]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/file.c | |
4 | * | |
5 | * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes | |
6 | * | |
7 | * Manage the dynamic fd arrays in the process files_struct. | |
8 | */ | |
9 | ||
fe17f22d | 10 | #include <linux/syscalls.h> |
630d9c47 | 11 | #include <linux/export.h> |
1da177e4 | 12 | #include <linux/fs.h> |
278a5fba | 13 | #include <linux/kernel.h> |
1da177e4 | 14 | #include <linux/mm.h> |
3f07c014 | 15 | #include <linux/sched/signal.h> |
1da177e4 | 16 | #include <linux/slab.h> |
1da177e4 | 17 | #include <linux/file.h> |
9f3acc31 | 18 | #include <linux/fdtable.h> |
1da177e4 | 19 | #include <linux/bitops.h> |
ab2af1f5 DS |
20 | #include <linux/spinlock.h> |
21 | #include <linux/rcupdate.h> | |
60997c3d | 22 | #include <linux/close_range.h> |
66590610 | 23 | #include <net/sock.h> |
ab2af1f5 | 24 | |
9b80a184 AD |
25 | unsigned int sysctl_nr_open __read_mostly = 1024*1024; |
26 | unsigned int sysctl_nr_open_min = BITS_PER_LONG; | |
752343be RV |
27 | /* our min() is unusable in constant expressions ;-/ */ |
28 | #define __const_min(x, y) ((x) < (y) ? (x) : (y)) | |
9b80a184 AD |
29 | unsigned int sysctl_nr_open_max = |
30 | __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG; | |
9cfe015a | 31 | |
a892e2d7 | 32 | static void __free_fdtable(struct fdtable *fdt) |
1da177e4 | 33 | { |
f6c0a192 AV |
34 | kvfree(fdt->fd); |
35 | kvfree(fdt->open_fds); | |
a892e2d7 | 36 | kfree(fdt); |
ab2af1f5 | 37 | } |
1da177e4 | 38 | |
7cf4dc3c | 39 | static void free_fdtable_rcu(struct rcu_head *rcu) |
ab2af1f5 | 40 | { |
ac3e3c5b | 41 | __free_fdtable(container_of(rcu, struct fdtable, rcu)); |
ab2af1f5 DS |
42 | } |
43 | ||
f3f86e33 LT |
44 | #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr)) |
45 | #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long)) | |
46 | ||
ab2af1f5 | 47 | /* |
ea5c58e7 EB |
48 | * Copy 'count' fd bits from the old table to the new table and clear the extra |
49 | * space if any. This does not copy the file pointers. Called with the files | |
50 | * spinlock held for write. | |
51 | */ | |
52 | static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt, | |
53 | unsigned int count) | |
54 | { | |
55 | unsigned int cpy, set; | |
56 | ||
57 | cpy = count / BITS_PER_BYTE; | |
58 | set = (nfdt->max_fds - count) / BITS_PER_BYTE; | |
59 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | |
60 | memset((char *)nfdt->open_fds + cpy, 0, set); | |
61 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | |
62 | memset((char *)nfdt->close_on_exec + cpy, 0, set); | |
63 | ||
64 | cpy = BITBIT_SIZE(count); | |
65 | set = BITBIT_SIZE(nfdt->max_fds) - cpy; | |
66 | memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy); | |
67 | memset((char *)nfdt->full_fds_bits + cpy, 0, set); | |
68 | } | |
69 | ||
70 | /* | |
71 | * Copy all file descriptors from the old table to the new, expanded table and | |
72 | * clear the extra space. Called with the files spinlock held for write. | |
ab2af1f5 | 73 | */ |
5466b456 | 74 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
ab2af1f5 | 75 | { |
4e89b721 | 76 | size_t cpy, set; |
ab2af1f5 | 77 | |
5466b456 | 78 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
5466b456 VL |
79 | |
80 | cpy = ofdt->max_fds * sizeof(struct file *); | |
81 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); | |
82 | memcpy(nfdt->fd, ofdt->fd, cpy); | |
ea5c58e7 | 83 | memset((char *)nfdt->fd + cpy, 0, set); |
5466b456 | 84 | |
ea5c58e7 | 85 | copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds); |
1da177e4 LT |
86 | } |
87 | ||
5466b456 | 88 | static struct fdtable * alloc_fdtable(unsigned int nr) |
1da177e4 | 89 | { |
5466b456 | 90 | struct fdtable *fdt; |
1fd36adc | 91 | void *data; |
1da177e4 | 92 | |
ab2af1f5 | 93 | /* |
5466b456 VL |
94 | * Figure out how many fds we actually want to support in this fdtable. |
95 | * Allocation steps are keyed to the size of the fdarray, since it | |
96 | * grows far faster than any of the other dynamic data. We try to fit | |
97 | * the fdarray into comfortable page-tuned chunks: starting at 1024B | |
98 | * and growing in powers of two from there on. | |
ab2af1f5 | 99 | */ |
5466b456 VL |
100 | nr /= (1024 / sizeof(struct file *)); |
101 | nr = roundup_pow_of_two(nr + 1); | |
102 | nr *= (1024 / sizeof(struct file *)); | |
5c598b34 AV |
103 | /* |
104 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open | |
105 | * had been set lower between the check in expand_files() and here. Deal | |
106 | * with that in caller, it's cheaper that way. | |
107 | * | |
108 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise | |
109 | * bitmaps handling below becomes unpleasant, to put it mildly... | |
110 | */ | |
111 | if (unlikely(nr > sysctl_nr_open)) | |
112 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; | |
bbea9f69 | 113 | |
5d097056 | 114 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT); |
5466b456 | 115 | if (!fdt) |
bbea9f69 | 116 | goto out; |
5466b456 | 117 | fdt->max_fds = nr; |
c823bd92 | 118 | data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT); |
5466b456 VL |
119 | if (!data) |
120 | goto out_fdt; | |
1fd36adc DH |
121 | fdt->fd = data; |
122 | ||
c823bd92 MH |
123 | data = kvmalloc(max_t(size_t, |
124 | 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES), | |
125 | GFP_KERNEL_ACCOUNT); | |
5466b456 VL |
126 | if (!data) |
127 | goto out_arr; | |
1fd36adc | 128 | fdt->open_fds = data; |
5466b456 | 129 | data += nr / BITS_PER_BYTE; |
