]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/file.c | |
3 | * | |
4 | * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes | |
5 | * | |
6 | * Manage the dynamic fd arrays in the process files_struct. | |
7 | */ | |
8 | ||
fe17f22d | 9 | #include <linux/syscalls.h> |
630d9c47 | 10 | #include <linux/export.h> |
1da177e4 LT |
11 | #include <linux/fs.h> |
12 | #include <linux/mm.h> | |
6d4831c2 | 13 | #include <linux/mmzone.h> |
1da177e4 | 14 | #include <linux/time.h> |
3f07c014 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/slab.h> |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/file.h> | |
9f3acc31 | 19 | #include <linux/fdtable.h> |
1da177e4 | 20 | #include <linux/bitops.h> |
ab2af1f5 DS |
21 | #include <linux/interrupt.h> |
22 | #include <linux/spinlock.h> | |
23 | #include <linux/rcupdate.h> | |
24 | #include <linux/workqueue.h> | |
25 | ||
9b80a184 AD |
26 | unsigned int sysctl_nr_open __read_mostly = 1024*1024; |
27 | unsigned int sysctl_nr_open_min = BITS_PER_LONG; | |
752343be RV |
28 | /* our min() is unusable in constant expressions ;-/ */ |
29 | #define __const_min(x, y) ((x) < (y) ? (x) : (y)) | |
9b80a184 AD |
30 | unsigned int sysctl_nr_open_max = |
31 | __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG; | |
9cfe015a | 32 | |
a892e2d7 | 33 | static void __free_fdtable(struct fdtable *fdt) |
1da177e4 | 34 | { |
f6c0a192 AV |
35 | kvfree(fdt->fd); |
36 | kvfree(fdt->open_fds); | |
a892e2d7 | 37 | kfree(fdt); |
ab2af1f5 | 38 | } |
1da177e4 | 39 | |
7cf4dc3c | 40 | static void free_fdtable_rcu(struct rcu_head *rcu) |
ab2af1f5 | 41 | { |
ac3e3c5b | 42 | __free_fdtable(container_of(rcu, struct fdtable, rcu)); |
ab2af1f5 DS |
43 | } |
44 | ||
f3f86e33 LT |
45 | #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr)) |
46 | #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long)) | |
47 | ||
ab2af1f5 | 48 | /* |
ea5c58e7 EB |
49 | * Copy 'count' fd bits from the old table to the new table and clear the extra |
50 | * space if any. This does not copy the file pointers. Called with the files | |
51 | * spinlock held for write. | |
52 | */ | |
53 | static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt, | |
54 | unsigned int count) | |
55 | { | |
56 | unsigned int cpy, set; | |
57 | ||
58 | cpy = count / BITS_PER_BYTE; | |
59 | set = (nfdt->max_fds - count) / BITS_PER_BYTE; | |
60 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | |
61 | memset((char *)nfdt->open_fds + cpy, 0, set); | |
62 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | |
63 | memset((char *)nfdt->close_on_exec + cpy, 0, set); | |
64 | ||
65 | cpy = BITBIT_SIZE(count); | |
66 | set = BITBIT_SIZE(nfdt->max_fds) - cpy; | |
67 | memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy); | |
68 | memset((char *)nfdt->full_fds_bits + cpy, 0, set); | |
69 | } | |
70 | ||
71 | /* | |
72 | * Copy all file descriptors from the old table to the new, expanded table and | |
73 | * clear the extra space. Called with the files spinlock held for write. | |
ab2af1f5 | 74 | */ |
5466b456 | 75 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
ab2af1f5 | 76 | { |
5466b456 | 77 | unsigned int cpy, set; |
ab2af1f5 | 78 | |
5466b456 | 79 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
5466b456 VL |
80 | |
81 | cpy = ofdt->max_fds * sizeof(struct file *); | |
82 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); | |
83 | memcpy(nfdt->fd, ofdt->fd, cpy); | |
ea5c58e7 | 84 | memset((char *)nfdt->fd + cpy, 0, set); |
5466b456 | 85 | |
ea5c58e7 | 86 | copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds); |
1da177e4 LT |
87 | } |
88 | ||
5466b456 | 89 | static struct fdtable * alloc_fdtable(unsigned int nr) |
1da177e4 | 90 | { |
5466b456 | 91 | struct fdtable *fdt; |
1fd36adc | 92 | void *data; |
1da177e4 | 93 | |
ab2af1f5 | 94 | /* |
5466b456 VL |
95 | * Figure out how many fds we actually want to support in this fdtable. |
96 | * Allocation steps are keyed to the size of the fdarray, since it | |
97 | * grows far faster than any of the other dynamic data. We try to fit | |
98 | * the fdarray into comfortable page-tuned chunks: starting at 1024B | |
99 | * and growing in powers of two from there on. | |
ab2af1f5 | 100 | */ |
5466b456 VL |
101 | nr /= (1024 / sizeof(struct file *)); |
102 | nr = roundup_pow_of_two(nr + 1); | |
103 | nr *= (1024 / sizeof(struct file *)); | |
5c598b34 AV |
104 | /* |
