1.. SPDX-License-Identifier: GPL-2.0
2
3======================
4The seq_file Interface
5======================
6
7	Copyright 2003 Jonathan Corbet <corbet@lwn.net>
8
9	This file is originally from the LWN.net Driver Porting series at
10	https://lwn.net/Articles/driver-porting/
11
12
13There are numerous ways for a device driver (or other kernel component) to
14provide information to the user or system administrator.  One useful
15technique is the creation of virtual files, in debugfs, /proc or elsewhere.
16Virtual files can provide human-readable output that is easy to get at
17without any special utility programs; they can also make life easier for
18script writers. It is not surprising that the use of virtual files has
19grown over the years.
20
21Creating those files correctly has always been a bit of a challenge,
22however. It is not that hard to make a virtual file which returns a
23string. But life gets trickier if the output is long - anything greater
24than an application is likely to read in a single operation.  Handling
25multiple reads (and seeks) requires careful attention to the reader's
26position within the virtual file - that position is, likely as not, in the
27middle of a line of output. The kernel has traditionally had a number of
28implementations that got this wrong.
29
30The 2.6 kernel contains a set of functions (implemented by Alexander Viro)
31which are designed to make it easy for virtual file creators to get it
32right.
33
34The seq_file interface is available via <linux/seq_file.h>. There are
35three aspects to seq_file:
36
37     * An iterator interface which lets a virtual file implementation
38       step through the objects it is presenting.
39
40     * Some utility functions for formatting objects for output without
41       needing to worry about things like output buffers.
42
43     * A set of canned file_operations which implement most operations on
44       the virtual file.
45
46We'll look at the seq_file interface via an extremely simple example: a
47loadable module which creates a file called /proc/sequence. The file, when
48read, simply produces a set of increasing integer values, one per line. The
49sequence will continue until the user loses patience and finds something
50better to do. The file is seekable, in that one can do something like the
51following::
52
53    dd if=/proc/sequence of=out1 count=1
54    dd if=/proc/sequence skip=1 of=out2 count=1
55
56Then concatenate the output files out1 and out2 and get the right
57result. Yes, it is a thoroughly useless module, but the point is to show
58how the mechanism works without getting lost in other details.  (Those
59wanting to see the full source for this module can find it at
60https://lwn.net/Articles/22359/).
61
62Deprecated create_proc_entry
63============================
64
65Note that the above article uses create_proc_entry which was removed in
66kernel 3.10. Current versions require the following update::
67
68    -	entry = create_proc_entry("sequence", 0, NULL);
69    -	if (entry)
70    -		entry->proc_fops = &ct_file_ops;
71    +	entry = proc_create("sequence", 0, NULL, &ct_file_ops);
72
73The iterator interface
74======================
75
76Modules implementing a virtual file with seq_file must implement an
77iterator object that allows stepping through the data of interest
78during a "session" (roughly one read() system call).  If the iterator
79is able to move to a specific position - like the file they implement,
80though with freedom to map the position number to a sequence location
81in whatever way is convenient - the iterator need only exist
82transiently during a session.  If the iterator cannot easily find a
83numerical position but works well with a first/next interface, the
84iterator can be stored in the private data area and continue from one
85session to the next.
86
87A seq_file implementation that is formatting firewall rules from a
88table, for example, could provide a simple iterator that interprets
89position N as the Nth rule in the chain.  A seq_file implementation
90that presents the content of a, potentially volatile, linked list
91might record a pointer into that list, providing that can be done
92without risk of the current location being removed.
93
94Positioning can thus be done in whatever way makes the most sense for
95the generator of the data, which need not be aware of how a position
96translates to an offset in the virtual file. The one obvious exception
97is that a position of zero should indicate the beginning of the file.
98
99The /proc/sequence iterator just uses the count of the next number it
100will output as its position.
101
102Four functions must be implemented to make the iterator work. The
103first, called start(), starts a session and takes a position as an
104argument, returning an iterator which will start reading at that
105position.  The pos passed to start() will always be either zero, or
106the most recent pos used in the previous session.
107
108For our simple sequence example,
109the start() function looks like::
110
111	static void *ct_seq_start(struct seq_file *s, loff_t *pos)
112	{
113	        loff_t *spos = kmalloc(sizeof(loff_t), GFP_KERNEL);
114	        if (! spos)
115	                return NULL;
116	        *spos = *pos;
117	        return spos;
118	}
119
120The entire data structure for this iterator is a single loff_t value
121holding the current position. There is no upper bound for the sequence
122iterator, but that will not be the case for most other seq_file
123implementations; in most cases the start() function should check for a
124"past end of file" condition and return NULL if need be.
