Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * inode.c - part of tracefs, a pseudo file system for activating tracing
4 *
5 * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
6 *
7 * Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
8 *
9 * tracefs is the file system that is used by the tracing infrastructure.
10 */
11
12#include <linux/module.h>
13#include <linux/fs.h>
14#include <linux/mount.h>
15#include <linux/kobject.h>
16#include <linux/namei.h>
17#include <linux/tracefs.h>
18#include <linux/fsnotify.h>
19#include <linux/security.h>
20#include <linux/seq_file.h>
21#include <linux/parser.h>
22#include <linux/magic.h>
23#include <linux/slab.h>
24
25#define TRACEFS_DEFAULT_MODE 0700
26
27static struct vfsmount *tracefs_mount;
28static int tracefs_mount_count;
29static bool tracefs_registered;
30
31static ssize_t default_read_file(struct file *file, char __user *buf,
32 size_t count, loff_t *ppos)
33{
34 return 0;
35}
36
37static ssize_t default_write_file(struct file *file, const char __user *buf,
38 size_t count, loff_t *ppos)
39{
40 return count;
41}
42
43static const struct file_operations tracefs_file_operations = {
44 .read = default_read_file,
45 .write = default_write_file,
46 .open = simple_open,
47 .llseek = noop_llseek,
48};
49
50static struct tracefs_dir_ops {
51 int (*mkdir)(const char *name);
52 int (*rmdir)(const char *name);
53} tracefs_ops __ro_after_init;
54
55static char *get_dname(struct dentry *dentry)
56{
57 const char *dname;
58 char *name;
59 int len = dentry->d_name.len;
60
61 dname = dentry->d_name.name;
62 name = kmalloc(len + 1, GFP_KERNEL);
63 if (!name)
64 return NULL;
65 memcpy(name, dname, len);
66 name[len] = 0;
67 return name;
68}
69
70static int tracefs_syscall_mkdir(struct inode *inode, struct dentry *dentry, umode_t mode)
71{
72 char *name;
73 int ret;
74
75 name = get_dname(dentry);
76 if (!name)
77 return -ENOMEM;
78
79 /*
80 * The mkdir call can call the generic functions that create
81 * the files within the tracefs system. It is up to the individual
82 * mkdir routine to handle races.
83 */
84 inode_unlock(inode);
85 ret = tracefs_ops.mkdir(name);
86 inode_lock(inode);
87
88 kfree(name);
89
90 return ret;
91}
92
93static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
94{
95 char *name;
96 int ret;
97
98 name = get_dname(dentry);
99 if (!name)
100 return -ENOMEM;
101
102 /*
103 * The rmdir call can call the generic functions that create
104 * the files within the tracefs system. It is up to the individual
105 * rmdir routine to handle races.
106 * This time we need to unlock not only the parent (inode) but
107 * also the directory that is being deleted.
108 */
109 inode_unlock(inode);
110 inode_unlock(dentry->d_inode);
111
112 ret = tracefs_ops.rmdir(name);
113
114 inode_lock_nested(inode, I_MUTEX_PARENT);
115 inode_lock(dentry->d_inode);
116
117 kfree(name);
118
119 return ret;
120}
121
122static const struct inode_operations tracefs_dir_inode_operations = {
123 .lookup = simple_lookup,
124 .mkdir = tracefs_syscall_mkdir,
125 .rmdir = tracefs_syscall_rmdir,
126};
127
128static struct inode *tracefs_get_inode(struct super_block *sb)
129{
130 struct inode *inode = new_inode(sb);
131 if (inode) {
132 inode->i_ino = get_next_ino();
133 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
134 }
135 return inode;
136}
137
138struct tracefs_mount_opts {
139 kuid_t uid;
140 kgid_t gid;
141 umode_t mode;
142};
143
144enum {
145 Opt_uid,
146 Opt_gid,
147 Opt_mode,
148 Opt_err
149};
150
151static const match_table_t tokens = {
152 {Opt_uid, "uid=%u"},
153 {Opt_gid, "gid=%u"},
154 {Opt_mode, "mode=%o"},
155 {Opt_err, NULL}
156};
157
158struct tracefs_fs_info {
159 struct tracefs_mount_opts mount_opts;
160};
161
162static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
163{
164 substring_t args[MAX_OPT_ARGS];
165 int option;
166 int token;
167 kuid_t uid;
168 kgid_t gid;
169 char *p;
170
171 opts->mode = TRACEFS_DEFAULT_MODE;