1fd36adc | 130 | fdt->close_on_exec = data; |
f3f86e33 LT |
131 | data += nr / BITS_PER_BYTE; |
132 | fdt->full_fds_bits = data; | |
5466b456 | 133 | |
ab2af1f5 | 134 | return fdt; |
5466b456 VL |
135 | |
136 | out_arr: | |
f6c0a192 | 137 | kvfree(fdt->fd); |
5466b456 | 138 | out_fdt: |
ab2af1f5 | 139 | kfree(fdt); |
5466b456 | 140 | out: |
ab2af1f5 DS |
141 | return NULL; |
142 | } | |
1da177e4 | 143 | |
ab2af1f5 | 144 | /* |
74d392aa VL |
145 | * Expand the file descriptor table. |
146 | * This function will allocate a new fdtable and both fd array and fdset, of | |
147 | * the given size. | |
148 | * Return <0 error code on error; 1 on successful completion. | |
149 | * The files->file_lock should be held on entry, and will be held on exit. | |
ab2af1f5 | 150 | */ |
9b80a184 | 151 | static int expand_fdtable(struct files_struct *files, unsigned int nr) |
ab2af1f5 DS |
152 | __releases(files->file_lock) |
153 | __acquires(files->file_lock) | |
154 | { | |
74d392aa | 155 | struct fdtable *new_fdt, *cur_fdt; |
ab2af1f5 DS |
156 | |
157 | spin_unlock(&files->file_lock); | |
74d392aa | 158 | new_fdt = alloc_fdtable(nr); |
8a81252b | 159 | |
d74ba04d | 160 | /* make sure all fd_install() have seen resize_in_progress |
8a81252b ED |
161 | * or have finished their rcu_read_lock_sched() section. |
162 | */ | |
163 | if (atomic_read(&files->count) > 1) | |
c93ffc15 | 164 | synchronize_rcu(); |
8a81252b | 165 | |
ab2af1f5 | 166 | spin_lock(&files->file_lock); |
74d392aa VL |
167 | if (!new_fdt) |
168 | return -ENOMEM; | |
5c598b34 AV |
169 | /* |
170 | * extremely unlikely race - sysctl_nr_open decreased between the check in | |
171 | * caller and alloc_fdtable(). Cheaper to catch it here... | |
172 | */ | |
173 | if (unlikely(new_fdt->max_fds <= nr)) { | |
a892e2d7 | 174 | __free_fdtable(new_fdt); |
5c598b34 AV |
175 | return -EMFILE; |
176 | } | |
74d392aa | 177 | cur_fdt = files_fdtable(files); |
8a81252b ED |
178 | BUG_ON(nr < cur_fdt->max_fds); |
179 | copy_fdtable(new_fdt, cur_fdt); | |
180 | rcu_assign_pointer(files->fdt, new_fdt); | |
181 | if (cur_fdt != &files->fdtab) | |
182 | call_rcu(&cur_fdt->rcu, free_fdtable_rcu); | |
d74ba04d | 183 | /* coupled with smp_rmb() in fd_install() */ |
8a81252b | 184 | smp_wmb(); |
74d392aa | 185 | return 1; |
1da177e4 LT |
186 | } |
187 | ||
188 | /* | |
189 | * Expand files. | |
74d392aa VL |
190 | * This function will expand the file structures, if the requested size exceeds |
191 | * the current capacity and there is room for expansion. | |
192 | * Return <0 error code on error; 0 when nothing done; 1 when files were | |
193 | * expanded and execution may have blocked. | |
194 | * The files->file_lock should be held on entry, and will be held on exit. | |
1da177e4 | 195 | */ |
9b80a184 | 196 | static int expand_files(struct files_struct *files, unsigned int nr) |
8a81252b ED |
197 | __releases(files->file_lock) |
198 | __acquires(files->file_lock) | |
1da177e4 | 199 | { |
badf1662 | 200 | struct fdtable *fdt; |
8a81252b | 201 | int expanded = 0; |
1da177e4 | 202 | |
8a81252b | 203 | repeat: |
badf1662 | 204 | fdt = files_fdtable(files); |
4e1e018e | 205 | |
74d392aa | 206 | /* Do we need to expand? */ |
bbea9f69 | 207 | if (nr < fdt->max_fds) |
8a81252b | 208 | return expanded; |
4e1e018e | 209 | |
74d392aa | 210 | /* Can we expand? */ |
9cfe015a | 211 | if (nr >= sysctl_nr_open) |
74d392aa VL |
212 | return -EMFILE; |
213 | ||
8a81252b ED |
214 | if (unlikely(files->resize_in_progress)) { |
215 | spin_unlock(&files->file_lock); | |
216 | expanded = 1; | |
217 | wait_event(files->resize_wait, !files->resize_in_progress); | |
218 | spin_lock(&files->file_lock); | |
219 | goto repeat; | |
220 | } | |
221 | ||
74d392aa | 222 | /* All good, so we try */ |
8a81252b ED |
223 | files->resize_in_progress = true; |
224 | expanded = expand_fdtable(files, nr); | |
225 | files->resize_in_progress = false; | |
226 | ||
227 | wake_up_all(&files->resize_wait); | |
228 | return expanded; | |
1da177e4 | 229 | } |
ab2af1f5 | 230 | |
9b80a184 | 231 | static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
232 | { |
233 | __set_bit(fd, fdt->close_on_exec); | |
234 | } | |
235 | ||
9b80a184 | 236 | static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt) |
b8318b01 | 237 | { |
fc90888d LT |
238 | if (test_bit(fd, fdt->close_on_exec)) |
239 | __clear_bit(fd, fdt->close_on_exec); | |
b8318b01 AV |
240 | } |
241 | ||
f3f86e33 | 242 | static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
243 | { |
244 | __set_bit(fd, fdt->open_fds); | |
f3f86e33 LT |
245 | fd /= BITS_PER_LONG; |
246 | if (!~fdt->open_fds[fd]) | |
247 | __set_bit(fd, fdt->full_fds_bits); | |
b8318b01 AV |
248 | } |
249 | ||
f3f86e33 | 250 | static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
251 | { |
252 | __clear_bit(fd, fdt->open_fds); | |
f3f86e33 | 253 | __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits); |
b8318b01 AV |
254 | } |
255 | ||
9b80a184 | 256 | static unsigned int count_open_files(struct fdtable *fdt) |
02afc626 | 257 | { |
9b80a184 AD |
258 | unsigned int size = fdt->max_fds; |
259 | unsigned int i; | |
02afc626 AV |
260 | |
261 | /* Find the last open fd */ | |
1fd36adc DH |
262 | for (i = size / BITS_PER_LONG; i > 0; ) { |
263 | if (fdt->open_fds[--i]) | |
02afc626 AV |
264 | break; |
265 | } | |
1fd36adc | 266 | i = (i + 1) * BITS_PER_LONG; |
02afc626 AV |
267 | return i; |
268 | } | |
269 | ||
60997c3d CB |
270 | static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds) |
