-/*
+/*
Unix SMB/CIFS implementation.
byte range locking code
Updated to handle range splits/merges.
Copyright (C) Andrew Tridgell 1992-2000
Copyright (C) Jeremy Allison 1992-2000
-
+
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
-
+
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
-
+
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
used. This allows us to provide the same semantics as NT */
#include "includes.h"
-#include "librpc/gen_ndr/messaging.h"
+#include "system/filesys.h"
+#include "locking/proto.h"
#include "smbd/globals.h"
-#include "dbwrap.h"
+#include "dbwrap/dbwrap.h"
+#include "dbwrap/dbwrap_open.h"
#include "serverid.h"
+#include "messages.h"
+#include "util_tdb.h"
#undef DBGC_CLASS
#define DBGC_CLASS DBGC_LOCKING
static struct db_context *brlock_db;
+struct byte_range_lock {
+ struct files_struct *fsp;
+ unsigned int num_locks;
+ bool modified;
+ bool have_read_oplocks;
+ struct lock_struct *lock_data;
+ struct db_record *record;
+};
+
/****************************************************************************
Debug info at level 10 for lock struct.
****************************************************************************/
-static void print_lock_struct(unsigned int i, struct lock_struct *pls)
+static void print_lock_struct(unsigned int i, const struct lock_struct *pls)
{
DEBUG(10,("[%u]: smblctx = %llu, tid = %u, pid = %s, ",
i,
(unsigned long long)pls->context.smblctx,
(unsigned int)pls->context.tid,
- procid_str(talloc_tos(), &pls->context.pid) ));
-
- DEBUG(10,("start = %.0f, size = %.0f, fnum = %d, %s %s\n",
+ server_id_str(talloc_tos(), &pls->context.pid) ));
+
+ DEBUG(10,("start = %.0f, size = %.0f, fnum = %llu, %s %s\n",
(double)pls->start,
(double)pls->size,
- pls->fnum,
+ (unsigned long long)pls->fnum,
lock_type_name(pls->lock_type),
lock_flav_name(pls->lock_flav) ));
}
+unsigned int brl_num_locks(const struct byte_range_lock *brl)
+{
+ return brl->num_locks;
+}
+
+struct files_struct *brl_fsp(struct byte_range_lock *brl)
+{
+ return brl->fsp;
+}
+
+bool brl_have_read_oplocks(const struct byte_range_lock *brl)
+{
+ return brl->have_read_oplocks;
+}
+
+void brl_set_have_read_oplocks(struct byte_range_lock *brl,
+ bool have_read_oplocks)
+{
+ DEBUG(10, ("Setting have_read_oplocks to %s\n",
+ have_read_oplocks ? "true" : "false"));
+ SMB_ASSERT(brl->record != NULL); /* otherwise we're readonly */
+ brl->have_read_oplocks = have_read_oplocks;
+ brl->modified = true;
+}
+
/****************************************************************************
See if two locking contexts are equal.
****************************************************************************/
-bool brl_same_context(const struct lock_context *ctx1,
+static bool brl_same_context(const struct lock_context *ctx1,
const struct lock_context *ctx2)
{
- return (procid_equal(&ctx1->pid, &ctx2->pid) &&
+ return (serverid_equal(&ctx1->pid, &ctx2->pid) &&
(ctx1->smblctx == ctx2->smblctx) &&
(ctx1->tid == ctx2->tid));
}
const struct lock_struct *lck2)
{
/* XXX Remove for Win7 compatibility. */
- /* this extra check is not redundent - it copes with locks
+ /* this extra check is not redundant - it copes with locks
that go beyond the end of 64 bit file space */
if (lck1->size != 0 &&
lck1->start == lck2->start &&
See if lock2 can be added when lock1 is in place.
****************************************************************************/
-static bool brl_conflict(const struct lock_struct *lck1,
+static bool brl_conflict(const struct lock_struct *lck1,
const struct lock_struct *lck2)
{
/* Ignore PENDING locks. */
}
return brl_overlap(lck1, lck2);
-}
+}
/****************************************************************************
See if lock2 can be added when lock1 is in place - when both locks are POSIX
know already match.
****************************************************************************/
-static bool brl_conflict_posix(const struct lock_struct *lck1,
+static bool brl_conflict_posix(const struct lock_struct *lck1,
const struct lock_struct *lck2)
{
#if defined(DEVELOPER)
/* One is read, the other write, or the context is different,
do they overlap ? */
return brl_overlap(lck1, lck2);
-}
+}
#if ZERO_ZERO
-static bool brl_conflict1(const struct lock_struct *lck1,
+static bool brl_conflict1(const struct lock_struct *lck1,
const struct lock_struct *lck2)
{
if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
lck2->start >= (lck1->start + lck1->size)) {
return False;
}
-
+
return True;
-}
+}
#endif
/****************************************************************************
if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
return False;
- if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
+ if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
return False;
/* POSIX flavour locks never conflict here - this is only called
}
return brl_overlap(lck1, lck2);
-}
+}
/****************************************************************************
Check if an unlock overlaps a pending lock.
app depends on this ?
