xref: /openbmc/linux/fs/tracefs/event_inode.c (revision f2a55d08)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  event_inode.c - part of tracefs, a pseudo file system for activating tracing
4  *
5  *  Copyright (C) 2020-23 VMware Inc, author: Steven Rostedt (VMware) <rostedt@goodmis.org>
6  *  Copyright (C) 2020-23 VMware Inc, author: Ajay Kaher <akaher@vmware.com>
7  *
8  *  eventfs is used to dynamically create inodes and dentries based on the
9  *  meta data provided by the tracing system.
10  *
11  *  eventfs stores the meta-data of files/dirs and holds off on creating
12  *  inodes/dentries of the files. When accessed, the eventfs will create the
13  *  inodes/dentries in a just-in-time (JIT) manner. The eventfs will clean up
14  *  and delete the inodes/dentries when they are no longer referenced.
15  */
16 #include <linux/fsnotify.h>
17 #include <linux/fs.h>
18 #include <linux/namei.h>
19 #include <linux/workqueue.h>
20 #include <linux/security.h>
21 #include <linux/tracefs.h>
22 #include <linux/kref.h>
23 #include <linux/delay.h>
24 #include "internal.h"
25 
26 struct eventfs_inode {
27 	struct list_head	e_top_files;
28 };
29 
30 /*
31  * struct eventfs_file - hold the properties of the eventfs files and
32  *                       directories.
33  * @name:	the name of the file or directory to create
34  * @d_parent:   holds parent's dentry
35  * @dentry:     once accessed holds dentry
36  * @list:	file or directory to be added to parent directory
37  * @ei:		list of files and directories within directory
38  * @fop:	file_operations for file or directory
39  * @iop:	inode_operations for file or directory
40  * @data:	something that the caller will want to get to later on
41  * @mode:	the permission that the file or directory should have
42  */
43 struct eventfs_file {
44 	const char			*name;
45 	struct dentry			*d_parent;
46 	struct dentry			*dentry;
47 	struct list_head		list;
48 	struct eventfs_inode		*ei;
49 	const struct file_operations	*fop;
50 	const struct inode_operations	*iop;
51 	/*
52 	 * Union - used for deletion
53 	 * @del_list:	list of eventfs_file to delete
54 	 * @rcu:	eventfs_file to delete in RCU
55 	 * @is_freed:	node is freed if one of the above is set
56 	 */
57 	union {
58 		struct list_head	del_list;
59 		struct rcu_head		rcu;
60 		unsigned long		is_freed;
61 	};
62 	void				*data;
63 	umode_t				mode;
64 };
65 
66 static DEFINE_MUTEX(eventfs_mutex);
67 DEFINE_STATIC_SRCU(eventfs_srcu);
68 
69 static struct dentry *eventfs_root_lookup(struct inode *dir,
70 					  struct dentry *dentry,
71 					  unsigned int flags);
72 static int dcache_dir_open_wrapper(struct inode *inode, struct file *file);
73 static int eventfs_release(struct inode *inode, struct file *file);
74 
75 static const struct inode_operations eventfs_root_dir_inode_operations = {
76 	.lookup		= eventfs_root_lookup,
77 };
78 
79 static const struct file_operations eventfs_file_operations = {
80 	.open		= dcache_dir_open_wrapper,
81 	.read		= generic_read_dir,
82 	.iterate_shared	= dcache_readdir,
83 	.llseek		= generic_file_llseek,
84 	.release	= eventfs_release,
85 };
86 
87 /**
88  * create_file - create a file in the tracefs filesystem
89  * @name: the name of the file to create.
90  * @mode: the permission that the file should have.
91  * @parent: parent dentry for this file.
92  * @data: something that the caller will want to get to later on.
93  * @fop: struct file_operations that should be used for this file.
94  *
95  * This is the basic "create a file" function for tracefs.  It allows for a
96  * wide range of flexibility in creating a file.
