ctdb-tests: Add explicit wait to the fork_helper()
[Samba.git] / source3 / smbd / oplock.c
blobbe20deee54cb5c546bc070847b1c9bdc5142ad30
1 /*
2 Unix SMB/CIFS implementation.
3 oplock processing
4 Copyright (C) Andrew Tridgell 1992-1998
5 Copyright (C) Jeremy Allison 1998 - 2001
6 Copyright (C) Volker Lendecke 2005
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 #define DBGC_CLASS DBGC_LOCKING
23 #include "includes.h"
24 #include "smbd/smbd.h"
25 #include "smbd/globals.h"
26 #include "messages.h"
27 #include "../librpc/gen_ndr/open_files.h"
30 * helper function used by the kernel oplock backends to post the break message
32 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
34 uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
36 /* Put the kernel break info into the message. */
37 push_file_id_24((char *)msg, &fsp->file_id);
38 SIVAL(msg,24,fsp->fh->gen_id);
40 /* Don't need to be root here as we're only ever
41 sending to ourselves. */
43 messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
44 MSG_SMB_KERNEL_BREAK,
45 msg, MSG_SMB_KERNEL_BREAK_SIZE);
48 /****************************************************************************
49 Attempt to set an oplock on a file. Succeeds if kernel oplocks are
50 disabled (just sets flags).
51 ****************************************************************************/
53 NTSTATUS set_file_oplock(files_struct *fsp)
55 struct smbd_server_connection *sconn = fsp->conn->sconn;
56 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
57 bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
59 if (fsp->oplock_type == LEVEL_II_OPLOCK) {
60 if (use_kernel &&
61 !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
62 DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
63 "don't support them\n"));
64 return NT_STATUS_NOT_SUPPORTED;
68 if ((fsp->oplock_type != NO_OPLOCK) &&
69 use_kernel &&
70 !koplocks->ops->set_oplock(koplocks, fsp, fsp->oplock_type))
72 return map_nt_error_from_unix(errno);
75 fsp->sent_oplock_break = NO_BREAK_SENT;
76 if (fsp->oplock_type == LEVEL_II_OPLOCK) {
77 sconn->oplocks.level_II_open++;
78 } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
79 sconn->oplocks.exclusive_open++;
82 DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
83 "tv_sec = %x, tv_usec = %x\n",
84 fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
85 fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
86 (int)fsp->open_time.tv_usec ));
88 return NT_STATUS_OK;
91 /****************************************************************************
92 Attempt to release an oplock on a file. Decrements oplock count.
93 ****************************************************************************/
95 static void release_file_oplock(files_struct *fsp)
97 struct smbd_server_connection *sconn = fsp->conn->sconn;
98 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
99 bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
101 if ((fsp->oplock_type != NO_OPLOCK) &&
102 use_kernel) {
103 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
106 if (fsp->oplock_type == LEVEL_II_OPLOCK) {
107 sconn->oplocks.level_II_open--;
108 } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
109 sconn->oplocks.exclusive_open--;
112 SMB_ASSERT(sconn->oplocks.exclusive_open>=0);
113 SMB_ASSERT(sconn->oplocks.level_II_open>=0);
115 fsp->oplock_type = NO_OPLOCK;
116 fsp->sent_oplock_break = NO_BREAK_SENT;
118 flush_write_cache(fsp, SAMBA_OPLOCK_RELEASE_FLUSH);
119 delete_write_cache(fsp);
121 TALLOC_FREE(fsp->oplock_timeout);
124 /****************************************************************************
125 Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
126 ****************************************************************************/
128 static void downgrade_file_oplock(files_struct *fsp)
130 struct smbd_server_connection *sconn = fsp->conn->sconn;
131 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
133 if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
134 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
135 return;
138 if (koplocks) {
139 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
141 fsp->oplock_type = LEVEL_II_OPLOCK;
142 sconn->oplocks.exclusive_open--;
143 sconn->oplocks.level_II_open++;
144 fsp->sent_oplock_break = NO_BREAK_SENT;
146 flush_write_cache(fsp, SAMBA_OPLOCK_RELEASE_FLUSH);
147 delete_write_cache(fsp);
149 TALLOC_FREE(fsp->oplock_timeout);
152 uint32_t map_oplock_to_lease_type(uint16_t op_type)
154 uint32_t ret;
156 switch(op_type) {
157 case BATCH_OPLOCK:
158 case BATCH_OPLOCK|EXCLUSIVE_OPLOCK:
159 ret = SMB2_LEASE_READ|SMB2_LEASE_WRITE|SMB2_LEASE_HANDLE;
160 break;
161 case EXCLUSIVE_OPLOCK:
