < n_iovec; i++) | |
915 | free(iovec[i].iov_base); | |
916 | free(iovec); | |
917 | ||
918 | return r; | |
919 | } | |
920 | ||
921 | static int send_iovec(const struct iovec iovec[], size_t n_iovec, int input_fd) { | |
922 | ||
923 | static const union sockaddr_union sa = { | |
924 | .un.sun_family = AF_UNIX, | |
925 | .un.sun_path = "/run/systemd/coredump", | |
926 | }; | |
927 | _cleanup_close_ int fd = -1; | |
928 | size_t i; | |
929 | int r; | |
930 | ||
931 | assert(iovec || n_iovec <= 0); | |
932 | assert(input_fd >= 0); | |
933 | ||
934 | fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0); | |
935 | if (fd < 0) | |
936 | return log_error_errno(errno, "Failed to create coredump socket: %m"); | |
937 | ||
fc2fffe7 | 938 | if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) |
3c171f0b LP |
939 | return log_error_errno(errno, "Failed to connect to coredump service: %m"); |
940 | ||
941 | for (i = 0; i < n_iovec; i++) { | |
fec603eb LP |
942 | struct msghdr mh = { |
943 | .msg_iov = (struct iovec*) iovec + i, | |
944 | .msg_iovlen = 1, | |
945 | }; | |
946 | struct iovec copy[2]; | |
947 | ||
948 | for (;;) { | |
949 | if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0) | |
950 | break; | |
951 | ||
952 | if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) { | |
953 | /* This field didn't fit? That's a pity. Given that this is just metadata, | |
954 | * let's truncate the field at half, and try again. We append three dots, in | |
955 | * order to show that this is truncated. */ | |
956 | ||
957 | if (mh.msg_iov != copy) { | |
958 | /* We don't want to modify the caller's iovec, hence let's create our | |
959 | * own array, consisting of two new iovecs, where the first is a | |
960 | * (truncated) copy of what we want to send, and the second one | |
961 | * contains the trailing dots. */ | |
962 | copy[0] = iovec[i]; | |
963 | copy[1] = (struct iovec) { | |
964 | .iov_base = (char[]) { '.', '.', '.' }, | |
965 | .iov_len = 3, | |
966 | }; | |
967 | ||
968 | mh.msg_iov = copy; | |
969 | mh.msg_iovlen = 2; | |
970 | } | |
971 | ||
972 | copy[0].iov_len /= 2; /* halve it, and try again */ | |
973 | continue; | |
974 | } | |
3c171f0b | 975 | |
3c171f0b | 976 | return log_error_errno(errno, "Failed to send coredump datagram: %m"); |
fec603eb | 977 | } |
1eef15b1 ZJS |
978 | } |
979 | ||
3c171f0b LP |
980 | r = send_one_fd(fd, input_fd, 0); |
981 | if (r < 0) | |
982 | return log_error_errno(r, "Failed to send coredump fd: %m"); | |
1eef15b1 | 983 | |
3c171f0b LP |
984 | return 0; |
985 | } | |
1eef15b1 | 986 | |
78f043f7 | 987 | static int process_special_crash(const char *context[], int input_fd) { |
3c171f0b LP |
988 | _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1; |
989 | _cleanup_free_ char *filename = NULL; | |
990 | uint64_t coredump_size; | |
991 | int r; | |
803a3464 | 992 | |
3c171f0b LP |
993 | assert(context); |
994 | assert(input_fd >= 0); | |
803a3464 | 995 | |
78f043f7 | 996 | /* If we are pid1 or journald, we cut things short, don't write to the journal, but still create a coredump. */ |
34c10968 | 997 | |
3c171f0b LP |
998 | if (arg_storage != COREDUMP_STORAGE_NONE) |