97  *
98  * This function will return a pointer to a dentry if it succeeds.  This
99  * pointer must be passed to the tracefs_remove() function when the file is
100  * to be removed (no automatic cleanup happens if your module is unloaded,
101  * you are responsible here.)  If an error occurs, %NULL will be returned.
102  *
103  * If tracefs is not enabled in the kernel, the value -%ENODEV will be
104  * returned.
105  */
106 static struct dentry *create_file(const char *name, umode_t mode,
107 				  struct dentry *parent, void *data,
108 				  const struct file_operations *fop)
109 {
110 	struct tracefs_inode *ti;
111 	struct dentry *dentry;
112 	struct inode *inode;
113 
114 	if (!(mode & S_IFMT))
115 		mode |= S_IFREG;
116 
117 	if (WARN_ON_ONCE(!S_ISREG(mode)))
118 		return NULL;
119 
120 	dentry = eventfs_start_creating(name, parent);
121 
122 	if (IS_ERR(dentry))
123 		return dentry;
124 
125 	inode = tracefs_get_inode(dentry->d_sb);
126 	if (unlikely(!inode))
127 		return eventfs_failed_creating(dentry);
128 
129 	inode->i_mode = mode;
130 	inode->i_fop = fop;
131 	inode->i_private = data;
132 
133 	ti = get_tracefs(inode);
134 	ti->flags |= TRACEFS_EVENT_INODE;
135 	d_instantiate(dentry, inode);
136 	fsnotify_create(dentry->d_parent->d_inode, dentry);
137 	return eventfs_end_creating(dentry);
138 };
139 
140 /**
141  * create_dir - create a dir in the tracefs filesystem
142  * @name: the name of the file to create.
143  * @parent: parent dentry for this file.
144  * @data: something that the caller will want to get to later on.
145  *
146  * This is the basic "create a dir" function for eventfs.  It allows for a
147  * wide range of flexibility in creating a dir.
148  *
149  * This function will return a pointer to a dentry if it succeeds.  This
150  * pointer must be passed to the tracefs_remove() function when the file is
151  * to be removed (no automatic cleanup happens if your module is unloaded,
152  * you are responsible here.)  If an error occurs, %NULL will be returned.
153  *
154  * If tracefs is not enabled in the kernel, the value -%ENODEV will be
155  * returned.
156  */
157 static struct dentry *create_dir(const char *name, struct dentry *parent, void *data)
158 {
159 	struct tracefs_inode *ti;
160 	struct dentry *dentry;
161 	struct inode *inode;
162 
163 	dentry = eventfs_start_creating(name, parent);
164 	if (IS_ERR(dentry))
165 		return dentry;
166 
167 	inode = tracefs_get_inode(dentry->d_sb);
168 	if (unlikely(!inode))
169 		return eventfs_failed_creating(dentry);
170 
171 	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
172 	inode->i_op = &eventfs_root_dir_inode_operations;
173 	inode->i_fop = &eventfs_file_operations;
174 	inode->i_private = data;
175 
176 	ti = get_tracefs(inode);
177 	ti->flags |= TRACEFS_EVENT_INODE;
178 
179 	inc_nlink(inode);
180 	d_instantiate(dentry, inode);
181 	inc_nlink(dentry->d_parent->d_inode);
182 	fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
183 	return eventfs_end_creating(dentry);
184 }
185 
186 /**
187  * eventfs_set_ef_status_free - set the ef->status to free
188  * @ti: the tracefs_inode of the dentry
189  * @dentry: dentry who's status to be freed
190  *
191  * eventfs_set_ef_status_free will be called if no more
192  * references remain
193  */
194 void eventfs_set_ef_status_free(struct tracefs_inode *ti, struct dentry *dentry)
195 {
196 	struct tracefs_inode *ti_parent;
197 	struct eventfs_inode *ei;
198 	struct eventfs_file *ef, *tmp;
199 
200 	/* The top level events directory may be freed by this */
201 	if (unlikely(ti->flags & TRACEFS_EVENT_TOP_INODE)) {
202 		LIST_HEAD(ef_del_list);
203 
