1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/kernel.h> 3 #include <linux/errno.h> 4 #include <linux/file.h> 5 #include <linux/io_uring/cmd.h> 6 #include <linux/io_uring/net.h> 7 #include <linux/security.h> 8 #include <linux/nospec.h> 9 #include <net/sock.h> 10 11 #include <uapi/linux/io_uring.h> 12 #include <asm/ioctls.h> 13 14 #include "io_uring.h" 15 #include "alloc_cache.h" 16 #include "rsrc.h" 17 #include "uring_cmd.h" 18 19 void io_cmd_cache_free(const void *entry) 20 { 21 struct io_async_cmd *ac = (struct io_async_cmd *)entry; 22 23 io_vec_free(&ac->vec); 24 kfree(ac); 25 } 26 27 static void io_req_uring_cleanup(struct io_kiocb *req, unsigned int issue_flags) 28 { 29 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 30 struct io_async_cmd *ac = req->async_data; 31 struct io_uring_cmd_data *cache = &ac->data; 32 33 if (cache->op_data) { 34 kfree(cache->op_data); 35 cache->op_data = NULL; 36 } 37 38 if (issue_flags & IO_URING_F_UNLOCKED) 39 return; 40 41 io_alloc_cache_vec_kasan(&ac->vec); 42 if (ac->vec.nr > IO_VEC_CACHE_SOFT_CAP) 43 io_vec_free(&ac->vec); 44 45 if (io_alloc_cache_put(&req->ctx->cmd_cache, cache)) { 46 ioucmd->sqe = NULL; 47 req->async_data = NULL; 48 req->flags &= ~(REQ_F_ASYNC_DATA|REQ_F_NEED_CLEANUP); 49 } 50 } 51 52 void io_uring_cmd_cleanup(struct io_kiocb *req) 53 { 54 io_req_uring_cleanup(req, 0); 55 } 56 57 bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx, 58 struct io_uring_task *tctx, bool cancel_all) 59 { 60 struct hlist_node *tmp; 61 struct io_kiocb *req; 62 bool ret = false; 63 64 lockdep_assert_held(&ctx->uring_lock); 65 66 hlist_for_each_entry_safe(req, tmp, &ctx->cancelable_uring_cmd, 67 hash_node) { 68 struct io_uring_cmd *cmd = io_kiocb_to_cmd(req, 69 struct io_uring_cmd); 70 struct file *file = req->file; 71 72 if (!cancel_all && req->tctx != tctx) 73 continue; 74 75 if (cmd->flags & IORING_URING_CMD_CANCELABLE) { 76 file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL | 77 IO_URING_F_COMPLETE_DEFER); 78 ret = true; 79 } 80 } 81 io_submit_flush_completions(ctx); 82 return ret; 83 } 84 85 static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd, 86 unsigned int issue_flags) 87 { 88 struct io_kiocb *req = cmd_to_io_kiocb(cmd); 89 struct io_ring_ctx *ctx = req->ctx; 90 91 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) 92 return; 93 94 cmd->flags &= ~IORING_URING_CMD_CANCELABLE; 95 io_ring_submit_lock(ctx, issue_flags); 96 hlist_del(&req->hash_node); 97 io_ring_submit_unlock(ctx, issue_flags); 98 } 99 100 /* 101 * Mark this command as concelable, then io_uring_try_cancel_uring_cmd() 102 * will try to cancel this issued command by sending ->uring_cmd() with 103 * issue_flags of IO_URING_F_CANCEL. 104 * 105 * The command is guaranteed to not be done when calling ->uring_cmd() 106 * with IO_URING_F_CANCEL, but it is driver's responsibility to deal 107 * with race between io_uring canceling and normal completion. 108 */ 109 void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd, 110 unsigned int issue_flags) 111 { 112 struct io_kiocb *req = cmd_to_io_kiocb(cmd); 113 struct io_ring_ctx *ctx = req->ctx; 114 115 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) { 116 cmd->flags |= IORING_URING_CMD_CANCELABLE; 117 io_ring_submit_lock(ctx, issue_flags); 118 hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd); 119 io_ring_submit_unlock(ctx, issue_flags); 120 } 121 } 122 EXPORT_SYMBOL_GPL(io_uring_cmd_mark_cancelable); 123 124 static void io_uring_cmd_work(struct io_kiocb *req, io_tw_token_t tw) 125 { 126 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 127 unsigned int flags = IO_URING_F_COMPLETE_DEFER; 128 129 if (io_should_terminate_tw()) 130 flags |= IO_URING_F_TASK_DEAD; 131 132 /* task_work executor