| 1 | |
| 2 | #include <linux/device.h>
|
| 3 | #include <linux/init.h>
|
| 4 | #include <linux/kernel.h>
|
| 5 | #include <linux/miscdevice.h>
|
| 6 | #include <linux/module.h>
|
| 7 | #include <linux/anon_inodes.h>
|
| 8 | #include <linux/ktime.h>
|
| 9 | #include <linux/hrtimer.h>
|
| 10 | #include <linux/poll.h>
|
| 11 |
|
| 12 | #include <linux/io_uring/cmd.h>
|
| 13 | #include <linux/io_uring_types.h>
|
| 14 | #include <uapi/linux/io_uring/mock_file.h>
|
| 15 |
|
| 16 | struct io_mock_iocb {
|
| 17 | struct kiocb *iocb;
|
| 18 | struct hrtimer timer;
|
| 19 | int res;
|
| 20 | };
|
| 21 |
|
| 22 | struct io_mock_file {
|
| 23 | size_t size;
|
| 24 | u64 rw_delay_ns;
|
| 25 | bool pollable;
|
| 26 | struct wait_queue_head poll_wq;
|
| 27 | };
|
| 28 |
|
| 29 | #define IO_VALID_COPY_CMD_FLAGS IORING_MOCK_COPY_FROM
|
| 30 |
|
| 31 | static int io_copy_regbuf(struct iov_iter *reg_iter, void __user *ubuf)
|
| 32 | {
|
| 33 | size_t ret, copied = 0;
|
| 34 | size_t buflen = PAGE_SIZE;
|
| 35 | void *tmp_buf;
|
| 36 |
|
| 37 | tmp_buf = kzalloc(buflen, GFP_KERNEL);
|
| 38 | if (!tmp_buf)
|
| 39 | return -ENOMEM;
|
| 40 |
|
| 41 | while (iov_iter_count(reg_iter)) {
|
| 42 | size_t len = min(iov_iter_count(reg_iter), buflen);
|
| 43 |
|
| 44 | if (iov_iter_rw(reg_iter) == ITER_SOURCE) {
|
| 45 | ret = copy_from_iter(tmp_buf, len, reg_iter);
|
| 46 | if (ret <= 0)
|
| 47 | break;
|
| 48 | if (copy_to_user(ubuf, tmp_buf, ret))
|
| 49 | break;
|
| 50 | } else {
|
| 51 | if (copy_from_user(tmp_buf, ubuf, len))
|
| 52 | break;
|
| 53 | ret = copy_to_iter(tmp_buf, len, reg_iter);
|
| 54 | if (ret <= 0)
|
| 55 | break;
|
| 56 | }
|
| 57 | ubuf += ret;
|
| 58 | copied += ret;
|
| 59 | }
|
| 60 |
|
| 61 | kfree(tmp_buf);
|
| 62 | return copied;
|
| 63 | }
|
| 64 |
|
| 65 | static int io_cmd_copy_regbuf(struct io_uring_cmd *cmd, unsigned int issue_flags)
|
| 66 | {
|
| 67 | const struct io_uring_sqe *sqe = cmd->sqe;
|
| 68 | const struct iovec __user *iovec;
|
| 69 | unsigned flags, iovec_len;
|
| 70 | struct iov_iter iter;
|
| 71 | void __user *ubuf;
|
| 72 | int dir, ret;
|
| 73 |
|
| 74 | ubuf = u64_to_user_ptr(READ_ONCE(sqe->addr3));
|
| 75 | iovec = u64_to_user_ptr(READ_ONCE(sqe->addr));
|
| 76 | iovec_len = READ_ONCE(sqe->len);
|
| 77 | flags = READ_ONCE(sqe->file_index);
|
| 78 |
|
| 79 | if (unlikely(sqe->ioprio || sqe->__pad1))
|
| 80 | return -EINVAL;
|
| 81 | if (flags & ~IO_VALID_COPY_CMD_FLAGS)
|
| 82 | return -EINVAL;
|
| 83 |
|
| 84 | dir = (flags & IORING_MOCK_COPY_FROM) ? ITER_SOURCE : ITER_DEST;
|
| 85 | ret = io_uring_cmd_import_fixed_vec(cmd, iovec, iovec_len, dir, &iter,
|
| 86 | issue_flags);
|
| 87 | if (ret)
|
| 88 | return ret;
|
| 89 | ret = io_copy_regbuf(&iter, ubuf);
|
| 90 | return ret ? ret : -EFAULT;
|
| 91 | }
|
| 92 |
|
| 93 | static int io_mock_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
