]>
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 | ||
26 | struct fdtable_defer { | |
27 | spinlock_t lock; | |
28 | struct work_struct wq; | |
ab2af1f5 DS |
29 | struct fdtable *next; |
30 | }; | |
31 | ||
9cfe015a | 32 | int sysctl_nr_open __read_mostly = 1024*1024; |
eceea0b3 AV |
33 | int sysctl_nr_open_min = BITS_PER_LONG; |
34 | int sysctl_nr_open_max = 1024 * 1024; /* raised later */ | |
9cfe015a | 35 | |
ab2af1f5 DS |
36 | /* |
37 | * We use this list to defer free fdtables that have vmalloced | |
38 | * sets/arrays. By keeping a per-cpu list, we avoid having to embed | |
39 | * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in | |
40 | * this per-task structure. | |
41 | */ | |
42 | static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list); | |
1da177e4 | 43 | |
1fd36adc | 44 | static void *alloc_fdmem(size_t size) |
1da177e4 | 45 | { |
6d4831c2 AM |
46 | /* |
47 | * Very large allocations can stress page reclaim, so fall back to | |
48 | * vmalloc() if the allocation size will be considered "large" by the VM. | |
49 | */ | |
50 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { | |
51 | void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN); | |
52 | if (data != NULL) | |
53 | return data; | |
54 | } | |
a892e2d7 | 55 | return vmalloc(size); |
1da177e4 LT |
56 | } |
57 | ||
a892e2d7 | 58 | static void free_fdmem(void *ptr) |
1da177e4 | 59 | { |
a892e2d7 | 60 | is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr); |
1da177e4 LT |
61 | } |
62 | ||
a892e2d7 | 63 | static void __free_fdtable(struct fdtable *fdt) |
1da177e4 | 64 | { |
a892e2d7 CG |
65 | free_fdmem(fdt->fd); |
66 | free_fdmem(fdt->open_fds); | |
67 | kfree(fdt); | |
ab2af1f5 | 68 | } |
1da177e4 | 69 | |
65f27f38 | 70 | static void free_fdtable_work(struct work_struct *work) |
ab2af1f5 | 71 | { |
65f27f38 DH |
72 | struct fdtable_defer *f = |
73 | container_of(work, struct fdtable_defer, wq); | |
ab2af1f5 | 74 | struct fdtable *fdt; |
1da177e4 | 75 | |
ab2af1f5 DS |
76 | spin_lock_bh(&f->lock); |
77 | fdt = f->next; | |
78 | f->next = NULL; | |
79 | spin_unlock_bh(&f->lock); | |
80 | while(fdt) { | |
81 | struct fdtable *next = fdt->next; | |
a892e2d7 CG |
82 | |
83 | __free_fdtable(fdt); | |
ab2af1f5 DS |
84 | fdt = next; |
85 | } | |
86 | } | |
1da177e4 | 87 | |
7cf4dc3c | 88 | static void free_fdtable_rcu(struct rcu_head *rcu) |
ab2af1f5 DS |
89 | { |
90 | struct fdtable *fdt = container_of(rcu, struct fdtable, rcu); | |
ab2af1f5 | 91 | struct fdtable_defer *fddef; |
1da177e4 | 92 | |
ab2af1f5 | 93 | BUG_ON(!fdt); |
1983e781 | 94 | BUG_ON(fdt->max_fds <= NR_OPEN_DEFAULT); |
ab2af1f5 | 95 | |
a892e2d7 | 96 | if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) { |
ab2af1f5 | 97 | kfree(fdt->fd); |
5466b456 | 98 | kfree(fdt->open_fds); |
ab2af1f5 | 99 | kfree(fdt); |
1da177e4 | 100 | } else { |
ab2af1f5 DS |
101 | fddef = &get_cpu_var(fdtable_defer_list); |
102 | spin_lock(&fddef->lock); | |
103 | fdt->next = fddef->next; | |
104 | fddef->next = fdt; | |
593be07a TH |
105 | /* vmallocs are handled from the workqueue context */ |
106 | schedule_work(&fddef->wq); | |
ab2af1f5 DS |
107 | spin_unlock(&fddef->lock); |
108 | put_cpu_var(fdtable_defer_list); | |
1da177e4 | 109 | } |
ab2af1f5 DS |
110 | } |
111 | ||
ab2af1f5 DS |
112 | /* |
113 | * Expand the fdset in the files_struct. Called with the files spinlock | |
114 | * held for write. | |
115 | */ | |
5466b456 | 116 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
ab2af1f5 | 117 | { |
5466b456 | 118 | unsigned int cpy, set; |
ab2af1f5 | 119 | |
