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