****************************************************************************/
-NTSTATUS brl_lock_failed(files_struct *fsp, const struct lock_struct *lock, bool blocking_lock)
+static NTSTATUS brl_lock_failed(files_struct *fsp,
+ const struct lock_struct *lock,
+ bool blocking_lock)
{
if (lock->start >= 0xEF000000 && (lock->start >> 63) == 0) {
/* amazing the little things you learn with a test
return NT_STATUS_FILE_LOCK_CONFLICT;
}
- if (procid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
+ if (serverid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
lock->context.tid == fsp->last_lock_failure.context.tid &&
lock->fnum == fsp->last_lock_failure.fnum &&
lock->start == fsp->last_lock_failure.start) {
brlock_db = db_open(NULL, lock_path("brlock.tdb"),
lp_open_files_db_hash_size(), tdb_flags,
- read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644 );
+ read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644,
+ DBWRAP_LOCK_ORDER_2);
if (!brlock_db) {
DEBUG(0,("Failed to open byte range locking database %s\n",
lock_path("brlock.tdb")));
Compare two locks for sorting.
****************************************************************************/
-static int lock_compare(const struct lock_struct *lck1,
+static int lock_compare(const struct lock_struct *lck1,
const struct lock_struct *lck2)
{
if (lck1->start != lck2->start) {
return brl_lock_failed(fsp,plock,blocking_lock);
}
#if ZERO_ZERO
- if (plock->start == 0 && plock->size == 0 &&
+ if (plock->start == 0 && plock->size == 0 &&
locks[i].size == 0) {
break;
}
}
/* no conflicts - add it to the list of locks */
- locks = (struct lock_struct *)SMB_REALLOC(locks, (br_lck->num_locks + 1) * sizeof(*locks));
+ locks = talloc_realloc(br_lck, locks, struct lock_struct,
+ (br_lck->num_locks + 1));
if (!locks) {
status = NT_STATUS_NO_MEMORY;
goto fail;
existing POSIX lock range into two, and add our lock,
so we need at most 2 more entries. */
- tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 2));
+ tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 2);
if (!tp) {
return NT_STATUS_NO_MEMORY;
}
/* Do any Windows flavour locks conflict ? */
if (brl_conflict(curr_lock, plock)) {
/* No games with error messages. */
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
/* Remember who blocked us. */
plock->context.smblctx = curr_lock->context.smblctx;
return NT_STATUS_FILE_LOCK_CONFLICT;
if (brl_conflict_posix(curr_lock, plock)) {
/* Can't block ourselves with POSIX locks. */
/* No games with error messages. */
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
/* Remember who blocked us. */
plock->context.smblctx = curr_lock->context.smblctx;
return NT_STATUS_FILE_LOCK_CONFLICT;
plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
if (errno_ret == EACCES || errno_ret == EAGAIN) {
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
status = NT_STATUS_FILE_LOCK_CONFLICT;
goto fail;
} else {
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
status = map_nt_error_from_unix(errno);
goto fail;
}
/* If we didn't use all the allocated size,
* Realloc so we don't leak entries per lock call. */
if (count < br_lck->num_locks + 2) {
- tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
+ tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
if (!tp) {
status = NT_STATUS_NO_MEMORY;
goto fail;
}
br_lck->num_locks = count;
- SAFE_FREE(br_lck->lock_data);
+ TALLOC_FREE(br_lck->lock_data);
br_lck->lock_data = tp;
locks = tp;
br_lck->modified = True;
struct blocking_lock_record *blr)
{
VFS_FIND(brl_lock_windows);
- return handle->fns->brl_lock_windows(handle, br_lck, plock,
- blocking_lock, blr);
+ return handle->fns->brl_lock_windows_fn(handle, br_lck, plock,
+ blocking_lock, blr);
}
/****************************************************************************
uint64_t smblctx,
struct server_id pid,
br_off start,
- br_off size,
+ br_off size,
enum brl_type lock_type,
enum brl_flavour lock_flav,
bool blocking_lock,
#ifdef DEVELOPER
/* Quieten valgrind on test. */
- memset(&lock, '\0', sizeof(lock));
+ ZERO_STRUCT(lock);
#endif
lock.context.smblctx = smblctx;
/* Actually delete the lock. */
if (i < br_lck->num_locks - 1) {
- memmove(&locks[i], &locks[i+1],
+ memmove(&locks[i], &locks[i+1],
sizeof(*locks)*((br_lck->num_locks-1) - i));
}
existing POSIX lock range into two, so we need at most
1 more entry. */
- tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 1));
+ tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 1);
if (!tp) {
DEBUG(10,("brl_unlock_posix: malloc fail\n"));
return False;
if (lock->lock_flav == WINDOWS_LOCK) {
/* Do any Windows flavour locks conflict ? */
if (brl_conflict(lock, plock)) {
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
return false;
}
/* Just copy the Windows lock into the new array. */
if (!overlap_found) {
/* Just ignore - no change. */
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
return True;
}
/* Realloc so we don't leak entries per unlock call. */
if (count) {
- tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
+ tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
if (!tp) {
DEBUG(10,("brl_unlock_posix: realloc fail\n"));
return False;
}
} else {
/* We deleted the last lock. */
- SAFE_FREE(tp);