271 | { | |
272 | unsigned int count; | |
273 | ||
274 | count = count_open_files(fdt); | |
275 | if (max_fds < NR_OPEN_DEFAULT) | |
276 | max_fds = NR_OPEN_DEFAULT; | |
277 | return min(count, max_fds); | |
278 | } | |
279 | ||
02afc626 AV |
280 | /* |
281 | * Allocate a new files structure and copy contents from the | |
282 | * passed in files structure. | |
283 | * errorp will be valid only when the returned files_struct is NULL. | |
284 | */ | |
60997c3d | 285 | struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp) |
02afc626 AV |
286 | { |
287 | struct files_struct *newf; | |
288 | struct file **old_fds, **new_fds; | |
9b80a184 | 289 | unsigned int open_files, i; |
02afc626 AV |
290 | struct fdtable *old_fdt, *new_fdt; |
291 | ||
292 | *errorp = -ENOMEM; | |
afbec7ff | 293 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
02afc626 AV |
294 | if (!newf) |
295 | goto out; | |
296 | ||
afbec7ff AV |
297 | atomic_set(&newf->count, 1); |
298 | ||
299 | spin_lock_init(&newf->file_lock); | |
8a81252b ED |
300 | newf->resize_in_progress = false; |
301 | init_waitqueue_head(&newf->resize_wait); | |
afbec7ff AV |
302 | newf->next_fd = 0; |
303 | new_fdt = &newf->fdtab; | |
304 | new_fdt->max_fds = NR_OPEN_DEFAULT; | |
1fd36adc DH |
305 | new_fdt->close_on_exec = newf->close_on_exec_init; |
306 | new_fdt->open_fds = newf->open_fds_init; | |
f3f86e33 | 307 | new_fdt->full_fds_bits = newf->full_fds_bits_init; |
afbec7ff | 308 | new_fdt->fd = &newf->fd_array[0]; |
afbec7ff | 309 | |
02afc626 AV |
310 | spin_lock(&oldf->file_lock); |
311 | old_fdt = files_fdtable(oldf); | |
60997c3d | 312 | open_files = sane_fdtable_size(old_fdt, max_fds); |
02afc626 AV |
313 | |
314 | /* | |
315 | * Check whether we need to allocate a larger fd array and fd set. | |
02afc626 | 316 | */ |
adbecb12 | 317 | while (unlikely(open_files > new_fdt->max_fds)) { |
02afc626 | 318 | spin_unlock(&oldf->file_lock); |
9dec3c4d | 319 | |
a892e2d7 CG |
320 | if (new_fdt != &newf->fdtab) |
321 | __free_fdtable(new_fdt); | |
adbecb12 | 322 | |
9dec3c4d AV |
323 | new_fdt = alloc_fdtable(open_files - 1); |
324 | if (!new_fdt) { | |
325 | *errorp = -ENOMEM; | |
326 | goto out_release; | |
327 | } | |
328 | ||
329 | /* beyond sysctl_nr_open; nothing to do */ | |
330 | if (unlikely(new_fdt->max_fds < open_files)) { | |
a892e2d7 | 331 | __free_fdtable(new_fdt); |
9dec3c4d | 332 | *errorp = -EMFILE; |
02afc626 | 333 | goto out_release; |
9dec3c4d | 334 | } |
9dec3c4d | 335 | |
02afc626 AV |
336 | /* |
337 | * Reacquire the oldf lock and a pointer to its fd table | |
338 | * who knows it may have a new bigger fd table. We need | |
339 | * the latest pointer. | |
340 | */ | |
341 | spin_lock(&oldf->file_lock); | |
342 | old_fdt = files_fdtable(oldf); | |
60997c3d | 343 | open_files = sane_fdtable_size(old_fdt, max_fds); |
02afc626 AV |
344 | } |
345 | ||
ea5c58e7 EB |
346 | copy_fd_bitmaps(new_fdt, old_fdt, open_files); |
347 | ||
02afc626 AV |
348 | old_fds = old_fdt->fd; |
349 | new_fds = new_fdt->fd; | |
350 | ||
02afc626 AV |
351 | for (i = open_files; i != 0; i--) { |
352 | struct file *f = *old_fds++; | |
353 | if (f) { | |
354 | get_file(f); | |
355 | } else { | |
356 | /* | |
357 | * The fd may be claimed in the fd bitmap but not yet | |
358 | * instantiated in the files array if a sibling thread | |
359 | * is partway through open(). So make sure that this | |
360 | * fd is available to the new process. | |
361 | */ | |
1dce27c5 | 362 | __clear_open_fd(open_files - i, new_fdt); |
02afc626 AV |
363 | } |
364 | rcu_assign_pointer(*new_fds++, f); | |
365 | } | |
366 | spin_unlock(&oldf->file_lock); | |
367 | ||
ea5c58e7 EB |
368 | /* clear the remainder */ |
369 | memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *)); | |
02afc626 | 370 | |
afbec7ff AV |
371 | rcu_assign_pointer(newf->fdt, new_fdt); |
372 | ||
02afc626 AV |
373 | return newf; |
374 | ||
375 | out_release: | |
376 | kmem_cache_free(files_cachep, newf); | |
377 | out: | |
378 | return NULL; | |
379 | } | |
380 | ||
ce08b62d | 381 | static struct fdtable *close_files(struct files_struct * files) |
7cf4dc3c | 382 | { |
7cf4dc3c AV |
383 | /* |
384 | * It is safe to dereference the fd table without RCU or | |
385 | * ->file_lock because this is the last reference to the | |
ce08b62d | 386 | * files structure. |
7cf4dc3c | 387 | */ |
ce08b62d | 388 | struct fdtable *fdt = rcu_dereference_raw(files->fdt); |
9b80a184 | 389 | unsigned int i, j = 0; |
ce08b62d | 390 | |
7cf4dc3c AV |
391 | for (;;) { |
392 | unsigned long set; | |
393 | i = j * BITS_PER_LONG; | |
394 | if (i >= fdt->max_fds) | |
395 | break; | |
396 | set = fdt->open_fds[j++]; | |
397 | while (set) { | |
398 | if (set & 1) { | |
399 | struct file * file = xchg(&fdt->fd[i], NULL); | |
400 | if (file) { | |
401 | filp_close(file, files); | |
388a4c88 | 402 | cond_resched(); |
7cf4dc3c AV |
403 | } |
404 | } | |
405 | i++; | |
406 | set >>= 1; | |
407 | } | |
408 | } | |
ce08b62d ON |
409 | |
410 | return fdt; | |
7cf4dc3c AV |
411 | } |
412 | ||
7cf4dc3c AV |
413 | void put_files_struct(struct files_struct *files) |
414 | { | |
7cf4dc3c | 415 | if (atomic_dec_and_test(&files->count)) { |
ce08b62d ON |
416 | struct fdtable *fdt = close_files(files); |
417 | ||
b9e02af0 AV |
418 | /* free the arrays if they are not embedded */ |
419 | if (fdt != &files->fdtab) | |
420 | __free_fdtable(fdt); | |
421 | kmem_cache_free(files_cachep, files); | |
7cf4dc3c AV |
422 | } |
423 | } | |
424 | ||
7cf4dc3c AV |