204 		mutex_lock(&eventfs_mutex);
205 
206 		ei = ti->private;
207 
208 		/* Record all the top level files */
209 		list_for_each_entry_srcu(ef, &ei->e_top_files, list,
210 					 lockdep_is_held(&eventfs_mutex)) {
211 			list_add_tail(&ef->del_list, &ef_del_list);
212 		}
213 
214 		/* Nothing should access this, but just in case! */
215 		ti->private = NULL;
216 
217 		mutex_unlock(&eventfs_mutex);
218 
219 		/* Now safely free the top level files and their children */
220 		list_for_each_entry_safe(ef, tmp, &ef_del_list, del_list) {
221 			list_del(&ef->del_list);
222 			eventfs_remove(ef);
223 		}
224 
225 		kfree(ei);
226 		return;
227 	}
228 
229 	mutex_lock(&eventfs_mutex);
230 
231 	ti_parent = get_tracefs(dentry->d_parent->d_inode);
232 	if (!ti_parent || !(ti_parent->flags & TRACEFS_EVENT_INODE))
233 		goto out;
234 
235 	ef = dentry->d_fsdata;
236 	if (!ef)
237 		goto out;
238 
239 	/*
240 	 * If ef was freed, then the LSB bit is set for d_fsdata.
241 	 * But this should not happen, as it should still have a
242 	 * ref count that prevents it. Warn in case it does.
243 	 */
244 	if (WARN_ON_ONCE((unsigned long)ef & 1))
245 		goto out;
246 
247 	dentry->d_fsdata = NULL;
248 	ef->dentry = NULL;
249 out:
250 	mutex_unlock(&eventfs_mutex);
251 }
252 
253 /**
254  * eventfs_post_create_dir - post create dir routine
255  * @ef: eventfs_file of recently created dir
256  *
257  * Map the meta-data of files within an eventfs dir to their parent dentry
258  */
259 static void eventfs_post_create_dir(struct eventfs_file *ef)
260 {
261 	struct eventfs_file *ef_child;
262 	struct tracefs_inode *ti;
263 
264 	/* srcu lock already held */
265 	/* fill parent-child relation */
266 	list_for_each_entry_srcu(ef_child, &ef->ei->e_top_files, list,
267 				 srcu_read_lock_held(&eventfs_srcu)) {
268 		ef_child->d_parent = ef->dentry;
269 	}
270 
271 	ti = get_tracefs(ef->dentry->d_inode);
272 	ti->private = ef->ei;
273 }
274 
275 /**
276  * create_dentry - helper function to create dentry
277  * @ef: eventfs_file of file or directory to create
278  * @parent: parent dentry
279  * @lookup: true if called from lookup routine
280  *
281  * Used to create a dentry for file/dir, executes post dentry creation routine
282  */
283 static struct dentry *
284 create_dentry(struct eventfs_file *ef, struct dentry *parent, bool lookup)
285 {
286 	bool invalidate = false;
287 	struct dentry *dentry;
288 
289 	mutex_lock(&eventfs_mutex);
290 	if (ef->is_freed) {
291 		mutex_unlock(&eventfs_mutex);
292 		return NULL;
293 	}
294 	if (ef->dentry) {
295 		dentry = ef->dentry;
296 		/* On dir open, up the ref count */
297 		if (!lookup)
298 			dget(dentry);
299 		mutex_unlock(&eventfs_mutex);
300 		return dentry;
301 	}
302 	mutex_unlock(&eventfs_mutex);
303 
304 	if (!lookup)
305 		inode_lock(parent->d_inode);
306 
307 	if (ef->ei)
308 		dentry = create_dir(ef->name, parent, ef->data);
309 	else
310 		dentry = create_file(ef->name, ef->mode, parent,
311 				     ef->data, ef->fop);
312 
313 	if (!lookup)
314 		inode_unlock(parent->d_inode);
315 
316 	mutex_lock(&eventfs_mutex);
317 	if (IS_ERR_OR_NULL(dentry)) {
318 		/* If the ef was already updated get it */
319 		dentry = ef->dentry;
320 		if (dentry && !lookup)
321 			dget(dentry);
322 		mutex_unlock(&eventfs_mutex);
323 		return dentry;
324 	}
325 
326 	if (!ef->dentry && !ef->is_freed) {
327 		ef->dentry = dentry;
328 		if (ef->ei)
329 			eventfs_post_create_dir(ef);
330 		dentry->d_fsdata = ef;
331 	} else {
332 		/* A race here, should try again (unless freed) */
333 		invalidate = true;
334 
335 		/*
336 		 * Should never happen unless we get here due to being freed.