162 ret = SMB2_LEASE_READ|SMB2_LEASE_WRITE;
163 break;
164 case LEVEL_II_OPLOCK:
165 ret = SMB2_LEASE_READ;
166 break;
167 default:
168 ret = SMB2_LEASE_NONE;
169 break;
171 return ret;
174 uint32_t get_lease_type(struct share_mode_data *d, struct share_mode_entry *e)
176 if (e->op_type == LEASE_OPLOCK) {
177 return d->leases[e->lease_idx].current_state;
179 return map_oplock_to_lease_type(e->op_type);
182 bool update_num_read_oplocks(files_struct *fsp, struct share_mode_lock *lck)
184 struct share_mode_data *d = lck->data;
185 struct byte_range_lock *br_lck;
186 uint32_t num_read_oplocks = 0;
187 uint32_t i;
189 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
191 * If we're the only one, we don't need a brlock entry
193 remove_stale_share_mode_entries(d);
194 SMB_ASSERT(d->num_share_modes == 1);
195 SMB_ASSERT(EXCLUSIVE_OPLOCK_TYPE(d->share_modes[0].op_type));
196 return true;
199 for (i=0; i<d->num_share_modes; i++) {
200 struct share_mode_entry *e = &d->share_modes[i];
201 uint32_t e_lease_type = get_lease_type(d, e);
203 if (e_lease_type & SMB2_LEASE_READ) {
204 num_read_oplocks += 1;
208 br_lck = brl_get_locks_readonly(fsp);
209 if (br_lck == NULL) {
210 return false;
212 if (brl_num_read_oplocks(br_lck) == num_read_oplocks) {
213 return true;
216 br_lck = brl_get_locks(talloc_tos(), fsp);
217 if (br_lck == NULL) {
218 return false;
220 brl_set_num_read_oplocks(br_lck, num_read_oplocks);
221 TALLOC_FREE(br_lck);
222 return true;
225 /****************************************************************************
226 Remove a file oplock. Copes with level II and exclusive.
227 Locks then unlocks the share mode lock. Client can decide to go directly
228 to none even if a "break-to-level II" was sent.
229 ****************************************************************************/
231 bool remove_oplock(files_struct *fsp)
233 bool ret;
234 struct share_mode_lock *lck;
236 DEBUG(10, ("remove_oplock called for %s\n",
237 fsp_str_dbg(fsp)));
239 /* Remove the oplock flag from the sharemode. */
240 lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
241 if (lck == NULL) {
242 DEBUG(0,("remove_oplock: failed to lock share entry for "
243 "file %s\n", fsp_str_dbg(fsp)));
244 return False;
247 ret = remove_share_oplock(lck, fsp);
248 if (!ret) {
249 DEBUG(0,("remove_oplock: failed to remove share oplock for "
250 "file %s, %s, %s\n",
251 fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
252 file_id_string_tos(&fsp->file_id)));
254 release_file_oplock(fsp);
256 ret = update_num_read_oplocks(fsp, lck);
257 if (!ret) {
258 DEBUG(0, ("%s: update_num_read_oplocks failed for "
259 "file %s, %s, %s\n",
260 __func__, fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
261 file_id_string_tos(&fsp->file_id)));
264 TALLOC_FREE(lck);
265 return ret;
269 * Deal with a reply when a break-to-level II was sent.
271 bool downgrade_oplock(files_struct *fsp)
273 bool ret;
274 struct share_mode_lock *lck;
276 DEBUG(10, ("downgrade_oplock called for %s\n",
277 fsp_str_dbg(fsp)));
279 lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
280 if (lck == NULL) {
281 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
282 "file %s\n", fsp_str_dbg(fsp)));
283 return False;
285 ret = downgrade_share_oplock(lck, fsp);
286 if (!ret) {
287 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
288 "for file %s, %s, file_id %s\n",
289 fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
290 file_id_string_tos(&fsp->file_id)));
292 downgrade_file_oplock(fsp);
294 ret = update_num_read_oplocks(fsp, lck);
295 if (!ret) {
296 DEBUG(0, ("%s: update_num_read_oplocks failed for "
297 "file %s, %s, %s\n",
298 __func__, fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
299 file_id_string_tos(&fsp->file_id)));
302 TALLOC_FREE(lck);
303 return ret;
306 static void lease_timeout_handler(struct tevent_context *ctx,
307 struct tevent_timer *te,
308 struct timeval now,
309 void *private_data)
311 struct fsp_lease *lease =
312 talloc_get_type_abort(private_data,
313 struct fsp_lease);
314 struct files_struct *fsp;
315 struct share_mode_lock *lck;
316 uint16_t old_epoch = lease->lease.lease_epoch;
318 fsp = file_find_one_fsp_from_lease_key(lease->sconn,
319 &lease->lease.lease_key);
320 if (fsp == NULL) {
321 /* race? */
322 TALLOC_FREE(lease->timeout);
323 return;
326 lck = get_existing_share_mode_lock(
327 talloc_tos(), fsp->file_id);
328 if (lck == NULL) {
329 /* race? */
330 TALLOC_FREE(lease->timeout);
331 return;
334 fsp_lease_update(lck, fsp_client_guid(fsp), lease);
336 if (lease->lease.lease_epoch != old_epoch) {
338 * If the epoch changed we need to wait for
339 * the next timeout to happen.
