Loading...
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/fs_context.h>
15#include <linux/fs_parser.h>
16#include <linux/kobject.h>
17#include <linux/namei.h>
18#include <linux/tracefs.h>
19#include <linux/fsnotify.h>
20#include <linux/security.h>
21#include <linux/seq_file.h>
22#include <linux/magic.h>
23#include <linux/slab.h>
24#include "internal.h"
25
26#define TRACEFS_DEFAULT_MODE 0700
27static struct kmem_cache *tracefs_inode_cachep __ro_after_init;
28
29static struct vfsmount *tracefs_mount;
30static int tracefs_mount_count;
31static bool tracefs_registered;
32
33/*
34 * Keep track of all tracefs_inodes in order to update their
35 * flags if necessary on a remount.
36 */
37static DEFINE_SPINLOCK(tracefs_inode_lock);
38static LIST_HEAD(tracefs_inodes);
39
40static struct inode *tracefs_alloc_inode(struct super_block *sb)
41{
42 struct tracefs_inode *ti;
43 unsigned long flags;
44
45 ti = alloc_inode_sb(sb, tracefs_inode_cachep, GFP_KERNEL);
46 if (!ti)
47 return NULL;
48
49 spin_lock_irqsave(&tracefs_inode_lock, flags);
50 list_add_rcu(&ti->list, &tracefs_inodes);
51 spin_unlock_irqrestore(&tracefs_inode_lock, flags);
52
53 return &ti->vfs_inode;
54}
55
56static void tracefs_free_inode(struct inode *inode)
57{
58 struct tracefs_inode *ti = get_tracefs(inode);
59
60 kmem_cache_free(tracefs_inode_cachep, ti);
61}
62
63static void tracefs_destroy_inode(struct inode *inode)
64{
65 struct tracefs_inode *ti = get_tracefs(inode);
66 unsigned long flags;
67
68 spin_lock_irqsave(&tracefs_inode_lock, flags);
69 list_del_rcu(&ti->list);
70 spin_unlock_irqrestore(&tracefs_inode_lock, flags);
71}
72
73static ssize_t default_read_file(struct file *file, char __user *buf,
74 size_t count, loff_t *ppos)
75{
76 return 0;
77}
78
79static ssize_t default_write_file(struct file *file, const char __user *buf,
80 size_t count, loff_t *ppos)
81{
82 return count;
83}
84
85static const struct file_operations tracefs_file_operations = {
86 .read = default_read_file,
87 .write = default_write_file,
88 .open = simple_open,
89 .llseek = noop_llseek,
90};
91
92static struct tracefs_dir_ops {
93 int (*mkdir)(const char *name);
94 int (*rmdir)(const char *name);
95} tracefs_ops __ro_after_init;
96
97static char *get_dname(struct dentry *dentry)
98{
99 const char *dname;
100 char *name;
101 int len = dentry->d_name.len;
102
103 dname = dentry->d_name.name;
104 name = kmalloc(len + 1, GFP_KERNEL);
105 if (!name)
106 return NULL;
107 memcpy(name, dname, len);
108 name[len] = 0;
109 return name;
110}
111
112static int tracefs_syscall_mkdir(struct mnt_idmap *idmap,
113 struct inode *inode, struct dentry *dentry,
114 umode_t mode)
115{
116 struct tracefs_inode *ti;
117 char *name;
118 int ret;
119
120 name = get_dname(dentry);
121 if (!name)
122 return -ENOMEM;
123
124 /*
125 * This is a new directory that does not take the default of
126 * the rootfs. It becomes the default permissions for all the
127 * files and directories underneath it.
128 */
129 ti = get_tracefs(inode);
130 ti->flags |= TRACEFS_INSTANCE_INODE;
131 ti->private = inode;
132
133 /*
134 * The mkdir call can call the generic functions that create
135 * the files within the tracefs system. It is up to the individual
136 * mkdir routine to handle races.
137 */
138 inode_unlock(inode);
139 ret = tracefs_ops.mkdir(name);
140 inode_lock(inode);
141
142 kfree(name);
143
144 return ret;
145}
146
147static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
148{
149 char *name;
150 int ret;
151
152 name = get_dname(dentry);
153 if (!name)
154 return -ENOMEM;
155
156 /*
157 * The rmdir call can call the generic functions that create
158 * the files within the tracefs system. It is up to the individual
159 * rmdir routine to handle races.
160 * This time we need to unlock not only the parent (inode) but
161 * also the directory that is being deleted.
162 */
163 inode_unlock(inode);
164 inode_unlock(d_inode(dentry));
165
166 ret = tracefs_ops.rmdir(name);
167
168 inode_lock_nested(inode, I_MUTEX_PARENT);
169 inode_lock(d_inode(dentry));
170
171 kfree(name);
172
173 return ret;
174}
175
176static void set_tracefs_inode_owner(struct inode *inode)
177{
178 struct tracefs_inode *ti = get_tracefs(inode);
179 struct inode *root_inode = ti->private;
180 kuid_t uid;
181 kgid_t gid;
182
183 uid = root_inode->i_uid;
184 gid = root_inode->i_gid;
185
186 /*
187 * If the root is not the mount point, then check the root's
188 * permissions. If it was never set, then default to the
189 * mount point.
