| 1 | /* |
| 2 | * Copyright (C) 2006-2008 Red Hat, Inc. All rights reserved. |
| 3 | * |
| 4 | * This copyrighted material is made available to anyone wishing to use, |
| 5 | * modify, copy, or redistribute it subject to the terms and conditions |
| 6 | * of the GNU General Public License v.2. |
| 7 | */ |
| 8 | |
| 9 | #include <linux/miscdevice.h> |
| 10 | #include <linux/init.h> |
| 11 | #include <linux/wait.h> |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/file.h> |
| 14 | #include <linux/fs.h> |
| 15 | #include <linux/poll.h> |
| 16 | #include <linux/signal.h> |
| 17 | #include <linux/spinlock.h> |
| 18 | #include <linux/smp_lock.h> |
| 19 | #include <linux/dlm.h> |
| 20 | #include <linux/dlm_device.h> |
| 21 | |
| 22 | #include "dlm_internal.h" |
| 23 | #include "lockspace.h" |
| 24 | #include "lock.h" |
| 25 | #include "lvb_table.h" |
| 26 | #include "user.h" |
| 27 | |
| 28 | static const char name_prefix[] = "dlm"; |
| 29 | static const struct file_operations device_fops; |
| 30 | |
| 31 | #ifdef CONFIG_COMPAT |
| 32 | |
| 33 | struct dlm_lock_params32 { |
| 34 | __u8 mode; |
| 35 | __u8 namelen; |
| 36 | __u16 unused; |
| 37 | __u32 flags; |
| 38 | __u32 lkid; |
| 39 | __u32 parent; |
| 40 | __u64 xid; |
| 41 | __u64 timeout; |
| 42 | __u32 castparam; |
| 43 | __u32 castaddr; |
| 44 | __u32 bastparam; |
| 45 | __u32 bastaddr; |
| 46 | __u32 lksb; |
| 47 | char lvb[DLM_USER_LVB_LEN]; |
| 48 | char name[0]; |
| 49 | }; |
| 50 | |
| 51 | struct dlm_write_request32 { |
| 52 | __u32 version[3]; |
| 53 | __u8 cmd; |
| 54 | __u8 is64bit; |
| 55 | __u8 unused[2]; |
| 56 | |
| 57 | union { |
| 58 | struct dlm_lock_params32 lock; |
| 59 | struct dlm_lspace_params lspace; |
| 60 | struct dlm_purge_params purge; |
| 61 | } i; |
| 62 | }; |
| 63 | |
| 64 | struct dlm_lksb32 { |
| 65 | __u32 sb_status; |
| 66 | __u32 sb_lkid; |
| 67 | __u8 sb_flags; |
| 68 | __u32 sb_lvbptr; |
| 69 | }; |
| 70 | |
| 71 | struct dlm_lock_result32 { |
| 72 | __u32 version[3]; |
| 73 | __u32 length; |
| 74 | __u32 user_astaddr; |
| 75 | __u32 user_astparam; |
| 76 | __u32 user_lksb; |
| 77 | struct dlm_lksb32 lksb; |
| 78 | __u8 bast_mode; |
| 79 | __u8 unused[3]; |
| 80 | /* Offsets may be zero if no data is present */ |
| 81 | __u32 lvb_offset; |
| 82 | }; |
| 83 | |
| 84 | static void compat_input(struct dlm_write_request *kb, |
| 85 | struct dlm_write_request32 *kb32, |
| 86 | size_t count) |
| 87 | { |
| 88 | kb->version[0] = kb32->version[0]; |
| 89 | kb->version[1] = kb32->version[1]; |
| 90 | kb->version[2] = kb32->version[2]; |
| 91 | |
| 92 | kb->cmd = kb32->cmd; |
| 93 | kb->is64bit = kb32->is64bit; |
| 94 | if (kb->cmd == DLM_USER_CREATE_LOCKSPACE || |
| 95 | kb->cmd == DLM_USER_REMOVE_LOCKSPACE) { |
| 96 | kb->i.lspace.flags = kb32->i.lspace.flags; |
| 97 | kb->i.lspace.minor = kb32->i.lspace.minor; |
| 98 | memcpy(kb->i.lspace.name, kb32->i.lspace.name, count - |
| 99 | offsetof(struct dlm_write_request32, i.lspace.name)); |
| 100 | } else if (kb->cmd == DLM_USER_PURGE) { |
| 101 | kb->i.purge.nodeid = kb32->i.purge.nodeid; |
| 102 | kb->i.purge.pid = kb32->i.purge.pid; |
| 103 | } else { |
| 104 | kb->i.lock.mode = kb32->i.lock.mode; |
| 105 | kb->i.lock.namelen = kb32->i.lock.namelen; |
| 106 | kb->i.lock.flags = kb32->i.lock.flags; |
| 107 | kb->i.lock.lkid = kb32->i.lock.lkid; |
| 108 | kb->i.lock.parent = kb32->i.lock.parent; |
| 109 | kb->i.lock.xid = kb32->i.lock.xid; |
| 110 | kb->i.lock.timeout = kb32->i.lock.timeout; |
| 111 | kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam; |
| 112 | kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr; |
| 113 | kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam; |
| 114 | kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr; |
| 115 | kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb; |
| 116 | memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN); |
