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 | ||
6fa3eb70 S |
26 | #define FD_OVER_CHECK |
27 | ||
9cfe015a | 28 | int sysctl_nr_open __read_mostly = 1024*1024; |
eceea0b3 AV |
29 | int sysctl_nr_open_min = BITS_PER_LONG; |
30 | int sysctl_nr_open_max = 1024 * 1024; /* raised later */ | |
9cfe015a | 31 | |
1fd36adc | 32 | static void *alloc_fdmem(size_t size) |
1da177e4 | 33 | { |
6d4831c2 AM |
34 | /* |
35 | * Very large allocations can stress page reclaim, so fall back to | |
36 | * vmalloc() if the allocation size will be considered "large" by the VM. | |
37 | */ | |
38 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { | |
434672c0 | 39 | void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN|__GFP_NORETRY); |
6d4831c2 AM |
40 | if (data != NULL) |
41 | return data; | |
42 | } | |
a892e2d7 | 43 | return vmalloc(size); |
1da177e4 LT |
44 | } |
45 | ||
a892e2d7 | 46 | static void free_fdmem(void *ptr) |
1da177e4 | 47 | { |
a892e2d7 | 48 | is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr); |
1da177e4 LT |
49 | } |
50 | ||
a892e2d7 | 51 | static void __free_fdtable(struct fdtable *fdt) |
1da177e4 | 52 | { |
a892e2d7 CG |
53 | free_fdmem(fdt->fd); |
54 | free_fdmem(fdt->open_fds); | |
55 | kfree(fdt); | |
ab2af1f5 | 56 | } |
1da177e4 | 57 | |
7cf4dc3c | 58 | static void free_fdtable_rcu(struct rcu_head *rcu) |
ab2af1f5 | 59 | { |
ac3e3c5b | 60 | __free_fdtable(container_of(rcu, struct fdtable, rcu)); |
ab2af1f5 DS |
61 | } |
62 | ||
ab2af1f5 DS |
63 | /* |
64 | * Expand the fdset in the files_struct. Called with the files spinlock | |
65 | * held for write. | |
66 | */ | |
5466b456 | 67 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
ab2af1f5 | 68 | { |
5466b456 | 69 | unsigned int cpy, set; |
ab2af1f5 | 70 | |
5466b456 | 71 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
5466b456 VL |
72 | |
73 | cpy = ofdt->max_fds * sizeof(struct file *); | |
74 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); | |
75 | memcpy(nfdt->fd, ofdt->fd, cpy); | |
76 | memset((char *)(nfdt->fd) + cpy, 0, set); | |
77 | ||
78 | cpy = ofdt->max_fds / BITS_PER_BYTE; | |
79 | set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; | |
80 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | |
81 | memset((char *)(nfdt->open_fds) + cpy, 0, set); | |
82 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | |
83 | memset((char *)(nfdt->close_on_exec) + cpy, 0, set); | |
1da177e4 LT |
84 | } |
85 | ||
5466b456 | 86 | static struct fdtable * alloc_fdtable(unsigned int nr) |
1da177e4 | 87 | { |
5466b456 | 88 | struct fdtable *fdt; |
1fd36adc | 89 | void *data; |
1da177e4 | 90 | |
ab2af1f5 | 91 | /* |
5466b456 VL |
92 | * Figure out how many fds we actually want to support in this fdtable. |
93 | * Allocation steps are keyed to the size of the fdarray, since it | |
94 | * grows far faster than any of the other dynamic data. We try to fit | |
95 | * the fdarray into comfortable page-tuned chunks: starting at 1024B | |
96 | * and growing in powers of two from there on. | |
ab2af1f5 | 97 | */ |
5466b456 VL |
98 | nr /= (1024 / sizeof(struct file *)); |
99 | nr = roundup_pow_of_two(nr + 1); | |
100 | nr *= (1024 / sizeof(struct file *)); | |
5c598b34 AV |
101 | /* |
102 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open | |
103 | * had been set lower between the check in expand_files() and here. Deal | |
104 | * with that in caller, it's cheaper that way. | |
105 | * | |
106 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise | |
107 | * bitmaps handling below becomes unpleasant, to put it mildly... | |
108 | */ | |
109 | if (unlikely(nr > sysctl_nr_open)) | |
110 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; | |
bbea9f69 | 111 | |
5466b456 VL |
112 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); |
113 | if (!fdt) | |
bbea9f69 | 114 | goto out; |
5466b456 VL |
115 | fdt->max_fds = nr; |
116 | data = alloc_fdmem(nr * sizeof(struct file *)); | |
117 | if (!data) | |
118 | goto out_fdt; | |
1fd36adc DH |
119 | fdt->fd = data; |
120 | ||
121 | data = alloc_fdmem(max_t(size_t, | |
5466b456 VL |
122 | 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES)); |
123 | if (!data) | |
124 | goto out_arr; | |
1fd36adc | 125 | fdt->open_fds = data; |
5466b456 | 126 | data += nr / BITS_PER_BYTE; |
1fd36adc | 127 | fdt->close_on_exec = data; |
5466b456 | 128 | |
ab2af1f5 | 129 | return fdt; |
5466b456 VL |
130 | |
131 | out_arr: | |
a892e2d7 | 132 | free_fdmem(fdt->fd); |
5466b456 | 133 | out_fdt: |
ab2af1f5 | 134 | kfree(fdt); |
5466b456 | 135 | out: |
ab2af1f5 DS |
136 | return NULL; |
137 | } | |
1da177e4 | 138 | |
