| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * fs/eventfd.c |
| 4 | * |
| 5 | * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org> |
| 6 | * |
| 7 | */ |
| 8 | |
| 9 | #include <linux/file.h> |
| 10 | #include <linux/poll.h> |
| 11 | #include <linux/init.h> |
| 12 | #include <linux/fs.h> |
| 13 | #include <linux/sched/signal.h> |
| 14 | #include <linux/kernel.h> |
| 15 | #include <linux/slab.h> |
| 16 | #include <linux/list.h> |
| 17 | #include <linux/spinlock.h> |
| 18 | #include <linux/anon_inodes.h> |
| 19 | #include <linux/syscalls.h> |
| 20 | #include <linux/export.h> |
| 21 | #include <linux/kref.h> |
| 22 | #include <linux/eventfd.h> |
| 23 | #include <linux/proc_fs.h> |
| 24 | #include <linux/seq_file.h> |
| 25 | #include <linux/idr.h> |
| 26 | #include <linux/uio.h> |
| 27 | |
| 28 | static DEFINE_IDA(eventfd_ida); |
| 29 | |
| 30 | struct eventfd_ctx { |
| 31 | struct kref kref; |
| 32 | wait_queue_head_t wqh; |
| 33 | /* |
| 34 | * Every time that a write(2) is performed on an eventfd, the |
| 35 | * value of the __u64 being written is added to "count" and a |
| 36 | * wakeup is performed on "wqh". If EFD_SEMAPHORE flag was not |
| 37 | * specified, a read(2) will return the "count" value to userspace, |
| 38 | * and will reset "count" to zero. The kernel side eventfd_signal() |
| 39 | * also, adds to the "count" counter and issue a wakeup. |
| 40 | */ |
| 41 | __u64 count; |
| 42 | unsigned int flags; |
| 43 | int id; |
| 44 | }; |
| 45 | |
| 46 | /** |
| 47 | * eventfd_signal_mask - Increment the event counter |
| 48 | * @ctx: [in] Pointer to the eventfd context. |
| 49 | * @mask: [in] poll mask |
| 50 | * |
| 51 | * This function is supposed to be called by the kernel in paths that do not |
| 52 | * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX |
| 53 | * value, and we signal this as overflow condition by returning a EPOLLERR |
| 54 | * to poll(2). |
| 55 | */ |
| 56 | void eventfd_signal_mask(struct eventfd_ctx *ctx, __poll_t mask) |
| 57 | { |
| 58 | unsigned long flags; |
| 59 | |
| 60 | /* |
| 61 | * Deadlock or stack overflow issues can happen if we recurse here |
| 62 | * through waitqueue wakeup handlers. If the caller users potentially |
| 63 | * nested waitqueues with custom wakeup handlers, then it should |
| 64 | * check eventfd_signal_allowed() before calling this function. If |
| 65 | * it returns false, the eventfd_signal() call should be deferred to a |
| 66 | * safe context. |
| 67 | */ |
| 68 | if (WARN_ON_ONCE(current->in_eventfd)) |
| 69 | return; |
| 70 | |
| 71 | spin_lock_irqsave(&ctx->wqh.lock, flags); |
| 72 | current->in_eventfd = 1; |
| 73 | if (ctx->count < ULLONG_MAX) |
| 74 | ctx->count++; |
| 75 | if (waitqueue_active(&ctx->wqh)) |
| 76 | wake_up_locked_poll(&ctx->wqh, EPOLLIN | mask); |
| 77 | current->in_eventfd = 0; |
| 78 | spin_unlock_irqrestore(&ctx->wqh.lock, flags); |
| 79 | } |