337 		 * Otherwise it means two dentries exist with the same name.
338 		 */
339 		WARN_ON_ONCE(!ef->is_freed);
340 	}
341 	mutex_unlock(&eventfs_mutex);
342 	if (invalidate)
343 		d_invalidate(dentry);
344 
345 	if (lookup || invalidate)
346 		dput(dentry);
347 
348 	return invalidate ? NULL : dentry;
349 }
350 
351 static bool match_event_file(struct eventfs_file *ef, const char *name)
352 {
353 	bool ret;
354 
355 	mutex_lock(&eventfs_mutex);
356 	ret = !ef->is_freed && strcmp(ef->name, name) == 0;
357 	mutex_unlock(&eventfs_mutex);
358 
359 	return ret;
360 }
361 
362 /**
363  * eventfs_root_lookup - lookup routine to create file/dir
364  * @dir: in which a lookup is being done
365  * @dentry: file/dir dentry
366  * @flags: to pass as flags parameter to simple lookup
367  *
368  * Used to create a dynamic file/dir within @dir. Use the eventfs_inode
369  * list of meta data to find the information needed to create the file/dir.
370  */
371 static struct dentry *eventfs_root_lookup(struct inode *dir,
372 					  struct dentry *dentry,
373 					  unsigned int flags)
374 {
375 	struct tracefs_inode *ti;
376 	struct eventfs_inode *ei;
377 	struct eventfs_file *ef;
378 	struct dentry *ret = NULL;
379 	int idx;
380 
381 	ti = get_tracefs(dir);
382 	if (!(ti->flags & TRACEFS_EVENT_INODE))
383 		return NULL;
384 
385 	ei = ti->private;
386 	idx = srcu_read_lock(&eventfs_srcu);
387 	list_for_each_entry_srcu(ef, &ei->e_top_files, list,
388 				 srcu_read_lock_held(&eventfs_srcu)) {
389 		if (!match_event_file(ef, dentry->d_name.name))
390 			continue;
391 		ret = simple_lookup(dir, dentry, flags);
392 		create_dentry(ef, ef->d_parent, true);
393 		break;
394 	}
395 	srcu_read_unlock(&eventfs_srcu, idx);
396 	return ret;
397 }
398 
399 /**
400  * eventfs_release - called to release eventfs file/dir
401  * @inode: inode to be released
402  * @file: file to be released (not used)
403  */
404 static int eventfs_release(struct inode *inode, struct file *file)
405 {
406 	struct tracefs_inode *ti;
407 	struct eventfs_inode *ei;
408 	struct eventfs_file *ef;
409 	struct dentry *dentry;
410 	int idx;
411 
412 	ti = get_tracefs(inode);
413 	if (!(ti->flags & TRACEFS_EVENT_INODE))
414 		return -EINVAL;
415 
416 	ei = ti->private;
417 	idx = srcu_read_lock(&eventfs_srcu);
418 	list_for_each_entry_srcu(ef, &ei->e_top_files, list,
419 				 srcu_read_lock_held(&eventfs_srcu)) {
420 		mutex_lock(&eventfs_mutex);
421 		dentry = ef->dentry;
422 		mutex_unlock(&eventfs_mutex);
423 		if (dentry)
424 			dput(dentry);
425 	}
426 	srcu_read_unlock(&eventfs_srcu, idx);
427 	return dcache_dir_close(inode, file);
428 }
429 
430 /**
431  * dcache_dir_open_wrapper - eventfs open wrapper
432  * @inode: not used
433  * @file: dir to be opened (to create its child)
434  *
435  * Used to dynamically create the file/dir within @file. @file is really a
436  * directory and all the files/dirs of the children within @file will be
437  * created. If any of the files/dirs have already been created, their
438  * reference count will be incremented.
