]>
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> |
d43c36dc | 15 | #include <linux/sched.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 | ||
9cfe015a | 26 | int sysctl_nr_open __read_mostly = 1024*1024; |
eceea0b3 | 27 | int sysctl_nr_open_min = BITS_PER_LONG; |
7f4b36f9 AV |
28 | /* our max() is unusable in constant expressions ;-/ */ |
29 | #define __const_max(x, y) ((x) < (y) ? (x) : (y)) | |
30 | int sysctl_nr_open_max = __const_max(INT_MAX, ~(size_t)0/sizeof(void *)) & | |
31 | -BITS_PER_LONG; | |
9cfe015a | 32 | |
1fd36adc | 33 | static void *alloc_fdmem(size_t size) |
1da177e4 | 34 | { |
6d4831c2 AM |
35 | /* |
36 | * Very large allocations can stress page reclaim, so fall back to | |
37 | * vmalloc() if the allocation size will be considered "large" by the VM. | |
38 | */ | |
39 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { | |
96c7a2ff | 40 | void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN|__GFP_NORETRY); |
6d4831c2 AM |
41 | if (data != NULL) |
42 | return data; | |
43 | } | |
a892e2d7 | 44 | return vmalloc(size); |
1da177e4 LT |
45 | } |
46 | ||
a892e2d7 | 47 | static void __free_fdtable(struct fdtable *fdt) |
1da177e4 | 48 | { |
f6c0a192 AV |
49 | kvfree(fdt->fd); |
50 | kvfree(fdt->open_fds); | |
a892e2d7 | 51 | kfree(fdt); |
ab2af1f5 | 52 | } |
1da177e4 | 53 | |
7cf4dc3c | 54 | static void free_fdtable_rcu(struct rcu_head *rcu) |
ab2af1f5 | 55 | { |
ac3e3c5b | 56 | __free_fdtable(container_of(rcu, struct fdtable, rcu)); |
ab2af1f5 DS |
57 | } |
58 | ||
f3f86e33 LT |
59 | #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr)) |
60 | #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long)) | |
61 | ||
ab2af1f5 DS |
62 | /* |
63 | * Expand the fdset in the files_struct. Called with the files spinlock | |
64 | * held for write. | |
65 | */ | |
5466b456 | 66 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
ab2af1f5 | 67 | { |
5466b456 | 68 | unsigned int cpy, set; |
ab2af1f5 | 69 | |
5466b456 | 70 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
5466b456 VL |
71 | |
72 | cpy = ofdt->max_fds * sizeof(struct file *); | |
73 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); | |
74 | memcpy(nfdt->fd, ofdt->fd, cpy); | |
75 | memset((char *)(nfdt->fd) + cpy, 0, set); | |
76 | ||
77 | cpy = ofdt->max_fds / BITS_PER_BYTE; | |
78 | set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; | |
79 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | |
80 | memset((char *)(nfdt->open_fds) + cpy, 0, set); | |
81 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | |
82 | memset((char *)(nfdt->close_on_exec) + cpy, 0, set); | |
f3f86e33 LT |
83 | |
84 | cpy = BITBIT_SIZE(ofdt->max_fds); | |
85 | set = BITBIT_SIZE(nfdt->max_fds) - cpy; | |
86 | memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy); | |
87 | memset(cpy+(char *)nfdt->full_fds_bits, 0, set); | |
1da177e4 LT |
88 | } |
89 | ||
5466b456 | 90 | static struct fdtable * alloc_fdtable(unsigned int nr) |
1da177e4 | 91 | { |
5466b456 | 92 | struct fdtable *fdt; |
1fd36adc | 93 | void *data; |
1da177e4 | 94 | |
ab2af1f5 | 95 | /* |
5466b456 VL |
96 | * Figure out how many fds we actually want to support in this fdtable. |
97 | * Allocation steps are keyed to the size of the fdarray, since it | |
98 | * grows far faster than any of the other dynamic data. We try to fit | |
99 | * the fdarray into comfortable page-tuned chunks: starting at 1024B | |
100 | * and growing in powers of two from there on. | |
ab2af1f5 | 101 | */ |
5466b456 VL |
102 | nr /= (1024 / sizeof(struct file *)); |
103 | nr = roundup_pow_of_two(nr + 1); | |
104 | nr *= (1024 / sizeof(struct file *)); | |
5c598b34 AV |
105 | /* |
106 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open | |