425 | void exit_files(struct task_struct *tsk) |
426 | { | |
427 | struct files_struct * files = tsk->files; | |
428 | ||
429 | if (files) { | |
430 | task_lock(tsk); | |
431 | tsk->files = NULL; | |
432 | task_unlock(tsk); | |
433 | put_files_struct(files); | |
434 | } | |
435 | } | |
436 | ||
f52111b1 AV |
437 | struct files_struct init_files = { |
438 | .count = ATOMIC_INIT(1), | |
439 | .fdt = &init_files.fdtab, | |
440 | .fdtab = { | |
441 | .max_fds = NR_OPEN_DEFAULT, | |
442 | .fd = &init_files.fd_array[0], | |
1fd36adc DH |
443 | .close_on_exec = init_files.close_on_exec_init, |
444 | .open_fds = init_files.open_fds_init, | |
f3f86e33 | 445 | .full_fds_bits = init_files.full_fds_bits_init, |
f52111b1 | 446 | }, |
eece09ec | 447 | .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock), |
5704a068 | 448 | .resize_wait = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait), |
f52111b1 | 449 | }; |
1027abe8 | 450 | |
9b80a184 | 451 | static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start) |
f3f86e33 | 452 | { |
9b80a184 AD |
453 | unsigned int maxfd = fdt->max_fds; |
454 | unsigned int maxbit = maxfd / BITS_PER_LONG; | |
455 | unsigned int bitbit = start / BITS_PER_LONG; | |
f3f86e33 LT |
456 | |
457 | bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG; | |
458 | if (bitbit > maxfd) | |
459 | return maxfd; | |
460 | if (bitbit > start) | |
461 | start = bitbit; | |
462 | return find_next_zero_bit(fdt->open_fds, maxfd, start); | |
463 | } | |
464 | ||
1027abe8 AV |
465 | /* |
466 | * allocate a file descriptor, mark it busy. | |
467 | */ | |
aa384d10 | 468 | static int alloc_fd(unsigned start, unsigned end, unsigned flags) |
1027abe8 | 469 | { |
aa384d10 | 470 | struct files_struct *files = current->files; |
1027abe8 AV |
471 | unsigned int fd; |
472 | int error; | |
473 | struct fdtable *fdt; | |
474 | ||
475 | spin_lock(&files->file_lock); | |
476 | repeat: | |
477 | fdt = files_fdtable(files); | |
478 | fd = start; | |
479 | if (fd < files->next_fd) | |
480 | fd = files->next_fd; | |
481 | ||
482 | if (fd < fdt->max_fds) | |
f3f86e33 | 483 | fd = find_next_fd(fdt, fd); |
1027abe8 | 484 | |
f33ff992 AV |
485 | /* |
486 | * N.B. For clone tasks sharing a files structure, this test | |
487 | * will limit the total number of files that can be opened. | |
488 | */ | |
489 | error = -EMFILE; | |
490 | if (fd >= end) | |
491 | goto out; | |
492 | ||
1027abe8 AV |
493 | error = expand_files(files, fd); |
494 | if (error < 0) | |
495 | goto out; | |
496 | ||
497 | /* | |
498 | * If we needed to expand the fs array we | |
499 | * might have blocked - try again. | |
500 | */ | |
501 | if (error) | |
502 | goto repeat; | |
503 | ||
504 | if (start <= files->next_fd) | |
505 | files->next_fd = fd + 1; | |
506 | ||
1dce27c5 | 507 | __set_open_fd(fd, fdt); |
1027abe8 | 508 | if (flags & O_CLOEXEC) |
1dce27c5 | 509 | __set_close_on_exec(fd, fdt); |
1027abe8 | 510 | else |
1dce27c5 | 511 | __clear_close_on_exec(fd, fdt); |
1027abe8 AV |
512 | error = fd; |
513 | #if 1 | |
514 | /* Sanity check */ | |
add1f099 | 515 | if (rcu_access_pointer(fdt->fd[fd]) != NULL) { |
1027abe8 AV |
516 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
517 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
518 | } | |
519 | #endif | |
520 | ||
521 | out: | |
522 | spin_unlock(&files->file_lock); | |
523 | return error; | |
524 | } | |
525 | ||
4022e7af JA |
526 | int __get_unused_fd_flags(unsigned flags, unsigned long nofile) |
527 | { | |
aa384d10 | 528 | return alloc_fd(0, nofile, flags); |
4022e7af JA |
529 | } |
530 | ||
1a7bd226 | 531 | int get_unused_fd_flags(unsigned flags) |
1027abe8 | 532 | { |
4022e7af | 533 | return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE)); |
1027abe8 | 534 | } |
1a7bd226 | 535 | EXPORT_SYMBOL(get_unused_fd_flags); |
56007cae AV |
536 | |
537 | static void __put_unused_fd(struct files_struct *files, unsigned int fd) | |
538 | { | |
539 | struct fdtable *fdt = files_fdtable(files); | |
540 | __clear_open_fd(fd, fdt); | |
541 | if (fd < files->next_fd) | |
542 | files->next_fd = fd; | |
543 | } | |
544 | ||
545 | void put_unused_fd(unsigned int fd) | |
546 | { | |
547 | struct files_struct *files = current->files; | |
548 | spin_lock(&files->file_lock); | |
549 | __put_unused_fd(files, fd); | |
550 | spin_unlock(&files->file_lock); | |
551 | } | |
552 | ||
553 | EXPORT_SYMBOL(put_unused_fd); | |
554 | ||
555 | /* | |
556 | * Install a file pointer in the fd array. | |
557 | * | |
558 | * The VFS is full of places where we drop the files lock between | |
559 | * setting the open_fds bitmap and installing the file in the file | |
560 | * array. At any such point, we are vulnerable to a dup2() race | |
561 | * installing a file in the array before us. We need to detect this and | |
562 | * fput() the struct file we are about to overwrite in this case. | |
563 | * | |
564 | * It should never happen - if we allow dup2() do it, _really_ bad things | |
565 | * will follow. | |
f869e8a7 | 566 | * |
d74ba04d EB |
567 | * This consumes the "file" refcount, so callers should treat it |
568 | * as if they had called fput(file). | |
56007cae AV |
569 | */ |
570 | ||
d74ba04d | 571 | void fd_install(unsigned int fd, struct file *file) |
56007cae | 572 | { |
d74ba04d | 573 | struct files_struct *files = current->files; |
56007cae | 574 | struct fdtable *fdt; |
8a81252b | 575 | |
8a81252b ED |
576 | rcu_read_lock_sched(); |
577 | ||
c02b1a9b | 578 | if (unlikely(files->resize_in_progress)) { |