|
| 94 | {
|
| 95 | switch (cmd->cmd_op) {
|
| 96 | case IORING_MOCK_CMD_COPY_REGBUF:
|
| 97 | return io_cmd_copy_regbuf(cmd, issue_flags);
|
| 98 | }
|
| 99 | return -ENOTSUPP;
|
| 100 | }
|
| 101 |
|
| 102 | static enum hrtimer_restart io_mock_rw_timer_expired(struct hrtimer *timer)
|
| 103 | {
|
| 104 | struct io_mock_iocb *mio = container_of(timer, struct io_mock_iocb, timer);
|
| 105 | struct kiocb *iocb = mio->iocb;
|
| 106 |
|
| 107 | WRITE_ONCE(iocb->private, NULL);
|
| 108 | iocb->ki_complete(iocb, mio->res);
|
| 109 | kfree(mio);
|
| 110 | return HRTIMER_NORESTART;
|
| 111 | }
|
| 112 |
|
| 113 | static ssize_t io_mock_delay_rw(struct kiocb *iocb, size_t len)
|
| 114 | {
|
| 115 | struct io_mock_file *mf = iocb->ki_filp->private_data;
|
| 116 | struct io_mock_iocb *mio;
|
| 117 |
|
| 118 | mio = kzalloc_obj(*mio);
|
| 119 | if (!mio)
|
| 120 | return -ENOMEM;
|
| 121 |
|
| 122 | mio->iocb = iocb;
|
| 123 | mio->res = len;
|
| 124 | hrtimer_setup(&mio->timer, io_mock_rw_timer_expired,
|
| 125 | CLOCK_MONOTONIC, HRTIMER_MODE_REL);
|
| 126 | hrtimer_start(&mio->timer, ns_to_ktime(mf->rw_delay_ns),
|
| 127 | HRTIMER_MODE_REL);
|
| 128 | return -EIOCBQUEUED;
|
| 129 | }
|
| 130 |
|
| 131 | static ssize_t io_mock_read_iter(struct kiocb *iocb, struct iov_iter *to)
|
| 132 | {
|
| 133 | struct io_mock_file *mf = iocb->ki_filp->private_data;
|
| 134 | size_t len = iov_iter_count(to);
|
| 135 | size_t nr_zeroed;
|
| 136 |
|
| 137 | if (iocb->ki_pos + len > mf->size)
|
| 138 | return -EINVAL;
|
| 139 | nr_zeroed = iov_iter_zero(len, to);
|
| 140 | if (!mf->rw_delay_ns || nr_zeroed != len)
|
| 141 | return nr_zeroed;
|
| 142 |
|
| 143 | return io_mock_delay_rw(iocb, len);
|
| 144 | }
|
| 145 |
|
| 146 | static ssize_t io_mock_write_iter(struct kiocb *iocb, struct iov_iter *from)
|
| 147 | {
|
| 148 | struct io_mock_file *mf = iocb->ki_filp->private_data;
|
| 149 | size_t len = iov_iter_count(from);
|
| 150 |
|
| 151 | if (iocb->ki_pos + len > mf->size)
|
| 152 | return -EINVAL;
|
| 153 | if (!mf->rw_delay_ns) {
|
| 154 | iov_iter_advance(from, len);
|
| 155 | return len;
|
| 156 | }
|
| 157 |
|
| 158 | return io_mock_delay_rw(iocb, len);
|
| 159 | }
|
| 160 |
|
| 161 | static loff_t io_mock_llseek(struct file *file, loff_t offset, int whence)
|
| 162 | {
|
| 163 | struct io_mock_file *mf = file->private_data;
|
| 164 |
|
| 165 | return fixed_size_llseek(file, offset, whence, mf->size);
|
| 166 | }
|
| 167 |
|
| 168 | static __poll_t io_mock_poll(struct file *file, struct poll_table_struct *pt)
|
| 169 | {
|
| 170 | struct io_mock_file *mf = file->private_data;
|
| 171 | __poll_t mask = 0;
|
| 172 |
|
| 173 | poll_wait(file, &mf->poll_wq, pt);
|
| 174 |
|
| 175 | mask |= EPOLLOUT | EPOLLWRNORM;
|
| 176 | mask |= EPOLLIN | EPOLLRDNORM;
|
| 177 | return mask;
|
| 178 | }
|
| 179 |
