Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * stack_user.c
6 *
7 * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
8 *
9 * Copyright (C) 2007 Oracle. All rights reserved.
10 */
11
12#include <linux/module.h>
13#include <linux/fs.h>
14#include <linux/miscdevice.h>
15#include <linux/mutex.h>
16#include <linux/slab.h>
17#include <linux/reboot.h>
18#include <linux/sched.h>
19#include <linux/uaccess.h>
20
21#include "stackglue.h"
22
23#include <linux/dlm_plock.h>
24
25/*
26 * The control protocol starts with a handshake. Until the handshake
27 * is complete, the control device will fail all write(2)s.
28 *
29 * The handshake is simple. First, the client reads until EOF. Each line
30 * of output is a supported protocol tag. All protocol tags are a single
31 * character followed by a two hex digit version number. Currently the
32 * only things supported is T01, for "Text-base version 0x01". Next, the
33 * client writes the version they would like to use, including the newline.
34 * Thus, the protocol tag is 'T01\n'. If the version tag written is
35 * unknown, -EINVAL is returned. Once the negotiation is complete, the
36 * client can start sending messages.
37 *
38 * The T01 protocol has three messages. First is the "SETN" message.
39 * It has the following syntax:
40 *
41 * SETN<space><8-char-hex-nodenum><newline>
42 *
43 * This is 14 characters.
44 *
45 * The "SETN" message must be the first message following the protocol.
46 * It tells ocfs2_control the local node number.
47 *
48 * Next comes the "SETV" message. It has the following syntax:
49 *
50 * SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
51 *
52 * This is 11 characters.
53 *
54 * The "SETV" message sets the filesystem locking protocol version as
55 * negotiated by the client. The client negotiates based on the maximum
56 * version advertised in /sys/fs/ocfs2/max_locking_protocol. The major
57 * number from the "SETV" message must match
58 * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
59 * must be less than or equal to ...sp_max_version.pv_minor.
60 *
61 * Once this information has been set, mounts will be allowed. From this
62 * point on, the "DOWN" message can be sent for node down notification.
63 * It has the following syntax:
64 *
65 * DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
66 *
67 * eg:
68 *
69 * DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
70 *
71 * This is 47 characters.
72 */
73
74/*
75 * Whether or not the client has done the handshake.
76 * For now, we have just one protocol version.
77 */
78#define OCFS2_CONTROL_PROTO "T01\n"
79#define OCFS2_CONTROL_PROTO_LEN 4
80
81/* Handshake states */
82#define OCFS2_CONTROL_HANDSHAKE_INVALID (0)
83#define OCFS2_CONTROL_HANDSHAKE_READ (1)
84#define OCFS2_CONTROL_HANDSHAKE_PROTOCOL (2)
85#define OCFS2_CONTROL_HANDSHAKE_VALID (3)
86
87/* Messages */
88#define OCFS2_CONTROL_MESSAGE_OP_LEN 4
89#define OCFS2_CONTROL_MESSAGE_SETNODE_OP "SETN"
90#define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN 14
91#define OCFS2_CONTROL_MESSAGE_SETVERSION_OP "SETV"
92#define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN 11
93#define OCFS2_CONTROL_MESSAGE_DOWN_OP "DOWN"
94#define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN 47
95#define OCFS2_TEXT_UUID_LEN 32
96#define OCFS2_CONTROL_MESSAGE_VERNUM_LEN 2
97#define OCFS2_CONTROL_MESSAGE_NODENUM_LEN 8
98#define VERSION_LOCK "version_lock"
99
100enum ocfs2_connection_type {
101 WITH_CONTROLD,
102 NO_CONTROLD
103};
104
105/*
106 * ocfs2_live_connection is refcounted because the filesystem and
107 * miscdevice sides can detach in different order. Let's just be safe.
108 */
109struct ocfs2_live_connection {
110 struct list_head oc_list;
111 struct ocfs2_cluster_connection *oc_conn;
112 enum ocfs2_connection_type oc_type;
113 atomic_t oc_this_node;
114 int oc_our_slot;
115 struct dlm_lksb oc_version_lksb;
116 char oc_lvb[DLM_LVB_LEN];
117 struct completion oc_sync_wait;
118 wait_queue_head_t oc_wait;
119};
120
121struct ocfs2_control_private {
122 struct list_head op_list;
123 int op_state;
124 int op_this_node;
125 struct ocfs2_protocol_version op_proto;
126};
127
128/* SETN<space><8-char-hex-nodenum><newline> */
129struct ocfs2_control_message_setn {
130 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
131 char space;
132 char nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
133 char newline;
134};
135
136/* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
137struct ocfs2_control_message_setv {
138 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
139 char space1;
140 char major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
141 char space2;
142 char minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
143 char newline;
144};
145
146/* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
147struct ocfs2_control_message_down {
148 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
149 char space1;
150 char uuid[OCFS2_TEXT_UUID_LEN];
151 char space2;
152 char nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
153 char newline;
154};
155
156union ocfs2_control_message {
157 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
158 struct ocfs2_control_message_setn u_setn;
159 struct ocfs2_control_message_setv u_setv;
160 struct ocfs2_control_message_down u_down;
161};
162
163static struct ocfs2_stack_plugin ocfs2_user_plugin;
164
165static atomic_t ocfs2_control_opened;
166static int ocfs2_control_this_node = -1;
167static struct ocfs2_protocol_version running_proto;
168
169static LIST_HEAD(ocfs2_live_connection_list);
170static LIST_HEAD(ocfs2_control_private_list);
171static DEFINE_MUTEX(ocfs2_control_lock);
172
173static inline void ocfs2_control_set_handshake_state(struct file *file,
174 int state)
175{
176 struct ocfs2_control_private *p = file->private_data;
177 p->op_state = state;
178}
179
180static inline int ocfs2_control_get_handshake_state(struct file *file)
181{
182 struct ocfs2_control_private *p = file->private_data;
183 return p->op_state;
184}
185
186static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
187{
188 size_t len = strlen(name);
189 struct ocfs2_live_connection *c;
190
191 BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
192
193 list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
194 if ((c->oc_conn->cc_namelen == len) &&
195 !strncmp(c->oc_conn->cc_name, name, len))
196 return c;
197 }
198
199 return NULL;
200}
201
202/*
203 * ocfs2_live_connection structures are created underneath the ocfs2
204 * mount path. Since the VFS prevents multiple calls to
205 * fill_super(), we can't get dupes here.
206 */
207static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
208 struct ocfs2_live_connection *c)
209{
210 int rc = 0;
211
212 mutex_lock(&ocfs2_control_lock);
213 c->oc_conn = conn;
214
215 if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
216 list_add(&c->oc_list, &ocfs2_live_connection_list);
217 else {
218 printk(KERN_ERR
219 "ocfs2: Userspace control daemon is not present\n");
220 rc = -ESRCH;
221 }
222
223 mutex_unlock(&ocfs2_control_lock);
224 return rc;
225}
226
227/*
228 * This function disconnects the cluster connection from ocfs2_control.
229 * Afterwards, userspace can't affect the cluster connection.