190 */
191 if (root_inode != d_inode(root_inode->i_sb->s_root)) {
192 struct tracefs_inode *rti;
193
194 rti = get_tracefs(root_inode);
195 root_inode = d_inode(root_inode->i_sb->s_root);
196
197 if (!(rti->flags & TRACEFS_UID_PERM_SET))
198 uid = root_inode->i_uid;
199
200 if (!(rti->flags & TRACEFS_GID_PERM_SET))
201 gid = root_inode->i_gid;
202 }
203
204 /*
205 * If this inode has never been referenced, then update
206 * the permissions to the superblock.
207 */
208 if (!(ti->flags & TRACEFS_UID_PERM_SET))
209 inode->i_uid = uid;
210
211 if (!(ti->flags & TRACEFS_GID_PERM_SET))
212 inode->i_gid = gid;
213}
214
215static int tracefs_permission(struct mnt_idmap *idmap,
216 struct inode *inode, int mask)
217{
218 set_tracefs_inode_owner(inode);
219 return generic_permission(idmap, inode, mask);
220}
221
222static int tracefs_getattr(struct mnt_idmap *idmap,
223 const struct path *path, struct kstat *stat,
224 u32 request_mask, unsigned int flags)
225{
226 struct inode *inode = d_backing_inode(path->dentry);
227
228 set_tracefs_inode_owner(inode);
229 generic_fillattr(idmap, request_mask, inode, stat);
230 return 0;
231}
232
233static int tracefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
234 struct iattr *attr)
235{
236 unsigned int ia_valid = attr->ia_valid;
237 struct inode *inode = d_inode(dentry);
238 struct tracefs_inode *ti = get_tracefs(inode);
239
240 if (ia_valid & ATTR_UID)
241 ti->flags |= TRACEFS_UID_PERM_SET;
242
243 if (ia_valid & ATTR_GID)
244 ti->flags |= TRACEFS_GID_PERM_SET;
245
246 return simple_setattr(idmap, dentry, attr);
247}
248
249static const struct inode_operations tracefs_instance_dir_inode_operations = {
250 .lookup = simple_lookup,
251 .mkdir = tracefs_syscall_mkdir,
252 .rmdir = tracefs_syscall_rmdir,
253 .permission = tracefs_permission,
254 .getattr = tracefs_getattr,
255 .setattr = tracefs_setattr,
256};
257
258static const struct inode_operations tracefs_dir_inode_operations = {
259 .lookup = simple_lookup,
260 .permission = tracefs_permission,
261 .getattr = tracefs_getattr,
262 .setattr = tracefs_setattr,
263};
264
265static const struct inode_operations tracefs_file_inode_operations = {
266 .permission = tracefs_permission,
267 .getattr = tracefs_getattr,
268 .setattr = tracefs_setattr,
269};
270
271struct inode *tracefs_get_inode(struct super_block *sb)
272{
273 struct inode *inode = new_inode(sb);
274 if (inode) {
275 inode->i_ino = get_next_ino();
276 simple_inode_init_ts(inode);
277 }
278 return inode;
279}
280
281struct tracefs_fs_info {
282 kuid_t uid;
283 kgid_t gid;
284 umode_t mode;
285 /* Opt_* bitfield. */
286 unsigned int opts;
287};
288
289enum {
290 Opt_uid,
291 Opt_gid,
292 Opt_mode,
293};
294
295static const struct fs_parameter_spec tracefs_param_specs[] = {
296 fsparam_gid ("gid", Opt_gid),
297 fsparam_u32oct ("mode", Opt_mode),
298 fsparam_uid ("uid", Opt_uid),
299 {}
300};
301
302static int tracefs_parse_param(struct fs_context *fc, struct fs_parameter *param)
303{
304 struct tracefs_fs_info *opts = fc->s_fs_info;
305 struct fs_parse_result result;
306 int opt;
307
308 opt = fs_parse(fc, tracefs_param_specs, param, &result);
309 if (opt < 0)
310 return opt;
311
312 switch (opt) {
313 case Opt_uid:
314 opts->uid = result.uid;
315 break;
316 case Opt_gid:
317 opts->gid = result.gid;
318 break;
319 case Opt_mode:
320 opts->mode = result.uint_32 & S_IALLUGO;
321 break;
322 /*
323 * We might like to report bad mount options here;
324 * but traditionally tracefs has ignored all mount options
325 */
326 }
327
328 opts->opts |= BIT(opt);
329
330 return 0;
331}
332
333static int tracefs_apply_options(struct super_block *sb, bool remount)
334{
335 struct tracefs_fs_info *fsi = sb->s_fs_info;
336 struct inode *inode = d_inode(sb->s_root);
337 struct tracefs_inode *ti;
338 bool update_uid, update_gid;
339 umode_t tmp_mode;
340
341 /*
342 * On remount, only reset mode/uid/gid if they were provided as mount
343 * options.