341 DEBUG(10, ("lease break timeout race (epoch) for file %s - ignoring\n",
342 fsp_str_dbg(fsp)));
343 TALLOC_FREE(lck);
344 return;
347 if (!(lease->lease.lease_flags & SMB2_LEASE_FLAG_BREAK_IN_PROGRESS)) {
349 * If the epoch changed we need to wait for
350 * the next timeout to happen.
352 DEBUG(10, ("lease break timeout race (flags) for file %s - ignoring\n",
353 fsp_str_dbg(fsp)));
354 TALLOC_FREE(lck);
355 return;
358 DEBUG(1, ("lease break timed out for file %s -- replying anyway\n",
359 fsp_str_dbg(fsp)));
360 (void)downgrade_lease(lease->sconn->client->connections,
362 &fsp->file_id,
363 &lease->lease.lease_key,
364 SMB2_LEASE_NONE);
366 TALLOC_FREE(lck);
369 bool fsp_lease_update(struct share_mode_lock *lck,
370 const struct GUID *client_guid,
371 struct fsp_lease *lease)
373 struct share_mode_data *d = lck->data;
374 int idx;
375 struct share_mode_lease *l = NULL;
377 idx = find_share_mode_lease(d, client_guid, &lease->lease.lease_key);
378 if (idx != -1) {
379 l = &d->leases[idx];
382 if (l == NULL) {
383 DEBUG(1, ("%s: Could not find lease entry\n", __func__));
384 TALLOC_FREE(lease->timeout);
385 lease->lease.lease_state = SMB2_LEASE_NONE;
386 lease->lease.lease_epoch += 1;
387 lease->lease.lease_flags = 0;
388 return false;
391 DEBUG(10,("%s: refresh lease state\n", __func__));
393 /* Ensure we're in sync with current lease state. */
394 if (lease->lease.lease_epoch != l->epoch) {
395 DEBUG(10,("%s: cancel outdated timeout\n", __func__));
396 TALLOC_FREE(lease->timeout);
398 lease->lease.lease_epoch = l->epoch;
399 lease->lease.lease_state = l->current_state;
401 if (l->breaking) {
402 lease->lease.lease_flags |= SMB2_LEASE_FLAG_BREAK_IN_PROGRESS;
404 if (lease->timeout == NULL) {
405 struct timeval t = timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0);
407 DEBUG(10,("%s: setup timeout handler\n", __func__));
409 lease->timeout = tevent_add_timer(lease->sconn->ev_ctx,
410 lease, t,
411 lease_timeout_handler,
412 lease);
413 if (lease->timeout == NULL) {
414 DEBUG(0, ("%s: Could not add lease timeout handler\n",
415 __func__));
418 } else {
419 lease->lease.lease_flags &= ~SMB2_LEASE_FLAG_BREAK_IN_PROGRESS;
420 TALLOC_FREE(lease->timeout);
423 return true;
426 struct downgrade_lease_additional_state {
427 struct tevent_immediate *im;
428 struct smbXsrv_connection *xconn;
429 uint32_t break_flags;
430 struct smb2_lease_key lease_key;
431 uint32_t break_from;
432 uint32_t break_to;
433 uint16_t new_epoch;
436 static void downgrade_lease_additional_trigger(struct tevent_context *ev,
437 struct tevent_immediate *im,
438 void *private_data)
440 struct downgrade_lease_additional_state *state =
441 talloc_get_type_abort(private_data,
442 struct downgrade_lease_additional_state);
443 struct smbXsrv_connection *xconn = state->xconn;
444 NTSTATUS status;
446 status = smbd_smb2_send_lease_break(xconn,
447 state->new_epoch,
448 state->break_flags,
449 &state->lease_key,
450 state->break_from,
451 state->break_to);
452 TALLOC_FREE(state);
453 if (!NT_STATUS_IS_OK(status)) {
454 smbd_server_connection_terminate(xconn,
455 nt_errstr(status));
456 return;
460 struct downgrade_lease_fsps_state {
461 struct file_id id;
462 struct share_mode_lock *lck;
463 const struct smb2_lease_key *key;
466 static struct files_struct *downgrade_lease_fsps(struct files_struct *fsp,
467 void *private_data)
469 struct downgrade_lease_fsps_state *state =
470 (struct downgrade_lease_fsps_state *)private_data;
472 if (fsp->oplock_type != LEASE_OPLOCK) {
473 return NULL;
475 if (!smb2_lease_key_equal(&fsp->lease->lease.lease_key, state->key)) {
476 return NULL;
478 if (!file_id_equal(&fsp->file_id, &state->id)) {
479 return NULL;
482 fsp_lease_update(state->lck, fsp_client_guid(fsp), fsp->lease);
484 return NULL;
487 NTSTATUS downgrade_lease(struct smbXsrv_connection *xconn,
488 uint32_t num_file_ids,
489 const struct file_id *ids,
490 const struct smb2_lease_key *key,
491 uint32_t lease_state)
493 struct smbd_server_connection *sconn = xconn->client->sconn;
494 struct share_mode_lock *lck;
495 struct share_mode_lease *l = NULL;
496 const struct file_id id = ids[0];
497 uint32_t i;
498 NTSTATUS status;
500 DEBUG(10, ("%s: Downgrading %s to %x\n", __func__,
501 file_id_string_tos(&id), (unsigned)lease_state));
503 lck = get_existing_share_mode_lock(talloc_tos(), id);