+ TALLOC_FREE(tp);
tp = NULL;
}
LEVEL2_CONTEND_POSIX_BRL);
br_lck->num_locks = count;
- SAFE_FREE(br_lck->lock_data);
+ TALLOC_FREE(br_lck->lock_data);
locks = tp;
br_lck->lock_data = tp;
br_lck->modified = True;
const struct lock_struct *plock)
{
VFS_FIND(brl_unlock_windows);
- return handle->fns->brl_unlock_windows(handle, msg_ctx, br_lck, plock);
+ return handle->fns->brl_unlock_windows_fn(handle, msg_ctx, br_lck,
+ plock);
}
/****************************************************************************
uint64_t smblctx,
struct server_id pid,
br_off start,
- br_off size,
+ br_off size,
enum brl_type lock_type,
enum brl_flavour lock_flav)
{
if(lp_posix_locking(fsp->conn->params) && (lock_flav == WINDOWS_LOCK)) {
ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
- DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
+ DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for %s file %s\n",
(double)start, (double)size, ret ? "locked" : "unlocked",
- fsp->fnum, fsp_str_dbg(fsp)));
+ fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
/* We need to return the inverse of is_posix_locked. */
ret = !ret;
uint64_t *psmblctx,
struct server_id pid,
br_off *pstart,
- br_off *psize,
+ br_off *psize,
enum brl_type *plock_type,
enum brl_flavour lock_flav)
{
if (exlock->lock_flav == WINDOWS_LOCK) {
conflict = brl_conflict(exlock, &lock);
- } else {
+ } else {
conflict = brl_conflict_posix(exlock, &lock);
}
if(lp_posix_locking(fsp->conn->params)) {
bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK);
- DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
+ DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for %s file %s\n",
(double)*pstart, (double)*psize, ret ? "locked" : "unlocked",
- fsp->fnum, fsp_str_dbg(fsp)));
+ fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
if (ret) {
/* Hmmm. No clue what to set smblctx to - use -1. */
struct blocking_lock_record *blr)
{
VFS_FIND(brl_cancel_windows);
- return handle->fns->brl_cancel_windows(handle, br_lck, plock, blr);
+ return handle->fns->brl_cancel_windows_fn(handle, br_lck, plock, blr);
}
/****************************************************************************
if (i < br_lck->num_locks - 1) {
/* Found this particular pending lock - delete it */
- memmove(&locks[i], &locks[i+1],
+ memmove(&locks[i], &locks[i+1],
sizeof(*locks)*((br_lck->num_locks-1) - i));
}
struct byte_range_lock *br_lck)
{
files_struct *fsp = br_lck->fsp;
- uint16 tid = fsp->conn->cnum;
- int fnum = fsp->fnum;
- unsigned int i, j, dcount=0;
- int num_deleted_windows_locks = 0;
+ uint32_t tid = fsp->conn->cnum;
+ uint64_t fnum = fsp->fnum;
+ unsigned int i;
struct lock_struct *locks = br_lck->lock_data;
- struct server_id pid = sconn_server_id(fsp->conn->sconn);
- bool unlock_individually = False;
- bool posix_level2_contention_ended = false;
+ struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
+ struct lock_struct *locks_copy;
+ unsigned int num_locks_copy;
+
+ /* Copy the current lock array. */
+ if (br_lck->num_locks) {
+ locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
+ if (!locks_copy) {
+ smb_panic("brl_close_fnum: talloc failed");
+ }
+ } else {
+ locks_copy = NULL;
+ }
- if(lp_posix_locking(fsp->conn->params)) {
+ num_locks_copy = br_lck->num_locks;
- /* Check if there are any Windows locks associated with this dev/ino
- pair that are not this fnum. If so we need to call unlock on each
- one in order to release the system POSIX locks correctly. */
+ for (i=0; i < num_locks_copy; i++) {
+ struct lock_struct *lock = &locks_copy[i];
- for (i=0; i < br_lck->num_locks; i++) {
- struct lock_struct *lock = &locks[i];
+ if (lock->context.tid == tid && serverid_equal(&lock->context.pid, &pid) &&
+ (lock->fnum == fnum)) {
+ brl_unlock(msg_ctx,
+ br_lck,
+ lock->context.smblctx,
+ pid,
+ lock->start,
+ lock->size,
+ lock->lock_flav);
+ }
+ }
+}
- if (!procid_equal(&lock->context.pid, &pid)) {
- continue;
- }
+bool brl_mark_disconnected(struct files_struct *fsp)
+{
+ uint32_t tid = fsp->conn->cnum;
+ uint64_t smblctx = fsp->op->global->open_persistent_id;
+ uint64_t fnum = fsp->fnum;
+ unsigned int i;
+ struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
+ struct byte_range_lock *br_lck = NULL;
- if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) {
- continue; /* Ignore pending. */
- }
+ if (!fsp->op->global->durable) {
+ return false;
+ }
- if (lock->context.tid != tid || lock->fnum != fnum) {
- unlock_individually = True;
- break;
- }
- }
+ if (fsp->current_lock_count == 0) {
+ return true;
+ }
- if (unlock_individually) {
- struct lock_struct *locks_copy;
- unsigned int num_locks_copy;
-
- /* Copy the current lock array. */
- if (br_lck->num_locks) {
- locks_copy = (struct lock_struct *)TALLOC_MEMDUP(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
- if (!locks_copy) {
- smb_panic("brl_close_fnum: talloc failed");
- }
- } else {
- locks_copy = NULL;
- }
+ br_lck = brl_get_locks(talloc_tos(), fsp);
+ if (br_lck == NULL) {
+ return false;
+ }
+
+ for (i=0; i < br_lck->num_locks; i++) {
+ struct lock_struct *lock = &br_lck->lock_data[i];
- num_locks_copy = br_lck->num_locks;
+ /*
+ * as this is a durable handle, we only expect locks
+ * of the current file handle!