| 117 | memcpy(kb->i.lock.name, kb32->i.lock.name, count - |
| 118 | offsetof(struct dlm_write_request32, i.lock.name)); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | static void compat_output(struct dlm_lock_result *res, |
| 123 | struct dlm_lock_result32 *res32) |
| 124 | { |
| 125 | res32->version[0] = res->version[0]; |
| 126 | res32->version[1] = res->version[1]; |
| 127 | res32->version[2] = res->version[2]; |
| 128 | |
| 129 | res32->user_astaddr = (__u32)(long)res->user_astaddr; |
| 130 | res32->user_astparam = (__u32)(long)res->user_astparam; |
| 131 | res32->user_lksb = (__u32)(long)res->user_lksb; |
| 132 | res32->bast_mode = res->bast_mode; |
| 133 | |
| 134 | res32->lvb_offset = res->lvb_offset; |
| 135 | res32->length = res->length; |
| 136 | |
| 137 | res32->lksb.sb_status = res->lksb.sb_status; |
| 138 | res32->lksb.sb_flags = res->lksb.sb_flags; |
| 139 | res32->lksb.sb_lkid = res->lksb.sb_lkid; |
| 140 | res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; |
| 141 | } |
| 142 | #endif |
| 143 | |
| 144 | /* Figure out if this lock is at the end of its life and no longer |
| 145 | available for the application to use. The lkb still exists until |
| 146 | the final ast is read. A lock becomes EOL in three situations: |
| 147 | 1. a noqueue request fails with EAGAIN |
| 148 | 2. an unlock completes with EUNLOCK |
| 149 | 3. a cancel of a waiting request completes with ECANCEL/EDEADLK |
| 150 | An EOL lock needs to be removed from the process's list of locks. |
| 151 | And we can't allow any new operation on an EOL lock. This is |
| 152 | not related to the lifetime of the lkb struct which is managed |
| 153 | entirely by refcount. */ |
| 154 | |
| 155 | static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type) |
| 156 | { |
| 157 | switch (sb_status) { |
| 158 | case -DLM_EUNLOCK: |
| 159 | return 1; |
| 160 | case -DLM_ECANCEL: |
| 161 | case -ETIMEDOUT: |
| 162 | case -EDEADLK: |
| 163 | if (lkb->lkb_grmode == DLM_LOCK_IV) |
| 164 | return 1; |
| 165 | break; |
| 166 | case -EAGAIN: |
| 167 | if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV) |
| 168 | return 1; |
| 169 | break; |
| 170 | } |
| 171 | return 0; |
| 172 | } |
| 173 | |
| 174 | /* we could possibly check if the cancel of an orphan has resulted in the lkb |
| 175 | being removed and then remove that lkb from the orphans list and free it */ |
| 176 | |
| 177 | void dlm_user_add_ast(struct dlm_lkb *lkb, int type) |
| 178 | { |
| 179 | struct dlm_ls *ls; |
| 180 | struct dlm_user_args *ua; |
| 181 | struct dlm_user_proc *proc; |
| 182 | int eol = 0, ast_type; |
| 183 | |
| 184 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
| 185 | return; |
| 186 | |
| 187 | ls = lkb->lkb_resource->res_ls; |
| 188 | mutex_lock(&ls->ls_clear_proc_locks); |
| 189 | |
| 190 | /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast |
| 191 | can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed |
| 192 | lkb->ua so we can't try to use it. This second check is necessary |
| 193 | for cases where a completion ast is received for an operation that |
| 194 | began before clear_proc_locks did its cancel/unlock. */ |
| 195 | |
| 196 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
| 197 | goto out; |
| 198 | |
| 199 | DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb);); |
| 200 | ua = lkb->lkb_ua; |
| 201 | proc = ua->proc; |
| 202 | |
| 203 | if (type == AST_BAST && ua->bastaddr == NULL) |
| 204 | goto out; |
| 205 | |
| 206 | spin_lock(&proc->asts_spin); |
| 207 | |
| 208 | ast_type = lkb->lkb_ast_type; |
| 209 | lkb->lkb_ast_type |= type; |
| 210 | |
| 211 | if (!ast_type) { |
| 212 | kref_get(&lkb->lkb_ref); |
| 213 | list_add_tail(&lkb->lkb_astqueue, &proc->asts); |
| 214 | wake_up_interruptible(&proc->wait); |
| 215 | } |
| 216 | if (type == AST_COMP && (ast_type & AST_COMP)) |
| 217 | log_debug(ls, "ast overlap %x status %x %x", |