172
173 while ((p = strsep(&data, ",")) != NULL) {
174 if (!*p)
175 continue;
176
177 token = match_token(p, tokens, args);
178 switch (token) {
179 case Opt_uid:
180 if (match_int(&args[0], &option))
181 return -EINVAL;
182 uid = make_kuid(current_user_ns(), option);
183 if (!uid_valid(uid))
184 return -EINVAL;
185 opts->uid = uid;
186 break;
187 case Opt_gid:
188 if (match_int(&args[0], &option))
189 return -EINVAL;
190 gid = make_kgid(current_user_ns(), option);
191 if (!gid_valid(gid))
192 return -EINVAL;
193 opts->gid = gid;
194 break;
195 case Opt_mode:
196 if (match_octal(&args[0], &option))
197 return -EINVAL;
198 opts->mode = option & S_IALLUGO;
199 break;
200 /*
201 * We might like to report bad mount options here;
202 * but traditionally tracefs has ignored all mount options
203 */
204 }
205 }
206
207 return 0;
208}
209
210static int tracefs_apply_options(struct super_block *sb)
211{
212 struct tracefs_fs_info *fsi = sb->s_fs_info;
213 struct inode *inode = sb->s_root->d_inode;
214 struct tracefs_mount_opts *opts = &fsi->mount_opts;
215
216 inode->i_mode &= ~S_IALLUGO;
217 inode->i_mode |= opts->mode;
218
219 inode->i_uid = opts->uid;
220 inode->i_gid = opts->gid;
221
222 return 0;
223}
224
225static int tracefs_remount(struct super_block *sb, int *flags, char *data)
226{
227 int err;
228 struct tracefs_fs_info *fsi = sb->s_fs_info;
229
230 sync_filesystem(sb);
231 err = tracefs_parse_options(data, &fsi->mount_opts);
232 if (err)
233 goto fail;
234
235 tracefs_apply_options(sb);
236
237fail:
238 return err;
239}
240
241static int tracefs_show_options(struct seq_file *m, struct dentry *root)
242{
243 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
244 struct tracefs_mount_opts *opts = &fsi->mount_opts;
245
246 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
247 seq_printf(m, ",uid=%u",
248 from_kuid_munged(&init_user_ns, opts->uid));
249 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
250 seq_printf(m, ",gid=%u",
251 from_kgid_munged(&init_user_ns, opts->gid));
252 if (opts->mode != TRACEFS_DEFAULT_MODE)
253 seq_printf(m, ",mode=%o", opts->mode);
254
255 return 0;
256}
257
258static const struct super_operations tracefs_super_operations = {
259 .statfs = simple_statfs,
260 .remount_fs = tracefs_remount,
261 .show_options = tracefs_show_options,
262};
263
264static int trace_fill_super(struct super_block *sb, void *data, int silent)
265{
266 static const struct tree_descr trace_files[] = {{""}};
267 struct tracefs_fs_info *fsi;
268 int err;
269
270 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
271 sb->s_fs_info = fsi;
272 if (!fsi) {
273 err = -ENOMEM;
274 goto fail;
275 }
276
277 err = tracefs_parse_options(data, &fsi->mount_opts);
278 if (err)
279 goto fail;
280
281 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
282 if (err)
283 goto fail;
284
285 sb->s_op = &tracefs_super_operations;
286
287 tracefs_apply_options(sb);
288
289 return 0;
290
291fail:
292 kfree(fsi);
293 sb->s_fs_info = NULL;
294 return err;
295}
296
297static struct dentry *trace_mount(struct file_system_type *fs_type,
298 int flags, const char *dev_name,
299 void *data)
300{
301 return mount_single(fs_type, flags, data, trace_fill_super);
302}
303
304static struct file_system_type trace_fs_type = {
305 .owner = THIS_MODULE,
306 .name = "tracefs",
307 .mount = trace_mount,
308 .kill_sb = kill_litter_super,
309};
310MODULE_ALIAS_FS("tracefs");
311
312static struct dentry *start_creating(const char *name, struct dentry *parent)
313{
314 struct dentry *dentry;
315 int error;
316
317 pr_debug("tracefs: creating file '%s'\n",name);
318
319 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
320 &tracefs_mount_count);
321 if (error)
322 return ERR_PTR(error);
323
324 /* If the parent is not specified, we create it in the root.
325 * We need the root dentry to do this, which is in the super
326 * block. A pointer to that is in the struct vfsmount that we
327 * have around.