+ */
- for (i=0; i < num_locks_copy; i++) {
- struct lock_struct *lock = &locks_copy[i];
+ if (lock->context.smblctx != smblctx) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
- if (lock->context.tid == tid && procid_equal(&lock->context.pid, &pid) &&
- (lock->fnum == fnum)) {
- brl_unlock(msg_ctx,
- br_lck,
- lock->context.smblctx,
- pid,
- lock->start,
- lock->size,
- lock->lock_flav);
- }
- }
- return;
+ if (lock->context.tid != tid) {
+ TALLOC_FREE(br_lck);
+ return false;
}
+
+ if (!serverid_equal(&lock->context.pid, &self)) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
+
+ if (lock->fnum != fnum) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
+
+ server_id_set_disconnected(&lock->context.pid);
+ lock->context.tid = TID_FIELD_INVALID;
+ lock->fnum = FNUM_FIELD_INVALID;
}
- /* We can bulk delete - any POSIX locks will be removed when the fd closes. */
+ br_lck->modified = true;
+ TALLOC_FREE(br_lck);
+ return true;
+}
- /* Remove any existing locks for this fnum (or any fnum if they're POSIX). */
+bool brl_reconnect_disconnected(struct files_struct *fsp)
+{
+ uint32_t tid = fsp->conn->cnum;
+ uint64_t smblctx = fsp->op->global->open_persistent_id;
+ uint64_t fnum = fsp->fnum;
+ unsigned int i;
+ struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
+ struct byte_range_lock *br_lck = NULL;
+
+ if (!fsp->op->global->durable) {
+ return false;
+ }
+
+ /*
+ * When reconnecting, we do not want to validate the brlock entries
+ * and thereby remove our own (disconnected) entries but reactivate
+ * them instead.
+ */
+ fsp->lockdb_clean = true;
+
+ br_lck = brl_get_locks(talloc_tos(), fsp);
+ if (br_lck == NULL) {
+ return false;
+ }
+
+ if (br_lck->num_locks == 0) {
+ TALLOC_FREE(br_lck);
+ return true;
+ }
for (i=0; i < br_lck->num_locks; i++) {
- struct lock_struct *lock = &locks[i];
- bool del_this_lock = False;
-
- if (lock->context.tid == tid && procid_equal(&lock->context.pid, &pid)) {
- if ((lock->lock_flav == WINDOWS_LOCK) && (lock->fnum == fnum)) {
- del_this_lock = True;
- num_deleted_windows_locks++;
- contend_level2_oplocks_end(br_lck->fsp,
- LEVEL2_CONTEND_WINDOWS_BRL);
- } else if (lock->lock_flav == POSIX_LOCK) {
- del_this_lock = True;
-
- /* Only end level2 contention once for posix */
- if (!posix_level2_contention_ended) {
- posix_level2_contention_ended = true;
- contend_level2_oplocks_end(br_lck->fsp,
- LEVEL2_CONTEND_POSIX_BRL);
- }
- }
- }
+ struct lock_struct *lock = &br_lck->lock_data[i];
- if (del_this_lock) {
- /* Send unlock messages to any pending waiters that overlap. */
- for (j=0; j < br_lck->num_locks; j++) {
- struct lock_struct *pend_lock = &locks[j];
+ /*
+ * as this is a durable handle we only expect locks
+ * of the current file handle!
+ */
- /* Ignore our own or non-pending locks. */
- if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
- continue;
- }
+ if (lock->context.smblctx != smblctx) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
- /* Optimisation - don't send to this fnum as we're
- closing it. */
- if (pend_lock->context.tid == tid &&
- procid_equal(&pend_lock->context.pid, &pid) &&
- pend_lock->fnum == fnum) {
- continue;
- }
+ if (lock->context.tid != TID_FIELD_INVALID) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
- /* We could send specific lock info here... */
- if (brl_pending_overlap(lock, pend_lock)) {
- messaging_send(msg_ctx, pend_lock->context.pid,
- MSG_SMB_UNLOCK, &data_blob_null);
- }
- }
+ if (!server_id_is_disconnected(&lock->context.pid)) {
+ TALLOC_FREE(br_lck);
+ return false;
+ }
- /* found it - delete it */
- if (br_lck->num_locks > 1 && i < br_lck->num_locks - 1) {
- memmove(&locks[i], &locks[i+1],
- sizeof(*locks)*((br_lck->num_locks-1) - i));
- }
- br_lck->num_locks--;
- br_lck->modified = True;
- i--;
- dcount++;
+ if (lock->fnum != FNUM_FIELD_INVALID) {
+ TALLOC_FREE(br_lck);
+ return false;
}
- }
- if(lp_posix_locking(fsp->conn->params) && num_deleted_windows_locks) {
- /* Reduce the Windows lock POSIX reference count on this dev/ino pair. */
- reduce_windows_lock_ref_count(fsp, num_deleted_windows_locks);
+ lock->context.pid = self;
+ lock->context.tid = tid;
+ lock->fnum = fnum;
}
+
+ fsp->current_lock_count = br_lck->num_locks;
+ br_lck->modified = true;
+ TALLOC_FREE(br_lck);
+ return true;
}
/****************************************************************************
Ensure this set of lock entries is valid.
