Root/
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 | |
9 | #include <linux/module.h> |
10 | #include <linux/fs.h> |
11 | #include <linux/mm.h> |
12 | #include <linux/time.h> |
13 | #include <linux/slab.h> |
14 | #include <linux/vmalloc.h> |
15 | #include <linux/file.h> |
16 | #include <linux/fdtable.h> |
17 | #include <linux/bitops.h> |
18 | #include <linux/interrupt.h> |
19 | #include <linux/spinlock.h> |
20 | #include <linux/rcupdate.h> |
21 | #include <linux/workqueue.h> |
22 | |
23 | struct fdtable_defer { |
24 | spinlock_t lock; |
25 | struct work_struct wq; |
26 | struct fdtable *next; |
27 | }; |
28 | |
29 | int sysctl_nr_open __read_mostly = 1024*1024; |
30 | int sysctl_nr_open_min = BITS_PER_LONG; |
31 | int sysctl_nr_open_max = 1024 * 1024; /* raised later */ |
32 | |
33 | /* |
34 | * We use this list to defer free fdtables that have vmalloced |
35 | * sets/arrays. By keeping a per-cpu list, we avoid having to embed |
36 | * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in |
37 | * this per-task structure. |
38 | */ |
39 | static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list); |
40 | |
41 | static inline void * alloc_fdmem(unsigned int size) |
42 | { |
43 | if (size <= PAGE_SIZE) |
44 | return kmalloc(size, GFP_KERNEL); |
45 | else |
46 | return vmalloc(size); |
47 | } |
48 | |
49 | static inline void free_fdarr(struct fdtable *fdt) |
50 | { |
51 | if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) |
52 | kfree(fdt->fd); |
53 | else |
54 | vfree(fdt->fd); |
55 | } |
56 | |
57 | static inline void free_fdset(struct fdtable *fdt) |
58 | { |
59 | if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2)) |
60 | kfree(fdt->open_fds); |
61 | else |
62 | vfree(fdt->open_fds); |
63 | } |
64 | |
65 | static void free_fdtable_work(struct work_struct *work) |
66 | { |
67 | struct fdtable_defer *f = |
68 | container_of(work, struct fdtable_defer, wq); |
69 | struct fdtable *fdt; |
70 | |
71 | spin_lock_bh(&f->lock); |
72 | fdt = f->next; |
73 | f->next = NULL; |
74 | spin_unlock_bh(&f->lock); |
75 | while(fdt) { |
76 | struct fdtable *next = fdt->next; |
77 | vfree(fdt->fd); |
78 | free_fdset(fdt); |
79 | kfree(fdt); |
80 | fdt = next; |
81 | } |
82 | } |
83 | |
84 | void free_fdtable_rcu(struct rcu_head *rcu) |
85 | { |
86 | struct fdtable *fdt = container_of(rcu, struct fdtable, rcu); |
87 | struct fdtable_defer *fddef; |
88 | |
89 | BUG_ON(!fdt); |
90 | |
91 | if (fdt->max_fds <= NR_OPEN_DEFAULT) { |
92 | /* |
93 | * This fdtable is embedded in the files structure and that |
94 | * structure itself is getting destroyed. |
95 | */ |
96 | kmem_cache_free(files_cachep, |
97 | container_of(fdt, struct files_struct, fdtab)); |
98 | return; |
99 | } |
100 | if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) { |
101 | kfree(fdt->fd); |
102 | kfree(fdt->open_fds); |
103 | kfree(fdt); |
104 | } else { |
105 | fddef = &get_cpu_var(fdtable_defer_list); |
106 | spin_lock(&fddef->lock); |
107 | fdt->next = fddef->next; |
108 | fddef->next = fdt; |
109 | /* vmallocs are handled from the workqueue context */ |
110 | schedule_work(&fddef->wq); |
111 | spin_unlock(&fddef->lock); |
112 | put_cpu_var(fdtable_defer_list); |
113 | } |
114 | } |
115 | |
116 | /* |
117 | * Expand the fdset in the files_struct. Called with the files spinlock |
118 | * held for write. |
119 | */ |
120 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
121 | { |
122 | unsigned int cpy, set; |
123 | |
124 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
125 | |
126 | cpy = ofdt->max_fds * sizeof(struct file *); |
127 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); |
128 | memcpy(nfdt->fd, ofdt->fd, cpy); |
129 | memset((char *)(nfdt->fd) + cpy, 0, set); |
130 | |
131 | cpy = ofdt->max_fds / BITS_PER_BYTE; |
132 | set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; |
133 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); |