344 */
345
346 if (!remount || fsi->opts & BIT(Opt_mode)) {
347 tmp_mode = READ_ONCE(inode->i_mode) & ~S_IALLUGO;
348 tmp_mode |= fsi->mode;
349 WRITE_ONCE(inode->i_mode, tmp_mode);
350 }
351
352 if (!remount || fsi->opts & BIT(Opt_uid))
353 inode->i_uid = fsi->uid;
354
355 if (!remount || fsi->opts & BIT(Opt_gid))
356 inode->i_gid = fsi->gid;
357
358 if (remount && (fsi->opts & BIT(Opt_uid) || fsi->opts & BIT(Opt_gid))) {
359
360 update_uid = fsi->opts & BIT(Opt_uid);
361 update_gid = fsi->opts & BIT(Opt_gid);
362
363 rcu_read_lock();
364 list_for_each_entry_rcu(ti, &tracefs_inodes, list) {
365 if (update_uid) {
366 ti->flags &= ~TRACEFS_UID_PERM_SET;
367 ti->vfs_inode.i_uid = fsi->uid;
368 }
369
370 if (update_gid) {
371 ti->flags &= ~TRACEFS_GID_PERM_SET;
372 ti->vfs_inode.i_gid = fsi->gid;
373 }
374
375 /*
376 * Note, the above ti->vfs_inode updates are
377 * used in eventfs_remount() so they must come
378 * before calling it.
379 */
380 if (ti->flags & TRACEFS_EVENT_INODE)
381 eventfs_remount(ti, update_uid, update_gid);
382 }
383 rcu_read_unlock();
384 }
385
386 return 0;
387}
388
389static int tracefs_reconfigure(struct fs_context *fc)
390{
391 struct super_block *sb = fc->root->d_sb;
392 struct tracefs_fs_info *sb_opts = sb->s_fs_info;
393 struct tracefs_fs_info *new_opts = fc->s_fs_info;
394
395 if (!new_opts)
396 return 0;
397
398 sync_filesystem(sb);
399 /* structure copy of new mount options to sb */
400 *sb_opts = *new_opts;
401
402 return tracefs_apply_options(sb, true);
403}
404
405static int tracefs_show_options(struct seq_file *m, struct dentry *root)
406{
407 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
408
409 if (!uid_eq(fsi->uid, GLOBAL_ROOT_UID))
410 seq_printf(m, ",uid=%u",
411 from_kuid_munged(&init_user_ns, fsi->uid));
412 if (!gid_eq(fsi->gid, GLOBAL_ROOT_GID))
413 seq_printf(m, ",gid=%u",
414 from_kgid_munged(&init_user_ns, fsi->gid));
415 if (fsi->mode != TRACEFS_DEFAULT_MODE)
416 seq_printf(m, ",mode=%o", fsi->mode);
417
418 return 0;
419}
420
421static int tracefs_drop_inode(struct inode *inode)
422{
423 struct tracefs_inode *ti = get_tracefs(inode);
424
425 /*
426 * This inode is being freed and cannot be used for
427 * eventfs. Clear the flag so that it doesn't call into
428 * eventfs during the remount flag updates. The eventfs_inode
429 * gets freed after an RCU cycle, so the content will still
430 * be safe if the iteration is going on now.
431 */
432 ti->flags &= ~TRACEFS_EVENT_INODE;
433
434 return 1;
435}
436
437static const struct super_operations tracefs_super_operations = {
438 .alloc_inode = tracefs_alloc_inode,
439 .free_inode = tracefs_free_inode,
440 .destroy_inode = tracefs_destroy_inode,
441 .drop_inode = tracefs_drop_inode,
442 .statfs = simple_statfs,
443 .show_options = tracefs_show_options,
444};
445
446/*
447 * It would be cleaner if eventfs had its own dentry ops.
448 *
449 * Note that d_revalidate is called potentially under RCU,
450 * so it can't take the eventfs mutex etc. It's fine - if
451 * we open a file just as it's marked dead, things will
452 * still work just fine, and just see the old stale case.
453 */
454static void tracefs_d_release(struct dentry *dentry)
455{
456 if (dentry->d_fsdata)
457 eventfs_d_release(dentry);
458}
459
460static int tracefs_d_revalidate(struct dentry *dentry, unsigned int flags)
461{
462 struct eventfs_inode *ei = dentry->d_fsdata;
463
464 return !(ei && ei->is_freed);
465}
466
467static const struct dentry_operations tracefs_dentry_operations = {
468 .d_revalidate = tracefs_d_revalidate,
469 .d_release = tracefs_d_release,
470};
471
472static int tracefs_fill_super(struct super_block *sb, struct fs_context *fc)
473{
474 static const struct tree_descr trace_files[] = {{""}};
475 int err;
476
477 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
478 if (err)
479 return err;
480
481 sb->s_op = &tracefs_super_operations;
482 sb->s_d_op = &tracefs_dentry_operations;
483
484 return 0;
485}
486
487static int tracefs_get_tree(struct fs_context *fc)
488{
489 int err = get_tree_single(fc, tracefs_fill_super);
490
491 if (err)
492 return err;
493
494 return tracefs_reconfigure(fc);
495}
496
497static void tracefs_free_fc(struct fs_context *fc)
498{
499 kfree(fc->s_fs_info);
500}
501
502static const struct fs_context_operations tracefs_context_ops = {
503 .free = tracefs_free_fc,
504 .parse_param = tracefs_parse_param,
505 .get_tree = tracefs_get_tree,
506 .reconfigure = tracefs_reconfigure,
507};
508
509static int tracefs_init_fs_context(struct fs_context *fc)
510{
511 struct tracefs_fs_info *fsi;
512
513 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
514 if (!fsi)
515 return -ENOMEM;
516
517 fsi->mode = TRACEFS_DEFAULT_MODE;
518
519 fc->s_fs_info = fsi;
520 fc->ops = &tracefs_context_ops;
521 return 0;
522}
523
524static struct file_system_type trace_fs_type = {
525 .owner = THIS_MODULE,
526 .name = "tracefs",
527 .init_fs_context = tracefs_init_fs_context,
528 .parameters = tracefs_param_specs,
529 .kill_sb = kill_litter_super,
530};
531MODULE_ALIAS_FS("tracefs");
532
533struct dentry *tracefs_start_creating(const char *name, struct dentry *parent)
534{
535 struct dentry *dentry;
536 int error;
537
538 pr_debug("tracefs: creating file '%s'\n",name);
539
540 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
541 &tracefs_mount_count);
542 if (error)
543 return ERR_PTR(error);
544
545 /* If the parent is not specified, we create it in the root.