105 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open | |
106 | * had been set lower between the check in expand_files() and here. Deal | |
107 | * with that in caller, it's cheaper that way. | |
108 | * | |
109 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise | |
110 | * bitmaps handling below becomes unpleasant, to put it mildly... | |
111 | */ | |
112 | if (unlikely(nr > sysctl_nr_open)) | |
113 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; | |
bbea9f69 | 114 | |
5d097056 | 115 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT); |
5466b456 | 116 | if (!fdt) |
bbea9f69 | 117 | goto out; |
5466b456 | 118 | fdt->max_fds = nr; |
c823bd92 | 119 | data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT); |
5466b456 VL |
120 | if (!data) |
121 | goto out_fdt; | |
1fd36adc DH |
122 | fdt->fd = data; |
123 | ||
c823bd92 MH |
124 | data = kvmalloc(max_t(size_t, |
125 | 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES), | |
126 | GFP_KERNEL_ACCOUNT); | |
5466b456 VL |
127 | if (!data) |
128 | goto out_arr; | |
1fd36adc | 129 | fdt->open_fds = data; |
5466b456 | 130 | data += nr / BITS_PER_BYTE; |
1fd36adc | 131 | fdt->close_on_exec = data; |
f3f86e33 LT |
132 | data += nr / BITS_PER_BYTE; |
133 | fdt->full_fds_bits = data; | |
5466b456 | 134 | |
ab2af1f5 | 135 | return fdt; |
5466b456 VL |
136 | |
137 | out_arr: | |
f6c0a192 | 138 | kvfree(fdt->fd); |
5466b456 | 139 | out_fdt: |
ab2af1f5 | 140 | kfree(fdt); |
5466b456 | 141 | out: |
ab2af1f5 DS |
142 | return NULL; |
143 | } | |
1da177e4 | 144 | |
ab2af1f5 | 145 | /* |
74d392aa VL |
146 | * Expand the file descriptor table. |
147 | * This function will allocate a new fdtable and both fd array and fdset, of | |
148 | * the given size. | |
149 | * Return <0 error code on error; 1 on successful completion. | |
150 | * The files->file_lock should be held on entry, and will be held on exit. | |
ab2af1f5 | 151 | */ |
9b80a184 | 152 | static int expand_fdtable(struct files_struct *files, unsigned int nr) |
ab2af1f5 DS |
153 | __releases(files->file_lock) |
154 | __acquires(files->file_lock) | |
155 | { | |
74d392aa | 156 | struct fdtable *new_fdt, *cur_fdt; |
ab2af1f5 DS |
157 | |
158 | spin_unlock(&files->file_lock); | |
74d392aa | 159 | new_fdt = alloc_fdtable(nr); |
8a81252b ED |
160 | |
161 | /* make sure all __fd_install() have seen resize_in_progress | |
162 | * or have finished their rcu_read_lock_sched() section. | |
163 | */ | |
164 | if (atomic_read(&files->count) > 1) | |
165 | synchronize_sched(); | |
166 | ||
ab2af1f5 | 167 | spin_lock(&files->file_lock); |
74d392aa VL |
168 | if (!new_fdt) |
169 | return -ENOMEM; | |
5c598b34 AV |
170 | /* |
171 | * extremely unlikely race - sysctl_nr_open decreased between the check in | |
172 | * caller and alloc_fdtable(). Cheaper to catch it here... | |
173 | */ | |
174 | if (unlikely(new_fdt->max_fds <= nr)) { | |
a892e2d7 | 175 | __free_fdtable(new_fdt); |
5c598b34 AV |
176 | return -EMFILE; |
177 | } | |
74d392aa | 178 | cur_fdt = files_fdtable(files); |
8a81252b ED |
179 | BUG_ON(nr < cur_fdt->max_fds); |
180 | copy_fdtable(new_fdt, cur_fdt); | |
181 | rcu_assign_pointer(files->fdt, new_fdt); | |
182 | if (cur_fdt != &files->fdtab) | |
183 | call_rcu(&cur_fdt->rcu, free_fdtable_rcu); | |
184 | /* coupled with smp_rmb() in __fd_install() */ | |
185 | smp_wmb(); | |
74d392aa | 186 | return 1; |
1da177e4 LT |
187 | } |
188 | ||
189 | /* | |
190 | * Expand files. | |
74d392aa VL |
191 | * This function will expand the file structures, if the requested size exceeds |
192 | * the current capacity and there is room for expansion. | |
193 | * Return <0 error code on error; 0 when nothing done; 1 when files were | |
194 | * expanded and execution may have blocked. | |
195 | * The files->file_lock should be held on entry, and will be held on exit. | |
1da177e4 | 196 | */ |
9b80a184 | 197 | static int expand_files(struct files_struct *files, unsigned int nr) |
8a81252b ED |
198 | __releases(files->file_lock) |
199 | __acquires(files->file_lock) | |
1da177e4 | 200 | { |
badf1662 | 201 | struct fdtable *fdt; |
8a81252b | 202 | int expanded = 0; |
1da177e4 | 203 | |
8a81252b | 204 | repeat: |
badf1662 | 205 | fdt = files_fdtable(files); |
4e1e018e | 206 | |
74d392aa | 207 | /* Do we need to expand? */ |
bbea9f69 | 208 | if (nr < fdt->max_fds) |
8a81252b | 209 | return expanded; |
4e1e018e | 210 | |