134 | memset((char *)(nfdt->open_fds) + cpy, 0, set); |
135 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); |
136 | memset((char *)(nfdt->close_on_exec) + cpy, 0, set); |
137 | } |
138 | |
139 | static struct fdtable * alloc_fdtable(unsigned int nr) |
140 | { |
141 | struct fdtable *fdt; |
142 | char *data; |
143 | |
144 | /* |
145 | * Figure out how many fds we actually want to support in this fdtable. |
146 | * Allocation steps are keyed to the size of the fdarray, since it |
147 | * grows far faster than any of the other dynamic data. We try to fit |
148 | * the fdarray into comfortable page-tuned chunks: starting at 1024B |
149 | * and growing in powers of two from there on. |
150 | */ |
151 | nr /= (1024 / sizeof(struct file *)); |
152 | nr = roundup_pow_of_two(nr + 1); |
153 | nr *= (1024 / sizeof(struct file *)); |
154 | /* |
155 | * Note that this can drive nr *below* what we had passed if sysctl_nr_open |
156 | * had been set lower between the check in expand_files() and here. Deal |
157 | * with that in caller, it's cheaper that way. |
158 | * |
159 | * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise |
160 | * bitmaps handling below becomes unpleasant, to put it mildly... |
161 | */ |
162 | if (unlikely(nr > sysctl_nr_open)) |
163 | nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; |
164 | |
165 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); |
166 | if (!fdt) |
167 | goto out; |
168 | fdt->max_fds = nr; |
169 | data = alloc_fdmem(nr * sizeof(struct file *)); |
170 | if (!data) |
171 | goto out_fdt; |
172 | fdt->fd = (struct file **)data; |
173 | data = alloc_fdmem(max_t(unsigned int, |
174 | 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES)); |
175 | if (!data) |
176 | goto out_arr; |
177 | fdt->open_fds = (fd_set *)data; |
178 | data += nr / BITS_PER_BYTE; |
179 | fdt->close_on_exec = (fd_set *)data; |
180 | INIT_RCU_HEAD(&fdt->rcu); |
181 | fdt->next = NULL; |
182 | |
183 | return fdt; |
184 | |
185 | out_arr: |
186 | free_fdarr(fdt); |
187 | out_fdt: |
188 | kfree(fdt); |
189 | out: |
190 | return NULL; |
191 | } |
192 | |
193 | /* |
194 | * Expand the file descriptor table. |
195 | * This function will allocate a new fdtable and both fd array and fdset, of |
196 | * the given size. |
197 | * Return <0 error code on error; 1 on successful completion. |
198 | * The files->file_lock should be held on entry, and will be held on exit. |
199 | */ |
200 | static int expand_fdtable(struct files_struct *files, int nr) |
201 | __releases(files->file_lock) |
202 | __acquires(files->file_lock) |
203 | { |
204 | struct fdtable *new_fdt, *cur_fdt; |
205 | |
206 | spin_unlock(&files->file_lock); |
207 | new_fdt = alloc_fdtable(nr); |
208 | spin_lock(&files->file_lock); |
209 | if (!new_fdt) |
210 | return -ENOMEM; |
211 | /* |
212 | * extremely unlikely race - sysctl_nr_open decreased between the check in |
213 | * caller and alloc_fdtable(). Cheaper to catch it here... |
214 | */ |
215 | if (unlikely(new_fdt->max_fds <= nr)) { |
216 | free_fdarr(new_fdt); |
217 | free_fdset(new_fdt); |
218 | kfree(new_fdt); |
219 | return -EMFILE; |
220 | } |
221 | /* |
222 | * Check again since another task may have expanded the fd table while |
223 | * we dropped the lock |
224 | */ |
225 | cur_fdt = files_fdtable(files); |
226 | if (nr >= cur_fdt->max_fds) { |
227 | /* Continue as planned */ |
228 | copy_fdtable(new_fdt, cur_fdt); |
229 | rcu_assign_pointer(files->fdt, new_fdt); |
230 | if (cur_fdt->max_fds > NR_OPEN_DEFAULT) |
231 | free_fdtable(cur_fdt); |
232 | } else { |
233 | /* Somebody else expanded, so undo our attempt */ |
234 | free_fdarr(new_fdt); |
235 | free_fdset(new_fdt); |
236 | kfree(new_fdt); |
237 | } |
238 | return 1; |
239 | } |
240 | |
241 | /* |
242 | * Expand files. |
243 | * This function will expand the file structures, if the requested size exceeds |