504 if (lck == NULL) {
505 return NT_STATUS_OBJECT_NAME_NOT_FOUND;
507 status = downgrade_share_lease(sconn, lck, key, lease_state, &l);
509 DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
510 file_id_string_tos(&id), (unsigned)lease_state, nt_errstr(status)));
512 if (NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_BREAK_IN_PROGRESS)) {
513 struct downgrade_lease_additional_state *state;
515 state = talloc_zero(xconn,
516 struct downgrade_lease_additional_state);
517 if (state == NULL) {
518 TALLOC_FREE(lck);
519 return NT_STATUS_NO_MEMORY;
522 state->im = tevent_create_immediate(state);
523 if (state->im == NULL) {
524 TALLOC_FREE(state);
525 TALLOC_FREE(lck);
526 return NT_STATUS_NO_MEMORY;
529 state->xconn = xconn;
530 if (l->current_state & (~SMB2_LEASE_READ)) {
531 state->break_flags = SMB2_NOTIFY_BREAK_LEASE_FLAG_ACK_REQUIRED;
533 state->lease_key = l->lease_key;
534 state->break_from = l->current_state;
535 state->break_to = l->breaking_to_requested;
536 if (l->lease_version > 1) {
537 state->new_epoch = l->epoch;
540 if (state->break_flags == 0) {
542 * This is an async break without
543 * SMB2_NOTIFY_BREAK_LEASE_FLAG_ACK_REQUIRED
545 * we need to store NONE state in the
546 * database.
548 l->current_state = 0;
549 l->breaking_to_requested = 0;
550 l->breaking_to_required = 0;
551 l->breaking = false;
553 lck->data->modified = true;
556 tevent_schedule_immediate(state->im, xconn->ev_ctx,
557 downgrade_lease_additional_trigger,
558 state);
562 struct downgrade_lease_fsps_state state = {
563 .id = id, .lck = lck, .key = key,
566 files_forall(sconn, downgrade_lease_fsps, &state);
569 TALLOC_FREE(lck);
570 DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
571 file_id_string_tos(&id), (unsigned)lease_state, nt_errstr(status)));
574 * Dynamic share case. Ensure other opens are copies.
575 * This will only be breaking to NONE.
578 for (i = 1; i < num_file_ids; i++) {
579 lck = get_existing_share_mode_lock(talloc_tos(), ids[i]);
580 if (lck == NULL) {
581 return NT_STATUS_OBJECT_NAME_NOT_FOUND;
585 struct downgrade_lease_fsps_state state = {
586 .id = ids[i], .lck = lck, .key = key,
589 files_forall(sconn, downgrade_lease_fsps, &state);
592 DEBUG(10, ("%s: Downgrading %s to %x => %s\n", __func__,
593 file_id_string_tos(&ids[i]), (unsigned)lease_state, nt_errstr(status)));
595 TALLOC_FREE(lck);
598 return status;
601 /****************************************************************************
602 Set up an oplock break message.
603 ****************************************************************************/
605 #define SMB1_BREAK_MESSAGE_LENGTH (smb_size + 8*2)
607 static void new_break_message_smb1(files_struct *fsp, int cmd,
608 char result[SMB1_BREAK_MESSAGE_LENGTH])
610 memset(result,'\0',smb_size);
611 srv_set_message(result,8,0,true);
612 SCVAL(result,smb_com,SMBlockingX);
613 SSVAL(result,smb_tid,fsp->conn->cnum);
614 SSVAL(result,smb_pid,0xFFFF);
615 SSVAL(result,smb_uid,0);
616 SSVAL(result,smb_mid,0xFFFF);
617 SCVAL(result,smb_vwv0,0xFF);
618 SSVAL(result,smb_vwv2,fsp->fnum);
619 SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
620 SCVAL(result,smb_vwv3+1,cmd);
623 /****************************************************************************
624 Function to do the waiting before sending a local break.
625 ****************************************************************************/
627 static void wait_before_sending_break(void)
629 long wait_time = (long)lp_oplock_break_wait_time();
631 if (wait_time) {
632 smb_msleep(wait_time);
636 /****************************************************************************
637 Ensure that we have a valid oplock.
638 ****************************************************************************/
640 static files_struct *initial_break_processing(
641 struct smbd_server_connection *sconn, struct file_id id,
642 unsigned long file_id)
644 files_struct *fsp = NULL;
646 DEBUG(3, ("initial_break_processing: called for %s/%u\n"
647 "Current oplocks_open (exclusive = %d, levelII = %d)\n",
648 file_id_string_tos(&id), (int)file_id,
649 sconn->oplocks.exclusive_open,
650 sconn->oplocks.level_II_open));
653 * We need to search the file open table for the
654 * entry containing this dev and inode, and ensure
655 * we have an oplock on it.