| 80 | EXPORT_SYMBOL_GPL(eventfd_signal_mask); |
| 81 | |
| 82 | static void eventfd_free_ctx(struct eventfd_ctx *ctx) |
| 83 | { |
| 84 | if (ctx->id >= 0) |
| 85 | ida_free(&eventfd_ida, ctx->id); |
| 86 | kfree(ctx); |
| 87 | } |
| 88 | |
| 89 | static void eventfd_free(struct kref *kref) |
| 90 | { |
| 91 | struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref); |
| 92 | |
| 93 | eventfd_free_ctx(ctx); |
| 94 | } |
| 95 | |
| 96 | /** |
| 97 | * eventfd_ctx_put - Releases a reference to the internal eventfd context. |
| 98 | * @ctx: [in] Pointer to eventfd context. |
| 99 | * |
| 100 | * The eventfd context reference must have been previously acquired either |
| 101 | * with eventfd_ctx_fdget() or eventfd_ctx_fileget(). |
| 102 | */ |
| 103 | void eventfd_ctx_put(struct eventfd_ctx *ctx) |
| 104 | { |
| 105 | kref_put(&ctx->kref, eventfd_free); |
| 106 | } |
| 107 | EXPORT_SYMBOL_GPL(eventfd_ctx_put); |
| 108 | |
| 109 | static int eventfd_release(struct inode *inode, struct file *file) |
| 110 | { |
| 111 | struct eventfd_ctx *ctx = file->private_data; |
| 112 | |
| 113 | wake_up_poll(&ctx->wqh, EPOLLHUP); |
| 114 | eventfd_ctx_put(ctx); |
| 115 | return 0; |
| 116 | } |
| 117 | |
| 118 | static __poll_t eventfd_poll(struct file *file, poll_table *wait) |
| 119 | { |
| 120 | struct eventfd_ctx *ctx = file->private_data; |
| 121 | __poll_t events = 0; |
| 122 | u64 count; |
| 123 | |
| 124 | poll_wait(file, &ctx->wqh, wait); |
| 125 | |
| 126 | /* |
| 127 | * All writes to ctx->count occur within ctx->wqh.lock. This read |
| 128 | * can be done outside ctx->wqh.lock because we know that poll_wait |
| 129 | * takes that lock (through add_wait_queue) if our caller will sleep. |
| 130 | * |
| 131 | * The read _can_ therefore seep into add_wait_queue's critical |
| 132 | * section, but cannot move above it! add_wait_queue's spin_lock acts |
| 133 | * as an acquire barrier and ensures that the read be ordered properly |
| 134 | * against the writes. The following CAN happen and is safe: |
| 135 | * |
| 136 | * poll write |
| 137 | * ----------------- ------------ |
| 138 | * lock ctx->wqh.lock (in poll_wait) |
| 139 | * count = ctx->count |
| 140 | * __add_wait_queue |
| 141 | * unlock ctx->wqh.lock |
| 142 | * lock ctx->qwh.lock |
| 143 | * ctx->count += n |
| 144 | * if (waitqueue_active) |
| 145 | * wake_up_locked_poll |
| 146 | * unlock ctx->qwh.lock |
| 147 | * eventfd_poll returns 0 |
| 148 | * |
| 149 | * but the following, which would miss a wakeup, cannot happen: |
| 150 | * |
| 151 | * poll write |
| 152 | * ----------------- ------------ |
| 153 | * count = ctx->count (INVALID!) |
| 154 | * lock ctx->qwh.lock |
| 155 | * ctx->count += n |
| 156 | * **waitqueue_active is false** |
| 157 | * **no wake_up_locked_poll!** |
| 158 | * unlock ctx->qwh.lock |
| 159 | * lock ctx->wqh.lock (in poll_wait) |
| 160 | * __add_wait_queue |
| 161 | * unlock ctx->wqh.lock |
| 162 | * eventfd_poll returns 0 |