5466b456 | 120 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
5466b456 VL |
121 | |
122 | cpy = ofdt->max_fds * sizeof(struct file *); | |
123 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); | |
124 | memcpy(nfdt->fd, ofdt->fd, cpy); | |
125 | memset((char *)(nfdt->fd) + cpy, 0, set); | |
126 | ||
127 | cpy = ofdt->max_fds / BITS_PER_BYTE; | |
128 | set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; | |
129 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | |
130 | memset((char *)(nfdt->open_fds) + cpy, 0, set); | |
131 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | |
132 | memset((char *)(nfdt->close_on_exec) + cpy, 0, set); | |
1da177e4 LT |
133 | } |
134 | ||
5466b456 | 135 | static struct fdtable * alloc_fdtable(unsigned int nr) |
1da177e4 | 136 | { |
5466b456 | 137 | struct fdtable *fdt; |
1fd36adc | 138 | void *data; |
1da177e4 | 139 | |
ab2af1f5 | 140 | /* |
5466b456 VL |
141 | * Figure out how many fds we actually want to support in this fdtable. |
142 | * Allocation steps are keyed to the size of the fdarray, since it | |
143 | * grows far faster than any of the other dynamic data. We try to fit | |
144 | * the fdarray into comfortable page-tuned chunks: starting at 1024B | |
145 | * and growing in powers of two from there on. | |
ab2af1f5 | 146 | */ |
5466b456 VL |
147 | nr /= (1024 / sizeof(struct file *)); |
148 | nr = roundup_pow_of_two(nr + 1); | |
149 | nr *= (1024 / sizeof(struct file *)); | |
5c598b34 AV |
150 | /* |
151 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open | |
152 | * had been set lower between the check in expand_files() and here. Deal | |
153 | * with that in caller, it's cheaper that way. | |
154 | * | |
155 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise | |
156 | * bitmaps handling below becomes unpleasant, to put it mildly... | |
157 | */ | |
158 | if (unlikely(nr > sysctl_nr_open)) | |
159 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; | |
bbea9f69 | 160 | |
5466b456 VL |
161 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); |
162 | if (!fdt) | |
bbea9f69 | 163 | goto out; |
5466b456 VL |
164 | fdt->max_fds = nr; |
165 | data = alloc_fdmem(nr * sizeof(struct file *)); | |
166 | if (!data) | |
167 | goto out_fdt; | |
1fd36adc DH |
168 | fdt->fd = data; |
169 | ||
170 | data = alloc_fdmem(max_t(size_t, | |
5466b456 VL |
171 | 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES)); |
172 | if (!data) | |
173 | goto out_arr; | |
1fd36adc | 174 | fdt->open_fds = data; |
5466b456 | 175 | data += nr / BITS_PER_BYTE; |
1fd36adc | 176 | fdt->close_on_exec = data; |
5466b456 VL |
177 | fdt->next = NULL; |
178 | ||
ab2af1f5 | 179 | return fdt; |
5466b456 VL |
180 | |
181 | out_arr: | |
a892e2d7 | 182 | free_fdmem(fdt->fd); |
5466b456 | 183 | out_fdt: |
ab2af1f5 | 184 | kfree(fdt); |
5466b456 | 185 | out: |
ab2af1f5 DS |
186 | return NULL; |
187 | } | |
1da177e4 | 188 | |
ab2af1f5 | 189 | /* |
74d392aa VL |
190 | * Expand the file descriptor table. |
191 | * This function will allocate a new fdtable and both fd array and fdset, of | |
192 | * the given size. | |
193 | * Return <0 error code on error; 1 on successful completion. | |
194 | * The files->file_lock should be held on entry, and will be held on exit. | |
ab2af1f5 DS |
195 | */ |
196 | static int expand_fdtable(struct files_struct *files, int nr) | |
197 | __releases(files->file_lock) | |
198 | __acquires(files->file_lock) | |
199 | { | |
74d392aa | 200 | struct fdtable *new_fdt, *cur_fdt; |
ab2af1f5 DS |
201 | |
202 | spin_unlock(&files->file_lock); | |