107 | * had been set lower between the check in expand_files() and here. Deal | |
108 | * with that in caller, it's cheaper that way. | |
109 | * | |
110 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise | |
111 | * bitmaps handling below becomes unpleasant, to put it mildly... | |
112 | */ | |
113 | if (unlikely(nr > sysctl_nr_open)) | |
114 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; | |
bbea9f69 | 115 | |
5466b456 VL |
116 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); |
117 | if (!fdt) | |
bbea9f69 | 118 | goto out; |
5466b456 VL |
119 | fdt->max_fds = nr; |
120 | data = alloc_fdmem(nr * sizeof(struct file *)); | |
121 | if (!data) | |
122 | goto out_fdt; | |
1fd36adc DH |
123 | fdt->fd = data; |
124 | ||
125 | data = alloc_fdmem(max_t(size_t, | |
f3f86e33 | 126 | 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES)); |
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 DS |
151 | */ |
152 | static int expand_fdtable(struct files_struct *files, int nr) | |
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 | */ |
ad47bd72 | 197 | static int expand_files(struct files_struct *files, 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 | |
b8318b01 AV |
232 | static inline void __set_close_on_exec(int fd, struct fdtable *fdt) |
233 | { | |
234 | __set_bit(fd, fdt->close_on_exec); | |
235 | } | |
236 | ||
237 | static inline void __clear_close_on_exec(int fd, struct fdtable *fdt) | |
238 | { | |
239 | __clear_bit(fd, fdt->close_on_exec); | |
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 | ||
02afc626 AV |
256 | static int count_open_files(struct fdtable *fdt) |
257 | { | |
258 | int size = fdt->max_fds; | |
259 | int i; | |
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 | ||
02afc626 AV |
270 | /* |
271 | * Allocate a new files structure and copy contents from the | |
272 | * passed in files structure. | |
273 | * errorp will be valid only when the returned files_struct is NULL. | |
274 | */ | |
275 | struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) | |
276 | { | |
277 | struct files_struct *newf; | |
278 | struct file **old_fds, **new_fds; | |
279 | int open_files, size, i; | |
280 | struct fdtable *old_fdt, *new_fdt; | |
281 | ||
282 | *errorp = -ENOMEM; | |
afbec7ff | 283 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
02afc626 AV |
284 | if (!newf) |
285 | goto out; | |
286 | ||
afbec7ff AV |
287 | atomic_set(&newf->count, 1); |
288 | ||
289 | spin_lock_init(&newf->file_lock); | |
8a81252b ED |
290 | newf->resize_in_progress = false; |
291 | init_waitqueue_head(&newf->resize_wait); | |
afbec7ff AV |
292 | newf->next_fd = 0; |
293 | new_fdt = &newf->fdtab; | |
294 | new_fdt->max_fds = NR_OPEN_DEFAULT; | |
1fd36adc DH |
295 | new_fdt->close_on_exec = newf->close_on_exec_init; |
296 | new_fdt->open_fds = newf->open_fds_init; | |
f3f86e33 | 297 | new_fdt->full_fds_bits = newf->full_fds_bits_init; |
afbec7ff | 298 | new_fdt->fd = &newf->fd_array[0]; |
afbec7ff | 299 | |
02afc626 AV |
300 | spin_lock(&oldf->file_lock); |
301 | old_fdt = files_fdtable(oldf); | |
02afc626 AV |
302 | open_files = count_open_files(old_fdt); |
303 | ||
304 | /* | |
305 | * Check whether we need to allocate a larger fd array and fd set. | |
02afc626 | 306 | */ |
adbecb12 | 307 | while (unlikely(open_files > new_fdt->max_fds)) { |
02afc626 | 308 | spin_unlock(&oldf->file_lock); |
9dec3c4d | 309 | |
a892e2d7 CG |
310 | if (new_fdt != &newf->fdtab) |
311 | __free_fdtable(new_fdt); | |
adbecb12 | 312 | |
9dec3c4d AV |
313 | new_fdt = alloc_fdtable(open_files - 1); |
314 | if (!new_fdt) { | |
315 | *errorp = -ENOMEM; | |
316 | goto out_release; | |
317 | } | |
318 | ||
319 | /* beyond sysctl_nr_open; nothing to do */ | |
320 | if (unlikely(new_fdt->max_fds < open_files)) { | |
