summaryrefslogtreecommitdiffstats
path: root/src/shared/data-fd-util.c
blob: adc7d7417e0f3459931353e09507a398122cc59a (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
/* SPDX-License-Identifier: LGPL-2.1-or-later */

#include <fcntl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#if HAVE_LINUX_MEMFD_H
#include <linux/memfd.h>
#endif

#include "alloc-util.h"
#include "copy.h"
#include "data-fd-util.h"
#include "fd-util.h"
#include "fs-util.h"
#include "io-util.h"
#include "memfd-util.h"
#include "missing_mman.h"
#include "missing_syscall.h"
#include "tmpfile-util.h"

/* When the data is smaller or equal to 64K, try to place the copy in a memfd/pipe */
#define DATA_FD_MEMORY_LIMIT (64U * U64_KB)

/* If memfd/pipe didn't work out, then let's use a file in /tmp up to a size of 1M. If it's large than that use /var/tmp instead. */
#define DATA_FD_TMP_LIMIT (1U * U64_MB)

int acquire_data_fd_full(const void *data, size_t size, DataFDFlags flags) {
        _cleanup_close_ int fd = -EBADF;
        ssize_t n;
        int r;

        assert(data || size == 0);

        /* Acquire a read-only file descriptor that when read from returns the specified data. This is much more
         * complex than I wish it was. But here's why:
         *
         * a) First we try to use memfds. They are the best option, as we can seal them nicely to make them
         *    read-only. Unfortunately they require kernel 3.17, and – at the time of writing – we still support 3.14.
         *
         * b) Then, we try classic pipes. They are the second best options, as we can close the writing side, retaining
         *    a nicely read-only fd in the reading side. However, they are by default quite small, and unprivileged
         *    clients can only bump their size to a system-wide limit, which might be quite low.
         *
         * c) Then, we try an O_TMPFILE file in /dev/shm (that dir is the only suitable one known to exist from
         *    earliest boot on). To make it read-only we open the fd a second time with O_RDONLY via
         *    /proc/self/<fd>. Unfortunately O_TMPFILE is not available on older kernels on tmpfs.
         *
         * d) Finally, we try creating a regular file in /dev/shm, which we then delete.
         *
         * It sucks a bit that depending on the situation we return very different objects here, but that's Linux I
         * figure. */

        if (size == SIZE_MAX)
                size = strlen(data);

        if (size == 0 && !FLAGS_SET(flags, ACQUIRE_NO_DEV_NULL))
                /* As a special case, return /dev/null if we have been called for an empty data block */
                return RET_NERRNO(open("/dev/null", O_RDONLY|O_CLOEXEC|O_NOCTTY));

        if (!FLAGS_SET(flags, ACQUIRE_NO_MEMFD)) {
                fd = memfd_new_and_seal("data-fd", data, size);
                if (fd < 0 && !ERRNO_IS_NOT_SUPPORTED(fd))
                        return fd;
                if (fd >= 0)
                        return TAKE_FD(fd);
        }

        if (!FLAGS_SET(flags, ACQUIRE_NO_PIPE)) {
                _cleanup_close_pair_ int pipefds[2] = EBADF_PAIR;
                int isz;

                if (pipe2(pipefds, O_CLOEXEC|O_NONBLOCK) < 0)
                        return -errno;

                isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0);
                if (isz < 0)
                        return -errno;

                if ((size_t) isz < size) {
                        isz = (int) size;
                        if (isz < 0 || (size_t) isz != size)
                                return -E2BIG;

                        /* Try to bump the pipe size */
                        (void) fcntl(pipefds[1], F_SETPIPE_SZ, isz);

                        /* See if that worked */
                        isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0);
                        if (isz < 0)
                                return -errno;

                        if ((size_t) isz < size)
                                goto try_dev_shm;
                }

                n = write(pipefds[1], data, size);
                if (n < 0)
                        return -errno;
                if ((size_t) n != size)
                        return -EIO;

                (void) fd_nonblock(pipefds[0], false);

                return TAKE_FD(pipefds[0]);
        }

try_dev_shm:
        if (!FLAGS_SET(flags, ACQUIRE_NO_TMPFILE)) {
                fd = open("/dev/shm", O_RDWR|O_TMPFILE|O_CLOEXEC, 0500);
                if (fd < 0)
                        goto try_dev_shm_without_o_tmpfile;

                n = write(fd, data, size);
                if (n < 0)
                        return -errno;
                if ((size_t) n != size)
                        return -EIO;