8a81252b | 579 | rcu_read_unlock_sched(); |
c02b1a9b MG |
580 | spin_lock(&files->file_lock); |
581 | fdt = files_fdtable(files); | |
582 | BUG_ON(fdt->fd[fd] != NULL); | |
583 | rcu_assign_pointer(fdt->fd[fd], file); | |
584 | spin_unlock(&files->file_lock); | |
585 | return; | |
8a81252b ED |
586 | } |
587 | /* coupled with smp_wmb() in expand_fdtable() */ | |
588 | smp_rmb(); | |
589 | fdt = rcu_dereference_sched(files->fdt); | |
56007cae AV |
590 | BUG_ON(fdt->fd[fd] != NULL); |
591 | rcu_assign_pointer(fdt->fd[fd], file); | |
8a81252b | 592 | rcu_read_unlock_sched(); |
56007cae AV |
593 | } |
594 | ||
595 | EXPORT_SYMBOL(fd_install); | |
0ee8cdfe | 596 | |
278a5fba | 597 | static struct file *pick_file(struct files_struct *files, unsigned fd) |
483ce1d4 | 598 | { |
278a5fba | 599 | struct file *file = NULL; |
483ce1d4 AV |
600 | struct fdtable *fdt; |
601 | ||
602 | spin_lock(&files->file_lock); | |
603 | fdt = files_fdtable(files); | |
604 | if (fd >= fdt->max_fds) | |
605 | goto out_unlock; | |
606 | file = fdt->fd[fd]; | |
607 | if (!file) | |
608 | goto out_unlock; | |
609 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
483ce1d4 | 610 | __put_unused_fd(files, fd); |
483ce1d4 AV |
611 | |
612 | out_unlock: | |
613 | spin_unlock(&files->file_lock); | |
278a5fba CB |
614 | return file; |
615 | } | |
616 | ||
8760c909 | 617 | int close_fd(unsigned fd) |
278a5fba | 618 | { |
8760c909 | 619 | struct files_struct *files = current->files; |
278a5fba CB |
620 | struct file *file; |
621 | ||
622 | file = pick_file(files, fd); | |
623 | if (!file) | |
624 | return -EBADF; | |
625 | ||
626 | return filp_close(file, files); | |
483ce1d4 | 627 | } |
8760c909 | 628 | EXPORT_SYMBOL(close_fd); /* for ksys_close() */ |
483ce1d4 | 629 | |
582f1fb6 GS |
630 | static inline void __range_cloexec(struct files_struct *cur_fds, |
631 | unsigned int fd, unsigned int max_fd) | |
632 | { | |
633 | struct fdtable *fdt; | |
634 | ||
635 | if (fd > max_fd) | |
636 | return; | |
637 | ||
638 | spin_lock(&cur_fds->file_lock); | |
639 | fdt = files_fdtable(cur_fds); | |
640 | bitmap_set(fdt->close_on_exec, fd, max_fd - fd + 1); | |
641 | spin_unlock(&cur_fds->file_lock); | |
642 | } | |
643 | ||
644 | static inline void __range_close(struct files_struct *cur_fds, unsigned int fd, | |
645 | unsigned int max_fd) | |
646 | { | |
647 | while (fd <= max_fd) { | |
648 | struct file *file; | |
649 | ||
650 | file = pick_file(cur_fds, fd++); | |
651 | if (!file) | |
652 | continue; | |
653 | ||
654 | filp_close(file, cur_fds); | |
655 | cond_resched(); | |
656 | } | |
657 | } | |
658 | ||
278a5fba CB |
659 | /** |
660 | * __close_range() - Close all file descriptors in a given range. | |
661 | * | |
662 | * @fd: starting file descriptor to close | |
663 | * @max_fd: last file descriptor to close | |
664 | * | |
665 | * This closes a range of file descriptors. All file descriptors | |
666 | * from @fd up to and including @max_fd are closed. | |
667 | */ | |
60997c3d | 668 | int __close_range(unsigned fd, unsigned max_fd, unsigned int flags) |
278a5fba CB |
669 | { |
670 | unsigned int cur_max; | |
60997c3d CB |
671 | struct task_struct *me = current; |
672 | struct files_struct *cur_fds = me->files, *fds = NULL; | |
673 | ||
582f1fb6 | 674 | if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC)) |
60997c3d | 675 | return -EINVAL; |
278a5fba CB |
676 | |
677 | if (fd > max_fd) | |
678 | return -EINVAL; | |
679 | ||
680 | rcu_read_lock(); | |
60997c3d | 681 | cur_max = files_fdtable(cur_fds)->max_fds; |
278a5fba CB |
682 | rcu_read_unlock(); |
683 | ||
684 | /* cap to last valid index into fdtable */ | |
685 | cur_max--; | |
686 | ||
60997c3d CB |
687 | if (flags & CLOSE_RANGE_UNSHARE) { |
688 | int ret; | |
689 | unsigned int max_unshare_fds = NR_OPEN_MAX; | |
690 | ||
691 | /* | |
692 | * If the requested range is greater than the current maximum, | |
693 | * we're closing everything so only copy all file descriptors | |
694 | * beneath the lowest file descriptor. | |
fec8a6a6 CB |
695 | * If the caller requested all fds to be made cloexec copy all |
696 | * of the file descriptors since they still want to use them. | |
60997c3d | 697 | */ |
fec8a6a6 | 698 | if (!(flags & CLOSE_RANGE_CLOEXEC) && (max_fd >= cur_max)) |
60997c3d CB |
699 | max_unshare_fds = fd; |
700 | ||
701 | ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds); | |
702 | if (ret) | |
703 | return ret; | |
704 | ||
705 | /* | |
706 | * We used to share our file descriptor table, and have now | |
707 | * created a private one, make sure we're using it below. | |
708 | */ | |
709 | if (fds) | |
710 | swap(cur_fds, fds); | |
711 | } | |
712 | ||
278a5fba | 713 | max_fd = min(max_fd, cur_max); |
278a5fba | 714 | |
582f1fb6 GS |
715 | if (flags & CLOSE_RANGE_CLOEXEC) |
716 | __range_cloexec(cur_fds, fd, max_fd); | |
717 | else | |
718 | __range_close(cur_fds, fd, max_fd); | |
278a5fba | 719 | |
60997c3d CB |
720 | if (fds) { |
721 | /* | |
722 | * We're done closing the files we were supposed to. Time to install | |
723 | * the new file descriptor table and drop the old one. | |
724 | */ | |
725 | task_lock(me); | |
726 | me->files = cur_fds; | |
727 | task_unlock(me); | |
728 | put_files_struct(fds); | |
729 | } | |
730 | ||
278a5fba CB |
731 | return 0; |
732 | } | |
733 | ||
80cd7956 | 734 | /* |
9fe83c43 | 735 | * variant of close_fd that gets a ref on the file for later fput. |
6e802a4b JA |
736 | * The caller must ensure that filp_close() called on the file, and then |
