| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved. |
| 4 |
*/ |
| 5 |
|
| 6 |
#include <linux/miscdevice.h> |
| 7 |
#include <linux/init.h> |
| 8 |
#include <linux/wait.h> |
| 9 |
#include <linux/file.h> |
| 10 |
#include <linux/fs.h> |
| 11 |
#include <linux/poll.h> |
| 12 |
#include <linux/signal.h> |
| 13 |
#include <linux/spinlock.h> |
| 14 |
#include <linux/dlm.h> |
| 15 |
#include <linux/dlm_device.h> |
| 16 |
#include <linux/slab.h> |
| 17 |
#include <linux/sched/signal.h> |
| 18 |
|
| 19 |
#include <trace/events/dlm.h> |
| 20 |
|
| 21 |
#include "dlm_internal.h" |
| 22 |
#include "lockspace.h" |
| 23 |
#include "lock.h" |
| 24 |
#include "lvb_table.h" |
| 25 |
#include "user.h" |
| 26 |
#include "ast.h" |
| 27 |
#include "config.h" |
| 28 |
#include "memory.h" |
| 29 |
|
| 30 |
static const char name_prefix[] = "dlm"; |
| 31 |
static const struct file_operations device_fops; |
| 32 |
static atomic_t dlm_monitor_opened; |
| 33 |
static int dlm_monitor_unused = 1; |
| 34 |
|
| 35 |
#ifdef CONFIG_COMPAT |
| 36 |
|
| 37 |
struct dlm_lock_params32 { |
| 38 |
__u8 mode; |
| 39 |
__u8 namelen; |
| 40 |
__u16 unused; |
| 41 |
__u32 flags; |
| 42 |
__u32 lkid; |
| 43 |
__u32 parent; |
| 44 |
__u64 xid; |
| 45 |
__u64 timeout; |
| 46 |
__u32 castparam; |
| 47 |
__u32 castaddr; |
| 48 |
__u32 bastparam; |
| 49 |
__u32 bastaddr; |
| 50 |
__u32 lksb; |
| 51 |
char lvb[DLM_USER_LVB_LEN]; |
| 52 |
char name[]; |
| 53 |
}; |
| 54 |
|
| 55 |
struct dlm_write_request32 { |
| 56 |
__u32 version[3]; |
| 57 |
__u8 cmd; |
| 58 |
__u8 is64bit; |
| 59 |
__u8 unused[2]; |
| 60 |
|
| 61 |
union { |
| 62 |
struct dlm_lock_params32 lock; |
| 63 |
struct dlm_lspace_params lspace; |
| 64 |
struct dlm_purge_params purge; |
| 65 |
} i; |
| 66 |
}; |
| 67 |
|
| 68 |
struct dlm_lksb32 { |
| 69 |
__u32 sb_status; |
| 70 |
__u32 sb_lkid; |
| 71 |
__u8 sb_flags; |
| 72 |
__u32 sb_lvbptr; |
| 73 |
}; |
| 74 |
|
| 75 |
struct dlm_lock_result32 { |
| 76 |
__u32 version[3]; |
| 77 |
__u32 length; |
| 78 |
__u32 user_astaddr; |
| 79 |
__u32 user_astparam; |
| 80 |
__u32 user_lksb; |
| 81 |
struct dlm_lksb32 lksb; |
| 82 |
__u8 bast_mode; |
| 83 |
__u8 unused[3]; |
| 84 |
/* Offsets may be zero if no data is present */ |
| 85 |
__u32 lvb_offset; |
| 86 |
}; |
| 87 |
|
| 88 |
static void compat_input(struct dlm_write_request *kb, |
| 89 |
struct dlm_write_request32 *kb32, |
| 90 |
int namelen) |
| 91 |
{ |
| 92 |
kb->version[0] = kb32->version[0]; |
| 93 |
kb->version[1] = kb32->version[1]; |
| 94 |
kb->version[2] = kb32->version[2]; |
| 95 |
|
| 96 |
kb->cmd = kb32->cmd; |
| 97 |
kb->is64bit = kb32->is64bit; |
| 98 |
if (kb->cmd == DLM_USER_CREATE_LOCKSPACE || |
| 99 |
kb->cmd == DLM_USER_REMOVE_LOCKSPACE) { |
| 100 |
kb->i.lspace.flags = kb32->i.lspace.flags; |
| 101 |
kb->i.lspace.minor = kb32->i.lspace.minor; |
| 102 |
memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen); |
| 103 |
} else if (kb->cmd == DLM_USER_PURGE) { |
| 104 |
kb->i.purge.nodeid = kb32->i.purge.nodeid; |
| 105 |
kb->i.purge.pid = kb32->i.purge.pid; |
| 106 |
} else { |
| 107 |
kb->i.lock.mode = kb32->i.lock.mode; |
| 108 |
kb->i.lock.namelen = kb32->i.lock.namelen; |
| 109 |
kb->i.lock.flags = kb32->i.lock.flags; |
| 110 |
kb->i.lock.lkid = kb32->i.lock.lkid; |
| 111 |
kb->i.lock.parent = kb32->i.lock.parent; |
| 112 |
kb->i.lock.xid = kb32->i.lock.xid; |
| 113 |
kb->i.lock.timeout = kb32->i.lock.timeout; |
| 114 |
kb->i.lock.castparam = (__user void *)(long)kb32->i.lock.castparam; |
| 115 |
kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr; |
| 116 |
kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam; |
| 117 |
kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr; |
| 118 |
kb->i.lock.lksb = (__user void *)(long)kb32->i.lock.lksb; |
| 119 |
memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN); |
| 120 |
memcpy(kb->i.lock.name, kb32->i.lock.name, namelen); |
| 121 |
} |
| 122 |
} |
| 123 |
|
| 124 |
static void compat_output(struct dlm_lock_result *res, |
| 125 |
struct dlm_lock_result32 *res32) |
| 126 |
{ |
| 127 |
memset(res32, 0, sizeof(*res32)); |