125
126For more complicated applications, the private field of the seq_file
127structure can be used to hold state from session to session.  There is
128also a special value which can be returned by the start() function
129called SEQ_START_TOKEN; it can be used if you wish to instruct your
130show() function (described below) to print a header at the top of the
131output. SEQ_START_TOKEN should only be used if the offset is zero,
132however.  SEQ_START_TOKEN has no special meaning to the core seq_file
133code.  It is provided as a convenience for a start() funciton to
134communicate with the next() and show() functions.
135
136The next function to implement is called, amazingly, next(); its job is to
137move the iterator forward to the next position in the sequence.  The
138example module can simply increment the position by one; more useful
139modules will do what is needed to step through some data structure. The
140next() function returns a new iterator, or NULL if the sequence is
141complete. Here's the example version::
142
143	static void *ct_seq_next(struct seq_file *s, void *v, loff_t *pos)
144	{
145	        loff_t *spos = v;
146	        *pos = ++*spos;
147	        return spos;
148	}
149
150The next() function should set ``*pos`` to a value that start() can use
151to find the new location in the sequence.  When the iterator is being
152stored in the private data area, rather than being reinitialized on each
153start(), it might seem sufficient to simply set ``*pos`` to any non-zero
154value (zero always tells start() to restart the sequence).  This is not
155sufficient due to historical problems.
156
157Historically, many next() functions have *not* updated ``*pos`` at
158end-of-file.  If the value is then used by start() to initialise the
159iterator, this can result in corner cases where the last entry in the
160sequence is reported twice in the file.  In order to discourage this bug
161from being resurrected, the core seq_file code now produces a warning if
162a next() function does not change the value of ``*pos``.  Consequently a
163next() function *must* change the value of ``*pos``, and of course must
164set it to a non-zero value.
165
166The stop() function closes a session; its job, of course, is to clean
167up. If dynamic memory is allocated for the iterator, stop() is the
168place to free it; if a lock was taken by start(), stop() must release
169that lock.  The value that ``*pos`` was set to by the last next() call
170before stop() is remembered, and used for the first start() call of
171the next session unless lseek() has been called on the file; in that
172case next start() will be asked to start at position zero::
173
174	static void ct_seq_stop(struct seq_file *s, void *v)
175	{
176	        kfree(v);
177	}
178
179Finally, the show() function should format the object currently pointed to
180by the iterator for output.  The example module's show() function is::
181
182	static int ct_seq_show(struct seq_file *s, void *v)
183	{
184	        loff_t *spos = v;
185	        seq_printf(s, "%lld\n", (long long)*spos);
186	        return 0;
187	}
188
189If all is well, the show() function should return zero.  A negative error
190code in the usual manner indicates that something went wrong; it will be
191passed back to user space.  This function can also return SEQ_SKIP, which
192causes the current item to be skipped; if the show() function has already
193generated output before returning SEQ_SKIP, that output will be dropped.
194
195We will look at seq_printf() in a moment. But first, the definition of the
196seq_file iterator is finished by creating a seq_operations structure with
197the four functions we have just defined::
198
199	static const struct seq_operations ct_seq_ops = {
200	        .start = ct_seq_start,
201	        .next  = ct_seq_next,
202	        .stop  = ct_seq_stop,
203	        .show  = ct_seq_show
204	};
205
206This structure will be needed to tie our iterator to the /proc file in
207a little bit.
208
209It's worth noting that the iterator value returned by start() and
210manipulated by the other functions is considered to be completely opaque by
211the seq_file code. It can thus be anything that is useful in stepping
212through the data to be output. Counters can be useful, but it could also be
213a direct pointer into an array or linked list. Anything goes, as long as
214the programmer is aware that things can happen between calls to the
215iterator function. However, the seq_file code (by design) will not sleep
216between the calls to start() and stop(), so holding a lock during that time
217is a reasonable thing to do. The seq_file code will also avoid taking any
218other locks while the iterator is active.
219
220The iterater value returned by start() or next() is guaranteed to be
221passed to a subsequent next() or stop() call.  This allows resources
222such as locks that were taken to be reliably released.  There is *no*
223guarantee that the iterator will be passed to show(), though in practice
224it often will be.
225
226
227Formatted output
228================
229
230The seq_file code manages positioning within the output created by the
231iterator and getting it into the user's buffer. But, for that to work, that
232output must be passed to the seq_file code. Some utility functions have
233been defined which make this task easy.
234
235Most code will simply use seq_printf(), which works pretty much like
236printk(), but which requires the seq_file pointer as an argument.
237
238For straight character output, the following functions may be used::
239
240	seq_putc(struct seq_file *m, char c);
241	seq_puts(struct seq_file *m, const char *s);
242	seq_escape(struct seq_file *m, const char *s, const char *esc);
243
244The first two output a single character and a string, just like one would
245expect. seq_escape() is like seq_puts(), except that any character in s
246which is in the string esc will be represented in octal form in the output.