74d392aa | 211 | /* Can we expand? */ |
9cfe015a | 212 | if (nr >= sysctl_nr_open) |
74d392aa VL |
213 | return -EMFILE; |
214 | ||
8a81252b ED |
215 | if (unlikely(files->resize_in_progress)) { |
216 | spin_unlock(&files->file_lock); | |
217 | expanded = 1; | |
218 | wait_event(files->resize_wait, !files->resize_in_progress); | |
219 | spin_lock(&files->file_lock); | |
220 | goto repeat; | |
221 | } | |
222 | ||
74d392aa | 223 | /* All good, so we try */ |
8a81252b ED |
224 | files->resize_in_progress = true; |
225 | expanded = expand_fdtable(files, nr); | |
226 | files->resize_in_progress = false; | |
227 | ||
228 | wake_up_all(&files->resize_wait); | |
229 | return expanded; | |
1da177e4 | 230 | } |
ab2af1f5 | 231 | |
9b80a184 | 232 | static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
233 | { |
234 | __set_bit(fd, fdt->close_on_exec); | |
235 | } | |
236 | ||
9b80a184 | 237 | static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt) |
b8318b01 | 238 | { |
fc90888d LT |
239 | if (test_bit(fd, fdt->close_on_exec)) |
240 | __clear_bit(fd, fdt->close_on_exec); | |
b8318b01 AV |
241 | } |
242 | ||
f3f86e33 | 243 | static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
244 | { |
245 | __set_bit(fd, fdt->open_fds); | |
f3f86e33 LT |
246 | fd /= BITS_PER_LONG; |
247 | if (!~fdt->open_fds[fd]) | |
248 | __set_bit(fd, fdt->full_fds_bits); | |
b8318b01 AV |
249 | } |
250 | ||
f3f86e33 | 251 | static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt) |
b8318b01 AV |
252 | { |
253 | __clear_bit(fd, fdt->open_fds); | |
f3f86e33 | 254 | __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits); |
b8318b01 AV |
255 | } |
256 | ||
9b80a184 | 257 | static unsigned int count_open_files(struct fdtable *fdt) |
02afc626 | 258 | { |
9b80a184 AD |
259 | unsigned int size = fdt->max_fds; |
260 | unsigned int i; | |
02afc626 AV |
261 | |
262 | /* Find the last open fd */ | |
1fd36adc DH |
263 | for (i = size / BITS_PER_LONG; i > 0; ) { |
264 | if (fdt->open_fds[--i]) | |
02afc626 AV |
265 | break; |
266 | } | |
1fd36adc | 267 | i = (i + 1) * BITS_PER_LONG; |
02afc626 AV |
268 | return i; |
269 | } | |
270 | ||
02afc626 AV |
271 | /* |
272 | * Allocate a new files structure and copy contents from the | |
273 | * passed in files structure. | |
274 | * errorp will be valid only when the returned files_struct is NULL. | |
275 | */ | |
276 | struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) | |
277 | { | |
278 | struct files_struct *newf; | |
279 | struct file **old_fds, **new_fds; | |
9b80a184 | 280 | unsigned int open_files, i; |
02afc626 AV |
281 | struct fdtable *old_fdt, *new_fdt; |
282 | ||
283 | *errorp = -ENOMEM; | |
afbec7ff | 284 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
02afc626 AV |
285 | if (!newf) |
286 | goto out; | |
287 | ||
afbec7ff AV |
288 | atomic_set(&newf->count, 1); |
289 | ||
290 | spin_lock_init(&newf->file_lock); | |
8a81252b ED |
291 | newf->resize_in_progress = false; |
292 | init_waitqueue_head(&newf->resize_wait); | |
afbec7ff AV |
293 | newf->next_fd = 0; |
294 | new_fdt = &newf->fdtab; | |
295 | new_fdt->max_fds = NR_OPEN_DEFAULT; | |
1fd36adc DH |
296 | new_fdt->close_on_exec = newf->close_on_exec_init; |
297 | new_fdt->open_fds = newf->open_fds_init; | |
f3f86e33 | 298 | new_fdt->full_fds_bits = newf->full_fds_bits_init; |
afbec7ff | 299 | new_fdt->fd = &newf->fd_array[0]; |
afbec7ff | 300 | |
02afc626 AV |
301 | spin_lock(&oldf->file_lock); |
302 | old_fdt = files_fdtable(oldf); | |
02afc626 AV |
303 | open_files = count_open_files(old_fdt); |
304 | ||
305 | /* | |
306 | * Check whether we need to allocate a larger fd array and fd set. | |
02afc626 | 307 | */ |
adbecb12 | 308 | while (unlikely(open_files > new_fdt->max_fds)) { |
02afc626 | 309 | spin_unlock(&oldf->file_lock); |
9dec3c4d | 310 | |
a892e2d7 CG |
311 | if (new_fdt != &newf->fdtab) |
312 | __free_fdtable(new_fdt); | |
adbecb12 | 313 | |
9dec3c4d AV |
314 | new_fdt = alloc_fdtable(open_files - 1); |
315 | if (!new_fdt) { | |
316 | *errorp = -ENOMEM; | |
317 | goto out_release; | |
318 | } | |
319 | ||
320 | /* beyond sysctl_nr_open; nothing to do */ | |
321 | if (unlikely(new_fdt->max_fds < open_files)) { | |
a892e2d7 | 322 | __free_fdtable(new_fdt); |
9dec3c4d | 323 | *errorp = -EMFILE; |
02afc626 | 324 | goto out_release; |
9dec3c4d | 325 | } |
9dec3c4d | 326 | |
02afc626 AV |
327 | /* |