439  */
440 static int dcache_dir_open_wrapper(struct inode *inode, struct file *file)
441 {
442 	struct tracefs_inode *ti;
443 	struct eventfs_inode *ei;
444 	struct eventfs_file *ef;
445 	struct dentry *dentry = file_dentry(file);
446 	struct inode *f_inode = file_inode(file);
447 	int idx;
448 
449 	ti = get_tracefs(f_inode);
450 	if (!(ti->flags & TRACEFS_EVENT_INODE))
451 		return -EINVAL;
452 
453 	ei = ti->private;
454 	idx = srcu_read_lock(&eventfs_srcu);
455 	list_for_each_entry_srcu(ef, &ei->e_top_files, list,
456 				 srcu_read_lock_held(&eventfs_srcu)) {
457 		create_dentry(ef, dentry, false);
458 	}
459 	srcu_read_unlock(&eventfs_srcu, idx);
460 	return dcache_dir_open(inode, file);
461 }
462 
463 /**
464  * eventfs_prepare_ef - helper function to prepare eventfs_file
465  * @name: the name of the file/directory to create.
466  * @mode: the permission that the file should have.
467  * @fop: struct file_operations that should be used for this file/directory.
468  * @iop: struct inode_operations that should be used for this file/directory.
469  * @data: something that the caller will want to get to later on. The
470  *        inode.i_private pointer will point to this value on the open() call.
471  *
472  * This function allocates and fills the eventfs_file structure.
473  */
474 static struct eventfs_file *eventfs_prepare_ef(const char *name, umode_t mode,
475 					const struct file_operations *fop,
476 					const struct inode_operations *iop,
477 					void *data)
478 {
479 	struct eventfs_file *ef;
480 
481 	ef = kzalloc(sizeof(*ef), GFP_KERNEL);
482 	if (!ef)
483 		return ERR_PTR(-ENOMEM);
484 
485 	ef->name = kstrdup(name, GFP_KERNEL);
486 	if (!ef->name) {
487 		kfree(ef);
488 		return ERR_PTR(-ENOMEM);
489 	}
490 
491 	if (S_ISDIR(mode)) {
492 		ef->ei = kzalloc(sizeof(*ef->ei), GFP_KERNEL);
493 		if (!ef->ei) {
494 			kfree(ef->name);
495 			kfree(ef);
496 			return ERR_PTR(-ENOMEM);
497 		}
498 		INIT_LIST_HEAD(&ef->ei->e_top_files);
499 	} else {
500 		ef->ei = NULL;
501 	}
502 
503 	ef->iop = iop;
504 	ef->fop = fop;
505 	ef->mode = mode;
506 	ef->data = data;
507 	return ef;
508 }
509 
510 /**
511  * eventfs_create_events_dir - create the trace event structure
512  * @name: the name of the directory to create.
513  * @parent: parent dentry for this file.  This should be a directory dentry
514  *          if set.  If this parameter is NULL, then the directory will be
515  *          created in the root of the tracefs filesystem.
516  *
517  * This function creates the top of the trace event directory.
