| 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/io_uring.h> |
| 10 | |
| 11 | #include <uapi/linux/io_uring.h> |
| 12 | |
| 13 | #include "../fs/internal.h" |
| 14 | |
| 15 | #include "io_uring.h" |
| 16 | #include "rsrc.h" |
| 17 | #include "openclose.h" |
| 18 | |
| 19 | struct io_open { |
| 20 | struct file *file; |
| 21 | int dfd; |
| 22 | u32 file_slot; |
| 23 | struct filename *filename; |
| 24 | struct open_how how; |
| 25 | unsigned long nofile; |
| 26 | }; |
| 27 | |
| 28 | struct io_close { |
| 29 | struct file *file; |
| 30 | int fd; |
| 31 | u32 file_slot; |
| 32 | }; |
| 33 | |
| 34 | static bool io_openat_force_async(struct io_open *open) |
| 35 | { |
| 36 | /* |
| 37 | * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open, |
| 38 | * it'll always -EAGAIN. Note that we test for __O_TMPFILE because |
| 39 | * O_TMPFILE includes O_DIRECTORY, which isn't a flag we need to force |
| 40 | * async for. |
| 41 | */ |
| 42 | return open->how.flags & (O_TRUNC | O_CREAT | __O_TMPFILE); |
| 43 | } |
| 44 | |
| 45 | static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 46 | { |
| 47 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 48 | const char __user *fname; |
| 49 | int ret; |
| 50 | |
| 51 | if (unlikely(sqe->buf_index)) |
| 52 | return -EINVAL; |
| 53 | if (unlikely(req->flags & REQ_F_FIXED_FILE)) |
| 54 | return -EBADF; |
| 55 | |
| 56 | /* open.how should be already initialised */ |
| 57 | if (!(open->how.flags & O_PATH) && force_o_largefile()) |
| 58 | open->how.flags |= O_LARGEFILE; |
| 59 | |
| 60 | open->dfd = READ_ONCE(sqe->fd); |
| 61 | fname = u64_to_user_ptr(READ_ONCE(sqe->addr)); |
| 62 | open->filename = getname(fname); |
| 63 | if (IS_ERR(open->filename)) { |
| 64 | ret = PTR_ERR(open->filename); |
| 65 | open->filename = NULL; |
| 66 | return ret; |
| 67 | } |
| 68 | |
| 69 | open->file_slot = READ_ONCE(sqe->file_index); |
| 70 | if (open->file_slot && (open->how.flags & O_CLOEXEC)) |
| 71 | return -EINVAL; |
| 72 | |
| 73 | open->nofile = rlimit(RLIMIT_NOFILE); |
| 74 | req->flags |= REQ_F_NEED_CLEANUP; |
| 75 | if (io_openat_force_async(open)) |
| 76 | req->flags |= REQ_F_FORCE_ASYNC; |
| 77 | return 0; |
| 78 | } |
| 79 | |
| 80 | int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 81 | { |
| 82 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 83 | u64 mode = READ_ONCE(sqe->len); |
| 84 | u64 flags = READ_ONCE(sqe->open_flags); |
| 85 | |
| 86 | open->how = build_open_how(flags, mode); |
| 87 | return __io_openat_prep(req, sqe); |
| 88 | } |
| 89 | |
| 90 | int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 91 | { |
| 92 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 93 | struct open_how __user *how; |
| 94 | size_t len; |
| 95 | int ret; |
| 96 | |
| 97 | how = u64_to_user_ptr(READ_ONCE(sqe->addr2)); |
| 98 | len = READ_ONCE(sqe->len); |
| 99 | if (len < OPEN_HOW_SIZE_VER0) |
| 100 | return -EINVAL; |
| 101 | |
| 102 | ret = copy_struct_from_user(&open->how, sizeof(open->how), how, len); |
| 103 | if (ret) |
| 104 | return ret; |
| 105 | |
| 106 | return __io_openat_prep(req, sqe); |
| 107 | } |
| 108 | |
| 109 | int io_openat2(struct io_kiocb *req, unsigned int issue_flags) |
| 110 | { |
| 111 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 112 | struct open_flags op; |
| 113 | struct file *file; |
| 114 | bool resolve_nonblock, nonblock_set; |
| 115 | bool fixed = !!open->file_slot; |
| 116 | int ret; |
| 117 | |
| 118 | ret = build_open_flags(&open->how, &op); |
| 119 | if (ret) |
| 120 | goto err; |
| 121 | nonblock_set = op.open_flag & O_NONBLOCK; |
| 122 | resolve_nonblock = open->how.resolve & RESOLVE_CACHED; |
| 123 | if (issue_flags & IO_URING_F_NONBLOCK) { |
| 124 | WARN_ON_ONCE(io_openat_force_async(open)); |
| 125 | op.lookup_flags |= LOOKUP_CACHED; |
| 126 | op.open_flag |= O_NONBLOCK; |
| 127 | } |
| 128 | |
| 129 | if (!fixed) { |
| 130 | ret = __get_unused_fd_flags(open->how.flags, open->nofile); |
| 131 | if (ret < 0) |
| 132 | goto err; |
| 133 | } |
| 134 | |
| 135 | file = do_filp_open(open->dfd, open->filename, &op); |
| 136 | if (IS_ERR(file)) { |
| 137 | /* |
