Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | * if the last task using a cpuset exits, and the cpuset was marked |
| 170 | * notify_on_release. In that case, the cpuset_sem is taken, the |
| 171 | * path to the released cpuset calculated, and a usermode call made |
| 172 | * to /sbin/cpuset_release_agent with the name of the cpuset (path |
| 173 | * relative to the root of cpuset file system) as the argument. |
| 174 | * |
| 175 | * A cpuset can only be deleted if both its 'count' of using tasks is |
| 176 | * zero, and its list of 'children' cpusets is empty. Since all tasks |
| 177 | * in the system use _some_ cpuset, and since there is always at least |
| 178 | * one task in the system (init, pid == 1), therefore, top_cpuset |
| 179 | * always has either children cpusets and/or using tasks. So no need |
| 180 | * for any special hack to ensure that top_cpuset cannot be deleted. |
| 181 | */ |
| 182 | |
| 183 | static DECLARE_MUTEX(cpuset_sem); |
| 184 | |
| 185 | /* |
| 186 | * A couple of forward declarations required, due to cyclic reference loop: |
| 187 | * cpuset_mkdir -> cpuset_create -> cpuset_populate_dir -> cpuset_add_file |
| 188 | * -> cpuset_create_file -> cpuset_dir_inode_operations -> cpuset_mkdir. |
| 189 | */ |
| 190 | |
| 191 | static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode); |
| 192 | static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry); |
| 193 | |
| 194 | static struct backing_dev_info cpuset_backing_dev_info = { |
| 195 | .ra_pages = 0, /* No readahead */ |
| 196 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, |
| 197 | }; |
| 198 | |
| 199 | static struct inode *cpuset_new_inode(mode_t mode) |
| 200 | { |
| 201 | struct inode *inode = new_inode(cpuset_sb); |
| 202 | |
| 203 | if (inode) { |
| 204 | inode->i_mode = mode; |
| 205 | inode->i_uid = current->fsuid; |
| 206 | inode->i_gid = current->fsgid; |
| 207 | inode->i_blksize = PAGE_CACHE_SIZE; |
| 208 | inode->i_blocks = 0; |
| 209 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
| 210 | inode->i_mapping->backing_dev_info = &cpuset_backing_dev_info; |
| 211 | } |
| 212 | return inode; |
| 213 | } |
| 214 | |
| 215 | static void cpuset_diput(struct dentry *dentry, struct inode *inode) |
| 216 | { |
| 217 | /* is dentry a directory ? if so, kfree() associated cpuset */ |
| 218 | if (S_ISDIR(inode->i_mode)) { |
| 219 | struct cpuset *cs = dentry->d_fsdata; |
| 220 | BUG_ON(!(is_removed(cs))); |
| 221 | kfree(cs); |
| 222 | } |
| 223 | iput(inode); |
| 224 | } |
| 225 | |
| 226 | static struct dentry_operations cpuset_dops = { |
| 227 | .d_iput = cpuset_diput, |
| 228 | }; |
| 229 | |
| 230 | static struct dentry *cpuset_get_dentry(struct dentry *parent, const char *name) |
| 231 | { |
| 232 | struct qstr qstr; |
| 233 | struct dentry *d; |
| 234 | |
| 235 | qstr.name = name; |
| 236 | qstr.len = strlen(name); |
| 237 | qstr.hash = full_name_hash(name, qstr.len); |
| 238 | d = lookup_hash(&qstr, parent); |
| 239 | if (!IS_ERR(d)) |
| 240 | d->d_op = &cpuset_dops; |
| 241 | return d; |
| 242 | } |
| 243 | |
| 244 | static void remove_dir(struct dentry *d) |
| 245 | { |
| 246 | struct dentry *parent = dget(d->d_parent); |
| 247 | |
| 248 | d_delete(d); |
| 249 | simple_rmdir(parent->d_inode, d); |
| 250 | dput(parent); |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * NOTE : the dentry must have been dget()'ed |
| 255 | */ |
| 256 | static void cpuset_d_remove_dir(struct dentry *dentry) |
| 257 | { |
| 258 | struct list_head *node; |
| 259 | |
| 260 | spin_lock(&dcache_lock); |
| 261 | node = dentry->d_subdirs.next; |
| 262 | while (node != &dentry->d_subdirs) { |
| 263 | struct dentry *d = list_entry(node, struct dentry, d_child); |
| 264 | list_del_init(node); |
| 265 | if (d->d_inode) { |
| 266 | d = dget_locked(d); |
| 267 | spin_unlock(&dcache_lock); |
| 268 | d_delete(d); |
| 269 | simple_unlink(dentry->d_inode, d); |
| 270 | dput(d); |
| 271 | spin_lock(&dcache_lock); |
| 272 | } |
| 273 | node = dentry->d_subdirs.next; |
| 274 | } |
| 275 | list_del_init(&dentry->d_child); |
| 276 | spin_unlock(&dcache_lock); |
| 277 | remove_dir(dentry); |
| 278 | } |
| 279 | |
| 280 | static struct super_operations cpuset_ops = { |
| 281 | .statfs = simple_statfs, |
| 282 | .drop_inode = generic_delete_inode, |
| 283 | }; |
| 284 | |
| 285 | static int cpuset_fill_super(struct super_block *sb, void *unused_data, |
| 286 | int unused_silent) |
| 287 | { |
| 288 | struct inode *inode; |
| 289 | struct dentry *root; |
| 290 | |
| 291 | sb->s_blocksize = PAGE_CACHE_SIZE; |
| 292 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; |
| 293 | sb->s_magic = CPUSET_SUPER_MAGIC; |
| 294 | sb->s_op = &cpuset_ops; |
| 295 | cpuset_sb = sb; |
| 296 | |
| 297 | inode = cpuset_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR); |
| 298 | if (inode) { |
| 299 | inode->i_op = &simple_dir_inode_operations; |
| 300 | inode->i_fop = &simple_dir_operations; |
| 301 | /* directories start off with i_nlink == 2 (for "." entry) */ |
| 302 | inode->i_nlink++; |
| 303 | } else { |
| 304 | return -ENOMEM; |
| 305 | } |
| 306 | |
| 307 | root = d_alloc_root(inode); |
| 308 | if (!root) { |
| 309 | iput(inode); |
| 310 | return -ENOMEM; |
| 311 | } |
| 312 | sb->s_root = root; |
| 313 | return 0; |
| 314 | } |
| 315 | |
| 316 | static struct super_block *cpuset_get_sb(struct file_system_type *fs_type, |
| 317 | int flags, const char *unused_dev_name, |
| 318 | void *data) |
| 319 | { |
| 320 | return get_sb_single(fs_type, flags, data, cpuset_fill_super); |
| 321 | } |
| 322 | |
| 323 | static struct file_system_type cpuset_fs_type = { |
| 324 | .name = "cpuset", |
| 325 | .get_sb = cpuset_get_sb, |
| 326 | .kill_sb = kill_litter_super, |
| 327 | }; |
| 328 | |
| 329 | /* struct cftype: |
| 330 | * |
| 331 | * The files in the cpuset filesystem mostly have a very simple read/write |
| 332 | * handling, some common function will take care of it. Nevertheless some cases |