| 218 | lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags); |
| 219 | |
| 220 | eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type); |
| 221 | if (eol) { |
| 222 | lkb->lkb_ast_type &= ~AST_BAST; |
| 223 | lkb->lkb_flags |= DLM_IFL_ENDOFLIFE; |
| 224 | } |
| 225 | |
| 226 | /* We want to copy the lvb to userspace when the completion |
| 227 | ast is read if the status is 0, the lock has an lvb and |
| 228 | lvb_ops says we should. We could probably have set_lvb_lock() |
| 229 | set update_user_lvb instead and not need old_mode */ |
| 230 | |
| 231 | if ((lkb->lkb_ast_type & AST_COMP) && |
| 232 | (lkb->lkb_lksb->sb_status == 0) && |
| 233 | lkb->lkb_lksb->sb_lvbptr && |
| 234 | dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1]) |
| 235 | ua->update_user_lvb = 1; |
| 236 | else |
| 237 | ua->update_user_lvb = 0; |
| 238 | |
| 239 | spin_unlock(&proc->asts_spin); |
| 240 | |
| 241 | if (eol) { |
| 242 | spin_lock(&proc->locks_spin); |
| 243 | if (!list_empty(&lkb->lkb_ownqueue)) { |
| 244 | list_del_init(&lkb->lkb_ownqueue); |
| 245 | dlm_put_lkb(lkb); |
| 246 | } |
| 247 | spin_unlock(&proc->locks_spin); |
| 248 | } |
| 249 | out: |
| 250 | mutex_unlock(&ls->ls_clear_proc_locks); |
| 251 | } |
| 252 | |
| 253 | static int device_user_lock(struct dlm_user_proc *proc, |
| 254 | struct dlm_lock_params *params) |
| 255 | { |
| 256 | struct dlm_ls *ls; |
| 257 | struct dlm_user_args *ua; |
| 258 | int error = -ENOMEM; |
| 259 | |
| 260 | ls = dlm_find_lockspace_local(proc->lockspace); |
| 261 | if (!ls) |
| 262 | return -ENOENT; |
| 263 | |
| 264 | if (!params->castaddr || !params->lksb) { |
| 265 | error = -EINVAL; |
| 266 | goto out; |
| 267 | } |
| 268 | |
| 269 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); |
| 270 | if (!ua) |
| 271 | goto out; |
| 272 | ua->proc = proc; |
| 273 | ua->user_lksb = params->lksb; |
| 274 | ua->castparam = params->castparam; |
| 275 | ua->castaddr = params->castaddr; |
| 276 | ua->bastparam = params->bastparam; |
| 277 | ua->bastaddr = params->bastaddr; |
| 278 | ua->xid = params->xid; |
| 279 | |
| 280 | if (params->flags & DLM_LKF_CONVERT) |
| 281 | error = dlm_user_convert(ls, ua, |
| 282 | params->mode, params->flags, |
| 283 | params->lkid, params->lvb, |
| 284 | (unsigned long) params->timeout); |
| 285 | else { |
| 286 | error = dlm_user_request(ls, ua, |
| 287 | params->mode, params->flags, |
| 288 | params->name, params->namelen, |
| 289 | (unsigned long) params->timeout); |
| 290 | if (!error) |
| 291 | error = ua->lksb.sb_lkid; |
| 292 | } |
| 293 | out: |
| 294 | dlm_put_lockspace(ls); |
| 295 | return error; |
| 296 | } |
| 297 | |
| 298 | static int device_user_unlock(struct dlm_user_proc *proc, |
| 299 | struct dlm_lock_params *params) |
| 300 | { |
| 301 | struct dlm_ls *ls; |
| 302 | struct dlm_user_args *ua; |
| 303 | int error = -ENOMEM; |
| 304 | |
| 305 | ls = dlm_find_lockspace_local(proc->lockspace); |
| 306 | if (!ls) |
| 307 | return -ENOENT; |
| 308 | |
| 309 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); |
| 310 | if (!ua) |
| 311 | goto out; |
| 312 | ua->proc = proc; |
| 313 | ua->user_lksb = params->lksb; |
| 314 | ua->castparam = params->castparam; |
| 315 | ua->castaddr = params->castaddr; |
| 316 | |
| 317 | if (params->flags & DLM_LKF_CANCEL) |
| 318 | error = dlm_user_cancel(ls, ua, params->flags, params->lkid); |
| 319 | else |
| 320 | error = dlm_user_unlock(ls, ua, params->flags, params->lkid, |
| 321 | params->lvb); |
| 322 | out: |
| 323 | dlm_put_lockspace(ls); |
| 324 | return error; |
| 325 | } |
| 326 | |
| 327 | static int device_user_deadlock(struct dlm_user_proc *proc, |
| 328 | struct dlm_lock_params *params) |
| 329 | { |
| 330 | struct dlm_ls *ls; |
| 331 | int error; |
| 332 | |
| 333 | ls = dlm_find_lockspace_local(proc->lockspace); |
| 334 | if (!ls) |
| 335 | return -ENOENT; |
| 336 | |