a892e2d7 | 321 | __free_fdtable(new_fdt); |
9dec3c4d | 322 | *errorp = -EMFILE; |
02afc626 | 323 | goto out_release; |
9dec3c4d | 324 | } |
9dec3c4d | 325 | |
02afc626 AV |
326 | /* |
327 | * Reacquire the oldf lock and a pointer to its fd table | |
328 | * who knows it may have a new bigger fd table. We need | |
329 | * the latest pointer. | |
330 | */ | |
331 | spin_lock(&oldf->file_lock); | |
332 | old_fdt = files_fdtable(oldf); | |
adbecb12 | 333 | open_files = count_open_files(old_fdt); |
02afc626 AV |
334 | } |
335 | ||
336 | old_fds = old_fdt->fd; | |
337 | new_fds = new_fdt->fd; | |
338 | ||
1fd36adc DH |
339 | memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8); |
340 | memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8); | |
f3f86e33 | 341 | memcpy(new_fdt->full_fds_bits, old_fdt->full_fds_bits, BITBIT_SIZE(open_files)); |
02afc626 AV |
342 | |
343 | for (i = open_files; i != 0; i--) { | |
344 | struct file *f = *old_fds++; | |
345 | if (f) { | |
346 | get_file(f); | |
347 | } else { | |
348 | /* | |
349 | * The fd may be claimed in the fd bitmap but not yet | |
350 | * instantiated in the files array if a sibling thread | |
351 | * is partway through open(). So make sure that this | |
352 | * fd is available to the new process. | |
353 | */ | |
1dce27c5 | 354 | __clear_open_fd(open_files - i, new_fdt); |
02afc626 AV |
355 | } |
356 | rcu_assign_pointer(*new_fds++, f); | |
357 | } | |
358 | spin_unlock(&oldf->file_lock); | |
359 | ||
360 | /* compute the remainder to be cleared */ | |
361 | size = (new_fdt->max_fds - open_files) * sizeof(struct file *); | |
362 | ||
363 | /* This is long word aligned thus could use a optimized version */ | |
364 | memset(new_fds, 0, size); | |
365 | ||
366 | if (new_fdt->max_fds > open_files) { | |
1fd36adc DH |
367 | int left = (new_fdt->max_fds - open_files) / 8; |
368 | int start = open_files / BITS_PER_LONG; | |
02afc626 | 369 | |
1fd36adc DH |
370 | memset(&new_fdt->open_fds[start], 0, left); |
371 | memset(&new_fdt->close_on_exec[start], 0, left); | |
02afc626 AV |
372 | } |
373 | ||
afbec7ff AV |
374 | rcu_assign_pointer(newf->fdt, new_fdt); |
375 | ||
02afc626 AV |
376 | return newf; |
377 | ||
378 | out_release: | |
379 | kmem_cache_free(files_cachep, newf); | |
380 | out: | |
381 | return NULL; | |
382 | } | |
383 | ||
ce08b62d | 384 | static struct fdtable *close_files(struct files_struct * files) |
7cf4dc3c | 385 | { |
7cf4dc3c AV |
386 | /* |
387 | * It is safe to dereference the fd table without RCU or | |
388 | * ->file_lock because this is the last reference to the | |
ce08b62d | 389 | * files structure. |
7cf4dc3c | 390 | */ |
ce08b62d ON |
391 | struct fdtable *fdt = rcu_dereference_raw(files->fdt); |
392 | int i, j = 0; | |
393 | ||
7cf4dc3c AV |
394 | for (;;) { |
395 | unsigned long set; | |
396 | i = j * BITS_PER_LONG; | |
397 | if (i >= fdt->max_fds) | |
398 | break; | |
399 | set = fdt->open_fds[j++]; | |
400 | while (set) { | |
401 | if (set & 1) { | |
402 | struct file * file = xchg(&fdt->fd[i], NULL); | |
403 | if (file) { | |
404 | filp_close(file, files); | |
bde6c3aa | 405 | cond_resched_rcu_qs(); |
7cf4dc3c AV |
406 | } |
407 | } | |
408 | i++; | |
409 | set >>= 1; | |
410 | } | |
411 | } | |
ce08b62d ON |
412 | |
413 | return fdt; | |
7cf4dc3c AV |
414 | } |
415 | ||
416 | struct files_struct *get_files_struct(struct task_struct *task) | |
417 | { | |
418 | struct files_struct *files; | |
419 | ||
420 | task_lock(task); | |
421 | files = task->files; | |
422 | if (files) | |
423 | atomic_inc(&files->count); | |
424 | task_unlock(task); | |
425 | ||
426 | return files; | |
427 | } | |
428 | ||
429 | void put_files_struct(struct files_struct *files) | |
430 | { | |
7cf4dc3c | 431 | if (atomic_dec_and_test(&files->count)) { |
