Contents of /alx-src/tags/kernel26-2.6.12-alx-r9/kernel/cpuset.c
Parent Directory | Revision Log
Revision 630 -
(show annotations)
(download)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 41060 byte(s)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 41060 byte(s)
Tag kernel26-2.6.12-alx-r9
1 | /* |
2 | * kernel/cpuset.c |
3 | * |
4 | * Processor and Memory placement constraints for sets of tasks. |
5 | * |
6 | * Copyright (C) 2003 BULL SA. |
7 | * Copyright (C) 2004 Silicon Graphics, Inc. |
8 | * |
9 | * Portions derived from Patrick Mochel's sysfs code. |
10 | * sysfs is Copyright (c) 2001-3 Patrick Mochel |
11 | * Portions Copyright (c) 2004 Silicon Graphics, Inc. |
12 | * |
13 | * 2003-10-10 Written by Simon Derr <simon.derr@bull.net> |
14 | * 2003-10-22 Updates by Stephen Hemminger. |
15 | * 2004 May-July Rework by Paul Jackson <pj@sgi.com> |
16 | * |
17 | * This file is subject to the terms and conditions of the GNU General Public |
18 | * License. See the file COPYING in the main directory of the Linux |
19 | * distribution for more details. |
20 | */ |
21 | |
22 | #include <linux/config.h> |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpumask.h> |
25 | #include <linux/cpuset.h> |
26 | #include <linux/err.h> |
27 | #include <linux/errno.h> |
28 | #include <linux/file.h> |
29 | #include <linux/fs.h> |
30 | #include <linux/init.h> |
31 | #include <linux/interrupt.h> |
32 | #include <linux/kernel.h> |
33 | #include <linux/kmod.h> |
34 | #include <linux/list.h> |
35 | #include <linux/mm.h> |
36 | #include <linux/module.h> |
37 | #include <linux/mount.h> |
38 | #include <linux/namei.h> |
39 | #include <linux/pagemap.h> |
40 | #include <linux/proc_fs.h> |
41 | #include <linux/sched.h> |
42 | #include <linux/seq_file.h> |
43 | #include <linux/slab.h> |
44 | #include <linux/smp_lock.h> |
45 | #include <linux/spinlock.h> |
46 | #include <linux/stat.h> |
47 | #include <linux/string.h> |
48 | #include <linux/time.h> |
49 | #include <linux/backing-dev.h> |
50 | #include <linux/sort.h> |
51 | |
52 | #include <asm/uaccess.h> |
53 | #include <asm/atomic.h> |
54 | #include <asm/semaphore.h> |
55 | |
56 | #define CPUSET_SUPER_MAGIC 0x27e0eb |
57 | |
58 | struct cpuset { |
59 | unsigned long flags; /* "unsigned long" so bitops work */ |
60 | cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
61 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
62 | |
63 | atomic_t count; /* count tasks using this cpuset */ |
64 | |
65 | /* |
66 | * We link our 'sibling' struct into our parents 'children'. |
67 | * Our children link their 'sibling' into our 'children'. |
68 | */ |
69 | struct list_head sibling; /* my parents children */ |
70 | struct list_head children; /* my children */ |
71 | |
72 | struct cpuset *parent; /* my parent */ |
73 | struct dentry *dentry; /* cpuset fs entry */ |
74 | |
75 | /* |
76 | * Copy of global cpuset_mems_generation as of the most |
77 | * recent time this cpuset changed its mems_allowed. |
78 | */ |
79 | int mems_generation; |
80 | }; |
81 | |
82 | /* bits in struct cpuset flags field */ |
83 | typedef enum { |
84 | CS_CPU_EXCLUSIVE, |
85 | CS_MEM_EXCLUSIVE, |
86 | CS_REMOVED, |
87 | CS_NOTIFY_ON_RELEASE |
88 | } cpuset_flagbits_t; |
89 | |
90 | /* convenient tests for these bits */ |
91 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
92 | { |
93 | return !!test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
94 | } |
95 | |
96 | static inline int is_mem_exclusive(const struct cpuset *cs) |
97 | { |
98 | return !!test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
99 | } |
100 | |
101 | static inline int is_removed(const struct cpuset *cs) |
102 | { |
103 | return !!test_bit(CS_REMOVED, &cs->flags); |
104 | } |
105 | |
106 | static inline int notify_on_release(const struct cpuset *cs) |
107 | { |
108 | return !!test_bit(CS_NOTIFY_ON_RELEASE, &cs->flags); |
109 | } |
110 | |
111 | /* |
112 | * Increment this atomic integer everytime any cpuset changes its |
113 | * mems_allowed value. Users of cpusets can track this generation |
114 | * number, and avoid having to lock and reload mems_allowed unless |
115 | * the cpuset they're using changes generation. |
116 | * |
117 | * A single, global generation is needed because attach_task() could |
118 | * reattach a task to a different cpuset, which must not have its |
119 | * generation numbers aliased with those of that tasks previous cpuset. |
120 | * |
121 | * Generations are needed for mems_allowed because one task cannot |
122 | * modify anothers memory placement. So we must enable every task, |
123 | * on every visit to __alloc_pages(), to efficiently check whether |
124 | * its current->cpuset->mems_allowed has changed, requiring an update |
125 | * of its current->mems_allowed. |
126 | */ |
127 | static atomic_t cpuset_mems_generation = ATOMIC_INIT(1); |
128 | |
129 | static struct cpuset top_cpuset = { |
130 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), |
131 | .cpus_allowed = CPU_MASK_ALL, |
132 | .mems_allowed = NODE_MASK_ALL, |
133 | .count = ATOMIC_INIT(0), |
134 | .sibling = LIST_HEAD_INIT(top_cpuset.sibling), |
135 | .children = LIST_HEAD_INIT(top_cpuset.children), |
136 | .parent = NULL, |
137 | .dentry = NULL, |
138 | .mems_generation = 0, |
139 | }; |
140 | |
141 | static struct vfsmount *cpuset_mount; |
142 | static struct super_block *cpuset_sb = NULL; |
143 | |
144 | /* |
145 | * cpuset_sem should be held by anyone who is depending on the children |
146 | * or sibling lists of any cpuset, or performing non-atomic operations |
147 | * on the flags or *_allowed values of a cpuset, such as raising the |
148 | * CS_REMOVED flag bit iff it is not already raised, or reading and |
149 | * conditionally modifying the *_allowed values. One kernel global |
150 | * cpuset semaphore should be sufficient - these things don't change |
151 | * that much. |
152 | * |
153 | * The code that modifies cpusets holds cpuset_sem across the entire |
154 | * operation, from cpuset_common_file_write() down, single threading |
155 | * all cpuset modifications (except for counter manipulations from |
156 | * fork and exit) across the system. This presumes that cpuset |
157 | * modifications are rare - better kept simple and safe, even if slow. |
158 | * |
159 | * The code that reads cpusets, such as in cpuset_common_file_read() |
160 | * and below, only holds cpuset_sem across small pieces of code, such |
161 | * as when reading out possibly multi-word cpumasks and nodemasks, as |
162 | * the risks are less, and the desire for performance a little greater. |