****************************************************************************/
-static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct **pplocks)
+static bool validate_lock_entries(TALLOC_CTX *mem_ctx,
+ unsigned int *pnum_entries, struct lock_struct **pplocks,
+ bool keep_disconnected)
{
unsigned int i;
unsigned int num_valid_entries = 0;
struct lock_struct *locks = *pplocks;
+ TALLOC_CTX *frame = talloc_stackframe();
+ struct server_id *ids;
+ bool *exists;
+
+ ids = talloc_array(frame, struct server_id, *pnum_entries);
+ if (ids == NULL) {
+ DEBUG(0, ("validate_lock_entries: "
+ "talloc_array(struct server_id, %u) failed\n",
+ *pnum_entries));
+ talloc_free(frame);
+ return false;
+ }
+
+ exists = talloc_array(frame, bool, *pnum_entries);
+ if (exists == NULL) {
+ DEBUG(0, ("validate_lock_entries: "
+ "talloc_array(bool, %u) failed\n",
+ *pnum_entries));
+ talloc_free(frame);
+ return false;
+ }
for (i = 0; i < *pnum_entries; i++) {
- struct lock_struct *lock_data = &locks[i];
- if (!serverid_exists(&lock_data->context.pid)) {
- /* This process no longer exists - mark this
- entry as invalid by zeroing it. */
- ZERO_STRUCTP(lock_data);
- } else {
+ ids[i] = locks[i].context.pid;
+ }
+
+ if (!serverids_exist(ids, *pnum_entries, exists)) {
+ DEBUG(3, ("validate_lock_entries: serverids_exists failed\n"));
+ talloc_free(frame);
+ return false;
+ }
+
+ for (i = 0; i < *pnum_entries; i++) {
+ if (exists[i]) {
num_valid_entries++;
+ continue;
}
+
+ if (keep_disconnected &&
+ server_id_is_disconnected(&ids[i]))
+ {
+ num_valid_entries++;
+ continue;
+ }
+
+ /* This process no longer exists - mark this
+ entry as invalid by zeroing it. */
+ ZERO_STRUCTP(&locks[i]);
}
+ TALLOC_FREE(frame);
if (num_valid_entries != *pnum_entries) {
struct lock_struct *new_lock_data = NULL;
if (num_valid_entries) {
- new_lock_data = SMB_MALLOC_ARRAY(struct lock_struct, num_valid_entries);
+ new_lock_data = talloc_array(
+ mem_ctx, struct lock_struct,
+ num_valid_entries);
if (!new_lock_data) {
DEBUG(3, ("malloc fail\n"));
return False;
}
}
- SAFE_FREE(*pplocks);
+ TALLOC_FREE(*pplocks);
*pplocks = new_lock_data;
*pnum_entries = num_valid_entries;
}
on each lock.
****************************************************************************/
-static int traverse_fn(struct db_record *rec, void *state)
+static int brl_traverse_fn(struct db_record *rec, void *state)
{
struct brl_forall_cb *cb = (struct brl_forall_cb *)state;
struct lock_struct *locks;
unsigned int i;
unsigned int num_locks = 0;
unsigned int orig_num_locks = 0;
+ TDB_DATA dbkey;
+ TDB_DATA value;
+
+ dbkey = dbwrap_record_get_key(rec);
+ value = dbwrap_record_get_value(rec);
/* In a traverse function we must make a copy of
dbuf before modifying it. */
- locks = (struct lock_struct *)memdup(rec->value.dptr,
- rec->value.dsize);
+ locks = (struct lock_struct *)talloc_memdup(
+ talloc_tos(), value.dptr, value.dsize);
if (!locks) {
return -1; /* Terminate traversal. */
}
- key = (struct file_id *)rec->key.dptr;
- orig_num_locks = num_locks = rec->value.dsize/sizeof(*locks);
+ key = (struct file_id *)dbkey.dptr;
+ orig_num_locks = num_locks = value.dsize/sizeof(*locks);
/* Ensure the lock db is clean of entries from invalid processes. */
- if (!validate_lock_entries(&num_locks, &locks)) {
- SAFE_FREE(locks);
+ if (!validate_lock_entries(talloc_tos(), &num_locks, &locks, true)) {
+ TALLOC_FREE(locks);
return -1; /* Terminate traversal */
}
TDB_DATA data;
data.dptr = (uint8_t *)locks;
data.dsize = num_locks*sizeof(struct lock_struct);
- rec->store(rec, data, TDB_REPLACE);
+ dbwrap_record_store(rec, data, TDB_REPLACE);
} else {
- rec->delete_rec(rec);
+ dbwrap_record_delete(rec);
}
}
}
}
- SAFE_FREE(locks);
+ TALLOC_FREE(locks);
return 0;
}
void *private_data)
{
struct brl_forall_cb cb;
+ NTSTATUS status;
+ int count = 0;
if (!brlock_db) {
return 0;
}
cb.fn = fn;
cb.private_data = private_data;
- return brlock_db->traverse(brlock_db, traverse_fn, &cb);
+ status = dbwrap_traverse(brlock_db, brl_traverse_fn, &cb, &count);
+
+ if (!NT_STATUS_IS_OK(status)) {
+ return -1;
+ } else {
+ return count;
+ }
}
/*******************************************************************
Unlock the record.