328 */
329 if (!parent)
330 parent = tracefs_mount->mnt_root;
331
332 inode_lock(parent->d_inode);
333 dentry = lookup_one_len(name, parent, strlen(name));
334 if (!IS_ERR(dentry) && dentry->d_inode) {
335 dput(dentry);
336 dentry = ERR_PTR(-EEXIST);
337 }
338
339 if (IS_ERR(dentry)) {
340 inode_unlock(parent->d_inode);
341 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
342 }
343
344 return dentry;
345}
346
347static struct dentry *failed_creating(struct dentry *dentry)
348{
349 inode_unlock(dentry->d_parent->d_inode);
350 dput(dentry);
351 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
352 return NULL;
353}
354
355static struct dentry *end_creating(struct dentry *dentry)
356{
357 inode_unlock(dentry->d_parent->d_inode);
358 return dentry;
359}
360
361/**
362 * tracefs_create_file - create a file in the tracefs filesystem
363 * @name: a pointer to a string containing the name of the file to create.
364 * @mode: the permission that the file should have.
365 * @parent: a pointer to the parent dentry for this file. This should be a
366 * directory dentry if set. If this parameter is NULL, then the
367 * file will be created in the root of the tracefs filesystem.
368 * @data: a pointer to something that the caller will want to get to later
369 * on. The inode.i_private pointer will point to this value on
370 * the open() call.
371 * @fops: a pointer to a struct file_operations that should be used for
372 * this file.
373 *
374 * This is the basic "create a file" function for tracefs. It allows for a
375 * wide range of flexibility in creating a file, or a directory (if you want
376 * to create a directory, the tracefs_create_dir() function is
377 * recommended to be used instead.)
378 *
379 * This function will return a pointer to a dentry if it succeeds. This
380 * pointer must be passed to the tracefs_remove() function when the file is
381 * to be removed (no automatic cleanup happens if your module is unloaded,
382 * you are responsible here.) If an error occurs, %NULL will be returned.
383 *
384 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
385 * returned.
386 */
387struct dentry *tracefs_create_file(const char *name, umode_t mode,
388 struct dentry *parent, void *data,
389 const struct file_operations *fops)
390{
391 struct dentry *dentry;
392 struct inode *inode;
393
394 if (security_locked_down(LOCKDOWN_TRACEFS))
395 return NULL;
396
397 if (!(mode & S_IFMT))
398 mode |= S_IFREG;
399 BUG_ON(!S_ISREG(mode));
400 dentry = start_creating(name, parent);
401
402 if (IS_ERR(dentry))
403 return NULL;
404
405 inode = tracefs_get_inode(dentry->d_sb);
406 if (unlikely(!inode))
407 return failed_creating(dentry);
408
409 inode->i_mode = mode;
410 inode->i_fop = fops ? fops : &tracefs_file_operations;
411 inode->i_private = data;
412 d_instantiate(dentry, inode);
413 fsnotify_create(dentry->d_parent->d_inode, dentry);
414 return end_creating(dentry);
415}
416
417static struct dentry *__create_dir(const char *name, struct dentry *parent,
418 const struct inode_operations *ops)
419{
420 struct dentry *dentry = start_creating(name, parent);
421 struct inode *inode;
422
423 if (IS_ERR(dentry))
424 return NULL;
425
426 inode = tracefs_get_inode(dentry->d_sb);
427 if (unlikely(!inode))
428 return failed_creating(dentry);
429
430 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
431 inode->i_op = ops;
432 inode->i_fop = &simple_dir_operations;
433
434 /* directory inodes start off with i_nlink == 2 (for "." entry) */
435 inc_nlink(inode);
436 d_instantiate(dentry, inode);
437 inc_nlink(dentry->d_parent->d_inode);
438 fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
439 return end_creating(dentry);
440}
441
442/**
443 * tracefs_create_dir - create a directory in the tracefs filesystem
444 * @name: a pointer to a string containing the name of the directory to
445 * create.
446 * @parent: a pointer to the parent dentry for this file. This should be a
447 * directory dentry if set. If this parameter is NULL, then the
448 * directory will be created in the root of the tracefs filesystem.
449 *
450 * This function creates a directory in tracefs with the given name.
451 *
452 * This function will return a pointer to a dentry if it succeeds. This
453 * pointer must be passed to the tracefs_remove() function when the file is
454 * to be removed. If an error occurs, %NULL will be returned.
455 *
456 * If tracing is not enabled in the kernel, the value -%ENODEV will be
457 * returned.
458 */
459struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
460{
461 return __create_dir(name, parent, &simple_dir_inode_operations);
462}
463
464/**
465 * tracefs_create_instance_dir - create the tracing instances directory
466 * @name: The name of the instances directory to create
467 * @parent: The parent directory that the instances directory will exist
468 * @mkdir: The function to call when a mkdir is performed.
469 * @rmdir: The function to call when a rmdir is performed.
470 *
471 * Only one instances directory is allowed.
472 *
473 * The instances directory is special as it allows for mkdir and rmdir to
474 * to be done by userspace. When a mkdir or rmdir is performed, the inode
475 * locks are released and the methhods passed in (@mkdir and @rmdir) are
476 * called without locks and with the name of the directory being created
477 * within the instances directory.