163 | * The proc_cpuset_show() routine needs to hold cpuset_sem to insure |
164 | * that no cs->dentry is NULL, as it walks up the cpuset tree to root. |
165 | * |
166 | * The hooks from fork and exit, cpuset_fork() and cpuset_exit(), don't |
167 | * (usually) grab cpuset_sem. These are the two most performance |
168 | * critical pieces of code here. The exception occurs on exit(), |
169 | * when a task in a notify_on_release cpuset exits. Then cpuset_sem |
170 | * is taken, and if the cpuset count is zero, a usermode call made |
171 | * to /sbin/cpuset_release_agent with the name of the cpuset (path |
172 | * relative to the root of cpuset file system) as the argument. |
173 | * |
174 | * A cpuset can only be deleted if both its 'count' of using tasks is |
175 | * zero, and its list of 'children' cpusets is empty. Since all tasks |
176 | * in the system use _some_ cpuset, and since there is always at least |
177 | * one task in the system (init, pid == 1), therefore, top_cpuset |
178 | * always has either children cpusets and/or using tasks. So no need |
179 | * for any special hack to ensure that top_cpuset cannot be deleted. |
180 | */ |
181 | |
182 | static DECLARE_MUTEX(cpuset_sem); |
183 | |
184 | /* |
185 | * A couple of forward declarations required, due to cyclic reference loop: |
186 | * cpuset_mkdir -> cpuset_create -> cpuset_populate_dir -> cpuset_add_file |
187 | * -> cpuset_create_file -> cpuset_dir_inode_operations -> cpuset_mkdir. |
188 | */ |
189 | |
190 | static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode); |
191 | static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry); |
192 | |
193 | static struct backing_dev_info cpuset_backing_dev_info = { |
194 | .ra_pages = 0, /* No readahead */ |
195 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, |
196 | }; |
197 | |
198 | static struct inode *cpuset_new_inode(mode_t mode) |
199 | { |
200 | struct inode *inode = new_inode(cpuset_sb); |
201 | |
202 | if (inode) { |
203 | inode->i_mode = mode; |
204 | inode->i_uid = current->fsuid; |
205 | inode->i_gid = current->fsgid; |
206 | inode->i_blksize = PAGE_CACHE_SIZE; |
207 | inode->i_blocks = 0; |
208 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
209 | inode->i_mapping->backing_dev_info = &cpuset_backing_dev_info; |
210 | } |
211 | return inode; |
212 | } |
213 | |
214 | static void cpuset_diput(struct dentry *dentry, struct inode *inode) |
215 | { |
216 | /* is dentry a directory ? if so, kfree() associated cpuset */ |
217 | if (S_ISDIR(inode->i_mode)) { |
218 | struct cpuset *cs = dentry->d_fsdata; |
219 | BUG_ON(!(is_removed(cs))); |
220 | kfree(cs); |
221 | } |
222 | iput(inode); |
223 | } |
224 | |
225 | static struct dentry_operations cpuset_dops = { |
226 | .d_iput = cpuset_diput, |
227 | }; |
228 | |
229 | static struct dentry *cpuset_get_dentry(struct dentry *parent, const char *name) |
230 | { |
231 | struct qstr qstr; |
232 | struct dentry *d; |
233 | |
234 | qstr.name = name; |
235 | qstr.len = strlen(name); |
236 | qstr.hash = full_name_hash(name, qstr.len); |
237 | d = lookup_hash(&qstr, parent); |
238 | if (!IS_ERR(d)) |
239 | d->d_op = &cpuset_dops; |
240 | return d; |
241 | } |
242 | |
243 | static void remove_dir(struct dentry *d) |
244 | { |
245 | struct dentry *parent = dget(d->d_parent); |
246 | |
247 | d_delete(d); |
248 | simple_rmdir(parent->d_inode, d); |
249 | dput(parent); |
250 | } |
251 | |
252 | /* |
253 | * NOTE : the dentry must have been dget()'ed |
254 | */ |
255 | static void cpuset_d_remove_dir(struct dentry *dentry) |
256 | { |
257 | struct list_head *node; |
258 | |
259 | spin_lock(&dcache_lock); |
260 | node = dentry->d_subdirs.next; |
261 | while (node != &dentry->d_subdirs) { |
262 | struct dentry *d = list_entry(node, struct dentry, d_child); |
263 | list_del_init(node); |
264 | if (d->d_inode) { |
265 | d = dget_locked(d); |
266 | spin_unlock(&dcache_lock); |
267 | d_delete(d); |
268 | simple_unlink(dentry->d_inode, d); |
269 | dput(d); |
270 | spin_lock(&dcache_lock); |
271 | } |
272 | node = dentry->d_subdirs.next; |
273 | } |
274 | list_del_init(&dentry->d_child); |
275 | spin_unlock(&dcache_lock); |
276 | remove_dir(dentry); |
277 | } |
278 | |
279 | static struct super_operations cpuset_ops = { |
280 | .statfs = simple_statfs, |
281 | .drop_inode = generic_delete_inode, |
282 | }; |
283 | |
284 | static int cpuset_fill_super(struct super_block *sb, void *unused_data, |
285 | int unused_silent) |
286 | { |
287 | struct inode *inode; |
288 | struct dentry *root; |
289 | |
290 | sb->s_blocksize = PAGE_CACHE_SIZE; |
291 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; |
292 | sb->s_magic = CPUSET_SUPER_MAGIC; |
293 | sb->s_op = &cpuset_ops; |
294 | cpuset_sb = sb; |
295 | |
296 | inode = cpuset_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR); |
297 | if (inode) { |
298 | inode->i_op = &simple_dir_inode_operations; |
299 | inode->i_fop = &simple_dir_operations; |
300 | /* directories start off with i_nlink == 2 (for "." entry) */ |
301 | inode->i_nlink++; |
302 | } else { |
303 | return -ENOMEM; |
304 | } |
305 | |
306 | root = d_alloc_root(inode); |
307 | if (!root) { |
308 | iput(inode); |
309 | return -ENOMEM; |
310 | } |
311 | sb->s_root = root; |
312 | return 0; |
313 | } |
314 | |
315 | static struct super_block *cpuset_get_sb(struct file_system_type *fs_type, |
316 | int flags, const char *unused_dev_name, |
317 | void *data) |
318 | { |
319 | return get_sb_single(fs_type, flags, data, cpuset_fill_super); |
320 | } |
321 | |
322 | static struct file_system_type cpuset_fs_type = { |
323 | .name = "cpuset", |
324 | .get_sb = cpuset_get_sb, |
325 | .kill_sb = kill_litter_super, |
326 | }; |
327 | |
328 | /* struct cftype: |
329 | * |
330 | * The files in the cpuset filesystem mostly have a very simple read/write |
331 | * handling, some common function will take care of it. Nevertheless some cases |
332 | * (read tasks) are special and therefore I define this structure for every |
333 | * kind of file. |
334 | * |
335 | * |
336 | * When reading/writing to a file: |
337 | * - the cpuset to use in file->f_dentry->d_parent->d_fsdata |
338 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
339 | */ |
340 | |
341 | struct cftype { |
342 | char *name; |
343 | int private; |
344 | int (*open) (struct inode *inode, struct file *file); |
345 | ssize_t (*read) (struct file *file, char __user *buf, size_t nbytes, |
346 | loff_t *ppos); |
347 | int (*write) (struct file *file, const char __user *buf, size_t nbytes, |