74d392aa | 203 | new_fdt = alloc_fdtable(nr); |
ab2af1f5 | 204 | spin_lock(&files->file_lock); |
74d392aa VL |
205 | if (!new_fdt) |
206 | return -ENOMEM; | |
5c598b34 AV |
207 | /* |
208 | * extremely unlikely race - sysctl_nr_open decreased between the check in | |
209 | * caller and alloc_fdtable(). Cheaper to catch it here... | |
210 | */ | |
211 | if (unlikely(new_fdt->max_fds <= nr)) { | |
a892e2d7 | 212 | __free_fdtable(new_fdt); |
5c598b34 AV |
213 | return -EMFILE; |
214 | } | |
ab2af1f5 | 215 | /* |
74d392aa VL |
216 | * Check again since another task may have expanded the fd table while |
217 | * we dropped the lock | |
ab2af1f5 | 218 | */ |
74d392aa | 219 | cur_fdt = files_fdtable(files); |
bbea9f69 | 220 | if (nr >= cur_fdt->max_fds) { |
74d392aa VL |
221 | /* Continue as planned */ |
222 | copy_fdtable(new_fdt, cur_fdt); | |
223 | rcu_assign_pointer(files->fdt, new_fdt); | |
4fd45812 | 224 | if (cur_fdt->max_fds > NR_OPEN_DEFAULT) |
1983e781 | 225 | call_rcu(&cur_fdt->rcu, free_fdtable_rcu); |
ab2af1f5 | 226 | } else { |
74d392aa | 227 | /* Somebody else expanded, so undo our attempt */ |
a892e2d7 | 228 | __free_fdtable(new_fdt); |
ab2af1f5 | 229 | } |
74d392aa | 230 | return 1; |
1da177e4 LT |
231 | } |
232 | ||
233 | /* | |
234 | * Expand files. | |
74d392aa VL |
235 | * This function will expand the file structures, if the requested size exceeds |
236 | * the current capacity and there is room for expansion. | |
237 | * Return <0 error code on error; 0 when nothing done; 1 when files were | |
238 | * expanded and execution may have blocked. | |
239 | * The files->file_lock should be held on entry, and will be held on exit. | |
1da177e4 LT |
240 | */ |
241 | int expand_files(struct files_struct *files, int nr) | |
242 | { | |
badf1662 | 243 | struct fdtable *fdt; |
1da177e4 | 244 | |
badf1662 | 245 | fdt = files_fdtable(files); |
4e1e018e | 246 | |
74d392aa | 247 | /* Do we need to expand? */ |
bbea9f69 | 248 | if (nr < fdt->max_fds) |
74d392aa | 249 | return 0; |
4e1e018e | 250 | |
74d392aa | 251 | /* Can we expand? */ |
9cfe015a | 252 | if (nr >= sysctl_nr_open) |
74d392aa VL |
253 | return -EMFILE; |
254 | ||
255 | /* All good, so we try */ | |
256 | return expand_fdtable(files, nr); | |
1da177e4 | 257 | } |
ab2af1f5 | 258 | |
02afc626 AV |
259 | static int count_open_files(struct fdtable *fdt) |
260 | { | |
261 | int size = fdt->max_fds; | |
262 | int i; | |
263 | ||
264 | /* Find the last open fd */ | |
1fd36adc DH |
265 | for (i = size / BITS_PER_LONG; i > 0; ) { |
266 | if (fdt->open_fds[--i]) | |
02afc626 AV |
267 | break; |
268 | } | |
1fd36adc | 269 | i = (i + 1) * BITS_PER_LONG; |
02afc626 AV |
270 | return i; |
271 | } | |
272 | ||
02afc626 AV |
273 | /* |
274 | * Allocate a new files structure and copy contents from the | |
275 | * passed in files structure. | |
276 | * errorp will be valid only when the returned files_struct is NULL. | |
277 | */ | |
278 | struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) | |
279 | { | |
280 | struct files_struct *newf; | |
281 | struct file **old_fds, **new_fds; | |
282 | int open_files, size, i; | |
283 | struct fdtable *old_fdt, *new_fdt; | |
284 | ||
285 | *errorp = -ENOMEM; | |
afbec7ff | 286 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
02afc626 AV |
287 | if (!newf) |
288 | goto out; | |
289 | ||
afbec7ff AV |
290 | atomic_set(&newf->count, 1); |
291 | ||
292 | spin_lock_init(&newf->file_lock); | |
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; | |
afbec7ff | 298 | new_fdt->fd = &newf->fd_array[0]; |
afbec7ff AV |
299 | new_fdt->next = NULL; |