658 fsp = file_find_dif(sconn, id, file_id);
660 if(fsp == NULL) {
661 /* The file could have been closed in the meantime - return success. */
662 DEBUG(3, ("initial_break_processing: cannot find open file "
663 "with file_id %s gen_id = %lu, allowing break to "
664 "succeed.\n", file_id_string_tos(&id), file_id));
665 return NULL;
668 /* Ensure we have an oplock on the file */
671 * There is a potential race condition in that an oplock could
672 * have been broken due to another udp request, and yet there are
673 * still oplock break messages being sent in the udp message
674 * queue for this file. So return true if we don't have an oplock,
675 * as we may have just freed it.
678 if(fsp->oplock_type == NO_OPLOCK) {
679 DEBUG(3, ("initial_break_processing: file %s (file_id = %s "
680 "gen_id = %lu) has no oplock. Allowing break to "
681 "succeed regardless.\n", fsp_str_dbg(fsp),
682 file_id_string_tos(&id), fsp->fh->gen_id));
683 return NULL;
686 return fsp;
689 static void oplock_timeout_handler(struct tevent_context *ctx,
690 struct tevent_timer *te,
691 struct timeval now,
692 void *private_data)
694 files_struct *fsp = (files_struct *)private_data;
696 SMB_ASSERT(fsp->sent_oplock_break != NO_BREAK_SENT);
698 /* Remove the timed event handler. */
699 TALLOC_FREE(fsp->oplock_timeout);
700 DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
701 fsp_str_dbg(fsp)));
702 remove_oplock(fsp);
705 /*******************************************************************
706 Add a timeout handler waiting for the client reply.
707 *******************************************************************/
709 static void add_oplock_timeout_handler(files_struct *fsp)
711 struct smbd_server_connection *sconn = fsp->conn->sconn;
712 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
715 * If kernel oplocks already notifies smbds when an oplock break times
716 * out, just return.
718 if (koplocks &&
719 (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
720 return;
723 if (fsp->oplock_timeout != NULL) {
724 DEBUG(0, ("Logic problem -- have an oplock event hanging "
725 "around\n"));
728 fsp->oplock_timeout =
729 tevent_add_timer(fsp->conn->sconn->ev_ctx, fsp,
730 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
731 oplock_timeout_handler, fsp);
733 if (fsp->oplock_timeout == NULL) {
734 DEBUG(0, ("Could not add oplock timeout handler\n"));
738 static void send_break_message_smb1(files_struct *fsp, int level)
740 struct smbXsrv_connection *xconn = NULL;
741 char break_msg[SMB1_BREAK_MESSAGE_LENGTH];
744 * For SMB1 we only have one connection
746 xconn = fsp->conn->sconn->client->connections;
748 new_break_message_smb1(fsp, level, break_msg);
750 show_msg(break_msg);
751 if (!srv_send_smb(xconn,
752 break_msg, false, 0,
753 IS_CONN_ENCRYPTED(fsp->conn),
754 NULL)) {
755 exit_server_cleanly("send_break_message_smb1: "
756 "srv_send_smb failed.");
760 /*******************************************************************
761 This handles the generic oplock break message from another smbd.
762 *******************************************************************/
764 static void process_oplock_break_message(struct messaging_context *msg_ctx,
765 void *private_data,
766 uint32_t msg_type,
767 struct server_id src,
768 DATA_BLOB *data)
770 struct share_mode_entry msg;
771 files_struct *fsp;
772 bool use_kernel;
773 struct smbd_server_connection *sconn =
774 talloc_get_type_abort(private_data,
775 struct smbd_server_connection);
776 struct server_id self = messaging_server_id(sconn->msg_ctx);
777 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
778 uint16_t break_from;
779 uint16_t break_to;
780 bool break_needed = true;
781 struct server_id_buf tmp;
783 if (data->data == NULL) {
784 DEBUG(0, ("Got NULL buffer\n"));
785 return;
788 if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
789 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
790 return;
793 /* De-linearize incoming message. */
794 message_to_share_mode_entry(&msg, (char *)data->data);
795 break_to = msg.op_type;
797 DEBUG(10, ("Got oplock break to %u message from pid %s: %s/%llu\n",
798 (unsigned)break_to, server_id_str_buf(src, &tmp),
799 file_id_string_tos(&msg.id),
800 (unsigned long long)msg.share_file_id));
802 fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
804 if (fsp == NULL) {
805 /* We hit a race here. Break messages are sent, and before we
806 * get to process this message, we have closed the file. */
807 DEBUG(3, ("Did not find fsp\n"));
808 return;
811 break_from = fsp_lease_type(fsp);
813 if (fsp->oplock_type != LEASE_OPLOCK) {
814 if (fsp->sent_oplock_break != NO_BREAK_SENT) {
816 * Nothing to do anymore
818 DEBUG(10, ("fsp->sent_oplock_break = %d\n",
819 fsp->sent_oplock_break));
820 return;
824 if (!(global_client_caps & CAP_LEVEL_II_OPLOCKS)) {
825 DEBUG(10, ("client_caps without level2 oplocks\n"));
826 break_to &= ~SMB2_LEASE_READ;
829 use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
830 if (use_kernel && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
831 DEBUG(10, ("Kernel oplocks don't allow level2\n"));
832 break_to &= ~SMB2_LEASE_READ;
835 if (!lp_level2_oplocks(SNUM(fsp->conn))) {
836 DEBUG(10, ("no level2 oplocks by config\n"));
837 break_to &= ~SMB2_LEASE_READ;
840 if (fsp->oplock_type == LEASE_OPLOCK) {
841 struct share_mode_lock *lck;
842 int idx;
844 lck = get_existing_share_mode_lock(
845 talloc_tos(), fsp->file_id);
846 if (lck == NULL) {
848 * We hit a race here. Break messages are sent, and
849 * before we get to process this message, we have closed
850 * the file.