| 128 |
|
| 129 |
res32->version[0] = res->version[0]; |
| 130 |
res32->version[1] = res->version[1]; |
| 131 |
res32->version[2] = res->version[2]; |
| 132 |
|
| 133 |
res32->user_astaddr = (__u32)(__force long)res->user_astaddr; |
| 134 |
res32->user_astparam = (__u32)(__force long)res->user_astparam; |
| 135 |
res32->user_lksb = (__u32)(__force long)res->user_lksb; |
| 136 |
res32->bast_mode = res->bast_mode; |
| 137 |
|
| 138 |
res32->lvb_offset = res->lvb_offset; |
| 139 |
res32->length = res->length; |
| 140 |
|
| 141 |
res32->lksb.sb_status = res->lksb.sb_status; |
| 142 |
res32->lksb.sb_flags = res->lksb.sb_flags; |
| 143 |
res32->lksb.sb_lkid = res->lksb.sb_lkid; |
| 144 |
res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; |
| 145 |
} |
| 146 |
#endif |
| 147 |
|
| 148 |
/* Figure out if this lock is at the end of its life and no longer |
| 149 |
available for the application to use. The lkb still exists until |
| 150 |
the final ast is read. A lock becomes EOL in three situations: |
| 151 |
1. a noqueue request fails with EAGAIN |
| 152 |
2. an unlock completes with EUNLOCK |
| 153 |
3. a cancel of a waiting request completes with ECANCEL/EDEADLK |
| 154 |
An EOL lock needs to be removed from the process's list of locks. |
| 155 |
And we can't allow any new operation on an EOL lock. This is |
| 156 |
not related to the lifetime of the lkb struct which is managed |
| 157 |
entirely by refcount. */ |
| 158 |
|
| 159 |
static int lkb_is_endoflife(int mode, int status) |
| 160 |
{ |
| 161 |
switch (status) { |
| 162 |
case -DLM_EUNLOCK: |
| 163 |
return 1; |
| 164 |
case -DLM_ECANCEL: |
| 165 |
case -ETIMEDOUT: |
| 166 |
case -EDEADLK: |
| 167 |
case -EAGAIN: |
| 168 |
if (mode == DLM_LOCK_IV) |
| 169 |
return 1; |
| 170 |
break; |
| 171 |
} |
| 172 |
return 0; |
| 173 |
} |
| 174 |
|
| 175 |
/* we could possibly check if the cancel of an orphan has resulted in the lkb |
| 176 |
being removed and then remove that lkb from the orphans list and free it */ |
| 177 |
|
| 178 |
void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode, |
| 179 |
int status, uint32_t sbflags) |
| 180 |
{ |
| 181 |
struct dlm_ls *ls; |
| 182 |
struct dlm_user_args *ua; |
| 183 |
struct dlm_user_proc *proc; |
| 184 |
struct dlm_callback *cb; |
| 185 |
int rv, copy_lvb; |
| 186 |
|
| 187 |
if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) || |
| 188 |
test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags)) |
| 189 |
return; |
| 190 |
|
| 191 |
ls = lkb->lkb_resource->res_ls; |
| 192 |
spin_lock_bh(&ls->ls_clear_proc_locks); |
| 193 |
|
| 194 |
/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast |
| 195 |
can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed |
| 196 |
lkb->ua so we can't try to use it. This second check is necessary |
| 197 |
for cases where a completion ast is received for an operation that |
| 198 |
began before clear_proc_locks did its cancel/unlock. */ |
| 199 |
|
| 200 |
if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) || |
| 201 |
test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags)) |
| 202 |
goto out; |
| 203 |
|
| 204 |
DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb);); |
| 205 |
ua = lkb->lkb_ua; |
| 206 |
proc = ua->proc; |
| 207 |
|
| 208 |
if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL) |
| 209 |
goto out; |
| 210 |
|
| 211 |
if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status)) |
| 212 |
set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags); |
| 213 |
|
| 214 |
spin_lock_bh(&proc->asts_spin); |
| 215 |
|
| 216 |
if (!dlm_may_skip_callback(lkb, flags, mode, status, sbflags, |
| 217 |
©_lvb)) { |
| 218 |
rv = dlm_get_cb(lkb, flags, mode, status, sbflags, &cb); |
| 219 |
if (!rv) { |
| 220 |
cb->copy_lvb = copy_lvb; |
| 221 |
cb->ua = *ua; |
| 222 |
cb->lkb_lksb = &cb->ua.lksb; |
| 223 |
if (copy_lvb) { |
| 224 |
memcpy(cb->lvbptr, ua->lksb.sb_lvbptr, |
| 225 |
DLM_USER_LVB_LEN); |
| 226 |
cb->lkb_lksb->sb_lvbptr = cb->lvbptr; |
| 227 |
} |
| 228 |
|
| 229 |
list_add_tail(&cb->list, &proc->asts); |
| 230 |
wake_up_interruptible(&proc->wait); |