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 | ||
337 | old_fds = old_fdt->fd; | |
338 | new_fds = new_fdt->fd; | |
339 | ||
1fd36adc DH |
340 | memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8); |
341 | memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8); | |
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 | ||
7cf4dc3c AV |
384 | static void close_files(struct files_struct * files) |
385 | { | |
386 | int i, j; | |
387 | struct fdtable *fdt; | |
388 | ||
389 | j = 0; | |
390 | ||
391 | /* | |
392 | * It is safe to dereference the fd table without RCU or | |
393 | * ->file_lock because this is the last reference to the | |
394 | * files structure. But use RCU to shut RCU-lockdep up. | |
395 | */ | |
396 | rcu_read_lock(); | |
397 | fdt = files_fdtable(files); | |
398 | rcu_read_unlock(); | |
399 | for (;;) { | |
400 | unsigned long set; | |
401 | i = j * BITS_PER_LONG; | |
402 | if (i >= fdt->max_fds) | |
403 | break; | |
404 | set = fdt->open_fds[j++]; | |
405 | while (set) { | |
406 | if (set & 1) { | |
407 | struct file * file = xchg(&fdt->fd[i], NULL); | |
408 | if (file) { | |
409 | filp_close(file, files); | |
410 | cond_resched(); | |
411 | } | |
412 | } | |
413 | i++; | |
414 | set >>= 1; | |
415 | } | |
416 | } | |
417 | } | |
418 | ||
419 | struct files_struct *get_files_struct(struct task_struct *task) | |
420 | { | |
421 | struct files_struct *files; | |
422 | ||
423 | task_lock(task); | |
424 | files = task->files; | |
425 | if (files) | |
426 | atomic_inc(&files->count); | |
427 | task_unlock(task); | |
428 | ||
429 | return files; | |
430 | } | |
431 | ||
432 | void put_files_struct(struct files_struct *files) | |
433 | { | |
434 | struct fdtable *fdt; | |
435 | ||
436 | if (atomic_dec_and_test(&files->count)) { | |
437 | close_files(files); | |
b9e02af0 | 438 | /* not really needed, since nobody can see us */ |
7cf4dc3c AV |
439 | rcu_read_lock(); |
440 | fdt = files_fdtable(files); | |
7cf4dc3c | 441 | rcu_read_unlock(); |
b9e02af0 AV |
442 | /* free the arrays if they are not embedded */ |
443 | if (fdt != &files->fdtab) | |
444 | __free_fdtable(fdt); | |
445 | kmem_cache_free(files_cachep, files); | |
7cf4dc3c AV |
446 | } |
447 | } | |
448 | ||
449 | void reset_files_struct(struct files_struct *files) | |
450 | { | |
451 | struct task_struct *tsk = current; | |
452 | struct files_struct *old; | |
453 | ||
454 | old = tsk->files; | |
455 | task_lock(tsk); | |
456 | tsk->files = files; | |
457 | task_unlock(tsk); | |
458 | put_files_struct(old); | |
459 | } | |
460 | ||
461 | void exit_files(struct task_struct *tsk) | |
462 | { | |
463 | struct files_struct * files = tsk->files; | |
464 | ||
465 | if (files) { | |
466 | task_lock(tsk); | |
467 | tsk->files = NULL; | |
468 | task_unlock(tsk); | |
469 | put_files_struct(files); | |
470 | } | |
471 | } | |
472 | ||
ab2af1f5 DS |
473 | static void __devinit fdtable_defer_list_init(int cpu) |
474 | { | |
475 | struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu); | |
476 | spin_lock_init(&fddef->lock); | |
65f27f38 | 477 | INIT_WORK(&fddef->wq, free_fdtable_work); |
ab2af1f5 DS |
478 | fddef->next = NULL; |
479 | } | |
480 | ||
481 | void __init files_defer_init(void) | |
482 | { | |
483 | int i; | |
0a945022 | 484 | for_each_possible_cpu(i) |
ab2af1f5 | 485 | fdtable_defer_list_init(i); |
eceea0b3 AV |
486 | sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) & |
487 | -BITS_PER_LONG; | |
ab2af1f5 | 488 | } |
f52111b1 AV |
489 | |
490 | struct files_struct init_files = { | |
491 | .count = ATOMIC_INIT(1), | |