230 */
231static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
232{
233 mutex_lock(&ocfs2_control_lock);
234 list_del_init(&c->oc_list);
235 c->oc_conn = NULL;
236 mutex_unlock(&ocfs2_control_lock);
237
238 kfree(c);
239}
240
241static int ocfs2_control_cfu(void *target, size_t target_len,
242 const char __user *buf, size_t count)
243{
244 /* The T01 expects write(2) calls to have exactly one command */
245 if ((count != target_len) ||
246 (count > sizeof(union ocfs2_control_message)))
247 return -EINVAL;
248
249 if (copy_from_user(target, buf, target_len))
250 return -EFAULT;
251
252 return 0;
253}
254
255static ssize_t ocfs2_control_validate_protocol(struct file *file,
256 const char __user *buf,
257 size_t count)
258{
259 ssize_t ret;
260 char kbuf[OCFS2_CONTROL_PROTO_LEN];
261
262 ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
263 buf, count);
264 if (ret)
265 return ret;
266
267 if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
268 return -EINVAL;
269
270 ocfs2_control_set_handshake_state(file,
271 OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
272
273 return count;
274}
275
276static void ocfs2_control_send_down(const char *uuid,
277 int nodenum)
278{
279 struct ocfs2_live_connection *c;
280
281 mutex_lock(&ocfs2_control_lock);
282
283 c = ocfs2_connection_find(uuid);
284 if (c) {
285 BUG_ON(c->oc_conn == NULL);
286 c->oc_conn->cc_recovery_handler(nodenum,
287 c->oc_conn->cc_recovery_data);
288 }
289
290 mutex_unlock(&ocfs2_control_lock);
291}
292
293/*
294 * Called whenever configuration elements are sent to /dev/ocfs2_control.
295 * If all configuration elements are present, try to set the global
296 * values. If there is a problem, return an error. Skip any missing
297 * elements, and only bump ocfs2_control_opened when we have all elements
298 * and are successful.
299 */
300static int ocfs2_control_install_private(struct file *file)
301{
302 int rc = 0;
303 int set_p = 1;
304 struct ocfs2_control_private *p = file->private_data;
305
306 BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
307
308 mutex_lock(&ocfs2_control_lock);
309
310 if (p->op_this_node < 0) {
311 set_p = 0;
312 } else if ((ocfs2_control_this_node >= 0) &&
313 (ocfs2_control_this_node != p->op_this_node)) {
314 rc = -EINVAL;
315 goto out_unlock;
316 }
317
318 if (!p->op_proto.pv_major) {
319 set_p = 0;
320 } else if (!list_empty(&ocfs2_live_connection_list) &&
321 ((running_proto.pv_major != p->op_proto.pv_major) ||
322 (running_proto.pv_minor != p->op_proto.pv_minor))) {
323 rc = -EINVAL;
324 goto out_unlock;
325 }
326
327 if (set_p) {
328 ocfs2_control_this_node = p->op_this_node;
329 running_proto.pv_major = p->op_proto.pv_major;
330 running_proto.pv_minor = p->op_proto.pv_minor;
331 }
332
333out_unlock:
334 mutex_unlock(&ocfs2_control_lock);
335
336 if (!rc && set_p) {
337 /* We set the global values successfully */
338 atomic_inc(&ocfs2_control_opened);
339 ocfs2_control_set_handshake_state(file,
340 OCFS2_CONTROL_HANDSHAKE_VALID);
341 }
342
343 return rc;
344}
345
346static int ocfs2_control_get_this_node(void)
347{
348 int rc;
349
350 mutex_lock(&ocfs2_control_lock);
351 if (ocfs2_control_this_node < 0)
352 rc = -EINVAL;
353 else
354 rc = ocfs2_control_this_node;
355 mutex_unlock(&ocfs2_control_lock);
356
357 return rc;
358}
359
360static int ocfs2_control_do_setnode_msg(struct file *file,
361 struct ocfs2_control_message_setn *msg)
362{
363 long nodenum;
364 char *ptr = NULL;
365 struct ocfs2_control_private *p = file->private_data;
366
367 if (ocfs2_control_get_handshake_state(file) !=
368 OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
369 return -EINVAL;
370
371 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
372 OCFS2_CONTROL_MESSAGE_OP_LEN))
373 return -EINVAL;
374
375 if ((msg->space != ' ') || (msg->newline != '\n'))
376 return -EINVAL;
377 msg->space = msg->newline = '\0';
378
379 nodenum = simple_strtol(msg->nodestr, &ptr, 16);
380 if (!ptr || *ptr)
381 return -EINVAL;
382
383 if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
384 (nodenum > INT_MAX) || (nodenum < 0))
385 return -ERANGE;
386 p->op_this_node = nodenum;
387
388 return ocfs2_control_install_private(file);
389}
390
391static int ocfs2_control_do_setversion_msg(struct file *file,
392 struct ocfs2_control_message_setv *msg)
393{
394 long major, minor;
395 char *ptr = NULL;
396 struct ocfs2_control_private *p = file->private_data;
397 struct ocfs2_protocol_version *max =
398 &ocfs2_user_plugin.sp_max_proto;
399
400 if (ocfs2_control_get_handshake_state(file) !=
401 OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
402 return -EINVAL;
403
404 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
405 OCFS2_CONTROL_MESSAGE_OP_LEN))
406 return -EINVAL;
407
408 if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
409 (msg->newline != '\n'))
410 return -EINVAL;
411 msg->space1 = msg->space2 = msg->newline = '\0';
412
413 major = simple_strtol(msg->major, &ptr, 16);
414 if (!ptr || *ptr)
415 return -EINVAL;
416 minor = simple_strtol(msg->minor, &ptr, 16);
417 if (!ptr || *ptr)
418 return -EINVAL;
419
420 /*
421 * The major must be between 1 and 255, inclusive. The minor
422 * must be between 0 and 255, inclusive. The version passed in
423 * must be within the maximum version supported by the filesystem.
424 */
425 if ((major == LONG_MIN) || (major == LONG_MAX) ||
426 (major > (u8)-1) || (major < 1))
427 return -ERANGE;
428 if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
429 (minor > (u8)-1) || (minor < 0))
430 return -ERANGE;
431 if ((major != max->pv_major) ||
432 (minor > max->pv_minor))
433 return -EINVAL;
434
435 p->op_proto.pv_major = major;
436 p->op_proto.pv_minor = minor;
437
438 return ocfs2_control_install_private(file);
439}
440
441static int ocfs2_control_do_down_msg(struct file *file,
442 struct ocfs2_control_message_down *msg)
443{
444 long nodenum;
445 char *p = NULL;
446
447 if (ocfs2_control_get_handshake_state(file) !=
448 OCFS2_CONTROL_HANDSHAKE_VALID)
449 return -EINVAL;
450
451 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
452 OCFS2_CONTROL_MESSAGE_OP_LEN))
453 return -EINVAL;
454
455 if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
456 (msg->newline != '\n'))
457 return -EINVAL;
458 msg->space1 = msg->space2 = msg->newline = '\0';
459
460 nodenum = simple_strtol(msg->nodestr, &p, 16);
461 if (!p || *p)
462 return -EINVAL;
463
464 if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
465 (nodenum > INT_MAX) || (nodenum < 0))
466 return -ERANGE;
467
468 ocfs2_control_send_down(msg->uuid, nodenum);
469
470 return 0;
471}
472
473static ssize_t ocfs2_control_message(struct file *file,
474 const char __user *buf,
475 size_t count)
476{
477 ssize_t ret;
478 union ocfs2_control_message msg;
479
480 /* Try to catch padding issues */
481 WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
482 (sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
483
484 memset(&msg, 0, sizeof(union ocfs2_control_message));
485 ret = ocfs2_control_cfu(&msg, count, buf, count);
486 if (ret)
487 goto out;
488
489 if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
490 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
491 OCFS2_CONTROL_MESSAGE_OP_LEN))
492 ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
493 else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
494 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
495 OCFS2_CONTROL_MESSAGE_OP_LEN))
496 ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
497 else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
498 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
499 OCFS2_CONTROL_MESSAGE_OP_LEN))
500 ret = ocfs2_control_do_down_msg(file, &msg.u_down);
501 else
502 ret = -EINVAL;
503
504out:
505 return ret ? ret : count;
506}
507
508static ssize_t ocfs2_control_write(struct file *file,
509 const char __user *buf,
510 size_t count,
511 loff_t *ppos)
512{
513 ssize_t ret;
514
515 switch (ocfs2_control_get_handshake_state(file)) {
516 case OCFS2_CONTROL_HANDSHAKE_INVALID:
517 ret = -EINVAL;
518 break;
519
520 case OCFS2_CONTROL_HANDSHAKE_READ:
521 ret = ocfs2_control_validate_protocol(file, buf,
522 count);
523 break;
524
525 case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
526 case OCFS2_CONTROL_HANDSHAKE_VALID:
527 ret = ocfs2_control_message(file, buf, count);
528 break;
529
530 default:
531 BUG();
532 ret = -EIO;
533 break;
534 }
535
536 return ret;
537}
538
539/*
540 * This is a naive version. If we ever have a new protocol, we'll expand
541 * it. Probably using seq_file.
