#include "includes.h"
#include "system/filesys.h"
+#include "lib/util/server_id.h"
#include "locking/proto.h"
#include "smbd/globals.h"
#include "dbwrap/dbwrap.h"
struct files_struct *fsp;
unsigned int num_locks;
bool modified;
+ uint32_t num_read_oplocks;
struct lock_struct *lock_data;
struct db_record *record;
};
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,
- 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,
- (unsigned long long)pls->fnum,
- lock_type_name(pls->lock_type),
- lock_flav_name(pls->lock_flav) ));
+ struct server_id_buf tmp;
+
+ DBG_DEBUG("[%u]: smblctx = %"PRIu64", tid = %"PRIu32", pid = %s, "
+ "start = %"PRIu64", size = %"PRIu64", fnum = %"PRIu64", "
+ "%s %s\n",
+ i,
+ pls->context.smblctx,
+ pls->context.tid,
+ server_id_str_buf(pls->context.pid, &tmp),
+ pls->start,
+ pls->size,
+ 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->fsp;
}
+uint32_t brl_num_read_oplocks(const struct byte_range_lock *brl)
+{
+ return brl->num_read_oplocks;
+}
+
+void brl_set_num_read_oplocks(struct byte_range_lock *brl,
+ uint32_t num_read_oplocks)
+{
+ DEBUG(10, ("Setting num_read_oplocks to %"PRIu32"\n",
+ num_read_oplocks));
+ SMB_ASSERT(brl->record != NULL); /* otherwise we're readonly */
+ brl->num_read_oplocks = num_read_oplocks;
+ brl->modified = true;
+}
+
/****************************************************************************
See if two locking contexts are equal.
****************************************************************************/
static bool brl_conflict(const struct lock_struct *lck1,
const struct lock_struct *lck2)
{
- /* Ignore PENDING locks. */
- if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
- return False;
-
/* Read locks never conflict. */
if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
return False;
SMB_ASSERT(lck2->lock_flav == POSIX_LOCK);
#endif
- /* Ignore PENDING locks. */
- if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
- return False;
-
/* Read locks never conflict. */
if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
return False;
}
- /* Locks on the same context con't conflict. Ignore fnum. */
+ /* Locks on the same context don't conflict. Ignore fnum. */
if (brl_same_context(&lck1->context, &lck2->context)) {
return False;
}
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))
- return False;
-
if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
return False;
}
This is never used in the POSIX lock case.
****************************************************************************/
-static bool brl_conflict_other(const struct lock_struct *lck1, const struct lock_struct *lck2)
+static bool brl_conflict_other(const struct lock_struct *lock,
+ const struct lock_struct *rw_probe)
{
- if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
+ if (lock->lock_type == READ_LOCK && rw_probe->lock_type == READ_LOCK) {
return False;
+ }
- if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
- return False;
-
- /* POSIX flavour locks never conflict here - this is only called
- in the read/write path. */
-
- if (lck1->lock_flav == POSIX_LOCK && lck2->lock_flav == POSIX_LOCK)
+ if (lock->lock_flav == POSIX_LOCK &&
+ rw_probe->lock_flav == POSIX_LOCK) {
+ /*
+ * POSIX flavour locks never conflict here - this is only called
+ * in the read/write path.
+ */
return False;
-
- /*
- * Incoming WRITE locks conflict with existing READ locks even
- * if the context is the same. JRA. See LOCKTEST7 in smbtorture.
- */
-
- if (!(lck2->lock_type == WRITE_LOCK && lck1->lock_type == READ_LOCK)) {
- if (brl_same_context(&lck1->context, &lck2->context) &&
- lck1->fnum == lck2->fnum)
- return False;
}
- return brl_overlap(lck1, lck2);
-}
-
-/****************************************************************************
- Check if an unlock overlaps a pending lock.
-****************************************************************************/
-
-static bool brl_pending_overlap(const struct lock_struct *lock, const struct lock_struct *pend_lock)
-{
- if ((lock->start <= pend_lock->start) && (lock->start + lock->size > pend_lock->start))
- return True;
- if ((lock->start >= pend_lock->start) && (lock->start <= pend_lock->start + pend_lock->size))
- return True;
- return False;
-}
-
-/****************************************************************************
- Amazingly enough, w2k3 "remembers" whether the last lock failure on a fnum
- is the same as this one and changes its error code. I wonder if any
- app depends on this ?