| 163 | */ |
| 164 | count = READ_ONCE(ctx->count); |
| 165 | |
| 166 | if (count > 0) |
| 167 | events |= EPOLLIN; |
| 168 | if (count == ULLONG_MAX) |
| 169 | events |= EPOLLERR; |
| 170 | if (ULLONG_MAX - 1 > count) |
| 171 | events |= EPOLLOUT; |
| 172 | |
| 173 | return events; |
| 174 | } |
| 175 | |
| 176 | void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt) |
| 177 | { |
| 178 | lockdep_assert_held(&ctx->wqh.lock); |
| 179 | |
| 180 | *cnt = ((ctx->flags & EFD_SEMAPHORE) && ctx->count) ? 1 : ctx->count; |
| 181 | ctx->count -= *cnt; |
| 182 | } |
| 183 | EXPORT_SYMBOL_GPL(eventfd_ctx_do_read); |
| 184 | |
| 185 | /** |
| 186 | * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue. |
| 187 | * @ctx: [in] Pointer to eventfd context. |
| 188 | * @wait: [in] Wait queue to be removed. |
| 189 | * @cnt: [out] Pointer to the 64-bit counter value. |
| 190 | * |
| 191 | * Returns %0 if successful, or the following error codes: |
| 192 | * |
| 193 | * -EAGAIN : The operation would have blocked. |
| 194 | * |
| 195 | * This is used to atomically remove a wait queue entry from the eventfd wait |
| 196 | * queue head, and read/reset the counter value. |
| 197 | */ |
| 198 | int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait, |
| 199 | __u64 *cnt) |
| 200 | { |
| 201 | unsigned long flags; |
| 202 | |
| 203 | spin_lock_irqsave(&ctx->wqh.lock, flags); |
| 204 | eventfd_ctx_do_read(ctx, cnt); |
| 205 | __remove_wait_queue(&ctx->wqh, wait); |
| 206 | if (*cnt != 0 && waitqueue_active(&ctx->wqh)) |
| 207 | wake_up_locked_poll(&ctx->wqh, EPOLLOUT); |
| 208 | spin_unlock_irqrestore(&ctx->wqh.lock, flags); |
| 209 | |
| 210 | return *cnt != 0 ? 0 : -EAGAIN; |
| 211 | } |
| 212 | EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue); |
| 213 | |
| 214 | static ssize_t eventfd_read(struct kiocb *iocb, struct iov_iter *to) |
| 215 | { |
| 216 | struct file *file = iocb->ki_filp; |
| 217 | struct eventfd_ctx *ctx = file->private_data; |
| 218 | __u64 ucnt = 0; |
| 219 | |
| 220 | if (iov_iter_count(to) < sizeof(ucnt)) |
| 221 | return -EINVAL; |
| 222 | spin_lock_irq(&ctx->wqh.lock); |
| 223 | if (!ctx->count) { |
| 224 | if ((file->f_flags & O_NONBLOCK) || |
| 225 | (iocb->ki_flags & IOCB_NOWAIT)) { |
| 226 | spin_unlock_irq(&ctx->wqh.lock); |
| 227 | return -EAGAIN; |
| 228 | } |
| 229 | |
| 230 | if (wait_event_interruptible_locked_irq(ctx->wqh, ctx->count)) { |
| 231 | spin_unlock_irq(&ctx->wqh.lock); |
| 232 | return -ERESTARTSYS; |
| 233 | } |
| 234 | } |
| 235 | eventfd_ctx_do_read(ctx, &ucnt); |
| 236 | current->in_eventfd = 1; |
| 237 | if (waitqueue_active(&ctx->wqh)) |
| 238 | wake_up_locked_poll(&ctx->wqh, EPOLLOUT); |
| 239 | current->in_eventfd = 0; |
| 240 | spin_unlock_irq(&ctx->wqh.lock); |
| 241 | if (unlikely(copy_to_iter(&ucnt, sizeof(ucnt), to) != sizeof(ucnt))) |
| 242 | return -EFAULT; |
| 243 | |
| 244 | return sizeof(ucnt); |
| 245 | } |
