| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/fs/file_table.c |
| 4 | * |
| 5 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 6 | * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) |
| 7 | */ |
| 8 | |
| 9 | #include <linux/string.h> |
| 10 | #include <linux/slab.h> |
| 11 | #include <linux/file.h> |
| 12 | #include <linux/fdtable.h> |
| 13 | #include <linux/init.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/fs.h> |
| 16 | #include <linux/filelock.h> |
| 17 | #include <linux/security.h> |
| 18 | #include <linux/cred.h> |
| 19 | #include <linux/eventpoll.h> |
| 20 | #include <linux/rcupdate.h> |
| 21 | #include <linux/mount.h> |
| 22 | #include <linux/capability.h> |
| 23 | #include <linux/cdev.h> |
| 24 | #include <linux/fsnotify.h> |
| 25 | #include <linux/sysctl.h> |
| 26 | #include <linux/percpu_counter.h> |
| 27 | #include <linux/percpu.h> |
| 28 | #include <linux/task_work.h> |
| 29 | #include <linux/ima.h> |
| 30 | #include <linux/swap.h> |
| 31 | #include <linux/kmemleak.h> |
| 32 | |
| 33 | #include <linux/atomic.h> |
| 34 | |
| 35 | #include "internal.h" |
| 36 | |
| 37 | /* sysctl tunables... */ |
| 38 | static struct files_stat_struct files_stat = { |
| 39 | .max_files = NR_FILE |
| 40 | }; |
| 41 | |
| 42 | /* SLAB cache for file structures */ |
| 43 | static struct kmem_cache *filp_cachep __ro_after_init; |
| 44 | |
| 45 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
| 46 | |
| 47 | /* Container for backing file with optional user path */ |
| 48 | struct backing_file { |
| 49 | struct file file; |
| 50 | struct path user_path; |
| 51 | }; |
| 52 | |
| 53 | static inline struct backing_file *backing_file(struct file *f) |
| 54 | { |
| 55 | return container_of(f, struct backing_file, file); |
| 56 | } |
| 57 | |
| 58 | struct path *backing_file_user_path(struct file *f) |
| 59 | { |
| 60 | return &backing_file(f)->user_path; |
| 61 | } |
| 62 | EXPORT_SYMBOL_GPL(backing_file_user_path); |
| 63 | |
| 64 | static inline void file_free(struct file *f) |
| 65 | { |
| 66 | security_file_free(f); |
| 67 | if (likely(!(f->f_mode & FMODE_NOACCOUNT))) |
| 68 | percpu_counter_dec(&nr_files); |
| 69 | put_cred(f->f_cred); |
| 70 | if (unlikely(f->f_mode & FMODE_BACKING)) { |
| 71 | path_put(backing_file_user_path(f)); |
| 72 | kfree(backing_file(f)); |
| 73 | } else { |
| 74 | kmem_cache_free(filp_cachep, f); |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | /* |
| 79 | * Return the total number of open files in the system |
| 80 | */ |
| 81 | static long get_nr_files(void) |
| 82 | { |
| 83 | return percpu_counter_read_positive(&nr_files); |
| 84 | } |
| 85 | |
| 86 | /* |
| 87 | * Return the maximum number of open files in the system |
| 88 | */ |
| 89 | unsigned long get_max_files(void) |
| 90 | { |
| 91 | return files_stat.max_files; |
| 92 | } |
| 93 | EXPORT_SYMBOL_GPL(get_max_files); |
| 94 | |
| 95 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) |
| 96 | |
| 97 | /* |
| 98 | * Handle nr_files sysctl |
| 99 | */ |
| 100 | static int proc_nr_files(struct ctl_table *table, int write, void *buffer, |
| 101 | size_t *lenp, loff_t *ppos) |
| 102 | { |
| 103 | files_stat.nr_files = get_nr_files(); |
| 104 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
| 105 | } |
| 106 | |
| 107 | static struct ctl_table fs_stat_sysctls[] = { |
| 108 | { |
| 109 | .procname = "file-nr", |
| 110 | .data = &files_stat, |
| 111 | .maxlen = sizeof(files_stat), |
| 112 | .mode = 0444, |
| 113 | .proc_handler = proc_nr_files, |
| 114 | }, |
| 115 | { |
| 116 | .procname = "file-max", |
| 117 | .data = &files_stat.max_files, |