492 | .fdt = &init_files.fdtab, | |
493 | .fdtab = { | |
494 | .max_fds = NR_OPEN_DEFAULT, | |
495 | .fd = &init_files.fd_array[0], | |
1fd36adc DH |
496 | .close_on_exec = init_files.close_on_exec_init, |
497 | .open_fds = init_files.open_fds_init, | |
f52111b1 AV |
498 | }, |
499 | .file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock), | |
500 | }; | |
1027abe8 AV |
501 | |
502 | /* | |
503 | * allocate a file descriptor, mark it busy. | |
504 | */ | |
dcfadfa4 AV |
505 | int __alloc_fd(struct files_struct *files, |
506 | unsigned start, unsigned end, unsigned flags) | |
1027abe8 | 507 | { |
1027abe8 AV |
508 | unsigned int fd; |
509 | int error; | |
510 | struct fdtable *fdt; | |
511 | ||
512 | spin_lock(&files->file_lock); | |
513 | repeat: | |
514 | fdt = files_fdtable(files); | |
515 | fd = start; | |
516 | if (fd < files->next_fd) | |
517 | fd = files->next_fd; | |
518 | ||
519 | if (fd < fdt->max_fds) | |
1fd36adc | 520 | fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd); |
1027abe8 | 521 | |
f33ff992 AV |
522 | /* |
523 | * N.B. For clone tasks sharing a files structure, this test | |
524 | * will limit the total number of files that can be opened. | |
525 | */ | |
526 | error = -EMFILE; | |
527 | if (fd >= end) | |
528 | goto out; | |
529 | ||
1027abe8 AV |
530 | error = expand_files(files, fd); |
531 | if (error < 0) | |
532 | goto out; | |
533 | ||
534 | /* | |
535 | * If we needed to expand the fs array we | |
536 | * might have blocked - try again. | |
537 | */ | |
538 | if (error) | |
539 | goto repeat; | |
540 | ||
541 | if (start <= files->next_fd) | |
542 | files->next_fd = fd + 1; | |
543 | ||
1dce27c5 | 544 | __set_open_fd(fd, fdt); |
1027abe8 | 545 | if (flags & O_CLOEXEC) |
1dce27c5 | 546 | __set_close_on_exec(fd, fdt); |
1027abe8 | 547 | else |
1dce27c5 | 548 | __clear_close_on_exec(fd, fdt); |
1027abe8 AV |
549 | error = fd; |
550 | #if 1 | |
551 | /* Sanity check */ | |
7dc52157 | 552 | if (rcu_dereference_raw(fdt->fd[fd]) != NULL) { |
1027abe8 AV |
553 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
554 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
555 | } | |
556 | #endif | |
557 | ||
558 | out: | |
559 | spin_unlock(&files->file_lock); | |
560 | return error; | |
561 | } | |
562 | ||
dcfadfa4 AV |
563 | int alloc_fd(unsigned start, unsigned flags) |
564 | { | |
565 | return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags); | |
566 | } | |
567 | ||
1a7bd226 | 568 | int get_unused_fd_flags(unsigned flags) |
1027abe8 | 569 | { |
dcfadfa4 | 570 | return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags); |
1027abe8 | 571 | } |
1a7bd226 | 572 | EXPORT_SYMBOL(get_unused_fd_flags); |
56007cae AV |
573 | |
574 | static void __put_unused_fd(struct files_struct *files, unsigned int fd) | |
575 | { | |
576 | struct fdtable *fdt = files_fdtable(files); | |
577 | __clear_open_fd(fd, fdt); | |
578 | if (fd < files->next_fd) | |
579 | files->next_fd = fd; | |
580 | } | |
581 | ||
582 | void put_unused_fd(unsigned int fd) | |
583 | { | |
584 | struct files_struct *files = current->files; | |
585 | spin_lock(&files->file_lock); | |
586 | __put_unused_fd(files, fd); | |
587 | spin_unlock(&files->file_lock); | |
588 | } | |
589 | ||
590 | EXPORT_SYMBOL(put_unused_fd); | |
591 | ||
592 | /* | |
593 | * Install a file pointer in the fd array. | |
594 | * | |
595 | * The VFS is full of places where we drop the files lock between | |
596 | * setting the open_fds bitmap and installing the file in the file | |
597 | * array. At any such point, we are vulnerable to a dup2() race | |