| 231 |
} |
| 232 |
} |
| 233 |
spin_unlock_bh(&proc->asts_spin); |
| 234 |
|
| 235 |
if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) { |
| 236 |
/* N.B. spin_lock locks_spin, not asts_spin */ |
| 237 |
spin_lock_bh(&proc->locks_spin); |
| 238 |
if (!list_empty(&lkb->lkb_ownqueue)) { |
| 239 |
list_del_init(&lkb->lkb_ownqueue); |
| 240 |
dlm_put_lkb(lkb); |
| 241 |
} |
| 242 |
spin_unlock_bh(&proc->locks_spin); |
| 243 |
} |
| 244 |
out: |
| 245 |
spin_unlock_bh(&ls->ls_clear_proc_locks); |
| 246 |
} |
| 247 |
|
| 248 |
static int device_user_lock(struct dlm_user_proc *proc, |
| 249 |
struct dlm_lock_params *params) |
| 250 |
{ |
| 251 |
struct dlm_ls *ls; |
| 252 |
struct dlm_user_args *ua; |
| 253 |
uint32_t lkid; |
| 254 |
int error = -ENOMEM; |
| 255 |
|
| 256 |
ls = dlm_find_lockspace_local(proc->lockspace); |
| 257 |
if (!ls) |
| 258 |
return -ENOENT; |
| 259 |
|
| 260 |
if (!params->castaddr || !params->lksb) { |
| 261 |
error = -EINVAL; |
| 262 |
goto out; |
| 263 |
} |
| 264 |
|
| 265 |
ua = kzalloc_obj(struct dlm_user_args, GFP_NOFS); |
| 266 |
if (!ua) |
| 267 |
goto out; |
| 268 |
ua->proc = proc; |
| 269 |
ua->user_lksb = params->lksb; |
| 270 |
ua->castparam = params->castparam; |
| 271 |
ua->castaddr = params->castaddr; |
| 272 |
ua->bastparam = params->bastparam; |
| 273 |
ua->bastaddr = params->bastaddr; |
| 274 |
ua->xid = params->xid; |
| 275 |
|
| 276 |
if (params->flags & DLM_LKF_CONVERT) { |
| 277 |
error = dlm_user_convert(ls, ua, |
| 278 |
params->mode, params->flags, |
| 279 |
params->lkid, params->lvb); |
| 280 |
} else if (params->flags & DLM_LKF_ORPHAN) { |
| 281 |
error = dlm_user_adopt_orphan(ls, ua, |
| 282 |
params->mode, params->flags, |
| 283 |
params->name, params->namelen, |
| 284 |
&lkid); |
| 285 |
if (!error) |
| 286 |
error = lkid; |
| 287 |
} else { |
| 288 |
error = dlm_user_request(ls, ua, |
| 289 |
params->mode, params->flags, |
| 290 |
params->name, params->namelen); |
| 291 |
if (!error) |
| 292 |
error = ua->lksb.sb_lkid; |
| 293 |
} |
| 294 |
out: |
| 295 |
dlm_put_lockspace(ls); |
| 296 |
return error; |
| 297 |
} |
| 298 |
|
| 299 |
static int device_user_unlock(struct dlm_user_proc *proc, |
| 300 |
struct dlm_lock_params *params) |
| 301 |
{ |
| 302 |
struct dlm_ls *ls; |
| 303 |
struct dlm_user_args *ua; |
| 304 |
int error = -ENOMEM; |
| 305 |
|
| 306 |
ls = dlm_find_lockspace_local(proc->lockspace); |
| 307 |
if (!ls) |
| 308 |
return -ENOENT; |
| 309 |
|
| 310 |
ua = kzalloc_obj(struct dlm_user_args, GFP_NOFS); |
| 311 |
if (!ua) |
| 312 |
goto out; |
| 313 |
ua->proc = proc; |
| 314 |
ua->user_lksb = params->lksb; |
| 315 |
ua->castparam = params->castparam; |
| 316 |
ua->castaddr = params->castaddr; |
| 317 |
|
| 318 |
if (params->flags & DLM_LKF_CANCEL) |
| 319 |
error = dlm_user_cancel(ls, ua, params->flags, params->lkid); |
| 320 |
else |
| 321 |
error = dlm_user_unlock(ls, ua, params->flags, params->lkid, |
| 322 |
params->lvb); |
| 323 |
out: |
| 324 |
dlm_put_lockspace(ls); |
| 325 |
return error; |
| 326 |
} |
| 327 |
|
| 328 |
static int device_user_deadlock(struct dlm_user_proc *proc, |
| 329 |
struct dlm_lock_params *params) |
| 330 |
{ |
| 331 |
struct dlm_ls *ls; |
| 332 |
int error; |
| 333 |
|
| 334 |
ls = dlm_find_lockspace_local(proc->lockspace); |
| 335 |
if (!ls) |
| 336 |
return -ENOENT; |
| 337 |
|
| 338 |
error = dlm_user_deadlock(ls, params->flags, params->lkid); |
| 339 |
|
| 340 |
dlm_put_lockspace(ls); |
| 341 |
return error; |
| 342 |
} |
| 343 |
|
| 344 |
static int dlm_device_register(struct dlm_ls *ls, char *name) |
| 345 |
{ |
| 346 |
int error, len; |
| 347 |
|
| 348 |
/* The device is already registered. This happens when the |
| 349 |
lockspace is created multiple times from userspace. */ |
| 350 |
if (ls->ls_device.name) |
| 351 |
return 0; |
| 352 |
|
| 353 |
error = -ENOMEM; |
| 354 |
len = strlen(name) + strlen(name_prefix) + 2; |
| 355 |
ls->ls_device.name = kzalloc(len, GFP_NOFS); |
| 356 |
if (!ls->ls_device.name) |
| 357 |
goto fail; |
| 358 |
|
| 359 |
snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, |