518  */
519 struct dentry *eventfs_create_events_dir(const char *name,
520 					 struct dentry *parent)
521 {
522 	struct dentry *dentry = tracefs_start_creating(name, parent);
523 	struct eventfs_inode *ei;
524 	struct tracefs_inode *ti;
525 	struct inode *inode;
526 
527 	if (security_locked_down(LOCKDOWN_TRACEFS))
528 		return NULL;
529 
530 	if (IS_ERR(dentry))
531 		return dentry;
532 
533 	ei = kzalloc(sizeof(*ei), GFP_KERNEL);
534 	if (!ei)
535 		return ERR_PTR(-ENOMEM);
536 	inode = tracefs_get_inode(dentry->d_sb);
537 	if (unlikely(!inode)) {
538 		kfree(ei);
539 		tracefs_failed_creating(dentry);
540 		return ERR_PTR(-ENOMEM);
541 	}
542 
543 	INIT_LIST_HEAD(&ei->e_top_files);
544 
545 	ti = get_tracefs(inode);
546 	ti->flags |= TRACEFS_EVENT_INODE | TRACEFS_EVENT_TOP_INODE;
547 	ti->private = ei;
548 
549 	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
550 	inode->i_op = &eventfs_root_dir_inode_operations;
551 	inode->i_fop = &eventfs_file_operations;
552 
553 	/* directory inodes start off with i_nlink == 2 (for "." entry) */
554 	inc_nlink(inode);
555 	d_instantiate(dentry, inode);
556 	inc_nlink(dentry->d_parent->d_inode);
557 	fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
558 	return tracefs_end_creating(dentry);
559 }
560 
561 /**
562  * eventfs_add_subsystem_dir - add eventfs subsystem_dir to list to create later
563  * @name: the name of the file to create.
564  * @parent: parent dentry for this dir.
565  *
566  * This function adds eventfs subsystem dir to list.
567  * And all these dirs are created on the fly when they are looked up,
568  * and the dentry and inodes will be removed when they are done.
569  */
570 struct eventfs_file *eventfs_add_subsystem_dir(const char *name,
571 					       struct dentry *parent)
572 {
573 	struct tracefs_inode *ti_parent;
574 	struct eventfs_inode *ei_parent;
575 	struct eventfs_file *ef;
576 
577 	if (security_locked_down(LOCKDOWN_TRACEFS))
578 		return NULL;
579 
580 	if (!parent)
581 		return ERR_PTR(-EINVAL);
582 
583 	ti_parent = get_tracefs(parent->d_inode);
584 	ei_parent = ti_parent->private;
585 
586 	ef = eventfs_prepare_ef(name, S_IFDIR, NULL, NULL, NULL);
587 	if (IS_ERR(ef))
588 		return ef;
589 
590 	mutex_lock(&eventfs_mutex);
591 	list_add_tail(&ef->list, &ei_parent->e_top_files);
592 	ef->d_parent = parent;
593 	mutex_unlock(&eventfs_mutex);
594 	return ef;
595 }
596 
597 /**
598  * eventfs_add_dir - add eventfs dir to list to create later
599  * @name: the name of the file to create.
600  * @ef_parent: parent eventfs_file for this dir.
601  *
602  * This function adds eventfs dir to list.
603  * And all these dirs are created on the fly when they are looked up,
604  * and the dentry and inodes will be removed when they are done.
605  */
606 struct eventfs_file *eventfs_add_dir(const char *name,
607 				     struct eventfs_file *ef_parent)
608 {
609 	struct eventfs_file *ef;
610 
611 	if (security_locked_down(LOCKDOWN_TRACEFS))
612 		return NULL;
613 
614 	if (!ef_parent)
615 		return ERR_PTR(-EINVAL);
616 
617 	ef = eventfs_prepare_ef(name, S_IFDIR, NULL, NULL, NULL);
618 	if (IS_ERR(ef))
619 		return ef;
620 
621 	mutex_lock(&eventfs_mutex);
622 	list_add_tail(&ef->list, &ef_parent->ei->e_top_files);
623 	ef->d_parent = ef_parent->dentry;
624 	mutex_unlock(&eventfs_mutex);
625 	return ef;
626 }
627 
628 /**
629  * eventfs_add_events_file - add the data needed to create a file for later reference
630  * @name: the name of the file to create.
631  * @mode: the permission that the file should have.
632  * @parent: parent dentry for this file.
633  * @data: something that the caller will want to get to later on.
634  * @fop: struct file_operations that should be used for this file.