********************************************************************/
-static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
+static void byte_range_lock_flush(struct byte_range_lock *br_lck)
{
- if (br_lck->read_only) {
- SMB_ASSERT(!br_lck->modified);
- }
-
+ size_t data_len;
if (!br_lck->modified) {
+ DEBUG(10, ("br_lck not modified\n"));
goto done;
}
- if (br_lck->num_locks == 0) {
+ data_len = br_lck->num_locks * sizeof(struct lock_struct);
+
+ if (br_lck->have_read_oplocks) {
+ data_len += 1;
+ }
+
+ DEBUG(10, ("data_len=%d\n", (int)data_len));
+
+ if (data_len == 0) {
/* No locks - delete this entry. */
- NTSTATUS status = br_lck->record->delete_rec(br_lck->record);
+ NTSTATUS status = dbwrap_record_delete(br_lck->record);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(0, ("delete_rec returned %s\n",
nt_errstr(status)));
TDB_DATA data;
NTSTATUS status;
- data.dptr = (uint8 *)br_lck->lock_data;
- data.dsize = br_lck->num_locks * sizeof(struct lock_struct);
+ data.dsize = data_len;
+ data.dptr = talloc_array(talloc_tos(), uint8_t, data_len);
+ SMB_ASSERT(data.dptr != NULL);
+
+ memcpy(data.dptr, br_lck->lock_data,
+ br_lck->num_locks * sizeof(struct lock_struct));
+
+ if (br_lck->have_read_oplocks) {
+ data.dptr[data_len-1] = 1;
+ }
- status = br_lck->record->store(br_lck->record, data,
- TDB_REPLACE);
+ status = dbwrap_record_store(br_lck->record, data, TDB_REPLACE);
+ TALLOC_FREE(data.dptr);
if (!NT_STATUS_IS_OK(status)) {
DEBUG(0, ("store returned %s\n", nt_errstr(status)));
smb_panic("Could not store byte range mode entry");
}
}
- done:
+ DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db)));
- SAFE_FREE(br_lck->lock_data);
+ done:
+ br_lck->modified = false;
TALLOC_FREE(br_lck->record);
+}
+
+static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
+{
+ byte_range_lock_flush(br_lck);
return 0;
}
TALLOC_FREE(brl) will release the lock in the destructor.
********************************************************************/
-static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
- files_struct *fsp, bool read_only)
+struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, files_struct *fsp)
{
TDB_DATA key, data;
- struct byte_range_lock *br_lck = TALLOC_P(mem_ctx, struct byte_range_lock);
+ struct byte_range_lock *br_lck = talloc(mem_ctx, struct byte_range_lock);
if (br_lck == NULL) {
return NULL;
br_lck->fsp = fsp;
br_lck->num_locks = 0;
+ br_lck->have_read_oplocks = false;
br_lck->modified = False;
- br_lck->key = fsp->file_id;
- key.dptr = (uint8 *)&br_lck->key;
+ key.dptr = (uint8 *)&fsp->file_id;
key.dsize = sizeof(struct file_id);
- if (!fsp->lockdb_clean) {
- /* We must be read/write to clean
- the dead entries. */
- read_only = False;
- }
+ br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key);
- if (read_only) {
- if (brlock_db->fetch(brlock_db, br_lck, key, &data) == -1) {
- DEBUG(3, ("Could not fetch byte range lock record\n"));
- TALLOC_FREE(br_lck);
- return NULL;
- }
- br_lck->record = NULL;
+ if (br_lck->record == NULL) {
+ DEBUG(3, ("Could not lock byte range lock entry\n"));
+ TALLOC_FREE(br_lck);
+ return NULL;
}
- else {
- br_lck->record = brlock_db->fetch_locked(brlock_db, br_lck, key);
- if (br_lck->record == NULL) {
- DEBUG(3, ("Could not lock byte range lock entry\n"));
- TALLOC_FREE(br_lck);
- return NULL;
- }
-
- data = br_lck->record->value;
- }
+ data = dbwrap_record_get_value(br_lck->record);
- br_lck->read_only = read_only;
br_lck->lock_data = NULL;
talloc_set_destructor(br_lck, byte_range_lock_destructor);
br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
if (br_lck->num_locks != 0) {
- br_lck->lock_data = SMB_MALLOC_ARRAY(struct lock_struct,
- br_lck->num_locks);
+ br_lck->lock_data = talloc_array(
+ br_lck, struct lock_struct, br_lck->num_locks);
if (br_lck->lock_data == NULL) {
DEBUG(0, ("malloc failed\n"));
TALLOC_FREE(br_lck);
return NULL;
}
- memcpy(br_lck->lock_data, data.dptr, data.dsize);
+ memcpy(br_lck->lock_data, data.dptr,
+ talloc_get_size(br_lck->lock_data));
}
-
+
+ DEBUG(10, ("data.dsize=%d\n", (int)data.dsize));
+
+ if ((data.dsize % sizeof(struct lock_struct)) == 1) {
+ br_lck->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
+ }
+
if (!fsp->lockdb_clean) {
int orig_num_locks = br_lck->num_locks;
- /* This is the first time we've accessed this. */
- /* Go through and ensure all entries exist - remove any that don't. */
- /* Makes the lockdb self cleaning at low cost. */
+ /*
+ * This is the first time we access the byte range lock
+ * record with this fsp. Go through and ensure all entries
+ * are valid - remove any that don't.