852 DEBUG(3, ("Did not find share_mode\n"));
853 return;
856 idx = find_share_mode_lease(
857 lck->data,
858 fsp_client_guid(fsp),
859 &fsp->lease->lease.lease_key);
860 if (idx != -1) {
861 struct share_mode_lease *l;
862 l = &lck->data->leases[idx];
864 break_from = l->current_state;
865 break_to &= l->current_state;
867 if (l->breaking) {
868 break_to &= l->breaking_to_required;
869 if (l->breaking_to_required != break_to) {
871 * Note we don't increment the epoch
872 * here, which might be a bug in
873 * Windows too...
875 l->breaking_to_required = break_to;
876 lck->data->modified = true;
878 break_needed = false;
879 } else if (l->current_state == break_to) {
880 break_needed = false;
881 } else if (l->current_state == SMB2_LEASE_READ) {
882 l->current_state = SMB2_LEASE_NONE;
883 /* Need to increment the epoch */
884 l->epoch += 1;
885 lck->data->modified = true;
886 } else {
887 l->breaking = true;
888 l->breaking_to_required = break_to;
889 l->breaking_to_requested = break_to;
890 /* Need to increment the epoch */
891 l->epoch += 1;
892 lck->data->modified = true;
895 /* Ensure we're in sync with current lease state. */
896 fsp_lease_update(lck, fsp_client_guid(fsp), fsp->lease);
899 TALLOC_FREE(lck);
902 if (!break_needed) {
903 DEBUG(10,("%s: skip break\n", __func__));
904 return;
907 if ((break_from == SMB2_LEASE_NONE) && !break_needed) {
908 DEBUG(3, ("Already downgraded oplock to none on %s: %s\n",
909 file_id_string_tos(&fsp->file_id),
910 fsp_str_dbg(fsp)));
911 return;
914 DEBUG(10, ("break_from=%u, break_to=%u\n",
915 (unsigned)break_from, (unsigned)break_to));
917 if ((break_from == break_to) && !break_needed) {
918 DEBUG(3, ("Already downgraded oplock to %u on %s: %s\n",
919 (unsigned)break_to,
920 file_id_string_tos(&fsp->file_id),
921 fsp_str_dbg(fsp)));
922 return;
925 /* Need to wait before sending a break
926 message if we sent ourselves this message. */
927 if (serverid_equal(&self, &src)) {
928 wait_before_sending_break();
931 if (sconn->using_smb2) {
932 send_break_message_smb2(fsp, break_from, break_to);
933 } else {
934 send_break_message_smb1(fsp, (break_to & SMB2_LEASE_READ) ?
935 OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
938 if ((break_from == SMB2_LEASE_READ) &&
939 (break_to == SMB2_LEASE_NONE)) {
941 * This is an async break without a reply and thus no timeout
943 * leases are handled above.
945 if (fsp->oplock_type != LEASE_OPLOCK) {
946 remove_oplock(fsp);
948 return;
950 if (fsp->oplock_type == LEASE_OPLOCK) {
951 return;
954 fsp->sent_oplock_break = (break_to & SMB2_LEASE_READ) ?
955 LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
957 add_oplock_timeout_handler(fsp);