ab2af1f5 | 139 | /* |
74d392aa VL |
140 | * Expand the file descriptor table. |
141 | * This function will allocate a new fdtable and both fd array and fdset, of | |
142 | * the given size. | |
143 | * Return <0 error code on error; 1 on successful completion. | |
144 | * The files->file_lock should be held on entry, and will be held on exit. | |
ab2af1f5 DS |
145 | */ |
146 | static int expand_fdtable(struct files_struct *files, int nr) | |
147 | __releases(files->file_lock) | |
148 | __acquires(files->file_lock) | |
149 | { | |
74d392aa | 150 | struct fdtable *new_fdt, *cur_fdt; |
ab2af1f5 DS |
151 | |
152 | spin_unlock(&files->file_lock); | |
74d392aa | 153 | new_fdt = alloc_fdtable(nr); |
ab2af1f5 | 154 | spin_lock(&files->file_lock); |
74d392aa VL |
155 | if (!new_fdt) |
156 | return -ENOMEM; | |
5c598b34 AV |
157 | /* |
158 | * extremely unlikely race - sysctl_nr_open decreased between the check in | |
159 | * caller and alloc_fdtable(). Cheaper to catch it here... | |
160 | */ | |
161 | if (unlikely(new_fdt->max_fds <= nr)) { | |
a892e2d7 | 162 | __free_fdtable(new_fdt); |
5c598b34 AV |
163 | return -EMFILE; |
164 | } | |
ab2af1f5 | 165 | /* |
74d392aa VL |
166 | * Check again since another task may have expanded the fd table while |
167 | * we dropped the lock | |
ab2af1f5 | 168 | */ |
74d392aa | 169 | cur_fdt = files_fdtable(files); |
bbea9f69 | 170 | if (nr >= cur_fdt->max_fds) { |
74d392aa VL |
171 | /* Continue as planned */ |
172 | copy_fdtable(new_fdt, cur_fdt); | |
173 | rcu_assign_pointer(files->fdt, new_fdt); | |
ac3e3c5b | 174 | if (cur_fdt != &files->fdtab) |
1983e781 | 175 | call_rcu(&cur_fdt->rcu, free_fdtable_rcu); |
ab2af1f5 | 176 | } else { |
74d392aa | 177 | /* Somebody else expanded, so undo our attempt */ |
a892e2d7 | 178 | __free_fdtable(new_fdt); |
ab2af1f5 | 179 | } |
74d392aa | 180 | return 1; |
1da177e4 LT |
181 | } |
182 | ||
183 | /* | |
184 | * Expand files. | |
74d392aa VL |
185 | * This function will expand the file structures, if the requested size exceeds |
186 | * the current capacity and there is room for expansion. | |
187 | * Return <0 error code on error; 0 when nothing done; 1 when files were | |
188 | * expanded and execution may have blocked. | |
189 | * The files->file_lock should be held on entry, and will be held on exit. | |
1da177e4 | 190 | */ |
ad47bd72 | 191 | static int expand_files(struct files_struct *files, int nr) |
1da177e4 | 192 | { |
badf1662 | 193 | struct fdtable *fdt; |
1da177e4 | 194 | |
badf1662 | 195 | fdt = files_fdtable(files); |
4e1e018e | 196 | |
74d392aa | 197 | /* Do we need to expand? */ |
bbea9f69 | 198 | if (nr < fdt->max_fds) |
74d392aa | 199 | return 0; |
4e1e018e | 200 | |
74d392aa | 201 | /* Can we expand? */ |
9cfe015a | 202 | if (nr >= sysctl_nr_open) |
74d392aa VL |
203 | return -EMFILE; |
204 | ||
205 | /* All good, so we try */ | |
206 | return expand_fdtable(files, nr); | |
1da177e4 | 207 | } |
ab2af1f5 | 208 | |
b8318b01 AV |
209 | static inline void __set_close_on_exec(int fd, struct fdtable *fdt) |
210 | { | |
211 | __set_bit(fd, fdt->close_on_exec); | |
212 | } | |
213 | ||
214 | static inline void __clear_close_on_exec(int fd, struct fdtable *fdt) | |
215 | { | |
216 | __clear_bit(fd, fdt->close_on_exec); | |
217 | } | |
218 | ||
219 | static inline void __set_open_fd(int fd, struct fdtable *fdt) | |
220 | { | |
221 | __set_bit(fd, fdt->open_fds); | |
222 | } | |
223 | ||
224 | static inline void __clear_open_fd(int fd, struct fdtable *fdt) | |
225 | { | |
226 | __clear_bit(fd, fdt->open_fds); | |
227 | } | |
228 | ||
02afc626 AV |
229 | static int count_open_files(struct fdtable *fdt) |
230 | { | |
231 | int size = fdt->max_fds; | |
232 | int i; | |
233 | ||
234 | /* Find the last open fd */ | |
1fd36adc DH |
235 | for (i = size / BITS_PER_LONG; i > 0; ) { |
236 | if (fdt->open_fds[--i]) | |
02afc626 AV |
237 | break; |
238 | } | |
1fd36adc | 239 | i = (i + 1) * BITS_PER_LONG; |
02afc626 AV |
240 | return i; |
241 | } | |
242 | ||
02afc626 AV |
243 | /* |
244 | * Allocate a new files structure and copy contents from the | |
245 | * passed in files structure. | |
246 | * errorp will be valid only when the returned files_struct is NULL. | |
247 | */ | |
248 | struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) | |
249 | { | |
250 | struct files_struct *newf; | |
251 | struct file **old_fds, **new_fds; | |
252 | int open_files, size, i; | |
253 | struct fdtable *old_fdt, *new_fdt; | |
254 | ||
255 | *errorp = -ENOMEM; | |
afbec7ff | 256 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
02afc626 AV |
257 | if (!newf) |
258 | goto out; | |