| 333 | * (read tasks) are special and therefore I define this structure for every |
| 334 | * kind of file. |
| 335 | * |
| 336 | * |
| 337 | * When reading/writing to a file: |
| 338 | * - the cpuset to use in file->f_dentry->d_parent->d_fsdata |
| 339 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
| 340 | */ |
| 341 | |
| 342 | struct cftype { |
| 343 | char *name; |
| 344 | int private; |
| 345 | int (*open) (struct inode *inode, struct file *file); |
| 346 | ssize_t (*read) (struct file *file, char __user *buf, size_t nbytes, |
| 347 | loff_t *ppos); |
| 348 | int (*write) (struct file *file, const char __user *buf, size_t nbytes, |
| 349 | loff_t *ppos); |
| 350 | int (*release) (struct inode *inode, struct file *file); |
| 351 | }; |
| 352 | |
| 353 | static inline struct cpuset *__d_cs(struct dentry *dentry) |
| 354 | { |
| 355 | return dentry->d_fsdata; |
| 356 | } |
| 357 | |
| 358 | static inline struct cftype *__d_cft(struct dentry *dentry) |
| 359 | { |
| 360 | return dentry->d_fsdata; |
| 361 | } |
| 362 | |
| 363 | /* |
| 364 | * Call with cpuset_sem held. Writes path of cpuset into buf. |
| 365 | * Returns 0 on success, -errno on error. |
| 366 | */ |
| 367 | |
| 368 | static int cpuset_path(const struct cpuset *cs, char *buf, int buflen) |
| 369 | { |
| 370 | char *start; |
| 371 | |
| 372 | start = buf + buflen; |
| 373 | |
| 374 | *--start = '\0'; |
| 375 | for (;;) { |
| 376 | int len = cs->dentry->d_name.len; |
| 377 | if ((start -= len) < buf) |
| 378 | return -ENAMETOOLONG; |
| 379 | memcpy(start, cs->dentry->d_name.name, len); |
| 380 | cs = cs->parent; |
| 381 | if (!cs) |
| 382 | break; |
| 383 | if (!cs->parent) |
| 384 | continue; |
| 385 | if (--start < buf) |
| 386 | return -ENAMETOOLONG; |
| 387 | *start = '/'; |
| 388 | } |
| 389 | memmove(buf, start, buf + buflen - start); |
| 390 | return 0; |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * Notify userspace when a cpuset is released, by running |
| 395 | * /sbin/cpuset_release_agent with the name of the cpuset (path |
| 396 | * relative to the root of cpuset file system) as the argument. |
| 397 | * |
| 398 | * Most likely, this user command will try to rmdir this cpuset. |
| 399 | * |
| 400 | * This races with the possibility that some other task will be |
| 401 | * attached to this cpuset before it is removed, or that some other |
| 402 | * user task will 'mkdir' a child cpuset of this cpuset. That's ok. |
| 403 | * The presumed 'rmdir' will fail quietly if this cpuset is no longer |
| 404 | * unused, and this cpuset will be reprieved from its death sentence, |
| 405 | * to continue to serve a useful existence. Next time it's released, |
| 406 | * we will get notified again, if it still has 'notify_on_release' set. |
| 407 | * |
| 408 | * Note final arg to call_usermodehelper() is 0 - that means |
| 409 | * don't wait. Since we are holding the global cpuset_sem here, |
| 410 | * and we are asking another thread (started from keventd) to rmdir a |
| 411 | * cpuset, we can't wait - or we'd deadlock with the removing thread |
| 412 | * on cpuset_sem. |
| 413 | */ |
| 414 | |
| 415 | static int cpuset_release_agent(char *cpuset_str) |
| 416 | { |
| 417 | char *argv[3], *envp[3]; |
| 418 | int i; |
| 419 | |
| 420 | i = 0; |
| 421 | argv[i++] = "/sbin/cpuset_release_agent"; |
| 422 | argv[i++] = cpuset_str; |
| 423 | argv[i] = NULL; |
| 424 | |
| 425 | i = 0; |
| 426 | /* minimal command environment */ |
| 427 | envp[i++] = "HOME=/"; |
| 428 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; |
| 429 | envp[i] = NULL; |
| 430 | |
| 431 | return call_usermodehelper(argv[0], argv, envp, 0); |
| 432 | } |
| 433 | |
| 434 | /* |
| 435 | * Either cs->count of using tasks transitioned to zero, or the |
| 436 | * cs->children list of child cpusets just became empty. If this |
| 437 | * cs is notify_on_release() and now both the user count is zero and |
| 438 | * the list of children is empty, send notice to user land. |
| 439 | */ |
| 440 | |
| 441 | static void check_for_release(struct cpuset *cs) |
| 442 | { |
| 443 | if (notify_on_release(cs) && atomic_read(&cs->count) == 0 && |
| 444 | list_empty(&cs->children)) { |
| 445 | char *buf; |
| 446 | |
| 447 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 448 | if (!buf) |
| 449 | return; |
| 450 | if (cpuset_path(cs, buf, PAGE_SIZE) < 0) |
| 451 | goto out; |
| 452 | cpuset_release_agent(buf); |
| 453 | out: |
| 454 | kfree(buf); |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | /* |
| 459 | * Return in *pmask the portion of a cpusets's cpus_allowed that |
| 460 | * are online. If none are online, walk up the cpuset hierarchy |
| 461 | * until we find one that does have some online cpus. If we get |
| 462 | * all the way to the top and still haven't found any online cpus, |
| 463 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing |
| 464 | * task, return cpu_online_map. |
| 465 | * |
| 466 | * One way or another, we guarantee to return some non-empty subset |
| 467 | * of cpu_online_map. |
| 468 | * |
| 469 | * Call with cpuset_sem held. |
| 470 | */ |
| 471 | |
| 472 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) |
| 473 | { |
| 474 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) |
| 475 | cs = cs->parent; |
| 476 | if (cs) |
| 477 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); |
| 478 | else |
| 479 | *pmask = cpu_online_map; |
| 480 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); |
| 481 | } |
| 482 | |
| 483 | /* |
| 484 | * Return in *pmask the portion of a cpusets's mems_allowed that |
| 485 | * are online. If none are online, walk up the cpuset hierarchy |
| 486 | * until we find one that does have some online mems. If we get |
| 487 | * all the way to the top and still haven't found any online mems, |
| 488 | * return node_online_map. |
| 489 | * |
| 490 | * One way or another, we guarantee to return some non-empty subset |
| 491 | * of node_online_map. |