-****************************************************************************/
+ if (!brl_overlap(lock, rw_probe)) {
+ /*
+ * I/O can only conflict when overlapping a lock, thus let it
+ * pass
+ */
+ return false;
+ }
-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
- suite. Locks beyond this offset (as a 64 bit
- number!) always generate the conflict error code,
- unless the top bit is set */
- if (!blocking_lock) {
- fsp->last_lock_failure = *lock;
- }
- return NT_STATUS_FILE_LOCK_CONFLICT;
+ if (!brl_same_context(&lock->context, &rw_probe->context)) {
+ /*
+ * Different process, conflict
+ */
+ return true;
}
- 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) {
- return NT_STATUS_FILE_LOCK_CONFLICT;
+ if (lock->fnum != rw_probe->fnum) {
+ /*
+ * Different file handle, conflict
+ */
+ return true;
}
- if (!blocking_lock) {
- fsp->last_lock_failure = *lock;
+ if ((lock->lock_type == READ_LOCK) &&
+ (rw_probe->lock_type == WRITE_LOCK)) {
+ /*
+ * Incoming WRITE locks conflict with existing READ locks even
+ * if the context is the same. JRA. See LOCKTEST7 in
+ * smbtorture.
+ */
+ return true;
}
- return NT_STATUS_LOCK_NOT_GRANTED;
+
+ /*
+ * I/O request compatible with existing lock, let it pass without
+ * conflict
+ */
+
+ return false;
}
/****************************************************************************
void brl_init(bool read_only)
{
int tdb_flags;
+ char *db_path;
if (brlock_db) {
return;
}
- tdb_flags = TDB_DEFAULT|TDB_VOLATILE|TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH;
+ tdb_flags =
+ TDB_DEFAULT|
+ TDB_VOLATILE|
+ TDB_CLEAR_IF_FIRST|
+ TDB_INCOMPATIBLE_HASH|
+ TDB_SEQNUM;
- if (!lp_clustering()) {
- /*
- * We can't use the SEQNUM trick to cache brlock
- * entries in the clustering case because ctdb seqnum
- * propagation has a delay.
- */
- tdb_flags |= TDB_SEQNUM;
+ db_path = lock_path(talloc_tos(), "brlock.tdb");
+ if (db_path == NULL) {
+ DEBUG(0, ("out of memory!\n"));
+ return;
}
- brlock_db = db_open(NULL, lock_path("brlock.tdb"),
- lp_open_files_db_hash_size(), tdb_flags,
+ brlock_db = db_open(NULL, db_path,
+ SMB_OPEN_DATABASE_TDB_HASH_SIZE, tdb_flags,
read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644,
- DBWRAP_LOCK_ORDER_2);
+ DBWRAP_LOCK_ORDER_2, DBWRAP_FLAG_NONE);
if (!brlock_db) {
DEBUG(0,("Failed to open byte range locking database %s\n",
- lock_path("brlock.tdb")));
+ db_path));
+ TALLOC_FREE(db_path);
return;
}
+ TALLOC_FREE(db_path);
}
/****************************************************************************
****************************************************************************/
NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
- struct lock_struct *plock, bool blocking_lock)
+ struct lock_struct *plock)
{
unsigned int i;
files_struct *fsp = br_lck->fsp;
for (i=0; i < br_lck->num_locks; i++) {
/* Do any Windows or POSIX locks conflict ? */
if (brl_conflict(&locks[i], plock)) {
+ if (!serverid_exists(&locks[i].context.pid)) {
+ locks[i].context.pid.pid = 0;
+ br_lck->modified = true;
+ continue;
+ }
/* Remember who blocked us. */
plock->context.smblctx = locks[i].context.smblctx;
- return brl_lock_failed(fsp,plock,blocking_lock);
+ return NT_STATUS_LOCK_NOT_GRANTED;
}
#if ZERO_ZERO
if (plock->start == 0 && plock->size == 0 &&
#endif
}
- if (!IS_PENDING_LOCK(plock->lock_type)) {
- contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
- }
+ contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
/* We can get the Windows lock, now see if it needs to
be mapped into a lower level POSIX one, and if so can
we get it ? */
- if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(fsp->conn->params)) {
+ if (lp_posix_locking(fsp->conn->params)) {
int errno_ret;
if (!set_posix_lock_windows_flavour(fsp,
plock->start,
return NT_STATUS_OK;
fail:
- if (!IS_PENDING_LOCK(plock->lock_type)) {
- contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
- }
+ contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
return status;
}
We must cope with range splits and merges.