ce08b62d ON |
432 | struct fdtable *fdt = close_files(files); |
433 | ||
b9e02af0 AV |
434 | /* free the arrays if they are not embedded */ |
435 | if (fdt != &files->fdtab) | |
436 | __free_fdtable(fdt); | |
437 | kmem_cache_free(files_cachep, files); | |
7cf4dc3c AV |
438 | } |
439 | } | |
440 | ||
441 | void reset_files_struct(struct files_struct *files) | |
442 | { | |
443 | struct task_struct *tsk = current; | |
444 | struct files_struct *old; | |
445 | ||
446 | old = tsk->files; | |
447 | task_lock(tsk); | |
448 | tsk->files = files; | |
449 | task_unlock(tsk); | |
450 | put_files_struct(old); | |
451 | } | |
452 | ||
453 | void exit_files(struct task_struct *tsk) | |
454 | { | |
455 | struct files_struct * files = tsk->files; | |
456 | ||
457 | if (files) { | |
458 | task_lock(tsk); | |
459 | tsk->files = NULL; | |
460 | task_unlock(tsk); | |
461 | put_files_struct(files); | |
462 | } | |
463 | } | |
464 | ||
f52111b1 AV |
465 | struct files_struct init_files = { |
466 | .count = ATOMIC_INIT(1), | |
467 | .fdt = &init_files.fdtab, | |
468 | .fdtab = { | |
469 | .max_fds = NR_OPEN_DEFAULT, | |
470 | .fd = &init_files.fd_array[0], | |
1fd36adc DH |
471 | .close_on_exec = init_files.close_on_exec_init, |
472 | .open_fds = init_files.open_fds_init, | |
f3f86e33 | 473 | .full_fds_bits = init_files.full_fds_bits_init, |
f52111b1 | 474 | }, |
eece09ec | 475 | .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock), |
f52111b1 | 476 | }; |
1027abe8 | 477 | |
f3f86e33 LT |
478 | static unsigned long find_next_fd(struct fdtable *fdt, unsigned long start) |
479 | { | |
480 | unsigned long maxfd = fdt->max_fds; | |
481 | unsigned long maxbit = maxfd / BITS_PER_LONG; | |
482 | unsigned long bitbit = start / BITS_PER_LONG; | |
483 | ||
484 | bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG; | |
485 | if (bitbit > maxfd) | |
486 | return maxfd; | |
487 | if (bitbit > start) | |
488 | start = bitbit; | |
489 | return find_next_zero_bit(fdt->open_fds, maxfd, start); | |
490 | } | |
491 | ||
1027abe8 AV |
492 | /* |
493 | * allocate a file descriptor, mark it busy. | |
494 | */ | |
dcfadfa4 AV |
495 | int __alloc_fd(struct files_struct *files, |
496 | unsigned start, unsigned end, unsigned flags) | |
1027abe8 | 497 | { |
1027abe8 AV |
498 | unsigned int fd; |
499 | int error; | |
500 | struct fdtable *fdt; | |
501 | ||
502 | spin_lock(&files->file_lock); | |
503 | repeat: | |
504 | fdt = files_fdtable(files); | |
505 | fd = start; | |
506 | if (fd < files->next_fd) | |
507 | fd = files->next_fd; | |
508 | ||
509 | if (fd < fdt->max_fds) | |
f3f86e33 | 510 | fd = find_next_fd(fdt, fd); |
1027abe8 | 511 | |
f33ff992 AV |
512 | /* |
513 | * N.B. For clone tasks sharing a files structure, this test | |
514 | * will limit the total number of files that can be opened. | |
515 | */ | |
516 | error = -EMFILE; | |
517 | if (fd >= end) | |
518 | goto out; | |
519 | ||
1027abe8 AV |
520 | error = expand_files(files, fd); |
521 | if (error < 0) | |
522 | goto out; | |
523 | ||
524 | /* | |
525 | * If we needed to expand the fs array we | |
526 | * might have blocked - try again. | |
527 | */ | |
528 | if (error) | |
529 | goto repeat; | |
530 | ||
531 | if (start <= files->next_fd) | |
532 | files->next_fd = fd + 1; | |
533 | ||
1dce27c5 | 534 | __set_open_fd(fd, fdt); |
1027abe8 | 535 | if (flags & O_CLOEXEC) |
1dce27c5 | 536 | __set_close_on_exec(fd, fdt); |
1027abe8 | 537 | else |
1dce27c5 | 538 | __clear_close_on_exec(fd, fdt); |
1027abe8 AV |
539 | error = fd; |
540 | #if 1 | |
541 | /* Sanity check */ | |
add1f099 | 542 | if (rcu_access_pointer(fdt->fd[fd]) != NULL) { |
1027abe8 AV |
543 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
544 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