| 337 | error = dlm_user_deadlock(ls, params->flags, params->lkid); |
| 338 | |
| 339 | dlm_put_lockspace(ls); |
| 340 | return error; |
| 341 | } |
| 342 | |
| 343 | static int dlm_device_register(struct dlm_ls *ls, char *name) |
| 344 | { |
| 345 | int error, len; |
| 346 | |
| 347 | /* The device is already registered. This happens when the |
| 348 | lockspace is created multiple times from userspace. */ |
| 349 | if (ls->ls_device.name) |
| 350 | return 0; |
| 351 | |
| 352 | error = -ENOMEM; |
| 353 | len = strlen(name) + strlen(name_prefix) + 2; |
| 354 | ls->ls_device.name = kzalloc(len, GFP_KERNEL); |
| 355 | if (!ls->ls_device.name) |
| 356 | goto fail; |
| 357 | |
| 358 | snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, |
| 359 | name); |
| 360 | ls->ls_device.fops = &device_fops; |
| 361 | ls->ls_device.minor = MISC_DYNAMIC_MINOR; |
| 362 | |
| 363 | error = misc_register(&ls->ls_device); |
| 364 | if (error) { |
| 365 | kfree(ls->ls_device.name); |
| 366 | } |
| 367 | fail: |
| 368 | return error; |
| 369 | } |
| 370 | |
| 371 | int dlm_device_deregister(struct dlm_ls *ls) |
| 372 | { |
| 373 | int error; |
| 374 | |
| 375 | /* The device is not registered. This happens when the lockspace |
| 376 | was never used from userspace, or when device_create_lockspace() |
| 377 | calls dlm_release_lockspace() after the register fails. */ |
| 378 | if (!ls->ls_device.name) |
| 379 | return 0; |
| 380 | |
| 381 | error = misc_deregister(&ls->ls_device); |
| 382 | if (!error) |
| 383 | kfree(ls->ls_device.name); |
| 384 | return error; |
| 385 | } |
| 386 | |
| 387 | static int device_user_purge(struct dlm_user_proc *proc, |
| 388 | struct dlm_purge_params *params) |
| 389 | { |
| 390 | struct dlm_ls *ls; |
| 391 | int error; |
| 392 | |
| 393 | ls = dlm_find_lockspace_local(proc->lockspace); |
| 394 | if (!ls) |
| 395 | return -ENOENT; |
| 396 | |
| 397 | error = dlm_user_purge(ls, proc, params->nodeid, params->pid); |
| 398 | |
| 399 | dlm_put_lockspace(ls); |
| 400 | return error; |
| 401 | } |
| 402 | |
| 403 | static int device_create_lockspace(struct dlm_lspace_params *params) |
| 404 | { |
| 405 | dlm_lockspace_t *lockspace; |
| 406 | struct dlm_ls *ls; |
| 407 | int error; |
| 408 | |
| 409 | if (!capable(CAP_SYS_ADMIN)) |
| 410 | return -EPERM; |
| 411 | |
| 412 | error = dlm_new_lockspace(params->name, strlen(params->name), |
| 413 | &lockspace, params->flags, DLM_USER_LVB_LEN); |
| 414 | if (error) |
| 415 | return error; |
| 416 | |
| 417 | ls = dlm_find_lockspace_local(lockspace); |
| 418 | if (!ls) |
| 419 | return -ENOENT; |
| 420 | |
| 421 | error = dlm_device_register(ls, params->name); |
| 422 | dlm_put_lockspace(ls); |
| 423 | |
| 424 | if (error) |
| 425 | dlm_release_lockspace(lockspace, 0); |
| 426 | else |
| 427 | error = ls->ls_device.minor; |
| 428 | |
| 429 | return error; |
| 430 | } |
| 431 | |
| 432 | static int device_remove_lockspace(struct dlm_lspace_params *params) |
| 433 | { |
| 434 | dlm_lockspace_t *lockspace; |
| 435 | struct dlm_ls *ls; |
| 436 | int error, force = 0; |
| 437 | |
| 438 | if (!capable(CAP_SYS_ADMIN)) |
| 439 | return -EPERM; |
| 440 | |
| 441 | ls = dlm_find_lockspace_device(params->minor); |
| 442 | if (!ls) |
| 443 | return -ENOENT; |
| 444 | |
| 445 | if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
| 446 | force = 2; |
| 447 | |
| 448 | lockspace = ls->ls_local_handle; |
| 449 | dlm_put_lockspace(ls); |
| 450 | |
| 451 | /* The final dlm_release_lockspace waits for references to go to |
| 452 | zero, so all processes will need to close their device for the |
| 453 | ls before the release will proceed. release also calls the |
| 454 | device_deregister above. Converting a positive return value |
| 455 | from release to zero means that userspace won't know when its |
| 456 | release was the final one, but it shouldn't need to know. */ |
| 457 | |