****************************************************************************/
-static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
- struct byte_range_lock *br_lck,
+static NTSTATUS brl_lock_posix(struct byte_range_lock *br_lck,
struct lock_struct *plock)
{
unsigned int i, count, posix_count;
struct lock_struct *locks = br_lck->lock_data;
struct lock_struct *tp;
- bool signal_pending_read = False;
bool break_oplocks = false;
NTSTATUS status;
for (i=0; i < br_lck->num_locks; i++) {
struct lock_struct *curr_lock = &locks[i];
- /* If we have a pending read lock, a lock downgrade should
- trigger a lock re-evaluation. */
- if (curr_lock->lock_type == PENDING_READ_LOCK &&
- brl_pending_overlap(plock, curr_lock)) {
- signal_pending_read = True;
- }
-
if (curr_lock->lock_flav == WINDOWS_LOCK) {
/* Do any Windows flavour locks conflict ? */
if (brl_conflict(curr_lock, plock)) {
+ if (!serverid_exists(&curr_lock->context.pid)) {
+ curr_lock->context.pid.pid = 0;
+ br_lck->modified = true;
+ continue;
+ }
/* No games with error messages. */
TALLOC_FREE(tp);
/* Remember who blocked us. */
/* POSIX conflict semantics are different. */
if (brl_conflict_posix(curr_lock, plock)) {
+ if (!serverid_exists(&curr_lock->context.pid)) {
+ curr_lock->context.pid.pid = 0;
+ br_lck->modified = true;
+ continue;
+ }
/* Can't block ourselves with POSIX locks. */
/* No games with error messages. */
TALLOC_FREE(tp);
* call contend_level2_oplocks_begin if this is the first POSIX brl on
* the file.
*/
- break_oplocks = (!IS_PENDING_LOCK(plock->lock_type) &&
- posix_count == 0);
+ break_oplocks = (posix_count == 0);
if (break_oplocks) {
contend_level2_oplocks_begin(br_lck->fsp,
LEVEL2_CONTEND_POSIX_BRL);
be mapped into a lower level POSIX one, and if so can
we get it ? */
- if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(br_lck->fsp->conn->params)) {
+ if (lp_posix_locking(br_lck->fsp->conn->params)) {
int errno_ret;
/* The lower layer just needs to attempt to
plock->start,
plock->size,
plock->lock_type,
+ &plock->context,
&errno_ret)) {
/* We don't know who blocked us. */
/* A successful downgrade from write to read lock can trigger a lock
re-evalutation where waiting readers can now proceed. */
- if (signal_pending_read) {
- /* Send unlock messages to any pending read waiters that overlap. */
- for (i=0; i < br_lck->num_locks; i++) {
- struct lock_struct *pend_lock = &locks[i];
-
- /* Ignore non-pending locks. */
- if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
- continue;
- }
-
- if (pend_lock->lock_type == PENDING_READ_LOCK &&
- brl_pending_overlap(plock, pend_lock)) {
- DEBUG(10,("brl_lock_posix: sending unlock message to pid %s\n",
- procid_str_static(&pend_lock->context.pid )));
-
- messaging_send(msg_ctx, pend_lock->context.pid,
- MSG_SMB_UNLOCK, &data_blob_null);
- }
- }
- }
-
return NT_STATUS_OK;
fail:
if (break_oplocks) {
NTSTATUS smb_vfs_call_brl_lock_windows(struct vfs_handle_struct *handle,
struct byte_range_lock *br_lck,
- struct lock_struct *plock,
- bool blocking_lock,
- struct blocking_lock_record *blr)
+ struct lock_struct *plock)
{
VFS_FIND(brl_lock_windows);
- return handle->fns->brl_lock_windows_fn(handle, br_lck, plock,
- blocking_lock, blr);
+ return handle->fns->brl_lock_windows_fn(handle, br_lck, plock);
}
/****************************************************************************
br_off size,
enum brl_type lock_type,
enum brl_flavour lock_flav,
- bool blocking_lock,
- uint64_t *psmblctx,
- struct blocking_lock_record *blr)
+ struct server_id *blocker_pid,
+ uint64_t *psmblctx)
{
NTSTATUS ret;