542 */
543static ssize_t ocfs2_control_read(struct file *file,
544 char __user *buf,
545 size_t count,
546 loff_t *ppos)
547{
548 ssize_t ret;
549
550 ret = simple_read_from_buffer(buf, count, ppos,
551 OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
552
553 /* Have we read the whole protocol list? */
554 if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
555 ocfs2_control_set_handshake_state(file,
556 OCFS2_CONTROL_HANDSHAKE_READ);
557
558 return ret;
559}
560
561static int ocfs2_control_release(struct inode *inode, struct file *file)
562{
563 struct ocfs2_control_private *p = file->private_data;
564
565 mutex_lock(&ocfs2_control_lock);
566
567 if (ocfs2_control_get_handshake_state(file) !=
568 OCFS2_CONTROL_HANDSHAKE_VALID)
569 goto out;
570
571 if (atomic_dec_and_test(&ocfs2_control_opened)) {
572 if (!list_empty(&ocfs2_live_connection_list)) {
573 /* XXX: Do bad things! */
574 printk(KERN_ERR
575 "ocfs2: Unexpected release of ocfs2_control!\n"
576 " Loss of cluster connection requires "
577 "an emergency restart!\n");
578 emergency_restart();
579 }
580 /*
581 * Last valid close clears the node number and resets
582 * the locking protocol version
583 */
584 ocfs2_control_this_node = -1;
585 running_proto.pv_major = 0;
586 running_proto.pv_minor = 0;
587 }
588
589out:
590 list_del_init(&p->op_list);
591 file->private_data = NULL;
592
593 mutex_unlock(&ocfs2_control_lock);
594
595 kfree(p);
596
597 return 0;
598}
599
600static int ocfs2_control_open(struct inode *inode, struct file *file)
601{
602 struct ocfs2_control_private *p;
603
604 p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
605 if (!p)
606 return -ENOMEM;
607 p->op_this_node = -1;
608
609 mutex_lock(&ocfs2_control_lock);
610 file->private_data = p;
611 list_add(&p->op_list, &ocfs2_control_private_list);
612 mutex_unlock(&ocfs2_control_lock);
613
614 return 0;
615}
616
617static const struct file_operations ocfs2_control_fops = {
618 .open = ocfs2_control_open,
619 .release = ocfs2_control_release,
620 .read = ocfs2_control_read,
621 .write = ocfs2_control_write,
622 .owner = THIS_MODULE,
623 .llseek = default_llseek,
624};
625
626static struct miscdevice ocfs2_control_device = {
627 .minor = MISC_DYNAMIC_MINOR,
628 .name = "ocfs2_control",
629 .fops = &ocfs2_control_fops,
630};
631
632static int ocfs2_control_init(void)
633{
634 int rc;
635
636 atomic_set(&ocfs2_control_opened, 0);
637
638 rc = misc_register(&ocfs2_control_device);
639 if (rc)
640 printk(KERN_ERR
641 "ocfs2: Unable to register ocfs2_control device "
642 "(errno %d)\n",
643 -rc);
644
645 return rc;
646}
647
648static void ocfs2_control_exit(void)
649{
650 misc_deregister(&ocfs2_control_device);
651}
652
653static void fsdlm_lock_ast_wrapper(void *astarg)
654{
655 struct ocfs2_dlm_lksb *lksb = astarg;
656 int status = lksb->lksb_fsdlm.sb_status;
657
658 /*
659 * For now we're punting on the issue of other non-standard errors
660 * where we can't tell if the unlock_ast or lock_ast should be called.
661 * The main "other error" that's possible is EINVAL which means the
662 * function was called with invalid args, which shouldn't be possible
663 * since the caller here is under our control. Other non-standard
664 * errors probably fall into the same category, or otherwise are fatal
665 * which means we can't carry on anyway.
666 */
667
668 if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
669 lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
670 else
671 lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
672}
673
674static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
675{
676 struct ocfs2_dlm_lksb *lksb = astarg;
677
678 lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
679}
680
681static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
682 int mode,
683 struct ocfs2_dlm_lksb *lksb,
684 u32 flags,
685 void *name,
686 unsigned int namelen)
687{
688 int ret;
689
690 if (!lksb->lksb_fsdlm.sb_lvbptr)
691 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
692 sizeof(struct dlm_lksb);
693
694 ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
695 flags|DLM_LKF_NODLCKWT, name, namelen, 0,
696 fsdlm_lock_ast_wrapper, lksb,
697 fsdlm_blocking_ast_wrapper);
698 return ret;
699}
700
701static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
702 struct ocfs2_dlm_lksb *lksb,
703 u32 flags)
704{
705 int ret;
706
707 ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
708 flags, &lksb->lksb_fsdlm, lksb);
709 return ret;
710}
711
712static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
713{
714 return lksb->lksb_fsdlm.sb_status;
715}
716
717static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
718{
719 int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
720
721 return !invalid;
722}
723
724static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
725{
726 if (!lksb->lksb_fsdlm.sb_lvbptr)
727 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
728 sizeof(struct dlm_lksb);
729 return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
730}
731
732static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
733{
734}
735
736static int user_plock(struct ocfs2_cluster_connection *conn,
737 u64 ino,
738 struct file *file,
739 int cmd,
740 struct file_lock *fl)
741{
742 /*
743 * This more or less just demuxes the plock request into any
744 * one of three dlm calls.
745 *
746 * Internally, fs/dlm will pass these to a misc device, which
747 * a userspace daemon will read and write to.
748 *
749 * For now, cancel requests (which happen internally only),
750 * are turned into unlocks. Most of this function taken from
751 * gfs2_lock.
752 */
753
754 if (cmd == F_CANCELLK) {
755 cmd = F_SETLK;
756 fl->fl_type = F_UNLCK;
757 }
758
759 if (IS_GETLK(cmd))
760 return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
761 else if (fl->fl_type == F_UNLCK)
762 return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
763 else
764 return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
765}
766
767/*
768 * Compare a requested locking protocol version against the current one.