| 118 | .maxlen = sizeof(files_stat.max_files), |
| 119 | .mode = 0644, |
| 120 | .proc_handler = proc_doulongvec_minmax, |
| 121 | .extra1 = SYSCTL_LONG_ZERO, |
| 122 | .extra2 = SYSCTL_LONG_MAX, |
| 123 | }, |
| 124 | { |
| 125 | .procname = "nr_open", |
| 126 | .data = &sysctl_nr_open, |
| 127 | .maxlen = sizeof(unsigned int), |
| 128 | .mode = 0644, |
| 129 | .proc_handler = proc_dointvec_minmax, |
| 130 | .extra1 = &sysctl_nr_open_min, |
| 131 | .extra2 = &sysctl_nr_open_max, |
| 132 | }, |
| 133 | }; |
| 134 | |
| 135 | static int __init init_fs_stat_sysctls(void) |
| 136 | { |
| 137 | register_sysctl_init("fs", fs_stat_sysctls); |
| 138 | if (IS_ENABLED(CONFIG_BINFMT_MISC)) { |
| 139 | struct ctl_table_header *hdr; |
| 140 | hdr = register_sysctl_mount_point("fs/binfmt_misc"); |
| 141 | kmemleak_not_leak(hdr); |
| 142 | } |
| 143 | return 0; |
| 144 | } |
| 145 | fs_initcall(init_fs_stat_sysctls); |
| 146 | #endif |
| 147 | |
| 148 | static int init_file(struct file *f, int flags, const struct cred *cred) |
| 149 | { |
| 150 | int error; |
| 151 | |
| 152 | f->f_cred = get_cred(cred); |
| 153 | error = security_file_alloc(f); |
| 154 | if (unlikely(error)) { |
| 155 | put_cred(f->f_cred); |
| 156 | return error; |
| 157 | } |
| 158 | |
| 159 | rwlock_init(&f->f_owner.lock); |
| 160 | spin_lock_init(&f->f_lock); |
| 161 | mutex_init(&f->f_pos_lock); |
| 162 | f->f_flags = flags; |
| 163 | f->f_mode = OPEN_FMODE(flags); |
| 164 | /* f->f_version: 0 */ |
| 165 | |
| 166 | /* |
| 167 | * We're SLAB_TYPESAFE_BY_RCU so initialize f_count last. While |
| 168 | * fget-rcu pattern users need to be able to handle spurious |
| 169 | * refcount bumps we should reinitialize the reused file first. |
| 170 | */ |
| 171 | atomic_long_set(&f->f_count, 1); |
| 172 | return 0; |
| 173 | } |
| 174 | |
| 175 | /* Find an unused file structure and return a pointer to it. |
| 176 | * Returns an error pointer if some error happend e.g. we over file |
| 177 | * structures limit, run out of memory or operation is not permitted. |
| 178 | * |
| 179 | * Be very careful using this. You are responsible for |
| 180 | * getting write access to any mount that you might assign |
| 181 | * to this filp, if it is opened for write. If this is not |
| 182 | * done, you will imbalance int the mount's writer count |
| 183 | * and a warning at __fput() time. |
| 184 | */ |
| 185 | struct file *alloc_empty_file(int flags, const struct cred *cred) |
| 186 | { |
| 187 | static long old_max; |
| 188 | struct file *f; |
| 189 | int error; |
| 190 | |
| 191 | /* |
| 192 | * Privileged users can go above max_files |
| 193 | */ |
| 194 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
| 195 | /* |
| 196 | * percpu_counters are inaccurate. Do an expensive check before |
| 197 | * we go and fail. |
| 198 | */ |
| 199 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
| 200 | goto over; |
| 201 | } |
| 202 | |
| 203 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
| 204 | if (unlikely(!f)) |
| 205 | return ERR_PTR(-ENOMEM); |
| 206 | |
| 207 | error = init_file(f, flags, cred); |
| 208 | if (unlikely(error)) { |
| 209 | kmem_cache_free(filp_cachep, f); |
| 210 | return ERR_PTR(error); |
| 211 | } |
| 212 | |
| 213 | percpu_counter_inc(&nr_files); |
| 214 | |
| 215 | return f; |
| 216 | |
| 217 | over: |
| 218 | /* Ran out of filps - report that */ |
| 219 | if (get_nr_files() > old_max) { |
| 220 | pr_info("VFS: file-max limit %lu reached\n", get_max_files()); |
| 221 | old_max = get_nr_files(); |
| 222 | } |
| 223 | return ERR_PTR(-ENFILE); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Variant of alloc_empty_file() that doesn't check and modify nr_files. |