328 | * Reacquire the oldf lock and a pointer to its fd table | |
329 | * who knows it may have a new bigger fd table. We need | |
330 | * the latest pointer. | |
331 | */ | |
332 | spin_lock(&oldf->file_lock); | |
333 | old_fdt = files_fdtable(oldf); | |
adbecb12 | 334 | open_files = count_open_files(old_fdt); |
02afc626 AV |
335 | } |
336 | ||
ea5c58e7 EB |
337 | copy_fd_bitmaps(new_fdt, old_fdt, open_files); |
338 | ||
02afc626 AV |
339 | old_fds = old_fdt->fd; |
340 | new_fds = new_fdt->fd; | |
341 | ||
02afc626 AV |
342 | for (i = open_files; i != 0; i--) { |
343 | struct file *f = *old_fds++; | |
344 | if (f) { | |
345 | get_file(f); | |
346 | } else { | |
347 | /* | |
348 | * The fd may be claimed in the fd bitmap but not yet | |
349 | * instantiated in the files array if a sibling thread | |
350 | * is partway through open(). So make sure that this | |
351 | * fd is available to the new process. | |
352 | */ | |
1dce27c5 | 353 | __clear_open_fd(open_files - i, new_fdt); |
02afc626 AV |
354 | } |
355 | rcu_assign_pointer(*new_fds++, f); | |
356 | } | |
357 | spin_unlock(&oldf->file_lock); | |
358 | ||
ea5c58e7 EB |
359 | /* clear the remainder */ |
360 | memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *)); | |
02afc626 | 361 | |
afbec7ff AV |
362 | rcu_assign_pointer(newf->fdt, new_fdt); |
363 | ||
02afc626 AV |
364 | return newf; |
365 | ||
366 | out_release: | |
367 | kmem_cache_free(files_cachep, newf); | |
368 | out: | |
369 | return NULL; | |
370 | } | |
371 | ||
ce08b62d | 372 | static struct fdtable *close_files(struct files_struct * files) |
7cf4dc3c | 373 | { |
7cf4dc3c AV |
374 | /* |
375 | * It is safe to dereference the fd table without RCU or | |
376 | * ->file_lock because this is the last reference to the | |
ce08b62d | 377 | * files structure. |
7cf4dc3c | 378 | */ |
ce08b62d | 379 | struct fdtable *fdt = rcu_dereference_raw(files->fdt); |
9b80a184 | 380 | unsigned int i, j = 0; |
ce08b62d | 381 | |
7cf4dc3c AV |
382 | for (;;) { |
383 | unsigned long set; | |
384 | i = j * BITS_PER_LONG; | |
385 | if (i >= fdt->max_fds) | |
386 | break; | |
387 | set = fdt->open_fds[j++]; | |
388 | while (set) { | |
389 | if (set & 1) { | |
390 | struct file * file = xchg(&fdt->fd[i], NULL); | |
391 | if (file) { | |
392 | filp_close(file, files); | |
bde6c3aa | 393 | cond_resched_rcu_qs(); |
7cf4dc3c AV |
394 | } |
395 | } | |
396 | i++; | |
397 | set >>= 1; | |
398 | } | |
399 | } | |
ce08b62d ON |
400 | |
401 | return fdt; | |
7cf4dc3c AV |
402 | } |
403 | ||
404 | struct files_struct *get_files_struct(struct task_struct *task) | |
405 | { | |
406 | struct files_struct *files; | |
407 | ||
408 | task_lock(task); | |
409 | files = task->files; | |
410 | if (files) | |
411 | atomic_inc(&files->count); | |
412 | task_unlock(task); | |
413 | ||
414 | return files; | |
415 | } | |
416 | ||
417 | void put_files_struct(struct files_struct *files) | |
418 | { | |
7cf4dc3c | 419 | if (atomic_dec_and_test(&files->count)) { |
ce08b62d ON |
420 | struct fdtable *fdt = close_files(files); |
421 | ||
b9e02af0 AV |
422 | /* free the arrays if they are not embedded */ |
423 | if (fdt != &files->fdtab) | |
424 | __free_fdtable(fdt); | |
425 | kmem_cache_free(files_cachep, files); | |
7cf4dc3c AV |
426 | } |
427 | } | |
428 | ||
429 | void reset_files_struct(struct files_struct *files) | |
430 | { | |
431 | struct task_struct *tsk = current; | |
432 | struct files_struct *old; | |
433 | ||
434 | old = tsk->files; | |
435 | task_lock(tsk); | |
436 | tsk->files = files; | |
437 | task_unlock(tsk); | |
438 | put_files_struct(old); | |
439 | } | |
440 | ||
441 | void exit_files(struct task_struct *tsk) | |
442 | { | |
443 | struct files_struct * files = tsk->files; | |
444 | ||
445 | if (files) { | |
446 | task_lock(tsk); | |
447 | tsk->files = NULL; | |
448 | task_unlock(tsk); | |
449 | put_files_struct(files); | |
450 | } | |
451 | } | |
452 | ||
f52111b1 AV |
453 | struct files_struct init_files = { |
454 | .count = ATOMIC_INIT(1), | |
455 | .fdt = &init_files.fdtab, | |
456 | .fdtab = { | |
457 | .max_fds = NR_OPEN_DEFAULT, | |
458 | .fd = &init_files.fd_array[0], | |
1fd36adc DH |
459 | .close_on_exec = init_files.close_on_exec_init, |
460 | .open_fds = init_files.open_fds_init, | |
f3f86e33 | 461 | .full_fds_bits = init_files.full_fds_bits_init, |
f52111b1 | 462 | }, |
eece09ec | 463 | .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock), |