545 | } | |
546 | #endif | |
547 | ||
548 | out: | |
549 | spin_unlock(&files->file_lock); | |
550 | return error; | |
551 | } | |
552 | ||
ad47bd72 | 553 | static int alloc_fd(unsigned start, unsigned flags) |
dcfadfa4 AV |
554 | { |
555 | return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags); | |
556 | } | |
557 | ||
1a7bd226 | 558 | int get_unused_fd_flags(unsigned flags) |
1027abe8 | 559 | { |
dcfadfa4 | 560 | return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags); |
1027abe8 | 561 | } |
1a7bd226 | 562 | EXPORT_SYMBOL(get_unused_fd_flags); |
56007cae AV |
563 | |
564 | static void __put_unused_fd(struct files_struct *files, unsigned int fd) | |
565 | { | |
566 | struct fdtable *fdt = files_fdtable(files); | |
567 | __clear_open_fd(fd, fdt); | |
568 | if (fd < files->next_fd) | |
569 | files->next_fd = fd; | |
570 | } | |
571 | ||
572 | void put_unused_fd(unsigned int fd) | |
573 | { | |
574 | struct files_struct *files = current->files; | |
575 | spin_lock(&files->file_lock); | |
576 | __put_unused_fd(files, fd); | |
577 | spin_unlock(&files->file_lock); | |
578 | } | |
579 | ||
580 | EXPORT_SYMBOL(put_unused_fd); | |
581 | ||
582 | /* | |
583 | * Install a file pointer in the fd array. | |
584 | * | |
585 | * The VFS is full of places where we drop the files lock between | |
586 | * setting the open_fds bitmap and installing the file in the file | |
587 | * array. At any such point, we are vulnerable to a dup2() race | |
588 | * installing a file in the array before us. We need to detect this and | |
589 | * fput() the struct file we are about to overwrite in this case. | |
590 | * | |
591 | * It should never happen - if we allow dup2() do it, _really_ bad things | |
592 | * will follow. | |
f869e8a7 AV |
593 | * |
594 | * NOTE: __fd_install() variant is really, really low-level; don't | |
595 | * use it unless you are forced to by truly lousy API shoved down | |
596 | * your throat. 'files' *MUST* be either current->files or obtained | |
597 | * by get_files_struct(current) done by whoever had given it to you, | |
598 | * or really bad things will happen. Normally you want to use | |
599 | * fd_install() instead. | |
56007cae AV |
600 | */ |
601 | ||
f869e8a7 AV |
602 | void __fd_install(struct files_struct *files, unsigned int fd, |
603 | struct file *file) | |
56007cae | 604 | { |
56007cae | 605 | struct fdtable *fdt; |
8a81252b ED |
606 | |
607 | might_sleep(); | |
608 | rcu_read_lock_sched(); | |
609 | ||
610 | while (unlikely(files->resize_in_progress)) { | |
611 | rcu_read_unlock_sched(); | |
612 | wait_event(files->resize_wait, !files->resize_in_progress); | |
613 | rcu_read_lock_sched(); | |
614 | } | |
615 | /* coupled with smp_wmb() in expand_fdtable() */ | |
616 | smp_rmb(); | |
617 | fdt = rcu_dereference_sched(files->fdt); | |
56007cae AV |
618 | BUG_ON(fdt->fd[fd] != NULL); |
619 | rcu_assign_pointer(fdt->fd[fd], file); | |
8a81252b | 620 | rcu_read_unlock_sched(); |
56007cae AV |
621 | } |
622 | ||
f869e8a7 AV |
623 | void fd_install(unsigned int fd, struct file *file) |
624 | { | |
625 | __fd_install(current->files, fd, file); | |
626 | } | |
627 | ||
56007cae | 628 | EXPORT_SYMBOL(fd_install); |
0ee8cdfe | 629 | |
483ce1d4 AV |
630 | /* |
631 | * The same warnings as for __alloc_fd()/__fd_install() apply here... | |
632 | */ | |
633 | int __close_fd(struct files_struct *files, unsigned fd) | |
634 | { | |
635 | struct file *file; | |
636 | struct fdtable *fdt; | |
637 | ||
638 | spin_lock(&files->file_lock); | |
639 | fdt = files_fdtable(files); | |
640 | if (fd >= fdt->max_fds) | |
641 | goto out_unlock; | |
642 | file = fdt->fd[fd]; | |
643 | if (!file) | |
644 | goto out_unlock; | |
645 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