999 | arg_storage = COREDUMP_STORAGE_EXTERNAL; | |
34c10968 | 1000 | |
3c171f0b LP |
1001 | r = save_external_coredump(context, input_fd, &filename, &coredump_node_fd, &coredump_fd, &coredump_size); |
1002 | if (r < 0) | |
1003 | return r; | |
34c10968 | 1004 | |
3c171f0b LP |
1005 | r = maybe_remove_external_coredump(filename, coredump_size); |
1006 | if (r < 0) | |
1007 | return r; | |
34c10968 | 1008 | |
78f043f7 LP |
1009 | log_notice("Detected coredump of the journal daemon or PID 1, diverted to %s.", filename); |
1010 | ||
3c171f0b LP |
1011 | return 0; |
1012 | } | |
1013 | ||
1014 | static int process_kernel(int argc, char* argv[]) { | |
1015 | ||
1016 | /* The small core field we allocate on the stack, to keep things simple */ | |
1017 | char | |
1018 | *core_pid = NULL, *core_uid = NULL, *core_gid = NULL, *core_signal = NULL, | |
1019 | *core_session = NULL, *core_exe = NULL, *core_comm = NULL, *core_cmdline = NULL, | |
1020 | *core_cgroup = NULL, *core_cwd = NULL, *core_root = NULL, *core_unit = NULL, | |
bdfd7b2c | 1021 | *core_user_unit = NULL, *core_slice = NULL, *core_timestamp = NULL, *core_rlimit = NULL; |
3c171f0b LP |
1022 | |
1023 | /* The larger ones we allocate on the heap */ | |
1024 | _cleanup_free_ char | |
1025 | *core_owner_uid = NULL, *core_open_fds = NULL, *core_proc_status = NULL, | |
d7032b1f | 1026 | *core_proc_maps = NULL, *core_proc_limits = NULL, *core_proc_cgroup = NULL, *core_environ = NULL, |
7ed03ce6 | 1027 | *core_proc_mountinfo = NULL, *core_container_cmdline = NULL; |
3c171f0b LP |
1028 | |
1029 | _cleanup_free_ char *exe = NULL, *comm = NULL; | |
1030 | const char *context[_CONTEXT_MAX]; | |
7ed03ce6 JF |
1031 | bool proc_self_root_is_slash; |
1032 | struct iovec iovec[27]; | |
3c171f0b LP |
1033 | size_t n_iovec = 0; |
1034 | uid_t owner_uid; | |
1035 | const char *p; | |
1036 | pid_t pid; | |
1037 | char *t; | |
1038 | int r; | |
1039 | ||
1040 | if (argc < CONTEXT_COMM + 1) { | |
1041 | log_error("Not enough arguments passed from kernel (%i, expected %i).", argc - 1, CONTEXT_COMM + 1 - 1); | |
1042 | return -EINVAL; | |
1043 | } | |
1044 | ||
1045 | r = parse_pid(argv[CONTEXT_PID + 1], &pid); | |
1046 | if (r < 0) | |
1047 | return log_error_errno(r, "Failed to parse PID."); | |
1048 | ||
1049 | r = get_process_comm(pid, &comm); | |
1050 | if (r < 0) { | |
1051 | log_warning_errno(r, "Failed to get COMM, falling back to the command line: %m"); | |
1052 | comm = strv_join(argv + CONTEXT_COMM + 1, " "); | |
1053 | if (!comm) | |
1054 | return log_oom(); | |
1055 | } | |
1056 | ||
1057 | r = get_process_exe(pid, &exe); | |
1058 | if (r < 0) | |
1059 | log_warning_errno(r, "Failed to get EXE, ignoring: %m"); | |
1060 | ||
1061 | context[CONTEXT_PID] = argv[CONTEXT_PID + 1]; | |
1062 | context[CONTEXT_UID] = argv[CONTEXT_UID + 1]; | |
1063 | context[CONTEXT_GID] = argv[CONTEXT_GID + 1]; | |
1064 | context[CONTEXT_SIGNAL] = argv[CONTEXT_SIGNAL + 1]; | |
1065 | context[CONTEXT_TIMESTAMP] = argv[CONTEXT_TIMESTAMP + 1]; | |
bdfd7b2c | 1066 | context[CONTEXT_RLIMIT] = argv[CONTEXT_RLIMIT + 1]; |
3c171f0b LP |
1067 | context[CONTEXT_COMM] = comm; |
1068 | context[CONTEXT_EXE] = exe; | |
1069 | ||
1070 | if (cg_pid_get_unit(pid, &t) >= 0) { | |