769 *
770 * If the major numbers are different, they are incompatible.
771 * If the current minor is greater than the request, they are incompatible.
772 * If the current minor is less than or equal to the request, they are
773 * compatible, and the requester should run at the current minor version.
774 */
775static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
776 struct ocfs2_protocol_version *request)
777{
778 if (existing->pv_major != request->pv_major)
779 return 1;
780
781 if (existing->pv_minor > request->pv_minor)
782 return 1;
783
784 if (existing->pv_minor < request->pv_minor)
785 request->pv_minor = existing->pv_minor;
786
787 return 0;
788}
789
790static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
791{
792 struct ocfs2_protocol_version *pv =
793 (struct ocfs2_protocol_version *)lvb;
794 /*
795 * ocfs2_protocol_version has two u8 variables, so we don't
796 * need any endian conversion.
797 */
798 ver->pv_major = pv->pv_major;
799 ver->pv_minor = pv->pv_minor;
800}
801
802static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
803{
804 struct ocfs2_protocol_version *pv =
805 (struct ocfs2_protocol_version *)lvb;
806 /*
807 * ocfs2_protocol_version has two u8 variables, so we don't
808 * need any endian conversion.
809 */
810 pv->pv_major = ver->pv_major;
811 pv->pv_minor = ver->pv_minor;
812}
813
814static void sync_wait_cb(void *arg)
815{
816 struct ocfs2_cluster_connection *conn = arg;
817 struct ocfs2_live_connection *lc = conn->cc_private;
818 complete(&lc->oc_sync_wait);
819}
820
821static int sync_unlock(struct ocfs2_cluster_connection *conn,
822 struct dlm_lksb *lksb, char *name)
823{
824 int error;
825 struct ocfs2_live_connection *lc = conn->cc_private;
826
827 error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
828 if (error) {
829 printk(KERN_ERR "%s lkid %x error %d\n",
830 name, lksb->sb_lkid, error);
831 return error;
832 }
833
834 wait_for_completion(&lc->oc_sync_wait);
835
836 if (lksb->sb_status != -DLM_EUNLOCK) {
837 printk(KERN_ERR "%s lkid %x status %d\n",
838 name, lksb->sb_lkid, lksb->sb_status);
839 return -1;
840 }
841 return 0;
842}
843
844static int sync_lock(struct ocfs2_cluster_connection *conn,
845 int mode, uint32_t flags,
846 struct dlm_lksb *lksb, char *name)
847{
848 int error, status;
849 struct ocfs2_live_connection *lc = conn->cc_private;
850
851 error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
852 name, strlen(name),
853 0, sync_wait_cb, conn, NULL);
854 if (error) {
855 printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
856 name, lksb->sb_lkid, flags, mode, error);
857 return error;
858 }
859
860 wait_for_completion(&lc->oc_sync_wait);
861
862 status = lksb->sb_status;
863
864 if (status && status != -EAGAIN) {
865 printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
866 name, lksb->sb_lkid, flags, mode, status);
867 }
868
869 return status;
870}
871
872
873static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
874 int flags)
875{
876 struct ocfs2_live_connection *lc = conn->cc_private;
877 return sync_lock(conn, mode, flags,
878 &lc->oc_version_lksb, VERSION_LOCK);
879}
880
881static int version_unlock(struct ocfs2_cluster_connection *conn)
882{
883 struct ocfs2_live_connection *lc = conn->cc_private;
884 return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
885}
886
887/* get_protocol_version()
888 *
889 * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
890 * The algorithm is:
891 * 1. Attempt to take the lock in EX mode (non-blocking).
892 * 2. If successful (which means it is the first mount), write the
893 * version number and downconvert to PR lock.
894 * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
895 * taking the PR lock.
896 */
897
898static int get_protocol_version(struct ocfs2_cluster_connection *conn)
899{
900 int ret;
901 struct ocfs2_live_connection *lc = conn->cc_private;
902 struct ocfs2_protocol_version pv;
903
904 running_proto.pv_major =
905 ocfs2_user_plugin.sp_max_proto.pv_major;
906 running_proto.pv_minor =
907 ocfs2_user_plugin.sp_max_proto.pv_minor;
908
909 lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
910 ret = version_lock(conn, DLM_LOCK_EX,
911 DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
912 if (!ret) {
913 conn->cc_version.pv_major = running_proto.pv_major;
914 conn->cc_version.pv_minor = running_proto.pv_minor;
915 version_to_lvb(&running_proto, lc->oc_lvb);
916 version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
917 } else if (ret == -EAGAIN) {
918 ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
919 if (ret)
920 goto out;
921 lvb_to_version(lc->oc_lvb, &pv);
922
923 if ((pv.pv_major != running_proto.pv_major) ||
924 (pv.pv_minor > running_proto.pv_minor)) {
925 ret = -EINVAL;
926 goto out;
927 }
928
929 conn->cc_version.pv_major = pv.pv_major;
930 conn->cc_version.pv_minor = pv.pv_minor;
931 }
932out:
933 return ret;
934}
935
936static void user_recover_prep(void *arg)
937{
938}
939
940static void user_recover_slot(void *arg, struct dlm_slot *slot)
941{
942 struct ocfs2_cluster_connection *conn = arg;
943 printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
944 slot->nodeid, slot->slot);
945 conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
946
947}
948
949static void user_recover_done(void *arg, struct dlm_slot *slots,
950 int num_slots, int our_slot,
951 uint32_t generation)
952{
953 struct ocfs2_cluster_connection *conn = arg;
954 struct ocfs2_live_connection *lc = conn->cc_private;
955 int i;
956
957 for (i = 0; i < num_slots; i++)
958 if (slots[i].slot == our_slot) {
959 atomic_set(&lc->oc_this_node, slots[i].nodeid);
960 break;
961 }
962
963 lc->oc_our_slot = our_slot;
964 wake_up(&lc->oc_wait);
965}
966
967static const struct dlm_lockspace_ops ocfs2_ls_ops = {
968 .recover_prep = user_recover_prep,
969 .recover_slot = user_recover_slot,
970 .recover_done = user_recover_done,
971};
972
973static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
974{
975 version_unlock(conn);
976 dlm_release_lockspace(conn->cc_lockspace, 2);
977 conn->cc_lockspace = NULL;
978 ocfs2_live_connection_drop(conn->cc_private);
979 conn->cc_private = NULL;
980 return 0;
981}
982
983static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
984{
985 dlm_lockspace_t *fsdlm;
986 struct ocfs2_live_connection *lc;
987 int rc, ops_rv;
988
989 BUG_ON(conn == NULL);
990
991 lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
992 if (!lc)
993 return -ENOMEM;
994
995 init_waitqueue_head(&lc->oc_wait);
996 init_completion(&lc->oc_sync_wait);
997 atomic_set(&lc->oc_this_node, 0);
998 conn->cc_private = lc;
999 lc->oc_type = NO_CONTROLD;
1000
1001 rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
1002 DLM_LSFL_FS | DLM_LSFL_NEWEXCL, DLM_LVB_LEN,
1003 &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
1004 if (rc) {
1005 if (rc == -EEXIST || rc == -EPROTO)
1006 printk(KERN_ERR "ocfs2: Unable to create the "
1007 "lockspace %s (%d), because a ocfs2-tools "
1008 "program is running on this file system "
1009 "with the same name lockspace\n",
1010 conn->cc_name, rc);
1011 goto out;
1012 }
1013
1014 if (ops_rv == -EOPNOTSUPP) {
1015 lc->oc_type = WITH_CONTROLD;
1016 printk(KERN_NOTICE "ocfs2: You seem to be using an older "
1017 "version of dlm_controld and/or ocfs2-tools."