646 | __clear_close_on_exec(fd, fdt); | |
647 | __put_unused_fd(files, fd); | |
648 | spin_unlock(&files->file_lock); | |
649 | return filp_close(file, files); | |
650 | ||
651 | out_unlock: | |
652 | spin_unlock(&files->file_lock); | |
653 | return -EBADF; | |
654 | } | |
655 | ||
6a6d27de AV |
656 | void do_close_on_exec(struct files_struct *files) |
657 | { | |
658 | unsigned i; | |
659 | struct fdtable *fdt; | |
660 | ||
661 | /* exec unshares first */ | |
6a6d27de AV |
662 | spin_lock(&files->file_lock); |
663 | for (i = 0; ; i++) { | |
664 | unsigned long set; | |
665 | unsigned fd = i * BITS_PER_LONG; | |
666 | fdt = files_fdtable(files); | |
667 | if (fd >= fdt->max_fds) | |
668 | break; | |
669 | set = fdt->close_on_exec[i]; | |
670 | if (!set) | |
671 | continue; | |
672 | fdt->close_on_exec[i] = 0; | |
673 | for ( ; set ; fd++, set >>= 1) { | |
674 | struct file *file; | |
675 | if (!(set & 1)) | |
676 | continue; | |
677 | file = fdt->fd[fd]; | |
678 | if (!file) | |
679 | continue; | |
680 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
681 | __put_unused_fd(files, fd); | |
682 | spin_unlock(&files->file_lock); | |
683 | filp_close(file, files); | |
684 | cond_resched(); | |
685 | spin_lock(&files->file_lock); | |
686 | } | |
687 | ||
688 | } | |
689 | spin_unlock(&files->file_lock); | |
690 | } | |
691 | ||
1deb46e2 | 692 | static struct file *__fget(unsigned int fd, fmode_t mask) |
0ee8cdfe | 693 | { |
0ee8cdfe | 694 | struct files_struct *files = current->files; |
1deb46e2 | 695 | struct file *file; |
0ee8cdfe AV |
696 | |
697 | rcu_read_lock(); | |
5ba97d28 | 698 | loop: |
0ee8cdfe AV |
699 | file = fcheck_files(files, fd); |
700 | if (file) { | |
5ba97d28 ED |
701 | /* File object ref couldn't be taken. |
702 | * dup2() atomicity guarantee is the reason | |
703 | * we loop to catch the new file (or NULL pointer) | |
704 | */ | |
705 | if (file->f_mode & mask) | |
0ee8cdfe | 706 | file = NULL; |
5ba97d28 ED |
707 | else if (!get_file_rcu(file)) |
708 | goto loop; | |
0ee8cdfe AV |
709 | } |
710 | rcu_read_unlock(); | |
711 | ||
712 | return file; | |
713 | } | |
714 | ||
1deb46e2 ON |
715 | struct file *fget(unsigned int fd) |
716 | { | |
717 | return __fget(fd, FMODE_PATH); | |
718 | } | |
0ee8cdfe AV |
719 | EXPORT_SYMBOL(fget); |
720 | ||
721 | struct file *fget_raw(unsigned int fd) | |
722 | { | |
1deb46e2 | 723 | return __fget(fd, 0); |
0ee8cdfe | 724 | } |
0ee8cdfe AV |
725 | EXPORT_SYMBOL(fget_raw); |
726 | ||
727 | /* | |
728 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
729 | * | |
730 | * You can use this instead of fget if you satisfy all of the following | |
731 | * conditions: | |
732 | * 1) You must call fput_light before exiting the syscall and returning control | |
733 | * to userspace (i.e. you cannot remember the returned struct file * after | |
734 | * returning to userspace). | |
735 | * 2) You must not call filp_close on the returned struct file * in between | |
736 | * calls to fget_light and fput_light. | |
737 | * 3) You must not clone the current task in between the calls to fget_light | |
738 | * and fput_light. | |
739 | * | |
740 | * The fput_needed flag returned by fget_light should be passed to the | |
741 | * corresponding fput_light. | |
742 | */ | |
bd2a31d5 | 743 | static unsigned long __fget_light(unsigned int fd, fmode_t mask) |
0ee8cdfe | 744 | { |
0ee8cdfe | 745 | struct files_struct *files = current->files; |
ad461834 | 746 | struct file *file; |
0ee8cdfe | 747 | |
0ee8cdfe | 748 | if (atomic_read(&files->count) == 1) { |
a8d4b834 | 749 | file = __fcheck_files(files, fd); |
bd2a31d5 AV |
750 | if (!file || unlikely(file->f_mode & mask)) |
751 | return 0; | |
752 | return (unsigned long)file; | |
0ee8cdfe | 753 | } else { |