struct lock_struct lock;
+ ZERO_STRUCT(lock);
+
#if !ZERO_ZERO
if (start == 0 && size == 0) {
DEBUG(0,("client sent 0/0 lock - please report this\n"));
}
#endif
-#ifdef DEVELOPER
- /* Quieten valgrind on test. */
- ZERO_STRUCT(lock);
-#endif
-
- lock.context.smblctx = smblctx;
- lock.context.pid = pid;
- lock.context.tid = br_lck->fsp->conn->cnum;
- lock.start = start;
- lock.size = size;
- lock.fnum = br_lck->fsp->fnum;
- lock.lock_type = lock_type;
- lock.lock_flav = lock_flav;
+ lock = (struct lock_struct) {
+ .context.smblctx = smblctx,
+ .context.pid = pid,
+ .context.tid = br_lck->fsp->conn->cnum,
+ .start = start,
+ .size = size,
+ .fnum = br_lck->fsp->fnum,
+ .lock_type = lock_type,
+ .lock_flav = lock_flav
+ };
if (lock_flav == WINDOWS_LOCK) {
- ret = SMB_VFS_BRL_LOCK_WINDOWS(br_lck->fsp->conn, br_lck,
- &lock, blocking_lock, blr);
+ ret = SMB_VFS_BRL_LOCK_WINDOWS(
+ br_lck->fsp->conn, br_lck, &lock);
} else {
- ret = brl_lock_posix(msg_ctx, br_lck, &lock);
+ ret = brl_lock_posix(br_lck, &lock);
}
#if ZERO_ZERO
/* sort the lock list */
TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare);
#endif
-
/* If we're returning an error, return who blocked us. */
if (!NT_STATUS_IS_OK(ret) && psmblctx) {
+ *blocker_pid = lock.context.pid;
*psmblctx = lock.context.smblctx;
}
return ret;
}
+static void brl_delete_lock_struct(struct lock_struct *locks,
+ unsigned num_locks,
+ unsigned del_idx)
+{
+ if (del_idx >= num_locks) {
+ return;
+ }
+ memmove(&locks[del_idx], &locks[del_idx+1],
+ sizeof(*locks) * (num_locks - del_idx - 1));
+}
+
/****************************************************************************
Unlock a range of bytes - Windows semantics.
****************************************************************************/
struct byte_range_lock *br_lck,
const struct lock_struct *plock)
{
- unsigned int i, j;
+ unsigned int i;
struct lock_struct *locks = br_lck->lock_data;
enum brl_type deleted_lock_type = READ_LOCK; /* shut the compiler up.... */
for (i = 0; i < br_lck->num_locks; i++) {
struct lock_struct *lock = &locks[i];
- if (IS_PENDING_LOCK(lock->lock_type)) {
- continue;
- }
-
/* Only remove our own locks that match in start, size, and flavour. */
if (brl_same_context(&lock->context, &plock->context) &&
lock->fnum == plock->fnum &&
unlock_continue:
#endif
- /* Actually delete the lock. */
- if (i < br_lck->num_locks - 1) {
- memmove(&locks[i], &locks[i+1],
- sizeof(*locks)*((br_lck->num_locks-1) - i));
- }
-
+ brl_delete_lock_struct(locks, br_lck->num_locks, i);
br_lck->num_locks -= 1;
br_lck->modified = True;
br_lck->num_locks);
}
- /* 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];
-
- /* Ignore non-pending locks. */
- if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
- continue;
- }
-
- /* We could send specific lock info here... */
- if (brl_pending_overlap(plock, pend_lock)) {
- DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
- procid_str_static(&pend_lock->context.pid )));
-
- messaging_send(msg_ctx, pend_lock->context.pid,
- MSG_SMB_UNLOCK, &data_blob_null);
- }
- }
-
contend_level2_oplocks_end(br_lck->fsp, LEVEL2_CONTEND_WINDOWS_BRL);
return True;
}
struct byte_range_lock *br_lck,
struct lock_struct *plock)
{
- unsigned int i, j, count;
+ unsigned int i, count;
struct lock_struct *tp;
struct lock_struct *locks = br_lck->lock_data;
bool overlap_found = False;
unsigned int tmp_count;
/* Only remove our own locks - ignore fnum. */
- if (IS_PENDING_LOCK(lock->lock_type) ||
- !brl_same_context(&lock->context, &plock->context)) {
+ if (!brl_same_context(&lock->context, &plock->context)) {