1018 " Please consider upgrading.\n");
1019 } else if (ops_rv) {
1020 rc = ops_rv;
1021 goto out;
1022 }
1023 conn->cc_lockspace = fsdlm;
1024
1025 rc = ocfs2_live_connection_attach(conn, lc);
1026 if (rc)
1027 goto out;
1028
1029 if (lc->oc_type == NO_CONTROLD) {
1030 rc = get_protocol_version(conn);
1031 if (rc) {
1032 printk(KERN_ERR "ocfs2: Could not determine"
1033 " locking version\n");
1034 user_cluster_disconnect(conn);
1035 goto out;
1036 }
1037 wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
1038 }
1039
1040 /*
1041 * running_proto must have been set before we allowed any mounts
1042 * to proceed.
1043 */
1044 if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
1045 printk(KERN_ERR
1046 "Unable to mount with fs locking protocol version "
1047 "%u.%u because negotiated protocol is %u.%u\n",
1048 conn->cc_version.pv_major, conn->cc_version.pv_minor,
1049 running_proto.pv_major, running_proto.pv_minor);
1050 rc = -EPROTO;
1051 ocfs2_live_connection_drop(lc);
1052 lc = NULL;
1053 }
1054
1055out:
1056 if (rc)
1057 kfree(lc);
1058 return rc;
1059}
1060
1061
1062static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
1063 unsigned int *this_node)
1064{
1065 int rc;
1066 struct ocfs2_live_connection *lc = conn->cc_private;
1067
1068 if (lc->oc_type == WITH_CONTROLD)
1069 rc = ocfs2_control_get_this_node();
1070 else if (lc->oc_type == NO_CONTROLD)
1071 rc = atomic_read(&lc->oc_this_node);
1072 else
1073 rc = -EINVAL;
1074
1075 if (rc < 0)
1076 return rc;
1077
1078 *this_node = rc;
1079 return 0;
1080}
1081
1082static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
1083 .connect = user_cluster_connect,
1084 .disconnect = user_cluster_disconnect,
1085 .this_node = user_cluster_this_node,
1086 .dlm_lock = user_dlm_lock,
1087 .dlm_unlock = user_dlm_unlock,
1088 .lock_status = user_dlm_lock_status,
1089 .lvb_valid = user_dlm_lvb_valid,
1090 .lock_lvb = user_dlm_lvb,
1091 .plock = user_plock,
1092 .dump_lksb = user_dlm_dump_lksb,
1093};
1094
1095static struct ocfs2_stack_plugin ocfs2_user_plugin = {
1096 .sp_name = "user",
1097 .sp_ops = &ocfs2_user_plugin_ops,
1098 .sp_owner = THIS_MODULE,
1099};
1100
1101
1102static int __init ocfs2_user_plugin_init(void)
1103{
1104 int rc;
1105
1106 rc = ocfs2_control_init();
1107 if (!rc) {
1108 rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
1109 if (rc)
1110 ocfs2_control_exit();
1111 }
1112
1113 return rc;
1114}
1115
1116static void __exit ocfs2_user_plugin_exit(void)
1117{
1118 ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
1119 ocfs2_control_exit();
1120}
1121
1122MODULE_AUTHOR("Oracle");
1123MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
1124MODULE_LICENSE("GPL");
1125module_init(ocfs2_user_plugin_init);
1126module_exit(ocfs2_user_plugin_exit);
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * stack_user.c
5 *
6 * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
7 *
8 * Copyright (C) 2007 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation, version 2.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 */
19
20#include <linux/module.h>
21#include <linux/fs.h>
22#include <linux/miscdevice.h>
23#include <linux/mutex.h>
24#include <linux/slab.h>
25#include <linux/reboot.h>
26#include <asm/uaccess.h>
27
28#include "stackglue.h"
29
30#include <linux/dlm_plock.h>
31
32/*
33 * The control protocol starts with a handshake. Until the handshake
34 * is complete, the control device will fail all write(2)s.
35 *
36 * The handshake is simple. First, the client reads until EOF. Each line
37 * of output is a supported protocol tag. All protocol tags are a single
38 * character followed by a two hex digit version number. Currently the
39 * only things supported is T01, for "Text-base version 0x01". Next, the
40 * client writes the version they would like to use, including the newline.
41 * Thus, the protocol tag is 'T01\n'. If the version tag written is
42 * unknown, -EINVAL is returned. Once the negotiation is complete, the
43 * client can start sending messages.
44 *
45 * The T01 protocol has three messages. First is the "SETN" message.
46 * It has the following syntax:
47 *
48 * SETN<space><8-char-hex-nodenum><newline>
49 *
50 * This is 14 characters.
51 *
52 * The "SETN" message must be the first message following the protocol.
53 * It tells ocfs2_control the local node number.
54 *
55 * Next comes the "SETV" message. It has the following syntax:
56 *
57 * SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
58 *
59 * This is 11 characters.
60 *
61 * The "SETV" message sets the filesystem locking protocol version as
62 * negotiated by the client. The client negotiates based on the maximum
63 * version advertised in /sys/fs/ocfs2/max_locking_protocol. The major
64 * number from the "SETV" message must match
65 * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
66 * must be less than or equal to ...sp_max_version.pv_minor.
67 *
68 * Once this information has been set, mounts will be allowed. From this
69 * point on, the "DOWN" message can be sent for node down notification.
70 * It has the following syntax:
71 *
72 * DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
73 *
74 * eg:
75 *
76 * DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
77 *
78 * This is 47 characters.
79 */
80
81/*
82 * Whether or not the client has done the handshake.
83 * For now, we have just one protocol version.
84 */
85#define OCFS2_CONTROL_PROTO "T01\n"
86#define OCFS2_CONTROL_PROTO_LEN 4
87
88/* Handshake states */
89#define OCFS2_CONTROL_HANDSHAKE_INVALID (0)
90#define OCFS2_CONTROL_HANDSHAKE_READ (1)
91#define OCFS2_CONTROL_HANDSHAKE_PROTOCOL (2)
92#define OCFS2_CONTROL_HANDSHAKE_VALID (3)
93
94/* Messages */
95#define OCFS2_CONTROL_MESSAGE_OP_LEN 4
96#define OCFS2_CONTROL_MESSAGE_SETNODE_OP "SETN"
97#define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN 14
98#define OCFS2_CONTROL_MESSAGE_SETVERSION_OP "SETV"
99#define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN 11
100#define OCFS2_CONTROL_MESSAGE_DOWN_OP "DOWN"
101#define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN 47
102#define OCFS2_TEXT_UUID_LEN 32
103#define OCFS2_CONTROL_MESSAGE_VERNUM_LEN 2
104#define OCFS2_CONTROL_MESSAGE_NODENUM_LEN 8
105
106/*
107 * ocfs2_live_connection is refcounted because the filesystem and
108 * miscdevice sides can detach in different order. Let's just be safe.