259 | ||
afbec7ff AV |
260 | atomic_set(&newf->count, 1); |
261 | ||
262 | spin_lock_init(&newf->file_lock); | |
263 | newf->next_fd = 0; | |
264 | new_fdt = &newf->fdtab; | |
265 | new_fdt->max_fds = NR_OPEN_DEFAULT; | |
1fd36adc DH |
266 | new_fdt->close_on_exec = newf->close_on_exec_init; |
267 | new_fdt->open_fds = newf->open_fds_init; | |
afbec7ff | 268 | new_fdt->fd = &newf->fd_array[0]; |
afbec7ff | 269 | |
02afc626 AV |
270 | spin_lock(&oldf->file_lock); |
271 | old_fdt = files_fdtable(oldf); | |
02afc626 AV |
272 | open_files = count_open_files(old_fdt); |
273 | ||
274 | /* | |
275 | * Check whether we need to allocate a larger fd array and fd set. | |
02afc626 | 276 | */ |
adbecb12 | 277 | while (unlikely(open_files > new_fdt->max_fds)) { |
02afc626 | 278 | spin_unlock(&oldf->file_lock); |
9dec3c4d | 279 | |
a892e2d7 CG |
280 | if (new_fdt != &newf->fdtab) |
281 | __free_fdtable(new_fdt); | |
adbecb12 | 282 | |
9dec3c4d AV |
283 | new_fdt = alloc_fdtable(open_files - 1); |
284 | if (!new_fdt) { | |
285 | *errorp = -ENOMEM; | |
286 | goto out_release; | |
287 | } | |
288 | ||
289 | /* beyond sysctl_nr_open; nothing to do */ | |
290 | if (unlikely(new_fdt->max_fds < open_files)) { | |
a892e2d7 | 291 | __free_fdtable(new_fdt); |
9dec3c4d | 292 | *errorp = -EMFILE; |
02afc626 | 293 | goto out_release; |
9dec3c4d | 294 | } |
9dec3c4d | 295 | |
02afc626 AV |
296 | /* |
297 | * Reacquire the oldf lock and a pointer to its fd table | |
298 | * who knows it may have a new bigger fd table. We need | |
299 | * the latest pointer. | |
300 | */ | |
301 | spin_lock(&oldf->file_lock); | |
302 | old_fdt = files_fdtable(oldf); | |
adbecb12 | 303 | open_files = count_open_files(old_fdt); |
02afc626 AV |
304 | } |
305 | ||
306 | old_fds = old_fdt->fd; | |
307 | new_fds = new_fdt->fd; | |
308 | ||
1fd36adc DH |
309 | memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8); |
310 | memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8); | |
02afc626 AV |
311 | |
312 | for (i = open_files; i != 0; i--) { | |
313 | struct file *f = *old_fds++; | |
314 | if (f) { | |
315 | get_file(f); | |
316 | } else { | |
317 | /* | |
318 | * The fd may be claimed in the fd bitmap but not yet | |
319 | * instantiated in the files array if a sibling thread | |
320 | * is partway through open(). So make sure that this | |
321 | * fd is available to the new process. | |
322 | */ | |
1dce27c5 | 323 | __clear_open_fd(open_files - i, new_fdt); |
02afc626 AV |
324 | } |
325 | rcu_assign_pointer(*new_fds++, f); | |
326 | } | |
327 | spin_unlock(&oldf->file_lock); | |
328 | ||
329 | /* compute the remainder to be cleared */ | |
330 | size = (new_fdt->max_fds - open_files) * sizeof(struct file *); | |
331 | ||
332 | /* This is long word aligned thus could use a optimized version */ | |
333 | memset(new_fds, 0, size); | |
334 | ||
335 | if (new_fdt->max_fds > open_files) { | |
1fd36adc DH |
336 | int left = (new_fdt->max_fds - open_files) / 8; |
337 | int start = open_files / BITS_PER_LONG; | |
02afc626 | 338 | |
1fd36adc DH |
339 | memset(&new_fdt->open_fds[start], 0, left); |
340 | memset(&new_fdt->close_on_exec[start], 0, left); | |
02afc626 AV |
341 | } |
342 | ||
afbec7ff AV |
343 | rcu_assign_pointer(newf->fdt, new_fdt); |
344 | ||
02afc626 AV |
345 | return newf; |
346 | ||
347 | out_release: | |
348 | kmem_cache_free(files_cachep, newf); | |
349 | out: | |
350 | return NULL; | |
351 | } | |
352 | ||
7cf4dc3c AV |
353 | static void close_files(struct files_struct * files) |
354 | { | |
355 | int i, j; | |
356 | struct fdtable *fdt; | |
357 | ||
358 | j = 0; | |
359 | ||
360 | /* | |
361 | * It is safe to dereference the fd table without RCU or | |
362 | * ->file_lock because this is the last reference to the | |
363 | * files structure. But use RCU to shut RCU-lockdep up. | |
364 | */ | |
365 | rcu_read_lock(); | |
366 | fdt = files_fdtable(files); | |
367 | rcu_read_unlock(); | |
368 | for (;;) { | |
369 | unsigned long set; | |
370 | i = j * BITS_PER_LONG; | |
371 | if (i >= fdt->max_fds) | |
372 | break; | |
373 | set = fdt->open_fds[j++]; | |
374 | while (set) { | |
375 | if (set & 1) { | |
376 | struct file * file = xchg(&fdt->fd[i], NULL); | |
377 | if (file) { | |
378 | filp_close(file, files); | |
379 | cond_resched(); | |
380 | } | |
381 | } | |
382 | i++; | |
383 | set >>= 1; | |
384 | } | |
385 | } | |
386 | } | |
387 | ||
388 | struct files_struct *get_files_struct(struct task_struct *task) | |
389 | { | |
390 | struct files_struct *files; | |
391 | ||
392 | task_lock(task); | |
393 | files = task->files; | |
394 | if (files) | |
395 | atomic_inc(&files->count); | |