| 360 |
name); |
| 361 |
ls->ls_device.fops = &device_fops; |
| 362 |
ls->ls_device.minor = MISC_DYNAMIC_MINOR; |
| 363 |
|
| 364 |
error = misc_register(&ls->ls_device); |
| 365 |
if (error) { |
| 366 |
kfree(ls->ls_device.name); |
| 367 |
/* this has to be set to NULL |
| 368 |
* to avoid a double-free in dlm_device_deregister |
| 369 |
*/ |
| 370 |
ls->ls_device.name = NULL; |
| 371 |
} |
| 372 |
fail: |
| 373 |
return error; |
| 374 |
} |
| 375 |
|
| 376 |
int dlm_device_deregister(struct dlm_ls *ls) |
| 377 |
{ |
| 378 |
/* The device is not registered. This happens when the lockspace |
| 379 |
was never used from userspace, or when device_create_lockspace() |
| 380 |
calls dlm_release_lockspace() after the register fails. */ |
| 381 |
if (!ls->ls_device.name) |
| 382 |
return 0; |
| 383 |
|
| 384 |
misc_deregister(&ls->ls_device); |
| 385 |
kfree(ls->ls_device.name); |
| 386 |
return 0; |
| 387 |
} |
| 388 |
|
| 389 |
static int device_user_purge(struct dlm_user_proc *proc, |
| 390 |
struct dlm_purge_params *params) |
| 391 |
{ |
| 392 |
struct dlm_ls *ls; |
| 393 |
int error; |
| 394 |
|
| 395 |
ls = dlm_find_lockspace_local(proc->lockspace); |
| 396 |
if (!ls) |
| 397 |
return -ENOENT; |
| 398 |
|
| 399 |
error = dlm_user_purge(ls, proc, params->nodeid, params->pid); |
| 400 |
|
| 401 |
dlm_put_lockspace(ls); |
| 402 |
return error; |
| 403 |
} |
| 404 |
|
| 405 |
static int device_create_lockspace(struct dlm_lspace_params *params) |
| 406 |
{ |
| 407 |
dlm_lockspace_t *lockspace; |
| 408 |
struct dlm_ls *ls; |
| 409 |
int error; |
| 410 |
|
| 411 |
if (!capable(CAP_SYS_ADMIN)) |
| 412 |
return -EPERM; |
| 413 |
|
| 414 |
error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name, |
| 415 |
params->flags, DLM_USER_LVB_LEN, NULL, |
| 416 |
NULL, NULL, &lockspace); |
| 417 |
if (error) |
| 418 |
return error; |
| 419 |
|
| 420 |
ls = dlm_find_lockspace_local(lockspace); |
| 421 |
if (!ls) |
| 422 |
return -ENOENT; |
| 423 |
|
| 424 |
error = dlm_device_register(ls, params->name); |
| 425 |
dlm_put_lockspace(ls); |
| 426 |
|
| 427 |
if (error) |
| 428 |
dlm_release_lockspace(lockspace, DLM_RELEASE_NO_LOCKS); |
| 429 |
else |
| 430 |
error = ls->ls_device.minor; |
| 431 |
|
| 432 |
return error; |
| 433 |
} |
| 434 |
|
| 435 |
static int device_remove_lockspace(struct dlm_lspace_params *params) |
| 436 |
{ |
| 437 |
dlm_lockspace_t *lockspace; |
| 438 |
struct dlm_ls *ls; |
| 439 |
int error, force = DLM_RELEASE_NO_LOCKS; |
| 440 |
|
| 441 |
if (!capable(CAP_SYS_ADMIN)) |
| 442 |
return -EPERM; |
| 443 |
|
| 444 |
ls = dlm_find_lockspace_device(params->minor); |
| 445 |
if (!ls) |
| 446 |
return -ENOENT; |
| 447 |
|
| 448 |
if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
| 449 |
force = DLM_RELEASE_NORMAL; |
| 450 |
|
| 451 |
lockspace = ls; |
| 452 |
dlm_put_lockspace(ls); |
| 453 |
|
| 454 |
/* The final dlm_release_lockspace waits for references to go to |
| 455 |
zero, so all processes will need to close their device for the |
| 456 |
ls before the release will proceed. release also calls the |
| 457 |
device_deregister above. Converting a positive return value |
| 458 |
from release to zero means that userspace won't know when its |
| 459 |
release was the final one, but it shouldn't need to know. */ |
| 460 |
|
| 461 |
error = dlm_release_lockspace(lockspace, force); |
| 462 |
if (error > 0) |
| 463 |
error = 0; |
| 464 |
return error; |
| 465 |
} |
| 466 |
|
| 467 |
/* Check the user's version matches ours */ |
| 468 |
static int check_version(struct dlm_write_request *req) |
| 469 |
{ |
| 470 |
if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || |
| 471 |
(req->version[0] == DLM_DEVICE_VERSION_MAJOR && |
| 472 |
req->version[1] > DLM_DEVICE_VERSION_MINOR)) { |
| 473 |
|
| 474 |
printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " |
| 475 |
"user (%d.%d.%d) kernel (%d.%d.%d)\n", |
| 476 |
current->comm, |
| 477 |
task_pid_nr(current), |
| 478 |
req->version[0], |
| 479 |
req->version[1], |
| 480 |
req->version[2], |
| 481 |
DLM_DEVICE_VERSION_MAJOR, |
| 482 |
DLM_DEVICE_VERSION_MINOR, |
| 483 |