| 228 | * |
| 229 | * This is only for kernel internal use, and the allocate file must not be |
| 230 | * installed into file tables or such. |
| 231 | */ |
| 232 | struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred) |
| 233 | { |
| 234 | struct file *f; |
| 235 | int error; |
| 236 | |
| 237 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
| 238 | if (unlikely(!f)) |
| 239 | return ERR_PTR(-ENOMEM); |
| 240 | |
| 241 | error = init_file(f, flags, cred); |
| 242 | if (unlikely(error)) { |
| 243 | kmem_cache_free(filp_cachep, f); |
| 244 | return ERR_PTR(error); |
| 245 | } |
| 246 | |
| 247 | f->f_mode |= FMODE_NOACCOUNT; |
| 248 | |
| 249 | return f; |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Variant of alloc_empty_file() that allocates a backing_file container |
| 254 | * and doesn't check and modify nr_files. |
| 255 | * |
| 256 | * This is only for kernel internal use, and the allocate file must not be |
| 257 | * installed into file tables or such. |
| 258 | */ |
| 259 | struct file *alloc_empty_backing_file(int flags, const struct cred *cred) |
| 260 | { |
| 261 | struct backing_file *ff; |
| 262 | int error; |
| 263 | |
| 264 | ff = kzalloc(sizeof(struct backing_file), GFP_KERNEL); |
| 265 | if (unlikely(!ff)) |
| 266 | return ERR_PTR(-ENOMEM); |
| 267 | |
| 268 | error = init_file(&ff->file, flags, cred); |
| 269 | if (unlikely(error)) { |
| 270 | kfree(ff); |
| 271 | return ERR_PTR(error); |
| 272 | } |
| 273 | |
| 274 | ff->file.f_mode |= FMODE_BACKING | FMODE_NOACCOUNT; |
| 275 | return &ff->file; |
| 276 | } |
| 277 | |
| 278 | /** |
| 279 | * file_init_path - initialize a 'struct file' based on path |
| 280 | * |
| 281 | * @file: the file to set up |
| 282 | * @path: the (dentry, vfsmount) pair for the new file |
| 283 | * @fop: the 'struct file_operations' for the new file |
| 284 | */ |
| 285 | static void file_init_path(struct file *file, const struct path *path, |
| 286 | const struct file_operations *fop) |
| 287 | { |
| 288 | file->f_path = *path; |
| 289 | file->f_inode = path->dentry->d_inode; |
| 290 | file->f_mapping = path->dentry->d_inode->i_mapping; |
| 291 | file->f_wb_err = filemap_sample_wb_err(file->f_mapping); |
| 292 | file->f_sb_err = file_sample_sb_err(file); |
| 293 | if (fop->llseek) |
| 294 | file->f_mode |= FMODE_LSEEK; |
| 295 | if ((file->f_mode & FMODE_READ) && |
| 296 | likely(fop->read || fop->read_iter)) |
| 297 | file->f_mode |= FMODE_CAN_READ; |
| 298 | if ((file->f_mode & FMODE_WRITE) && |
| 299 | likely(fop->write || fop->write_iter)) |
| 300 | file->f_mode |= FMODE_CAN_WRITE; |
| 301 | file->f_iocb_flags = iocb_flags(file); |
| 302 | file->f_mode |= FMODE_OPENED; |
| 303 | file->f_op = fop; |
| 304 | if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
| 305 | i_readcount_inc(path->dentry->d_inode); |
| 306 | } |
| 307 | |
| 308 | /** |
| 309 | * alloc_file - allocate and initialize a 'struct file' |
| 310 | * |
| 311 | * @path: the (dentry, vfsmount) pair for the new file |
| 312 | * @flags: O_... flags with which the new file will be opened |
| 313 | * @fop: the 'struct file_operations' for the new file |
| 314 | */ |
| 315 | static struct file *alloc_file(const struct path *path, int flags, |
| 316 | const struct file_operations *fop) |
| 317 | { |
| 318 | struct file *file; |
| 319 | |
| 320 | file = alloc_empty_file(flags, current_cred()); |
| 321 | if (!IS_ERR(file)) |
| 322 | file_init_path(file, path, fop); |
| 323 | return file; |
| 324 | } |
| 325 | |
| 326 | static inline int alloc_path_pseudo(const char *name, struct inode *inode, |
| 327 | struct vfsmount *mnt, struct path *path) |