memcpy(&tp[count], lock, sizeof(struct lock_struct));
count++;
continue;
br_lck->lock_data = tp;
br_lck->modified = True;
- /* 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];
-
- /* Ignore non-pending locks. */
- if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
- continue;
- }
-
- /* We could send specific lock info here... */
- if (brl_pending_overlap(plock, pend_lock)) {
- DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
- procid_str_static(&pend_lock->context.pid )));
-
- messaging_send(msg_ctx, pend_lock->context.pid,
- MSG_SMB_UNLOCK, &data_blob_null);
- }
- }
-
return True;
}
****************************************************************************/
bool brl_locktest(struct byte_range_lock *br_lck,
- uint64_t smblctx,
- struct server_id pid,
- br_off start,
- br_off size,
- enum brl_type lock_type,
- enum brl_flavour lock_flav)
+ const struct lock_struct *rw_probe)
{
bool ret = True;
unsigned int i;
- struct lock_struct lock;
- const struct lock_struct *locks = br_lck->lock_data;
+ struct lock_struct *locks = br_lck->lock_data;
files_struct *fsp = br_lck->fsp;
- lock.context.smblctx = smblctx;
- lock.context.pid = pid;
- lock.context.tid = br_lck->fsp->conn->cnum;
- lock.start = start;
- lock.size = size;
- lock.fnum = fsp->fnum;
- lock.lock_type = lock_type;
- lock.lock_flav = lock_flav;
-
/* Make sure existing locks don't conflict */
for (i=0; i < br_lck->num_locks; i++) {
/*
* Our own locks don't conflict.
*/
- if (brl_conflict_other(&locks[i], &lock)) {
+ if (brl_conflict_other(&locks[i], rw_probe)) {
+ if (br_lck->record == NULL) {
+ /* readonly */
+ return false;
+ }
+
+ if (!serverid_exists(&locks[i].context.pid)) {
+ locks[i].context.pid.pid = 0;
+ br_lck->modified = true;
+ continue;
+ }
+
return False;
}
}
* This only conflicts with Windows locks, not POSIX locks.
*/
- if(lp_posix_locking(fsp->conn->params) && (lock_flav == WINDOWS_LOCK)) {
+ if(lp_posix_locking(fsp->conn->params) &&
+ (rw_probe->lock_flav == WINDOWS_LOCK)) {
+ /*
+ * Make copies -- is_posix_locked might modify the values
+ */
+
+ br_off start = rw_probe->start;
+ br_off size = rw_probe->size;
+ enum brl_type lock_type = rw_probe->lock_type;
+
ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
- DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for %s file %s\n",
- (double)start, (double)size, ret ? "locked" : "unlocked",
- fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
+ DEBUG(10, ("brl_locktest: posix start=%ju len=%ju %s for %s "
+ "file %s\n", (uintmax_t)start, (uintmax_t)size,
+ ret ? "locked" : "unlocked",
+ fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
/* We need to return the inverse of is_posix_locked. */
ret = !ret;
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 %s file %s\n",
- (double)*pstart, (double)*psize, ret ? "locked" : "unlocked",
- fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
+ DEBUG(10, ("brl_lockquery: posix start=%ju len=%ju %s for %s "
+ "file %s\n", (uintmax_t)*pstart,
+ (uintmax_t)*psize, ret ? "locked" : "unlocked",
+ fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
if (ret) {
/* Hmmm. No clue what to set smblctx to - use -1. */
}
-bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct *handle,
- struct byte_range_lock *br_lck,
- struct lock_struct *plock,
- struct blocking_lock_record *blr)
-{
- VFS_FIND(brl_cancel_windows);
- return handle->fns->brl_cancel_windows_fn(handle, br_lck, plock, blr);
-}
-
-/****************************************************************************
- Remove a particular pending lock.