109 */
110struct ocfs2_live_connection {
111 struct list_head oc_list;
112 struct ocfs2_cluster_connection *oc_conn;
113};
114
115struct ocfs2_control_private {
116 struct list_head op_list;
117 int op_state;
118 int op_this_node;
119 struct ocfs2_protocol_version op_proto;
120};
121
122/* SETN<space><8-char-hex-nodenum><newline> */
123struct ocfs2_control_message_setn {
124 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
125 char space;
126 char nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
127 char newline;
128};
129
130/* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
131struct ocfs2_control_message_setv {
132 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
133 char space1;
134 char major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
135 char space2;
136 char minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
137 char newline;
138};
139
140/* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
141struct ocfs2_control_message_down {
142 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
143 char space1;
144 char uuid[OCFS2_TEXT_UUID_LEN];
145 char space2;
146 char nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
147 char newline;
148};
149
150union ocfs2_control_message {
151 char tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
152 struct ocfs2_control_message_setn u_setn;
153 struct ocfs2_control_message_setv u_setv;
154 struct ocfs2_control_message_down u_down;
155};
156
157static struct ocfs2_stack_plugin ocfs2_user_plugin;
158
159static atomic_t ocfs2_control_opened;
160static int ocfs2_control_this_node = -1;
161static struct ocfs2_protocol_version running_proto;
162
163static LIST_HEAD(ocfs2_live_connection_list);
164static LIST_HEAD(ocfs2_control_private_list);
165static DEFINE_MUTEX(ocfs2_control_lock);
166
167static inline void ocfs2_control_set_handshake_state(struct file *file,
168 int state)
169{
170 struct ocfs2_control_private *p = file->private_data;
171 p->op_state = state;
172}
173
174static inline int ocfs2_control_get_handshake_state(struct file *file)
175{
176 struct ocfs2_control_private *p = file->private_data;
177 return p->op_state;
178}
179
180static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
181{
182 size_t len = strlen(name);
183 struct ocfs2_live_connection *c;
184
185 BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
186
187 list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
188 if ((c->oc_conn->cc_namelen == len) &&
189 !strncmp(c->oc_conn->cc_name, name, len))
190 return c;
191 }
192
193 return NULL;
194}
195
196/*
197 * ocfs2_live_connection structures are created underneath the ocfs2
198 * mount path. Since the VFS prevents multiple calls to
199 * fill_super(), we can't get dupes here.
200 */
201static int ocfs2_live_connection_new(struct ocfs2_cluster_connection *conn,
202 struct ocfs2_live_connection **c_ret)
203{
204 int rc = 0;
205 struct ocfs2_live_connection *c;
206
207 c = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
208 if (!c)
209 return -ENOMEM;
210
211 mutex_lock(&ocfs2_control_lock);
212 c->oc_conn = conn;
213
214 if (atomic_read(&ocfs2_control_opened))
215 list_add(&c->oc_list, &ocfs2_live_connection_list);
216 else {
217 printk(KERN_ERR
218 "ocfs2: Userspace control daemon is not present\n");
219 rc = -ESRCH;
220 }
221
222 mutex_unlock(&ocfs2_control_lock);
223
224 if (!rc)
225 *c_ret = c;
226 else
227 kfree(c);
228
229 return rc;
230}
231
232/*
233 * This function disconnects the cluster connection from ocfs2_control.
234 * Afterwards, userspace can't affect the cluster connection.
235 */
236static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
237{
238 mutex_lock(&ocfs2_control_lock);
239 list_del_init(&c->oc_list);
240 c->oc_conn = NULL;
241 mutex_unlock(&ocfs2_control_lock);
242
243 kfree(c);
244}
245
246static int ocfs2_control_cfu(void *target, size_t target_len,
247 const char __user *buf, size_t count)
248{
249 /* The T01 expects write(2) calls to have exactly one command */
250 if ((count != target_len) ||
251 (count > sizeof(union ocfs2_control_message)))
252 return -EINVAL;
253
254 if (copy_from_user(target, buf, target_len))
255 return -EFAULT;
256
257 return 0;
258}
259
260static ssize_t ocfs2_control_validate_protocol(struct file *file,
261 const char __user *buf,
262 size_t count)
263{
264 ssize_t ret;
265 char kbuf[OCFS2_CONTROL_PROTO_LEN];
266
267 ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
268 buf, count);
269 if (ret)
270 return ret;
271
272 if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
273 return -EINVAL;
274
275 ocfs2_control_set_handshake_state(file,
276 OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
277
278 return count;
279}
280
281static void ocfs2_control_send_down(const char *uuid,
282 int nodenum)
283{
284 struct ocfs2_live_connection *c;
285
286 mutex_lock(&ocfs2_control_lock);
287
288 c = ocfs2_connection_find(uuid);
289 if (c) {
290 BUG_ON(c->oc_conn == NULL);
291 c->oc_conn->cc_recovery_handler(nodenum,
292 c->oc_conn->cc_recovery_data);
293 }
294
295 mutex_unlock(&ocfs2_control_lock);
296}
297
298/*
299 * Called whenever configuration elements are sent to /dev/ocfs2_control.
300 * If all configuration elements are present, try to set the global
301 * values. If there is a problem, return an error. Skip any missing
302 * elements, and only bump ocfs2_control_opened when we have all elements
303 * and are successful.
304 */
305static int ocfs2_control_install_private(struct file *file)
306{
307 int rc = 0;
308 int set_p = 1;
309 struct ocfs2_control_private *p = file->private_data;
310
311 BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
312
313 mutex_lock(&ocfs2_control_lock);
314
315 if (p->op_this_node < 0) {
316 set_p = 0;
317 } else if ((ocfs2_control_this_node >= 0) &&
318 (ocfs2_control_this_node != p->op_this_node)) {
319 rc = -EINVAL;
320 goto out_unlock;
321 }
322
323 if (!p->op_proto.pv_major) {
324 set_p = 0;
325 } else if (!list_empty(&ocfs2_live_connection_list) &&
326 ((running_proto.pv_major != p->op_proto.pv_major) ||
327 (running_proto.pv_minor != p->op_proto.pv_minor))) {
328 rc = -EINVAL;
329 goto out_unlock;
330 }
331
332 if (set_p) {
333 ocfs2_control_this_node = p->op_this_node;
334 running_proto.pv_major = p->op_proto.pv_major;
335 running_proto.pv_minor = p->op_proto.pv_minor;
336 }
337
338out_unlock:
339 mutex_unlock(&ocfs2_control_lock);
340
341 if (!rc && set_p) {
342 /* We set the global values successfully */
343 atomic_inc(&ocfs2_control_opened);
344 ocfs2_control_set_handshake_state(file,
345 OCFS2_CONTROL_HANDSHAKE_VALID);
346 }
347
348 return rc;
349}
350
351static int ocfs2_control_get_this_node(void)
352{
353 int rc;
354
355 mutex_lock(&ocfs2_control_lock);
356 if (ocfs2_control_this_node < 0)
357 rc = -EINVAL;
358 else
359 rc = ocfs2_control_this_node;
360 mutex_unlock(&ocfs2_control_lock);
361
362 return rc;
363}
364
365static int ocfs2_control_do_setnode_msg(struct file *file,
366 struct ocfs2_control_message_setn *msg)
367{
368 long nodenum;
369 char *ptr = NULL;
370 struct ocfs2_control_private *p = file->private_data;
371
372 if (ocfs2_control_get_handshake_state(file) !=
373 OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
374 return -EINVAL;
375
376 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
377 OCFS2_CONTROL_MESSAGE_OP_LEN))
378 return -EINVAL;
379
380 if ((msg->space != ' ') || (msg->newline != '\n'))
381 return -EINVAL;
382 msg->space = msg->newline = '\0';
383
384 nodenum = simple_strtol(msg->nodestr, &ptr, 16);
385 if (!ptr || *ptr)
386 return -EINVAL;
387
388 if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
389 (nodenum > INT_MAX) || (nodenum < 0))
390 return -ERANGE;
391 p->op_this_node = nodenum;
392
393 return ocfs2_control_install_private(file);
394}
395
396static int ocfs2_control_do_setversion_msg(struct file *file,
397 struct ocfs2_control_message_setv *msg)
398 {
399 long major, minor;
400 char *ptr = NULL;
401 struct ocfs2_control_private *p = file->private_data;
402 struct ocfs2_protocol_version *max =
403 &ocfs2_user_plugin.sp_max_proto;
404
405 if (ocfs2_control_get_handshake_state(file) !=
406 OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
407 return -EINVAL;
408
409 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
410 OCFS2_CONTROL_MESSAGE_OP_LEN))
411 return -EINVAL;
412
413 if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
414 (msg->newline != '\n'))
415 return -EINVAL;
416 msg->space1 = msg->space2 = msg->newline = '\0';
417
418 major = simple_strtol(msg->major, &ptr, 16);
419 if (!ptr || *ptr)
420 return -EINVAL;
421 minor = simple_strtol(msg->minor, &ptr, 16);
422 if (!ptr || *ptr)
423 return -EINVAL;
424
425 /*
426 * The major must be between 1 and 255, inclusive. The minor
427 * must be between 0 and 255, inclusive. The version passed in
428 * must be within the maximum version supported by the filesystem.