635  *
636  * This function is used to add the information needed to create a
637  * dentry/inode within the top level events directory. The file created
638  * will have the @mode permissions. The @data will be used to fill the
639  * inode.i_private when the open() call is done. The dentry and inodes are
640  * all created when they are referenced, and removed when they are no
641  * longer referenced.
642  */
643 int eventfs_add_events_file(const char *name, umode_t mode,
644 			 struct dentry *parent, void *data,
645 			 const struct file_operations *fop)
646 {
647 	struct tracefs_inode *ti;
648 	struct eventfs_inode *ei;
649 	struct eventfs_file *ef;
650 
651 	if (security_locked_down(LOCKDOWN_TRACEFS))
652 		return -ENODEV;
653 
654 	if (!parent)
655 		return -EINVAL;
656 
657 	if (!(mode & S_IFMT))
658 		mode |= S_IFREG;
659 
660 	if (!parent->d_inode)
661 		return -EINVAL;
662 
663 	ti = get_tracefs(parent->d_inode);
664 	if (!(ti->flags & TRACEFS_EVENT_INODE))
665 		return -EINVAL;
666 
667 	ei = ti->private;
668 	ef = eventfs_prepare_ef(name, mode, fop, NULL, data);
669 
670 	if (IS_ERR(ef))
671 		return -ENOMEM;
672 
673 	mutex_lock(&eventfs_mutex);
674 	list_add_tail(&ef->list, &ei->e_top_files);
675 	ef->d_parent = parent;
676 	mutex_unlock(&eventfs_mutex);
677 	return 0;
678 }
679 
680 /**
681  * eventfs_add_file - add eventfs file to list to create later
682  * @name: the name of the file to create.
683  * @mode: the permission that the file should have.
684  * @ef_parent: parent eventfs_file for this file.
685  * @data: something that the caller will want to get to later on.
686  * @fop: struct file_operations that should be used for this file.
687  *
688  * This function is used to add the information needed to create a
689  * file within a subdirectory of the events directory. The file created
690  * will have the @mode permissions. The @data will be used to fill the
691  * inode.i_private when the open() call is done. The dentry and inodes are
692  * all created when they are referenced, and removed when they are no
693  * longer referenced.
694  */
695 int eventfs_add_file(const char *name, umode_t mode,
696 		     struct eventfs_file *ef_parent,
697 		     void *data,
698 		     const struct file_operations *fop)
699 {
700 	struct eventfs_file *ef;
701 
702 	if (security_locked_down(LOCKDOWN_TRACEFS))
703 		return -ENODEV;
704 
705 	if (!ef_parent)
706 		return -EINVAL;
707 
708 	if (!(mode & S_IFMT))
709 		mode |= S_IFREG;
710 
711 	ef = eventfs_prepare_ef(name, mode, fop, NULL, data);
712 	if (IS_ERR(ef))
713 		return -ENOMEM;
714 
715 	mutex_lock(&eventfs_mutex);
716 	list_add_tail(&ef->list, &ef_parent->ei->e_top_files);
717 	ef->d_parent = ef_parent->dentry;
718 	mutex_unlock(&eventfs_mutex);
719 	return 0;
720 }
721 
722 static void free_ef(struct rcu_head *head)
723 {
724 	struct eventfs_file *ef = container_of(head, struct eventfs_file, rcu);
725 
726 	kfree(ef->name);
727 	kfree(ef->ei);
728 	kfree(ef);
729 }
730 
731 /**
732  * eventfs_remove_rec - remove eventfs dir or file from list
733  * @ef: eventfs_file to be removed.
734  * @head: to create list of eventfs_file to be deleted
735  * @level: to check recursion depth
736  *
737  * The helper function eventfs_remove_rec() is used to clean up and free the
738  * associated data from eventfs for both of the added functions.