1071 | ||
c8091d92 LP |
1072 | /* If this is PID 1 disable coredump collection, we'll unlikely be able to process it later on. */ |
1073 | if (streq(t, SPECIAL_INIT_SCOPE)) { | |
1074 | log_notice("Due to PID 1 having crashed coredump collection will now be turned off."); | |
1075 | (void) write_string_file("/proc/sys/kernel/core_pattern", "|/bin/false", 0); | |
1076 | } | |
1077 | ||
78f043f7 LP |
1078 | /* Let's avoid dead-locks when processing journald and init crashes, as socket activation and logging |
1079 | * are unlikely to work then. */ | |
1080 | if (STR_IN_SET(t, SPECIAL_JOURNALD_SERVICE, SPECIAL_INIT_SCOPE)) { | |
3c171f0b | 1081 | free(t); |
78f043f7 | 1082 | return process_special_crash(context, STDIN_FILENO); |
803a3464 LP |
1083 | } |
1084 | ||
63c372cb | 1085 | core_unit = strjoina("COREDUMP_UNIT=", t); |
8c8549db LP |
1086 | free(t); |
1087 | ||
3c171f0b | 1088 | IOVEC_SET_STRING(iovec[n_iovec++], core_unit); |
8c8549db | 1089 | } |
803a3464 | 1090 | |
3c171f0b | 1091 | /* OK, now we know it's not the journal, hence we can make use of it now. */ |
803a3464 LP |
1092 | log_set_target(LOG_TARGET_JOURNAL_OR_KMSG); |
1093 | log_open(); | |
1094 | ||
3c171f0b LP |
1095 | if (cg_pid_get_user_unit(pid, &t) >= 0) { |
1096 | core_user_unit = strjoina("COREDUMP_USER_UNIT=", t); | |
1097 | free(t); | |
1098 | ||
1099 | IOVEC_SET_STRING(iovec[n_iovec++], core_user_unit); | |
1100 | } | |
1101 | ||
1102 | core_pid = strjoina("COREDUMP_PID=", context[CONTEXT_PID]); | |
1103 | IOVEC_SET_STRING(iovec[n_iovec++], core_pid); | |
f5e04665 | 1104 | |
3c171f0b LP |
1105 | core_uid = strjoina("COREDUMP_UID=", context[CONTEXT_UID]); |
1106 | IOVEC_SET_STRING(iovec[n_iovec++], core_uid); | |
f5e04665 | 1107 | |
3c171f0b LP |
1108 | core_gid = strjoina("COREDUMP_GID=", context[CONTEXT_GID]); |
1109 | IOVEC_SET_STRING(iovec[n_iovec++], core_gid); | |
f5e04665 | 1110 | |
3c171f0b LP |
1111 | core_signal = strjoina("COREDUMP_SIGNAL=", context[CONTEXT_SIGNAL]); |
1112 | IOVEC_SET_STRING(iovec[n_iovec++], core_signal); | |
f5e04665 | 1113 | |
bdfd7b2c LP |
1114 | core_rlimit = strjoina("COREDUMP_RLIMIT=", context[CONTEXT_RLIMIT]); |
1115 | IOVEC_SET_STRING(iovec[n_iovec++], core_rlimit); | |
1116 | ||
f5e04665 | 1117 | if (sd_pid_get_session(pid, &t) >= 0) { |
63c372cb | 1118 | core_session = strjoina("COREDUMP_SESSION=", t); |
f5e04665 LP |
1119 | free(t); |
1120 | ||
3c171f0b | 1121 | IOVEC_SET_STRING(iovec[n_iovec++], core_session); |
f5e04665 LP |
1122 | } |
1123 | ||
a035f819 | 1124 | if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) { |
3c171f0b | 1125 | r = asprintf(&core_owner_uid, "COREDUMP_OWNER_UID=" UID_FMT, owner_uid); |
7de80bfe | 1126 | if (r > 0) |
3c171f0b | 1127 | IOVEC_SET_STRING(iovec[n_iovec++], core_owner_uid); |
a035f819 LP |
1128 | } |
1129 | ||
1130 | if (sd_pid_get_slice(pid, &t) >= 0) { | |
63c372cb | 1131 | core_slice = strjoina("COREDUMP_SLICE=", t); |
a035f819 LP |
1132 | free(t); |
1133 | ||
3c171f0b | 1134 | IOVEC_SET_STRING(iovec[n_iovec++], core_slice); |
a035f819 LP |
1135 | } |
1136 | ||
1eef15b1 | 1137 | if (comm) { |
63c372cb | 1138 | core_comm = strjoina("COREDUMP_COMM=", comm); |