244 | * the current capacity and there is room for expansion. |
245 | * Return <0 error code on error; 0 when nothing done; 1 when files were |
246 | * expanded and execution may have blocked. |
247 | * The files->file_lock should be held on entry, and will be held on exit. |
248 | */ |
249 | int expand_files(struct files_struct *files, int nr) |
250 | { |
251 | struct fdtable *fdt; |
252 | |
253 | fdt = files_fdtable(files); |
254 | |
255 | /* |
256 | * N.B. For clone tasks sharing a files structure, this test |
257 | * will limit the total number of files that can be opened. |
258 | */ |
259 | if (nr >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur) |
260 | return -EMFILE; |
261 | |
262 | /* Do we need to expand? */ |
263 | if (nr < fdt->max_fds) |
264 | return 0; |
265 | |
266 | /* Can we expand? */ |
267 | if (nr >= sysctl_nr_open) |
268 | return -EMFILE; |
269 | |
270 | /* All good, so we try */ |
271 | return expand_fdtable(files, nr); |
272 | } |
273 | EXPORT_SYMBOL_GPL(expand_files); |
274 | |
275 | static int count_open_files(struct fdtable *fdt) |
276 | { |
277 | int size = fdt->max_fds; |
278 | int i; |
279 | |
280 | /* Find the last open fd */ |
281 | for (i = size/(8*sizeof(long)); i > 0; ) { |
282 | if (fdt->open_fds->fds_bits[--i]) |
283 | break; |
284 | } |
285 | i = (i+1) * 8 * sizeof(long); |
286 | return i; |
287 | } |
288 | |
289 | /* |
290 | * Allocate a new files structure and copy contents from the |
291 | * passed in files structure. |
292 | * errorp will be valid only when the returned files_struct is NULL. |
293 | */ |
294 | struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) |
295 | { |
296 | struct files_struct *newf; |
297 | struct file **old_fds, **new_fds; |
298 | int open_files, size, i; |
299 | struct fdtable *old_fdt, *new_fdt; |
300 | |
301 | *errorp = -ENOMEM; |
302 | newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); |
303 | if (!newf) |
304 | goto out; |
305 | |
306 | atomic_set(&newf->count, 1); |
307 | |
308 | spin_lock_init(&newf->file_lock); |
309 | newf->next_fd = 0; |
310 | new_fdt = &newf->fdtab; |
311 | new_fdt->max_fds = NR_OPEN_DEFAULT; |
312 | new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init; |
313 | new_fdt->open_fds = (fd_set *)&newf->open_fds_init; |
314 | new_fdt->fd = &newf->fd_array[0]; |
315 | INIT_RCU_HEAD(&new_fdt->rcu); |
316 | new_fdt->next = NULL; |
317 | |
318 | spin_lock(&oldf->file_lock); |
319 | old_fdt = files_fdtable(oldf); |
320 | open_files = count_open_files(old_fdt); |
321 | |
322 | /* |
323 | * Check whether we need to allocate a larger fd array and fd set. |
324 | */ |
325 | while (unlikely(open_files > new_fdt->max_fds)) { |
326 | spin_unlock(&oldf->file_lock); |
327 | |
328 | if (new_fdt != &newf->fdtab) { |
329 | free_fdarr(new_fdt); |
330 | free_fdset(new_fdt); |
331 | kfree(new_fdt); |
332 | } |
333 | |
334 | new_fdt = alloc_fdtable(open_files - 1); |
335 | if (!new_fdt) { |
336 | *errorp = -ENOMEM; |
337 | goto out_release; |
338 | } |
339 | |
340 | /* beyond sysctl_nr_open; nothing to do */ |
341 | if (unlikely(new_fdt->max_fds < open_files)) { |
342 | free_fdarr(new_fdt); |
343 | free_fdset(new_fdt); |
344 | kfree(new_fdt); |
345 | *errorp = -EMFILE; |
346 | goto out_release; |
347 | } |
348 | |
349 | /* |
350 | * Reacquire the oldf lock and a pointer to its fd table |
351 | * who knows it may have a new bigger fd table. We need |
352 | * the latest pointer. |
353 | */ |
354 | spin_lock(&oldf->file_lock); |
355 | old_fdt = files_fdtable(oldf); |
356 | open_files = count_open_files(old_fdt); |
357 | } |
358 | |
359 | old_fds = old_fdt->fd; |
360 | new_fds = new_fdt->fd; |
361 | |
362 | memcpy(new_fdt->open_fds->fds_bits, |
363 | old_fdt->open_fds->fds_bits, open_files/8); |
364 | memcpy(new_fdt->close_on_exec->fds_bits, |
365 | old_fdt->close_on_exec->fds_bits, open_files/8); |
366 | |