348 | loff_t *ppos); |
349 | int (*release) (struct inode *inode, struct file *file); |
350 | }; |
351 | |
352 | static inline struct cpuset *__d_cs(struct dentry *dentry) |
353 | { |
354 | return dentry->d_fsdata; |
355 | } |
356 | |
357 | static inline struct cftype *__d_cft(struct dentry *dentry) |
358 | { |
359 | return dentry->d_fsdata; |
360 | } |
361 | |
362 | /* |
363 | * Call with cpuset_sem held. Writes path of cpuset into buf. |
364 | * Returns 0 on success, -errno on error. |
365 | */ |
366 | |
367 | static int cpuset_path(const struct cpuset *cs, char *buf, int buflen) |
368 | { |
369 | char *start; |
370 | |
371 | start = buf + buflen; |
372 | |
373 | *--start = '\0'; |
374 | for (;;) { |
375 | int len = cs->dentry->d_name.len; |
376 | if ((start -= len) < buf) |
377 | return -ENAMETOOLONG; |
378 | memcpy(start, cs->dentry->d_name.name, len); |
379 | cs = cs->parent; |
380 | if (!cs) |
381 | break; |
382 | if (!cs->parent) |
383 | continue; |
384 | if (--start < buf) |
385 | return -ENAMETOOLONG; |
386 | *start = '/'; |
387 | } |
388 | memmove(buf, start, buf + buflen - start); |
389 | return 0; |
390 | } |
391 | |
392 | /* |
393 | * Notify userspace when a cpuset is released, by running |
394 | * /sbin/cpuset_release_agent with the name of the cpuset (path |
395 | * relative to the root of cpuset file system) as the argument. |
396 | * |
397 | * Most likely, this user command will try to rmdir this cpuset. |
398 | * |
399 | * This races with the possibility that some other task will be |
400 | * attached to this cpuset before it is removed, or that some other |
401 | * user task will 'mkdir' a child cpuset of this cpuset. That's ok. |
402 | * The presumed 'rmdir' will fail quietly if this cpuset is no longer |
403 | * unused, and this cpuset will be reprieved from its death sentence, |
404 | * to continue to serve a useful existence. Next time it's released, |
405 | * we will get notified again, if it still has 'notify_on_release' set. |
406 | * |
407 | * Note final arg to call_usermodehelper() is 0 - that means |
408 | * don't wait. Since we are holding the global cpuset_sem here, |
409 | * and we are asking another thread (started from keventd) to rmdir a |
410 | * cpuset, we can't wait - or we'd deadlock with the removing thread |
411 | * on cpuset_sem. |
412 | */ |
413 | |
414 | static int cpuset_release_agent(char *cpuset_str) |
415 | { |
416 | char *argv[3], *envp[3]; |
417 | int i; |
418 | |
419 | i = 0; |
420 | argv[i++] = "/sbin/cpuset_release_agent"; |
421 | argv[i++] = cpuset_str; |
422 | argv[i] = NULL; |
423 | |
424 | i = 0; |
425 | /* minimal command environment */ |
426 | envp[i++] = "HOME=/"; |
427 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; |
428 | envp[i] = NULL; |
429 | |
430 | return call_usermodehelper(argv[0], argv, envp, 0); |
431 | } |
432 | |
433 | /* |
434 | * Either cs->count of using tasks transitioned to zero, or the |
435 | * cs->children list of child cpusets just became empty. If this |
436 | * cs is notify_on_release() and now both the user count is zero and |
437 | * the list of children is empty, send notice to user land. |
438 | */ |
439 | |
440 | static void check_for_release(struct cpuset *cs) |
441 | { |
442 | if (notify_on_release(cs) && atomic_read(&cs->count) == 0 && |
443 | list_empty(&cs->children)) { |
444 | char *buf; |
445 | |
446 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
447 | if (!buf) |
448 | return; |
449 | if (cpuset_path(cs, buf, PAGE_SIZE) < 0) |
450 | goto out; |
451 | cpuset_release_agent(buf); |
452 | out: |
453 | kfree(buf); |
454 | } |
455 | } |
456 | |
457 | /* |
458 | * Return in *pmask the portion of a cpusets's cpus_allowed that |
459 | * are online. If none are online, walk up the cpuset hierarchy |
460 | * until we find one that does have some online cpus. If we get |
461 | * all the way to the top and still haven't found any online cpus, |
462 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing |
463 | * task, return cpu_online_map. |
464 | * |
465 | * One way or another, we guarantee to return some non-empty subset |
466 | * of cpu_online_map. |
467 | * |
468 | * Call with cpuset_sem held. |
469 | */ |
470 | |
471 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) |
472 | { |
473 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) |
474 | cs = cs->parent; |
475 | if (cs) |
476 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); |
477 | else |
478 | *pmask = cpu_online_map; |
479 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); |
480 | } |
481 | |
482 | /* |
483 | * Return in *pmask the portion of a cpusets's mems_allowed that |
484 | * are online. If none are online, walk up the cpuset hierarchy |
485 | * until we find one that does have some online mems. If we get |
486 | * all the way to the top and still haven't found any online mems, |
487 | * return node_online_map. |
488 | * |
489 | * One way or another, we guarantee to return some non-empty subset |
490 | * of node_online_map. |
491 | * |
492 | * Call with cpuset_sem held. |
493 | */ |
494 | |
495 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
496 | { |
497 | while (cs && !nodes_intersects(cs->mems_allowed, node_online_map)) |
498 | cs = cs->parent; |
499 | if (cs) |
500 | nodes_and(*pmask, cs->mems_allowed, node_online_map); |
501 | else |
502 | *pmask = node_online_map; |
503 | BUG_ON(!nodes_intersects(*pmask, node_online_map)); |
504 | } |
505 | |
506 | /* |
507 | * Refresh current tasks mems_allowed and mems_generation from |
508 | * current tasks cpuset. Call with cpuset_sem held. |
509 | * |
510 | * Be sure to call refresh_mems() on any cpuset operation which |
511 | * (1) holds cpuset_sem, and (2) might possibly alloc memory. |
512 | * Call after obtaining cpuset_sem lock, before any possible |
513 | * allocation. Otherwise one risks trying to allocate memory |
514 | * while the task cpuset_mems_generation is not the same as |
515 | * the mems_generation in its cpuset, which would deadlock on |
516 | * cpuset_sem in cpuset_update_current_mems_allowed(). |
517 | * |
518 | * Since we hold cpuset_sem, once refresh_mems() is called, the |
519 | * test (current->cpuset_mems_generation != cs->mems_generation) |
520 | * in cpuset_update_current_mems_allowed() will remain false, |
521 | * until we drop cpuset_sem. Anyone else who would change our |
522 | * cpusets mems_generation needs to lock cpuset_sem first. |
523 | */ |
524 | |
525 | static void refresh_mems(void) |
526 | { |
527 | struct cpuset *cs = current->cpuset; |
528 | |