546 * We need the root dentry to do this, which is in the super
547 * block. A pointer to that is in the struct vfsmount that we
548 * have around.
549 */
550 if (!parent)
551 parent = tracefs_mount->mnt_root;
552
553 inode_lock(d_inode(parent));
554 if (unlikely(IS_DEADDIR(d_inode(parent))))
555 dentry = ERR_PTR(-ENOENT);
556 else
557 dentry = lookup_one_len(name, parent, strlen(name));
558 if (!IS_ERR(dentry) && d_inode(dentry)) {
559 dput(dentry);
560 dentry = ERR_PTR(-EEXIST);
561 }
562
563 if (IS_ERR(dentry)) {
564 inode_unlock(d_inode(parent));
565 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
566 }
567
568 return dentry;
569}
570
571struct dentry *tracefs_failed_creating(struct dentry *dentry)
572{
573 inode_unlock(d_inode(dentry->d_parent));
574 dput(dentry);
575 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
576 return NULL;
577}
578
579struct dentry *tracefs_end_creating(struct dentry *dentry)
580{
581 inode_unlock(d_inode(dentry->d_parent));
582 return dentry;
583}
584
585/* Find the inode that this will use for default */
586static struct inode *instance_inode(struct dentry *parent, struct inode *inode)
587{
588 struct tracefs_inode *ti;
589
590 /* If parent is NULL then use root inode */
591 if (!parent)
592 return d_inode(inode->i_sb->s_root);
593
594 /* Find the inode that is flagged as an instance or the root inode */
595 while (!IS_ROOT(parent)) {
596 ti = get_tracefs(d_inode(parent));
597 if (ti->flags & TRACEFS_INSTANCE_INODE)
598 break;
599 parent = parent->d_parent;
600 }
601
602 return d_inode(parent);
603}
604
605/**
606 * tracefs_create_file - create a file in the tracefs filesystem
607 * @name: a pointer to a string containing the name of the file to create.
608 * @mode: the permission that the file should have.
609 * @parent: a pointer to the parent dentry for this file. This should be a
610 * directory dentry if set. If this parameter is NULL, then the
611 * file will be created in the root of the tracefs filesystem.
612 * @data: a pointer to something that the caller will want to get to later
613 * on. The inode.i_private pointer will point to this value on
614 * the open() call.
615 * @fops: a pointer to a struct file_operations that should be used for
616 * this file.
617 *
618 * This is the basic "create a file" function for tracefs. It allows for a
619 * wide range of flexibility in creating a file, or a directory (if you want
620 * to create a directory, the tracefs_create_dir() function is
621 * recommended to be used instead.)
622 *
623 * This function will return a pointer to a dentry if it succeeds. This
624 * pointer must be passed to the tracefs_remove() function when the file is
625 * to be removed (no automatic cleanup happens if your module is unloaded,
626 * you are responsible here.) If an error occurs, %NULL will be returned.
627 *
628 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
629 * returned.