396 | task_unlock(task); | |
397 | ||
398 | return files; | |
399 | } | |
400 | ||
401 | void put_files_struct(struct files_struct *files) | |
402 | { | |
403 | struct fdtable *fdt; | |
404 | ||
405 | if (atomic_dec_and_test(&files->count)) { | |
406 | close_files(files); | |
b9e02af0 | 407 | /* not really needed, since nobody can see us */ |
7cf4dc3c AV |
408 | rcu_read_lock(); |
409 | fdt = files_fdtable(files); | |
7cf4dc3c | 410 | rcu_read_unlock(); |
b9e02af0 AV |
411 | /* free the arrays if they are not embedded */ |
412 | if (fdt != &files->fdtab) | |
413 | __free_fdtable(fdt); | |
414 | kmem_cache_free(files_cachep, files); | |
7cf4dc3c AV |
415 | } |
416 | } | |
417 | ||
418 | void reset_files_struct(struct files_struct *files) | |
419 | { | |
420 | struct task_struct *tsk = current; | |
421 | struct files_struct *old; | |
422 | ||
423 | old = tsk->files; | |
424 | task_lock(tsk); | |
425 | tsk->files = files; | |
426 | task_unlock(tsk); | |
427 | put_files_struct(old); | |
428 | } | |
429 | ||
430 | void exit_files(struct task_struct *tsk) | |
431 | { | |
432 | struct files_struct * files = tsk->files; | |
433 | ||
434 | if (files) { | |
435 | task_lock(tsk); | |
436 | tsk->files = NULL; | |
437 | task_unlock(tsk); | |
438 | put_files_struct(files); | |
439 | } | |
440 | } | |
441 | ||
ab2af1f5 DS |
442 | void __init files_defer_init(void) |
443 | { | |
eceea0b3 AV |
444 | sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) & |
445 | -BITS_PER_LONG; | |
ab2af1f5 | 446 | } |
f52111b1 AV |
447 | |
448 | struct files_struct init_files = { | |
449 | .count = ATOMIC_INIT(1), | |
450 | .fdt = &init_files.fdtab, | |
451 | .fdtab = { | |
452 | .max_fds = NR_OPEN_DEFAULT, | |
453 | .fd = &init_files.fd_array[0], | |
1fd36adc DH |
454 | .close_on_exec = init_files.close_on_exec_init, |
455 | .open_fds = init_files.open_fds_init, | |
f52111b1 | 456 | }, |
eece09ec | 457 | .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock), |
f52111b1 | 458 | }; |
1027abe8 | 459 | |
6fa3eb70 S |
460 | #ifdef FD_OVER_CHECK |
461 | #define FD_CHECK_NAME_SIZE 256 | |
462 | // Declare a radix tree to construct fd set tree | |
463 | static RADIX_TREE(over_fd_tree, GFP_KERNEL); | |
464 | static LIST_HEAD(fd_listhead); | |
465 | static DEFINE_MUTEX(over_fd_mutex); | |
466 | struct over_fd_entry | |
467 | { | |
468 | int num_of_fd; | |
469 | char name[FD_CHECK_NAME_SIZE]; | |
470 | int hash; | |
471 | struct list_head fd_link; | |
472 | }; | |
473 | ||
474 | /* | |
475 | * Get File Name from FD value | |
476 | */ | |
477 | long get_file_name_from_fd(struct files_struct *files, int fd, int procid, struct over_fd_entry *res_name) | |
478 | { | |
479 | char *tmp; | |
480 | char *pathname; | |
481 | struct file *file; | |
482 | struct path path; | |
483 | spin_lock(&files->file_lock); | |
484 | file = fget(fd); | |
485 | if (!file) { | |
486 | spin_unlock(&files->file_lock); | |
487 | return (long)NULL; | |
488 | } | |
489 | path_get(&file->f_path); | |
490 | path = file->f_path; | |
491 | fput(file); | |
492 | spin_unlock(&files->file_lock); | |
493 | tmp = (char *)__get_free_page(GFP_TEMPORARY); | |
494 | if (!tmp) { | |
495 | return (long)NULL; | |
496 | } | |
497 | pathname = d_path(&path, tmp, PAGE_SIZE); | |
498 | ||
499 | path_put(&path); | |
500 | if (IS_ERR(pathname)) | |
501 | { | |
502 | free_page((unsigned long)tmp); | |
503 | return PTR_ERR(pathname); | |
504 | } /* do something here with pathname */ | |
505 | if(pathname!=NULL) | |
506 | { | |
507 | strncpy(res_name->name, pathname, FD_CHECK_NAME_SIZE - 1); | |
508 | } | |
509 | free_page((unsigned long)tmp); | |
510 | return 1; | |
511 | } | |
512 | ||
513 | unsigned int get_hash(char *name) | |
514 | { | |
515 | return full_name_hash(name, strlen(name)); | |
516 | } | |
517 | ||
518 | static struct over_fd_entry* fd_lookup(unsigned int hash) | |
519 | { | |
520 | return radix_tree_lookup(&over_fd_tree, hash); | |
521 | } | |
522 | ||
523 | static void fd_insert(struct over_fd_entry *entry) | |
524 | { | |
525 | unsigned int hash = get_hash(entry->name); | |
526 | struct over_fd_entry *find_entry = fd_lookup(hash); | |
527 | ||
528 | if(!find_entry) // Can't find the element, just add the element | |
529 | { | |
530 | entry->num_of_fd = 1; | |
531 | entry->hash = hash; | |
532 | list_add_tail(&entry->fd_link, &fd_listhead); | |
533 | radix_tree_insert(&over_fd_tree, hash, (void *)entry); | |
534 | } | |
535 | else // Cover the original element | |
536 | { | |
537 | find_entry->num_of_fd = find_entry->num_of_fd+1; | |
538 | kfree(entry); | |
539 | } | |
540 | } | |
541 | ||
