| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | #include <linux/kernel.h> |
| 3 | #include <linux/errno.h> |
| 4 | #include <linux/fs.h> |
| 5 | #include <linux/file.h> |
| 6 | #include <linux/fdtable.h> |
| 7 | #include <linux/fsnotify.h> |
| 8 | #include <linux/namei.h> |
| 9 | #include <linux/pipe_fs_i.h> |
| 10 | #include <linux/watch_queue.h> |
| 11 | #include <linux/io_uring.h> |
| 12 | |
| 13 | #include <uapi/linux/io_uring.h> |
| 14 | |
| 15 | #include "../fs/internal.h" |
| 16 | |
| 17 | #include "filetable.h" |
| 18 | #include "io_uring.h" |
| 19 | #include "rsrc.h" |
| 20 | #include "openclose.h" |
| 21 | |
| 22 | struct io_open { |
| 23 | struct file *file; |
| 24 | int dfd; |
| 25 | u32 file_slot; |
| 26 | struct filename *filename; |
| 27 | struct open_how how; |
| 28 | unsigned long nofile; |
| 29 | }; |
| 30 | |
| 31 | struct io_close { |
| 32 | struct file *file; |
| 33 | int fd; |
| 34 | u32 file_slot; |
| 35 | }; |
| 36 | |
| 37 | struct io_fixed_install { |
| 38 | struct file *file; |
| 39 | unsigned int o_flags; |
| 40 | }; |
| 41 | |
| 42 | static bool io_openat_force_async(struct io_open *open) |
| 43 | { |
| 44 | /* |
| 45 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, |
| 46 | * it'll always -EAGAIN. Note that we test for __O_TMPFILE because |
| 47 | * O_TMPFILE includes O_DIRECTORY, which isn't a flag we need to force |
| 48 | * async for. |
| 49 | */ |
| 50 | return open->how.flags & (O_TRUNC | O_CREAT | __O_TMPFILE); |
| 51 | } |
| 52 | |
| 53 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 54 | { |
| 55 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 56 | const char __user *fname; |
| 57 | int ret; |
| 58 | |
| 59 | if (unlikely(sqe->buf_index)) |
| 60 | return -EINVAL; |
| 61 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
| 62 | return -EBADF; |
| 63 | |
| 64 | /* open.how should be already initialised */ |
| 65 | if (!(open->how.flags & O_PATH) && force_o_largefile()) |
| 66 | open->how.flags |= O_LARGEFILE; |
| 67 | |
| 68 | open->dfd = READ_ONCE(sqe->fd); |
| 69 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
| 70 | open->filename = getname(name: fname); |
| 71 | if (IS_ERR(ptr: open->filename)) { |
| 72 | ret = PTR_ERR(ptr: open->filename); |
| 73 | open->filename = NULL; |
| 74 | return ret; |
| 75 | } |
| 76 | |
| 77 | open->file_slot = READ_ONCE(sqe->file_index); |
| 78 | if (open->file_slot && (open->how.flags & O_CLOEXEC)) |
| 79 | return -EINVAL; |
| 80 | |
| 81 | open->nofile = rlimit(RLIMIT_NOFILE); |
| 82 | req->flags |= REQ_F_NEED_CLEANUP; |
| 83 | if (io_openat_force_async(open)) |
| 84 | req->flags |= REQ_F_FORCE_ASYNC; |
| 85 | return 0; |
| 86 | } |
| 87 | |
| 88 | int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 89 | { |
| 90 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 91 | u64 mode = READ_ONCE(sqe->len); |
| 92 | u64 flags = READ_ONCE(sqe->open_flags); |
| 93 | |
| 94 | open->how = build_open_how(flags, mode); |
| 95 | return __io_openat_prep(req, sqe); |
| 96 | } |
| 97 | |
| 98 | int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 99 | { |
| 100 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 101 | struct open_how __user *how; |
| 102 | size_t len; |
| 103 | int ret; |
| 104 | |
| 105 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
| 106 | len = READ_ONCE(sqe->len); |
| 107 | if (len < OPEN_HOW_SIZE_VER0) |
| 108 | return -EINVAL; |
| 109 | |
| 110 | ret = copy_struct_from_user(dst: &open->how, ksize: sizeof(open->how), src: how, usize: len); |