f52111b1 | 464 | }; |
1027abe8 | 465 | |
9b80a184 | 466 | static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start) |
f3f86e33 | 467 | { |
9b80a184 AD |
468 | unsigned int maxfd = fdt->max_fds; |
469 | unsigned int maxbit = maxfd / BITS_PER_LONG; | |
470 | unsigned int bitbit = start / BITS_PER_LONG; | |
f3f86e33 LT |
471 | |
472 | bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG; | |
473 | if (bitbit > maxfd) | |
474 | return maxfd; | |
475 | if (bitbit > start) | |
476 | start = bitbit; | |
477 | return find_next_zero_bit(fdt->open_fds, maxfd, start); | |
478 | } | |
479 | ||
1027abe8 AV |
480 | /* |
481 | * allocate a file descriptor, mark it busy. | |
482 | */ | |
dcfadfa4 AV |
483 | int __alloc_fd(struct files_struct *files, |
484 | unsigned start, unsigned end, unsigned flags) | |
1027abe8 | 485 | { |
1027abe8 AV |
486 | unsigned int fd; |
487 | int error; | |
488 | struct fdtable *fdt; | |
489 | ||
490 | spin_lock(&files->file_lock); | |
491 | repeat: | |
492 | fdt = files_fdtable(files); | |
493 | fd = start; | |
494 | if (fd < files->next_fd) | |
495 | fd = files->next_fd; | |
496 | ||
497 | if (fd < fdt->max_fds) | |
f3f86e33 | 498 | fd = find_next_fd(fdt, fd); |
1027abe8 | 499 | |
f33ff992 AV |
500 | /* |
501 | * N.B. For clone tasks sharing a files structure, this test | |
502 | * will limit the total number of files that can be opened. | |
503 | */ | |
504 | error = -EMFILE; | |
505 | if (fd >= end) | |
506 | goto out; | |
507 | ||
1027abe8 AV |
508 | error = expand_files(files, fd); |
509 | if (error < 0) | |
510 | goto out; | |
511 | ||
512 | /* | |
513 | * If we needed to expand the fs array we | |
514 | * might have blocked - try again. | |
515 | */ | |
516 | if (error) | |
517 | goto repeat; | |
518 | ||
519 | if (start <= files->next_fd) | |
520 | files->next_fd = fd + 1; | |
521 | ||
1dce27c5 | 522 | __set_open_fd(fd, fdt); |
1027abe8 | 523 | if (flags & O_CLOEXEC) |
1dce27c5 | 524 | __set_close_on_exec(fd, fdt); |
1027abe8 | 525 | else |
1dce27c5 | 526 | __clear_close_on_exec(fd, fdt); |
1027abe8 AV |
527 | error = fd; |
528 | #if 1 | |
529 | /* Sanity check */ | |
add1f099 | 530 | if (rcu_access_pointer(fdt->fd[fd]) != NULL) { |
1027abe8 AV |
531 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
532 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
533 | } | |
534 | #endif | |
535 | ||
536 | out: | |
537 | spin_unlock(&files->file_lock); | |
538 | return error; | |
539 | } | |
540 | ||
ad47bd72 | 541 | static int alloc_fd(unsigned start, unsigned flags) |
dcfadfa4 AV |
542 | { |
543 | return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags); | |
544 | } | |
545 | ||
1a7bd226 | 546 | int get_unused_fd_flags(unsigned flags) |
1027abe8 | 547 | { |
dcfadfa4 | 548 | return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags); |
1027abe8 | 549 | } |
1a7bd226 | 550 | EXPORT_SYMBOL(get_unused_fd_flags); |
56007cae AV |
551 | |
552 | static void __put_unused_fd(struct files_struct *files, unsigned int fd) | |
553 | { | |
554 | struct fdtable *fdt = files_fdtable(files); | |
555 | __clear_open_fd(fd, fdt); | |
556 | if (fd < files->next_fd) | |
557 | files->next_fd = fd; | |
558 | } | |
559 | ||
560 | void put_unused_fd(unsigned int fd) | |
561 | { | |
562 | struct files_struct *files = current->files; | |
563 | spin_lock(&files->file_lock); | |
564 | __put_unused_fd(files, fd); | |
565 | spin_unlock(&files->file_lock); | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL(put_unused_fd); | |
569 | ||
570 | /* | |
571 | * Install a file pointer in the fd array. | |
572 | * | |
573 | * The VFS is full of places where we drop the files lock between | |
574 | * setting the open_fds bitmap and installing the file in the file | |
575 | * array. At any such point, we are vulnerable to a dup2() race | |
576 | * installing a file in the array before us. We need to detect this and | |
577 | * fput() the struct file we are about to overwrite in this case. | |
578 | * | |
579 | * It should never happen - if we allow dup2() do it, _really_ bad things | |
580 | * will follow. | |
f869e8a7 AV |
581 | * |
582 | * NOTE: __fd_install() variant is really, really low-level; don't | |
583 | * use it unless you are forced to by truly lousy API shoved down | |
584 | * your throat. 'files' *MUST* be either current->files or obtained | |