                /* Let's reopen the thing, in order to get an O_RDONLY fd for the original O_RDWR one */
                return fd_reopen(fd, O_RDONLY|O_CLOEXEC);
        }

try_dev_shm_without_o_tmpfile:
        if (!FLAGS_SET(flags, ACQUIRE_NO_REGULAR)) {
                char pattern[] = "/dev/shm/data-fd-XXXXXX";

                fd = mkostemp_safe(pattern);
                if (fd < 0)
                        return fd;

                n = write(fd, data, size);
                if (n < 0) {
                        r = -errno;
                        goto unlink_and_return;
                }
                if ((size_t) n != size) {
                        r = -EIO;
                        goto unlink_and_return;
                }

                /* Let's reopen the thing, in order to get an O_RDONLY fd for the original O_RDWR one */
                r = fd_reopen(fd, O_RDONLY|O_CLOEXEC);

        unlink_and_return:
                (void) unlink(pattern);
                return r;
        }

        return -EOPNOTSUPP;
}

int copy_data_fd(int fd) {
        _cleanup_close_ int copy_fd = -EBADF, tmp_fd = -EBADF;
        _cleanup_free_ void *remains = NULL;
        size_t remains_size = 0;
        const char *td;
        struct stat st;
        int r;

        /* Creates a 'data' fd from the specified source fd, containing all the same data in a read-only fashion, but
         * independent of it (i.e. the source fd can be closed and unmounted after this call succeeded). Tries to be
         * somewhat smart about where to place the data. In the best case uses a memfd(). If memfd() are not supported
         * uses a pipe instead. For larger data will use an unlinked file in /tmp, and for even larger data one in
         * /var/tmp. */

        if (fstat(fd, &st) < 0)
                return -errno;

        /* For now, let's only accept regular files, sockets, pipes and char devices */
        if (S_ISDIR(st.st_mode))
                return -EISDIR;
        if (S_ISLNK(st.st_mode))
                return -ELOOP;
        if (!S_ISREG(st.st_mode) && !S_ISSOCK(st.st_mode) && !S_ISFIFO(st.st_mode) && !S_ISCHR(st.st_mode))
                return -EBADFD;

        /* If we have reason to believe the data is bounded in size, then let's use memfds or pipes as backing fd. Note
         * that we use the reported regular file size only as a hint, given that there are plenty special files in
         * /proc and /sys which report a zero file size but can be read from. */

        if (!S_ISREG(st.st_mode) || (uint64_t) st.st_size < DATA_FD_MEMORY_LIMIT) {

                /* Try a memfd first */
                copy_fd = memfd_new("data-fd");
                if (copy_fd >= 0) {
                        off_t f;

                        r = copy_bytes(fd, copy_fd, DATA_FD_MEMORY_LIMIT, 0);
                        if (r < 0)
                                return r;

                        f = lseek(copy_fd, 0, SEEK_SET);
                        if (f != 0)
                                return -errno;

                        if (r == 0) {
                                /* Did it fit into the limit? If so, we are done. */
                                r = memfd_set_sealed(copy_fd);
                                if (r < 0)
                                        return r;

                                return TAKE_FD(copy_fd);
                        }

                        /* Hmm, pity, this didn't fit. Let's fall back to /tmp then, see below */

                } else {
                        _cleanup_close_pair_ int pipefds[2] = EBADF_PAIR;
                        int isz;

                        /* If memfds aren't available, use a pipe. Set O_NONBLOCK so that we will get EAGAIN rather
                         * then block indefinitely when we hit the pipe size limit */

                        if (pipe2(pipefds, O_CLOEXEC|O_NONBLOCK) < 0)
                                return -errno;

                        isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0);
                        if (isz < 0)
                                return -errno;

                        /* Try to enlarge the pipe size if necessary */
                        if ((size_t) isz < DATA_FD_MEMORY_LIMIT) {

                                (void) fcntl(pipefds[1], F_SETPIPE_SZ, DATA_FD_MEMORY_LIMIT);

                                isz = fcntl(pipefds[1], F_GETPIPE_SZ, 0);
                                if (isz < 0)
                                        return -errno;
                        }

                        if ((size_t) isz >= DATA_FD_MEMORY_LIMIT) {

                                r = copy_bytes_full(fd, pipefds[1], DATA_FD_MEMORY_LIMIT, 0, &remains, &remains_size, NULL, NULL);
                                if (r < 0 && r != -EAGAIN)
                                        return r; /* If we get EAGAIN it could be because of the source or because of
                                                   * the destination fd, we can't know, as sendfile() and friends won't
                                                   * tell us. Hence, treat this as reason to fall back, just to be
                                                   * sure. */
                                if (r == 0) {
                                        /* Everything fit in, yay! */
                                        (void) fd_nonblock(pipefds[0], false);

                                        return TAKE_FD(pipefds[0]);
                                }

                                /* Things didn't fit in. But we read data into the pipe, let's remember that, so that
                                 * when writing the new file we incorporate this first. */
                                copy_fd = TAKE_FD(pipefds[0]);
                        }
                }
        }