| 246 | |
| 247 | static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count, |
| 248 | loff_t *ppos) |
| 249 | { |
| 250 | struct eventfd_ctx *ctx = file->private_data; |
| 251 | ssize_t res; |
| 252 | __u64 ucnt; |
| 253 | |
| 254 | if (count < sizeof(ucnt)) |
| 255 | return -EINVAL; |
| 256 | if (copy_from_user(&ucnt, buf, sizeof(ucnt))) |
| 257 | return -EFAULT; |
| 258 | if (ucnt == ULLONG_MAX) |
| 259 | return -EINVAL; |
| 260 | spin_lock_irq(&ctx->wqh.lock); |
| 261 | res = -EAGAIN; |
| 262 | if (ULLONG_MAX - ctx->count > ucnt) |
| 263 | res = sizeof(ucnt); |
| 264 | else if (!(file->f_flags & O_NONBLOCK)) { |
| 265 | res = wait_event_interruptible_locked_irq(ctx->wqh, |
| 266 | ULLONG_MAX - ctx->count > ucnt); |
| 267 | if (!res) |
| 268 | res = sizeof(ucnt); |
| 269 | } |
| 270 | if (likely(res > 0)) { |
| 271 | ctx->count += ucnt; |
| 272 | current->in_eventfd = 1; |
| 273 | if (waitqueue_active(&ctx->wqh)) |
| 274 | wake_up_locked_poll(&ctx->wqh, EPOLLIN); |
| 275 | current->in_eventfd = 0; |
| 276 | } |
| 277 | spin_unlock_irq(&ctx->wqh.lock); |
| 278 | |
| 279 | return res; |
| 280 | } |
| 281 | |
| 282 | #ifdef CONFIG_PROC_FS |
| 283 | static void eventfd_show_fdinfo(struct seq_file *m, struct file *f) |
| 284 | { |
| 285 | struct eventfd_ctx *ctx = f->private_data; |
| 286 | |
| 287 | spin_lock_irq(&ctx->wqh.lock); |
| 288 | seq_printf(m, "eventfd-count: %16llx\n", |
| 289 | (unsigned long long)ctx->count); |
| 290 | spin_unlock_irq(&ctx->wqh.lock); |
| 291 | seq_printf(m, "eventfd-id: %d\n", ctx->id); |
| 292 | seq_printf(m, "eventfd-semaphore: %d\n", |
| 293 | !!(ctx->flags & EFD_SEMAPHORE)); |
| 294 | } |
| 295 | #endif |
| 296 | |
| 297 | static const struct file_operations eventfd_fops = { |
| 298 | #ifdef CONFIG_PROC_FS |
| 299 | .show_fdinfo = eventfd_show_fdinfo, |
| 300 | #endif |
| 301 | .release = eventfd_release, |
| 302 | .poll = eventfd_poll, |
| 303 | .read_iter = eventfd_read, |
| 304 | .write = eventfd_write, |
| 305 | .llseek = noop_llseek, |
| 306 | }; |
| 307 | |
| 308 | /** |
| 309 | * eventfd_fget - Acquire a reference of an eventfd file descriptor. |
| 310 | * @fd: [in] Eventfd file descriptor. |
| 311 | * |
| 312 | * Returns a pointer to the eventfd file structure in case of success, or the |
| 313 | * following error pointer: |
| 314 | * |
| 315 | * -EBADF : Invalid @fd file descriptor. |
| 316 | * -EINVAL : The @fd file descriptor is not an eventfd file. |
| 317 | */ |
| 318 | struct file *eventfd_fget(int fd) |
| 319 | { |
| 320 | struct file *file; |
| 321 | |
| 322 | file = fget(fd); |
| 323 | if (!file) |
| 324 | return ERR_PTR(-EBADF); |
| 325 | if (file->f_op != &eventfd_fops) { |
| 326 | fput(file); |
| 327 | return ERR_PTR(-EINVAL); |
| 328 | } |
| 329 | |
| 330 | return file; |
| 331 | } |
| 332 | EXPORT_SYMBOL_GPL(eventfd_fget); |
| 333 | |
| 334 | /** |
| 335 | * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context. |