| 111 | if (ret) |
| 112 | return ret; |
| 113 | |
| 114 | return __io_openat_prep(req, sqe); |
| 115 | } |
| 116 | |
| 117 | int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
| 118 | { |
| 119 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 120 | struct open_flags op; |
| 121 | struct file *file; |
| 122 | bool resolve_nonblock, nonblock_set; |
| 123 | bool fixed = !!open->file_slot; |
| 124 | int ret; |
| 125 | |
| 126 | ret = build_open_flags(how: &open->how, op: &op); |
| 127 | if (ret) |
| 128 | goto err; |
| 129 | nonblock_set = op.open_flag & O_NONBLOCK; |
| 130 | resolve_nonblock = open->how.resolve & RESOLVE_CACHED; |
| 131 | if (issue_flags & IO_URING_F_NONBLOCK) { |
| 132 | WARN_ON_ONCE(io_openat_force_async(open)); |
| 133 | op.lookup_flags |= LOOKUP_CACHED; |
| 134 | op.open_flag |= O_NONBLOCK; |
| 135 | } |
| 136 | |
| 137 | if (!fixed) { |
| 138 | ret = __get_unused_fd_flags(flags: open->how.flags, nofile: open->nofile); |
| 139 | if (ret < 0) |
| 140 | goto err; |
| 141 | } |
| 142 | |
| 143 | file = do_filp_open(dfd: open->dfd, pathname: open->filename, op: &op); |
| 144 | if (IS_ERR(ptr: file)) { |
| 145 | /* |
| 146 | * We could hang on to this 'fd' on retrying, but seems like |
| 147 | * marginal gain for something that is now known to be a slower |
| 148 | * path. So just put it, and we'll get a new one when we retry. |
| 149 | */ |
| 150 | if (!fixed) |
| 151 | put_unused_fd(fd: ret); |
| 152 | |
| 153 | ret = PTR_ERR(ptr: file); |
| 154 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
| 155 | if (ret == -EAGAIN && |
| 156 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) |
| 157 | return -EAGAIN; |
| 158 | goto err; |
| 159 | } |
| 160 | |
| 161 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
| 162 | file->f_flags &= ~O_NONBLOCK; |
| 163 | |
| 164 | if (!fixed) |
| 165 | fd_install(fd: ret, file); |
| 166 | else |
| 167 | ret = io_fixed_fd_install(req, issue_flags, file, |
| 168 | file_slot: open->file_slot); |
| 169 | err: |
| 170 | putname(name: open->filename); |
| 171 | req->flags &= ~REQ_F_NEED_CLEANUP; |
| 172 | if (ret < 0) |
| 173 | req_set_fail(req); |
| 174 | io_req_set_res(req, res: ret, cflags: 0); |
| 175 | return IOU_COMPLETE; |
| 176 | } |
| 177 | |
| 178 | int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
| 179 | { |
| 180 | return io_openat2(req, issue_flags); |
| 181 | } |
| 182 | |
| 183 | void io_open_cleanup(struct io_kiocb *req) |
| 184 | { |
| 185 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 186 | |
| 187 | if (open->filename) |
| 188 | putname(name: open->filename); |
| 189 | } |
| 190 | |
| 191 | int __io_close_fixed(struct io_ring_ctx *ctx, unsigned int issue_flags, |
| 192 | unsigned int offset) |
| 193 | { |
| 194 | int ret; |
| 195 | |
| 196 | io_ring_submit_lock(ctx, issue_flags); |
| 197 | ret = io_fixed_fd_remove(ctx, offset); |
| 198 | io_ring_submit_unlock(ctx, issue_flags); |
| 199 | |
| 200 | return ret; |
| 201 | } |
| 202 | |
| 203 | static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) |
| 204 | { |
| 205 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 206 | |
| 207 | return __io_close_fixed(ctx: req->ctx, issue_flags, offset: close->file_slot - 1); |
| 208 | } |
| 209 | |
| 210 | int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 211 | { |
| 212 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 213 | |
| 214 | if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) |
| 215 | return -EINVAL; |
| 216 | if (req->flags & REQ_F_FIXED_FILE) |
| 217 | return -EBADF; |