478 *
479 * Returns the dentry of the instances directory.
480 */
481__init struct dentry *tracefs_create_instance_dir(const char *name,
482 struct dentry *parent,
483 int (*mkdir)(const char *name),
484 int (*rmdir)(const char *name))
485{
486 struct dentry *dentry;
487
488 /* Only allow one instance of the instances directory. */
489 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
490 return NULL;
491
492 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
493 if (!dentry)
494 return NULL;
495
496 tracefs_ops.mkdir = mkdir;
497 tracefs_ops.rmdir = rmdir;
498
499 return dentry;
500}
501
502static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
503{
504 int ret = 0;
505
506 if (simple_positive(dentry)) {
507 if (dentry->d_inode) {
508 dget(dentry);
509 switch (dentry->d_inode->i_mode & S_IFMT) {
510 case S_IFDIR:
511 ret = simple_rmdir(parent->d_inode, dentry);
512 if (!ret)
513 fsnotify_rmdir(parent->d_inode, dentry);
514 break;
515 default:
516 simple_unlink(parent->d_inode, dentry);
517 fsnotify_unlink(parent->d_inode, dentry);
518 break;
519 }
520 if (!ret)
521 d_delete(dentry);
522 dput(dentry);
523 }
524 }
525 return ret;
526}
527
528/**
529 * tracefs_remove - removes a file or directory from the tracefs filesystem
530 * @dentry: a pointer to a the dentry of the file or directory to be
531 * removed.
532 *
533 * This function removes a file or directory in tracefs that was previously
534 * created with a call to another tracefs function (like
535 * tracefs_create_file() or variants thereof.)
536 */
537void tracefs_remove(struct dentry *dentry)
538{
539 struct dentry *parent;
540 int ret;
541
542 if (IS_ERR_OR_NULL(dentry))
543 return;
544
545 parent = dentry->d_parent;
546 inode_lock(parent->d_inode);
547 ret = __tracefs_remove(dentry, parent);
548 inode_unlock(parent->d_inode);
549 if (!ret)
550 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
551}
552
553/**
554 * tracefs_remove_recursive - recursively removes a directory
555 * @dentry: a pointer to a the dentry of the directory to be removed.
556 *
557 * This function recursively removes a directory tree in tracefs that
558 * was previously created with a call to another tracefs function
559 * (like tracefs_create_file() or variants thereof.)
560 */
561void tracefs_remove_recursive(struct dentry *dentry)
562{
563 struct dentry *child, *parent;
564
565 if (IS_ERR_OR_NULL(dentry))
566 return;
567
568 parent = dentry;
569 down:
570 inode_lock(parent->d_inode);
571 loop:
572 /*
573 * The parent->d_subdirs is protected by the d_lock. Outside that
574 * lock, the child can be unlinked and set to be freed which can
575 * use the d_u.d_child as the rcu head and corrupt this list.
576 */
577 spin_lock(&parent->d_lock);
578 list_for_each_entry(child, &parent->d_subdirs, d_child) {
579 if (!simple_positive(child))
580 continue;
581
582 /* perhaps simple_empty(child) makes more sense */
583 if (!list_empty(&child->d_subdirs)) {
584 spin_unlock(&parent->d_lock);
585 inode_unlock(parent->d_inode);
586 parent = child;
587 goto down;
588 }
589
590 spin_unlock(&parent->d_lock);
591
592 if (!__tracefs_remove(child, parent))
593 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
594
595 /*
596 * The parent->d_lock protects agaist child from unlinking
597 * from d_subdirs. When releasing the parent->d_lock we can
598 * no longer trust that the next pointer is valid.
599 * Restart the loop. We'll skip this one with the
600 * simple_positive() check.
601 */
602 goto loop;
603 }
604 spin_unlock(&parent->d_lock);
605
606 inode_unlock(parent->d_inode);
607 child = parent;
608 parent = parent->d_parent;
609 inode_lock(parent->d_inode);
610
611 if (child != dentry)
612 /* go up */
613 goto loop;
614
615 if (!__tracefs_remove(child, parent))
616 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
617 inode_unlock(parent->d_inode);
618}
619
620/**
621 * tracefs_initialized - Tells whether tracefs has been registered
622 */
623bool tracefs_initialized(void)
624{
625 return tracefs_registered;
626}
627
628static int __init tracefs_init(void)
629{
630 int retval;
631
632 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
633 if (retval)
634 return -EINVAL;
635
636 retval = register_filesystem(&trace_fs_type);
637 if (!retval)
638 tracefs_registered = true;
639
640 return retval;
641}
642core_initcall(tracefs_init);