+ * This makes the lockdb self cleaning at low cost.
+ *
+ * Note: Disconnected entries belong to disconnected
+ * durable handles. So at this point, we have a new
+ * handle on the file and the disconnected durable has
+ * already been closed (we are not a durable reconnect).
+ * So we need to clean the disconnected brl entry.
+ */
- if (!validate_lock_entries(&br_lck->num_locks,
- &br_lck->lock_data)) {
- SAFE_FREE(br_lck->lock_data);
+ if (!validate_lock_entries(br_lck, &br_lck->num_locks,
+ &br_lck->lock_data, false)) {
TALLOC_FREE(br_lck);
return NULL;
}
print_lock_struct(i, &locks[i]);
}
}
+
return br_lck;
}
-struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx,
- files_struct *fsp)
+struct brl_get_locks_readonly_state {
+ TALLOC_CTX *mem_ctx;
+ struct byte_range_lock **br_lock;
+};
+
+static void brl_get_locks_readonly_parser(TDB_DATA key, TDB_DATA data,
+ void *private_data)
{
- return brl_get_locks_internal(mem_ctx, fsp, False);
+ struct brl_get_locks_readonly_state *state =
+ (struct brl_get_locks_readonly_state *)private_data;
+ struct byte_range_lock *br_lock;
+
+ br_lock = talloc_pooled_object(
+ state->mem_ctx, struct byte_range_lock, 1, data.dsize);
+ if (br_lock == NULL) {
+ *state->br_lock = NULL;
+ return;
+ }
+ br_lock->lock_data = (struct lock_struct *)talloc_memdup(
+ br_lock, data.dptr, data.dsize);
+ br_lock->num_locks = data.dsize / sizeof(struct lock_struct);
+
+ if ((data.dsize % sizeof(struct lock_struct)) == 1) {
+ br_lock->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
+ }
+
+ DEBUG(10, ("Got %d bytes, have_read_oplocks: %s\n", (int)data.dsize,
+ br_lock->have_read_oplocks ? "true" : "false"));
+
+ *state->br_lock = br_lock;
}
struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
{
- struct byte_range_lock *br_lock;
+ struct byte_range_lock *br_lock = NULL;
+ struct byte_range_lock *rw = NULL;
- if (lp_clustering()) {
- return brl_get_locks_internal(talloc_tos(), fsp, true);
- }
+ DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
+ dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum));
if ((fsp->brlock_rec != NULL)
- && (brlock_db->get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
+ && (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
+ /*
+ * We have cached the brlock_rec and the database did not
+ * change.
+ */
return fsp->brlock_rec;
}
- TALLOC_FREE(fsp->brlock_rec);
-
- br_lock = brl_get_locks_internal(talloc_tos(), fsp, false);
- if (br_lock == NULL) {
- return NULL;
+ if (!fsp->lockdb_clean) {
+ /*
+ * Fetch the record in R/W mode to give validate_lock_entries
+ * a chance to kick in once.
+ */
+ rw = brl_get_locks(talloc_tos(), fsp);
+ if (rw == NULL) {
+ return NULL;
+ }
+ fsp->lockdb_clean = true;
}
- fsp->brlock_seqnum = brlock_db->get_seqnum(brlock_db);
- fsp->brlock_rec = talloc_zero(fsp, struct byte_range_lock);
- if (fsp->brlock_rec == NULL) {
- goto fail;
+ if (rw != NULL) {
+ size_t lock_data_size;
+
+ /*
+ * Make a copy of the already retrieved and sanitized rw record
+ */
+ lock_data_size = rw->num_locks * sizeof(struct lock_struct);
+ br_lock = talloc_pooled_object(
+ fsp, struct byte_range_lock, 1, lock_data_size);
+ if (br_lock == NULL) {
+ goto fail;
+ }
+ br_lock->have_read_oplocks = rw->have_read_oplocks;
+ br_lock->num_locks = rw->num_locks;
+ br_lock->lock_data = (struct lock_struct *)talloc_memdup(
+ br_lock, rw->lock_data, lock_data_size);
+ } else {
+ struct brl_get_locks_readonly_state state;
+ NTSTATUS status;
+
+ /*
+ * Parse the record fresh from the database
+ */
+
+ state.mem_ctx = fsp;
+ state.br_lock = &br_lock;
+
+ status = dbwrap_parse_record(
+ brlock_db,
+ make_tdb_data((uint8_t *)&fsp->file_id,
+ sizeof(fsp->file_id)),
+ brl_get_locks_readonly_parser, &state);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3, ("Could not parse byte range lock record: "
+ "%s\n", nt_errstr(status)));
+ goto fail;
+ }
+ if (br_lock == NULL) {
+ goto fail;
+ }
}
- fsp->brlock_rec->fsp = fsp;
- fsp->brlock_rec->num_locks = br_lock->num_locks;
- fsp->brlock_rec->read_only = true;
- fsp->brlock_rec->key = br_lock->key;
- fsp->brlock_rec->lock_data = (struct lock_struct *)
- talloc_memdup(fsp->brlock_rec, br_lock->lock_data,
- sizeof(struct lock_struct) * br_lock->num_locks);
- if (fsp->brlock_rec->lock_data == NULL) {
- goto fail;
+ br_lock->fsp = fsp;
+ br_lock->modified = false;
+ br_lock->record = NULL;
+
+ if (lp_clustering()) {
+ /*
+ * In the cluster case we can't cache the brlock struct
+ * because dbwrap_get_seqnum does not work reliably over
+ * ctdb. Thus we have to throw away the brlock struct soon.