| 328 | { |
| 329 | struct qstr this = QSTR_INIT(name, strlen(name)); |
| 330 | |
| 331 | path->dentry = d_alloc_pseudo(mnt->mnt_sb, &this); |
| 332 | if (!path->dentry) |
| 333 | return -ENOMEM; |
| 334 | path->mnt = mntget(mnt); |
| 335 | d_instantiate(path->dentry, inode); |
| 336 | return 0; |
| 337 | } |
| 338 | |
| 339 | struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt, |
| 340 | const char *name, int flags, |
| 341 | const struct file_operations *fops) |
| 342 | { |
| 343 | int ret; |
| 344 | struct path path; |
| 345 | struct file *file; |
| 346 | |
| 347 | ret = alloc_path_pseudo(name, inode, mnt, &path); |
| 348 | if (ret) |
| 349 | return ERR_PTR(ret); |
| 350 | |
| 351 | file = alloc_file(&path, flags, fops); |
| 352 | if (IS_ERR(file)) { |
| 353 | ihold(inode); |
| 354 | path_put(&path); |
| 355 | } |
| 356 | return file; |
| 357 | } |
| 358 | EXPORT_SYMBOL(alloc_file_pseudo); |
| 359 | |
| 360 | struct file *alloc_file_pseudo_noaccount(struct inode *inode, |
| 361 | struct vfsmount *mnt, const char *name, |
| 362 | int flags, |
| 363 | const struct file_operations *fops) |
| 364 | { |
| 365 | int ret; |
| 366 | struct path path; |
| 367 | struct file *file; |
| 368 | |
| 369 | ret = alloc_path_pseudo(name, inode, mnt, &path); |
| 370 | if (ret) |
| 371 | return ERR_PTR(ret); |
| 372 | |
| 373 | file = alloc_empty_file_noaccount(flags, current_cred()); |
| 374 | if (IS_ERR(file)) { |
| 375 | ihold(inode); |
| 376 | path_put(&path); |
| 377 | return file; |
| 378 | } |
| 379 | file_init_path(file, &path, fops); |
| 380 | return file; |
| 381 | } |
| 382 | EXPORT_SYMBOL_GPL(alloc_file_pseudo_noaccount); |
| 383 | |
| 384 | struct file *alloc_file_clone(struct file *base, int flags, |
| 385 | const struct file_operations *fops) |
| 386 | { |
| 387 | struct file *f = alloc_file(&base->f_path, flags, fops); |
| 388 | if (!IS_ERR(f)) { |
| 389 | path_get(&f->f_path); |
| 390 | f->f_mapping = base->f_mapping; |
| 391 | } |
| 392 | return f; |
| 393 | } |
| 394 | |
| 395 | /* the real guts of fput() - releasing the last reference to file |
| 396 | */ |
| 397 | static void __fput(struct file *file) |
| 398 | { |
| 399 | struct dentry *dentry = file->f_path.dentry; |
| 400 | struct vfsmount *mnt = file->f_path.mnt; |
| 401 | struct inode *inode = file->f_inode; |
| 402 | fmode_t mode = file->f_mode; |
| 403 | |
| 404 | if (unlikely(!(file->f_mode & FMODE_OPENED))) |
| 405 | goto out; |
| 406 | |
| 407 | might_sleep(); |
| 408 | |
| 409 | fsnotify_close(file); |
| 410 | /* |
| 411 | * The function eventpoll_release() should be the first called |
| 412 | * in the file cleanup chain. |
| 413 | */ |
| 414 | eventpoll_release(file); |
| 415 | locks_remove_file(file); |
| 416 | |
| 417 | ima_file_free(file); |
| 418 | if (unlikely(file->f_flags & FASYNC)) { |
| 419 | if (file->f_op->fasync) |
| 420 | file->f_op->fasync(-1, file, 0); |
| 421 | } |
| 422 | if (file->f_op->release) |
| 423 | file->f_op->release(inode, file); |
| 424 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && |
| 425 | !(mode & FMODE_PATH))) { |
| 426 | cdev_put(inode->i_cdev); |
| 427 | } |
| 428 | fops_put(file->f_op); |
| 429 | put_pid(file->f_owner.pid); |
| 430 | put_file_access(file); |
| 431 | dput(dentry); |
| 432 | if (unlikely(mode & FMODE_NEED_UNMOUNT)) |
| 433 | dissolve_on_fput(mnt); |
| 434 | mntput(mnt); |
| 435 | out: |
| 436 | file_free(file); |
| 437 | } |
| 438 | |
| 439 | static LLIST_HEAD(delayed_fput_list); |
| 440 | static void delayed_fput(struct work_struct *unused) |
| 441 | { |
| 442 | struct llist_node *node = llist_del_all(&delayed_fput_list); |