| 458 | error = dlm_release_lockspace(lockspace, force); |
| 459 | if (error > 0) |
| 460 | error = 0; |
| 461 | return error; |
| 462 | } |
| 463 | |
| 464 | /* Check the user's version matches ours */ |
| 465 | static int check_version(struct dlm_write_request *req) |
| 466 | { |
| 467 | if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || |
| 468 | (req->version[0] == DLM_DEVICE_VERSION_MAJOR && |
| 469 | req->version[1] > DLM_DEVICE_VERSION_MINOR)) { |
| 470 | |
| 471 | printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " |
| 472 | "user (%d.%d.%d) kernel (%d.%d.%d)\n", |
| 473 | current->comm, |
| 474 | task_pid_nr(current), |
| 475 | req->version[0], |
| 476 | req->version[1], |
| 477 | req->version[2], |
| 478 | DLM_DEVICE_VERSION_MAJOR, |
| 479 | DLM_DEVICE_VERSION_MINOR, |
| 480 | DLM_DEVICE_VERSION_PATCH); |
| 481 | return -EINVAL; |
| 482 | } |
| 483 | return 0; |
| 484 | } |
| 485 | |
| 486 | /* |
| 487 | * device_write |
| 488 | * |
| 489 | * device_user_lock |
| 490 | * dlm_user_request -> request_lock |
| 491 | * dlm_user_convert -> convert_lock |
| 492 | * |
| 493 | * device_user_unlock |
| 494 | * dlm_user_unlock -> unlock_lock |
| 495 | * dlm_user_cancel -> cancel_lock |
| 496 | * |
| 497 | * device_create_lockspace |
| 498 | * dlm_new_lockspace |
| 499 | * |
| 500 | * device_remove_lockspace |
| 501 | * dlm_release_lockspace |
| 502 | */ |
| 503 | |
| 504 | /* a write to a lockspace device is a lock or unlock request, a write |
| 505 | to the control device is to create/remove a lockspace */ |
| 506 | |
| 507 | static ssize_t device_write(struct file *file, const char __user *buf, |
| 508 | size_t count, loff_t *ppos) |
| 509 | { |
| 510 | struct dlm_user_proc *proc = file->private_data; |
| 511 | struct dlm_write_request *kbuf; |
| 512 | sigset_t tmpsig, allsigs; |
| 513 | int error; |
| 514 | |
| 515 | #ifdef CONFIG_COMPAT |
| 516 | if (count < sizeof(struct dlm_write_request32)) |
| 517 | #else |
| 518 | if (count < sizeof(struct dlm_write_request)) |
| 519 | #endif |
| 520 | return -EINVAL; |
| 521 | |
| 522 | kbuf = kzalloc(count + 1, GFP_KERNEL); |
| 523 | if (!kbuf) |
| 524 | return -ENOMEM; |
| 525 | |
| 526 | if (copy_from_user(kbuf, buf, count)) { |
| 527 | error = -EFAULT; |
| 528 | goto out_free; |
| 529 | } |
| 530 | |
| 531 | if (check_version(kbuf)) { |
| 532 | error = -EBADE; |
| 533 | goto out_free; |
| 534 | } |
| 535 | |
| 536 | #ifdef CONFIG_COMPAT |
| 537 | if (!kbuf->is64bit) { |
| 538 | struct dlm_write_request32 *k32buf; |
| 539 | k32buf = (struct dlm_write_request32 *)kbuf; |
| 540 | kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) - |
| 541 | sizeof(struct dlm_write_request32)), GFP_KERNEL); |
| 542 | if (!kbuf) { |
| 543 | kfree(k32buf); |
| 544 | return -ENOMEM; |
| 545 | } |
| 546 | |
| 547 | if (proc) |
| 548 | set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); |
| 549 | compat_input(kbuf, k32buf, count + 1); |
| 550 | kfree(k32buf); |
| 551 | } |
| 552 | #endif |
| 553 | |
| 554 | /* do we really need this? can a write happen after a close? */ |
| 555 | if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && |
| 556 | (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) { |
| 557 | error = -EINVAL; |
| 558 | goto out_free; |
| 559 | } |
| 560 | |
| 561 | sigfillset(&allsigs); |
| 562 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); |
| 563 | |
| 564 | error = -EINVAL; |
| 565 | |
| 566 | switch (kbuf->cmd) |
| 567 | { |
| 568 | case DLM_USER_LOCK: |
| 569 | if (!proc) { |
| 570 | log_print("no locking on control device"); |
| 571 | goto out_sig; |
| 572 | } |
| 573 | error = device_user_lock(proc, &kbuf->i.lock); |
| 574 | break; |
| 575 | |
| 576 | case DLM_USER_UNLOCK: |
| 577 | if (!proc) { |
| 578 | log_print("no locking on control device"); |
| 579 | goto out_sig; |
| 580 | } |
| 581 | error = device_user_unlock(proc, &kbuf->i.lock); |
| 582 | break; |