| 492 | * |
| 493 | * Call with cpuset_sem held. |
| 494 | */ |
| 495 | |
| 496 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
| 497 | { |
| 498 | while (cs && !nodes_intersects(cs->mems_allowed, node_online_map)) |
| 499 | cs = cs->parent; |
| 500 | if (cs) |
| 501 | nodes_and(*pmask, cs->mems_allowed, node_online_map); |
| 502 | else |
| 503 | *pmask = node_online_map; |
| 504 | BUG_ON(!nodes_intersects(*pmask, node_online_map)); |
| 505 | } |
| 506 | |
| 507 | /* |
| 508 | * Refresh current tasks mems_allowed and mems_generation from |
| 509 | * current tasks cpuset. Call with cpuset_sem held. |
| 510 | * |
| 511 | * Be sure to call refresh_mems() on any cpuset operation which |
| 512 | * (1) holds cpuset_sem, and (2) might possibly alloc memory. |
| 513 | * Call after obtaining cpuset_sem lock, before any possible |
| 514 | * allocation. Otherwise one risks trying to allocate memory |
| 515 | * while the task cpuset_mems_generation is not the same as |
| 516 | * the mems_generation in its cpuset, which would deadlock on |
| 517 | * cpuset_sem in cpuset_update_current_mems_allowed(). |
| 518 | * |
| 519 | * Since we hold cpuset_sem, once refresh_mems() is called, the |
| 520 | * test (current->cpuset_mems_generation != cs->mems_generation) |
| 521 | * in cpuset_update_current_mems_allowed() will remain false, |
| 522 | * until we drop cpuset_sem. Anyone else who would change our |
| 523 | * cpusets mems_generation needs to lock cpuset_sem first. |
| 524 | */ |
| 525 | |
| 526 | static void refresh_mems(void) |
| 527 | { |
| 528 | struct cpuset *cs = current->cpuset; |
| 529 | |
| 530 | if (current->cpuset_mems_generation != cs->mems_generation) { |
| 531 | guarantee_online_mems(cs, ¤t->mems_allowed); |
| 532 | current->cpuset_mems_generation = cs->mems_generation; |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | /* |
| 537 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? |
| 538 | * |
| 539 | * One cpuset is a subset of another if all its allowed CPUs and |
| 540 | * Memory Nodes are a subset of the other, and its exclusive flags |
| 541 | * are only set if the other's are set. |
| 542 | */ |
| 543 | |
| 544 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) |
| 545 | { |
| 546 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && |
| 547 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
| 548 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && |
| 549 | is_mem_exclusive(p) <= is_mem_exclusive(q); |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * validate_change() - Used to validate that any proposed cpuset change |
| 554 | * follows the structural rules for cpusets. |
| 555 | * |
| 556 | * If we replaced the flag and mask values of the current cpuset |
| 557 | * (cur) with those values in the trial cpuset (trial), would |
| 558 | * our various subset and exclusive rules still be valid? Presumes |
| 559 | * cpuset_sem held. |
| 560 | * |
| 561 | * 'cur' is the address of an actual, in-use cpuset. Operations |
| 562 | * such as list traversal that depend on the actual address of the |
| 563 | * cpuset in the list must use cur below, not trial. |
| 564 | * |
| 565 | * 'trial' is the address of bulk structure copy of cur, with |
| 566 | * perhaps one or more of the fields cpus_allowed, mems_allowed, |
| 567 | * or flags changed to new, trial values. |
| 568 | * |
| 569 | * Return 0 if valid, -errno if not. |
| 570 | */ |
| 571 | |
| 572 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) |
| 573 | { |
| 574 | struct cpuset *c, *par; |
| 575 | |
| 576 | /* Each of our child cpusets must be a subset of us */ |
| 577 | list_for_each_entry(c, &cur->children, sibling) { |
| 578 | if (!is_cpuset_subset(c, trial)) |
| 579 | return -EBUSY; |
| 580 | } |
| 581 | |
| 582 | /* Remaining checks don't apply to root cpuset */ |
| 583 | if ((par = cur->parent) == NULL) |
| 584 | return 0; |
| 585 | |
| 586 | /* We must be a subset of our parent cpuset */ |
| 587 | if (!is_cpuset_subset(trial, par)) |
| 588 | return -EACCES; |
| 589 | |
| 590 | /* If either I or some sibling (!= me) is exclusive, we can't overlap */ |
| 591 | list_for_each_entry(c, &par->children, sibling) { |
| 592 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
| 593 | c != cur && |
| 594 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) |
| 595 | return -EINVAL; |
| 596 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
| 597 | c != cur && |
| 598 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) |
| 599 | return -EINVAL; |
| 600 | } |
| 601 | |
| 602 | return 0; |
| 603 | } |
| 604 | |
| 605 | static int update_cpumask(struct cpuset *cs, char *buf) |
| 606 | { |
| 607 | struct cpuset trialcs; |
| 608 | int retval; |
| 609 | |
| 610 | trialcs = *cs; |
| 611 | retval = cpulist_parse(buf, trialcs.cpus_allowed); |
| 612 | if (retval < 0) |
| 613 | return retval; |
| 614 | cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map); |
| 615 | if (cpus_empty(trialcs.cpus_allowed)) |
| 616 | return -ENOSPC; |
| 617 | retval = validate_change(cs, &trialcs); |
| 618 | if (retval == 0) |
| 619 | cs->cpus_allowed = trialcs.cpus_allowed; |
| 620 | return retval; |
| 621 | } |
| 622 | |
| 623 | static int update_nodemask(struct cpuset *cs, char *buf) |
| 624 | { |
| 625 | struct cpuset trialcs; |
| 626 | int retval; |
| 627 | |
| 628 | trialcs = *cs; |
| 629 | retval = nodelist_parse(buf, trialcs.mems_allowed); |
| 630 | if (retval < 0) |
| 631 | return retval; |
| 632 | nodes_and(trialcs.mems_allowed, trialcs.mems_allowed, node_online_map); |
| 633 | if (nodes_empty(trialcs.mems_allowed)) |
| 634 | return -ENOSPC; |
| 635 | retval = validate_change(cs, &trialcs); |
| 636 | if (retval == 0) { |
| 637 | cs->mems_allowed = trialcs.mems_allowed; |
| 638 | atomic_inc(&cpuset_mems_generation); |
| 639 | cs->mems_generation = atomic_read(&cpuset_mems_generation); |
| 640 | } |
| 641 | return retval; |
| 642 | } |
| 643 | |
| 644 | /* |