DLM_DEVICE_VERSION_PATCH); |
| 484 |
return -EINVAL; |
| 485 |
} |
| 486 |
return 0; |
| 487 |
} |
| 488 |
|
| 489 |
/* |
| 490 |
* device_write |
| 491 |
* |
| 492 |
* device_user_lock |
| 493 |
* dlm_user_request -> request_lock |
| 494 |
* dlm_user_convert -> convert_lock |
| 495 |
* |
| 496 |
* device_user_unlock |
| 497 |
* dlm_user_unlock -> unlock_lock |
| 498 |
* dlm_user_cancel -> cancel_lock |
| 499 |
* |
| 500 |
* device_create_lockspace |
| 501 |
* dlm_new_lockspace |
| 502 |
* |
| 503 |
* device_remove_lockspace |
| 504 |
* dlm_release_lockspace |
| 505 |
*/ |
| 506 |
|
| 507 |
/* a write to a lockspace device is a lock or unlock request, a write |
| 508 |
to the control device is to create/remove a lockspace */ |
| 509 |
|
| 510 |
static ssize_t device_write(struct file *file, const char __user *buf, |
| 511 |
size_t count, loff_t *ppos) |
| 512 |
{ |
| 513 |
struct dlm_user_proc *proc = file->private_data; |
| 514 |
struct dlm_write_request *kbuf; |
| 515 |
int error; |
| 516 |
|
| 517 |
#ifdef CONFIG_COMPAT |
| 518 |
if (count < sizeof(struct dlm_write_request32)) |
| 519 |
#else |
| 520 |
if (count < sizeof(struct dlm_write_request)) |
| 521 |
#endif |
| 522 |
return -EINVAL; |
| 523 |
|
| 524 |
/* |
| 525 |
* can't compare against COMPAT/dlm_write_request32 because |
| 526 |
* we don't yet know if is64bit is zero |
| 527 |
*/ |
| 528 |
if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN) |
| 529 |
return -EINVAL; |
| 530 |
|
| 531 |
kbuf = memdup_user_nul(buf, count); |
| 532 |
if (IS_ERR(kbuf)) |
| 533 |
return PTR_ERR(kbuf); |
| 534 |
|
| 535 |
if (check_version(kbuf)) { |
| 536 |
error = -EBADE; |
| 537 |
goto out_free; |
| 538 |
} |
| 539 |
|
| 540 |
#ifdef CONFIG_COMPAT |
| 541 |
if (!kbuf->is64bit) { |
| 542 |
struct dlm_write_request32 *k32buf; |
| 543 |
int namelen = 0; |
| 544 |
|
| 545 |
if (count > sizeof(struct dlm_write_request32)) |
| 546 |
namelen = count - sizeof(struct dlm_write_request32); |
| 547 |
|
| 548 |
k32buf = (struct dlm_write_request32 *)kbuf; |
| 549 |
|
| 550 |
/* add 1 after namelen so that the name string is terminated */ |
| 551 |
kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1, |
| 552 |
GFP_NOFS); |
| 553 |
if (!kbuf) { |
| 554 |
kfree(k32buf); |
| 555 |
return -ENOMEM; |
| 556 |
} |
| 557 |
|
| 558 |
if (proc) |
| 559 |
set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); |
| 560 |
|
| 561 |
compat_input(kbuf, k32buf, namelen); |
| 562 |
kfree(k32buf); |
| 563 |
} |
| 564 |
#endif |
| 565 |
|
| 566 |
/* do we really need this? can a write happen after a close? */ |
| 567 |
if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && |
| 568 |
(proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) { |
| 569 |
error = -EINVAL; |
| 570 |
goto out_free; |
| 571 |
} |
| 572 |
|
| 573 |
error = -EINVAL; |
| 574 |
|
| 575 |
switch (kbuf->cmd) |
| 576 |
{ |
| 577 |
case DLM_USER_LOCK: |
| 578 |
if (!proc) { |
| 579 |
log_print("no locking on control device"); |
| 580 |
goto out_free; |
| 581 |
} |
| 582 |
error = device_user_lock(proc, &kbuf->i.lock); |
| 583 |
break; |
| 584 |
|
| 585 |
case DLM_USER_UNLOCK: |
| 586 |
if (!proc) { |
| 587 |
log_print("no locking on control device"); |
| 588 |
goto out_free; |
| 589 |
} |
| 590 |
error = device_user_unlock(proc, &kbuf->i.lock); |
| 591 |
break; |
| 592 |
|
| 593 |
case DLM_USER_DEADLOCK: |
| 594 |
if (!proc) { |
| 595 |
log_print("no locking on control device"); |
| 596 |
goto out_free; |
| 597 |
} |
| 598 |
error = device_user_deadlock(proc, &kbuf->i.lock); |
| 599 |
break; |
| 600 |
|
| 601 |
case DLM_USER_CREATE_LOCKSPACE: |
| 602 |
if (proc) { |
| 603 |
log_print("create/remove only on control device"); |
| 604 |
goto out_free; |
| 605 |
} |
| 606 |
error = device_create_lockspace(&kbuf->i.lspace); |
| 607 |
break; |
| 608 |
|
| 609 |
case DLM_USER_REMOVE_LOCKSPACE: |
| 610 |
if (proc) { |
| 611 |
log_print("create/remove only on control device"); |
| 612 |
goto out_free; |
| 613 |
} |
| 614 |
error = device_remove_lockspace(&kbuf->i.lspace); |
| 615 |
break; |
| 616 |
|
| 617 |
case DLM_USER_PURGE: |
| 618 |