e6ff9a9f | 754 | file = __fget(fd, mask); |
bd2a31d5 AV |
755 | if (!file) |
756 | return 0; | |
757 | return FDPUT_FPUT | (unsigned long)file; | |
0ee8cdfe | 758 | } |
0ee8cdfe | 759 | } |
bd2a31d5 | 760 | unsigned long __fdget(unsigned int fd) |
ad461834 | 761 | { |
bd2a31d5 | 762 | return __fget_light(fd, FMODE_PATH); |
ad461834 | 763 | } |
bd2a31d5 | 764 | EXPORT_SYMBOL(__fdget); |
0ee8cdfe | 765 | |
bd2a31d5 | 766 | unsigned long __fdget_raw(unsigned int fd) |
0ee8cdfe | 767 | { |
bd2a31d5 | 768 | return __fget_light(fd, 0); |
0ee8cdfe | 769 | } |
fe17f22d | 770 | |
bd2a31d5 AV |
771 | unsigned long __fdget_pos(unsigned int fd) |
772 | { | |
99aea681 EB |
773 | unsigned long v = __fdget(fd); |
774 | struct file *file = (struct file *)(v & ~3); | |
bd2a31d5 | 775 | |
99aea681 | 776 | if (file && (file->f_mode & FMODE_ATOMIC_POS)) { |
bd2a31d5 AV |
777 | if (file_count(file) > 1) { |
778 | v |= FDPUT_POS_UNLOCK; | |
779 | mutex_lock(&file->f_pos_lock); | |
780 | } | |
781 | } | |
99aea681 | 782 | return v; |
bd2a31d5 AV |
783 | } |
784 | ||
785 | /* | |
786 | * We only lock f_pos if we have threads or if the file might be | |
787 | * shared with another process. In both cases we'll have an elevated | |
788 | * file count (done either by fdget() or by fork()). | |
789 | */ | |
790 | ||
fe17f22d AV |
791 | void set_close_on_exec(unsigned int fd, int flag) |
792 | { | |
793 | struct files_struct *files = current->files; | |
794 | struct fdtable *fdt; | |
795 | spin_lock(&files->file_lock); | |
796 | fdt = files_fdtable(files); | |
797 | if (flag) | |
798 | __set_close_on_exec(fd, fdt); | |
799 | else | |
800 | __clear_close_on_exec(fd, fdt); | |
801 | spin_unlock(&files->file_lock); | |
802 | } | |
803 | ||
804 | bool get_close_on_exec(unsigned int fd) | |
805 | { | |
806 | struct files_struct *files = current->files; | |
807 | struct fdtable *fdt; | |
808 | bool res; | |
809 | rcu_read_lock(); | |
810 | fdt = files_fdtable(files); | |
811 | res = close_on_exec(fd, fdt); | |
812 | rcu_read_unlock(); | |
813 | return res; | |
814 | } | |
815 | ||
8280d161 AV |
816 | static int do_dup2(struct files_struct *files, |
817 | struct file *file, unsigned fd, unsigned flags) | |
e983094d | 818 | __releases(&files->file_lock) |
fe17f22d | 819 | { |
8280d161 | 820 | struct file *tofree; |
fe17f22d AV |
821 | struct fdtable *fdt; |
822 | ||
fe17f22d AV |
823 | /* |
824 | * We need to detect attempts to do dup2() over allocated but still | |
825 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
826 | * extra work in their equivalent of fget() - they insert struct | |
827 | * file immediately after grabbing descriptor, mark it larval if | |
828 | * more work (e.g. actual opening) is needed and make sure that | |
829 | * fget() treats larval files as absent. Potentially interesting, | |
830 | * but while extra work in fget() is trivial, locking implications | |
831 | * and amount of surgery on open()-related paths in VFS are not. | |
832 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
833 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
834 | * scope of POSIX or SUS, since neither considers shared descriptor | |
835 | * tables and this condition does not arise without those. | |
836 | */ | |
fe17f22d | 837 | fdt = files_fdtable(files); |
8280d161 AV |
838 | tofree = fdt->fd[fd]; |
839 | if (!tofree && fd_is_open(fd, fdt)) | |
840 | goto Ebusy; | |
fe17f22d | 841 | get_file(file); |
8280d161 AV |
842 | rcu_assign_pointer(fdt->fd[fd], file); |
843 | __set_open_fd(fd, fdt); | |
fe17f22d | 844 | if (flags & O_CLOEXEC) |
8280d161 | 845 | __set_close_on_exec(fd, fdt); |
fe17f22d | 846 | else |
8280d161 | 847 | __clear_close_on_exec(fd, fdt); |
fe17f22d AV |
848 | spin_unlock(&files->file_lock); |
849 | ||
850 | if (tofree) | |