| 218 | |
| 219 | close->fd = READ_ONCE(sqe->fd); |
| 220 | close->file_slot = READ_ONCE(sqe->file_index); |
| 221 | if (close->file_slot && close->fd) |
| 222 | return -EINVAL; |
| 223 | |
| 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | int io_close(struct io_kiocb *req, unsigned int issue_flags) |
| 228 | { |
| 229 | struct files_struct *files = current->files; |
| 230 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 231 | struct file *file; |
| 232 | int ret = -EBADF; |
| 233 | |
| 234 | if (close->file_slot) { |
| 235 | ret = io_close_fixed(req, issue_flags); |
| 236 | goto err; |
| 237 | } |
| 238 | |
| 239 | spin_lock(lock: &files->file_lock); |
| 240 | file = files_lookup_fd_locked(files, fd: close->fd); |
| 241 | if (!file || io_is_uring_fops(file)) { |
| 242 | spin_unlock(lock: &files->file_lock); |
| 243 | goto err; |
| 244 | } |
| 245 | |
| 246 | /* if the file has a flush method, be safe and punt to async */ |
| 247 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
| 248 | spin_unlock(lock: &files->file_lock); |
| 249 | return -EAGAIN; |
| 250 | } |
| 251 | |
| 252 | file = file_close_fd_locked(files, fd: close->fd); |
| 253 | spin_unlock(lock: &files->file_lock); |
| 254 | if (!file) |
| 255 | goto err; |
| 256 | |
| 257 | /* No ->flush() or already async, safely close from here */ |
| 258 | ret = filp_close(file, current->files); |
| 259 | err: |
| 260 | if (ret < 0) |
| 261 | req_set_fail(req); |
| 262 | io_req_set_res(req, res: ret, cflags: 0); |
| 263 | return IOU_COMPLETE; |
| 264 | } |
| 265 | |
| 266 | int io_install_fixed_fd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 267 | { |
| 268 | struct io_fixed_install *ifi; |
| 269 | unsigned int flags; |
| 270 | |
| 271 | if (sqe->off || sqe->addr || sqe->len || sqe->buf_index || |
| 272 | sqe->splice_fd_in || sqe->addr3) |
| 273 | return -EINVAL; |
| 274 | |
| 275 | /* must be a fixed file */ |
| 276 | if (!(req->flags & REQ_F_FIXED_FILE)) |
| 277 | return -EBADF; |
| 278 | |
| 279 | flags = READ_ONCE(sqe->install_fd_flags); |
| 280 | if (flags & ~IORING_FIXED_FD_NO_CLOEXEC) |
| 281 | return -EINVAL; |
| 282 | |
| 283 | /* ensure the task's creds are used when installing/receiving fds */ |
| 284 | if (req->flags & REQ_F_CREDS) |
| 285 | return -EPERM; |
| 286 | |
| 287 | /* default to O_CLOEXEC, disable if IORING_FIXED_FD_NO_CLOEXEC is set */ |
| 288 | ifi = io_kiocb_to_cmd(req, struct io_fixed_install); |
| 289 | ifi->o_flags = O_CLOEXEC; |
| 290 | if (flags & IORING_FIXED_FD_NO_CLOEXEC) |
| 291 | ifi->o_flags = 0; |
| 292 | |
| 293 | return 0; |
| 294 | } |
| 295 | |
| 296 | int io_install_fixed_fd(struct io_kiocb *req, unsigned int issue_flags) |
| 297 | { |
| 298 | struct io_fixed_install *ifi; |
| 299 | int ret; |
| 300 | |
| 301 | ifi = io_kiocb_to_cmd(req, struct io_fixed_install); |
| 302 | ret = receive_fd(file: req->file, NULL, o_flags: ifi->o_flags); |
| 303 | if (ret < 0) |
| 304 | req_set_fail(req); |
| 305 | io_req_set_res(req, res: ret, cflags: 0); |
| 306 | return IOU_COMPLETE; |
| 307 | } |
| 308 | |
| 309 | struct io_pipe { |
| 310 | struct file *file; |
| 311 | int __user *fds; |
| 312 | int flags; |
| 313 | int file_slot; |
| 314 | unsigned long nofile; |
| 315 | }; |
| 316 | |
| 317 | int io_pipe_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 318 | { |
| 319 | struct io_pipe *p = io_kiocb_to_cmd(req, struct io_pipe); |
| 320 | |
| 321 | if (sqe->fd || sqe->off || sqe->addr3) |
| 322 | return -EINVAL; |
| 323 | |
| 324 | p->fds = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