542 | static void fd_delete(unsigned int hash) | |
543 | { | |
544 | radix_tree_delete(&over_fd_tree, hash); | |
545 | } | |
546 | ||
547 | void fd_show_open_files(pid_t pid, struct files_struct *files, struct fdtable *fdt) | |
548 | { | |
549 | int i=0; | |
550 | struct over_fd_entry *lentry; | |
551 | long result; | |
552 | int num_of_entry; | |
553 | int sum_fds_of_pid = 0; | |
554 | ||
555 | mutex_lock(&over_fd_mutex); | |
556 | //printk(KERN_ERR "(PID:%d)Max FD Number:%d", current->pid, fdt->max_fds); | |
557 | for(i=0; i<fdt->max_fds; i++) { | |
558 | struct over_fd_entry *entry = (struct over_fd_entry*)kzalloc(sizeof(struct over_fd_entry), GFP_KERNEL); | |
559 | if(!entry) { | |
560 | pr_debug("[FD_LEAK](PID:%d)Empty FD:%d", pid, i); | |
561 | } | |
562 | else { | |
563 | memset(entry->name, 0, sizeof entry->name); | |
564 | result = get_file_name_from_fd(files, i, pid, entry); | |
565 | if(result==1) { | |
566 | fd_insert(entry); | |
567 | sum_fds_of_pid++; | |
568 | } | |
569 | } | |
570 | } | |
571 | for(;;) { | |
572 | if(list_empty(&fd_listhead)) { | |
573 | break; | |
574 | } | |
575 | lentry = list_entry((&fd_listhead)->next, struct over_fd_entry, fd_link); | |
576 | num_of_entry = lentry->num_of_fd; | |
577 | if(lentry != NULL && lentry->name!=NULL) | |
578 | pr_debug("[FD_LEAK]OverAllocFDError(PID:%d fileName:%s Num:%d)\n", pid, lentry->name, num_of_entry); | |
579 | else | |
580 | pr_debug("[FD_LEAK]OverAllocFDError(PID:%d fileName:%s Num:%d)\n", pid, "NULL", num_of_entry); | |
581 | list_del((&fd_listhead)->next); | |
582 | fd_delete(lentry->hash); | |
583 | kfree(lentry); | |
584 | } | |
585 | if(sum_fds_of_pid) { | |
586 | pr_debug("[FD_LEAK]OverAllocFDError(PID:%d totalFDs:%d)\n", pid, sum_fds_of_pid); | |
587 | } | |
588 | mutex_unlock(&over_fd_mutex); | |
589 | } | |
590 | #endif | |
591 | ||
1027abe8 AV |
592 | /* |
593 | * allocate a file descriptor, mark it busy. | |
594 | */ | |
dcfadfa4 AV |
595 | int __alloc_fd(struct files_struct *files, |
596 | unsigned start, unsigned end, unsigned flags) | |
1027abe8 | 597 | { |
1027abe8 AV |
598 | unsigned int fd; |
599 | int error; | |
600 | struct fdtable *fdt; | |
601 | ||
602 | spin_lock(&files->file_lock); | |
603 | repeat: | |
604 | fdt = files_fdtable(files); | |
605 | fd = start; | |
606 | if (fd < files->next_fd) | |
607 | fd = files->next_fd; | |
608 | ||
609 | if (fd < fdt->max_fds) | |
1fd36adc | 610 | fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd); |
1027abe8 | 611 | |
f33ff992 AV |
612 | /* |
613 | * N.B. For clone tasks sharing a files structure, this test | |
614 | * will limit the total number of files that can be opened. | |
615 | */ | |
616 | error = -EMFILE; | |
617 | if (fd >= end) | |
618 | goto out; | |
619 | ||
1027abe8 AV |
620 | error = expand_files(files, fd); |
621 | if (error < 0) | |
622 | goto out; | |
623 | ||
624 | /* | |
625 | * If we needed to expand the fs array we | |
626 | * might have blocked - try again. | |
627 | */ | |
628 | if (error) | |
629 | goto repeat; | |
630 | ||
631 | if (start <= files->next_fd) | |
632 | files->next_fd = fd + 1; | |
633 | ||
1dce27c5 | 634 | __set_open_fd(fd, fdt); |
1027abe8 | 635 | if (flags & O_CLOEXEC) |
1dce27c5 | 636 | __set_close_on_exec(fd, fdt); |
1027abe8 | 637 | else |
1dce27c5 | 638 | __clear_close_on_exec(fd, fdt); |
1027abe8 AV |
639 | error = fd; |
640 | #if 1 | |
641 | /* Sanity check */ | |
7dc52157 | 642 | if (rcu_dereference_raw(fdt->fd[fd]) != NULL) { |
1027abe8 AV |
643 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
644 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
645 | } | |
646 | #endif | |
647 | ||
648 | out: | |
649 | spin_unlock(&files->file_lock); | |
6fa3eb70 S |
650 | #ifdef FD_OVER_CHECK |
651 | if(error == -EMFILE) { | |
652 | static int dump_current_open_files = 0; | |
653 | if(!dump_current_open_files && | |
654 | strcmp(current->comm, "Backbone")) { /*add Backbone into FD white list for skype*/ | |
655 | dump_current_open_files = 0x1; | |
656 | pr_debug("[FD_LEAK](PID:%d)fd over RLIMIT_NOFILE:%ld", current->pid, rlimit(RLIMIT_NOFILE)); | |
657 | fd_show_open_files(current->pid, files, fdt); | |
658 | } | |
659 | } | |
660 | #endif | |
1027abe8 AV |
661 | return error; |
662 | } | |
663 | ||
ad47bd72 | 664 | static int alloc_fd(unsigned start, unsigned flags) |
dcfadfa4 AV |
665 | { |
666 | return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags); | |
667 | } | |
668 | ||
1a7bd226 | 669 | int get_unused_fd_flags(unsigned flags) |
1027abe8 | 670 | { |
dcfadfa4 | 671 | return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags); |
1027abe8 | 672 | } |