630 */
631struct dentry *tracefs_create_file(const char *name, umode_t mode,
632 struct dentry *parent, void *data,
633 const struct file_operations *fops)
634{
635 struct tracefs_inode *ti;
636 struct dentry *dentry;
637 struct inode *inode;
638
639 if (security_locked_down(LOCKDOWN_TRACEFS))
640 return NULL;
641
642 if (!(mode & S_IFMT))
643 mode |= S_IFREG;
644 BUG_ON(!S_ISREG(mode));
645 dentry = tracefs_start_creating(name, parent);
646
647 if (IS_ERR(dentry))
648 return NULL;
649
650 inode = tracefs_get_inode(dentry->d_sb);
651 if (unlikely(!inode))
652 return tracefs_failed_creating(dentry);
653
654 ti = get_tracefs(inode);
655 ti->private = instance_inode(parent, inode);
656
657 inode->i_mode = mode;
658 inode->i_op = &tracefs_file_inode_operations;
659 inode->i_fop = fops ? fops : &tracefs_file_operations;
660 inode->i_private = data;
661 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
662 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
663 d_instantiate(dentry, inode);
664 fsnotify_create(d_inode(dentry->d_parent), dentry);
665 return tracefs_end_creating(dentry);
666}
667
668static struct dentry *__create_dir(const char *name, struct dentry *parent,
669 const struct inode_operations *ops)
670{
671 struct tracefs_inode *ti;
672 struct dentry *dentry = tracefs_start_creating(name, parent);
673 struct inode *inode;
674
675 if (IS_ERR(dentry))
676 return NULL;
677
678 inode = tracefs_get_inode(dentry->d_sb);
679 if (unlikely(!inode))
680 return tracefs_failed_creating(dentry);
681
682 /* Do not set bits for OTH */
683 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
684 inode->i_op = ops;
685 inode->i_fop = &simple_dir_operations;
686 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
687 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
688
689 ti = get_tracefs(inode);
690 ti->private = instance_inode(parent, inode);
691
692 /* directory inodes start off with i_nlink == 2 (for "." entry) */
693 inc_nlink(inode);
694 d_instantiate(dentry, inode);
695 inc_nlink(d_inode(dentry->d_parent));
696 fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
697 return tracefs_end_creating(dentry);
698}
699
700/**
701 * tracefs_create_dir - create a directory in the tracefs filesystem
702 * @name: a pointer to a string containing the name of the directory to
703 * create.
704 * @parent: a pointer to the parent dentry for this file. This should be a
705 * directory dentry if set. If this parameter is NULL, then the
706 * directory will be created in the root of the tracefs filesystem.
707 *
708 * This function creates a directory in tracefs with the given name.
709 *
710 * This function will return a pointer to a dentry if it succeeds. This
711 * pointer must be passed to the tracefs_remove() function when the file is
712 * to be removed. If an error occurs, %NULL will be returned.
713 *
714 * If tracing is not enabled in the kernel, the value -%ENODEV will be
715 * returned.
716 */
717struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
718{
719 if (security_locked_down(LOCKDOWN_TRACEFS))
720 return NULL;
721
722 return __create_dir(name, parent, &tracefs_dir_inode_operations);
723}
724
725/**
726 * tracefs_create_instance_dir - create the tracing instances directory
727 * @name: The name of the instances directory to create
728 * @parent: The parent directory that the instances directory will exist
729 * @mkdir: The function to call when a mkdir is performed.
730 * @rmdir: The function to call when a rmdir is performed.
731 *
732 * Only one instances directory is allowed.
733 *
734 * The instances directory is special as it allows for mkdir and rmdir
735 * to be done by userspace. When a mkdir or rmdir is performed, the inode
736 * locks are released and the methods passed in (@mkdir and @rmdir) are
737 * called without locks and with the name of the directory being created
738 * within the instances directory.
739 *
740 * Returns the dentry of the instances directory.
741 */
742__init struct dentry *tracefs_create_instance_dir(const char *name,
743 struct dentry *parent,
744 int (*mkdir)(const char *name),
745 int (*rmdir)(const char *name))
746{
747 struct dentry *dentry;
748
749 /* Only allow one instance of the instances directory. */
750 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
751 return NULL;
752
753 dentry = __create_dir(name, parent, &tracefs_instance_dir_inode_operations);
754 if (!dentry)
755 return NULL;
756
757 tracefs_ops.mkdir = mkdir;
758 tracefs_ops.rmdir = rmdir;
759
760 return dentry;
761}
762
763static void remove_one(struct dentry *victim)
764{
765 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
766}
767
768/**
769 * tracefs_remove - recursively removes a directory
770 * @dentry: a pointer to a the dentry of the directory to be removed.
771 *
772 * This function recursively removes a directory tree in tracefs that
773 * was previously created with a call to another tracefs function
774 * (like tracefs_create_file() or variants thereof.)
775 */
776void tracefs_remove(struct dentry *dentry)
777{
778 if (IS_ERR_OR_NULL(dentry))
779 return;
780
781 simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
782 simple_recursive_removal(dentry, remove_one);
783 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
784}
785
786/**
787 * tracefs_initialized - Tells whether tracefs has been registered
788 */
789bool tracefs_initialized(void)
790{
791 return tracefs_registered;
792}
793
794static void init_once(void *foo)
795{
796 struct tracefs_inode *ti = (struct tracefs_inode *) foo;
797
798 /* inode_init_once() calls memset() on the vfs_inode portion */
799 inode_init_once(&ti->vfs_inode);
800
801 /* Zero out the rest */
802 memset_after(ti, 0, vfs_inode);
803}
804
805static int __init tracefs_init(void)
806{
807 int retval;
808
809 tracefs_inode_cachep = kmem_cache_create("tracefs_inode_cache",
810 sizeof(struct tracefs_inode),
811 0, (SLAB_RECLAIM_ACCOUNT|
812 SLAB_ACCOUNT),
813 init_once);
814 if (!tracefs_inode_cachep)
815 return -ENOMEM;
816
817 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
818 if (retval)
819 return -EINVAL;
820
821 retval = register_filesystem(&trace_fs_type);
822 if (!retval)
823 tracefs_registered = true;
824
825 return retval;
826}
827core_initcall(tracefs_init);
1/*
2 * inode.c - part of tracefs, a pseudo file system for activating tracing
3 *
4 * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
5 *
6 * Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version
10 * 2 as published by the Free Software Foundation.