| 645 | * update_flag - read a 0 or a 1 in a file and update associated flag |
| 646 | * bit: the bit to update (CS_CPU_EXCLUSIVE, CS_MEM_EXCLUSIVE, |
| 647 | * CS_NOTIFY_ON_RELEASE) |
| 648 | * cs: the cpuset to update |
| 649 | * buf: the buffer where we read the 0 or 1 |
| 650 | */ |
| 651 | |
| 652 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, char *buf) |
| 653 | { |
| 654 | int turning_on; |
| 655 | struct cpuset trialcs; |
| 656 | int err; |
| 657 | |
| 658 | turning_on = (simple_strtoul(buf, NULL, 10) != 0); |
| 659 | |
| 660 | trialcs = *cs; |
| 661 | if (turning_on) |
| 662 | set_bit(bit, &trialcs.flags); |
| 663 | else |
| 664 | clear_bit(bit, &trialcs.flags); |
| 665 | |
| 666 | err = validate_change(cs, &trialcs); |
| 667 | if (err == 0) { |
| 668 | if (turning_on) |
| 669 | set_bit(bit, &cs->flags); |
| 670 | else |
| 671 | clear_bit(bit, &cs->flags); |
| 672 | } |
| 673 | return err; |
| 674 | } |
| 675 | |
| 676 | static int attach_task(struct cpuset *cs, char *buf) |
| 677 | { |
| 678 | pid_t pid; |
| 679 | struct task_struct *tsk; |
| 680 | struct cpuset *oldcs; |
| 681 | cpumask_t cpus; |
| 682 | |
| 683 | if (sscanf(buf, "%d", &pid) != 1) |
| 684 | return -EIO; |
| 685 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
| 686 | return -ENOSPC; |
| 687 | |
| 688 | if (pid) { |
| 689 | read_lock(&tasklist_lock); |
| 690 | |
| 691 | tsk = find_task_by_pid(pid); |
| 692 | if (!tsk) { |
| 693 | read_unlock(&tasklist_lock); |
| 694 | return -ESRCH; |
| 695 | } |
| 696 | |
| 697 | get_task_struct(tsk); |
| 698 | read_unlock(&tasklist_lock); |
| 699 | |
| 700 | if ((current->euid) && (current->euid != tsk->uid) |
| 701 | && (current->euid != tsk->suid)) { |
| 702 | put_task_struct(tsk); |
| 703 | return -EACCES; |
| 704 | } |
| 705 | } else { |
| 706 | tsk = current; |
| 707 | get_task_struct(tsk); |
| 708 | } |
| 709 | |
| 710 | task_lock(tsk); |
| 711 | oldcs = tsk->cpuset; |
| 712 | if (!oldcs) { |
| 713 | task_unlock(tsk); |
| 714 | put_task_struct(tsk); |
| 715 | return -ESRCH; |
| 716 | } |
| 717 | atomic_inc(&cs->count); |
| 718 | tsk->cpuset = cs; |
| 719 | task_unlock(tsk); |
| 720 | |
| 721 | guarantee_online_cpus(cs, &cpus); |
| 722 | set_cpus_allowed(tsk, cpus); |
| 723 | |
| 724 | put_task_struct(tsk); |
| 725 | if (atomic_dec_and_test(&oldcs->count)) |
| 726 | check_for_release(oldcs); |
| 727 | return 0; |
| 728 | } |
| 729 | |
| 730 | /* The various types of files and directories in a cpuset file system */ |
| 731 | |
| 732 | typedef enum { |
| 733 | FILE_ROOT, |
| 734 | FILE_DIR, |
| 735 | FILE_CPULIST, |
| 736 | FILE_MEMLIST, |
| 737 | FILE_CPU_EXCLUSIVE, |
| 738 | FILE_MEM_EXCLUSIVE, |
| 739 | FILE_NOTIFY_ON_RELEASE, |
| 740 | FILE_TASKLIST, |
| 741 | } cpuset_filetype_t; |
| 742 | |
| 743 | static ssize_t cpuset_common_file_write(struct file *file, const char __user *userbuf, |
| 744 | size_t nbytes, loff_t *unused_ppos) |
| 745 | { |
| 746 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
| 747 | struct cftype *cft = __d_cft(file->f_dentry); |
| 748 | cpuset_filetype_t type = cft->private; |
| 749 | char *buffer; |
| 750 | int retval = 0; |
| 751 | |
| 752 | /* Crude upper limit on largest legitimate cpulist user might write. */ |
| 753 | if (nbytes > 100 + 6 * NR_CPUS) |
| 754 | return -E2BIG; |
| 755 | |
| 756 | /* +1 for nul-terminator */ |
| 757 | if ((buffer = kmalloc(nbytes + 1, GFP_KERNEL)) == 0) |
| 758 | return -ENOMEM; |
| 759 | |
| 760 | if (copy_from_user(buffer, userbuf, nbytes)) { |
| 761 | retval = -EFAULT; |
| 762 | goto out1; |
| 763 | } |
| 764 | buffer[nbytes] = 0; /* nul-terminate */ |
| 765 | |
| 766 | down(&cpuset_sem); |
| 767 | |
| 768 | if (is_removed(cs)) { |
| 769 | retval = -ENODEV; |
| 770 | goto out2; |
| 771 | } |
| 772 | |
| 773 | switch (type) { |
| 774 | case FILE_CPULIST: |
| 775 | retval = update_cpumask(cs, buffer); |
| 776 | break; |
| 777 | case FILE_MEMLIST: |
| 778 | retval = update_nodemask(cs, buffer); |
| 779 | break; |
| 780 | case FILE_CPU_EXCLUSIVE: |
| 781 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, buffer); |
| 782 | break; |
| 783 | case FILE_MEM_EXCLUSIVE: |
| 784 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, buffer); |
| 785 | break; |
| 786 | case FILE_NOTIFY_ON_RELEASE: |
| 787 | retval = update_flag(CS_NOTIFY_ON_RELEASE, cs, buffer); |
| 788 | break; |
| 789 | case FILE_TASKLIST: |
| 790 | retval = attach_task(cs, buffer); |
| 791 | break; |
| 792 | default: |
| 793 | retval = -EINVAL; |
| 794 | goto out2; |
| 795 | } |
| 796 | |
| 797 | if (retval == 0) |
| 798 | retval = nbytes; |
| 799 | out2: |
| 800 | up(&cpuset_sem); |
| 801 | out1: |
| 802 | kfree(buffer); |
| 803 | return retval; |
| 804 | } |
| 805 | |
| 806 | static ssize_t cpuset_file_write(struct file *file, const char __user *buf, |
| 807 | size_t nbytes, loff_t *ppos) |
| 808 | { |
| 809 | ssize_t retval = 0; |
| 810 | struct cftype *cft = __d_cft(file->f_dentry); |
| 811 | if (!cft) |
| 812 | return -ENODEV; |
| 813 | |
| 814 | /* special function ? */ |
| 815 | if (cft->write) |
| 816 | retval = cft->write(file, buf, nbytes, ppos); |
| 817 | else |
| 818 | retval = cpuset_common_file_write(file, buf, nbytes, ppos); |
| 819 | |
| 820 | return retval; |
| 821 | } |
| 822 | |
| 823 | /* |
| 824 | * These ascii lists should be read in a single call, by using a user |
| 825 | * buffer large enough to hold the entire map. If read in smaller |
| 826 | * chunks, there is no guarantee of atomicity. Since the display format |
| 827 | * used, list of ranges of sequential numbers, is variable length, |
| 828 | * and since these maps can change value dynamically, one could read |
| 829 | * gibberish by doing partial reads while a list was changing. |
| 830 | * A single large read to a buffer that crosses a page boundary is |
| 831 | * ok, because the result being copied to user land is not recomputed |
| 832 | * across a page fault. |
| 833 | */ |
| 834 | |
| 835 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