529 | if (current->cpuset_mems_generation != cs->mems_generation) { |
530 | guarantee_online_mems(cs, ¤t->mems_allowed); |
531 | current->cpuset_mems_generation = cs->mems_generation; |
532 | } |
533 | } |
534 | |
535 | /* |
536 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? |
537 | * |
538 | * One cpuset is a subset of another if all its allowed CPUs and |
539 | * Memory Nodes are a subset of the other, and its exclusive flags |
540 | * are only set if the other's are set. |
541 | */ |
542 | |
543 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) |
544 | { |
545 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && |
546 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
547 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && |
548 | is_mem_exclusive(p) <= is_mem_exclusive(q); |
549 | } |
550 | |
551 | /* |
552 | * validate_change() - Used to validate that any proposed cpuset change |
553 | * follows the structural rules for cpusets. |
554 | * |
555 | * If we replaced the flag and mask values of the current cpuset |
556 | * (cur) with those values in the trial cpuset (trial), would |
557 | * our various subset and exclusive rules still be valid? Presumes |
558 | * cpuset_sem held. |
559 | * |
560 | * 'cur' is the address of an actual, in-use cpuset. Operations |
561 | * such as list traversal that depend on the actual address of the |
562 | * cpuset in the list must use cur below, not trial. |
563 | * |
564 | * 'trial' is the address of bulk structure copy of cur, with |
565 | * perhaps one or more of the fields cpus_allowed, mems_allowed, |
566 | * or flags changed to new, trial values. |
567 | * |
568 | * Return 0 if valid, -errno if not. |
569 | */ |
570 | |
571 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) |
572 | { |
573 | struct cpuset *c, *par; |
574 | |
575 | /* Each of our child cpusets must be a subset of us */ |
576 | list_for_each_entry(c, &cur->children, sibling) { |
577 | if (!is_cpuset_subset(c, trial)) |
578 | return -EBUSY; |
579 | } |
580 | |
581 | /* Remaining checks don't apply to root cpuset */ |
582 | if ((par = cur->parent) == NULL) |
583 | return 0; |
584 | |
585 | /* We must be a subset of our parent cpuset */ |
586 | if (!is_cpuset_subset(trial, par)) |
587 | return -EACCES; |
588 | |
589 | /* If either I or some sibling (!= me) is exclusive, we can't overlap */ |
590 | list_for_each_entry(c, &par->children, sibling) { |
591 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
592 | c != cur && |
593 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) |
594 | return -EINVAL; |
595 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
596 | c != cur && |
597 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) |
598 | return -EINVAL; |
599 | } |
600 | |
601 | return 0; |
602 | } |
603 | |
604 | static int update_cpumask(struct cpuset *cs, char *buf) |
605 | { |
606 | struct cpuset trialcs; |
607 | int retval; |
608 | |
609 | trialcs = *cs; |
610 | retval = cpulist_parse(buf, trialcs.cpus_allowed); |
611 | if (retval < 0) |
612 | return retval; |
613 | cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map); |
614 | if (cpus_empty(trialcs.cpus_allowed)) |
615 | return -ENOSPC; |
616 | retval = validate_change(cs, &trialcs); |
617 | if (retval == 0) |
618 | cs->cpus_allowed = trialcs.cpus_allowed; |
619 | return retval; |
620 | } |
621 | |
622 | static int update_nodemask(struct cpuset *cs, char *buf) |
623 | { |
624 | struct cpuset trialcs; |
625 | int retval; |
626 | |
627 | trialcs = *cs; |
628 | retval = nodelist_parse(buf, trialcs.mems_allowed); |
629 | if (retval < 0) |
630 | return retval; |
631 | nodes_and(trialcs.mems_allowed, trialcs.mems_allowed, node_online_map); |
632 | if (nodes_empty(trialcs.mems_allowed)) |
633 | return -ENOSPC; |
634 | retval = validate_change(cs, &trialcs); |
635 | if (retval == 0) { |
636 | cs->mems_allowed = trialcs.mems_allowed; |
637 | atomic_inc(&cpuset_mems_generation); |
638 | cs->mems_generation = atomic_read(&cpuset_mems_generation); |
639 | } |
640 | return retval; |
641 | } |
642 | |
643 | /* |
644 | * update_flag - read a 0 or a 1 in a file and update associated flag |
645 | * bit: the bit to update (CS_CPU_EXCLUSIVE, CS_MEM_EXCLUSIVE, |
646 | * CS_NOTIFY_ON_RELEASE) |
647 | * cs: the cpuset to update |
648 | * buf: the buffer where we read the 0 or 1 |
649 | */ |
650 | |
651 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, char *buf) |
652 | { |
653 | int turning_on; |
654 | struct cpuset trialcs; |
655 | int err; |
656 | |
657 | turning_on = (simple_strtoul(buf, NULL, 10) != 0); |
658 | |
659 | trialcs = *cs; |
660 | if (turning_on) |
661 | set_bit(bit, &trialcs.flags); |
662 | else |
663 | clear_bit(bit, &trialcs.flags); |
664 | |
665 | err = validate_change(cs, &trialcs); |
666 | if (err == 0) { |
667 | if (turning_on) |
668 | set_bit(bit, &cs->flags); |
669 | else |
670 | clear_bit(bit, &cs->flags); |
671 | } |
672 | return err; |
673 | } |
674 | |
675 | static int attach_task(struct cpuset *cs, char *buf) |
676 | { |
677 | pid_t pid; |
678 | struct task_struct *tsk; |
679 | struct cpuset *oldcs; |
680 | cpumask_t cpus; |
681 | |
682 | if (sscanf(buf, "%d", &pid) != 1) |
683 | return -EIO; |
684 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
685 | return -ENOSPC; |
686 | |
687 | if (pid) { |
688 | read_lock(&tasklist_lock); |
689 | |
690 | tsk = find_task_by_pid(pid); |
691 | if (!tsk) { |
692 | read_unlock(&tasklist_lock); |
693 | return -ESRCH; |
694 | } |
695 | |
696 | get_task_struct(tsk); |
697 | read_unlock(&tasklist_lock); |
698 | |
699 | if ((current->euid) && (current->euid != tsk->uid) |
700 | && (current->euid != tsk->suid)) { |
701 | put_task_struct(tsk); |
702 | return -EACCES; |
703 | } |
704 | } else { |
705 | tsk = current; |
706 | get_task_struct(tsk); |
707 | } |
708 | |
709 | task_lock(tsk); |
710 | oldcs = tsk->cpuset; |
711 | if (!oldcs) { |
712 | task_unlock(tsk); |
713 | put_task_struct(tsk); |
714 | return -ESRCH; |
715 | } |
716 | atomic_inc(&cs->count); |
717 | tsk->cpuset = cs; |
718 | task_unlock(tsk); |
719 | |
720 | guarantee_online_cpus(cs, &cpus); |
721 | set_cpus_allowed(tsk, cpus); |
722 | |
723 | put_task_struct(tsk); |
724 | if (atomic_dec_and_test(&oldcs->count)) |
725 | check_for_release(oldcs); |
726 | return 0; |
727 | } |
728 | |
729 | /* The various types of files and directories in a cpuset file system */ |
730 | |
731 | typedef enum { |
732 | FILE_ROOT, |
733 | FILE_DIR, |
734 | FILE_CPULIST, |
735 | FILE_MEMLIST, |
736 | FILE_CPU_EXCLUSIVE, |
737 | FILE_MEM_EXCLUSIVE, |