| 583 | |
| 584 | case DLM_USER_DEADLOCK: |
| 585 | if (!proc) { |
| 586 | log_print("no locking on control device"); |
| 587 | goto out_sig; |
| 588 | } |
| 589 | error = device_user_deadlock(proc, &kbuf->i.lock); |
| 590 | break; |
| 591 | |
| 592 | case DLM_USER_CREATE_LOCKSPACE: |
| 593 | if (proc) { |
| 594 | log_print("create/remove only on control device"); |
| 595 | goto out_sig; |
| 596 | } |
| 597 | error = device_create_lockspace(&kbuf->i.lspace); |
| 598 | break; |
| 599 | |
| 600 | case DLM_USER_REMOVE_LOCKSPACE: |
| 601 | if (proc) { |
| 602 | log_print("create/remove only on control device"); |
| 603 | goto out_sig; |
| 604 | } |
| 605 | error = device_remove_lockspace(&kbuf->i.lspace); |
| 606 | break; |
| 607 | |
| 608 | case DLM_USER_PURGE: |
| 609 | if (!proc) { |
| 610 | log_print("no locking on control device"); |
| 611 | goto out_sig; |
| 612 | } |
| 613 | error = device_user_purge(proc, &kbuf->i.purge); |
| 614 | break; |
| 615 | |
| 616 | default: |
| 617 | log_print("Unknown command passed to DLM device : %d\n", |
| 618 | kbuf->cmd); |
| 619 | } |
| 620 | |
| 621 | out_sig: |
| 622 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); |
| 623 | recalc_sigpending(); |
| 624 | out_free: |
| 625 | kfree(kbuf); |
| 626 | return error; |
| 627 | } |
| 628 | |
| 629 | /* Every process that opens the lockspace device has its own "proc" structure |
| 630 | hanging off the open file that's used to keep track of locks owned by the |
| 631 | process and asts that need to be delivered to the process. */ |
| 632 | |
| 633 | static int device_open(struct inode *inode, struct file *file) |
| 634 | { |
| 635 | struct dlm_user_proc *proc; |
| 636 | struct dlm_ls *ls; |
| 637 | |
| 638 | lock_kernel(); |
| 639 | ls = dlm_find_lockspace_device(iminor(inode)); |
| 640 | if (!ls) { |
| 641 | unlock_kernel(); |
| 642 | return -ENOENT; |
| 643 | } |
| 644 | |
| 645 | proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL); |
| 646 | if (!proc) { |
| 647 | dlm_put_lockspace(ls); |
| 648 | unlock_kernel(); |
| 649 | return -ENOMEM; |
| 650 | } |
| 651 | |
| 652 | proc->lockspace = ls->ls_local_handle; |
| 653 | INIT_LIST_HEAD(&proc->asts); |
| 654 | INIT_LIST_HEAD(&proc->locks); |
| 655 | INIT_LIST_HEAD(&proc->unlocking); |
| 656 | spin_lock_init(&proc->asts_spin); |
| 657 | spin_lock_init(&proc->locks_spin); |
| 658 | init_waitqueue_head(&proc->wait); |
| 659 | file->private_data = proc; |
| 660 | unlock_kernel(); |
| 661 | |
| 662 | return 0; |
| 663 | } |
| 664 | |
| 665 | static int device_close(struct inode *inode, struct file *file) |
| 666 | { |
| 667 | struct dlm_user_proc *proc = file->private_data; |
| 668 | struct dlm_ls *ls; |
| 669 | sigset_t tmpsig, allsigs; |
| 670 | |
| 671 | ls = dlm_find_lockspace_local(proc->lockspace); |
| 672 | if (!ls) |
| 673 | return -ENOENT; |
| 674 | |
| 675 | sigfillset(&allsigs); |
| 676 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); |
| 677 | |
| 678 | set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); |
| 679 | |
| 680 | dlm_clear_proc_locks(ls, proc); |
| 681 | |
| 682 | /* at this point no more lkb's should exist for this lockspace, |
| 683 | so there's no chance of dlm_user_add_ast() being called and |
| 684 | looking for lkb->ua->proc */ |
| 685 | |
| 686 | kfree(proc); |
| 687 | file->private_data = NULL; |
| 688 | |
| 689 | dlm_put_lockspace(ls); |
| 690 | dlm_put_lockspace(ls); /* for the find in device_open() */ |
| 691 | |
| 692 | /* FIXME: AUTOFREE: if this ls is no longer used do |
| 693 | device_remove_lockspace() */ |
| 694 | |
| 695 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); |
| 696 | recalc_sigpending(); |
| 697 | |
| 698 | return 0; |
| 699 | } |
| 700 | |
| 701 | static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type, |
| 702 | int bmode, char __user *buf, size_t count) |
| 703 | { |
| 704 | #ifdef CONFIG_COMPAT |