| 443 | struct file *f, *t; |
| 444 | |
| 445 | llist_for_each_entry_safe(f, t, node, f_llist) |
| 446 | __fput(f); |
| 447 | } |
| 448 | |
| 449 | static void ____fput(struct callback_head *work) |
| 450 | { |
| 451 | __fput(container_of(work, struct file, f_task_work)); |
| 452 | } |
| 453 | |
| 454 | /* |
| 455 | * If kernel thread really needs to have the final fput() it has done |
| 456 | * to complete, call this. The only user right now is the boot - we |
| 457 | * *do* need to make sure our writes to binaries on initramfs has |
| 458 | * not left us with opened struct file waiting for __fput() - execve() |
| 459 | * won't work without that. Please, don't add more callers without |
| 460 | * very good reasons; in particular, never call that with locks |
| 461 | * held and never call that from a thread that might need to do |
| 462 | * some work on any kind of umount. |
| 463 | */ |
| 464 | void flush_delayed_fput(void) |
| 465 | { |
| 466 | delayed_fput(NULL); |
| 467 | } |
| 468 | EXPORT_SYMBOL_GPL(flush_delayed_fput); |
| 469 | |
| 470 | static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput); |
| 471 | |
| 472 | void fput(struct file *file) |
| 473 | { |
| 474 | if (atomic_long_dec_and_test(&file->f_count)) { |
| 475 | struct task_struct *task = current; |
| 476 | |
| 477 | if (unlikely(!(file->f_mode & (FMODE_BACKING | FMODE_OPENED)))) { |
| 478 | file_free(file); |
| 479 | return; |
| 480 | } |
| 481 | if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) { |
| 482 | init_task_work(&file->f_task_work, ____fput); |
| 483 | if (!task_work_add(task, &file->f_task_work, TWA_RESUME)) |
| 484 | return; |
| 485 | /* |
| 486 | * After this task has run exit_task_work(), |
| 487 | * task_work_add() will fail. Fall through to delayed |
| 488 | * fput to avoid leaking *file. |
| 489 | */ |
| 490 | } |
| 491 | |
| 492 | if (llist_add(&file->f_llist, &delayed_fput_list)) |
| 493 | schedule_delayed_work(&delayed_fput_work, 1); |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | /* |
| 498 | * synchronous analog of fput(); for kernel threads that might be needed |
| 499 | * in some umount() (and thus can't use flush_delayed_fput() without |
| 500 | * risking deadlocks), need to wait for completion of __fput() and know |
| 501 | * for this specific struct file it won't involve anything that would |
| 502 | * need them. Use only if you really need it - at the very least, |
| 503 | * don't blindly convert fput() by kernel thread to that. |
| 504 | */ |
| 505 | void __fput_sync(struct file *file) |
| 506 | { |
| 507 | if (atomic_long_dec_and_test(&file->f_count)) |
| 508 | __fput(file); |
| 509 | } |
| 510 | |
| 511 | EXPORT_SYMBOL(fput); |
| 512 | EXPORT_SYMBOL(__fput_sync); |
| 513 | |
| 514 | void __init files_init(void) |
| 515 | { |
| 516 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, |
| 517 | SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN | |
| 518 | SLAB_PANIC | SLAB_ACCOUNT, NULL); |
| 519 | percpu_counter_init(&nr_files, 0, GFP_KERNEL); |
| 520 | } |
| 521 | |
| 522 | /* |
| 523 | * One file with associated inode and dcache is very roughly 1K. Per default |
| 524 | * do not use more than 10% of our memory for files. |
| 525 | */ |
| 526 | void __init files_maxfiles_init(void) |
| 527 | { |
| 528 | unsigned long n; |
| 529 | unsigned long nr_pages = totalram_pages(); |
| 530 | unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2; |
| 531 | |
| 532 | memreserve = min(memreserve, nr_pages - 1); |
| 533 | n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10; |
| 534 | |
| 535 | files_stat.max_files = max_t(unsigned long, n, NR_FILE); |
| 536 | } |