11 *
12 * tracefs is the file system that is used by the tracing infrastructure.
13 *
14 */
15
16#include <linux/module.h>
17#include <linux/fs.h>
18#include <linux/mount.h>
19#include <linux/kobject.h>
20#include <linux/namei.h>
21#include <linux/tracefs.h>
22#include <linux/fsnotify.h>
23#include <linux/seq_file.h>
24#include <linux/parser.h>
25#include <linux/magic.h>
26#include <linux/slab.h>
27
28#define TRACEFS_DEFAULT_MODE 0700
29
30static struct vfsmount *tracefs_mount;
31static int tracefs_mount_count;
32static bool tracefs_registered;
33
34static ssize_t default_read_file(struct file *file, char __user *buf,
35 size_t count, loff_t *ppos)
36{
37 return 0;
38}
39
40static ssize_t default_write_file(struct file *file, const char __user *buf,
41 size_t count, loff_t *ppos)
42{
43 return count;
44}
45
46static const struct file_operations tracefs_file_operations = {
47 .read = default_read_file,
48 .write = default_write_file,
49 .open = simple_open,
50 .llseek = noop_llseek,
51};
52
53static struct tracefs_dir_ops {
54 int (*mkdir)(const char *name);
55 int (*rmdir)(const char *name);
56} tracefs_ops;
57
58static char *get_dname(struct dentry *dentry)
59{
60 const char *dname;
61 char *name;
62 int len = dentry->d_name.len;
63
64 dname = dentry->d_name.name;
65 name = kmalloc(len + 1, GFP_KERNEL);
66 if (!name)
67 return NULL;
68 memcpy(name, dname, len);
69 name[len] = 0;
70 return name;
71}
72
73static int tracefs_syscall_mkdir(struct inode *inode, struct dentry *dentry, umode_t mode)
74{
75 char *name;
76 int ret;
77
78 name = get_dname(dentry);
79 if (!name)
80 return -ENOMEM;
81
82 /*
83 * The mkdir call can call the generic functions that create
84 * the files within the tracefs system. It is up to the individual
85 * mkdir routine to handle races.
86 */
87 inode_unlock(inode);
88 ret = tracefs_ops.mkdir(name);
89 inode_lock(inode);
90
91 kfree(name);
92
93 return ret;
94}
95
96static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
97{
98 char *name;
99 int ret;
100
101 name = get_dname(dentry);
102 if (!name)
103 return -ENOMEM;
104
105 /*
106 * The rmdir call can call the generic functions that create
107 * the files within the tracefs system. It is up to the individual
108 * rmdir routine to handle races.
109 * This time we need to unlock not only the parent (inode) but
110 * also the directory that is being deleted.
111 */
112 inode_unlock(inode);
113 inode_unlock(dentry->d_inode);
114
115 ret = tracefs_ops.rmdir(name);
116
117 inode_lock_nested(inode, I_MUTEX_PARENT);
118 inode_lock(dentry->d_inode);
119
120 kfree(name);
121
122 return ret;
123}
124
125static const struct inode_operations tracefs_dir_inode_operations = {
126 .lookup = simple_lookup,
127 .mkdir = tracefs_syscall_mkdir,
128 .rmdir = tracefs_syscall_rmdir,
129};
130
131static struct inode *tracefs_get_inode(struct super_block *sb)
132{
133 struct inode *inode = new_inode(sb);
134 if (inode) {
135 inode->i_ino = get_next_ino();
136 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
137 }
138 return inode;
139}
140
141struct tracefs_mount_opts {
142 kuid_t uid;
143 kgid_t gid;
144 umode_t mode;
145};
146
147enum {
148 Opt_uid,
149 Opt_gid,
150 Opt_mode,
151 Opt_err
152};
153
154static const match_table_t tokens = {
155 {Opt_uid, "uid=%u"},
156 {Opt_gid, "gid=%u"},
157 {Opt_mode, "mode=%o"},
158 {Opt_err, NULL}
159};
160
161struct tracefs_fs_info {
162 struct tracefs_mount_opts mount_opts;
163};
164
165static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
166{
167 substring_t args[MAX_OPT_ARGS];
168 int option;
169 int token;
170 kuid_t uid;
171 kgid_t gid;
172 char *p;
173
174 opts->mode = TRACEFS_DEFAULT_MODE;
175
176 while ((p = strsep(&data, ",")) != NULL) {
177 if (!*p)
178 continue;
179
180 token = match_token(p, tokens, args);
181 switch (token) {
182 case Opt_uid:
183 if (match_int(&args[0], &option))
184 return -EINVAL;
185 uid = make_kuid(current_user_ns(), option);
186 if (!uid_valid(uid))
187 return -EINVAL;
188 opts->uid = uid;
189 break;
190 case Opt_gid:
191 if (match_int(&args[0], &option))
192 return -EINVAL;
193 gid = make_kgid(current_user_ns(), option);
194 if (!gid_valid(gid))
195 return -EINVAL;
196 opts->gid = gid;
197 break;
198 case Opt_mode:
199 if (match_octal(&args[0], &option))
200 return -EINVAL;
201 opts->mode = option & S_IALLUGO;
202 break;
203 /*
204 * We might like to report bad mount options here;
205 * but traditionally tracefs has ignored all mount options
206 */
207 }
208 }
209