738 | FILE_NOTIFY_ON_RELEASE, |
739 | FILE_TASKLIST, |
740 | } cpuset_filetype_t; |
741 | |
742 | static ssize_t cpuset_common_file_write(struct file *file, const char __user *userbuf, |
743 | size_t nbytes, loff_t *unused_ppos) |
744 | { |
745 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
746 | struct cftype *cft = __d_cft(file->f_dentry); |
747 | cpuset_filetype_t type = cft->private; |
748 | char *buffer; |
749 | int retval = 0; |
750 | |
751 | /* Crude upper limit on largest legitimate cpulist user might write. */ |
752 | if (nbytes > 100 + 6 * NR_CPUS) |
753 | return -E2BIG; |
754 | |
755 | /* +1 for nul-terminator */ |
756 | if ((buffer = kmalloc(nbytes + 1, GFP_KERNEL)) == 0) |
757 | return -ENOMEM; |
758 | |
759 | if (copy_from_user(buffer, userbuf, nbytes)) { |
760 | retval = -EFAULT; |
761 | goto out1; |
762 | } |
763 | buffer[nbytes] = 0; /* nul-terminate */ |
764 | |
765 | down(&cpuset_sem); |
766 | |
767 | if (is_removed(cs)) { |
768 | retval = -ENODEV; |
769 | goto out2; |
770 | } |
771 | |
772 | switch (type) { |
773 | case FILE_CPULIST: |
774 | retval = update_cpumask(cs, buffer); |
775 | break; |
776 | case FILE_MEMLIST: |
777 | retval = update_nodemask(cs, buffer); |
778 | break; |
779 | case FILE_CPU_EXCLUSIVE: |
780 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, buffer); |
781 | break; |
782 | case FILE_MEM_EXCLUSIVE: |
783 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, buffer); |
784 | break; |
785 | case FILE_NOTIFY_ON_RELEASE: |
786 | retval = update_flag(CS_NOTIFY_ON_RELEASE, cs, buffer); |
787 | break; |
788 | case FILE_TASKLIST: |
789 | retval = attach_task(cs, buffer); |
790 | break; |
791 | default: |
792 | retval = -EINVAL; |
793 | goto out2; |
794 | } |
795 | |
796 | if (retval == 0) |
797 | retval = nbytes; |
798 | out2: |
799 | up(&cpuset_sem); |
800 | out1: |
801 | kfree(buffer); |
802 | return retval; |
803 | } |
804 | |
805 | static ssize_t cpuset_file_write(struct file *file, const char __user *buf, |
806 | size_t nbytes, loff_t *ppos) |
807 | { |
808 | ssize_t retval = 0; |
809 | struct cftype *cft = __d_cft(file->f_dentry); |
810 | if (!cft) |
811 | return -ENODEV; |
812 | |
813 | /* special function ? */ |
814 | if (cft->write) |
815 | retval = cft->write(file, buf, nbytes, ppos); |
816 | else |
817 | retval = cpuset_common_file_write(file, buf, nbytes, ppos); |
818 | |
819 | return retval; |
820 | } |
821 | |
822 | /* |
823 | * These ascii lists should be read in a single call, by using a user |
824 | * buffer large enough to hold the entire map. If read in smaller |
825 | * chunks, there is no guarantee of atomicity. Since the display format |
826 | * used, list of ranges of sequential numbers, is variable length, |
827 | * and since these maps can change value dynamically, one could read |
828 | * gibberish by doing partial reads while a list was changing. |
829 | * A single large read to a buffer that crosses a page boundary is |
830 | * ok, because the result being copied to user land is not recomputed |
831 | * across a page fault. |
832 | */ |
833 | |
834 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
835 | { |
836 | cpumask_t mask; |
837 | |
838 | down(&cpuset_sem); |
839 | mask = cs->cpus_allowed; |
840 | up(&cpuset_sem); |
841 | |
842 | return cpulist_scnprintf(page, PAGE_SIZE, mask); |
843 | } |
844 | |
845 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
846 | { |
847 | nodemask_t mask; |
848 | |
849 | down(&cpuset_sem); |
850 | mask = cs->mems_allowed; |
851 | up(&cpuset_sem); |
852 | |
853 | return nodelist_scnprintf(page, PAGE_SIZE, mask); |
854 | } |
855 | |
856 | static ssize_t cpuset_common_file_read(struct file *file, char __user *buf, |
857 | size_t nbytes, loff_t *ppos) |
858 | { |
859 | struct cftype *cft = __d_cft(file->f_dentry); |
860 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
861 | cpuset_filetype_t type = cft->private; |
862 | char *page; |
863 | ssize_t retval = 0; |
864 | char *s; |
865 | char *start; |
866 | size_t n; |
867 | |
868 | if (!(page = (char *)__get_free_page(GFP_KERNEL))) |
869 | return -ENOMEM; |
870 | |
871 | s = page; |
872 | |
873 | switch (type) { |
874 | case FILE_CPULIST: |
875 | s += cpuset_sprintf_cpulist(s, cs); |
876 | break; |
877 | case FILE_MEMLIST: |
878 | s += cpuset_sprintf_memlist(s, cs); |
879 | break; |
880 | case FILE_CPU_EXCLUSIVE: |
881 | *s++ = is_cpu_exclusive(cs) ? '1' : '0'; |
882 | break; |
883 | case FILE_MEM_EXCLUSIVE: |
884 | *s++ = is_mem_exclusive(cs) ? '1' : '0'; |
885 | break; |
886 | case FILE_NOTIFY_ON_RELEASE: |
887 | *s++ = notify_on_release(cs) ? '1' : '0'; |
888 | break; |
889 | default: |
890 | retval = -EINVAL; |
891 | goto out; |
892 | } |
893 | *s++ = '\n'; |
894 | *s = '\0'; |
895 | |
896 | start = page + *ppos; |
897 | n = s - start; |
898 | retval = n - copy_to_user(buf, start, min(n, nbytes)); |
899 | *ppos += retval; |
900 | out: |
901 | free_page((unsigned long)page); |
902 | return retval; |
903 | } |
904 | |
905 | static ssize_t cpuset_file_read(struct file *file, char __user *buf, size_t nbytes, |
906 | loff_t *ppos) |
907 | { |
908 | ssize_t retval = 0; |
909 | struct cftype *cft = __d_cft(file->f_dentry); |
910 | if (!cft) |
911 | return -ENODEV; |
912 | |
913 | /* special function ? */ |
914 | if (cft->read) |
915 | retval = cft->read(file, buf, nbytes, ppos); |
916 | else |
917 | retval = cpuset_common_file_read(file, buf, nbytes, ppos); |
918 | |
919 | return retval; |
920 | } |
921 | |
922 | static int cpuset_file_open(struct inode *inode, struct file *file) |
923 | { |
924 | int err; |
925 | struct cftype *cft; |
926 | |
927 | err = generic_file_open(inode, file); |
928 | if (err) |
929 | return err; |
930 | |
931 | cft = __d_cft(file->f_dentry); |
932 | if (!cft) |
933 | return -ENODEV; |
934 | if (cft->open) |
935 | err = cft->open(inode, file); |
936 | else |
937 | err = 0; |
938 | |
939 | return err; |
940 | } |
941 | |
942 | static int cpuset_file_release(struct inode *inode, struct file *file) |
943 | { |
944 | struct cftype *cft = __d_cft(file->f_dentry); |
945 | if (cft->release) |
946 | return cft->release(inode, file); |
947 | return 0; |
948 | } |
949 | |
950 | static struct file_operations cpuset_file_operations = { |
951 | .read = cpuset_file_read, |
952 | .write = cpuset_file_write, |
953 | .llseek = generic_file_llseek, |
954 | .open = cpuset_file_open, |
955 | .release = cpuset_file_release, |
956 | }; |
957 | |
958 | static struct inode_operations cpuset_dir_inode_operations = { |
959 | .lookup = simple_lookup, |