| 336 | * @fd: [in] Eventfd file descriptor. |
| 337 | * |
| 338 | * Returns a pointer to the internal eventfd context, otherwise the error |
| 339 | * pointers returned by the following functions: |
| 340 | * |
| 341 | * eventfd_fget |
| 342 | */ |
| 343 | struct eventfd_ctx *eventfd_ctx_fdget(int fd) |
| 344 | { |
| 345 | struct eventfd_ctx *ctx; |
| 346 | struct fd f = fdget(fd); |
| 347 | if (!f.file) |
| 348 | return ERR_PTR(-EBADF); |
| 349 | ctx = eventfd_ctx_fileget(f.file); |
| 350 | fdput(f); |
| 351 | return ctx; |
| 352 | } |
| 353 | EXPORT_SYMBOL_GPL(eventfd_ctx_fdget); |
| 354 | |
| 355 | /** |
| 356 | * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context. |
| 357 | * @file: [in] Eventfd file pointer. |
| 358 | * |
| 359 | * Returns a pointer to the internal eventfd context, otherwise the error |
| 360 | * pointer: |
| 361 | * |
| 362 | * -EINVAL : The @fd file descriptor is not an eventfd file. |
| 363 | */ |
| 364 | struct eventfd_ctx *eventfd_ctx_fileget(struct file *file) |
| 365 | { |
| 366 | struct eventfd_ctx *ctx; |
| 367 | |
| 368 | if (file->f_op != &eventfd_fops) |
| 369 | return ERR_PTR(-EINVAL); |
| 370 | |
| 371 | ctx = file->private_data; |
| 372 | kref_get(&ctx->kref); |
| 373 | return ctx; |
| 374 | } |
| 375 | EXPORT_SYMBOL_GPL(eventfd_ctx_fileget); |
| 376 | |
| 377 | static int do_eventfd(unsigned int count, int flags) |
| 378 | { |
| 379 | struct eventfd_ctx *ctx; |
| 380 | struct file *file; |
| 381 | int fd; |
| 382 | |
| 383 | /* Check the EFD_* constants for consistency. */ |
| 384 | BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC); |
| 385 | BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK); |
| 386 | |
| 387 | if (flags & ~EFD_FLAGS_SET) |
| 388 | return -EINVAL; |
| 389 | |
| 390 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
| 391 | if (!ctx) |
| 392 | return -ENOMEM; |
| 393 | |
| 394 | kref_init(&ctx->kref); |
| 395 | init_waitqueue_head(&ctx->wqh); |
| 396 | ctx->count = count; |
| 397 | ctx->flags = flags; |
| 398 | ctx->id = ida_alloc(&eventfd_ida, GFP_KERNEL); |
| 399 | |
| 400 | flags &= EFD_SHARED_FCNTL_FLAGS; |
| 401 | flags |= O_RDWR; |
| 402 | fd = get_unused_fd_flags(flags); |
| 403 | if (fd < 0) |
| 404 | goto err; |
| 405 | |
| 406 | file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, flags); |
| 407 | if (IS_ERR(file)) { |
| 408 | put_unused_fd(fd); |
| 409 | fd = PTR_ERR(file); |
| 410 | goto err; |
| 411 | } |
| 412 | |
| 413 | file->f_mode |= FMODE_NOWAIT; |
| 414 | fd_install(fd, file); |
| 415 | return fd; |
| 416 | err: |
| 417 | eventfd_free_ctx(ctx); |
| 418 | return fd; |
| 419 | } |
| 420 | |
| 421 | SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags) |
| 422 | { |
| 423 | return do_eventfd(count, flags); |
| 424 | } |
| 425 | |
| 426 | SYSCALL_DEFINE1(eventfd, unsigned int, count) |
| 427 | { |
| 428 | return do_eventfd(count, 0); |
| 429 | } |
| 430 | |