if (!proc) { |
| 619 |
log_print("no locking on control device"); |
| 620 |
goto out_free; |
| 621 |
} |
| 622 |
error = device_user_purge(proc, &kbuf->i.purge); |
| 623 |
break; |
| 624 |
|
| 625 |
default: |
| 626 |
log_print("Unknown command passed to DLM device : %d\n", |
| 627 |
kbuf->cmd); |
| 628 |
} |
| 629 |
|
| 630 |
out_free: |
| 631 |
kfree(kbuf); |
| 632 |
return error; |
| 633 |
} |
| 634 |
|
| 635 |
/* Every process that opens the lockspace device has its own "proc" structure |
| 636 |
hanging off the open file that's used to keep track of locks owned by the |
| 637 |
process and asts that need to be delivered to the process. */ |
| 638 |
|
| 639 |
static int device_open(struct inode *inode, struct file *file) |
| 640 |
{ |
| 641 |
struct dlm_user_proc *proc; |
| 642 |
struct dlm_ls *ls; |
| 643 |
|
| 644 |
ls = dlm_find_lockspace_device(iminor(inode)); |
| 645 |
if (!ls) |
| 646 |
return -ENOENT; |
| 647 |
|
| 648 |
proc = kzalloc_obj(struct dlm_user_proc, GFP_NOFS); |
| 649 |
if (!proc) { |
| 650 |
dlm_put_lockspace(ls); |
| 651 |
return -ENOMEM; |
| 652 |
} |
| 653 |
|
| 654 |
proc->lockspace = ls; |
| 655 |
INIT_LIST_HEAD(&proc->asts); |
| 656 |
INIT_LIST_HEAD(&proc->locks); |
| 657 |
INIT_LIST_HEAD(&proc->unlocking); |
| 658 |
spin_lock_init(&proc->asts_spin); |
| 659 |
spin_lock_init(&proc->locks_spin); |
| 660 |
init_waitqueue_head(&proc->wait); |
| 661 |
file->private_data = proc; |
| 662 |
|
| 663 |
return 0; |
| 664 |
} |
| 665 |
|
| 666 |
static int device_close(struct inode *inode, struct file *file) |
| 667 |
{ |
| 668 |
struct dlm_user_proc *proc = file->private_data; |
| 669 |
struct dlm_ls *ls; |
| 670 |
|
| 671 |
ls = dlm_find_lockspace_local(proc->lockspace); |
| 672 |
if (!ls) |
| 673 |
return -ENOENT; |
| 674 |
|
| 675 |
set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); |
| 676 |
|
| 677 |
dlm_clear_proc_locks(ls, proc); |
| 678 |
|
| 679 |
/* at this point no more lkb's should exist for this lockspace, |
| 680 |
so there's no chance of dlm_user_add_ast() being called and |
| 681 |
looking for lkb->ua->proc */ |
| 682 |
|
| 683 |
kfree(proc); |
| 684 |
file->private_data = NULL; |
| 685 |
|
| 686 |
dlm_put_lockspace(ls); |
| 687 |
dlm_put_lockspace(ls); /* for the find in device_open() */ |
| 688 |
|
| 689 |
/* FIXME: AUTOFREE: if this ls is no longer used do |
| 690 |
device_remove_lockspace() */ |
| 691 |
|
| 692 |
return 0; |
| 693 |
} |
| 694 |
|
| 695 |
static int copy_result_to_user(struct dlm_user_args *ua, int compat, |
| 696 |
uint32_t flags, int mode, int copy_lvb, |
| 697 |
char __user *buf, size_t count) |
| 698 |
{ |
| 699 |
#ifdef CONFIG_COMPAT |
| 700 |
struct dlm_lock_result32 result32; |
| 701 |
#endif |
| 702 |
struct dlm_lock_result result; |
| 703 |
void *resultptr; |
| 704 |
int error=0; |
| 705 |
int len; |
| 706 |
int struct_len; |
| 707 |
|
| 708 |
memset(&result, 0, sizeof(struct dlm_lock_result)); |
| 709 |
result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
| 710 |
result.version[1] = DLM_DEVICE_VERSION_MINOR; |
| 711 |
result.version[2] = DLM_DEVICE_VERSION_PATCH; |
| 712 |
memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr)); |
| 713 |
result.user_lksb = ua->user_lksb; |
| 714 |
|
| 715 |
/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated |
| 716 |
in a conversion unless the conversion is successful. See code |
| 717 |
in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, |
| 718 |
notes that a new blocking AST address and parameter are set even if |
| 719 |
the conversion fails, so maybe we should just do that. */ |
| 720 |
|
| 721 |
if (flags & DLM_CB_BAST) { |
| 722 |
result.user_astaddr = ua->bastaddr; |
| 723 |
result.user_astparam = ua->bastparam; |
| 724 |
result.bast_mode = mode; |
| 725 |
} else { |
| 726 |
result.user_astaddr = ua->castaddr; |
| 727 |
result.user_astparam = ua->castparam; |
| 728 |
} |
| 729 |
|
| 730 |
#ifdef CONFIG_COMPAT |
| 731 |
if (compat) |
| 732 |
len = sizeof(struct dlm_lock_result32); |
| 733 |
else |
| 734 |
#endif |
| 735 |
len = sizeof(struct dlm_lock_result); |
| 736 |
struct_len = len; |