        /* If we have reason to believe this will fit fine in /tmp, then use that as first fallback. */
        if ((!S_ISREG(st.st_mode) || (uint64_t) st.st_size < DATA_FD_TMP_LIMIT) &&
            (DATA_FD_MEMORY_LIMIT + remains_size) < DATA_FD_TMP_LIMIT) {
                off_t f;

                tmp_fd = open_tmpfile_unlinkable(NULL /* NULL as directory means /tmp */, O_RDWR|O_CLOEXEC);
                if (tmp_fd < 0)
                        return tmp_fd;

                if (copy_fd >= 0) {
                        /* If we tried a memfd/pipe first and it ended up being too large, then copy this into the
                         * temporary file first. */

                        r = copy_bytes(copy_fd, tmp_fd, UINT64_MAX, 0);
                        if (r < 0)
                                return r;

                        assert(r == 0);
                }

                if (remains_size > 0) {
                        /* If there were remaining bytes (i.e. read into memory, but not written out yet) from the
                         * failed copy operation, let's flush them out next. */

                        r = loop_write(tmp_fd, remains, remains_size);
                        if (r < 0)
                                return r;
                }

                r = copy_bytes(fd, tmp_fd, DATA_FD_TMP_LIMIT - DATA_FD_MEMORY_LIMIT - remains_size, COPY_REFLINK);
                if (r < 0)
                        return r;
                if (r == 0)
                        goto finish;  /* Yay, it fit in */

                /* It didn't fit in. Let's not forget to use what we already used */
                f = lseek(tmp_fd, 0, SEEK_SET);
                if (f != 0)
                        return -errno;

                close_and_replace(copy_fd, tmp_fd);

                remains = mfree(remains);
                remains_size = 0;
        }

        /* As last fallback use /var/tmp */
        r = var_tmp_dir(&td);
        if (r < 0)
                return r;

        tmp_fd = open_tmpfile_unlinkable(td, O_RDWR|O_CLOEXEC);
        if (tmp_fd < 0)
                return tmp_fd;

        if (copy_fd >= 0) {
                /* If we tried a memfd/pipe first, or a file in /tmp, and it ended up being too large, than copy this
                 * into the temporary file first. */
                r = copy_bytes(copy_fd, tmp_fd, UINT64_MAX, COPY_REFLINK);
                if (r < 0)
                        return r;

                assert(r == 0);
        }

        if (remains_size > 0) {
                /* Then, copy in any read but not yet written bytes. */
                r = loop_write(tmp_fd, remains, remains_size);
                if (r < 0)
                        return r;
        }

        /* Copy in the rest */
        r = copy_bytes(fd, tmp_fd, UINT64_MAX, COPY_REFLINK);
        if (r < 0)
                return r;

        assert(r == 0);

finish:
        /* Now convert the O_RDWR file descriptor into an O_RDONLY one (and as side effect seek to the beginning of the
         * file again */

        return fd_reopen(tmp_fd, O_RDONLY|O_CLOEXEC);
}

int memfd_clone_fd(int fd, const char *name, int mode) {
        _cleanup_close_ int mfd = -EBADF;
        struct stat st;
        bool ro, exec;
        int r;

        /* Creates a clone of a regular file in a memfd. Unlike copy_data_fd() this returns strictly a memfd
         * (and if it can't it will fail). Thus the resulting fd is seekable, and definitely reports as
         * S_ISREG. */

        assert(fd >= 0);
        assert(name);
        assert(IN_SET(mode & O_ACCMODE, O_RDONLY, O_RDWR));
        assert((mode & ~(O_RDONLY|O_RDWR|O_CLOEXEC)) == 0);

        if (fstat(fd, &st) < 0)
                return -errno;

        ro = (mode & O_ACCMODE) == O_RDONLY;
        exec = st.st_mode & 0111;

        mfd = memfd_create_wrapper(name,
                                   ((FLAGS_SET(mode, O_CLOEXEC) || ro) ? MFD_CLOEXEC : 0) |
                                   (ro ? MFD_ALLOW_SEALING : 0) |
                                   (exec ? MFD_EXEC : MFD_NOEXEC_SEAL));
        if (mfd < 0)
                return mfd;

        r = copy_bytes(fd, mfd, UINT64_MAX, COPY_REFLINK);
        if (r < 0)
                return r;

        if (ro) {
                _cleanup_close_ int rfd = -EBADF;

                r = memfd_set_sealed(mfd);
                if (r < 0)
                        return r;

                rfd = fd_reopen(mfd, mode);
                if (rfd < 0)
                        return rfd;

                return TAKE_FD(rfd);
        }

        off_t f = lseek(mfd, 0, SEEK_SET);
        if (f < 0)
                return -errno;

        return TAKE_FD(mfd);
}