| 138 | * We could hang on to this 'fd' on retrying, but seems like |
| 139 | * marginal gain for something that is now known to be a slower |
| 140 | * path. So just put it, and we'll get a new one when we retry. |
| 141 | */ |
| 142 | if (!fixed) |
| 143 | put_unused_fd(ret); |
| 144 | |
| 145 | ret = PTR_ERR(file); |
| 146 | /* only retry if RESOLVE_CACHED wasn't already set by application */ |
| 147 | if (ret == -EAGAIN && |
| 148 | (!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK))) |
| 149 | return -EAGAIN; |
| 150 | goto err; |
| 151 | } |
| 152 | |
| 153 | if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set) |
| 154 | file->f_flags &= ~O_NONBLOCK; |
| 155 | |
| 156 | if (!fixed) |
| 157 | fd_install(ret, file); |
| 158 | else |
| 159 | ret = io_fixed_fd_install(req, issue_flags, file, |
| 160 | open->file_slot); |
| 161 | err: |
| 162 | putname(open->filename); |
| 163 | req->flags &= ~REQ_F_NEED_CLEANUP; |
| 164 | if (ret < 0) |
| 165 | req_set_fail(req); |
| 166 | io_req_set_res(req, ret, 0); |
| 167 | return IOU_OK; |
| 168 | } |
| 169 | |
| 170 | int io_openat(struct io_kiocb *req, unsigned int issue_flags) |
| 171 | { |
| 172 | return io_openat2(req, issue_flags); |
| 173 | } |
| 174 | |
| 175 | void io_open_cleanup(struct io_kiocb *req) |
| 176 | { |
| 177 | struct io_open *open = io_kiocb_to_cmd(req, struct io_open); |
| 178 | |
| 179 | if (open->filename) |
| 180 | putname(open->filename); |
| 181 | } |
| 182 | |
| 183 | int __io_close_fixed(struct io_ring_ctx *ctx, unsigned int issue_flags, |
| 184 | unsigned int offset) |
| 185 | { |
| 186 | int ret; |
| 187 | |
| 188 | io_ring_submit_lock(ctx, issue_flags); |
| 189 | ret = io_fixed_fd_remove(ctx, offset); |
| 190 | io_ring_submit_unlock(ctx, issue_flags); |
| 191 | |
| 192 | return ret; |
| 193 | } |
| 194 | |
| 195 | static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) |
| 196 | { |
| 197 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 198 | |
| 199 | return __io_close_fixed(req->ctx, issue_flags, close->file_slot - 1); |
| 200 | } |
| 201 | |
| 202 | int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) |
| 203 | { |
| 204 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 205 | |
| 206 | if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) |
| 207 | return -EINVAL; |
| 208 | if (req->flags & REQ_F_FIXED_FILE) |
| 209 | return -EBADF; |
| 210 | |
| 211 | close->fd = READ_ONCE(sqe->fd); |
| 212 | close->file_slot = READ_ONCE(sqe->file_index); |
| 213 | if (close->file_slot && close->fd) |
| 214 | return -EINVAL; |
| 215 | |
| 216 | return 0; |
| 217 | } |
| 218 | |
| 219 | int io_close(struct io_kiocb *req, unsigned int issue_flags) |
| 220 | { |
| 221 | struct files_struct *files = current->files; |
| 222 | struct io_close *close = io_kiocb_to_cmd(req, struct io_close); |
| 223 | struct fdtable *fdt; |
| 224 | struct file *file; |
| 225 | int ret = -EBADF; |
| 226 | |
| 227 | if (close->file_slot) { |
| 228 | ret = io_close_fixed(req, issue_flags); |
| 229 | goto err; |
| 230 | } |
| 231 | |
| 232 | spin_lock(&files->file_lock); |
| 233 | fdt = files_fdtable(files); |
| 234 | if (close->fd >= fdt->max_fds) { |
| 235 | spin_unlock(&files->file_lock); |
| 236 | goto err; |
| 237 | } |
| 238 | file = rcu_dereference_protected(fdt->fd[close->fd], |
| 239 | lockdep_is_held(&files->file_lock)); |
| 240 | if (!file || io_is_uring_fops(file)) { |
| 241 | spin_unlock(&files->file_lock); |
| 242 | goto err; |
| 243 | } |
| 244 | |
| 245 | /* if the file has a flush method, be safe and punt to async */ |
| 246 | if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) { |
| 247 | spin_unlock(&files->file_lock); |
| 248 | return -EAGAIN; |
| 249 | } |
| 250 | |
| 251 | file = __close_fd_get_file(close->fd); |
| 252 | spin_unlock(&files->file_lock); |
| 253 | if (!file) |
| 254 | goto err; |
| 255 | |
| 256 | /* No ->flush() or already async, safely close from here */ |
| 257 | ret = filp_close(file, current->files); |
| 258 | err: |
| 259 | if (ret < 0) |
| 260 | req_set_fail(req); |
| 261 | io_req_set_res(req, ret, 0); |
| 262 | return IOU_OK; |
| 263 | } |