+ */
+ talloc_steal(talloc_tos(), br_lock);
+ } else {
+ /*
+ * Cache the brlock struct, invalidated when the dbwrap_seqnum
+ * changes. See beginning of this routine.
+ */
+ TALLOC_FREE(fsp->brlock_rec);
+ fsp->brlock_rec = br_lock;
+ fsp->brlock_seqnum = dbwrap_get_seqnum(brlock_db);
}
- TALLOC_FREE(br_lock);
- return fsp->brlock_rec;
fail:
- TALLOC_FREE(br_lock);
- TALLOC_FREE(fsp->brlock_rec);
- return NULL;
+ TALLOC_FREE(rw);
+ return br_lock;
}
struct brl_revalidate_state {
static int compare_procids(const void *p1, const void *p2)
{
- const struct server_id *i1 = (struct server_id *)p1;
- const struct server_id *i2 = (struct server_id *)p2;
+ const struct server_id *i1 = (const struct server_id *)p1;
+ const struct server_id *i2 = (const struct server_id *)p2;
if (i1->pid < i2->pid) return -1;
if (i2->pid > i2->pid) return 1;
* array, then qsort that array and only send to non-dupes.
*/
-static void brl_revalidate(struct messaging_context *msg_ctx,
- void *private_data,
- uint32_t msg_type,
- struct server_id server_id,
- DATA_BLOB *data)
+void brl_revalidate(struct messaging_context *msg_ctx,
+ void *private_data,
+ uint32_t msg_type,
+ struct server_id server_id,
+ DATA_BLOB *data)
{
struct brl_revalidate_state *state;
uint32 i;
struct server_id last_pid;
- if (!(state = TALLOC_ZERO_P(NULL, struct brl_revalidate_state))) {
+ if (!(state = talloc_zero(NULL, struct brl_revalidate_state))) {
DEBUG(0, ("talloc failed\n"));
return;
}
ZERO_STRUCT(last_pid);
for (i=0; i<state->num_pids; i++) {
- if (procid_equal(&last_pid, &state->pids[i])) {
+ if (serverid_equal(&last_pid, &state->pids[i])) {
/*
* We've seen that one already
*/
return;
}
-void brl_register_msgs(struct messaging_context *msg_ctx)
+bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id)
{
- messaging_register(msg_ctx, NULL, MSG_SMB_BRL_VALIDATE,
- brl_revalidate);
+ bool ret = false;
+ TALLOC_CTX *frame = talloc_stackframe();
+ TDB_DATA key, val;
+ struct db_record *rec;
+ struct lock_struct *lock;
+ unsigned n, num;
+ NTSTATUS status;
+
+ key = make_tdb_data((void*)&fid, sizeof(fid));
+
+ rec = dbwrap_fetch_locked(brlock_db, frame, key);
+ if (rec == NULL) {
+ DEBUG(5, ("brl_cleanup_disconnected: failed to fetch record "
+ "for file %s\n", file_id_string(frame, &fid)));
+ goto done;
+ }
+
+ val = dbwrap_record_get_value(rec);
+ lock = (struct lock_struct*)val.dptr;
+ num = val.dsize / sizeof(struct lock_struct);
+ if (lock == NULL) {
+ DEBUG(10, ("brl_cleanup_disconnected: no byte range locks for "
+ "file %s\n", file_id_string(frame, &fid)));
+ ret = true;
+ goto done;
+ }
+
+ for (n=0; n<num; n++) {
+ struct lock_context *ctx = &lock[n].context;
+
+ if (!server_id_is_disconnected(&ctx->pid)) {
+ DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
+ "%s used by server %s, do not cleanup\n",
+ file_id_string(frame, &fid),
+ server_id_str(frame, &ctx->pid)));
+ goto done;
+ }
+
+ if (ctx->smblctx != open_persistent_id) {
+ DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
+ "%s expected smblctx %llu but found %llu"
+ ", do not cleanup\n",
+ file_id_string(frame, &fid),
+ (unsigned long long)open_persistent_id,
+ (unsigned long long)ctx->smblctx));
+ goto done;
+ }
+ }
+
+ status = dbwrap_record_delete(rec);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(5, ("brl_cleanup_disconnected: failed to delete record "
+ "for file %s from %s, open %llu: %s\n",
+ file_id_string(frame, &fid), dbwrap_name(brlock_db),
+ (unsigned long long)open_persistent_id,
+ nt_errstr(status)));
+ goto done;
+ }
+
+ DEBUG(10, ("brl_cleanup_disconnected: "
+ "file %s cleaned up %u entries from open %llu\n",
+ file_id_string(frame, &fid), num,
+ (unsigned long long)open_persistent_id));
+
+ ret = true;
+done:
+ talloc_free(frame);
+ return ret;
}