| 737 |
|
| 738 |
/* copy lvb to userspace if there is one, it's been updated, and |
| 739 |
the user buffer has space for it */ |
| 740 |
|
| 741 |
if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) { |
| 742 |
if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, |
| 743 |
DLM_USER_LVB_LEN)) { |
| 744 |
error = -EFAULT; |
| 745 |
goto out; |
| 746 |
} |
| 747 |
|
| 748 |
result.lvb_offset = len; |
| 749 |
len += DLM_USER_LVB_LEN; |
| 750 |
} |
| 751 |
|
| 752 |
result.length = len; |
| 753 |
resultptr = &result; |
| 754 |
#ifdef CONFIG_COMPAT |
| 755 |
if (compat) { |
| 756 |
compat_output(&result, &result32); |
| 757 |
resultptr = &result32; |
| 758 |
} |
| 759 |
#endif |
| 760 |
|
| 761 |
if (copy_to_user(buf, resultptr, struct_len)) |
| 762 |
error = -EFAULT; |
| 763 |
else |
| 764 |
error = len; |
| 765 |
out: |
| 766 |
return error; |
| 767 |
} |
| 768 |
|
| 769 |
static int copy_version_to_user(char __user *buf, size_t count) |
| 770 |
{ |
| 771 |
struct dlm_device_version ver; |
| 772 |
|
| 773 |
memset(&ver, 0, sizeof(struct dlm_device_version)); |
| 774 |
ver.version[0] = DLM_DEVICE_VERSION_MAJOR; |
| 775 |
ver.version[1] = DLM_DEVICE_VERSION_MINOR; |
| 776 |
ver.version[2] = DLM_DEVICE_VERSION_PATCH; |
| 777 |
|
| 778 |
if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) |
| 779 |
return -EFAULT; |
| 780 |
return sizeof(struct dlm_device_version); |
| 781 |
} |
| 782 |
|
| 783 |
/* a read returns a single ast described in a struct dlm_lock_result */ |
| 784 |
|
| 785 |
static ssize_t device_read(struct file *file, char __user *buf, size_t count, |
| 786 |
loff_t *ppos) |
| 787 |
{ |
| 788 |
struct dlm_user_proc *proc = file->private_data; |
| 789 |
DECLARE_WAITQUEUE(wait, current); |
| 790 |
struct dlm_callback *cb; |
| 791 |
int rv, ret; |
| 792 |
|
| 793 |
if (count == sizeof(struct dlm_device_version)) { |
| 794 |
rv = copy_version_to_user(buf, count); |
| 795 |
return rv; |
| 796 |
} |
| 797 |
|
| 798 |
if (!proc) { |
| 799 |
log_print("non-version read from control device %zu", count); |
| 800 |
return -EINVAL; |
| 801 |
} |
| 802 |
|
| 803 |
#ifdef CONFIG_COMPAT |
| 804 |
if (count < sizeof(struct dlm_lock_result32)) |
| 805 |
#else |
| 806 |
if (count < sizeof(struct dlm_lock_result)) |
| 807 |
#endif |
| 808 |
return -EINVAL; |
| 809 |
|
| 810 |
/* do we really need this? can a read happen after a close? */ |
| 811 |
if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) |
| 812 |
return -EINVAL; |
| 813 |
|
| 814 |
spin_lock_bh(&proc->asts_spin); |
| 815 |
if (list_empty(&proc->asts)) { |
| 816 |
if (file->f_flags & O_NONBLOCK) { |
| 817 |
spin_unlock_bh(&proc->asts_spin); |
| 818 |
return -EAGAIN; |
| 819 |
} |
| 820 |
|
| 821 |
add_wait_queue(&proc->wait, &wait); |
| 822 |
|
| 823 |
repeat: |
| 824 |
set_current_state(TASK_INTERRUPTIBLE); |
| 825 |
if (list_empty(&proc->asts) && !signal_pending(current)) { |
| 826 |
spin_unlock_bh(&proc->asts_spin); |
| 827 |
schedule(); |
| 828 |
spin_lock_bh(&proc->asts_spin); |
| 829 |
goto repeat; |
| 830 |
} |
| 831 |
set_current_state(TASK_RUNNING); |
| 832 |
remove_wait_queue(&proc->wait, &wait); |
| 833 |
|
| 834 |
if (signal_pending(current)) { |
| 835 |
spin_unlock_bh(&proc->asts_spin); |
| 836 |
return -ERESTARTSYS; |
| 837 |
} |
| 838 |
} |
| 839 |
|
| 840 |
/* if we empty lkb_callbacks, we don't want to unlock the spinlock |
| 841 |
without removing lkb_cb_list; so empty lkb_cb_list is always |
| 842 |
consistent with empty lkb_callbacks */ |
| 843 |
|
| 844 |
cb = list_first_entry(&proc->asts, struct dlm_callback, list); |
| 845 |
list_del(&cb->list); |
| 846 |
spin_unlock_bh(&proc->asts_spin); |
| 847 |
|
| 848 |
if (cb->flags & DLM_CB_BAST) { |
| 849 |
trace_dlm_bast(cb->ls_id, cb->lkb_id, cb->mode, cb->res_name, |
| 850 |
cb->res_length); |
| 851 |
} else if (cb->flags & DLM_CB_CAST) { |
| 852 |
cb->lkb_lksb->sb_status = cb->sb_status; |
| 853 |
cb->lkb_lksb->sb_flags = cb->sb_flags; |
| 854 |
trace_dlm_ast(cb->ls_id, cb->lkb_id, cb->sb_status, |
| 855 |
cb->sb_flags, cb->res_name, cb->res_length); |
| 856 |
} |
| 857 |
|
| 858 |