210 return 0;
211}
212
213static int tracefs_apply_options(struct super_block *sb)
214{
215 struct tracefs_fs_info *fsi = sb->s_fs_info;
216 struct inode *inode = sb->s_root->d_inode;
217 struct tracefs_mount_opts *opts = &fsi->mount_opts;
218
219 inode->i_mode &= ~S_IALLUGO;
220 inode->i_mode |= opts->mode;
221
222 inode->i_uid = opts->uid;
223 inode->i_gid = opts->gid;
224
225 return 0;
226}
227
228static int tracefs_remount(struct super_block *sb, int *flags, char *data)
229{
230 int err;
231 struct tracefs_fs_info *fsi = sb->s_fs_info;
232
233 sync_filesystem(sb);
234 err = tracefs_parse_options(data, &fsi->mount_opts);
235 if (err)
236 goto fail;
237
238 tracefs_apply_options(sb);
239
240fail:
241 return err;
242}
243
244static int tracefs_show_options(struct seq_file *m, struct dentry *root)
245{
246 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
247 struct tracefs_mount_opts *opts = &fsi->mount_opts;
248
249 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
250 seq_printf(m, ",uid=%u",
251 from_kuid_munged(&init_user_ns, opts->uid));
252 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
253 seq_printf(m, ",gid=%u",
254 from_kgid_munged(&init_user_ns, opts->gid));
255 if (opts->mode != TRACEFS_DEFAULT_MODE)
256 seq_printf(m, ",mode=%o", opts->mode);
257
258 return 0;
259}
260
261static const struct super_operations tracefs_super_operations = {
262 .statfs = simple_statfs,
263 .remount_fs = tracefs_remount,
264 .show_options = tracefs_show_options,
265};
266
267static int trace_fill_super(struct super_block *sb, void *data, int silent)
268{
269 static const struct tree_descr trace_files[] = {{""}};
270 struct tracefs_fs_info *fsi;
271 int err;
272
273 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
274 sb->s_fs_info = fsi;
275 if (!fsi) {
276 err = -ENOMEM;
277 goto fail;
278 }
279
280 err = tracefs_parse_options(data, &fsi->mount_opts);
281 if (err)
282 goto fail;
283
284 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
285 if (err)
286 goto fail;
287
288 sb->s_op = &tracefs_super_operations;
289
290 tracefs_apply_options(sb);
291
292 return 0;
293
294fail:
295 kfree(fsi);
296 sb->s_fs_info = NULL;
297 return err;
298}
299
300static struct dentry *trace_mount(struct file_system_type *fs_type,
301 int flags, const char *dev_name,
302 void *data)
303{
304 return mount_single(fs_type, flags, data, trace_fill_super);
305}
306
307static struct file_system_type trace_fs_type = {
308 .owner = THIS_MODULE,
309 .name = "tracefs",
310 .mount = trace_mount,
311 .kill_sb = kill_litter_super,
312};
313MODULE_ALIAS_FS("tracefs");
314
315static struct dentry *start_creating(const char *name, struct dentry *parent)
316{
317 struct dentry *dentry;
318 int error;
319
320 pr_debug("tracefs: creating file '%s'\n",name);
321
322 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
323 &tracefs_mount_count);
324 if (error)
325 return ERR_PTR(error);
326
327 /* If the parent is not specified, we create it in the root.
328 * We need the root dentry to do this, which is in the super
329 * block. A pointer to that is in the struct vfsmount that we
330 * have around.
331 */
332 if (!parent)
333 parent = tracefs_mount->mnt_root;
334
335 inode_lock(parent->d_inode);
336 dentry = lookup_one_len(name, parent, strlen(name));
337 if (!IS_ERR(dentry) && dentry->d_inode) {
338 dput(dentry);
339 dentry = ERR_PTR(-EEXIST);
340 }
341
342 if (IS_ERR(dentry)) {
343 inode_unlock(parent->d_inode);
344 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
345 }
346
347 return dentry;
348}
349
350static struct dentry *failed_creating(struct dentry *dentry)
351{
352 inode_unlock(dentry->d_parent->d_inode);
353 dput(dentry);
354 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
355 return NULL;
356}
357
358static struct dentry *end_creating(struct dentry *dentry)
359{
360 inode_unlock(dentry->d_parent->d_inode);
361 return dentry;
362}
363
364/**
365 * tracefs_create_file - create a file in the tracefs filesystem
366 * @name: a pointer to a string containing the name of the file to create.
367 * @mode: the permission that the file should have.
368 * @parent: a pointer to the parent dentry for this file. This should be a
369 * directory dentry if set. If this parameter is NULL, then the
370 * file will be created in the root of the tracefs filesystem.
371 * @data: a pointer to something that the caller will want to get to later
372 * on. The inode.i_private pointer will point to this value on
373 * the open() call.