737 | * an fput(). | |
80cd7956 | 738 | */ |
9fe83c43 | 739 | int close_fd_get_file(unsigned int fd, struct file **res) |
80cd7956 TK |
740 | { |
741 | struct files_struct *files = current->files; | |
742 | struct file *file; | |
743 | struct fdtable *fdt; | |
744 | ||
745 | spin_lock(&files->file_lock); | |
746 | fdt = files_fdtable(files); | |
747 | if (fd >= fdt->max_fds) | |
748 | goto out_unlock; | |
749 | file = fdt->fd[fd]; | |
750 | if (!file) | |
751 | goto out_unlock; | |
752 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
753 | __put_unused_fd(files, fd); | |
754 | spin_unlock(&files->file_lock); | |
755 | get_file(file); | |
756 | *res = file; | |
6e802a4b | 757 | return 0; |
80cd7956 TK |
758 | |
759 | out_unlock: | |
760 | spin_unlock(&files->file_lock); | |
761 | *res = NULL; | |
762 | return -ENOENT; | |
763 | } | |
764 | ||
6a6d27de AV |
765 | void do_close_on_exec(struct files_struct *files) |
766 | { | |
767 | unsigned i; | |
768 | struct fdtable *fdt; | |
769 | ||
770 | /* exec unshares first */ | |
6a6d27de AV |
771 | spin_lock(&files->file_lock); |
772 | for (i = 0; ; i++) { | |
773 | unsigned long set; | |
774 | unsigned fd = i * BITS_PER_LONG; | |
775 | fdt = files_fdtable(files); | |
776 | if (fd >= fdt->max_fds) | |
777 | break; | |
778 | set = fdt->close_on_exec[i]; | |
779 | if (!set) | |
780 | continue; | |
781 | fdt->close_on_exec[i] = 0; | |
782 | for ( ; set ; fd++, set >>= 1) { | |
783 | struct file *file; | |
784 | if (!(set & 1)) | |
785 | continue; | |
786 | file = fdt->fd[fd]; | |
787 | if (!file) | |
788 | continue; | |
789 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
790 | __put_unused_fd(files, fd); | |
791 | spin_unlock(&files->file_lock); | |
792 | filp_close(file, files); | |
793 | cond_resched(); | |
794 | spin_lock(&files->file_lock); | |
795 | } | |
796 | ||
797 | } | |
798 | spin_unlock(&files->file_lock); | |
799 | } | |
800 | ||
5e876fb4 SD |
801 | static struct file *__fget_files(struct files_struct *files, unsigned int fd, |
802 | fmode_t mask, unsigned int refs) | |
0ee8cdfe | 803 | { |
1deb46e2 | 804 | struct file *file; |
0ee8cdfe AV |
805 | |
806 | rcu_read_lock(); | |
5ba97d28 | 807 | loop: |
f36c2943 | 808 | file = files_lookup_fd_rcu(files, fd); |
0ee8cdfe | 809 | if (file) { |
5ba97d28 ED |
810 | /* File object ref couldn't be taken. |
811 | * dup2() atomicity guarantee is the reason | |
812 | * we loop to catch the new file (or NULL pointer) | |
813 | */ | |
814 | if (file->f_mode & mask) | |
0ee8cdfe | 815 | file = NULL; |
091141a4 | 816 | else if (!get_file_rcu_many(file, refs)) |
5ba97d28 | 817 | goto loop; |
0ee8cdfe AV |
818 | } |
819 | rcu_read_unlock(); | |
820 | ||
821 | return file; | |
822 | } | |
823 | ||
5e876fb4 SD |
824 | static inline struct file *__fget(unsigned int fd, fmode_t mask, |
825 | unsigned int refs) | |
826 | { | |
827 | return __fget_files(current->files, fd, mask, refs); | |
828 | } | |
829 | ||
091141a4 JA |
830 | struct file *fget_many(unsigned int fd, unsigned int refs) |
831 | { | |
832 | return __fget(fd, FMODE_PATH, refs); | |
833 | } | |
834 | ||
1deb46e2 ON |
835 | struct file *fget(unsigned int fd) |
836 | { | |
091141a4 | 837 | return __fget(fd, FMODE_PATH, 1); |
1deb46e2 | 838 | } |
0ee8cdfe AV |
839 | EXPORT_SYMBOL(fget); |
840 | ||
841 | struct file *fget_raw(unsigned int fd) | |
842 | { | |
091141a4 | 843 | return __fget(fd, 0, 1); |
0ee8cdfe | 844 | } |
0ee8cdfe AV |
845 | EXPORT_SYMBOL(fget_raw); |
846 | ||
5e876fb4 SD |
847 | struct file *fget_task(struct task_struct *task, unsigned int fd) |
848 | { | |
849 | struct file *file = NULL; | |
850 | ||
851 | task_lock(task); | |
852 | if (task->files) | |
853 | file = __fget_files(task->files, fd, 0, 1); | |
854 | task_unlock(task); | |
855 | ||
856 | return file; | |
857 | } | |
858 | ||
3a879fb3 EB |
859 | struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd) |
860 | { | |
861 | /* Must be called with rcu_read_lock held */ | |
862 | struct files_struct *files; | |
863 | struct file *file = NULL; | |
864 | ||
865 | task_lock(task); | |
866 | files = task->files; | |
867 | if (files) | |
868 | file = files_lookup_fd_rcu(files, fd); | |
869 | task_unlock(task); | |
870 | ||
871 | return file; | |
872 | } | |
873 | ||
e9a53aeb EB |
874 | struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *ret_fd) |
875 | { | |
876 | /* Must be called with rcu_read_lock held */ | |
877 | struct files_struct *files; | |
878 | unsigned int fd = *ret_fd; | |
879 | struct file *file = NULL; | |
880 | ||
881 | task_lock(task); | |
882 | files = task->files; | |
883 | if (files) { | |
884 | for (; fd < files_fdtable(files)->max_fds; fd++) { | |
885 | file = files_lookup_fd_rcu(files, fd); | |
886 | if (file) | |
887 | break; | |
888 | } | |
889 | } | |
890 | task_unlock(task); | |
891 | *ret_fd = fd; | |
892 | return file; | |
893 | } | |
894 | ||
0ee8cdfe AV |
895 | /* |
896 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
897 | * | |
898 | * You can use this instead of fget if you satisfy all of the following | |
899 | * conditions: | |
900 | * 1) You must call fput_light before exiting the syscall and returning control | |
901 | * to userspace (i.e. you cannot remember the returned struct file * after | |
902 | * returning to userspace). | |
903 | * 2) You must not call filp_close on the returned struct file * in between | |
904 | * calls to fget_light and fput_light. | |
905 | * 3) You must not clone the current task in between the calls to fget_light | |