ret = copy_result_to_user(&cb->ua, |
| 859 |
test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), |
| 860 |
cb->flags, cb->mode, cb->copy_lvb, buf, count); |
| 861 |
dlm_free_cb(cb); |
| 862 |
return ret; |
| 863 |
} |
| 864 |
|
| 865 |
static __poll_t device_poll(struct file *file, poll_table *wait) |
| 866 |
{ |
| 867 |
struct dlm_user_proc *proc = file->private_data; |
| 868 |
|
| 869 |
poll_wait(file, &proc->wait, wait); |
| 870 |
|
| 871 |
spin_lock_bh(&proc->asts_spin); |
| 872 |
if (!list_empty(&proc->asts)) { |
| 873 |
spin_unlock_bh(&proc->asts_spin); |
| 874 |
return EPOLLIN | EPOLLRDNORM; |
| 875 |
} |
| 876 |
spin_unlock_bh(&proc->asts_spin); |
| 877 |
return 0; |
| 878 |
} |
| 879 |
|
| 880 |
int dlm_user_daemon_available(void) |
| 881 |
{ |
| 882 |
/* dlm_controld hasn't started (or, has started, but not |
| 883 |
properly populated configfs) */ |
| 884 |
|
| 885 |
if (!dlm_our_nodeid()) |
| 886 |
return 0; |
| 887 |
|
| 888 |
/* This is to deal with versions of dlm_controld that don't |
| 889 |
know about the monitor device. We assume that if the |
| 890 |
dlm_controld was started (above), but the monitor device |
| 891 |
was never opened, that it's an old version. dlm_controld |
| 892 |
should open the monitor device before populating configfs. */ |
| 893 |
|
| 894 |
if (dlm_monitor_unused) |
| 895 |
return 1; |
| 896 |
|
| 897 |
return atomic_read(&dlm_monitor_opened) ? 1 : 0; |
| 898 |
} |
| 899 |
|
| 900 |
static int ctl_device_open(struct inode *inode, struct file *file) |
| 901 |
{ |
| 902 |
file->private_data = NULL; |
| 903 |
return 0; |
| 904 |
} |
| 905 |
|
| 906 |
static int ctl_device_close(struct inode *inode, struct file *file) |
| 907 |
{ |
| 908 |
return 0; |
| 909 |
} |
| 910 |
|
| 911 |
static int monitor_device_open(struct inode *inode, struct file *file) |
| 912 |
{ |
| 913 |
atomic_inc(&dlm_monitor_opened); |
| 914 |
dlm_monitor_unused = 0; |
| 915 |
return 0; |
| 916 |
} |
| 917 |
|
| 918 |
static int monitor_device_close(struct inode *inode, struct file *file) |
| 919 |
{ |
| 920 |
if (atomic_dec_and_test(&dlm_monitor_opened)) |
| 921 |
dlm_stop_lockspaces(); |
| 922 |
return 0; |
| 923 |
} |
| 924 |
|
| 925 |
static const struct file_operations device_fops = { |
| 926 |
.open = device_open, |
| 927 |
.release = device_close, |
| 928 |
.read = device_read, |
| 929 |
.write = device_write, |
| 930 |
.poll = device_poll, |
| 931 |
.owner = THIS_MODULE, |
| 932 |
.llseek = noop_llseek, |
| 933 |
}; |
| 934 |
|
| 935 |
static const struct file_operations ctl_device_fops = { |
| 936 |
.open = ctl_device_open, |
| 937 |
.release = ctl_device_close, |
| 938 |
.read = device_read, |
| 939 |
.write = device_write, |
| 940 |
.owner = THIS_MODULE, |
| 941 |
.llseek = noop_llseek, |
| 942 |
}; |
| 943 |
|
| 944 |
static struct miscdevice ctl_device = { |
| 945 |
.name = "dlm-control", |
| 946 |
.fops = &ctl_device_fops, |
| 947 |
.minor = MISC_DYNAMIC_MINOR, |
| 948 |
}; |
| 949 |
|
| 950 |
static const struct file_operations monitor_device_fops = { |
| 951 |
.open = monitor_device_open, |
| 952 |
.release = monitor_device_close, |
| 953 |
.owner = THIS_MODULE, |
| 954 |
.llseek = noop_llseek, |
| 955 |
}; |
| 956 |
|
| 957 |
static struct miscdevice monitor_device = { |
| 958 |
.name = "dlm-monitor", |
| 959 |
.fops = &monitor_device_fops, |
| 960 |
.minor = MISC_DYNAMIC_MINOR, |
| 961 |
}; |
| 962 |
|
| 963 |
int __init dlm_user_init(void) |
| 964 |
{ |
| 965 |
int error; |
| 966 |
|
| 967 |
atomic_set(&dlm_monitor_opened, 0); |
| 968 |
|
| 969 |
error = misc_register(&ctl_device); |
| 970 |
if (error) { |
| 971 |
log_print("misc_register failed for control device"); |
| 972 |
goto out; |
| 973 |
} |
| 974 |
|
| 975 |
error = misc_register(&monitor_device); |
| 976 |
if (error) { |
| 977 |
log_print("misc_register failed for monitor device"); |
| 978 |
misc_deregister(&ctl_device); |
| 979 |
} |
| 980 |
out: |
| 981 |
return error; |
| 982 |
} |
| 983 |
|
| 984 |
void dlm_user_exit(void) |
| 985 |
{ |
| 986 |
misc_deregister(&ctl_device); |
| 987 |
misc_deregister(&monitor_device); |
| 988 |
} |
| 989 |
|