| 836 | { |
| 837 | cpumask_t mask; |
| 838 | |
| 839 | down(&cpuset_sem); |
| 840 | mask = cs->cpus_allowed; |
| 841 | up(&cpuset_sem); |
| 842 | |
| 843 | return cpulist_scnprintf(page, PAGE_SIZE, mask); |
| 844 | } |
| 845 | |
| 846 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
| 847 | { |
| 848 | nodemask_t mask; |
| 849 | |
| 850 | down(&cpuset_sem); |
| 851 | mask = cs->mems_allowed; |
| 852 | up(&cpuset_sem); |
| 853 | |
| 854 | return nodelist_scnprintf(page, PAGE_SIZE, mask); |
| 855 | } |
| 856 | |
| 857 | static ssize_t cpuset_common_file_read(struct file *file, char __user *buf, |
| 858 | size_t nbytes, loff_t *ppos) |
| 859 | { |
| 860 | struct cftype *cft = __d_cft(file->f_dentry); |
| 861 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
| 862 | cpuset_filetype_t type = cft->private; |
| 863 | char *page; |
| 864 | ssize_t retval = 0; |
| 865 | char *s; |
| 866 | char *start; |
| 867 | size_t n; |
| 868 | |
| 869 | if (!(page = (char *)__get_free_page(GFP_KERNEL))) |
| 870 | return -ENOMEM; |
| 871 | |
| 872 | s = page; |
| 873 | |
| 874 | switch (type) { |
| 875 | case FILE_CPULIST: |
| 876 | s += cpuset_sprintf_cpulist(s, cs); |
| 877 | break; |
| 878 | case FILE_MEMLIST: |
| 879 | s += cpuset_sprintf_memlist(s, cs); |
| 880 | break; |
| 881 | case FILE_CPU_EXCLUSIVE: |
| 882 | *s++ = is_cpu_exclusive(cs) ? '1' : '0'; |
| 883 | break; |
| 884 | case FILE_MEM_EXCLUSIVE: |
| 885 | *s++ = is_mem_exclusive(cs) ? '1' : '0'; |
| 886 | break; |
| 887 | case FILE_NOTIFY_ON_RELEASE: |
| 888 | *s++ = notify_on_release(cs) ? '1' : '0'; |
| 889 | break; |
| 890 | default: |
| 891 | retval = -EINVAL; |
| 892 | goto out; |
| 893 | } |
| 894 | *s++ = '\n'; |
| 895 | *s = '\0'; |
| 896 | |
| 897 | start = page + *ppos; |
| 898 | n = s - start; |
| 899 | retval = n - copy_to_user(buf, start, min(n, nbytes)); |
| 900 | *ppos += retval; |
| 901 | out: |
| 902 | free_page((unsigned long)page); |
| 903 | return retval; |
| 904 | } |
| 905 | |
| 906 | static ssize_t cpuset_file_read(struct file *file, char __user *buf, size_t nbytes, |
| 907 | loff_t *ppos) |
| 908 | { |
| 909 | ssize_t retval = 0; |
| 910 | struct cftype *cft = __d_cft(file->f_dentry); |
| 911 | if (!cft) |
| 912 | return -ENODEV; |
| 913 | |
| 914 | /* special function ? */ |
| 915 | if (cft->read) |
| 916 | retval = cft->read(file, buf, nbytes, ppos); |
| 917 | else |
| 918 | retval = cpuset_common_file_read(file, buf, nbytes, ppos); |
| 919 | |
| 920 | return retval; |
| 921 | } |
| 922 | |
| 923 | static int cpuset_file_open(struct inode *inode, struct file *file) |
| 924 | { |
| 925 | int err; |
| 926 | struct cftype *cft; |
| 927 | |
| 928 | err = generic_file_open(inode, file); |
| 929 | if (err) |
| 930 | return err; |
| 931 | |
| 932 | cft = __d_cft(file->f_dentry); |
| 933 | if (!cft) |
| 934 | return -ENODEV; |
| 935 | if (cft->open) |
| 936 | err = cft->open(inode, file); |
| 937 | else |
| 938 | err = 0; |
| 939 | |
| 940 | return err; |
| 941 | } |
| 942 | |
| 943 | static int cpuset_file_release(struct inode *inode, struct file *file) |
| 944 | { |
| 945 | struct cftype *cft = __d_cft(file->f_dentry); |
| 946 | if (cft->release) |
| 947 | return cft->release(inode, file); |
| 948 | return 0; |
| 949 | } |
| 950 | |
| 951 | static struct file_operations cpuset_file_operations = { |
| 952 | .read = cpuset_file_read, |
| 953 | .write = cpuset_file_write, |
| 954 | .llseek = generic_file_llseek, |
| 955 | .open = cpuset_file_open, |
| 956 | .release = cpuset_file_release, |
| 957 | }; |
| 958 | |
| 959 | static struct inode_operations cpuset_dir_inode_operations = { |
| 960 | .lookup = simple_lookup, |
| 961 | .mkdir = cpuset_mkdir, |
| 962 | .rmdir = cpuset_rmdir, |
| 963 | }; |
| 964 | |
| 965 | static int cpuset_create_file(struct dentry *dentry, int mode) |
| 966 | { |
| 967 | struct inode *inode; |
| 968 | |
| 969 | if (!dentry) |
| 970 | return -ENOENT; |
| 971 | if (dentry->d_inode) |
| 972 | return -EEXIST; |
| 973 | |
| 974 | inode = cpuset_new_inode(mode); |
| 975 | if (!inode) |
| 976 | return -ENOMEM; |
| 977 | |
| 978 | if (S_ISDIR(mode)) { |
| 979 | inode->i_op = &cpuset_dir_inode_operations; |
| 980 | inode->i_fop = &simple_dir_operations; |
| 981 | |
| 982 | /* start off with i_nlink == 2 (for "." entry) */ |
| 983 | inode->i_nlink++; |
| 984 | } else if (S_ISREG(mode)) { |
| 985 | inode->i_size = 0; |
| 986 | inode->i_fop = &cpuset_file_operations; |
| 987 | } |
| 988 | |
| 989 | d_instantiate(dentry, inode); |
| 990 | dget(dentry); /* Extra count - pin the dentry in core */ |
| 991 | return 0; |
| 992 | } |
| 993 | |
| 994 | /* |
| 995 | * cpuset_create_dir - create a directory for an object. |
| 996 | * cs: the cpuset we create the directory for. |
| 997 | * It must have a valid ->parent field |
| 998 | * And we are going to fill its ->dentry field. |
| 999 | * name: The name to give to the cpuset directory. Will be copied. |
| 1000 | * mode: mode to set on new directory. |
| 1001 | */ |
| 1002 | |
| 1003 | static int cpuset_create_dir(struct cpuset *cs, const char *name, int mode) |
| 1004 | { |
| 1005 | struct dentry *dentry = NULL; |
| 1006 | struct dentry *parent; |
| 1007 | int error = 0; |
| 1008 | |
| 1009 | parent = cs->parent->dentry; |
| 1010 | dentry = cpuset_get_dentry(parent, name); |
| 1011 | if (IS_ERR(dentry)) |
| 1012 | return PTR_ERR(dentry); |
| 1013 | error = cpuset_create_file(dentry, S_IFDIR | mode); |
| 1014 | if (!error) { |
| 1015 | dentry->d_fsdata = cs; |
| 1016 | parent->d_inode->i_nlink++; |
| 1017 | cs->dentry = dentry; |
| 1018 | } |
| 1019 | dput(dentry); |
| 1020 | |
| 1021 | return error; |
| 1022 | } |
| 1023 | |
| 1024 | static int cpuset_add_file(struct dentry *dir, const struct cftype *cft) |
| 1025 | { |
| 1026 | struct dentry *dentry; |
| 1027 | int error; |
| 1028 | |
| 1029 | down(&dir->d_inode->i_sem); |
| 1030 | dentry = cpuset_get_dentry(dir, cft->name); |
| 1031 | if (!IS_ERR(dentry)) { |