-****************************************************************************/
-bool brl_lock_cancel(struct byte_range_lock *br_lck,
- uint64_t smblctx,
- struct server_id pid,
- br_off start,
- br_off size,
- enum brl_flavour lock_flav,
- struct blocking_lock_record *blr)
-{
- bool ret;
- struct lock_struct lock;
-
- lock.context.smblctx = smblctx;
- lock.context.pid = pid;
- lock.context.tid = br_lck->fsp->conn->cnum;
- lock.start = start;
- lock.size = size;
- lock.fnum = br_lck->fsp->fnum;
- lock.lock_flav = lock_flav;
- /* lock.lock_type doesn't matter */
-
- if (lock_flav == WINDOWS_LOCK) {
- ret = SMB_VFS_BRL_CANCEL_WINDOWS(br_lck->fsp->conn, br_lck,
- &lock, blr);
- } else {
- ret = brl_lock_cancel_default(br_lck, &lock);
- }
-
- return ret;
-}
-
-bool brl_lock_cancel_default(struct byte_range_lock *br_lck,
- struct lock_struct *plock)
-{
- unsigned int i;
- struct lock_struct *locks = br_lck->lock_data;
-
- SMB_ASSERT(plock);
-
- for (i = 0; i < br_lck->num_locks; i++) {
- struct lock_struct *lock = &locks[i];
-
- /* For pending locks we *always* care about the fnum. */
- if (brl_same_context(&lock->context, &plock->context) &&
- lock->fnum == plock->fnum &&
- IS_PENDING_LOCK(lock->lock_type) &&
- lock->lock_flav == plock->lock_flav &&
- lock->start == plock->start &&
- lock->size == plock->size) {
- break;
- }
- }
-
- if (i == br_lck->num_locks) {
- /* Didn't find it. */
- return False;
- }
-
- if (i < br_lck->num_locks - 1) {
- /* Found this particular pending lock - delete it */
- memmove(&locks[i], &locks[i+1],
- sizeof(*locks)*((br_lck->num_locks-1) - i));
- }
-
- br_lck->num_locks -= 1;
- br_lck->modified = True;
- return True;
-}
-
/****************************************************************************
Remove any locks associated with a open file.
We return True if this process owns any other Windows locks on this
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 smblctx;
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 == NULL) {
+ return false;
+ }
+
+ smblctx = fsp->op->global->open_persistent_id;
+
if (!fsp->op->global->durable) {
return false;
}
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 smblctx;
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 == NULL) {
+ return false;
+ }
+
+ smblctx = fsp->op->global->open_persistent_id;
+
if (!fsp->op->global->durable) {
return false;
}
* 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 true;
}
-/****************************************************************************
- Ensure this set of lock entries is valid.
-****************************************************************************/
-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++) {
- 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 = talloc_array(
- mem_ctx, struct lock_struct,
- num_valid_entries);
- if (!new_lock_data) {
- DEBUG(3, ("malloc fail\n"));
- return False;
- }
-
- num_valid_entries = 0;
- for (i = 0; i < *pnum_entries; i++) {
- struct lock_struct *lock_data = &locks[i];
- if (lock_data->context.smblctx &&
- lock_data->context.tid) {
- /* Valid (nonzero) entry - copy it. */
- memcpy(&new_lock_data[num_valid_entries],
- lock_data, sizeof(struct lock_struct));
- num_valid_entries++;
- }
- }
- }
-
- TALLOC_FREE(*pplocks);
- *pplocks = new_lock_data;
- *pnum_entries = num_valid_entries;
- }
-
- return True;
-}
-
struct brl_forall_cb {
void (*fn)(struct file_id id, struct server_id pid,
enum brl_type lock_type,
struct file_id *key;
unsigned int i;
unsigned int num_locks = 0;
- unsigned int orig_num_locks = 0;
TDB_DATA dbkey;
TDB_DATA value;
}
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(talloc_tos(), &num_locks, &locks, true)) {
- TALLOC_FREE(locks);
- return -1; /* Terminate traversal */
- }
-
- if (orig_num_locks != num_locks) {
- if (num_locks) {
- TDB_DATA data;
- data.dptr = (uint8_t *)locks;
- data.dsize = num_locks*sizeof(struct lock_struct);
- dbwrap_record_store(rec, data, TDB_REPLACE);
- } else {
- dbwrap_record_delete(rec);
- }
- }
+ num_locks = value.dsize/sizeof(*locks);
if (cb->fn) {
for ( i=0; i<num_locks; i++) {
static void byte_range_lock_flush(struct byte_range_lock *br_lck)
{
+ unsigned i;
+ struct lock_struct *locks = br_lck->lock_data;
+
if (!br_lck->modified) {
+ DEBUG(10, ("br_lck not modified\n"));
goto done;
}
- if (br_lck->num_locks == 0) {
+ i = 0;
+
+ while (i < br_lck->num_locks) {
+ if (locks[i].context.pid.pid == 0) {
+ /*
+ * Autocleanup, the process conflicted and does not
+ * exist anymore.
