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 static struct uring_cache *io_uring_async_get(struct io_kiocb *req) 20 { 21 struct io_ring_ctx *ctx = req->ctx; 22 struct uring_cache *cache; 23 24 cache = io_alloc_cache_get(&ctx->uring_cache); 25 if (cache) { 26 req->flags |= REQ_F_ASYNC_DATA; 27 req->async_data = cache; 28 return cache; 29 } 30 if (!io_alloc_async_data(req)) 31 return req->async_data; 32 return NULL; 33 } 34 35 static void io_req_uring_cleanup(struct io_kiocb *req, unsigned int issue_flags) 36 { 37 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 38 struct uring_cache *cache = req->async_data; 39 40 if (issue_flags & IO_URING_F_UNLOCKED) 41 return; 42 if (io_alloc_cache_put(&req->ctx->uring_cache, cache)) { 43 ioucmd->sqe = NULL; 44 req->async_data = NULL; 45 req->flags &= ~REQ_F_ASYNC_DATA; 46 } 47 } 48 49 bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx, 50 struct io_uring_task *tctx, bool cancel_all) 51 { 52 struct hlist_node *tmp; 53 struct io_kiocb *req; 54 bool ret = false; 55 56 lockdep_assert_held(&ctx->uring_lock); 57 58 hlist_for_each_entry_safe(req, tmp, &ctx->cancelable_uring_cmd, 59 hash_node) { 60 struct io_uring_cmd *cmd = io_kiocb_to_cmd(req, 61 struct io_uring_cmd); 62 struct file *file = req->file; 63 64 if (!cancel_all && req->tctx != tctx) 65 continue; 66 67 if (cmd->flags & IORING_URING_CMD_CANCELABLE) { 68 /* ->sqe isn't available if no async data */ 69 if (!req_has_async_data(req)) 70 cmd->sqe = NULL; 71 file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL | 72 IO_URING_F_COMPLETE_DEFER); 73 ret = true; 74 } 75 } 76 io_submit_flush_completions(ctx); 77 return ret; 78 } 79 80 static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd, 81 unsigned int issue_flags) 82 { 83 struct io_kiocb *req = cmd_to_io_kiocb(cmd); 84 struct io_ring_ctx *ctx = req->ctx; 85 86 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) 87 return; 88 89 cmd->flags &= ~IORING_URING_CMD_CANCELABLE; 90 io_ring_submit_lock(ctx, issue_flags); 91 hlist_del(&req->hash_node); 92 io_ring_submit_unlock(ctx, issue_flags); 93 } 94 95 /* 96 * Mark this command as concelable, then io_uring_try_cancel_uring_cmd() 97 * will try to cancel this issued command by sending ->uring_cmd() with 98 * issue_flags of IO_URING_F_CANCEL. 99 * 100 * The command is guaranteed to not be done when calling ->uring_cmd() 101 * with IO_URING_F_CANCEL, but it is driver's responsibility to deal 102 * with race between io_uring canceling and normal completion. 103 */ 104 void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd, 105 unsigned int issue_flags) 106 { 107 struct io_kiocb *req = cmd_to_io_kiocb(cmd); 108 struct io_ring_ctx *ctx = req->ctx; 109 110 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) { 111 cmd->flags |= IORING_URING_CMD_CANCELABLE; 112 io_ring_submit_lock(ctx, issue_flags); 113 hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd); 114 io_ring_submit_unlock(ctx, issue_flags); 115 } 116 } 117 EXPORT_SYMBOL_GPL(io_uring_cmd_mark_cancelable); 118 119 static void io_uring_cmd_work(struct io_kiocb *req, struct io_tw_state *ts) 120 { 121 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 122 unsigned int flags = IO_URING_F_COMPLETE_DEFER; 123 124 if (current->flags & (PF_EXITING | PF_KTHREAD)) 125 flags |= IO_URING_F_TASK_DEAD; 126 127 /* task_work executor checks the deffered list completion */ 128 ioucmd->task_work_cb(ioucmd, flags); 129 } 130 131 void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd, 132 void (*task_work_cb)(struct io_uring_cmd *, unsigned), 133 unsigned flags) 134 { 135 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 136 137 ioucmd->task_work_cb = task_work_cb; 138 req->io_task_work.func = io_uring_cmd_work; 139 __io_req_task_work_add(req, flags); 140 } 141 EXPORT_SYMBOL_GPL(__io_uring_cmd_do_in_task); 142 143 static inline void io_req_set_cqe32_extra(struct io_kiocb *req, 144 u64 extra1, u64 extra2) 145 { 146 req->big_cqe.extra1 = extra1; 147 req->big_cqe.extra2 = extra2; 148 } 149 150 /* 151 * Called by consumers of io_uring_cmd, if they originally returned 152 * -EIOCBQUEUED upon receiving the command. 153 */ 154 void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, u64 res2, 155 unsigned issue_flags) 156 { 157 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 158 159 io_uring_cmd_del_cancelable(ioucmd, issue_flags); 160 161 if (ret < 0) 162 req_set_fail(req); 163 164 io_req_set_res(req, ret, 0); 165 if (req->ctx->flags & IORING_SETUP_CQE32) 166 io_req_set_cqe32_extra(req, res2, 0); 167 io_req_uring_cleanup(req, issue_flags); 168 if (req->ctx->flags & IORING_SETUP_IOPOLL) { 169 /* order with io_iopoll_req_issued() checking ->iopoll_complete */ 170 smp_store_release(&req->iopoll_completed, 1); 171 } else if (issue_flags & IO_URING_F_COMPLETE_DEFER) { 172 if (WARN_ON_ONCE(issue_flags & IO_URING_F_UNLOCKED)) 173 return; 174 io_req_complete_defer(req); 175 } else { 176 req->io_task_work.func = io_req_task_complete; 177 io_req_task_work_add(req); 178 } 179 } 180 EXPORT_SYMBOL_GPL(io_uring_cmd_done); 181 182 static int io_uring_cmd_prep_setup(struct io_kiocb *req, 183 const struct io_uring_sqe *sqe) 184 { 185 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 186 struct uring_cache *cache; 187 188 cache = io_uring_async_get(req); 189 if (unlikely(!cache)) 190 return -ENOMEM; 191 192 if (!(req->flags & REQ_F_FORCE_ASYNC)) { 193 /* defer memcpy until we need it */ 194 ioucmd->sqe = sqe; 195 return 0; 196 } 197 198 memcpy(req->async_data, sqe, uring_sqe_size(req->ctx)); 199 ioucmd->sqe = req->async_data; 200 return 0; 201 } 202 203 int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) 204 { 205 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 206 207 if (sqe->__pad1) 208 return -EINVAL; 209 210 ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags); 211 if (ioucmd->flags & ~IORING_URING_CMD_MASK) 212 return -EINVAL; 213 214 if (ioucmd->flags & IORING_URING_CMD_FIXED) { 215 struct io_ring_ctx *ctx = req->ctx; 216 struct io_rsrc_node *node; 217 u16 index = READ_ONCE(sqe->buf_index); 218 219 node = io_rsrc_node_lookup(&ctx->buf_table, index); 220 if (unlikely(!node)) 221 return -EFAULT; 222 /* 223 * Pi node upfront, prior to io_uring_cmd_import_fixed() 224 * being called. This prevents destruction of the mapped buffer 225 * we'll need at actual import time. 226 */ 227 io_req_assign_buf_node(req, node); 228 } 229 ioucmd->cmd_op = READ_ONCE(sqe->cmd_op); 230 231 return io_uring_cmd_prep_setup(req, sqe); 232 } 233 234 int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags) 235 { 236 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd); 237 struct io_ring_ctx *ctx = req->ctx; 238 struct file *file = req->file; 239 int ret; 240 241 if (!file->f_op->uring_cmd) 242 return -EOPNOTSUPP; 243 244 ret = security_uring_cmd(ioucmd); 245 if (ret) 246 return ret; 247 248 if (ctx->flags & IORING_SETUP_SQE128) 249 issue_flags |= IO_URING_F_SQE128; 250 if (ctx->flags & IORING_SETUP_CQE32) 251 issue_flags |= IO_URING_F_CQE32; 252 if (ctx->compat) 