851 | filp_close(tofree, files); | |
852 | ||
8280d161 AV |
853 | return fd; |
854 | ||
855 | Ebusy: | |
856 | spin_unlock(&files->file_lock); | |
857 | return -EBUSY; | |
858 | } | |
859 | ||
860 | int replace_fd(unsigned fd, struct file *file, unsigned flags) | |
861 | { | |
862 | int err; | |
863 | struct files_struct *files = current->files; | |
864 | ||
865 | if (!file) | |
866 | return __close_fd(files, fd); | |
867 | ||
868 | if (fd >= rlimit(RLIMIT_NOFILE)) | |
08f05c49 | 869 | return -EBADF; |
8280d161 AV |
870 | |
871 | spin_lock(&files->file_lock); | |
872 | err = expand_files(files, fd); | |
873 | if (unlikely(err < 0)) | |
874 | goto out_unlock; | |
875 | return do_dup2(files, file, fd, flags); | |
876 | ||
877 | out_unlock: | |
878 | spin_unlock(&files->file_lock); | |
879 | return err; | |
880 | } | |
881 | ||
882 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) | |
883 | { | |
884 | int err = -EBADF; | |
885 | struct file *file; | |
886 | struct files_struct *files = current->files; | |
887 | ||
888 | if ((flags & ~O_CLOEXEC) != 0) | |
889 | return -EINVAL; | |
890 | ||
aed97647 RJ |
891 | if (unlikely(oldfd == newfd)) |
892 | return -EINVAL; | |
893 | ||
8280d161 | 894 | if (newfd >= rlimit(RLIMIT_NOFILE)) |
08f05c49 | 895 | return -EBADF; |
8280d161 AV |
896 | |
897 | spin_lock(&files->file_lock); | |
898 | err = expand_files(files, newfd); | |
899 | file = fcheck(oldfd); | |
900 | if (unlikely(!file)) | |
901 | goto Ebadf; | |
902 | if (unlikely(err < 0)) { | |
903 | if (err == -EMFILE) | |
904 | goto Ebadf; | |
905 | goto out_unlock; | |
906 | } | |
907 | return do_dup2(files, file, newfd, flags); | |
fe17f22d AV |
908 | |
909 | Ebadf: | |
910 | err = -EBADF; | |
911 | out_unlock: | |
912 | spin_unlock(&files->file_lock); | |
913 | return err; | |
914 | } | |
915 | ||
916 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) | |
917 | { | |
918 | if (unlikely(newfd == oldfd)) { /* corner case */ | |
919 | struct files_struct *files = current->files; | |
920 | int retval = oldfd; | |
921 | ||
922 | rcu_read_lock(); | |
923 | if (!fcheck_files(files, oldfd)) | |
924 | retval = -EBADF; | |
925 | rcu_read_unlock(); | |
926 | return retval; | |
927 | } | |
928 | return sys_dup3(oldfd, newfd, 0); | |
929 | } | |
930 | ||
931 | SYSCALL_DEFINE1(dup, unsigned int, fildes) | |
932 | { | |
933 | int ret = -EBADF; | |
934 | struct file *file = fget_raw(fildes); | |
935 | ||
936 | if (file) { | |
8d10a035 | 937 | ret = get_unused_fd_flags(0); |
fe17f22d AV |
938 | if (ret >= 0) |
939 | fd_install(ret, file); | |
940 | else | |
941 | fput(file); | |
942 | } | |
943 | return ret; | |
944 | } | |
945 | ||
946 | int f_dupfd(unsigned int from, struct file *file, unsigned flags) | |
947 | { | |
948 | int err; | |
949 | if (from >= rlimit(RLIMIT_NOFILE)) | |
950 | return -EINVAL; | |
951 | err = alloc_fd(from, flags); | |
952 | if (err >= 0) { | |
953 | get_file(file); | |
954 | fd_install(err, file); | |
955 | } | |
956 | return err; | |
957 | } | |
c3c073f8 AV |
958 | |
959 | int iterate_fd(struct files_struct *files, unsigned n, | |
960 | int (*f)(const void *, struct file *, unsigned), | |
961 | const void *p) | |
962 | { | |
963 | struct fdtable *fdt; | |
c3c073f8 AV |
964 | int res = 0; |
965 | if (!files) | |
966 | return 0; | |
967 | spin_lock(&files->file_lock); | |
a77cfcb4 AV |
968 | for (fdt = files_fdtable(files); n < fdt->max_fds; n++) { |
969 | struct file *file; | |
970 | file = rcu_dereference_check_fdtable(files, fdt->fd[n]); | |
971 | if (!file) | |
972 | continue; | |
973 | res = f(p, file, n); | |
974 | if (res) | |
975 | break; | |
c3c073f8 AV |
976 | } |
977 | spin_unlock(&files->file_lock); | |
978 | return res; | |
979 | } | |
980 | EXPORT_SYMBOL(iterate_fd); |