960 /*******************************************************************
961 This handles the kernel oplock break message.
962 *******************************************************************/
964 static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
965 void *private_data,
966 uint32_t msg_type,
967 struct server_id src,
968 DATA_BLOB *data)
970 struct file_id id;
971 unsigned long file_id;
972 files_struct *fsp;
973 struct smbd_server_connection *sconn =
974 talloc_get_type_abort(private_data,
975 struct smbd_server_connection);
976 struct server_id_buf tmp;
978 if (data->data == NULL) {
979 DEBUG(0, ("Got NULL buffer\n"));
980 return;
983 if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
984 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
985 return;
988 /* Pull the data from the message. */
989 pull_file_id_24((char *)data->data, &id);
990 file_id = (unsigned long)IVAL(data->data, 24);
992 DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
993 server_id_str_buf(src, &tmp), file_id_string_tos(&id),
994 (unsigned int)file_id));
996 fsp = initial_break_processing(sconn, id, file_id);
998 if (fsp == NULL) {
999 DEBUG(3, ("Got a kernel oplock break message for a file "
1000 "I don't know about\n"));
1001 return;
1004 if (fsp->sent_oplock_break != NO_BREAK_SENT) {
1005 /* This is ok, kernel oplocks come in completely async */
1006 DEBUG(3, ("Got a kernel oplock request while waiting for a "
1007 "break reply\n"));
1008 return;
1011 if (sconn->using_smb2) {
1012 send_break_message_smb2(fsp, 0, OPLOCKLEVEL_NONE);
1013 } else {
1014 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
1017 fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
1019 add_oplock_timeout_handler(fsp);
1022 struct break_to_none_state {
1023 struct smbd_server_connection *sconn;
1024 struct file_id id;
1025 struct smb2_lease_key lease_key;
1026 struct GUID client_guid;
1028 static void do_break_to_none(struct tevent_context *ctx,
1029 struct tevent_immediate *im,
1030 void *private_data);
1032 /****************************************************************************
1033 This function is called on any file modification or lock request. If a file
1034 is level 2 oplocked then it must tell all other level 2 holders to break to
1035 none.
1036 ****************************************************************************/
1038 static void contend_level2_oplocks_begin_default(files_struct *fsp,
1039 enum level2_contention_type type)
1041 struct smbd_server_connection *sconn = fsp->conn->sconn;
1042 struct tevent_immediate *im;
1043 struct break_to_none_state *state;
1044 struct byte_range_lock *brl;
1045 uint32_t num_read_oplocks = 0;
1048 * If this file is level II oplocked then we need
1049 * to grab the shared memory lock and inform all
1050 * other files with a level II lock that they need
1051 * to flush their read caches. We keep the lock over
1052 * the shared memory area whilst doing this.
1055 if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
1057 * There can't be any level2 oplocks, we're alone.
1059 return;
1062 brl = brl_get_locks_readonly(fsp);
1063 if (brl != NULL) {
1064 num_read_oplocks = brl_num_read_oplocks(brl);
1067 DEBUG(10, ("num_read_oplocks = %"PRIu32"\n", num_read_oplocks));
1069 if (num_read_oplocks == 0) {
1070 DEBUG(10, ("No read oplocks around\n"));
1071 return;
1075 * When we get here we might have a brlock entry locked. Also
1076 * locking the share mode entry would violate the locking
1077 * order. Breaking level2 oplocks to none is asynchronous
1078 * anyway, so we postpone this into an immediate event.
1081 state = talloc_zero(sconn, struct break_to_none_state);
1082 if (state == NULL) {
1083 DEBUG(1, ("talloc failed\n"));
1084 return;
1086 state->sconn = sconn;
1087 state->id = fsp->file_id;
1089 if (fsp->oplock_type == LEASE_OPLOCK) {
1090 state->client_guid = *fsp_client_guid(fsp);
1091 state->lease_key = fsp->lease->lease.lease_key;
1092 DEBUG(10, ("Breaking through lease key %"PRIu64"/%"PRIu64"\n",
1093 state->lease_key.data[0],
1094 state->lease_key.data[1]));
1097 im = tevent_create_immediate(state);
1098 if (im == NULL) {
1099 DEBUG(1, ("tevent_create_immediate failed\n"));
1100 TALLOC_FREE(state);
1101 return;
1103 tevent_schedule_immediate(im, sconn->ev_ctx, do_break_to_none, state);
1106 static void send_break_to_none(struct messaging_context *msg_ctx,
1107 const struct share_mode_entry *e)
1109 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
1111 share_mode_entry_to_message(msg, e);
1112 /* Overload entry->op_type */
1113 SSVAL(msg, OP_BREAK_MSG_OP_TYPE_OFFSET, NO_OPLOCK);
1115 messaging_send_buf(msg_ctx, e->pid, MSG_SMB_BREAK_REQUEST,
1116 (uint8_t *)msg, sizeof(msg));
1119 static void do_break_to_none(struct tevent_context *ctx,
1120 struct tevent_immediate *im,
1121 void *private_data)
1123 struct break_to_none_state *state = talloc_get_type_abort(
1124 private_data, struct break_to_none_state);
1125 uint32_t i;
1126 struct share_mode_lock *lck;
1127 struct share_mode_data *d;
1129 lck = get_existing_share_mode_lock(talloc_tos(), state->id);
1130 if (lck == NULL) {
1131 DEBUG(1, ("%s: failed to lock share mode entry for file %s.\n",
1132 __func__, file_id_string_tos(&state->id)));
1133 goto done;
1135 d = lck->data;
1138 * Walk leases and oplocks separately: We have to send one break per
1139 * lease. If we have multiple share_mode_entry having a common lease,
1140 * we would break the lease twice if we don't walk the leases list
1141 * separately.
