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