253 issue_flags |= IO_URING_F_COMPAT; 254 if (ctx->flags & IORING_SETUP_IOPOLL) { 255 if (!file->f_op->uring_cmd_iopoll) 256 return -EOPNOTSUPP; 257 issue_flags |= IO_URING_F_IOPOLL; 258 req->iopoll_completed = 0; 259 } 260 261 ret = file->f_op->uring_cmd(ioucmd, issue_flags); 262 if (ret == -EAGAIN) { 263 struct uring_cache *cache = req->async_data; 264 265 if (ioucmd->sqe != (void *) cache) 266 memcpy(cache, ioucmd->sqe, uring_sqe_size(req->ctx)); 267 return -EAGAIN; 268 } else if (ret == -EIOCBQUEUED) { 269 return -EIOCBQUEUED; 270 } 271 272 if (ret < 0) 273 req_set_fail(req); 274 io_req_uring_cleanup(req, issue_flags); 275 io_req_set_res(req, ret, 0); 276 return IOU_OK; 277 } 278 279 int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw, 280 struct iov_iter *iter, void *ioucmd) 281 { 282 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 283 struct io_rsrc_node *node = req->buf_node; 284 285 /* Must have had rsrc_node assigned at prep time */ 286 if (node) 287 return io_import_fixed(rw, iter, node->buf, ubuf, len); 288 289 return -EFAULT; 290 } 291 EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed); 292 293 void io_uring_cmd_issue_blocking(struct io_uring_cmd *ioucmd) 294 { 295 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd); 296 297 io_req_queue_iowq(req); 298 } 299 300 static inline int io_uring_cmd_getsockopt(struct socket *sock, 301 struct io_uring_cmd *cmd, 302 unsigned int issue_flags) 303 { 304 bool compat = !!(issue_flags & IO_URING_F_COMPAT); 305 int optlen, optname, level, err; 306 void __user *optval; 307 308 level = READ_ONCE(cmd->sqe->level); 309 if (level != SOL_SOCKET) 310 return -EOPNOTSUPP; 311 312 optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval)); 313 optname = READ_ONCE(cmd->sqe->optname); 314 optlen = READ_ONCE(cmd->sqe->optlen); 315 316 err = do_sock_getsockopt(sock, compat, level, optname, 317 USER_SOCKPTR(optval), 318 KERNEL_SOCKPTR(&optlen)); 319 if (err) 320 return err; 321 322 /* On success, return optlen */ 323 return optlen; 324 } 325 326 static inline int io_uring_cmd_setsockopt(struct socket *sock, 327 struct io_uring_cmd *cmd, 328 unsigned int issue_flags) 329 { 330 bool compat = !!(issue_flags & IO_URING_F_COMPAT); 331 int optname, optlen, level; 332 void __user *optval; 333 sockptr_t optval_s; 334 335 optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval)); 336 optname = READ_ONCE(cmd->sqe->optname); 337 optlen = READ_ONCE(cmd->sqe->optlen); 338 level = READ_ONCE(cmd->sqe->level); 339 optval_s = USER_SOCKPTR(optval); 340 341 return do_sock_setsockopt(sock, compat, level, optname, optval_s, 342 optlen); 343 } 344 345 #if defined(CONFIG_NET) 346 int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags) 347 { 348 struct socket *sock = cmd->file->private_data; 349 struct sock *sk = sock->sk; 350 struct proto *prot = READ_ONCE(sk->sk_prot); 351 int ret, arg = 0; 352 353 if (!prot || !prot->ioctl) 354 return -EOPNOTSUPP; 355 356 switch (cmd->sqe->cmd_op) { 357 case SOCKET_URING_OP_SIOCINQ: 358 ret = prot->ioctl(sk, SIOCINQ, &arg); 359 if (ret) 360 return ret; 361 return arg; 362 case SOCKET_URING_OP_SIOCOUTQ: 363 ret = prot->ioctl(sk, SIOCOUTQ, &arg); 364 if (ret) 365 return ret; 366 return arg; 367 case SOCKET_URING_OP_GETSOCKOPT: 368 return io_uring_cmd_getsockopt(sock, cmd, issue_flags); 369 case SOCKET_URING_OP_SETSOCKOPT: 370 return io_uring_cmd_setsockopt(sock, cmd, issue_flags); 371 default: 372 return -EOPNOTSUPP; 373 } 374 } 375 EXPORT_SYMBOL_GPL(io_uring_cmd_sock); 376 #endif 377