| 1032 | error = cpuset_create_file(dentry, 0644 | S_IFREG); |
| 1033 | if (!error) |
| 1034 | dentry->d_fsdata = (void *)cft; |
| 1035 | dput(dentry); |
| 1036 | } else |
| 1037 | error = PTR_ERR(dentry); |
| 1038 | up(&dir->d_inode->i_sem); |
| 1039 | return error; |
| 1040 | } |
| 1041 | |
| 1042 | /* |
| 1043 | * Stuff for reading the 'tasks' file. |
| 1044 | * |
| 1045 | * Reading this file can return large amounts of data if a cpuset has |
| 1046 | * *lots* of attached tasks. So it may need several calls to read(), |
| 1047 | * but we cannot guarantee that the information we produce is correct |
| 1048 | * unless we produce it entirely atomically. |
| 1049 | * |
| 1050 | * Upon tasks file open(), a struct ctr_struct is allocated, that |
| 1051 | * will have a pointer to an array (also allocated here). The struct |
| 1052 | * ctr_struct * is stored in file->private_data. Its resources will |
| 1053 | * be freed by release() when the file is closed. The array is used |
| 1054 | * to sprintf the PIDs and then used by read(). |
| 1055 | */ |
| 1056 | |
| 1057 | /* cpusets_tasks_read array */ |
| 1058 | |
| 1059 | struct ctr_struct { |
| 1060 | char *buf; |
| 1061 | int bufsz; |
| 1062 | }; |
| 1063 | |
| 1064 | /* |
| 1065 | * Load into 'pidarray' up to 'npids' of the tasks using cpuset 'cs'. |
| 1066 | * Return actual number of pids loaded. |
| 1067 | */ |
| 1068 | static inline int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs) |
| 1069 | { |
| 1070 | int n = 0; |
| 1071 | struct task_struct *g, *p; |
| 1072 | |
| 1073 | read_lock(&tasklist_lock); |
| 1074 | |
| 1075 | do_each_thread(g, p) { |
| 1076 | if (p->cpuset == cs) { |
| 1077 | pidarray[n++] = p->pid; |
| 1078 | if (unlikely(n == npids)) |
| 1079 | goto array_full; |
| 1080 | } |
| 1081 | } while_each_thread(g, p); |
| 1082 | |
| 1083 | array_full: |
| 1084 | read_unlock(&tasklist_lock); |
| 1085 | return n; |
| 1086 | } |
| 1087 | |
| 1088 | static int cmppid(const void *a, const void *b) |
| 1089 | { |
| 1090 | return *(pid_t *)a - *(pid_t *)b; |
| 1091 | } |
| 1092 | |
| 1093 | /* |
| 1094 | * Convert array 'a' of 'npids' pid_t's to a string of newline separated |
| 1095 | * decimal pids in 'buf'. Don't write more than 'sz' chars, but return |
| 1096 | * count 'cnt' of how many chars would be written if buf were large enough. |
| 1097 | */ |
| 1098 | static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids) |
| 1099 | { |
| 1100 | int cnt = 0; |
| 1101 | int i; |
| 1102 | |
| 1103 | for (i = 0; i < npids; i++) |
| 1104 | cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]); |
| 1105 | return cnt; |
| 1106 | } |
| 1107 | |
| 1108 | static int cpuset_tasks_open(struct inode *unused, struct file *file) |
| 1109 | { |
| 1110 | struct cpuset *cs = __d_cs(file->f_dentry->d_parent); |
| 1111 | struct ctr_struct *ctr; |
| 1112 | pid_t *pidarray; |
| 1113 | int npids; |
| 1114 | char c; |
| 1115 | |
| 1116 | if (!(file->f_mode & FMODE_READ)) |
| 1117 | return 0; |
| 1118 | |
| 1119 | ctr = kmalloc(sizeof(*ctr), GFP_KERNEL); |
| 1120 | if (!ctr) |
| 1121 | goto err0; |
| 1122 | |
| 1123 | /* |
| 1124 | * If cpuset gets more users after we read count, we won't have |
| 1125 | * enough space - tough. This race is indistinguishable to the |
| 1126 | * caller from the case that the additional cpuset users didn't |
| 1127 | * show up until sometime later on. |
| 1128 | */ |
| 1129 | npids = atomic_read(&cs->count); |
| 1130 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); |
| 1131 | if (!pidarray) |
| 1132 | goto err1; |
| 1133 | |
| 1134 | npids = pid_array_load(pidarray, npids, cs); |
| 1135 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); |
| 1136 | |
| 1137 | /* Call pid_array_to_buf() twice, first just to get bufsz */ |
| 1138 | ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1; |
| 1139 | ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL); |
| 1140 | if (!ctr->buf) |
| 1141 | goto err2; |
| 1142 | ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids); |
| 1143 | |
| 1144 | kfree(pidarray); |
| 1145 | file->private_data = ctr; |
| 1146 | return 0; |
| 1147 | |
| 1148 | err2: |
| 1149 | kfree(pidarray); |
| 1150 | err1: |
| 1151 | kfree(ctr); |
| 1152 | err0: |
| 1153 | return -ENOMEM; |
| 1154 | } |
| 1155 | |
| 1156 | static ssize_t cpuset_tasks_read(struct file *file, char __user *buf, |
| 1157 | size_t nbytes, loff_t *ppos) |
| 1158 | { |
| 1159 | struct ctr_struct *ctr = file->private_data; |
| 1160 | |
| 1161 | if (*ppos + nbytes > ctr->bufsz) |
| 1162 | nbytes = ctr->bufsz - *ppos; |
| 1163 | if (copy_to_user(buf, ctr->buf + *ppos, nbytes)) |
| 1164 | return -EFAULT; |
| 1165 | *ppos += nbytes; |
| 1166 | return nbytes; |
| 1167 | } |
| 1168 | |
| 1169 | static int cpuset_tasks_release(struct inode *unused_inode, struct file *file) |
| 1170 | { |
| 1171 | struct ctr_struct *ctr; |
| 1172 | |
| 1173 | if (file->f_mode & FMODE_READ) { |
| 1174 | ctr = file->private_data; |
| 1175 | kfree(ctr->buf); |
| 1176 | kfree(ctr); |
| 1177 | } |
| 1178 | return 0; |
| 1179 | } |
| 1180 | |
| 1181 | /* |
| 1182 | * for the common functions, 'private' gives the type of file |
| 1183 | */ |
| 1184 | |
| 1185 | static struct cftype cft_tasks = { |
| 1186 | .name = "tasks", |
| 1187 | .open = cpuset_tasks_open, |
| 1188 | .read = cpuset_tasks_read, |
| 1189 | .release = cpuset_tasks_release, |
| 1190 | .private = FILE_TASKLIST, |
| 1191 | }; |
| 1192 | |
| 1193 | static struct cftype cft_cpus = { |
| 1194 | .name = "cpus", |
| 1195 | .private = FILE_CPULIST, |
| 1196 | }; |
| 1197 | |
| 1198 | static struct cftype cft_mems = { |
| 1199 | .name = "mems", |
| 1200 | .private = FILE_MEMLIST, |
| 1201 | }; |
| 1202 | |
| 1203 | static struct cftype cft_cpu_exclusive = { |
| 1204 | .name = "cpu_exclusive", |
| 1205 | .private = FILE_CPU_EXCLUSIVE, |
| 1206 | }; |
| 1207 | |
| 1208 | static struct cftype cft_mem_exclusive = { |
| 1209 | .name = "mem_exclusive", |