906 | * and fput_light. | |
907 | * | |
908 | * The fput_needed flag returned by fget_light should be passed to the | |
909 | * corresponding fput_light. | |
910 | */ | |
bd2a31d5 | 911 | static unsigned long __fget_light(unsigned int fd, fmode_t mask) |
0ee8cdfe | 912 | { |
0ee8cdfe | 913 | struct files_struct *files = current->files; |
ad461834 | 914 | struct file *file; |
0ee8cdfe | 915 | |
0ee8cdfe | 916 | if (atomic_read(&files->count) == 1) { |
bebf684b | 917 | file = files_lookup_fd_raw(files, fd); |
bd2a31d5 AV |
918 | if (!file || unlikely(file->f_mode & mask)) |
919 | return 0; | |
920 | return (unsigned long)file; | |
0ee8cdfe | 921 | } else { |
091141a4 | 922 | file = __fget(fd, mask, 1); |
bd2a31d5 AV |
923 | if (!file) |
924 | return 0; | |
925 | return FDPUT_FPUT | (unsigned long)file; | |
0ee8cdfe | 926 | } |
0ee8cdfe | 927 | } |
bd2a31d5 | 928 | unsigned long __fdget(unsigned int fd) |
ad461834 | 929 | { |
bd2a31d5 | 930 | return __fget_light(fd, FMODE_PATH); |
ad461834 | 931 | } |
bd2a31d5 | 932 | EXPORT_SYMBOL(__fdget); |
0ee8cdfe | 933 | |
bd2a31d5 | 934 | unsigned long __fdget_raw(unsigned int fd) |
0ee8cdfe | 935 | { |
bd2a31d5 | 936 | return __fget_light(fd, 0); |
0ee8cdfe | 937 | } |
fe17f22d | 938 | |
bd2a31d5 AV |
939 | unsigned long __fdget_pos(unsigned int fd) |
940 | { | |
99aea681 EB |
941 | unsigned long v = __fdget(fd); |
942 | struct file *file = (struct file *)(v & ~3); | |
bd2a31d5 | 943 | |
2be7d348 | 944 | if (file && (file->f_mode & FMODE_ATOMIC_POS)) { |
bd2a31d5 AV |
945 | if (file_count(file) > 1) { |
946 | v |= FDPUT_POS_UNLOCK; | |
947 | mutex_lock(&file->f_pos_lock); | |
948 | } | |
949 | } | |
99aea681 | 950 | return v; |
bd2a31d5 AV |
951 | } |
952 | ||
63b6df14 AV |
953 | void __f_unlock_pos(struct file *f) |
954 | { | |
955 | mutex_unlock(&f->f_pos_lock); | |
956 | } | |
957 | ||
bd2a31d5 AV |
958 | /* |
959 | * We only lock f_pos if we have threads or if the file might be | |
960 | * shared with another process. In both cases we'll have an elevated | |
961 | * file count (done either by fdget() or by fork()). | |
962 | */ | |
963 | ||
fe17f22d AV |
964 | void set_close_on_exec(unsigned int fd, int flag) |
965 | { | |
966 | struct files_struct *files = current->files; | |
967 | struct fdtable *fdt; | |
968 | spin_lock(&files->file_lock); | |
969 | fdt = files_fdtable(files); | |
970 | if (flag) | |
971 | __set_close_on_exec(fd, fdt); | |
972 | else | |
973 | __clear_close_on_exec(fd, fdt); | |
974 | spin_unlock(&files->file_lock); | |
975 | } | |
976 | ||
977 | bool get_close_on_exec(unsigned int fd) | |
978 | { | |
979 | struct files_struct *files = current->files; | |
980 | struct fdtable *fdt; | |
981 | bool res; | |
982 | rcu_read_lock(); | |
983 | fdt = files_fdtable(files); | |
984 | res = close_on_exec(fd, fdt); | |
985 | rcu_read_unlock(); | |
986 | return res; | |
987 | } | |
988 | ||
8280d161 AV |
989 | static int do_dup2(struct files_struct *files, |
990 | struct file *file, unsigned fd, unsigned flags) | |
e983094d | 991 | __releases(&files->file_lock) |
fe17f22d | 992 | { |
8280d161 | 993 | struct file *tofree; |
fe17f22d AV |
994 | struct fdtable *fdt; |
995 | ||
fe17f22d AV |
996 | /* |
997 | * We need to detect attempts to do dup2() over allocated but still | |
998 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
999 | * extra work in their equivalent of fget() - they insert struct | |
1000 | * file immediately after grabbing descriptor, mark it larval if | |
1001 | * more work (e.g. actual opening) is needed and make sure that | |
1002 | * fget() treats larval files as absent. Potentially interesting, | |
1003 | * but while extra work in fget() is trivial, locking implications | |
1004 | * and amount of surgery on open()-related paths in VFS are not. | |
1005 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
1006 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
1007 | * scope of POSIX or SUS, since neither considers shared descriptor | |
1008 | * tables and this condition does not arise without those. | |
1009 | */ | |
fe17f22d | 1010 | fdt = files_fdtable(files); |
8280d161 AV |
1011 | tofree = fdt->fd[fd]; |
1012 | if (!tofree && fd_is_open(fd, fdt)) | |
1013 | goto Ebusy; | |
fe17f22d | 1014 | get_file(file); |
8280d161 AV |
1015 | rcu_assign_pointer(fdt->fd[fd], file); |
1016 | __set_open_fd(fd, fdt); | |
fe17f22d | 1017 | if (flags & O_CLOEXEC) |
8280d161 | 1018 | __set_close_on_exec(fd, fdt); |
fe17f22d | 1019 | else |
8280d161 | 1020 | __clear_close_on_exec(fd, fdt); |
fe17f22d AV |
1021 | spin_unlock(&files->file_lock); |
1022 | ||
1023 | if (tofree) | |
1024 | filp_close(tofree, files); | |
1025 | ||
8280d161 AV |
1026 | return fd; |
1027 | ||
1028 | Ebusy: | |
1029 | spin_unlock(&files->file_lock); | |
1030 | return -EBUSY; | |
1031 | } | |
1032 | ||
1033 | int replace_fd(unsigned fd, struct file *file, unsigned flags) | |
1034 | { | |
1035 | int err; | |
1036 | struct files_struct *files = current->files; | |
1037 | ||
1038 | if (!file) | |
8760c909 | 1039 | return close_fd(fd); |
8280d161 AV |
1040 | |
1041 | if (fd >= rlimit(RLIMIT_NOFILE)) | |
08f05c49 | 1042 | return -EBADF; |
8280d161 AV |
1043 | |
1044 | spin_lock(&files->file_lock); | |
1045 | err = expand_files(files, fd); | |
1046 | if (unlikely(err < 0)) | |
1047 | goto out_unlock; | |
1048 | return do_dup2(files, file, fd, flags); | |
1049 | ||
1050 | out_unlock: | |
1051 | spin_unlock(&files->file_lock); | |
1052 | return err; | |