| 705 | struct dlm_lock_result32 result32; |
| 706 | #endif |
| 707 | struct dlm_lock_result result; |
| 708 | void *resultptr; |
| 709 | int error=0; |
| 710 | int len; |
| 711 | int struct_len; |
| 712 | |
| 713 | memset(&result, 0, sizeof(struct dlm_lock_result)); |
| 714 | result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
| 715 | result.version[1] = DLM_DEVICE_VERSION_MINOR; |
| 716 | result.version[2] = DLM_DEVICE_VERSION_PATCH; |
| 717 | memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb)); |
| 718 | result.user_lksb = ua->user_lksb; |
| 719 | |
| 720 | /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated |
| 721 | in a conversion unless the conversion is successful. See code |
| 722 | in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, |
| 723 | notes that a new blocking AST address and parameter are set even if |
| 724 | the conversion fails, so maybe we should just do that. */ |
| 725 | |
| 726 | if (type == AST_BAST) { |
| 727 | result.user_astaddr = ua->bastaddr; |
| 728 | result.user_astparam = ua->bastparam; |
| 729 | result.bast_mode = bmode; |
| 730 | } else { |
| 731 | result.user_astaddr = ua->castaddr; |
| 732 | result.user_astparam = ua->castparam; |
| 733 | } |
| 734 | |
| 735 | #ifdef CONFIG_COMPAT |
| 736 | if (compat) |
| 737 | len = sizeof(struct dlm_lock_result32); |
| 738 | else |
| 739 | #endif |
| 740 | len = sizeof(struct dlm_lock_result); |
| 741 | struct_len = len; |
| 742 | |
| 743 | /* copy lvb to userspace if there is one, it's been updated, and |
| 744 | the user buffer has space for it */ |
| 745 | |
| 746 | if (ua->update_user_lvb && ua->lksb.sb_lvbptr && |
| 747 | count >= len + DLM_USER_LVB_LEN) { |
| 748 | if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, |
| 749 | DLM_USER_LVB_LEN)) { |
| 750 | error = -EFAULT; |
| 751 | goto out; |
| 752 | } |
| 753 | |
| 754 | result.lvb_offset = len; |
| 755 | len += DLM_USER_LVB_LEN; |
| 756 | } |
| 757 | |
| 758 | result.length = len; |
| 759 | resultptr = &result; |
| 760 | #ifdef CONFIG_COMPAT |
| 761 | if (compat) { |
| 762 | compat_output(&result, &result32); |
| 763 | resultptr = &result32; |
| 764 | } |
| 765 | #endif |
| 766 | |
| 767 | if (copy_to_user(buf, resultptr, struct_len)) |
| 768 | error = -EFAULT; |
| 769 | else |
| 770 | error = len; |
| 771 | out: |
| 772 | return error; |
| 773 | } |
| 774 | |
| 775 | static int copy_version_to_user(char __user *buf, size_t count) |
| 776 | { |
| 777 | struct dlm_device_version ver; |
| 778 | |
| 779 | memset(&ver, 0, sizeof(struct dlm_device_version)); |
| 780 | ver.version[0] = DLM_DEVICE_VERSION_MAJOR; |
| 781 | ver.version[1] = DLM_DEVICE_VERSION_MINOR; |
| 782 | ver.version[2] = DLM_DEVICE_VERSION_PATCH; |
| 783 | |
| 784 | if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) |
| 785 | return -EFAULT; |
| 786 | return sizeof(struct dlm_device_version); |
| 787 | } |
| 788 | |
| 789 | /* a read returns a single ast described in a struct dlm_lock_result */ |
| 790 | |
| 791 | static ssize_t device_read(struct file *file, char __user *buf, size_t count, |
| 792 | loff_t *ppos) |
| 793 | { |
| 794 | struct dlm_user_proc *proc = file->private_data; |
| 795 | struct dlm_lkb *lkb; |
| 796 | DECLARE_WAITQUEUE(wait, current); |
| 797 | int error, type=0, bmode=0, removed = 0; |
| 798 | |
| 799 | if (count == sizeof(struct dlm_device_version)) { |
| 800 | error = copy_version_to_user(buf, count); |
| 801 | return error; |
| 802 | } |
| 803 | |
| 804 | if (!proc) { |
| 805 | log_print("non-version read from control device %zu", count); |
| 806 | return -EINVAL; |
| 807 | } |
| 808 | |
| 809 | #ifdef CONFIG_COMPAT |
| 810 | if (count < sizeof(struct dlm_lock_result32)) |
| 811 | #else |
| 812 | if (count < sizeof(struct dlm_lock_result)) |
| 813 | #endif |
| 814 | return -EINVAL; |
| 815 | |
| 816 | /* do we really need this? can a read happen after a close? */ |