367 | for (i = open_files; i != 0; i--) { |
368 | struct file *f = *old_fds++; |
369 | if (f) { |
370 | get_file(f); |
371 | } else { |
372 | /* |
373 | * The fd may be claimed in the fd bitmap but not yet |
374 | * instantiated in the files array if a sibling thread |
375 | * is partway through open(). So make sure that this |
376 | * fd is available to the new process. |
377 | */ |
378 | FD_CLR(open_files - i, new_fdt->open_fds); |
379 | } |
380 | rcu_assign_pointer(*new_fds++, f); |
381 | } |
382 | spin_unlock(&oldf->file_lock); |
383 | |
384 | /* compute the remainder to be cleared */ |
385 | size = (new_fdt->max_fds - open_files) * sizeof(struct file *); |
386 | |
387 | /* This is long word aligned thus could use a optimized version */ |
388 | memset(new_fds, 0, size); |
389 | |
390 | if (new_fdt->max_fds > open_files) { |
391 | int left = (new_fdt->max_fds-open_files)/8; |
392 | int start = open_files / (8 * sizeof(unsigned long)); |
393 | |
394 | memset(&new_fdt->open_fds->fds_bits[start], 0, left); |
395 | memset(&new_fdt->close_on_exec->fds_bits[start], 0, left); |
396 | } |
397 | |
398 | rcu_assign_pointer(newf->fdt, new_fdt); |
399 | |
400 | return newf; |
401 | |
402 | out_release: |
403 | kmem_cache_free(files_cachep, newf); |
404 | out: |
405 | return NULL; |
406 | } |
407 | |
408 | static void __devinit fdtable_defer_list_init(int cpu) |
409 | { |
410 | struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu); |
411 | spin_lock_init(&fddef->lock); |
412 | INIT_WORK(&fddef->wq, free_fdtable_work); |
413 | fddef->next = NULL; |
414 | } |
415 | |
416 | void __init files_defer_init(void) |
417 | { |
418 | int i; |
419 | for_each_possible_cpu(i) |
420 | fdtable_defer_list_init(i); |
421 | sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) & |
422 | -BITS_PER_LONG; |
423 | } |
424 | |
425 | struct files_struct init_files = { |
426 | .count = ATOMIC_INIT(1), |
427 | .fdt = &init_files.fdtab, |
428 | .fdtab = { |
429 | .max_fds = NR_OPEN_DEFAULT, |
430 | .fd = &init_files.fd_array[0], |
431 | .close_on_exec = (fd_set *)&init_files.close_on_exec_init, |
432 | .open_fds = (fd_set *)&init_files.open_fds_init, |
433 | .rcu = RCU_HEAD_INIT, |
434 | }, |
435 | .file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock), |
436 | }; |
437 | |
438 | /* |
439 | * allocate a file descriptor, mark it busy. |
440 | */ |
441 | int alloc_fd(unsigned start, unsigned flags) |
442 | { |
443 | struct files_struct *files = current->files; |
444 | unsigned int fd; |
445 | int error; |
446 | struct fdtable *fdt; |
447 | |
448 | spin_lock(&files->file_lock); |
449 | repeat: |
450 | fdt = files_fdtable(files); |
451 | fd = start; |
452 | if (fd < files->next_fd) |
453 | fd = files->next_fd; |
454 | |
455 | if (fd < fdt->max_fds) |
456 | fd = find_next_zero_bit(fdt->open_fds->fds_bits, |
457 | fdt->max_fds, fd); |
458 | |
459 | error = expand_files(files, fd); |
460 | if (error < 0) |
461 | goto out; |
462 | |
463 | /* |
464 | * If we needed to expand the fs array we |
465 | * might have blocked - try again. |
466 | */ |
467 | if (error) |
468 | goto repeat; |
469 | |
470 | if (start <= files->next_fd) |
471 | files->next_fd = fd + 1; |
472 | |
473 | FD_SET(fd, fdt->open_fds); |
474 | if (flags & O_CLOEXEC) |
475 | FD_SET(fd, fdt->close_on_exec); |
476 | else |
477 | FD_CLR(fd, fdt->close_on_exec); |
478 | error = fd; |
479 | #if 1 |
480 | /* Sanity check */ |
481 | if (rcu_dereference(fdt->fd[fd]) != NULL) { |
482 | printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); |
483 | rcu_assign_pointer(fdt->fd[fd], NULL); |
484 | } |
485 | #endif |
486 | |
487 | out: |
488 | spin_unlock(&files->file_lock); |
489 | return error; |
490 | } |
491 | |
492 | int get_unused_fd(void) |
493 | { |
494 | return alloc_fd(0, 0); |
495 | } |
496 | EXPORT_SYMBOL(get_unused_fd); |
497 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9