1053 | } | |
1054 | ||
66590610 KC |
1055 | /** |
1056 | * __receive_fd() - Install received file into file descriptor table | |
1057 | * | |
17381715 | 1058 | * @fd: fd to install into (if negative, a new fd will be allocated) |
66590610 KC |
1059 | * @file: struct file that was received from another process |
1060 | * @ufd: __user pointer to write new fd number to | |
1061 | * @o_flags: the O_* flags to apply to the new fd entry | |
1062 | * | |
1063 | * Installs a received file into the file descriptor table, with appropriate | |
deefa7f3 KC |
1064 | * checks and count updates. Optionally writes the fd number to userspace, if |
1065 | * @ufd is non-NULL. | |
66590610 KC |
1066 | * |
1067 | * This helper handles its own reference counting of the incoming | |
1068 | * struct file. | |
1069 | * | |
deefa7f3 | 1070 | * Returns newly install fd or -ve on error. |
66590610 | 1071 | */ |
17381715 | 1072 | int __receive_fd(int fd, struct file *file, int __user *ufd, unsigned int o_flags) |
66590610 KC |
1073 | { |
1074 | int new_fd; | |
1075 | int error; | |
1076 | ||
1077 | error = security_file_receive(file); | |
1078 | if (error) | |
1079 | return error; | |
1080 | ||
17381715 KC |
1081 | if (fd < 0) { |
1082 | new_fd = get_unused_fd_flags(o_flags); | |
1083 | if (new_fd < 0) | |
1084 | return new_fd; | |
1085 | } else { | |
1086 | new_fd = fd; | |
1087 | } | |
66590610 | 1088 | |
deefa7f3 KC |
1089 | if (ufd) { |
1090 | error = put_user(new_fd, ufd); | |
1091 | if (error) { | |
17381715 KC |
1092 | if (fd < 0) |
1093 | put_unused_fd(new_fd); | |
deefa7f3 KC |
1094 | return error; |
1095 | } | |
66590610 KC |
1096 | } |
1097 | ||
17381715 KC |
1098 | if (fd < 0) { |
1099 | fd_install(new_fd, get_file(file)); | |
1100 | } else { | |
1101 | error = replace_fd(new_fd, file, o_flags); | |
1102 | if (error) | |
1103 | return error; | |
1104 | } | |
1105 | ||
66590610 KC |
1106 | /* Bump the sock usage counts, if any. */ |
1107 | __receive_sock(file); | |
deefa7f3 | 1108 | return new_fd; |
66590610 KC |
1109 | } |
1110 | ||
c7248321 | 1111 | static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags) |
8280d161 AV |
1112 | { |
1113 | int err = -EBADF; | |
1114 | struct file *file; | |
1115 | struct files_struct *files = current->files; | |
1116 | ||
1117 | if ((flags & ~O_CLOEXEC) != 0) | |
1118 | return -EINVAL; | |
1119 | ||
aed97647 RJ |
1120 | if (unlikely(oldfd == newfd)) |
1121 | return -EINVAL; | |
1122 | ||
8280d161 | 1123 | if (newfd >= rlimit(RLIMIT_NOFILE)) |
08f05c49 | 1124 | return -EBADF; |
8280d161 AV |
1125 | |
1126 | spin_lock(&files->file_lock); | |
1127 | err = expand_files(files, newfd); | |
120ce2b0 | 1128 | file = files_lookup_fd_locked(files, oldfd); |
8280d161 AV |
1129 | if (unlikely(!file)) |
1130 | goto Ebadf; | |
1131 | if (unlikely(err < 0)) { | |
1132 | if (err == -EMFILE) | |
1133 | goto Ebadf; | |
1134 | goto out_unlock; | |
1135 | } | |
1136 | return do_dup2(files, file, newfd, flags); | |
fe17f22d AV |
1137 | |
1138 | Ebadf: | |
1139 | err = -EBADF; | |
1140 | out_unlock: | |
1141 | spin_unlock(&files->file_lock); | |
1142 | return err; | |
1143 | } | |
1144 | ||
c7248321 DB |
1145 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) |
1146 | { | |
1147 | return ksys_dup3(oldfd, newfd, flags); | |
1148 | } | |
1149 | ||
fe17f22d AV |
1150 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) |
1151 | { | |
1152 | if (unlikely(newfd == oldfd)) { /* corner case */ | |
1153 | struct files_struct *files = current->files; | |
1154 | int retval = oldfd; | |
1155 | ||
1156 | rcu_read_lock(); | |
f36c2943 | 1157 | if (!files_lookup_fd_rcu(files, oldfd)) |
fe17f22d AV |
1158 | retval = -EBADF; |
1159 | rcu_read_unlock(); | |
1160 | return retval; | |
1161 | } | |
c7248321 | 1162 | return ksys_dup3(oldfd, newfd, 0); |
fe17f22d AV |
1163 | } |
1164 | ||
bc1cd99a | 1165 | SYSCALL_DEFINE1(dup, unsigned int, fildes) |
fe17f22d AV |
1166 | { |
1167 | int ret = -EBADF; | |
1168 | struct file *file = fget_raw(fildes); | |
1169 | ||
1170 | if (file) { | |
8d10a035 | 1171 | ret = get_unused_fd_flags(0); |
fe17f22d AV |
1172 | if (ret >= 0) |
1173 | fd_install(ret, file); | |
1174 | else | |
1175 | fput(file); | |
1176 | } | |
1177 | return ret; | |
1178 | } | |
1179 | ||
1180 | int f_dupfd(unsigned int from, struct file *file, unsigned flags) | |
1181 | { | |
e06b53c2 | 1182 | unsigned long nofile = rlimit(RLIMIT_NOFILE); |
fe17f22d | 1183 | int err; |
e06b53c2 | 1184 | if (from >= nofile) |
fe17f22d | 1185 | return -EINVAL; |
e06b53c2 | 1186 | err = alloc_fd(from, nofile, flags); |
fe17f22d AV |
1187 | if (err >= 0) { |
1188 | get_file(file); | |
1189 | fd_install(err, file); | |
1190 | } | |
1191 | return err; | |
1192 | } | |
c3c073f8 AV |
1193 | |
1194 | int iterate_fd(struct files_struct *files, unsigned n, | |
1195 | int (*f)(const void *, struct file *, unsigned), | |
1196 | const void *p) | |
1197 | { | |
1198 | struct fdtable *fdt; | |
c3c073f8 AV |
1199 | int res = 0; |
1200 | if (!files) | |
1201 | return 0; | |
1202 | spin_lock(&files->file_lock); | |
a77cfcb4 AV |
1203 | for (fdt = files_fdtable(files); n < fdt->max_fds; n++) { |
1204 | struct file *file; | |
1205 | file = rcu_dereference_check_fdtable(files, fdt->fd[n]); | |
1206 | if (!file) | |
1207 | continue; | |
1208 | res = f(p, file, n); | |
1209 | if (res) | |
1210 | break; | |
c3c073f8 AV |
1211 | } |
1212 | spin_unlock(&files->file_lock); | |
1213 | return res; | |
1214 | } | |
1215 | EXPORT_SYMBOL(iterate_fd); |