585 | * by get_files_struct(current) done by whoever had given it to you, | |
586 | * or really bad things will happen. Normally you want to use | |
587 | * fd_install() instead. | |
56007cae AV |
588 | */ |
589 | ||
f869e8a7 AV |
590 | void __fd_install(struct files_struct *files, unsigned int fd, |
591 | struct file *file) | |
56007cae | 592 | { |
56007cae | 593 | struct fdtable *fdt; |
8a81252b ED |
594 | |
595 | might_sleep(); | |
596 | rcu_read_lock_sched(); | |
597 | ||
598 | while (unlikely(files->resize_in_progress)) { | |
599 | rcu_read_unlock_sched(); | |
600 | wait_event(files->resize_wait, !files->resize_in_progress); | |
601 | rcu_read_lock_sched(); | |
602 | } | |
603 | /* coupled with smp_wmb() in expand_fdtable() */ | |
604 | smp_rmb(); | |
605 | fdt = rcu_dereference_sched(files->fdt); | |
56007cae AV |
606 | BUG_ON(fdt->fd[fd] != NULL); |
607 | rcu_assign_pointer(fdt->fd[fd], file); | |
8a81252b | 608 | rcu_read_unlock_sched(); |
56007cae AV |
609 | } |
610 | ||
f869e8a7 AV |
611 | void fd_install(unsigned int fd, struct file *file) |
612 | { | |
613 | __fd_install(current->files, fd, file); | |
614 | } | |
615 | ||
56007cae | 616 | EXPORT_SYMBOL(fd_install); |
0ee8cdfe | 617 | |
483ce1d4 AV |
618 | /* |
619 | * The same warnings as for __alloc_fd()/__fd_install() apply here... | |
620 | */ | |
621 | int __close_fd(struct files_struct *files, unsigned fd) | |
622 | { | |
623 | struct file *file; | |
624 | struct fdtable *fdt; | |
625 | ||
626 | spin_lock(&files->file_lock); | |
627 | fdt = files_fdtable(files); | |
628 | if (fd >= fdt->max_fds) | |
629 | goto out_unlock; | |
630 | file = fdt->fd[fd]; | |
631 | if (!file) | |
632 | goto out_unlock; | |
633 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
634 | __clear_close_on_exec(fd, fdt); | |
635 | __put_unused_fd(files, fd); | |
636 | spin_unlock(&files->file_lock); | |
637 | return filp_close(file, files); | |
638 | ||
639 | out_unlock: | |
640 | spin_unlock(&files->file_lock); | |
641 | return -EBADF; | |
642 | } | |
643 | ||
6a6d27de AV |
644 | void do_close_on_exec(struct files_struct *files) |
645 | { | |
646 | unsigned i; | |
647 | struct fdtable *fdt; | |
648 | ||
649 | /* exec unshares first */ | |
6a6d27de AV |
650 | spin_lock(&files->file_lock); |
651 | for (i = 0; ; i++) { | |
652 | unsigned long set; | |
653 | unsigned fd = i * BITS_PER_LONG; | |
654 | fdt = files_fdtable(files); | |
655 | if (fd >= fdt->max_fds) | |
656 | break; | |
657 | set = fdt->close_on_exec[i]; | |
658 | if (!set) | |
659 | continue; | |
660 | fdt->close_on_exec[i] = 0; | |
661 | for ( ; set ; fd++, set >>= 1) { | |
662 | struct file *file; | |
663 | if (!(set & 1)) | |
664 | continue; | |
665 | file = fdt->fd[fd]; | |
666 | if (!file) | |
667 | continue; | |
668 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
669 | __put_unused_fd(files, fd); | |
670 | spin_unlock(&files->file_lock); | |
671 | filp_close(file, files); | |
672 | cond_resched(); | |
673 | spin_lock(&files->file_lock); | |
674 | } | |
675 | ||
676 | } | |
677 | spin_unlock(&files->file_lock); | |
678 | } | |
679 | ||
1deb46e2 | 680 | static struct file *__fget(unsigned int fd, fmode_t mask) |
0ee8cdfe | 681 | { |
0ee8cdfe | 682 | struct files_struct *files = current->files; |
1deb46e2 | 683 | struct file *file; |
0ee8cdfe AV |
684 | |
685 | rcu_read_lock(); | |
5ba97d28 | 686 | loop: |
0ee8cdfe AV |
687 | file = fcheck_files(files, fd); |
688 | if (file) { | |
5ba97d28 ED |
689 | /* File object ref couldn't be taken. |
690 | * dup2() atomicity guarantee is the reason | |
691 | * we loop to catch the new file (or NULL pointer) | |
692 | */ | |
693 | if (file->f_mode & mask) | |
0ee8cdfe | 694 | file = NULL; |
5ba97d28 ED |
695 | else if (!get_file_rcu(file)) |
696 | goto loop; | |
0ee8cdfe AV |
697 | } |
698 | rcu_read_unlock(); | |
699 | ||
700 | return file; | |
701 | } | |
702 | ||
1deb46e2 ON |
703 | struct file *fget(unsigned int fd) |
704 | { | |
705 | return __fget(fd, FMODE_PATH); | |
706 | } | |
0ee8cdfe AV |
707 | EXPORT_SYMBOL(fget); |
708 | ||
709 | struct file *fget_raw(unsigned int fd) | |
710 | { | |
1deb46e2 | 711 | return __fget(fd, 0); |
0ee8cdfe | 712 | } |
0ee8cdfe AV |
713 | EXPORT_SYMBOL(fget_raw); |
714 | ||
715 | /* | |
716 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
717 | * | |
718 | * You can use this instead of fget if you satisfy all of the following | |
719 | * conditions: | |