|
| 180 | static int io_mock_release(struct inode *inode, struct file *file)
|
| 181 | {
|
| 182 | struct io_mock_file *mf = file->private_data;
|
| 183 |
|
| 184 | kfree(mf);
|
| 185 | return 0;
|
| 186 | }
|
| 187 |
|
| 188 | static const struct file_operations io_mock_fops = {
|
| 189 | .owner = THIS_MODULE,
|
| 190 | .release = io_mock_release,
|
| 191 | .uring_cmd = io_mock_cmd,
|
| 192 | .read_iter = io_mock_read_iter,
|
| 193 | .write_iter = io_mock_write_iter,
|
| 194 | .llseek = io_mock_llseek,
|
| 195 | };
|
| 196 |
|
| 197 | static const struct file_operations io_mock_poll_fops = {
|
| 198 | .owner = THIS_MODULE,
|
| 199 | .release = io_mock_release,
|
| 200 | .uring_cmd = io_mock_cmd,
|
| 201 | .read_iter = io_mock_read_iter,
|
| 202 | .write_iter = io_mock_write_iter,
|
| 203 | .llseek = io_mock_llseek,
|
| 204 | .poll = io_mock_poll,
|
| 205 | };
|
| 206 |
|
| 207 | #define IO_VALID_CREATE_FLAGS (IORING_MOCK_CREATE_F_SUPPORT_NOWAIT | \
|
| 208 | IORING_MOCK_CREATE_F_POLL)
|
| 209 |
|
| 210 | static int io_create_mock_file(struct io_uring_cmd *cmd, unsigned int issue_flags)
|
| 211 | {
|
| 212 | const struct file_operations *fops = &io_mock_fops;
|
| 213 | const struct io_uring_sqe *sqe = cmd->sqe;
|
| 214 | struct io_uring_mock_create mc, __user *uarg;
|
| 215 | struct file *file;
|
| 216 | struct io_mock_file *mf __free(kfree) = NULL;
|
| 217 | size_t uarg_size;
|
| 218 |
|
| 219 | |
| 220 | * It's a testing only driver that allows exercising edge cases
|
| 221 | * that wouldn't be possible to hit otherwise.
|
| 222 |
|
| 223 | add_taint(TAINT_TEST, LOCKDEP_STILL_OK);
|
| 224 |
|
| 225 | uarg = u64_to_user_ptr(READ_ONCE(sqe->addr));
|
| 226 | uarg_size = READ_ONCE(sqe->len);
|
| 227 |
|
| 228 | if (sqe->ioprio || sqe->__pad1 || sqe->addr3 || sqe->file_index)
|
| 229 | return -EINVAL;
|
| 230 | if (uarg_size != sizeof(mc))
|
| 231 | return -EINVAL;
|
| 232 |
|
| 233 | memset(&mc, 0, sizeof(mc));
|
| 234 | if (copy_from_user(&mc, uarg, uarg_size))
|
| 235 | return -EFAULT;
|
| 236 | if (!mem_is_zero(mc.__resv, sizeof(mc.__resv)))
|
| 237 | return -EINVAL;
|
| 238 | if (mc.flags & ~IO_VALID_CREATE_FLAGS)
|
| 239 | return -EINVAL;
|
| 240 | if (mc.file_size > SZ_1G)
|
| 241 | return -EINVAL;
|
| 242 | if (mc.rw_delay_ns > NSEC_PER_SEC)
|
| 243 | return -EINVAL;
|
| 244 |
|
| 245 | mf = kzalloc_obj(*mf, GFP_KERNEL_ACCOUNT);
|
| 246 | if (!mf)
|
| 247 | return -ENOMEM;
|
| 248 |
|
| 249 | init_waitqueue_head(&mf->poll_wq);
|
| 250 | mf->size = mc.file_size;
|
| 251 | mf->rw_delay_ns = mc.rw_delay_ns;
|
| 252 | if (mc.flags & IORING_MOCK_CREATE_F_POLL) {
|
| 253 | fops = &io_mock_poll_fops;
|
| 254 | mf->pollable = true;
|
| 255 | }
|
| 256 |
|
| 257 | FD_PREPARE(fdf, O_RDWR | O_CLOEXEC,
|
| 258 | anon_inode_create_getfile("[io_uring_mock]", fops, mf,
|
| 259 | O_RDWR | O_CLOEXEC, NULL));