429 */
430 if ((major == LONG_MIN) || (major == LONG_MAX) ||
431 (major > (u8)-1) || (major < 1))
432 return -ERANGE;
433 if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
434 (minor > (u8)-1) || (minor < 0))
435 return -ERANGE;
436 if ((major != max->pv_major) ||
437 (minor > max->pv_minor))
438 return -EINVAL;
439
440 p->op_proto.pv_major = major;
441 p->op_proto.pv_minor = minor;
442
443 return ocfs2_control_install_private(file);
444}
445
446static int ocfs2_control_do_down_msg(struct file *file,
447 struct ocfs2_control_message_down *msg)
448{
449 long nodenum;
450 char *p = NULL;
451
452 if (ocfs2_control_get_handshake_state(file) !=
453 OCFS2_CONTROL_HANDSHAKE_VALID)
454 return -EINVAL;
455
456 if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
457 OCFS2_CONTROL_MESSAGE_OP_LEN))
458 return -EINVAL;
459
460 if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
461 (msg->newline != '\n'))
462 return -EINVAL;
463 msg->space1 = msg->space2 = msg->newline = '\0';
464
465 nodenum = simple_strtol(msg->nodestr, &p, 16);
466 if (!p || *p)
467 return -EINVAL;
468
469 if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
470 (nodenum > INT_MAX) || (nodenum < 0))
471 return -ERANGE;
472
473 ocfs2_control_send_down(msg->uuid, nodenum);
474
475 return 0;
476}
477
478static ssize_t ocfs2_control_message(struct file *file,
479 const char __user *buf,
480 size_t count)
481{
482 ssize_t ret;
483 union ocfs2_control_message msg;
484
485 /* Try to catch padding issues */
486 WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
487 (sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
488
489 memset(&msg, 0, sizeof(union ocfs2_control_message));
490 ret = ocfs2_control_cfu(&msg, count, buf, count);
491 if (ret)
492 goto out;
493
494 if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
495 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
496 OCFS2_CONTROL_MESSAGE_OP_LEN))
497 ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
498 else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
499 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
500 OCFS2_CONTROL_MESSAGE_OP_LEN))
501 ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
502 else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
503 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
504 OCFS2_CONTROL_MESSAGE_OP_LEN))
505 ret = ocfs2_control_do_down_msg(file, &msg.u_down);
506 else
507 ret = -EINVAL;
508
509out:
510 return ret ? ret : count;
511}
512
513static ssize_t ocfs2_control_write(struct file *file,
514 const char __user *buf,
515 size_t count,
516 loff_t *ppos)
517{
518 ssize_t ret;
519
520 switch (ocfs2_control_get_handshake_state(file)) {
521 case OCFS2_CONTROL_HANDSHAKE_INVALID:
522 ret = -EINVAL;
523 break;
524
525 case OCFS2_CONTROL_HANDSHAKE_READ:
526 ret = ocfs2_control_validate_protocol(file, buf,
527 count);
528 break;
529
530 case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
531 case OCFS2_CONTROL_HANDSHAKE_VALID:
532 ret = ocfs2_control_message(file, buf, count);
533 break;
534
535 default:
536 BUG();
537 ret = -EIO;
538 break;
539 }
540
541 return ret;
542}
543
544/*
545 * This is a naive version. If we ever have a new protocol, we'll expand
546 * it. Probably using seq_file.
547 */
548static ssize_t ocfs2_control_read(struct file *file,
549 char __user *buf,
550 size_t count,
551 loff_t *ppos)
552{
553 ssize_t ret;
554
555 ret = simple_read_from_buffer(buf, count, ppos,
556 OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
557
558 /* Have we read the whole protocol list? */
559 if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
560 ocfs2_control_set_handshake_state(file,
561 OCFS2_CONTROL_HANDSHAKE_READ);
562
563 return ret;
564}
565
566static int ocfs2_control_release(struct inode *inode, struct file *file)
567{
568 struct ocfs2_control_private *p = file->private_data;
569
570 mutex_lock(&ocfs2_control_lock);
571
572 if (ocfs2_control_get_handshake_state(file) !=
573 OCFS2_CONTROL_HANDSHAKE_VALID)
574 goto out;
575
576 if (atomic_dec_and_test(&ocfs2_control_opened)) {
577 if (!list_empty(&ocfs2_live_connection_list)) {
578 /* XXX: Do bad things! */
579 printk(KERN_ERR
580 "ocfs2: Unexpected release of ocfs2_control!\n"
581 " Loss of cluster connection requires "
582 "an emergency restart!\n");
583 emergency_restart();
584 }
585 /*
586 * Last valid close clears the node number and resets
587 * the locking protocol version
588 */
589 ocfs2_control_this_node = -1;
590 running_proto.pv_major = 0;
591 running_proto.pv_major = 0;
592 }
593
594out:
595 list_del_init(&p->op_list);
596 file->private_data = NULL;
597
598 mutex_unlock(&ocfs2_control_lock);
599
600 kfree(p);
601
602 return 0;
603}
604
605static int ocfs2_control_open(struct inode *inode, struct file *file)
606{
607 struct ocfs2_control_private *p;
608
609 p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
610 if (!p)
611 return -ENOMEM;
612 p->op_this_node = -1;
613
614 mutex_lock(&ocfs2_control_lock);
615 file->private_data = p;
616 list_add(&p->op_list, &ocfs2_control_private_list);
617 mutex_unlock(&ocfs2_control_lock);
618
619 return 0;
620}
621
622static const struct file_operations ocfs2_control_fops = {
623 .open = ocfs2_control_open,
624 .release = ocfs2_control_release,
625 .read = ocfs2_control_read,
626 .write = ocfs2_control_write,
627 .owner = THIS_MODULE,
628 .llseek = default_llseek,
629};
630
631static struct miscdevice ocfs2_control_device = {
632 .minor = MISC_DYNAMIC_MINOR,
633 .name = "ocfs2_control",
634 .fops = &ocfs2_control_fops,
635};
636
637static int ocfs2_control_init(void)
638{
639 int rc;
640
641 atomic_set(&ocfs2_control_opened, 0);
642
643 rc = misc_register(&ocfs2_control_device);
644 if (rc)
645 printk(KERN_ERR
646 "ocfs2: Unable to register ocfs2_control device "
647 "(errno %d)\n",
648 -rc);
649
650 return rc;
651}
652
653static void ocfs2_control_exit(void)
654{
655 int rc;
656
657 rc = misc_deregister(&ocfs2_control_device);
658 if (rc)
659 printk(KERN_ERR
660 "ocfs2: Unable to deregister ocfs2_control device "
661 "(errno %d)\n",
662 -rc);
663}
664
665static void fsdlm_lock_ast_wrapper(void *astarg)
666{
667 struct ocfs2_dlm_lksb *lksb = astarg;
668 int status = lksb->lksb_fsdlm.sb_status;
669
670 /*
671 * For now we're punting on the issue of other non-standard errors
672 * where we can't tell if the unlock_ast or lock_ast should be called.