checks the deffered list completion */ 133 ioucmd->task_work_cb(ioucmd, flags); 134 } 135 136 void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd, 137 void (*task_work_cb)(struct io_uring_cmd *, unsigned), 138 unsigned flags) 139 { 140 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 141 142 ioucmd->task_work_cb = task_work_cb; 143 req->io_task_work.func = io_uring_cmd_work; 144 __io_req_task_work_add(req, flags); 145 } 146 EXPORT_SYMBOL_GPL(__io_uring_cmd_do_in_task); 147 148 static inline void io_req_set_cqe32_extra(struct io_kiocb *req, 149 u64 extra1, u64 extra2) 150 { 151 req->big_cqe.extra1 = extra1; 152 req->big_cqe.extra2 = extra2; 153 } 154 155 /* 156 * Called by consumers of io_uring_cmd, if they originally returned 157 * -EIOCBQUEUED upon receiving the command. 158 */ 159 void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, u64 res2, 160 unsigned issue_flags) 161 { 162 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 163 164 io_uring_cmd_del_cancelable(ioucmd, issue_flags); 165 166 if (ret < 0) 167 req_set_fail(req); 168 169 io_req_set_res(req, ret, 0); 170 if (req->ctx->flags & IORING_SETUP_CQE32) 171 io_req_set_cqe32_extra(req, res2, 0); 172 io_req_uring_cleanup(req, issue_flags); 173 if (req->ctx->flags & IORING_SETUP_IOPOLL) { 174 /* order with io_iopoll_req_issued() checking ->iopoll_complete */ 175 smp_store_release(&req->iopoll_completed, 1); 176 } else if (issue_flags & IO_URING_F_COMPLETE_DEFER) { 177 if (WARN_ON_ONCE(issue_flags & IO_URING_F_UNLOCKED)) 178 return; 179 io_req_complete_defer(req); 180 } else { 181 req->io_task_work.func = io_req_task_complete; 182 io_req_task_work_add(req); 183 } 184 } 185 EXPORT_SYMBOL_GPL(io_uring_cmd_done); 186 187 static int io_uring_cmd_prep_setup(struct io_kiocb *req, 188 const struct io_uring_sqe *sqe) 189 { 190 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 191 struct io_async_cmd *ac; 192 193 /* see io_uring_cmd_get_async_data() */ 194 BUILD_BUG_ON(offsetof(struct io_async_cmd, data) != 0); 195 196 ac = io_uring_alloc_async_data(&req->ctx->cmd_cache, req); 197 if (!ac) 198 return -ENOMEM; 199 ac->data.op_data = NULL; 200 201 /* 202 * Unconditionally cache the SQE for now - this is only needed for 203 * requests that go async, but prep handlers must ensure that any 204 * sqe data is stable beyond prep. Since uring_cmd is special in 205 * that it doesn't read in per-op data, play it safe and ensure that 206 * any SQE data is stable beyond prep. This can later get relaxed. 207 */ 208 memcpy(ac->sqes, sqe, uring_sqe_size(req->ctx)); 209 ioucmd->sqe = ac->sqes; 210 return 0; 211 } 212 213 int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 214 { 215 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 216 217 if (sqe->__pad1) 218 return -EINVAL; 219 220 ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags); 221 if (ioucmd->flags & ~IORING_URING_CMD_MASK) 222 return -EINVAL; 223 224 if (ioucmd->flags & IORING_URING_CMD_FIXED) 225 req->buf_index = READ_ONCE(sqe->buf_index); 226 227 ioucmd->cmd_op = READ_ONCE(sqe->cmd_op); 228 229 return io_uring_cmd_prep_setup(req, sqe); 230 } 231 232 int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags) 233 { 234 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 235 struct io_ring_ctx *ctx = req->ctx; 236 struct file *file = req->file; 237 int ret; 238 239 if (!file->f_op->uring_cmd) 240 return -EOPNOTSUPP; 241 242 ret = security_uring_cmd(ioucmd); 243 if (ret) 244 return ret; 245 246 if (ctx->flags & IORING_SETUP_SQE128) 247 issue_flags |= IO_URING_F_SQE128; 248 if (ctx->flags & IORING_SETUP_CQE32) 249 issue_flags |= IO_URING_F_CQE32; 250 if (io_is_compat(ctx)) 251 issue_flags |= IO_URING_F_COMPAT; 252 if (ctx->flags & IORING_SETUP_IOPOLL) { 253 