598 | * installing a file in the array before us. We need to detect this and | |
599 | * fput() the struct file we are about to overwrite in this case. | |
600 | * | |
601 | * It should never happen - if we allow dup2() do it, _really_ bad things | |
602 | * will follow. | |
f869e8a7 AV |
603 | * |
604 | * NOTE: __fd_install() variant is really, really low-level; don't | |
605 | * use it unless you are forced to by truly lousy API shoved down | |
606 | * your throat. 'files' *MUST* be either current->files or obtained | |
607 | * by get_files_struct(current) done by whoever had given it to you, | |
608 | * or really bad things will happen. Normally you want to use | |
609 | * fd_install() instead. | |
56007cae AV |
610 | */ |
611 | ||
f869e8a7 AV |
612 | void __fd_install(struct files_struct *files, unsigned int fd, |
613 | struct file *file) | |
56007cae | 614 | { |
56007cae AV |
615 | struct fdtable *fdt; |
616 | spin_lock(&files->file_lock); | |
617 | fdt = files_fdtable(files); | |
618 | BUG_ON(fdt->fd[fd] != NULL); | |
619 | rcu_assign_pointer(fdt->fd[fd], file); | |
620 | spin_unlock(&files->file_lock); | |
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 */ | |
662 | BUG_ON(atomic_read(&files->count) != 1); | |
663 | spin_lock(&files->file_lock); | |
664 | for (i = 0; ; i++) { | |
665 | unsigned long set; | |
666 | unsigned fd = i * BITS_PER_LONG; | |
667 | fdt = files_fdtable(files); | |
668 | if (fd >= fdt->max_fds) | |
669 | break; | |
670 | set = fdt->close_on_exec[i]; | |
671 | if (!set) | |
672 | continue; | |
673 | fdt->close_on_exec[i] = 0; | |
674 | for ( ; set ; fd++, set >>= 1) { | |
675 | struct file *file; | |
676 | if (!(set & 1)) | |
677 | continue; | |
678 | file = fdt->fd[fd]; | |
679 | if (!file) | |
680 | continue; | |
681 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
682 | __put_unused_fd(files, fd); | |
683 | spin_unlock(&files->file_lock); | |
684 | filp_close(file, files); | |
685 | cond_resched(); | |
686 | spin_lock(&files->file_lock); | |
687 | } | |
688 | ||
689 | } | |
690 | spin_unlock(&files->file_lock); | |
691 | } | |
692 | ||
0ee8cdfe AV |
693 | struct file *fget(unsigned int fd) |
694 | { | |
695 | struct file *file; | |
696 | struct files_struct *files = current->files; | |
697 | ||
698 | rcu_read_lock(); | |
699 | file = fcheck_files(files, fd); | |
700 | if (file) { | |
701 | /* File object ref couldn't be taken */ | |
702 | if (file->f_mode & FMODE_PATH || | |
703 | !atomic_long_inc_not_zero(&file->f_count)) | |
704 | file = NULL; | |
705 | } | |
706 | rcu_read_unlock(); | |
707 | ||
708 | return file; | |
709 | } | |
710 | ||
711 | EXPORT_SYMBOL(fget); | |
712 | ||
713 | struct file *fget_raw(unsigned int fd) | |
714 | { | |
715 | struct file *file; | |
716 | struct files_struct *files = current->files; | |
717 | ||
718 | rcu_read_lock(); | |
719 | file = fcheck_files(files, fd); | |
720 | if (file) { | |
721 | /* File object ref couldn't be taken */ | |
722 | if (!atomic_long_inc_not_zero(&file->f_count)) | |
723 | file = NULL; | |
724 | } | |
725 | rcu_read_unlock(); | |
726 | ||
727 | return file; | |
728 | } | |
729 | ||
730 | EXPORT_SYMBOL(fget_raw); | |
731 | ||
732 | /* | |
733 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
734 | * | |
735 | * You can use this instead of fget if you satisfy all of the following | |
736 | * conditions: | |
737 | * 1) You must call fput_light before exiting the syscall and returning control | |
738 | * to userspace (i.e. you cannot remember the returned struct file * after | |
739 | * returning to userspace). | |