960 | .mkdir = cpuset_mkdir, |
961 | .rmdir = cpuset_rmdir, |
962 | }; |
963 | |
964 | static int cpuset_create_file(struct dentry *dentry, int mode) |
965 | { |
966 | struct inode *inode; |
967 | |
968 | if (!dentry) |
969 | return -ENOENT; |
970 | if (dentry->d_inode) |
971 | return -EEXIST; |
972 | |
973 | inode = cpuset_new_inode(mode); |
974 | if (!inode) |
975 | return -ENOMEM; |
976 | |
977 | if (S_ISDIR(mode)) { |
978 | inode->i_op = &cpuset_dir_inode_operations; |
979 | inode->i_fop = &simple_dir_operations; |
980 | |
981 | /* start off with i_nlink == 2 (for "." entry) */ |
982 | inode->i_nlink++; |
983 | } else if (S_ISREG(mode)) { |
984 | inode->i_size = 0; |
985 | inode->i_fop = &cpuset_file_operations; |
986 | } |
987 | |
988 | d_instantiate(dentry, inode); |
989 | dget(dentry); /* Extra count - pin the dentry in core */ |
990 | return 0; |
991 | } |
992 | |
993 | /* |
994 | * cpuset_create_dir - create a directory for an object. |
995 | * cs: the cpuset we create the directory for. |
996 | * It must have a valid ->parent field |
997 | * And we are going to fill its ->dentry field. |
998 | * name: The name to give to the cpuset directory. Will be copied. |
999 | * mode: mode to set on new directory. |
1000 | */ |
1001 | |
1002 | static int cpuset_create_dir(struct cpuset *cs, const char *name, int mode) |
1003 | { |
1004 | struct dentry *dentry = NULL; |
1005 | struct dentry *parent; |
1006 | int error = 0; |
1007 | |
1008 | parent = cs->parent->dentry; |
1009 | dentry = cpuset_get_dentry(parent, name); |
1010 | if (IS_ERR(dentry)) |
1011 | return PTR_ERR(dentry); |
1012 | error = cpuset_create_file(dentry, S_IFDIR | mode); |
1013 | if (!error) { |
1014 | dentry->d_fsdata = cs; |
1015 | parent->d_inode->i_nlink++; |
1016 | cs->dentry = dentry; |
1017 | } |
1018 | dput(dentry); |
1019 | |
1020 | return error; |
1021 | } |
1022 | |
1023 | static int cpuset_add_file(struct dentry *dir, const struct cftype *cft) |
1024 | { |
1025 | struct dentry *dentry; |
1026 | int error; |
1027 | |
1028 | down(&dir->d_inode->i_sem); |
1029 | dentry = cpuset_get_dentry(dir, cft->name); |
1030 | if (!IS_ERR(dentry)) { |
1031 | error = cpuset_create_file(dentry, 0644 | S_IFREG); |
1032 | if (!error) |
1033 | dentry->d_fsdata = (void *)cft; |
1034 | dput(dentry); |
1035 | } else |
1036 | error = PTR_ERR(dentry); |
1037 | up(&dir->d_inode->i_sem); |
1038 | return error; |
1039 | } |
1040 | |
1041 | /* |
1042 | * Stuff for reading the 'tasks' file. |
1043 | * |
1044 | * Reading this file can return large amounts of data if a cpuset has |
1045 | * *lots* of attached tasks. So it may need several calls to read(), |
1046 | * but we cannot guarantee that the information we produce is correct |
1047 | * unless we produce it entirely atomically. |
1048 | * |
1049 | * Upon tasks file open(), a struct ctr_struct is allocated, that |
1050 | * will have a pointer to an array (also allocated here). The struct |
1051 | * ctr_struct * is stored in file->private_data. Its resources will |
1052 | * be freed by release() when the file is closed. The array is used |
1053 | * to sprintf the PIDs and then used by read(). |
1054 | */ |
1055 | |
1056 | /* cpusets_tasks_read array */ |
1057 | |
1058 | struct ctr_struct { |
1059 | char *buf; |
1060 | int bufsz; |
1061 | }; |
1062 | |
1063 | /* |
1064 | * Load into 'pidarray' up to 'npids' of the tasks using cpuset 'cs'. |
1065 | * Return actual number of pids loaded. |
1066 | */ |
1067 | static inline int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs) |
1068 | { |
1069 | int n = 0; |
1070 | struct task_struct *g, *p; |
1071 | |
1072 | read_lock(&tasklist_lock); |
1073 | |
1074 | do_each_thread(g, p) { |
1075 | if (p->cpuset == cs) { |
1076 | pidarray[n++] = p->pid; |
1077 | if (unlikely(n == npids)) |
1078 | goto array_full; |
1079 | } |
1080 | } while_each_thread(g, p); |
1081 | |
1082 | array_full: |
1083 | read_unlock(&tasklist_lock); |
1084 | return n; |
1085 | } |
1086 | |
1087 | static int cmppid(const void *a, const void *b) |
1088 | { |
1089 | return *(pid_t *)a - *(pid_t *)b; |
1090 | } |
1091 | |
1092 | /* |
1093 | * Convert array 'a' of 'npids' pid_t's to a string of newline separated |
1094 | * decimal pids in 'buf'. Don't write more than 'sz' chars, but return |
1095 | * count 'cnt' of how many chars would be written if buf were large enough. |
1096 | */ |
1097 | static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids) |
1098 | { |
1099 | int cnt = 0; |
1100 | int i; |
1101 | |
1102 | for (i = 0; i < npids; i++) |
1103 | cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]); |
1104 | return cnt; |
1105 | } |
1106 | |
1107 | static int cpuset_tasks_open(struct inode *unused, struct file *file) |
1108 | { |
1109 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
1110 | struct ctr_struct *ctr; |
1111 | pid_t *pidarray; |
1112 | int npids; |
1113 | char c; |
1114 | |
1115 | if (!(file->f_mode & FMODE_READ)) |
1116 | return 0; |
1117 | |
1118 | ctr = kmalloc(sizeof(*ctr), GFP_KERNEL); |
1119 | if (!ctr) |
1120 | goto err0; |
1121 | |
1122 | /* |
1123 | * If cpuset gets more users after we read count, we won't have |
1124 | * enough space - tough. This race is indistinguishable to the |
1125 | * caller from the case that the additional cpuset users didn't |
1126 | * show up until sometime later on. |
1127 | */ |
1128 | npids = atomic_read(&cs->count); |
1129 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); |
1130 | if (!pidarray) |
1131 | goto err1; |
1132 | |
1133 | npids = pid_array_load(pidarray, npids, cs); |
1134 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); |
1135 | |
1136 | /* Call pid_array_to_buf() twice, first just to get bufsz */ |
1137 | ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1; |
1138 | ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL); |
1139 | if (!ctr->buf) |
1140 | goto err2; |
1141 | ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids); |
1142 | |
1143 | kfree(pidarray); |
1144 | file->private_data = ctr; |
1145 | return 0; |
1146 | |
1147 | err2: |
1148 | kfree(pidarray); |
1149 | err1: |
1150 | kfree(ctr); |
1151 | err0: |
1152 | return -ENOMEM; |
1153 | } |
1154 | |
1155 | static ssize_t cpuset_tasks_read(struct file *file, char __user *buf, |
1156 | size_t nbytes, loff_t *ppos) |
1157 | { |
1158 | struct ctr_struct *ctr = file->private_data; |
1159 | |
1160 | if (*ppos + nbytes > ctr->bufsz) |
1161 | nbytes = ctr->bufsz - *ppos; |
1162 | if (copy_to_user(buf, ctr->buf + *ppos, nbytes)) |
1163 | return -EFAULT; |