1a7bd226 | 673 | EXPORT_SYMBOL(get_unused_fd_flags); |
56007cae AV |
674 | |
675 | static void __put_unused_fd(struct files_struct *files, unsigned int fd) | |
676 | { | |
677 | struct fdtable *fdt = files_fdtable(files); | |
678 | __clear_open_fd(fd, fdt); | |
679 | if (fd < files->next_fd) | |
680 | files->next_fd = fd; | |
681 | } | |
682 | ||
683 | void put_unused_fd(unsigned int fd) | |
684 | { | |
685 | struct files_struct *files = current->files; | |
686 | spin_lock(&files->file_lock); | |
687 | __put_unused_fd(files, fd); | |
688 | spin_unlock(&files->file_lock); | |
689 | } | |
690 | ||
691 | EXPORT_SYMBOL(put_unused_fd); | |
692 | ||
693 | /* | |
694 | * Install a file pointer in the fd array. | |
695 | * | |
696 | * The VFS is full of places where we drop the files lock between | |
697 | * setting the open_fds bitmap and installing the file in the file | |
698 | * array. At any such point, we are vulnerable to a dup2() race | |
699 | * installing a file in the array before us. We need to detect this and | |
700 | * fput() the struct file we are about to overwrite in this case. | |
701 | * | |
702 | * It should never happen - if we allow dup2() do it, _really_ bad things | |
703 | * will follow. | |
f869e8a7 AV |
704 | * |
705 | * NOTE: __fd_install() variant is really, really low-level; don't | |
706 | * use it unless you are forced to by truly lousy API shoved down | |
707 | * your throat. 'files' *MUST* be either current->files or obtained | |
708 | * by get_files_struct(current) done by whoever had given it to you, | |
709 | * or really bad things will happen. Normally you want to use | |
710 | * fd_install() instead. | |
56007cae AV |
711 | */ |
712 | ||
f869e8a7 AV |
713 | void __fd_install(struct files_struct *files, unsigned int fd, |
714 | struct file *file) | |
56007cae | 715 | { |
56007cae AV |
716 | struct fdtable *fdt; |
717 | spin_lock(&files->file_lock); | |
718 | fdt = files_fdtable(files); | |
719 | BUG_ON(fdt->fd[fd] != NULL); | |
720 | rcu_assign_pointer(fdt->fd[fd], file); | |
721 | spin_unlock(&files->file_lock); | |
722 | } | |
723 | ||
f869e8a7 AV |
724 | void fd_install(unsigned int fd, struct file *file) |
725 | { | |
726 | __fd_install(current->files, fd, file); | |
727 | } | |
728 | ||
56007cae | 729 | EXPORT_SYMBOL(fd_install); |
0ee8cdfe | 730 | |
483ce1d4 AV |
731 | /* |
732 | * The same warnings as for __alloc_fd()/__fd_install() apply here... | |
733 | */ | |
734 | int __close_fd(struct files_struct *files, unsigned fd) | |
735 | { | |
736 | struct file *file; | |
737 | struct fdtable *fdt; | |
738 | ||
739 | spin_lock(&files->file_lock); | |
740 | fdt = files_fdtable(files); | |
741 | if (fd >= fdt->max_fds) | |
742 | goto out_unlock; | |
743 | file = fdt->fd[fd]; | |
744 | if (!file) | |
745 | goto out_unlock; | |
746 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
747 | __clear_close_on_exec(fd, fdt); | |
748 | __put_unused_fd(files, fd); | |
749 | spin_unlock(&files->file_lock); | |
750 | return filp_close(file, files); | |
751 | ||
752 | out_unlock: | |
753 | spin_unlock(&files->file_lock); | |
754 | return -EBADF; | |
755 | } | |
756 | ||
6a6d27de AV |
757 | void do_close_on_exec(struct files_struct *files) |
758 | { | |
759 | unsigned i; | |
760 | struct fdtable *fdt; | |
761 | ||
762 | /* exec unshares first */ | |
6a6d27de AV |
763 | spin_lock(&files->file_lock); |
764 | for (i = 0; ; i++) { | |
765 | unsigned long set; | |
766 | unsigned fd = i * BITS_PER_LONG; | |
767 | fdt = files_fdtable(files); | |
768 | if (fd >= fdt->max_fds) | |
769 | break; | |
770 | set = fdt->close_on_exec[i]; | |
771 | if (!set) | |
772 | continue; | |
773 | fdt->close_on_exec[i] = 0; | |
774 | for ( ; set ; fd++, set >>= 1) { | |
775 | struct file *file; | |
776 | if (!(set & 1)) | |
777 | continue; | |
778 | file = fdt->fd[fd]; | |
779 | if (!file) | |
780 | continue; | |
781 | rcu_assign_pointer(fdt->fd[fd], NULL); | |
782 | __put_unused_fd(files, fd); | |
783 | spin_unlock(&files->file_lock); | |
784 | filp_close(file, files); | |
785 | cond_resched(); | |
786 | spin_lock(&files->file_lock); | |
787 | } | |
788 | ||
789 | } | |
790 | spin_unlock(&files->file_lock); | |
791 | } | |
792 | ||
0ee8cdfe AV |
793 | struct file *fget(unsigned int fd) |
794 | { | |
795 | struct file *file; | |
796 | struct files_struct *files = current->files; | |
797 | ||
798 | rcu_read_lock(); | |
799 | file = fcheck_files(files, fd); | |
800 | if (file) { | |
801 | /* File object ref couldn't be taken */ | |
802 | if (file->f_mode & FMODE_PATH || | |
803 | !atomic_long_inc_not_zero(&file->f_count)) | |
804 | file = NULL; | |