740 | * 2) You must not call filp_close on the returned struct file * in between | |
741 | * calls to fget_light and fput_light. | |
742 | * 3) You must not clone the current task in between the calls to fget_light | |
743 | * and fput_light. | |
744 | * | |
745 | * The fput_needed flag returned by fget_light should be passed to the | |
746 | * corresponding fput_light. | |
747 | */ | |
748 | struct file *fget_light(unsigned int fd, int *fput_needed) | |
749 | { | |
750 | struct file *file; | |
751 | struct files_struct *files = current->files; | |
752 | ||
753 | *fput_needed = 0; | |
754 | if (atomic_read(&files->count) == 1) { | |
755 | file = fcheck_files(files, fd); | |
756 | if (file && (file->f_mode & FMODE_PATH)) | |
757 | file = NULL; | |
758 | } else { | |
759 | rcu_read_lock(); | |
760 | file = fcheck_files(files, fd); | |
761 | if (file) { | |
762 | if (!(file->f_mode & FMODE_PATH) && | |
763 | atomic_long_inc_not_zero(&file->f_count)) | |
764 | *fput_needed = 1; | |
765 | else | |
766 | /* Didn't get the reference, someone's freed */ | |
767 | file = NULL; | |
768 | } | |
769 | rcu_read_unlock(); | |
770 | } | |
771 | ||
772 | return file; | |
773 | } | |
774 | ||
775 | struct file *fget_raw_light(unsigned int fd, int *fput_needed) | |
776 | { | |
777 | struct file *file; | |
778 | struct files_struct *files = current->files; | |
779 | ||
780 | *fput_needed = 0; | |
781 | if (atomic_read(&files->count) == 1) { | |
782 | file = fcheck_files(files, fd); | |
783 | } else { | |
784 | rcu_read_lock(); | |
785 | file = fcheck_files(files, fd); | |
786 | if (file) { | |
787 | if (atomic_long_inc_not_zero(&file->f_count)) | |
788 | *fput_needed = 1; | |
789 | else | |
790 | /* Didn't get the reference, someone's freed */ | |
791 | file = NULL; | |
792 | } | |
793 | rcu_read_unlock(); | |
794 | } | |
795 | ||
796 | return file; | |
797 | } | |
fe17f22d AV |
798 | |
799 | void set_close_on_exec(unsigned int fd, int flag) | |
800 | { | |
801 | struct files_struct *files = current->files; | |
802 | struct fdtable *fdt; | |
803 | spin_lock(&files->file_lock); | |
804 | fdt = files_fdtable(files); | |
805 | if (flag) | |
806 | __set_close_on_exec(fd, fdt); | |
807 | else | |
808 | __clear_close_on_exec(fd, fdt); | |
809 | spin_unlock(&files->file_lock); | |
810 | } | |
811 | ||
812 | bool get_close_on_exec(unsigned int fd) | |
813 | { | |
814 | struct files_struct *files = current->files; | |
815 | struct fdtable *fdt; | |
816 | bool res; | |
817 | rcu_read_lock(); | |
818 | fdt = files_fdtable(files); | |
819 | res = close_on_exec(fd, fdt); | |
820 | rcu_read_unlock(); | |
821 | return res; | |
822 | } | |
823 | ||
8280d161 AV |
824 | static int do_dup2(struct files_struct *files, |
825 | struct file *file, unsigned fd, unsigned flags) | |
fe17f22d | 826 | { |
8280d161 | 827 | struct file *tofree; |
fe17f22d AV |
828 | struct fdtable *fdt; |
829 | ||
fe17f22d AV |
830 | /* |
831 | * We need to detect attempts to do dup2() over allocated but still | |
832 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
833 | * extra work in their equivalent of fget() - they insert struct | |
834 | * file immediately after grabbing descriptor, mark it larval if | |
835 | * more work (e.g. actual opening) is needed and make sure that | |
836 | * fget() treats larval files as absent. Potentially interesting, | |
837 | * but while extra work in fget() is trivial, locking implications | |
838 | * and amount of surgery on open()-related paths in VFS are not. | |
839 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