+ */
+ locks[i] = locks[br_lck->num_locks-1];
+ br_lck->num_locks -= 1;
+ } else {
+ i += 1;
+ }
+ }
+
+ if ((br_lck->num_locks == 0) && (br_lck->num_read_oplocks == 0)) {
/* No locks - delete this entry. */
NTSTATUS status = dbwrap_record_delete(br_lck->record);
if (!NT_STATUS_IS_OK(status)) {
smb_panic("Could not delete byte range lock entry");
}
} else {
+ size_t lock_len, data_len;
TDB_DATA data;
NTSTATUS status;
- data.dptr = (uint8 *)br_lck->lock_data;
- data.dsize = br_lck->num_locks * sizeof(struct lock_struct);
+ lock_len = br_lck->num_locks * sizeof(struct lock_struct);
+ data_len = lock_len + sizeof(br_lck->num_read_oplocks);
+
+ data.dsize = data_len;
+ data.dptr = talloc_array(talloc_tos(), uint8_t, data_len);
+ SMB_ASSERT(data.dptr != NULL);
+
+ if (lock_len > 0) {
+ memcpy(data.dptr, br_lck->lock_data, lock_len);
+ }
+ memcpy(data.dptr + lock_len, &br_lck->num_read_oplocks,
+ sizeof(br_lck->num_read_oplocks));
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");
}
}
+ DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db)));
+
done:
br_lck->modified = false;
TALLOC_FREE(br_lck->record);
return 0;
}
+static bool brl_parse_data(struct byte_range_lock *br_lck, TDB_DATA data)
+{
+ size_t data_len;
+
+ if (data.dsize == 0) {
+ return true;
+ }
+ if (data.dsize % sizeof(struct lock_struct) !=
+ sizeof(br_lck->num_read_oplocks)) {
+ DEBUG(1, ("Invalid data size: %u\n", (unsigned)data.dsize));
+ return false;
+ }
+
+ br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
+ data_len = br_lck->num_locks * sizeof(struct lock_struct);
+
+ br_lck->lock_data = talloc_memdup(br_lck, data.dptr, data_len);
+ if (br_lck->lock_data == NULL) {
+ DEBUG(1, ("talloc_memdup failed\n"));
+ return false;
+ }
+ memcpy(&br_lck->num_read_oplocks, data.dptr + data_len,
+ sizeof(br_lck->num_read_oplocks));
+ return true;
+}
+
/*******************************************************************
Fetch a set of byte range lock data from the database.
Leave the record locked.
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(mem_ctx, struct byte_range_lock);
+ struct byte_range_lock *br_lck;
+ br_lck = talloc_zero(mem_ctx, struct byte_range_lock);
if (br_lck == NULL) {
return NULL;
}
br_lck->fsp = fsp;
- br_lck->num_locks = 0;
- br_lck->modified = False;
- key.dptr = (uint8 *)&fsp->file_id;
+ key.dptr = (uint8_t *)&fsp->file_id;
key.dsize = sizeof(struct file_id);
br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key);
data = dbwrap_record_get_value(br_lck->record);
- if ((data.dsize % sizeof(struct lock_struct)) != 0) {
- DEBUG(3, ("Got invalid brlock data\n"));
+ if (!brl_parse_data(br_lck, data)) {
TALLOC_FREE(br_lck);
return NULL;
}
- 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 = 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);
- }
-
- if (!fsp->lockdb_clean) {
- int orig_num_locks = br_lck->num_locks;
-
- /*
- * 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, &br_lck->num_locks,
- &br_lck->lock_data, false)) {
- TALLOC_FREE(br_lck);
- return NULL;
- }
-
- /* Ensure invalid locks are cleaned up in the destructor. */
- if (orig_num_locks != br_lck->num_locks) {
- br_lck->modified = True;
- }
-
- /* Mark the lockdb as "clean" as seen from this open file. */
- fsp->lockdb_clean = True;
- }
-
if (DEBUGLEVEL >= 10) {
unsigned int i;
struct lock_struct *locks = br_lck->lock_data;
{
struct brl_get_locks_readonly_state *state =
(struct brl_get_locks_readonly_state *)private_data;
- struct byte_range_lock *br_lock;
+ struct byte_range_lock *br_lck;
- br_lock = talloc_pooled_object(
+ br_lck = talloc_pooled_object(
state->mem_ctx, struct byte_range_lock, 1, data.dsize);
- if (br_lock == NULL) {
+ if (br_lck == 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);
-
- *state->br_lock = br_lock;
+ *br_lck = (struct byte_range_lock) { 0 };
+ if (!brl_parse_data(br_lck, data)) {
+ *state->br_lock = NULL;
+ return;
+ }
+ *state->br_lock = br_lck;
}
struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
{
struct byte_range_lock *br_lock = NULL;
- struct byte_range_lock *rw = NULL;
+ struct brl_get_locks_readonly_state state;
+ NTSTATUS status;
+
+ DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
+ dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum));
if ((fsp->brlock_rec != NULL)
&& (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
return fsp->brlock_rec;
}
- 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;
- }
+ /*
+ * Parse the record fresh from the database
+ */
- if (rw != NULL) {
- size_t lock_data_size;
+ 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_EQUAL(status,NT_STATUS_NOT_FOUND)) {
/*
- * Make a copy of the already retrieved and sanitized rw record
+ * No locks on this file. Return an empty br_lock.