3c171f0b | 1139 | IOVEC_SET_STRING(iovec[n_iovec++], core_comm); |
1eef15b1 ZJS |
1140 | } |
1141 | ||
1142 | if (exe) { | |
63c372cb | 1143 | core_exe = strjoina("COREDUMP_EXE=", exe); |
3c171f0b | 1144 | IOVEC_SET_STRING(iovec[n_iovec++], core_exe); |
f5e04665 LP |
1145 | } |
1146 | ||
9bdbc2e2 | 1147 | if (get_process_cmdline(pid, 0, false, &t) >= 0) { |
63c372cb | 1148 | core_cmdline = strjoina("COREDUMP_CMDLINE=", t); |
f5e04665 LP |
1149 | free(t); |
1150 | ||
3c171f0b | 1151 | IOVEC_SET_STRING(iovec[n_iovec++], core_cmdline); |
f5e04665 LP |
1152 | } |
1153 | ||
a035f819 | 1154 | if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0) { |
63c372cb | 1155 | core_cgroup = strjoina("COREDUMP_CGROUP=", t); |
a035f819 LP |
1156 | free(t); |
1157 | ||
3c171f0b | 1158 | IOVEC_SET_STRING(iovec[n_iovec++], core_cgroup); |
a035f819 LP |
1159 | } |
1160 | ||
3f132692 JF |
1161 | if (compose_open_fds(pid, &t) >= 0) { |
1162 | core_open_fds = strappend("COREDUMP_OPEN_FDS=", t); | |
1163 | free(t); | |
1164 | ||
1165 | if (core_open_fds) | |
3c171f0b | 1166 | IOVEC_SET_STRING(iovec[n_iovec++], core_open_fds); |
3f132692 JF |
1167 | } |
1168 | ||
1169 | p = procfs_file_alloca(pid, "status"); | |
1170 | if (read_full_file(p, &t, NULL) >= 0) { | |
1171 | core_proc_status = strappend("COREDUMP_PROC_STATUS=", t); | |
1172 | free(t); | |
1173 | ||
1174 | if (core_proc_status) | |
3c171f0b | 1175 | IOVEC_SET_STRING(iovec[n_iovec++], core_proc_status); |
3f132692 JF |
1176 | } |
1177 | ||
1178 | p = procfs_file_alloca(pid, "maps"); | |
1179 | if (read_full_file(p, &t, NULL) >= 0) { | |
1180 | core_proc_maps = strappend("COREDUMP_PROC_MAPS=", t); | |
1181 | free(t); | |
1182 | ||
1183 | if (core_proc_maps) | |
3c171f0b | 1184 | IOVEC_SET_STRING(iovec[n_iovec++], core_proc_maps); |
3f132692 JF |
1185 | } |
1186 | ||
1187 | p = procfs_file_alloca(pid, "limits"); | |
1188 | if (read_full_file(p, &t, NULL) >= 0) { | |
1189 | core_proc_limits = strappend("COREDUMP_PROC_LIMITS=", t); | |
1190 | free(t); | |
1191 | ||
1192 | if (core_proc_limits) | |
3c171f0b | 1193 | IOVEC_SET_STRING(iovec[n_iovec++], core_proc_limits); |
3f132692 JF |
1194 | } |
1195 | ||
1196 | p = procfs_file_alloca(pid, "cgroup"); | |
1197 | if (read_full_file(p, &t, NULL) >=0) { | |
1198 | core_proc_cgroup = strappend("COREDUMP_PROC_CGROUP=", t); | |
1199 | free(t); | |
1200 | ||
1201 | if (core_proc_cgroup) | |
3c171f0b | 1202 | IOVEC_SET_STRING(iovec[n_iovec++], core_proc_cgroup); |
3f132692 JF |
1203 | } |
1204 | ||
d7032b1f JF |
1205 | p = procfs_file_alloca(pid, "mountinfo"); |
1206 | if (read_full_file(p, &t, NULL) >=0) { | |
1207 | core_proc_mountinfo = strappend("COREDUMP_PROC_MOUNTINFO=", t); | |
1208 | free(t); | |
1209 | ||
1210 | if (core_proc_mountinfo) | |
1211 | IOVEC_SET_STRING(iovec[n_iovec++], core_proc_mountinfo); | |
1212 | } | |
1213 | ||
3f132692 | 1214 | if (get_process_cwd(pid, &t) >= 0) { |
63c372cb | 1215 | core_cwd = strjoina("COREDUMP_CWD=", t); |
3f132692 JF |
1216 | free(t); |
1217 | ||
3c171f0b | 1218 | IOVEC_SET_STRING(iovec[n_iovec++], core_cwd); |
3f132692 JF |
1219 | } |
1220 | ||
1221 | if (get_process_root(pid, &t) >= 0) { | |