| 817 | if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) |
| 818 | return -EINVAL; |
| 819 | |
| 820 | spin_lock(&proc->asts_spin); |
| 821 | if (list_empty(&proc->asts)) { |
| 822 | if (file->f_flags & O_NONBLOCK) { |
| 823 | spin_unlock(&proc->asts_spin); |
| 824 | return -EAGAIN; |
| 825 | } |
| 826 | |
| 827 | add_wait_queue(&proc->wait, &wait); |
| 828 | |
| 829 | repeat: |
| 830 | set_current_state(TASK_INTERRUPTIBLE); |
| 831 | if (list_empty(&proc->asts) && !signal_pending(current)) { |
| 832 | spin_unlock(&proc->asts_spin); |
| 833 | schedule(); |
| 834 | spin_lock(&proc->asts_spin); |
| 835 | goto repeat; |
| 836 | } |
| 837 | set_current_state(TASK_RUNNING); |
| 838 | remove_wait_queue(&proc->wait, &wait); |
| 839 | |
| 840 | if (signal_pending(current)) { |
| 841 | spin_unlock(&proc->asts_spin); |
| 842 | return -ERESTARTSYS; |
| 843 | } |
| 844 | } |
| 845 | |
| 846 | /* there may be both completion and blocking asts to return for |
| 847 | the lkb, don't remove lkb from asts list unless no asts remain */ |
| 848 | |
| 849 | lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue); |
| 850 | |
| 851 | if (lkb->lkb_ast_type & AST_COMP) { |
| 852 | lkb->lkb_ast_type &= ~AST_COMP; |
| 853 | type = AST_COMP; |
| 854 | } else if (lkb->lkb_ast_type & AST_BAST) { |
| 855 | lkb->lkb_ast_type &= ~AST_BAST; |
| 856 | type = AST_BAST; |
| 857 | bmode = lkb->lkb_bastmode; |
| 858 | } |
| 859 | |
| 860 | if (!lkb->lkb_ast_type) { |
| 861 | list_del(&lkb->lkb_astqueue); |
| 862 | removed = 1; |
| 863 | } |
| 864 | spin_unlock(&proc->asts_spin); |
| 865 | |
| 866 | error = copy_result_to_user(lkb->lkb_ua, |
| 867 | test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), |
| 868 | type, bmode, buf, count); |
| 869 | |
| 870 | /* removes reference for the proc->asts lists added by |
| 871 | dlm_user_add_ast() and may result in the lkb being freed */ |
| 872 | if (removed) |
| 873 | dlm_put_lkb(lkb); |
| 874 | |
| 875 | return error; |
| 876 | } |
| 877 | |
| 878 | static unsigned int device_poll(struct file *file, poll_table *wait) |
| 879 | { |
| 880 | struct dlm_user_proc *proc = file->private_data; |
| 881 | |
| 882 | poll_wait(file, &proc->wait, wait); |
| 883 | |
| 884 | spin_lock(&proc->asts_spin); |
| 885 | if (!list_empty(&proc->asts)) { |
| 886 | spin_unlock(&proc->asts_spin); |
| 887 | return POLLIN | POLLRDNORM; |
| 888 | } |
| 889 | spin_unlock(&proc->asts_spin); |
| 890 | return 0; |
| 891 | } |
| 892 | |
| 893 | static int ctl_device_open(struct inode *inode, struct file *file) |
| 894 | { |
| 895 | cycle_kernel_lock(); |
| 896 | file->private_data = NULL; |
| 897 | return 0; |
| 898 | } |
| 899 | |
| 900 | static int ctl_device_close(struct inode *inode, struct file *file) |
| 901 | { |
| 902 | return 0; |
| 903 | } |
| 904 | |
| 905 | static const struct file_operations device_fops = { |
| 906 | .open = device_open, |
| 907 | .release = device_close, |
| 908 | .read = device_read, |
| 909 | .write = device_write, |
| 910 | .poll = device_poll, |
| 911 | .owner = THIS_MODULE, |
| 912 | }; |
| 913 | |
| 914 | static const struct file_operations ctl_device_fops = { |
| 915 | .open = ctl_device_open, |
| 916 | .release = ctl_device_close, |
| 917 | .read = device_read, |
| 918 | .write = device_write, |
| 919 | .owner = THIS_MODULE, |
| 920 | }; |
| 921 | |
| 922 | static struct miscdevice ctl_device = { |
| 923 | .name = "dlm-control", |
| 924 | .fops = &ctl_device_fops, |
| 925 | .minor = MISC_DYNAMIC_MINOR, |
| 926 | }; |
| 927 | |
| 928 | int __init dlm_user_init(void) |
| 929 | { |
| 930 | int error; |
| 931 | |
| 932 | error = misc_register(&ctl_device); |
| 933 | if (error) |
| 934 | log_print("misc_register failed for control device"); |
| 935 | |
| 936 | return error; |
| 937 | } |
| 938 | |
| 939 | void dlm_user_exit(void) |
| 940 | { |
| 941 | misc_deregister(&ctl_device); |
| 942 | } |
| 943 | |