| 1210 | .private = FILE_MEM_EXCLUSIVE, |
| 1211 | }; |
| 1212 | |
| 1213 | static struct cftype cft_notify_on_release = { |
| 1214 | .name = "notify_on_release", |
| 1215 | .private = FILE_NOTIFY_ON_RELEASE, |
| 1216 | }; |
| 1217 | |
| 1218 | static int cpuset_populate_dir(struct dentry *cs_dentry) |
| 1219 | { |
| 1220 | int err; |
| 1221 | |
| 1222 | if ((err = cpuset_add_file(cs_dentry, &cft_cpus)) < 0) |
| 1223 | return err; |
| 1224 | if ((err = cpuset_add_file(cs_dentry, &cft_mems)) < 0) |
| 1225 | return err; |
| 1226 | if ((err = cpuset_add_file(cs_dentry, &cft_cpu_exclusive)) < 0) |
| 1227 | return err; |
| 1228 | if ((err = cpuset_add_file(cs_dentry, &cft_mem_exclusive)) < 0) |
| 1229 | return err; |
| 1230 | if ((err = cpuset_add_file(cs_dentry, &cft_notify_on_release)) < 0) |
| 1231 | return err; |
| 1232 | if ((err = cpuset_add_file(cs_dentry, &cft_tasks)) < 0) |
| 1233 | return err; |
| 1234 | return 0; |
| 1235 | } |
| 1236 | |
| 1237 | /* |
| 1238 | * cpuset_create - create a cpuset |
| 1239 | * parent: cpuset that will be parent of the new cpuset. |
| 1240 | * name: name of the new cpuset. Will be strcpy'ed. |
| 1241 | * mode: mode to set on new inode |
| 1242 | * |
| 1243 | * Must be called with the semaphore on the parent inode held |
| 1244 | */ |
| 1245 | |
| 1246 | static long cpuset_create(struct cpuset *parent, const char *name, int mode) |
| 1247 | { |
| 1248 | struct cpuset *cs; |
| 1249 | int err; |
| 1250 | |
| 1251 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
| 1252 | if (!cs) |
| 1253 | return -ENOMEM; |
| 1254 | |
| 1255 | down(&cpuset_sem); |
| 1256 | refresh_mems(); |
| 1257 | cs->flags = 0; |
| 1258 | if (notify_on_release(parent)) |
| 1259 | set_bit(CS_NOTIFY_ON_RELEASE, &cs->flags); |
| 1260 | cs->cpus_allowed = CPU_MASK_NONE; |
| 1261 | cs->mems_allowed = NODE_MASK_NONE; |
| 1262 | atomic_set(&cs->count, 0); |
| 1263 | INIT_LIST_HEAD(&cs->sibling); |
| 1264 | INIT_LIST_HEAD(&cs->children); |
| 1265 | atomic_inc(&cpuset_mems_generation); |
| 1266 | cs->mems_generation = atomic_read(&cpuset_mems_generation); |
| 1267 | |
| 1268 | cs->parent = parent; |
| 1269 | |
| 1270 | list_add(&cs->sibling, &cs->parent->children); |
| 1271 | |
| 1272 | err = cpuset_create_dir(cs, name, mode); |
| 1273 | if (err < 0) |
| 1274 | goto err; |
| 1275 | |
| 1276 | /* |
| 1277 | * Release cpuset_sem before cpuset_populate_dir() because it |
| 1278 | * will down() this new directory's i_sem and if we race with |
| 1279 | * another mkdir, we might deadlock. |
| 1280 | */ |
| 1281 | up(&cpuset_sem); |
| 1282 | |
| 1283 | err = cpuset_populate_dir(cs->dentry); |
| 1284 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
| 1285 | return 0; |
| 1286 | err: |
| 1287 | list_del(&cs->sibling); |
| 1288 | up(&cpuset_sem); |
| 1289 | kfree(cs); |
| 1290 | return err; |
| 1291 | } |
| 1292 | |
| 1293 | static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
| 1294 | { |
| 1295 | struct cpuset *c_parent = dentry->d_parent->d_fsdata; |
| 1296 | |
| 1297 | /* the vfs holds inode->i_sem already */ |
| 1298 | return cpuset_create(c_parent, dentry->d_name.name, mode | S_IFDIR); |
| 1299 | } |
| 1300 | |
| 1301 | static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry) |
| 1302 | { |
| 1303 | struct cpuset *cs = dentry->d_fsdata; |
| 1304 | struct dentry *d; |
| 1305 | struct cpuset *parent; |
| 1306 | |
| 1307 | /* the vfs holds both inode->i_sem already */ |
| 1308 | |
| 1309 | down(&cpuset_sem); |
| 1310 | refresh_mems(); |
| 1311 | if (atomic_read(&cs->count) > 0) { |
| 1312 | up(&cpuset_sem); |
| 1313 | return -EBUSY; |
| 1314 | } |
| 1315 | if (!list_empty(&cs->children)) { |
| 1316 | up(&cpuset_sem); |
| 1317 | return -EBUSY; |
| 1318 | } |
| 1319 | spin_lock(&cs->dentry->d_lock); |
| 1320 | parent = cs->parent; |
| 1321 | set_bit(CS_REMOVED, &cs->flags); |
| 1322 | list_del(&cs->sibling); /* delete my sibling from parent->children */ |
| 1323 | if (list_empty(&parent->children)) |
| 1324 | check_for_release(parent); |
| 1325 | d = dget(cs->dentry); |
| 1326 | cs->dentry = NULL; |
| 1327 | spin_unlock(&d->d_lock); |
| 1328 | cpuset_d_remove_dir(d); |
| 1329 | dput(d); |
| 1330 | up(&cpuset_sem); |
| 1331 | return 0; |
| 1332 | } |
| 1333 | |
| 1334 | /** |
| 1335 | * cpuset_init - initialize cpusets at system boot |
| 1336 | * |
| 1337 | * Description: Initialize top_cpuset and the cpuset internal file system, |
| 1338 | **/ |
| 1339 | |
| 1340 | int __init cpuset_init(void) |
| 1341 | { |
| 1342 | struct dentry *root; |
| 1343 | int err; |
| 1344 | |
| 1345 | top_cpuset.cpus_allowed = CPU_MASK_ALL; |
| 1346 | top_cpuset.mems_allowed = NODE_MASK_ALL; |
| 1347 | |
| 1348 | atomic_inc(&cpuset_mems_generation); |
| 1349 | top_cpuset.mems_generation = atomic_read(&cpuset_mems_generation); |
| 1350 | |
| 1351 | init_task.cpuset = &top_cpuset; |
| 1352 | |
| 1353 | err = register_filesystem(&cpuset_fs_type); |
| 1354 | if (err < 0) |
| 1355 | goto out; |
| 1356 | cpuset_mount = kern_mount(&cpuset_fs_type); |
| 1357 | if (IS_ERR(cpuset_mount)) { |
| 1358 | printk(KERN_ERR "cpuset: could not mount!\n"); |
| 1359 | err = PTR_ERR(cpuset_mount); |
| 1360 | cpuset_mount = NULL; |
| 1361 | goto out; |
| 1362 | } |
| 1363 | root = cpuset_mount->mnt_sb->s_root; |
| 1364 | root->d_fsdata = &top_cpuset; |
| 1365 | root->d_inode->i_nlink++; |
| 1366 | top_cpuset.dentry = root; |
| 1367 | root->d_inode->i_op = &cpuset_dir_inode_operations; |
| 1368 | err = cpuset_populate_dir(root); |
| 1369 | out: |
| 1370 | return err; |
| 1371 | } |
| 1372 | |
| 1373 | /** |
| 1374 | * cpuset_init_smp - initialize cpus_allowed |
| 1375 | * |
| 1376 | * Description: Finish top cpuset after cpu, node maps are initialized |
| 1377 | **/ |
| 1378 | |
| 1379 | void __init cpuset_init_smp(void) |
| 1380 | { |
| 1381 | top_cpuset.cpus_allowed = cpu_online_map; |
| 1382 | top_cpuset.mems_allowed = node_online_map; |
| 1383 | } |
| 1384 | |
| 1385 | /** |