805 | } | |
806 | rcu_read_unlock(); | |
807 | ||
808 | return file; | |
809 | } | |
810 | ||
811 | EXPORT_SYMBOL(fget); | |
812 | ||
813 | struct file *fget_raw(unsigned int fd) | |
814 | { | |
815 | struct file *file; | |
816 | struct files_struct *files = current->files; | |
817 | ||
818 | rcu_read_lock(); | |
819 | file = fcheck_files(files, fd); | |
820 | if (file) { | |
821 | /* File object ref couldn't be taken */ | |
822 | if (!atomic_long_inc_not_zero(&file->f_count)) | |
823 | file = NULL; | |
824 | } | |
825 | rcu_read_unlock(); | |
826 | ||
827 | return file; | |
828 | } | |
829 | ||
830 | EXPORT_SYMBOL(fget_raw); | |
831 | ||
832 | /* | |
833 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
834 | * | |
835 | * You can use this instead of fget if you satisfy all of the following | |
836 | * conditions: | |
837 | * 1) You must call fput_light before exiting the syscall and returning control | |
838 | * to userspace (i.e. you cannot remember the returned struct file * after | |
839 | * returning to userspace). | |
840 | * 2) You must not call filp_close on the returned struct file * in between | |
841 | * calls to fget_light and fput_light. | |
842 | * 3) You must not clone the current task in between the calls to fget_light | |
843 | * and fput_light. | |
844 | * | |
845 | * The fput_needed flag returned by fget_light should be passed to the | |
846 | * corresponding fput_light. | |
847 | */ | |
848 | struct file *fget_light(unsigned int fd, int *fput_needed) | |
849 | { | |
850 | struct file *file; | |
851 | struct files_struct *files = current->files; | |
852 | ||
853 | *fput_needed = 0; | |
854 | if (atomic_read(&files->count) == 1) { | |
855 | file = fcheck_files(files, fd); | |
856 | if (file && (file->f_mode & FMODE_PATH)) | |
857 | file = NULL; | |
858 | } else { | |
859 | rcu_read_lock(); | |
860 | file = fcheck_files(files, fd); | |
861 | if (file) { | |
862 | if (!(file->f_mode & FMODE_PATH) && | |
863 | atomic_long_inc_not_zero(&file->f_count)) | |
864 | *fput_needed = 1; | |
865 | else | |
866 | /* Didn't get the reference, someone's freed */ | |
867 | file = NULL; | |
868 | } | |
869 | rcu_read_unlock(); | |
870 | } | |
871 | ||
872 | return file; | |
873 | } | |
4557c669 | 874 | EXPORT_SYMBOL(fget_light); |
0ee8cdfe AV |
875 | |
876 | struct file *fget_raw_light(unsigned int fd, int *fput_needed) | |
877 | { | |
878 | struct file *file; | |
879 | struct files_struct *files = current->files; | |
880 | ||
881 | *fput_needed = 0; | |
882 | if (atomic_read(&files->count) == 1) { | |
883 | file = fcheck_files(files, fd); | |
884 | } else { | |
885 | rcu_read_lock(); | |
886 | file = fcheck_files(files, fd); | |
887 | if (file) { | |
888 | if (atomic_long_inc_not_zero(&file->f_count)) | |
889 | *fput_needed = 1; | |
890 | else | |
891 | /* Didn't get the reference, someone's freed */ | |
892 | file = NULL; | |
893 | } | |
894 | rcu_read_unlock(); | |
895 | } | |
896 | ||
897 | return file; | |
898 | } | |
fe17f22d AV |
899 | |
900 | void set_close_on_exec(unsigned int fd, int flag) | |
901 | { | |
902 | struct files_struct *files = current->files; | |
903 | struct fdtable *fdt; | |
904 | spin_lock(&files->file_lock); | |
905 | fdt = files_fdtable(files); | |
906 | if (flag) | |
907 | __set_close_on_exec(fd, fdt); | |
908 | else | |
909 | __clear_close_on_exec(fd, fdt); | |
910 | spin_unlock(&files->file_lock); | |
911 | } | |
912 | ||
913 | bool get_close_on_exec(unsigned int fd) | |
914 | { | |
915 | struct files_struct *files = current->files; | |
916 | struct fdtable *fdt; | |
917 | bool res; | |
918 | rcu_read_lock(); | |
919 | fdt = files_fdtable(files); | |
920 | res = close_on_exec(fd, fdt); | |
921 | rcu_read_unlock(); | |
922 | return res; | |
923 | } | |
924 | ||
8280d161 AV |
925 | static int do_dup2(struct files_struct *files, |
926 | struct file *file, unsigned fd, unsigned flags) | |
fe17f22d | 927 | { |
8280d161 | 928 | struct file *tofree; |
fe17f22d AV |
929 | struct fdtable *fdt; |
930 | ||
fe17f22d AV |
931 | /* |
932 | * We need to detect attempts to do dup2() over allocated but still | |
933 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
934 | * extra work in their equivalent of fget() - they insert struct | |
935 | * file immediately after grabbing descriptor, mark it larval if | |
936 | * more work (e.g. actual opening) is needed and make sure that | |
937 | * fget() treats larval files as absent. Potentially interesting, | |
938 | * but while extra work in fget() is trivial, locking implications | |
939 | * and amount of surgery on open()-related paths in VFS are not. | |