720 | * 1) You must call fput_light before exiting the syscall and returning control | |
721 | * to userspace (i.e. you cannot remember the returned struct file * after | |
722 | * returning to userspace). | |
723 | * 2) You must not call filp_close on the returned struct file * in between | |
724 | * calls to fget_light and fput_light. | |
725 | * 3) You must not clone the current task in between the calls to fget_light | |
726 | * and fput_light. | |
727 | * | |
728 | * The fput_needed flag returned by fget_light should be passed to the | |
729 | * corresponding fput_light. | |
730 | */ | |
bd2a31d5 | 731 | static unsigned long __fget_light(unsigned int fd, fmode_t mask) |
0ee8cdfe | 732 | { |
0ee8cdfe | 733 | struct files_struct *files = current->files; |
ad461834 | 734 | struct file *file; |
0ee8cdfe | 735 | |
0ee8cdfe | 736 | if (atomic_read(&files->count) == 1) { |
a8d4b834 | 737 | file = __fcheck_files(files, fd); |
bd2a31d5 AV |
738 | if (!file || unlikely(file->f_mode & mask)) |
739 | return 0; | |
740 | return (unsigned long)file; | |
0ee8cdfe | 741 | } else { |
e6ff9a9f | 742 | file = __fget(fd, mask); |
bd2a31d5 AV |
743 | if (!file) |
744 | return 0; | |
745 | return FDPUT_FPUT | (unsigned long)file; | |
0ee8cdfe | 746 | } |
0ee8cdfe | 747 | } |
bd2a31d5 | 748 | unsigned long __fdget(unsigned int fd) |
ad461834 | 749 | { |
bd2a31d5 | 750 | return __fget_light(fd, FMODE_PATH); |
ad461834 | 751 | } |
bd2a31d5 | 752 | EXPORT_SYMBOL(__fdget); |
0ee8cdfe | 753 | |
bd2a31d5 | 754 | unsigned long __fdget_raw(unsigned int fd) |
0ee8cdfe | 755 | { |
bd2a31d5 | 756 | return __fget_light(fd, 0); |
0ee8cdfe | 757 | } |
fe17f22d | 758 | |
bd2a31d5 AV |
759 | unsigned long __fdget_pos(unsigned int fd) |
760 | { | |
99aea681 EB |
761 | unsigned long v = __fdget(fd); |
762 | struct file *file = (struct file *)(v & ~3); | |
bd2a31d5 | 763 | |
99aea681 | 764 | if (file && (file->f_mode & FMODE_ATOMIC_POS)) { |
bd2a31d5 AV |
765 | if (file_count(file) > 1) { |
766 | v |= FDPUT_POS_UNLOCK; | |
767 | mutex_lock(&file->f_pos_lock); | |
768 | } | |
769 | } | |
99aea681 | 770 | return v; |
bd2a31d5 AV |
771 | } |
772 | ||
63b6df14 AV |
773 | void __f_unlock_pos(struct file *f) |
774 | { | |
775 | mutex_unlock(&f->f_pos_lock); | |
776 | } | |
777 | ||
bd2a31d5 AV |
778 | /* |
779 | * We only lock f_pos if we have threads or if the file might be | |
780 | * shared with another process. In both cases we'll have an elevated | |
781 | * file count (done either by fdget() or by fork()). | |
782 | */ | |
783 | ||
fe17f22d AV |
784 | void set_close_on_exec(unsigned int fd, int flag) |
785 | { | |
786 | struct files_struct *files = current->files; | |
787 | struct fdtable *fdt; | |
788 | spin_lock(&files->file_lock); | |
789 | fdt = files_fdtable(files); | |
790 | if (flag) | |
791 | __set_close_on_exec(fd, fdt); | |
792 | else | |
793 | __clear_close_on_exec(fd, fdt); | |
794 | spin_unlock(&files->file_lock); | |
795 | } | |
796 | ||
797 | bool get_close_on_exec(unsigned int fd) | |
798 | { | |
799 | struct files_struct *files = current->files; | |
800 | struct fdtable *fdt; | |
801 | bool res; | |
802 | rcu_read_lock(); | |
803 | fdt = files_fdtable(files); | |
804 | res = close_on_exec(fd, fdt); | |
805 | rcu_read_unlock(); | |
806 | return res; | |
807 | } | |
808 | ||
8280d161 AV |
809 | static int do_dup2(struct files_struct *files, |
810 | struct file *file, unsigned fd, unsigned flags) | |
e983094d | 811 | __releases(&files->file_lock) |
fe17f22d | 812 | { |
8280d161 | 813 | struct file *tofree; |
fe17f22d AV |
814 | struct fdtable *fdt; |
815 | ||
fe17f22d AV |
816 | /* |
817 | * We need to detect attempts to do dup2() over allocated but still | |
818 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
819 | * extra work in their equivalent of fget() - they insert struct | |
820 | * file immediately after grabbing descriptor, mark it larval if | |
821 | * more work (e.g. actual opening) is needed and make sure that | |
822 | * fget() treats larval files as absent. Potentially interesting, | |
823 | * but while extra work in fget() is trivial, locking implications | |
824 | * and amount of surgery on open()-related paths in VFS are not. | |
825 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
826 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
827 | * scope of POSIX or SUS, since neither considers shared descriptor | |