1144 for (i=0; i<d->num_leases; i++) {
1145 struct share_mode_lease *l = &d->leases[i];
1146 struct share_mode_entry *e = NULL;
1147 uint32_t j;
1149 if ((l->current_state & SMB2_LEASE_READ) == 0) {
1150 continue;
1152 if (smb2_lease_equal(&state->client_guid,
1153 &state->lease_key,
1154 &l->client_guid,
1155 &l->lease_key)) {
1156 DEBUG(10, ("Don't break our own lease\n"));
1157 continue;
1160 for (j=0; j<d->num_share_modes; j++) {
1161 e = &d->share_modes[j];
1163 if (!is_valid_share_mode_entry(e)) {
1164 continue;
1166 if (e->lease_idx == i) {
1167 break;
1170 if (j == d->num_share_modes) {
1171 DEBUG(0, ("leases[%"PRIu32"] has no share mode\n",
1172 i));
1173 continue;
1176 DEBUG(10, ("Breaking lease# %"PRIu32" with share_entry# "
1177 "%"PRIu32"\n", i, j));
1179 send_break_to_none(state->sconn->msg_ctx, e);
1182 for(i = 0; i < d->num_share_modes; i++) {
1183 struct share_mode_entry *e = &d->share_modes[i];
1185 if (!is_valid_share_mode_entry(e)) {
1186 continue;
1188 if (e->op_type == LEASE_OPLOCK) {
1190 * Took care of those in the loop above
1192 continue;
1196 * As there could have been multiple writes waiting at the
1197 * lock_share_entry gate we may not be the first to
1198 * enter. Hence the state of the op_types in the share mode
1199 * entries may be partly NO_OPLOCK and partly LEVEL_II
1200 * oplock. It will do no harm to re-send break messages to
1201 * those smbd's that are still waiting their turn to remove
1202 * their LEVEL_II state, and also no harm to ignore existing
1203 * NO_OPLOCK states. JRA.
1206 DEBUG(10, ("%s: share_entry[%i]->op_type == %d\n", __func__,
1207 i, e->op_type ));
1209 if (e->op_type == NO_OPLOCK) {
1210 continue;
1213 /* Paranoia .... */
1214 if (EXCLUSIVE_OPLOCK_TYPE(e->op_type)) {
1215 DEBUG(0,("%s: PANIC. "
1216 "share mode entry %d is an exclusive "
1217 "oplock !\n", __func__, i ));
1218 TALLOC_FREE(lck);
1219 abort();
1222 send_break_to_none(state->sconn->msg_ctx, e);
1225 /* We let the message receivers handle removing the oplock state
1226 in the share mode lock db. */
1228 TALLOC_FREE(lck);
1229 done:
1230 TALLOC_FREE(state);
1231 return;
1234 void smbd_contend_level2_oplocks_begin(files_struct *fsp,
1235 enum level2_contention_type type)
1237 struct smbd_server_connection *sconn = fsp->conn->sconn;
1238 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
1240 if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
1241 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
1242 return;
1245 contend_level2_oplocks_begin_default(fsp, type);
1248 void smbd_contend_level2_oplocks_end(files_struct *fsp,
1249 enum level2_contention_type type)
1251 struct smbd_server_connection *sconn = fsp->conn->sconn;
1252 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
1254 /* Only kernel oplocks implement this so far */
1255 if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
1256 koplocks->ops->contend_level2_oplocks_end(fsp, type);
1260 /****************************************************************************
1261 Linearize a share mode entry struct to an internal oplock break message.
1262 ****************************************************************************/
1264 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
1266 SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32_t)e->pid.pid);
1267 SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
1268 SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
1269 SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
1270 SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
1271 SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
1272 SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
1273 SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
1274 push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
1275 SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
1276 SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
1277 SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
1278 SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
1279 SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
1282 /****************************************************************************
1283 De-linearize an internal oplock break message to a share mode entry struct.
1284 ****************************************************************************/
1286 void message_to_share_mode_entry(struct share_mode_entry *e, const char *msg)
1288 e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
1289 e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
1290 e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
1291 e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
1292 e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
1293 e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
1294 e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
1295 e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
1296 pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
1297 e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
1298 e->uid = (uint32_t)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
1299 e->flags = (uint16_t)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
1300 e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
1301 e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
1304 /****************************************************************************
1305 Setup oplocks for this process.
1306 ****************************************************************************/
1308 bool init_oplocks(struct smbd_server_connection *sconn)
1310 DEBUG(3,("init_oplocks: initializing messages.\n"));
1312 messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_REQUEST,
1313 process_oplock_break_message);
1314 messaging_register(sconn->msg_ctx, sconn, MSG_SMB_KERNEL_BREAK,
1315 process_kernel_oplock_break);
1316 return true;
1319 void init_kernel_oplocks(struct smbd_server_connection *sconn)
1321 struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
1323 /* only initialize once */
1324 if (koplocks == NULL) {
1325 #if HAVE_KERNEL_OPLOCKS_IRIX
1326 koplocks = irix_init_kernel_oplocks(sconn);
1327 #elif HAVE_KERNEL_OPLOCKS_LINUX
1328 koplocks = linux_init_kernel_oplocks(sconn);
1329 #endif
1330 sconn->oplocks.kernel_ops = koplocks;