63c372cb | 1222 | core_root = strjoina("COREDUMP_ROOT=", t); |
3f132692 | 1223 | |
3c171f0b | 1224 | IOVEC_SET_STRING(iovec[n_iovec++], core_root); |
7ed03ce6 JF |
1225 | |
1226 | /* If the process' root is "/", then there is a chance it has | |
1227 | * mounted own root and hence being containerized. */ | |
1228 | proc_self_root_is_slash = strcmp(t, "/") == 0; | |
1229 | free(t); | |
1230 | if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0) { | |
1231 | core_container_cmdline = strappend("COREDUMP_CONTAINER_CMDLINE=", t); | |
1232 | free(t); | |
1233 | ||
1234 | if (core_container_cmdline) | |
1235 | IOVEC_SET_STRING(iovec[n_iovec++], core_container_cmdline); | |
1236 | } | |
3f132692 JF |
1237 | } |
1238 | ||
1239 | if (get_process_environ(pid, &t) >= 0) { | |
1240 | core_environ = strappend("COREDUMP_ENVIRON=", t); | |
1241 | free(t); | |
1242 | ||
1243 | if (core_environ) | |
3c171f0b | 1244 | IOVEC_SET_STRING(iovec[n_iovec++], core_environ); |
3f132692 JF |
1245 | } |
1246 | ||
81d62103 | 1247 | core_timestamp = strjoina("COREDUMP_TIMESTAMP=", context[CONTEXT_TIMESTAMP], "000000"); |
3c171f0b | 1248 | IOVEC_SET_STRING(iovec[n_iovec++], core_timestamp); |
f5e04665 | 1249 | |
3c171f0b | 1250 | IOVEC_SET_STRING(iovec[n_iovec++], "MESSAGE_ID=fc2e22bc6ee647b6b90729ab34a250b1"); |
4850d39a LP |
1251 | |
1252 | assert_cc(2 == LOG_CRIT); | |
3c171f0b | 1253 | IOVEC_SET_STRING(iovec[n_iovec++], "PRIORITY=2"); |
0dc5d23c | 1254 | |
3c171f0b | 1255 | assert(n_iovec <= ELEMENTSOF(iovec)); |
34c10968 | 1256 | |
3c171f0b LP |
1257 | return send_iovec(iovec, n_iovec, STDIN_FILENO); |
1258 | } | |
34c10968 | 1259 | |
3c171f0b LP |
1260 | int main(int argc, char *argv[]) { |
1261 | int r; | |
fee80f69 | 1262 | |
3c171f0b LP |
1263 | /* First, log to a safe place, since we don't know what crashed and it might be journald which we'd rather not |
1264 | * log to then. */ | |
8d4e028f | 1265 | |
3c171f0b LP |
1266 | log_set_target(LOG_TARGET_KMSG); |
1267 | log_open(); | |
8d4e028f | 1268 | |
3c171f0b LP |
1269 | /* Make sure we never enter a loop */ |
1270 | (void) prctl(PR_SET_DUMPABLE, 0); | |
8d4e028f | 1271 | |
3c171f0b LP |
1272 | /* Ignore all parse errors */ |
1273 | (void) parse_config(); | |
fee80f69 | 1274 | |
3c171f0b LP |
1275 | log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage)); |
1276 | log_debug("Selected compression %s.", yes_no(arg_compress)); | |
fee80f69 | 1277 | |
3c171f0b LP |
1278 | r = sd_listen_fds(false); |
1279 | if (r < 0) { | |
1280 | log_error_errno(r, "Failed to determine number of file descriptor: %m"); | |
1281 | goto finish; | |
fee80f69 LP |
1282 | } |
1283 | ||
3c171f0b LP |
1284 | /* If we got an fd passed, we are running in coredumpd mode. Otherwise we are invoked from the kernel as |
1285 | * coredump handler */ | |
1286 | if (r == 0) | |
1287 | r = process_kernel(argc, argv); | |
1288 | else if (r == 1) | |
1289 | r = process_socket(SD_LISTEN_FDS_START); | |
1290 | else { | |
1291 | log_error("Received unexpected number of file descriptors."); | |
1292 | r = -EINVAL; | |
1293 | } | |
f5e04665 LP |
1294 | |
1295 | finish: | |
f5e04665 LP |
1296 | return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS; |
1297 | } |