247
248There are also a pair of functions for printing filenames::
249
250	int seq_path(struct seq_file *m, const struct path *path,
251		     const char *esc);
252	int seq_path_root(struct seq_file *m, const struct path *path,
253			  const struct path *root, const char *esc)
254
255Here, path indicates the file of interest, and esc is a set of characters
256which should be escaped in the output.  A call to seq_path() will output
257the path relative to the current process's filesystem root.  If a different
258root is desired, it can be used with seq_path_root().  If it turns out that
259path cannot be reached from root, seq_path_root() returns SEQ_SKIP.
260
261A function producing complicated output may want to check::
262
263	bool seq_has_overflowed(struct seq_file *m);
264
265and avoid further seq_<output> calls if true is returned.
266
267A true return from seq_has_overflowed means that the seq_file buffer will
268be discarded and the seq_show function will attempt to allocate a larger
269buffer and retry printing.
270
271
272Making it all work
273==================
274
275So far, we have a nice set of functions which can produce output within the
276seq_file system, but we have not yet turned them into a file that a user
277can see. Creating a file within the kernel requires, of course, the
278creation of a set of file_operations which implement the operations on that
279file. The seq_file interface provides a set of canned operations which do
280most of the work. The virtual file author still must implement the open()
281method, however, to hook everything up. The open function is often a single
282line, as in the example module::
283
284	static int ct_open(struct inode *inode, struct file *file)
285	{
286		return seq_open(file, &ct_seq_ops);
287	}
288
289Here, the call to seq_open() takes the seq_operations structure we created
290before, and gets set up to iterate through the virtual file.
291
292On a successful open, seq_open() stores the struct seq_file pointer in
293file->private_data. If you have an application where the same iterator can
294be used for more than one file, you can store an arbitrary pointer in the
295private field of the seq_file structure; that value can then be retrieved
296by the iterator functions.
297
298There is also a wrapper function to seq_open() called seq_open_private(). It
299kmallocs a zero filled block of memory and stores a pointer to it in the
300private field of the seq_file structure, returning 0 on success. The
301block size is specified in a third parameter to the function, e.g.::
302
303	static int ct_open(struct inode *inode, struct file *file)
304	{
305		return seq_open_private(file, &ct_seq_ops,
306					sizeof(struct mystruct));
307	}
308
309There is also a variant function, __seq_open_private(), which is functionally
310identical except that, if successful, it returns the pointer to the allocated
311memory block, allowing further initialisation e.g.::
312
313	static int ct_open(struct inode *inode, struct file *file)
314	{
315		struct mystruct *p =
316			__seq_open_private(file, &ct_seq_ops, sizeof(*p));
317
318		if (!p)
319			return -ENOMEM;
320
321		p->foo = bar; /* initialize my stuff */
322			...
323		p->baz = true;
324
325		return 0;
326	}
327
328A corresponding close function, seq_release_private() is available which
329frees the memory allocated in the corresponding open.
330
331The other operations of interest - read(), llseek(), and release() - are
332all implemented by the seq_file code itself. So a virtual file's
333file_operations structure will look like::
334
335	static const struct file_operations ct_file_ops = {
336	        .owner   = THIS_MODULE,
337	        .open    = ct_open,
338	        .read    = seq_read,
339	        .llseek  = seq_lseek,
340	        .release = seq_release
341	};
342
343There is also a seq_release_private() which passes the contents of the
344seq_file private field to kfree() before releasing the structure.
345
346The final step is the creation of the /proc file itself. In the example
347code, that is done in the initialization code in the usual way::
348
349	static int ct_init(void)
350	{
351	        struct proc_dir_entry *entry;
352
353	        proc_create("sequence", 0, NULL, &ct_file_ops);
354	        return 0;
355	}
356
357	module_init(ct_init);
358
359And that is pretty much it.
360
361
362seq_list
363========
364
365If your file will be iterating through a linked list, you may find these
366routines useful::
367
368	struct list_head *seq_list_start(struct list_head *head,
369	       		 		 loff_t pos);
370	struct list_head *seq_list_start_head(struct list_head *head,
371			 		      loff_t pos);
372	struct list_head *seq_list_next(void *v, struct list_head *head,
373					loff_t *ppos);
374
375These helpers will interpret pos as a position within the list and iterate
376accordingly.  Your start() and next() functions need only invoke the
377``seq_list_*`` helpers with a pointer to the appropriate list_head structure.
378
379
380The extra-simple version
381========================
382
383For extremely simple virtual files, there is an even easier interface.  A
384module can define only the show() function, which should create all the
385output that the virtual file will contain. The file's open() method then
386calls::
387
388	int single_open(struct file *file,
389	                int (*show)(struct seq_file *m, void *p),
390	                void *data);
391
392When output time comes, the show() function will be called once. The data
393value given to single_open() can be found in the private field of the
394seq_file structure. When using single_open(), the programmer should use
395single_release() instead of seq_release() in the file_operations structure
396to avoid a memory leak.
397