828 | * tables and this condition does not arise without those. | |
829 | */ | |
fe17f22d | 830 | fdt = files_fdtable(files); |
8280d161 AV |
831 | tofree = fdt->fd[fd]; |
832 | if (!tofree && fd_is_open(fd, fdt)) | |
833 | goto Ebusy; | |
fe17f22d | 834 | get_file(file); |
8280d161 AV |
835 | rcu_assign_pointer(fdt->fd[fd], file); |
836 | __set_open_fd(fd, fdt); | |
fe17f22d | 837 | if (flags & O_CLOEXEC) |
8280d161 | 838 | __set_close_on_exec(fd, fdt); |
fe17f22d | 839 | else |
8280d161 | 840 | __clear_close_on_exec(fd, fdt); |
fe17f22d AV |
841 | spin_unlock(&files->file_lock); |
842 | ||
843 | if (tofree) | |
844 | filp_close(tofree, files); | |
845 | ||
8280d161 AV |
846 | return fd; |
847 | ||
848 | Ebusy: | |
849 | spin_unlock(&files->file_lock); | |
850 | return -EBUSY; | |
851 | } | |
852 | ||
853 | int replace_fd(unsigned fd, struct file *file, unsigned flags) | |
854 | { | |
855 | int err; | |
856 | struct files_struct *files = current->files; | |
857 | ||
858 | if (!file) | |
859 | return __close_fd(files, fd); | |
860 | ||
861 | if (fd >= rlimit(RLIMIT_NOFILE)) | |
08f05c49 | 862 | return -EBADF; |
8280d161 AV |
863 | |
864 | spin_lock(&files->file_lock); | |
865 | err = expand_files(files, fd); | |
866 | if (unlikely(err < 0)) | |
867 | goto out_unlock; | |
868 | return do_dup2(files, file, fd, flags); | |
869 | ||
870 | out_unlock: | |
871 | spin_unlock(&files->file_lock); | |
872 | return err; | |
873 | } | |
874 | ||
875 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) | |
876 | { | |
877 | int err = -EBADF; | |
878 | struct file *file; | |
879 | struct files_struct *files = current->files; | |
880 | ||
881 | if ((flags & ~O_CLOEXEC) != 0) | |
882 | return -EINVAL; | |
883 | ||
aed97647 RJ |
884 | if (unlikely(oldfd == newfd)) |
885 | return -EINVAL; | |
886 | ||
8280d161 | 887 | if (newfd >= rlimit(RLIMIT_NOFILE)) |
08f05c49 | 888 | return -EBADF; |
8280d161 AV |
889 | |
890 | spin_lock(&files->file_lock); | |
891 | err = expand_files(files, newfd); | |
892 | file = fcheck(oldfd); | |
893 | if (unlikely(!file)) | |
894 | goto Ebadf; | |
895 | if (unlikely(err < 0)) { | |
896 | if (err == -EMFILE) | |
897 | goto Ebadf; | |
898 | goto out_unlock; | |
899 | } | |
900 | return do_dup2(files, file, newfd, flags); | |
fe17f22d AV |
901 | |
902 | Ebadf: | |
903 | err = -EBADF; | |
904 | out_unlock: | |
905 | spin_unlock(&files->file_lock); | |
906 | return err; | |
907 | } | |
908 | ||
909 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) | |
910 | { | |
911 | if (unlikely(newfd == oldfd)) { /* corner case */ | |
912 | struct files_struct *files = current->files; | |
913 | int retval = oldfd; | |
914 | ||
915 | rcu_read_lock(); | |
916 | if (!fcheck_files(files, oldfd)) | |
917 | retval = -EBADF; | |
918 | rcu_read_unlock(); | |
919 | return retval; | |
920 | } | |
921 | return sys_dup3(oldfd, newfd, 0); | |
922 | } | |
923 | ||
924 | SYSCALL_DEFINE1(dup, unsigned int, fildes) | |
925 | { | |
926 | int ret = -EBADF; | |
927 | struct file *file = fget_raw(fildes); | |
928 | ||
929 | if (file) { | |
8d10a035 | 930 | ret = get_unused_fd_flags(0); |
fe17f22d AV |
931 | if (ret >= 0) |
932 | fd_install(ret, file); | |
933 | else | |
934 | fput(file); | |
935 | } | |
936 | return ret; | |
937 | } | |
938 | ||
939 | int f_dupfd(unsigned int from, struct file *file, unsigned flags) | |
940 | { | |
941 | int err; | |
942 | if (from >= rlimit(RLIMIT_NOFILE)) | |
943 | return -EINVAL; | |
944 | err = alloc_fd(from, flags); | |
945 | if (err >= 0) { | |
946 | get_file(file); | |
947 | fd_install(err, file); | |
948 | } | |
949 | return err; | |
950 | } | |
c3c073f8 AV |
951 | |
952 | int iterate_fd(struct files_struct *files, unsigned n, | |
953 | int (*f)(const void *, struct file *, unsigned), | |
954 | const void *p) | |
955 | { | |
956 | struct fdtable *fdt; | |
c3c073f8 AV |
957 | int res = 0; |
958 | if (!files) | |
959 | return 0; | |
960 | spin_lock(&files->file_lock); | |
a77cfcb4 AV |
961 | for (fdt = files_fdtable(files); n < fdt->max_fds; n++) { |
962 | struct file *file; | |
963 | file = rcu_dereference_check_fdtable(files, fdt->fd[n]); | |
964 | if (!file) | |
965 | continue; | |
966 | res = f(p, file, n); | |
967 | if (res) | |
968 | break; | |
c3c073f8 AV |
969 | } |
970 | spin_unlock(&files->file_lock); | |
971 | return res; | |
972 | } | |
973 | EXPORT_SYMBOL(iterate_fd); |