840 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
841 | * scope of POSIX or SUS, since neither considers shared descriptor | |
842 | * tables and this condition does not arise without those. | |
843 | */ | |
fe17f22d | 844 | fdt = files_fdtable(files); |
8280d161 AV |
845 | tofree = fdt->fd[fd]; |
846 | if (!tofree && fd_is_open(fd, fdt)) | |
847 | goto Ebusy; | |
fe17f22d | 848 | get_file(file); |
8280d161 AV |
849 | rcu_assign_pointer(fdt->fd[fd], file); |
850 | __set_open_fd(fd, fdt); | |
fe17f22d | 851 | if (flags & O_CLOEXEC) |
8280d161 | 852 | __set_close_on_exec(fd, fdt); |
fe17f22d | 853 | else |
8280d161 | 854 | __clear_close_on_exec(fd, fdt); |
fe17f22d AV |
855 | spin_unlock(&files->file_lock); |
856 | ||
857 | if (tofree) | |
858 | filp_close(tofree, files); | |
859 | ||
8280d161 AV |
860 | return fd; |
861 | ||
862 | Ebusy: | |
863 | spin_unlock(&files->file_lock); | |
864 | return -EBUSY; | |
865 | } | |
866 | ||
867 | int replace_fd(unsigned fd, struct file *file, unsigned flags) | |
868 | { | |
869 | int err; | |
870 | struct files_struct *files = current->files; | |
871 | ||
872 | if (!file) | |
873 | return __close_fd(files, fd); | |
874 | ||
875 | if (fd >= rlimit(RLIMIT_NOFILE)) | |
876 | return -EMFILE; | |
877 | ||
878 | spin_lock(&files->file_lock); | |
879 | err = expand_files(files, fd); | |
880 | if (unlikely(err < 0)) | |
881 | goto out_unlock; | |
882 | return do_dup2(files, file, fd, flags); | |
883 | ||
884 | out_unlock: | |
885 | spin_unlock(&files->file_lock); | |
886 | return err; | |
887 | } | |
888 | ||
889 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) | |
890 | { | |
891 | int err = -EBADF; | |
892 | struct file *file; | |
893 | struct files_struct *files = current->files; | |
894 | ||
895 | if ((flags & ~O_CLOEXEC) != 0) | |
896 | return -EINVAL; | |
897 | ||
898 | if (newfd >= rlimit(RLIMIT_NOFILE)) | |
899 | return -EMFILE; | |
900 | ||
901 | spin_lock(&files->file_lock); | |
902 | err = expand_files(files, newfd); | |
903 | file = fcheck(oldfd); | |
904 | if (unlikely(!file)) | |
905 | goto Ebadf; | |
906 | if (unlikely(err < 0)) { | |
907 | if (err == -EMFILE) | |
908 | goto Ebadf; | |
909 | goto out_unlock; | |
910 | } | |
911 | return do_dup2(files, file, newfd, flags); | |
fe17f22d AV |
912 | |
913 | Ebadf: | |
914 | err = -EBADF; | |
915 | out_unlock: | |
916 | spin_unlock(&files->file_lock); | |
917 | return err; | |
918 | } | |
919 | ||
920 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) | |
921 | { | |
922 | if (unlikely(newfd == oldfd)) { /* corner case */ | |
923 | struct files_struct *files = current->files; | |
924 | int retval = oldfd; | |
925 | ||
926 | rcu_read_lock(); | |
927 | if (!fcheck_files(files, oldfd)) | |
928 | retval = -EBADF; | |
929 | rcu_read_unlock(); | |
930 | return retval; | |
931 | } | |
932 | return sys_dup3(oldfd, newfd, 0); | |
933 | } | |
934 | ||
935 | SYSCALL_DEFINE1(dup, unsigned int, fildes) | |
936 | { | |
937 | int ret = -EBADF; | |
938 | struct file *file = fget_raw(fildes); | |
939 | ||
940 | if (file) { | |
941 | ret = get_unused_fd(); | |
942 | if (ret >= 0) | |
943 | fd_install(ret, file); | |
944 | else | |
945 | fput(file); | |
946 | } | |
947 | return ret; | |
948 | } | |
949 | ||
950 | int f_dupfd(unsigned int from, struct file *file, unsigned flags) | |
951 | { | |
952 | int err; | |
953 | if (from >= rlimit(RLIMIT_NOFILE)) | |
954 | return -EINVAL; | |
955 | err = alloc_fd(from, flags); | |
956 | if (err >= 0) { | |
957 | get_file(file); | |
958 | fd_install(err, file); | |
959 | } | |
960 | return err; | |
961 | } |