*/
- lock_data_size = rw->num_locks * sizeof(struct lock_struct);
- br_lock = talloc_pooled_object(
- fsp, struct byte_range_lock, 1, lock_data_size);
+ br_lock = talloc(fsp, struct byte_range_lock);
if (br_lock == NULL) {
- goto fail;
+ return NULL;
}
- 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;
+ br_lock->num_read_oplocks = 0;
+ br_lock->num_locks = 0;
+ br_lock->lock_data = NULL;
- 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;
- }
+ } else if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(3, ("Could not parse byte range lock record: "
+ "%s\n", nt_errstr(status)));
+ return NULL;
+ }
+ if (br_lock == NULL) {
+ return NULL;
}
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);
- }
+ /*
+ * 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);
-fail:
- TALLOC_FREE(rw);
return br_lock;
}
-struct brl_revalidate_state {
- ssize_t array_size;
- uint32 num_pids;
- struct server_id *pids;
-};
-
-/*
- * Collect PIDs of all processes with pending entries
- */
-
-static void brl_revalidate_collect(struct file_id id, struct server_id pid,
- enum brl_type lock_type,
- enum brl_flavour lock_flav,
- br_off start, br_off size,
- void *private_data)
-{
- struct brl_revalidate_state *state =
- (struct brl_revalidate_state *)private_data;
-
- if (!IS_PENDING_LOCK(lock_type)) {
- return;
- }
-
- add_to_large_array(state, sizeof(pid), (void *)&pid,
- &state->pids, &state->num_pids,
- &state->array_size);
-}
-
-/*
- * qsort callback to sort the processes
- */
-
-static int compare_procids(const void *p1, const void *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;
- return 0;
-}
-
-/*
- * Send a MSG_SMB_UNLOCK message to all processes with pending byte range
- * locks so that they retry. Mainly used in the cluster code after a node has
- * died.
- *
- * Done in two steps to avoid double-sends: First we collect all entries in an
- * array, then qsort that array and only send to non-dupes.
- */
-
-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(NULL, struct brl_revalidate_state))) {
- DEBUG(0, ("talloc failed\n"));
- return;
- }
-
- brl_forall(brl_revalidate_collect, state);
-
- if (state->array_size == -1) {
- DEBUG(0, ("talloc failed\n"));
- goto done;
- }
-
- if (state->num_pids == 0) {
- goto done;
- }
-
- TYPESAFE_QSORT(state->pids, state->num_pids, compare_procids);
-
- ZERO_STRUCT(last_pid);
-
- for (i=0; i<state->num_pids; i++) {
- if (serverid_equal(&last_pid, &state->pids[i])) {
- /*
- * We've seen that one already
- */
- continue;
- }
-
- messaging_send(msg_ctx, state->pids[i], MSG_SMB_UNLOCK,
- &data_blob_null);
- last_pid = state->pids[i];
- }
-
- done:
- TALLOC_FREE(state);
- return;
-}
-
bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id)
{
bool ret = false;
struct lock_context *ctx = &lock[n].context;
if (!server_id_is_disconnected(&ctx->pid)) {
+ struct server_id_buf tmp;
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)));
+ server_id_str_buf(ctx->pid, &tmp)));
goto done;
}
git.samba.org / samba.git / blobdiff