1164 | *ppos += nbytes; |
1165 | return nbytes; |
1166 | } |
1167 | |
1168 | static int cpuset_tasks_release(struct inode *unused_inode, struct file *file) |
1169 | { |
1170 | struct ctr_struct *ctr; |
1171 | |
1172 | if (file->f_mode & FMODE_READ) { |
1173 | ctr = file->private_data; |
1174 | kfree(ctr->buf); |
1175 | kfree(ctr); |
1176 | } |
1177 | return 0; |
1178 | } |
1179 | |
1180 | /* |
1181 | * for the common functions, 'private' gives the type of file |
1182 | */ |
1183 | |
1184 | static struct cftype cft_tasks = { |
1185 | .name = "tasks", |
1186 | .open = cpuset_tasks_open, |
1187 | .read = cpuset_tasks_read, |
1188 | .release = cpuset_tasks_release, |
1189 | .private = FILE_TASKLIST, |
1190 | }; |
1191 | |
1192 | static struct cftype cft_cpus = { |
1193 | .name = "cpus", |
1194 | .private = FILE_CPULIST, |
1195 | }; |
1196 | |
1197 | static struct cftype cft_mems = { |
1198 | .name = "mems", |
1199 | .private = FILE_MEMLIST, |
1200 | }; |
1201 | |
1202 | static struct cftype cft_cpu_exclusive = { |
1203 | .name = "cpu_exclusive", |
1204 | .private = FILE_CPU_EXCLUSIVE, |
1205 | }; |
1206 | |
1207 | static struct cftype cft_mem_exclusive = { |
1208 | .name = "mem_exclusive", |
1209 | .private = FILE_MEM_EXCLUSIVE, |
1210 | }; |
1211 | |
1212 | static struct cftype cft_notify_on_release = { |
1213 | .name = "notify_on_release", |
1214 | .private = FILE_NOTIFY_ON_RELEASE, |
1215 | }; |
1216 | |
1217 | static int cpuset_populate_dir(struct dentry *cs_dentry) |
1218 | { |
1219 | int err; |
1220 | |
1221 | if ((err = cpuset_add_file(cs_dentry, &cft_cpus)) < 0) |
1222 | return err; |
1223 | if ((err = cpuset_add_file(cs_dentry, &cft_mems)) < 0) |
1224 | return err; |
1225 | if ((err = cpuset_add_file(cs_dentry, &cft_cpu_exclusive)) < 0) |
1226 | return err; |
1227 | if ((err = cpuset_add_file(cs_dentry, &cft_mem_exclusive)) < 0) |
1228 | return err; |
1229 | if ((err = cpuset_add_file(cs_dentry, &cft_notify_on_release)) < 0) |
1230 | return err; |
1231 | if ((err = cpuset_add_file(cs_dentry, &cft_tasks)) < 0) |
1232 | return err; |
1233 | return 0; |
1234 | } |
1235 | |
1236 | /* |
1237 | * cpuset_create - create a cpuset |
1238 | * parent: cpuset that will be parent of the new cpuset. |
1239 | * name: name of the new cpuset. Will be strcpy'ed. |
1240 | * mode: mode to set on new inode |
1241 | * |
1242 | * Must be called with the semaphore on the parent inode held |
1243 | */ |
1244 | |
1245 | static long cpuset_create(struct cpuset *parent, const char *name, int mode) |
1246 | { |
1247 | struct cpuset *cs; |
1248 | int err; |
1249 | |
1250 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1251 | if (!cs) |
1252 | return -ENOMEM; |
1253 | |
1254 | down(&cpuset_sem); |
1255 | refresh_mems(); |
1256 | cs->flags = 0; |
1257 | if (notify_on_release(parent)) |
1258 | set_bit(CS_NOTIFY_ON_RELEASE, &cs->flags); |
1259 | cs->cpus_allowed = CPU_MASK_NONE; |
1260 | cs->mems_allowed = NODE_MASK_NONE; |
1261 | atomic_set(&cs->count, 0); |
1262 | INIT_LIST_HEAD(&cs->sibling); |
1263 | INIT_LIST_HEAD(&cs->children); |
1264 | atomic_inc(&cpuset_mems_generation); |
1265 | cs->mems_generation = atomic_read(&cpuset_mems_generation); |
1266 | |
1267 | cs->parent = parent; |
1268 | |
1269 | list_add(&cs->sibling, &cs->parent->children); |
1270 | |
1271 | err = cpuset_create_dir(cs, name, mode); |
1272 | if (err < 0) |
1273 | goto err; |
1274 | |
1275 | /* |
1276 | * Release cpuset_sem before cpuset_populate_dir() because it |
1277 | * will down() this new directory's i_sem and if we race with |
1278 | * another mkdir, we might deadlock. |
1279 | */ |
1280 | up(&cpuset_sem); |
1281 | |
1282 | err = cpuset_populate_dir(cs->dentry); |
1283 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
1284 | return 0; |
1285 | err: |
1286 | list_del(&cs->sibling); |
1287 | up(&cpuset_sem); |
1288 | kfree(cs); |
1289 | return err; |
1290 | } |
1291 | |
1292 | static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
1293 | { |
1294 | struct cpuset *c_parent = dentry->d_parent->d_fsdata; |
1295 | |
1296 | /* the vfs holds inode->i_sem already */ |
1297 | return cpuset_create(c_parent, dentry->d_name.name, mode | S_IFDIR); |
1298 | } |
1299 | |
1300 | static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry) |
1301 | { |
1302 | struct cpuset *cs = dentry->d_fsdata; |
1303 | struct dentry *d; |
1304 | struct cpuset *parent; |
1305 | |
1306 | /* the vfs holds both inode->i_sem already */ |
1307 | |
1308 | down(&cpuset_sem); |
1309 | refresh_mems(); |
1310 | if (atomic_read(&cs->count) > 0) { |
1311 | up(&cpuset_sem); |
1312 | return -EBUSY; |
1313 | } |
1314 | if (!list_empty(&cs->children)) { |
1315 | up(&cpuset_sem); |
1316 | return -EBUSY; |
1317 | } |
1318 | spin_lock(&cs->dentry->d_lock); |
1319 | parent = cs->parent; |
1320 | set_bit(CS_REMOVED, &cs->flags); |
1321 | list_del(&cs->sibling); /* delete my sibling from parent->children */ |
1322 | if (list_empty(&parent->children)) |
1323 | check_for_release(parent); |
1324 | d = dget(cs->dentry); |
1325 | cs->dentry = NULL; |
1326 | spin_unlock(&d->d_lock); |
1327 | cpuset_d_remove_dir(d); |
1328 | dput(d); |
1329 | up(&cpuset_sem); |
1330 | return 0; |
1331 | } |
1332 | |
1333 | /** |
1334 | * cpuset_init - initialize cpusets at system boot |
1335 | * |
1336 | * Description: Initialize top_cpuset and the cpuset internal file system, |
1337 | **/ |
1338 | |
1339 | int __init cpuset_init(void) |
1340 | { |
1341 | struct dentry *root; |
1342 | int err; |
1343 | |
1344 | top_cpuset.cpus_allowed = CPU_MASK_ALL; |
1345 | top_cpuset.mems_allowed = NODE_MASK_ALL; |
1346 | |
1347 | atomic_inc(&cpuset_mems_generation); |
1348 | top_cpuset.mems_generation = atomic_read(&cpuset_mems_generation); |
1349 | |
1350 | init_task.cpuset = &top_cpuset; |
1351 | |
1352 | err = register_filesystem(&cpuset_fs_type); |
1353 | if (err < 0) |
1354 | goto out; |
1355 | cpuset_mount = kern_mount(&cpuset_fs_type); |
1356 | if (IS_ERR(cpuset_mount)) { |
1357 | printk(KERN_ERR "cpuset: could not mount!\n"); |
1358 | err = PTR_ERR(cpuset_mount); |
1359 | cpuset_mount = NULL; |
1360 | goto out; |
1361 | } |
1362 | root = cpuset_mount->mnt_sb->s_root; |
1363 | root->d_fsdata = &top_cpuset; |
1364 | root->d_inode->i_nlink++; |
1365 | top_cpuset.dentry = root; |
1366 | root->d_inode->i_op = &cpuset_dir_inode_operations; |
1367 | err = cpuset_populate_dir(root); |
1368 | out: |
1369 | return err; |
1370 | } |
1371 | |
1372 | /** |
1373 | * cpuset_init_smp - initialize cpus_allowed |
1374 | * |
1375 | * Description: Finish top cpuset after cpu, node maps are initialized |