if (!file->f_op->uring_cmd_iopoll) 254 return -EOPNOTSUPP; 255 issue_flags |= IO_URING_F_IOPOLL; 256 req->iopoll_completed = 0; 257 } 258 259 ret = file->f_op->uring_cmd(ioucmd, issue_flags); 260 if (ret == -EAGAIN || ret == -EIOCBQUEUED) 261 return ret; 262 if (ret < 0) 263 req_set_fail(req); 264 io_req_uring_cleanup(req, issue_flags); 265 io_req_set_res(req, ret, 0); 266 return IOU_OK; 267 } 268 269 int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw, 270 struct iov_iter *iter, 271 struct io_uring_cmd *ioucmd, 272 unsigned int issue_flags) 273 { 274 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 275 276 return io_import_reg_buf(req, iter, ubuf, len, rw, issue_flags); 277 } 278 EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed); 279 280 int io_uring_cmd_import_fixed_vec(struct io_uring_cmd *ioucmd, 281 const struct iovec __user *uvec, 282 size_t uvec_segs, 283 int ddir, struct iov_iter *iter, 284 unsigned issue_flags) 285 { 286 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 287 struct io_async_cmd *ac = req->async_data; 288 int ret; 289 290 ret = io_prep_reg_iovec(req, &ac->vec, uvec, uvec_segs); 291 if (ret) 292 return ret; 293 294 return io_import_reg_vec(ddir, iter, req, &ac->vec, uvec_segs, 295 issue_flags); 296 } 297 EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed_vec); 298 299 void io_uring_cmd_issue_blocking(struct io_uring_cmd *ioucmd) 300 { 301 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 302 303 io_req_queue_iowq(req); 304 } 305 306 static inline int io_uring_cmd_getsockopt(struct socket *sock, 307 struct io_uring_cmd *cmd, 308 unsigned int issue_flags) 309 { 310 const struct io_uring_sqe *sqe = cmd->sqe; 311 bool compat = !!(issue_flags & IO_URING_F_COMPAT); 312 int optlen, optname, level, err; 313 void __user *optval; 314 315 level = READ_ONCE(sqe->level); 316 if (level != SOL_SOCKET) 317 return -EOPNOTSUPP; 318 319 optval = u64_to_user_ptr(READ_ONCE(sqe->optval)); 320 optname = READ_ONCE(sqe->optname); 321 optlen = READ_ONCE(sqe->optlen); 322 323 err = do_sock_getsockopt(sock, compat, level, optname, 324 USER_SOCKPTR(optval), 325 KERNEL_SOCKPTR(&optlen)); 326 if (err) 327 return err; 328 329 /* On success, return optlen */ 330 return optlen; 331 } 332 333 static inline int io_uring_cmd_setsockopt(struct socket *sock, 334 struct io_uring_cmd *cmd, 335 unsigned int issue_flags) 336 { 337 const struct io_uring_sqe *sqe = cmd->sqe; 338 bool compat = !!(issue_flags & IO_URING_F_COMPAT); 339 int optname, optlen, level; 340 void __user *optval; 341 sockptr_t optval_s; 342 343 optval = u64_to_user_ptr(READ_ONCE(sqe->optval)); 344 optname = READ_ONCE(sqe->optname); 345 optlen = READ_ONCE(sqe->optlen); 346 level = READ_ONCE(sqe->level); 347 optval_s = USER_SOCKPTR(optval); 348 349 return do_sock_setsockopt(sock, compat, level, optname, optval_s, 350 optlen); 351 } 352 353 #if defined(CONFIG_NET) 354 int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags) 355 { 356 struct socket *sock = cmd->file->private_data; 357 struct sock *sk = sock->sk; 358 struct proto *prot = READ_ONCE(sk->sk_prot); 359 int ret, arg = 0; 360 361 if (!prot || !prot->ioctl) 362 return -EOPNOTSUPP; 363 364 switch (cmd->cmd_op) { 365 case SOCKET_URING_OP_SIOCINQ: 366 ret = prot->ioctl(sk, SIOCINQ, &arg); 367 if (ret) 368 return ret; 369 return arg; 370 case SOCKET_URING_OP_SIOCOUTQ: 371 ret = prot->ioctl(sk, SIOCOUTQ, &arg); 372 if (ret) 373 return ret; 374 return arg; 375 case SOCKET_URING_OP_GETSOCKOPT: 376 return io_uring_cmd_getsockopt(sock, cmd, issue_flags); 377 case SOCKET_URING_OP_SETSOCKOPT: 378 return io_uring_cmd_setsockopt(sock, cmd, issue_flags); 379 default: 380 return -EOPNOTSUPP; 381 } 382 } 383 EXPORT_SYMBOL_GPL(io_uring_cmd_sock); 384 #endif 385