| 1386 | * cpuset_fork - attach newly forked task to its parents cpuset. |
| 1387 | * @p: pointer to task_struct of forking parent process. |
| 1388 | * |
| 1389 | * Description: By default, on fork, a task inherits its |
| 1390 | * parents cpuset. The pointer to the shared cpuset is |
| 1391 | * automatically copied in fork.c by dup_task_struct(). |
| 1392 | * This cpuset_fork() routine need only increment the usage |
| 1393 | * counter in that cpuset. |
| 1394 | **/ |
| 1395 | |
| 1396 | void cpuset_fork(struct task_struct *tsk) |
| 1397 | { |
| 1398 | atomic_inc(&tsk->cpuset->count); |
| 1399 | } |
| 1400 | |
| 1401 | /** |
| 1402 | * cpuset_exit - detach cpuset from exiting task |
| 1403 | * @tsk: pointer to task_struct of exiting process |
| 1404 | * |
| 1405 | * Description: Detach cpuset from @tsk and release it. |
| 1406 | * |
| 1407 | **/ |
| 1408 | |
| 1409 | void cpuset_exit(struct task_struct *tsk) |
| 1410 | { |
| 1411 | struct cpuset *cs; |
| 1412 | |
| 1413 | task_lock(tsk); |
| 1414 | cs = tsk->cpuset; |
| 1415 | tsk->cpuset = NULL; |
| 1416 | task_unlock(tsk); |
| 1417 | |
| 1418 | if (atomic_dec_and_test(&cs->count)) { |
| 1419 | down(&cpuset_sem); |
| 1420 | check_for_release(cs); |
| 1421 | up(&cpuset_sem); |
| 1422 | } |
| 1423 | } |
| 1424 | |
| 1425 | /** |
| 1426 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
| 1427 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. |
| 1428 | * |
| 1429 | * Description: Returns the cpumask_t cpus_allowed of the cpuset |
| 1430 | * attached to the specified @tsk. Guaranteed to return some non-empty |
| 1431 | * subset of cpu_online_map, even if this means going outside the |
| 1432 | * tasks cpuset. |
| 1433 | **/ |
| 1434 | |
Benoit Boissinot | 9a84889 | 2005-04-16 15:25:59 -0700 | [diff] [blame] | 1435 | cpumask_t cpuset_cpus_allowed(const struct task_struct *tsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | { |
| 1437 | cpumask_t mask; |
| 1438 | |
| 1439 | down(&cpuset_sem); |
| 1440 | task_lock((struct task_struct *)tsk); |
| 1441 | guarantee_online_cpus(tsk->cpuset, &mask); |
| 1442 | task_unlock((struct task_struct *)tsk); |
| 1443 | up(&cpuset_sem); |
| 1444 | |
| 1445 | return mask; |
| 1446 | } |
| 1447 | |
| 1448 | void cpuset_init_current_mems_allowed(void) |
| 1449 | { |
| 1450 | current->mems_allowed = NODE_MASK_ALL; |
| 1451 | } |
| 1452 | |
| 1453 | /* |
| 1454 | * If the current tasks cpusets mems_allowed changed behind our backs, |
| 1455 | * update current->mems_allowed and mems_generation to the new value. |
| 1456 | * Do not call this routine if in_interrupt(). |
| 1457 | */ |
| 1458 | |
| 1459 | void cpuset_update_current_mems_allowed(void) |
| 1460 | { |
| 1461 | struct cpuset *cs = current->cpuset; |
| 1462 | |
| 1463 | if (!cs) |
| 1464 | return; /* task is exiting */ |
| 1465 | if (current->cpuset_mems_generation != cs->mems_generation) { |
| 1466 | down(&cpuset_sem); |
| 1467 | refresh_mems(); |
| 1468 | up(&cpuset_sem); |
| 1469 | } |
| 1470 | } |
| 1471 | |
| 1472 | void cpuset_restrict_to_mems_allowed(unsigned long *nodes) |
| 1473 | { |
| 1474 | bitmap_and(nodes, nodes, nodes_addr(current->mems_allowed), |
| 1475 | MAX_NUMNODES); |
| 1476 | } |
| 1477 | |
| 1478 | /* |
| 1479 | * Are any of the nodes on zonelist zl allowed in current->mems_allowed? |
| 1480 | */ |
| 1481 | int cpuset_zonelist_valid_mems_allowed(struct zonelist *zl) |
| 1482 | { |
| 1483 | int i; |
| 1484 | |
| 1485 | for (i = 0; zl->zones[i]; i++) { |
| 1486 | int nid = zl->zones[i]->zone_pgdat->node_id; |
| 1487 | |
| 1488 | if (node_isset(nid, current->mems_allowed)) |
| 1489 | return 1; |
| 1490 | } |
| 1491 | return 0; |
| 1492 | } |
| 1493 | |
| 1494 | /* |
| 1495 | * Is 'current' valid, and is zone z allowed in current->mems_allowed? |
| 1496 | */ |
| 1497 | int cpuset_zone_allowed(struct zone *z) |
| 1498 | { |
| 1499 | return in_interrupt() || |
| 1500 | node_isset(z->zone_pgdat->node_id, current->mems_allowed); |
| 1501 | } |
| 1502 | |
| 1503 | /* |
| 1504 | * proc_cpuset_show() |
| 1505 | * - Print tasks cpuset path into seq_file. |
| 1506 | * - Used for /proc/<pid>/cpuset. |
| 1507 | */ |
| 1508 | |
| 1509 | static int proc_cpuset_show(struct seq_file *m, void *v) |
| 1510 | { |
| 1511 | struct cpuset *cs; |
| 1512 | struct task_struct *tsk; |
| 1513 | char *buf; |
| 1514 | int retval = 0; |
| 1515 | |
| 1516 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 1517 | if (!buf) |
| 1518 | return -ENOMEM; |
| 1519 | |
| 1520 | tsk = m->private; |
| 1521 | down(&cpuset_sem); |
| 1522 | task_lock(tsk); |
| 1523 | cs = tsk->cpuset; |
| 1524 | task_unlock(tsk); |
| 1525 | if (!cs) { |
| 1526 | retval = -EINVAL; |
| 1527 | goto out; |
| 1528 | } |
| 1529 | |
| 1530 | retval = cpuset_path(cs, buf, PAGE_SIZE); |
| 1531 | if (retval < 0) |
| 1532 | goto out; |
| 1533 | seq_puts(m, buf); |
| 1534 | seq_putc(m, '\n'); |
| 1535 | out: |
| 1536 | up(&cpuset_sem); |
| 1537 | kfree(buf); |
| 1538 | return retval; |
| 1539 | } |
| 1540 | |
| 1541 | static int cpuset_open(struct inode *inode, struct file *file) |
| 1542 | { |
| 1543 | struct task_struct *tsk = PROC_I(inode)->task; |
| 1544 | return single_open(file, proc_cpuset_show, tsk); |
| 1545 | } |
| 1546 | |
| 1547 | struct file_operations proc_cpuset_operations = { |
| 1548 | .open = cpuset_open, |
| 1549 | .read = seq_read, |
| 1550 | .llseek = seq_lseek, |
| 1551 | .release = single_release, |
| 1552 | }; |
| 1553 | |
| 1554 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ |
| 1555 | char *cpuset_task_status_allowed(struct task_struct *task, char *buffer) |
| 1556 | { |
| 1557 | buffer += sprintf(buffer, "Cpus_allowed:\t"); |
| 1558 | buffer += cpumask_scnprintf(buffer, PAGE_SIZE, task->cpus_allowed); |
| 1559 | buffer += sprintf(buffer, "\n"); |
| 1560 | buffer += sprintf(buffer, "Mems_allowed:\t"); |
| 1561 | buffer += nodemask_scnprintf(buffer, PAGE_SIZE, task->mems_allowed); |
| 1562 | buffer += sprintf(buffer, "\n"); |
| 1563 | return buffer; |
| 1564 | } |