1376 | **/ |
1377 | |
1378 | void __init cpuset_init_smp(void) |
1379 | { |
1380 | top_cpuset.cpus_allowed = cpu_online_map; |
1381 | top_cpuset.mems_allowed = node_online_map; |
1382 | } |
1383 | |
1384 | /** |
1385 | * cpuset_fork - attach newly forked task to its parents cpuset. |
1386 | * @p: pointer to task_struct of forking parent process. |
1387 | * |
1388 | * Description: By default, on fork, a task inherits its |
1389 | * parents cpuset. The pointer to the shared cpuset is |
1390 | * automatically copied in fork.c by dup_task_struct(). |
1391 | * This cpuset_fork() routine need only increment the usage |
1392 | * counter in that cpuset. |
1393 | **/ |
1394 | |
1395 | void cpuset_fork(struct task_struct *tsk) |
1396 | { |
1397 | atomic_inc(&tsk->cpuset->count); |
1398 | } |
1399 | |
1400 | /** |
1401 | * cpuset_exit - detach cpuset from exiting task |
1402 | * @tsk: pointer to task_struct of exiting process |
1403 | * |
1404 | * Description: Detach cpuset from @tsk and release it. |
1405 | * |
1406 | * Note that cpusets marked notify_on_release force every task |
1407 | * in them to take the global cpuset_sem semaphore when exiting. |
1408 | * This could impact scaling on very large systems. Be reluctant |
1409 | * to use notify_on_release cpusets where very high task exit |
1410 | * scaling is required on large systems. |
1411 | * |
1412 | * Don't even think about derefencing 'cs' after the cpuset use |
1413 | * count goes to zero, except inside a critical section guarded |
1414 | * by the cpuset_sem semaphore. If you don't hold cpuset_sem, |
1415 | * then a zero cpuset use count is a license to any other task to |
1416 | * nuke the cpuset immediately. |
1417 | * |
1418 | **/ |
1419 | |
1420 | void cpuset_exit(struct task_struct *tsk) |
1421 | { |
1422 | struct cpuset *cs; |
1423 | |
1424 | task_lock(tsk); |
1425 | cs = tsk->cpuset; |
1426 | tsk->cpuset = NULL; |
1427 | task_unlock(tsk); |
1428 | |
1429 | if (notify_on_release(cs)) { |
1430 | down(&cpuset_sem); |
1431 | if (atomic_dec_and_test(&cs->count)) |
1432 | check_for_release(cs); |
1433 | up(&cpuset_sem); |
1434 | } else { |
1435 | atomic_dec(&cs->count); |
1436 | } |
1437 | } |
1438 | |
1439 | /** |
1440 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
1441 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. |
1442 | * |
1443 | * Description: Returns the cpumask_t cpus_allowed of the cpuset |
1444 | * attached to the specified @tsk. Guaranteed to return some non-empty |
1445 | * subset of cpu_online_map, even if this means going outside the |
1446 | * tasks cpuset. |
1447 | **/ |
1448 | |
1449 | cpumask_t cpuset_cpus_allowed(const struct task_struct *tsk) |
1450 | { |
1451 | cpumask_t mask; |
1452 | |
1453 | down(&cpuset_sem); |
1454 | task_lock((struct task_struct *)tsk); |
1455 | guarantee_online_cpus(tsk->cpuset, &mask); |
1456 | task_unlock((struct task_struct *)tsk); |
1457 | up(&cpuset_sem); |
1458 | |
1459 | return mask; |
1460 | } |
1461 | |
1462 | void cpuset_init_current_mems_allowed(void) |
1463 | { |
1464 | current->mems_allowed = NODE_MASK_ALL; |
1465 | } |
1466 | |
1467 | /* |
1468 | * If the current tasks cpusets mems_allowed changed behind our backs, |
1469 | * update current->mems_allowed and mems_generation to the new value. |
1470 | * Do not call this routine if in_interrupt(). |
1471 | */ |
1472 | |
1473 | void cpuset_update_current_mems_allowed(void) |
1474 | { |
1475 | struct cpuset *cs = current->cpuset; |
1476 | |
1477 | if (!cs) |
1478 | return; /* task is exiting */ |
1479 | if (current->cpuset_mems_generation != cs->mems_generation) { |
1480 | down(&cpuset_sem); |
1481 | refresh_mems(); |
1482 | up(&cpuset_sem); |
1483 | } |
1484 | } |
1485 | |
1486 | void cpuset_restrict_to_mems_allowed(unsigned long *nodes) |
1487 | { |
1488 | bitmap_and(nodes, nodes, nodes_addr(current->mems_allowed), |
1489 | MAX_NUMNODES); |
1490 | } |
1491 | |
1492 | /* |
1493 | * Are any of the nodes on zonelist zl allowed in current->mems_allowed? |
1494 | */ |
1495 | int cpuset_zonelist_valid_mems_allowed(struct zonelist *zl) |
1496 | { |
1497 | int i; |
1498 | |
1499 | for (i = 0; zl->zones[i]; i++) { |
1500 | int nid = zl->zones[i]->zone_pgdat->node_id; |
1501 | |
1502 | if (node_isset(nid, current->mems_allowed)) |
1503 | return 1; |
1504 | } |
1505 | return 0; |
1506 | } |
1507 | |
1508 | /* |
1509 | * Is 'current' valid, and is zone z allowed in current->mems_allowed? |
1510 | */ |
1511 | int cpuset_zone_allowed(struct zone *z) |
1512 | { |
1513 | return in_interrupt() || |
1514 | node_isset(z->zone_pgdat->node_id, current->mems_allowed); |
1515 | } |
1516 | |
1517 | /* |
1518 | * proc_cpuset_show() |
1519 | * - Print tasks cpuset path into seq_file. |
1520 | * - Used for /proc/<pid>/cpuset. |
1521 | */ |
1522 | |
1523 | static int proc_cpuset_show(struct seq_file *m, void *v) |
1524 | { |
1525 | struct cpuset *cs; |
1526 | struct task_struct *tsk; |
1527 | char *buf; |
1528 | int retval = 0; |
1529 | |
1530 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
1531 | if (!buf) |
1532 | return -ENOMEM; |
1533 | |
1534 | tsk = m->private; |
1535 | down(&cpuset_sem); |
1536 | task_lock(tsk); |
1537 | cs = tsk->cpuset; |
1538 | task_unlock(tsk); |
1539 | if (!cs) { |
1540 | retval = -EINVAL; |
1541 | goto out; |
1542 | } |
1543 | |
1544 | retval = cpuset_path(cs, buf, PAGE_SIZE); |
1545 | if (retval < 0) |
1546 | goto out; |
1547 | seq_puts(m, buf); |
1548 | seq_putc(m, '\n'); |
1549 | out: |
1550 | up(&cpuset_sem); |
1551 | kfree(buf); |
1552 | return retval; |
1553 | } |
1554 | |
1555 | static int cpuset_open(struct inode *inode, struct file *file) |
1556 | { |
1557 | struct task_struct *tsk = PROC_I(inode)->task; |
1558 | return single_open(file, proc_cpuset_show, tsk); |
1559 | } |
1560 | |
1561 | struct file_operations proc_cpuset_operations = { |
1562 | .open = cpuset_open, |
1563 | .read = seq_read, |
1564 | .llseek = seq_lseek, |
1565 | .release = single_release, |
1566 | }; |
1567 | |
1568 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ |
1569 | char *cpuset_task_status_allowed(struct task_struct *task, char *buffer) |
1570 | { |
1571 | buffer += sprintf(buffer, "Cpus_allowed:\t"); |
1572 | buffer += cpumask_scnprintf(buffer, PAGE_SIZE, task->cpus_allowed); |
1573 | buffer += sprintf(buffer, "\n"); |
1574 | buffer += sprintf(buffer, "Mems_allowed:\t"); |
1575 | buffer += nodemask_scnprintf(buffer, PAGE_SIZE, task->mems_allowed); |
1576 | buffer += sprintf(buffer, "\n"); |
1577 | return buffer; |
1578 | } |