|
| 260 | if (fdf->fd < 0)
|
| 261 | return fdf->fd;
|
| 262 |
|
| 263 | retain_and_null_ptr(mf);
|
| 264 | file = fdf->file;
|
| 265 | file->f_mode |= FMODE_READ | FMODE_CAN_READ | FMODE_WRITE |
|
| 266 | FMODE_CAN_WRITE | FMODE_LSEEK;
|
| 267 | if (mc.flags & IORING_MOCK_CREATE_F_SUPPORT_NOWAIT)
|
| 268 | file->f_mode |= FMODE_NOWAIT;
|
| 269 |
|
| 270 | mc.out_fd = fdf->fd;
|
| 271 | if (copy_to_user(uarg, &mc, uarg_size))
|
| 272 | return -EFAULT;
|
| 273 |
|
| 274 | fd_publish(fdf);
|
| 275 | return 0;
|
| 276 | }
|
| 277 |
|
| 278 | static int io_probe_mock(struct io_uring_cmd *cmd)
|
| 279 | {
|
| 280 | const struct io_uring_sqe *sqe = cmd->sqe;
|
| 281 | struct io_uring_mock_probe mp, __user *uarg;
|
| 282 | size_t uarg_size;
|
| 283 |
|
| 284 | uarg = u64_to_user_ptr(READ_ONCE(sqe->addr));
|
| 285 | uarg_size = READ_ONCE(sqe->len);
|
| 286 |
|
| 287 | if (sqe->ioprio || sqe->__pad1 || sqe->addr3 || sqe->file_index ||
|
| 288 | uarg_size != sizeof(mp))
|
| 289 | return -EINVAL;
|
| 290 |
|
| 291 | memset(&mp, 0, sizeof(mp));
|
| 292 | if (copy_from_user(&mp, uarg, uarg_size))
|
| 293 | return -EFAULT;
|
| 294 | if (!mem_is_zero(&mp, sizeof(mp)))
|
| 295 | return -EINVAL;
|
| 296 |
|
| 297 | mp.features = IORING_MOCK_FEAT_END;
|
| 298 |
|
| 299 | if (copy_to_user(uarg, &mp, uarg_size))
|
| 300 | return -EFAULT;
|
| 301 | return 0;
|
| 302 | }
|
| 303 |
|
| 304 | static int iou_mock_mgr_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
|
| 305 | {
|
| 306 | if (!capable(CAP_SYS_ADMIN))
|
| 307 | return -EPERM;
|
| 308 |
|
| 309 | switch (cmd->cmd_op) {
|
| 310 | case IORING_MOCK_MGR_CMD_PROBE:
|
| 311 | return io_probe_mock(cmd);
|
| 312 | case IORING_MOCK_MGR_CMD_CREATE:
|
| 313 | return io_create_mock_file(cmd, issue_flags);
|
| 314 | }
|
| 315 | return -EOPNOTSUPP;
|
| 316 | }
|
| 317 |
|
| 318 | static const struct file_operations iou_mock_dev_fops = {
|
| 319 | .owner = THIS_MODULE,
|
| 320 | .uring_cmd = iou_mock_mgr_cmd,
|
| 321 | };
|
| 322 |
|
| 323 | static struct miscdevice iou_mock_miscdev = {
|
| 324 | .minor = MISC_DYNAMIC_MINOR,
|
| 325 | .name = "io_uring_mock",
|
| 326 | .fops = &iou_mock_dev_fops,
|
| 327 | };
|
| 328 |
|
| 329 | static int __init io_mock_init(void)
|
| 330 | {
|
| 331 | int ret;
|
| 332 |
|
| 333 | ret = misc_register(&iou_mock_miscdev);
|
| 334 | if (ret < 0) {
|
| 335 | pr_err("Could not initialize io_uring mock device\n");
|
| 336 | return ret;
|
| 337 | }
|
| 338 | return 0;
|
| 339 | }
|
| 340 |
|
| 341 | static void __exit io_mock_exit(void)
|
| 342 | {
|
| 343 | misc_deregister(&iou_mock_miscdev);
|
| 344 | }
|
| 345 |
|
| 346 | module_init(io_mock_init)
|
| 347 | module_exit(io_mock_exit)
|
| 348 |
|
| 349 | MODULE_AUTHOR("Pavel Begunkov <[email protected]>");
|
| 350 | MODULE_DESCRIPTION("io_uring mock file");
|
| 351 | MODULE_LICENSE("GPL");
|