940 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
941 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
942 | * scope of POSIX or SUS, since neither considers shared descriptor | |
943 | * tables and this condition does not arise without those. | |
944 | */ | |
fe17f22d | 945 | fdt = files_fdtable(files); |
8280d161 AV |
946 | tofree = fdt->fd[fd]; |
947 | if (!tofree && fd_is_open(fd, fdt)) | |
948 | goto Ebusy; | |
fe17f22d | 949 | get_file(file); |
8280d161 AV |
950 | rcu_assign_pointer(fdt->fd[fd], file); |
951 | __set_open_fd(fd, fdt); | |
fe17f22d | 952 | if (flags & O_CLOEXEC) |
8280d161 | 953 | __set_close_on_exec(fd, fdt); |
fe17f22d | 954 | else |
8280d161 | 955 | __clear_close_on_exec(fd, fdt); |
fe17f22d AV |
956 | spin_unlock(&files->file_lock); |
957 | ||
958 | if (tofree) | |
959 | filp_close(tofree, files); | |
960 | ||
8280d161 AV |
961 | return fd; |
962 | ||
963 | Ebusy: | |
964 | spin_unlock(&files->file_lock); | |
965 | return -EBUSY; | |
966 | } | |
967 | ||
968 | int replace_fd(unsigned fd, struct file *file, unsigned flags) | |
969 | { | |
970 | int err; | |
971 | struct files_struct *files = current->files; | |
972 | ||
973 | if (!file) | |
974 | return __close_fd(files, fd); | |
975 | ||
976 | if (fd >= rlimit(RLIMIT_NOFILE)) | |
08f05c49 | 977 | return -EBADF; |
8280d161 AV |
978 | |
979 | spin_lock(&files->file_lock); | |
980 | err = expand_files(files, fd); | |
981 | if (unlikely(err < 0)) | |
982 | goto out_unlock; | |
983 | return do_dup2(files, file, fd, flags); | |
984 | ||
985 | out_unlock: | |
986 | spin_unlock(&files->file_lock); | |
987 | return err; | |
988 | } | |
989 | ||
990 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) | |
991 | { | |
992 | int err = -EBADF; | |
993 | struct file *file; | |
994 | struct files_struct *files = current->files; | |
995 | ||
996 | if ((flags & ~O_CLOEXEC) != 0) | |
997 | return -EINVAL; | |
998 | ||
aed97647 RJ |
999 | if (unlikely(oldfd == newfd)) |
1000 | return -EINVAL; | |
1001 | ||
8280d161 | 1002 | if (newfd >= rlimit(RLIMIT_NOFILE)) |
08f05c49 | 1003 | return -EBADF; |
8280d161 AV |
1004 | |
1005 | spin_lock(&files->file_lock); | |
1006 | err = expand_files(files, newfd); | |
1007 | file = fcheck(oldfd); | |
1008 | if (unlikely(!file)) | |
1009 | goto Ebadf; | |
1010 | if (unlikely(err < 0)) { | |
1011 | if (err == -EMFILE) | |
1012 | goto Ebadf; | |
1013 | goto out_unlock; | |
1014 | } | |
1015 | return do_dup2(files, file, newfd, flags); | |
fe17f22d AV |
1016 | |
1017 | Ebadf: | |
1018 | err = -EBADF; | |
1019 | out_unlock: | |
1020 | spin_unlock(&files->file_lock); | |
1021 | return err; | |
1022 | } | |
1023 | ||
1024 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) | |
1025 | { | |
1026 | if (unlikely(newfd == oldfd)) { /* corner case */ | |
1027 | struct files_struct *files = current->files; | |
1028 | int retval = oldfd; | |
1029 | ||
1030 | rcu_read_lock(); | |
1031 | if (!fcheck_files(files, oldfd)) | |
1032 | retval = -EBADF; | |
1033 | rcu_read_unlock(); | |
1034 | return retval; | |
1035 | } | |
1036 | return sys_dup3(oldfd, newfd, 0); | |
1037 | } | |
1038 | ||
1039 | SYSCALL_DEFINE1(dup, unsigned int, fildes) | |
1040 | { | |
1041 | int ret = -EBADF; | |
1042 | struct file *file = fget_raw(fildes); | |
1043 | ||
1044 | if (file) { | |
1045 | ret = get_unused_fd(); | |
1046 | if (ret >= 0) | |
1047 | fd_install(ret, file); | |
1048 | else | |
1049 | fput(file); | |
1050 | } | |
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | int f_dupfd(unsigned int from, struct file *file, unsigned flags) | |
1055 | { | |
1056 | int err; | |
1057 | if (from >= rlimit(RLIMIT_NOFILE)) | |
1058 | return -EINVAL; | |
1059 | err = alloc_fd(from, flags); | |
1060 | if (err >= 0) { | |
1061 | get_file(file); | |
1062 | fd_install(err, file); | |
1063 | } | |
1064 | return err; | |
1065 | } | |
c3c073f8 AV |
1066 | |
1067 | int iterate_fd(struct files_struct *files, unsigned n, | |
1068 | int (*f)(const void *, struct file *, unsigned), | |
1069 | const void *p) | |
1070 | { | |
1071 | struct fdtable *fdt; | |
c3c073f8 AV |
1072 | int res = 0; |
1073 | if (!files) | |
1074 | return 0; | |
1075 | spin_lock(&files->file_lock); | |
a77cfcb4 AV |
1076 | for (fdt = files_fdtable(files); n < fdt->max_fds; n++) { |
1077 | struct file *file; | |
1078 | file = rcu_dereference_check_fdtable(files, fdt->fd[n]); | |
1079 | if (!file) | |
1080 | continue; | |
1081 | res = f(p, file, n); | |
1082 | if (res) | |
1083 | break; | |
c3c073f8 AV |
1084 | } |
1085 | spin_unlock(&files->file_lock); | |
1086 | return res; | |
1087 | } | |
1088 | EXPORT_SYMBOL(iterate_fd); |