| 325 | p->flags = READ_ONCE(sqe->pipe_flags); |
| 326 | if (p->flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT | O_NOTIFICATION_PIPE)) |
| 327 | return -EINVAL; |
| 328 | |
| 329 | p->file_slot = READ_ONCE(sqe->file_index); |
| 330 | p->nofile = rlimit(RLIMIT_NOFILE); |
| 331 | return 0; |
| 332 | } |
| 333 | |
| 334 | static int io_pipe_fixed(struct io_kiocb *req, struct file **files, |
| 335 | unsigned int issue_flags) |
| 336 | { |
| 337 | struct io_pipe *p = io_kiocb_to_cmd(req, struct io_pipe); |
| 338 | struct io_ring_ctx *ctx = req->ctx; |
| 339 | int ret, fds[2] = { -1, -1 }; |
| 340 | int slot = p->file_slot; |
| 341 | |
| 342 | if (p->flags & O_CLOEXEC) |
| 343 | return -EINVAL; |
| 344 | |
| 345 | io_ring_submit_lock(ctx, issue_flags); |
| 346 | |
| 347 | ret = __io_fixed_fd_install(ctx, file: files[0], file_slot: slot); |
| 348 | if (ret < 0) |
| 349 | goto err; |
| 350 | fds[0] = ret; |
| 351 | files[0] = NULL; |
| 352 | |
| 353 | /* |
| 354 | * If a specific slot is given, next one will be used for |
| 355 | * the write side. |
| 356 | */ |
| 357 | if (slot != IORING_FILE_INDEX_ALLOC) |
| 358 | slot++; |
| 359 | |
| 360 | ret = __io_fixed_fd_install(ctx, file: files[1], file_slot: slot); |
| 361 | if (ret < 0) |
| 362 | goto err; |
| 363 | fds[1] = ret; |
| 364 | files[1] = NULL; |
| 365 | |
| 366 | io_ring_submit_unlock(ctx, issue_flags); |
| 367 | |
| 368 | if (!copy_to_user(to: p->fds, from: fds, n: sizeof(fds))) |
| 369 | return 0; |
| 370 | |
| 371 | ret = -EFAULT; |
| 372 | io_ring_submit_lock(ctx, issue_flags); |
| 373 | err: |
| 374 | if (fds[0] != -1) |
| 375 | io_fixed_fd_remove(ctx, offset: fds[0]); |
| 376 | if (fds[1] != -1) |
| 377 | io_fixed_fd_remove(ctx, offset: fds[1]); |
| 378 | io_ring_submit_unlock(ctx, issue_flags); |
| 379 | return ret; |
| 380 | } |
| 381 | |
| 382 | static int io_pipe_fd(struct io_kiocb *req, struct file **files) |
| 383 | { |
| 384 | struct io_pipe *p = io_kiocb_to_cmd(req, struct io_pipe); |
| 385 | int ret, fds[2] = { -1, -1 }; |
| 386 | |
| 387 | ret = __get_unused_fd_flags(flags: p->flags, nofile: p->nofile); |
| 388 | if (ret < 0) |
| 389 | goto err; |
| 390 | fds[0] = ret; |
| 391 | |
| 392 | ret = __get_unused_fd_flags(flags: p->flags, nofile: p->nofile); |
| 393 | if (ret < 0) |
| 394 | goto err; |
| 395 | fds[1] = ret; |
| 396 | |
| 397 | if (!copy_to_user(to: p->fds, from: fds, n: sizeof(fds))) { |
| 398 | fd_install(fd: fds[0], file: files[0]); |
| 399 | fd_install(fd: fds[1], file: files[1]); |
| 400 | return 0; |
| 401 | } |
| 402 | ret = -EFAULT; |
| 403 | err: |
| 404 | if (fds[0] != -1) |
| 405 | put_unused_fd(fd: fds[0]); |
| 406 | if (fds[1] != -1) |
| 407 | put_unused_fd(fd: fds[1]); |
| 408 | return ret; |
| 409 | } |
| 410 | |
| 411 | int io_pipe(struct io_kiocb *req, unsigned int issue_flags) |
| 412 | { |
| 413 | struct io_pipe *p = io_kiocb_to_cmd(req, struct io_pipe); |
| 414 | struct file *files[2]; |
| 415 | int ret; |
| 416 | |
| 417 | ret = create_pipe_files(files, p->flags); |
| 418 | if (ret) |
| 419 | return ret; |
| 420 | |
| 421 | if (!!p->file_slot) |
| 422 | ret = io_pipe_fixed(req, files, issue_flags); |
| 423 | else |
| 424 | ret = io_pipe_fd(req, files); |
| 425 | |
| 426 | io_req_set_res(req, res: ret, cflags: 0); |
| 427 | if (!ret) |
| 428 | return IOU_COMPLETE; |
| 429 | |
| 430 | req_set_fail(req); |
| 431 | if (files[0]) |
| 432 | fput(files[0]); |
| 433 | if (files[1]) |
| 434 | fput(files[1]); |
| 435 | return ret; |
| 436 | } |
| 437 | |