673 * The main "other error" that's possible is EINVAL which means the
674 * function was called with invalid args, which shouldn't be possible
675 * since the caller here is under our control. Other non-standard
676 * errors probably fall into the same category, or otherwise are fatal
677 * which means we can't carry on anyway.
678 */
679
680 if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
681 lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
682 else
683 lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
684}
685
686static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
687{
688 struct ocfs2_dlm_lksb *lksb = astarg;
689
690 lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
691}
692
693static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
694 int mode,
695 struct ocfs2_dlm_lksb *lksb,
696 u32 flags,
697 void *name,
698 unsigned int namelen)
699{
700 int ret;
701
702 if (!lksb->lksb_fsdlm.sb_lvbptr)
703 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
704 sizeof(struct dlm_lksb);
705
706 ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
707 flags|DLM_LKF_NODLCKWT, name, namelen, 0,
708 fsdlm_lock_ast_wrapper, lksb,
709 fsdlm_blocking_ast_wrapper);
710 return ret;
711}
712
713static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
714 struct ocfs2_dlm_lksb *lksb,
715 u32 flags)
716{
717 int ret;
718
719 ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
720 flags, &lksb->lksb_fsdlm, lksb);
721 return ret;
722}
723
724static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
725{
726 return lksb->lksb_fsdlm.sb_status;
727}
728
729static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
730{
731 int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
732
733 return !invalid;
734}
735
736static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
737{
738 if (!lksb->lksb_fsdlm.sb_lvbptr)
739 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
740 sizeof(struct dlm_lksb);
741 return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
742}
743
744static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
745{
746}
747
748static int user_plock(struct ocfs2_cluster_connection *conn,
749 u64 ino,
750 struct file *file,
751 int cmd,
752 struct file_lock *fl)
753{
754 /*
755 * This more or less just demuxes the plock request into any
756 * one of three dlm calls.
757 *
758 * Internally, fs/dlm will pass these to a misc device, which
759 * a userspace daemon will read and write to.
760 *
761 * For now, cancel requests (which happen internally only),
762 * are turned into unlocks. Most of this function taken from
763 * gfs2_lock.
764 */
765
766 if (cmd == F_CANCELLK) {
767 cmd = F_SETLK;
768 fl->fl_type = F_UNLCK;
769 }
770
771 if (IS_GETLK(cmd))
772 return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
773 else if (fl->fl_type == F_UNLCK)
774 return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
775 else
776 return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
777}
778
779/*
780 * Compare a requested locking protocol version against the current one.
781 *
782 * If the major numbers are different, they are incompatible.
783 * If the current minor is greater than the request, they are incompatible.
784 * If the current minor is less than or equal to the request, they are
785 * compatible, and the requester should run at the current minor version.
786 */
787static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
788 struct ocfs2_protocol_version *request)
789{
790 if (existing->pv_major != request->pv_major)
791 return 1;
792
793 if (existing->pv_minor > request->pv_minor)
794 return 1;
795
796 if (existing->pv_minor < request->pv_minor)
797 request->pv_minor = existing->pv_minor;
798
799 return 0;
800}
801
802static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
803{
804 dlm_lockspace_t *fsdlm;
805 struct ocfs2_live_connection *uninitialized_var(control);
806 int rc = 0;
807
808 BUG_ON(conn == NULL);
809
810 rc = ocfs2_live_connection_new(conn, &control);
811 if (rc)
812 goto out;
813
814 /*
815 * running_proto must have been set before we allowed any mounts
816 * to proceed.
817 */
818 if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
819 printk(KERN_ERR
820 "Unable to mount with fs locking protocol version "
821 "%u.%u because the userspace control daemon has "
822 "negotiated %u.%u\n",
823 conn->cc_version.pv_major, conn->cc_version.pv_minor,
824 running_proto.pv_major, running_proto.pv_minor);
825 rc = -EPROTO;
826 ocfs2_live_connection_drop(control);
827 goto out;
828 }
829
830 rc = dlm_new_lockspace(conn->cc_name, strlen(conn->cc_name),
831 &fsdlm, DLM_LSFL_FS, DLM_LVB_LEN);
832 if (rc) {
833 ocfs2_live_connection_drop(control);
834 goto out;
835 }
836
837 conn->cc_private = control;
838 conn->cc_lockspace = fsdlm;
839out:
840 return rc;
841}
842
843static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
844{
845 dlm_release_lockspace(conn->cc_lockspace, 2);
846 conn->cc_lockspace = NULL;
847 ocfs2_live_connection_drop(conn->cc_private);
848 conn->cc_private = NULL;
849 return 0;
850}
851
852static int user_cluster_this_node(unsigned int *this_node)
853{
854 int rc;
855
856 rc = ocfs2_control_get_this_node();
857 if (rc < 0)
858 return rc;
859
860 *this_node = rc;
861 return 0;
862}
863
864static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
865 .connect = user_cluster_connect,
866 .disconnect = user_cluster_disconnect,
867 .this_node = user_cluster_this_node,
868 .dlm_lock = user_dlm_lock,
869 .dlm_unlock = user_dlm_unlock,
870 .lock_status = user_dlm_lock_status,
871 .lvb_valid = user_dlm_lvb_valid,
872 .lock_lvb = user_dlm_lvb,
873 .plock = user_plock,
874 .dump_lksb = user_dlm_dump_lksb,
875};
876
877static struct ocfs2_stack_plugin ocfs2_user_plugin = {
878 .sp_name = "user",
879 .sp_ops = &ocfs2_user_plugin_ops,
880 .sp_owner = THIS_MODULE,
881};
882
883
884static int __init ocfs2_user_plugin_init(void)
885{
886 int rc;
887
888 rc = ocfs2_control_init();
889 if (!rc) {
890 rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
891 if (rc)
892 ocfs2_control_exit();
893 }
894
895 return rc;
896}
897
898static void __exit ocfs2_user_plugin_exit(void)
899{
900 ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
901 ocfs2_control_exit();
902}
903
904MODULE_AUTHOR("Oracle");
905MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
906MODULE_LICENSE("GPL");
907module_init(ocfs2_user_plugin_init);
908module_exit(ocfs2_user_plugin_exit);