739  */
740 static void eventfs_remove_rec(struct eventfs_file *ef, struct list_head *head, int level)
741 {
742 	struct eventfs_file *ef_child;
743 
744 	if (!ef)
745 		return;
746 	/*
747 	 * Check recursion depth. It should never be greater than 3:
748 	 * 0 - events/
749 	 * 1 - events/group/
750 	 * 2 - events/group/event/
751 	 * 3 - events/group/event/file
752 	 */
753 	if (WARN_ON_ONCE(level > 3))
754 		return;
755 
756 	if (ef->ei) {
757 		/* search for nested folders or files */
758 		list_for_each_entry_srcu(ef_child, &ef->ei->e_top_files, list,
759 					 lockdep_is_held(&eventfs_mutex)) {
760 			eventfs_remove_rec(ef_child, head, level + 1);
761 		}
762 	}
763 
764 	list_del_rcu(&ef->list);
765 	list_add_tail(&ef->del_list, head);
766 }
767 
768 /**
769  * eventfs_remove - remove eventfs dir or file from list
770  * @ef: eventfs_file to be removed.
771  *
772  * This function acquire the eventfs_mutex lock and call eventfs_remove_rec()
773  */
774 void eventfs_remove(struct eventfs_file *ef)
775 {
776 	struct eventfs_file *tmp;
777 	LIST_HEAD(ef_del_list);
778 	struct dentry *dentry_list = NULL;
779 	struct dentry *dentry;
780 
781 	if (!ef)
782 		return;
783 
784 	mutex_lock(&eventfs_mutex);
785 	eventfs_remove_rec(ef, &ef_del_list, 0);
786 	list_for_each_entry_safe(ef, tmp, &ef_del_list, del_list) {
787 		if (ef->dentry) {
788 			unsigned long ptr = (unsigned long)dentry_list;
789 
790 			/* Keep the dentry from being freed yet */
791 			dget(ef->dentry);
792 
793 			/*
794 			 * Paranoid: The dget() above should prevent the dentry
795 			 * from being freed and calling eventfs_set_ef_status_free().
796 			 * But just in case, set the link list LSB pointer to 1
797 			 * and have eventfs_set_ef_status_free() check that to
798 			 * make sure that if it does happen, it will not think
799 			 * the d_fsdata is an event_file.
800 			 *
801 			 * For this to work, no event_file should be allocated
802 			 * on a odd space, as the ef should always be allocated
803 			 * to be at least word aligned. Check for that too.
804 			 */
805 			WARN_ON_ONCE(ptr & 1);
806 
807 			ef->dentry->d_fsdata = (void *)(ptr | 1);
808 			dentry_list = ef->dentry;
809 			ef->dentry = NULL;
810 		}
811 		call_srcu(&eventfs_srcu, &ef->rcu, free_ef);
812 	}
813 	mutex_unlock(&eventfs_mutex);
814 
815 	while (dentry_list) {
816 		unsigned long ptr;
817 
818 		dentry = dentry_list;
819 		ptr = (unsigned long)dentry->d_fsdata & ~1UL;
820 		dentry_list = (struct dentry *)ptr;
821 		dentry->d_fsdata = NULL;
822 		d_invalidate(dentry);
823 		mutex_lock(&eventfs_mutex);
824 		/* dentry should now have at least a single reference */
825 		WARN_ONCE((int)d_count(dentry) < 1,
826 			  "dentry %p less than one reference (%d) after invalidate\n",
827 			  dentry, d_count(dentry));
828 		mutex_unlock(&eventfs_mutex);
829 		dput(dentry);
830 	}
831 }
832 
833 /**
834  * eventfs_remove_events_dir - remove eventfs dir or file from list
835  * @dentry: events's dentry to be removed.
836  *
837  * This function remove events main directory
838  */
839 void eventfs_remove_events_dir(struct dentry *dentry)
840 {
841 	struct tracefs_inode *ti;
842 
843 	if (!dentry || !dentry->d_inode)
844 		return;
845 
846 	ti = get_tracefs(dentry->d_inode);
847 	if (!ti || !(ti->flags & TRACEFS_EVENT_INODE))
848 		return;
849 
850 	d_invalidate(dentry);
851 	dput(dentry);
852 }
853