374 * @fops: a pointer to a struct file_operations that should be used for
375 * this file.
376 *
377 * This is the basic "create a file" function for tracefs. It allows for a
378 * wide range of flexibility in creating a file, or a directory (if you want
379 * to create a directory, the tracefs_create_dir() function is
380 * recommended to be used instead.)
381 *
382 * This function will return a pointer to a dentry if it succeeds. This
383 * pointer must be passed to the tracefs_remove() function when the file is
384 * to be removed (no automatic cleanup happens if your module is unloaded,
385 * you are responsible here.) If an error occurs, %NULL will be returned.
386 *
387 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
388 * returned.
389 */
390struct dentry *tracefs_create_file(const char *name, umode_t mode,
391 struct dentry *parent, void *data,
392 const struct file_operations *fops)
393{
394 struct dentry *dentry;
395 struct inode *inode;
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 */
481struct dentry *tracefs_create_instance_dir(const char *name, struct dentry *parent,
482 int (*mkdir)(const char *name),
483 int (*rmdir)(const char *name))
484{
485 struct dentry *dentry;
486
487 /* Only allow one instance of the instances directory. */
488 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
489 return NULL;
490
491 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
492 if (!dentry)
493 return NULL;
494
495 tracefs_ops.mkdir = mkdir;
496 tracefs_ops.rmdir = rmdir;
497
498 return dentry;
499}
500
501static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
502{
503 int ret = 0;
504
505 if (simple_positive(dentry)) {
506 if (dentry->d_inode) {
507 dget(dentry);
508 switch (dentry->d_inode->i_mode & S_IFMT) {
509 case S_IFDIR:
510 ret = simple_rmdir(parent->d_inode, dentry);
511 break;
512 default:
513 simple_unlink(parent->d_inode, dentry);
514 break;
515 }
516 if (!ret)
517 d_delete(dentry);
518 dput(dentry);
519 }
520 }
521 return ret;
522}
523
524/**
525 * tracefs_remove - removes a file or directory from the tracefs filesystem
526 * @dentry: a pointer to a the dentry of the file or directory to be
527 * removed.
528 *
529 * This function removes a file or directory in tracefs that was previously
530 * created with a call to another tracefs function (like
531 * tracefs_create_file() or variants thereof.)
532 */
533void tracefs_remove(struct dentry *dentry)
534{
535 struct dentry *parent;
536 int ret;
537
538 if (IS_ERR_OR_NULL(dentry))
539 return;
540
541 parent = dentry->d_parent;
542 inode_lock(parent->d_inode);
543 ret = __tracefs_remove(dentry, parent);
544 inode_unlock(parent->d_inode);
545 if (!ret)
546 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
547}
548
549/**
550 * tracefs_remove_recursive - recursively removes a directory
551 * @dentry: a pointer to a the dentry of the directory to be removed.
552 *
553 * This function recursively removes a directory tree in tracefs that
554 * was previously created with a call to another tracefs function
555 * (like tracefs_create_file() or variants thereof.)
556 */
557void tracefs_remove_recursive(struct dentry *dentry)
558{
559 struct dentry *child, *parent;
560
561 if (IS_ERR_OR_NULL(dentry))
562 return;
563
564 parent = dentry;
565 down:
566 inode_lock(parent->d_inode);
567 loop:
568 /*
569 * The parent->d_subdirs is protected by the d_lock. Outside that
570 * lock, the child can be unlinked and set to be freed which can
571 * use the d_u.d_child as the rcu head and corrupt this list.
572 */
573 spin_lock(&parent->d_lock);
574 list_for_each_entry(child, &parent->d_subdirs, d_child) {
575 if (!simple_positive(child))
576 continue;
577
578 /* perhaps simple_empty(child) makes more sense */
579 if (!list_empty(&child->d_subdirs)) {
580 spin_unlock(&parent->d_lock);
581 inode_unlock(parent->d_inode);
582 parent = child;
583 goto down;
584 }
585
586 spin_unlock(&parent->d_lock);
587
588 if (!__tracefs_remove(child, parent))
589 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
590
591 /*
592 * The parent->d_lock protects agaist child from unlinking
593 * from d_subdirs. When releasing the parent->d_lock we can
594 * no longer trust that the next pointer is valid.
595 * Restart the loop. We'll skip this one with the
596 * simple_positive() check.
597 */
598 goto loop;
599 }
600 spin_unlock(&parent->d_lock);
601
602 inode_unlock(parent->d_inode);
603 child = parent;
604 parent = parent->d_parent;
605 inode_lock(parent->d_inode);
606
607 if (child != dentry)
608 /* go up */
609 goto loop;
610
611 if (!__tracefs_remove(child, parent))
612 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
613 inode_unlock(parent->d_inode);
614}
615
616/**
617 * tracefs_initialized - Tells whether tracefs has been registered
618 */
619bool tracefs_initialized(void)
620{
621 return tracefs_registered;
622}
623
624static int __init tracefs_init(void)
625{
626 int retval;
627
628 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
629 if (retval)
630 return -EINVAL;
631
632 retval = register_filesystem(&trace_fs_type);
633 if (!retval)
634 tracefs_registered = true;
635
636 return retval;
637}
638core_initcall(tracefs_init);