📄 locking.c
字号:
/* Unix SMB/CIFS implementation. Locking functions Copyright (C) Andrew Tridgell 1992-2000 Copyright (C) Jeremy Allison 1992-2000 Copyright (C) Volker Lendecke 2005 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 2 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, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. Revision History: 12 aug 96: Erik.Devriendt@te6.siemens.be added support for shared memory implementation of share mode locking May 1997. Jeremy Allison (jallison@whistle.com). Modified share mode locking to deal with multiple share modes per open file. September 1997. Jeremy Allison (jallison@whistle.com). Added oplock support. rewrtten completely to use new tdb code. Tridge, Dec '99 Added POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.*/#include "includes.h"uint16 global_smbpid;#undef DBGC_CLASS#define DBGC_CLASS DBGC_LOCKING/* the locking database handle */static TDB_CONTEXT *tdb;struct locking_data { union { struct { int num_share_mode_entries; BOOL delete_on_close; } s; struct share_mode_entry dummy; /* Needed for alignment. */ } u; /* the following two entries are implicit struct share_mode_entry modes[num_share_mode_entries]; char file_name[]; */};/**************************************************************************** Debugging aid :-).****************************************************************************/static const char *lock_type_name(enum brl_type lock_type){ return (lock_type == READ_LOCK) ? "READ" : "WRITE";}/**************************************************************************** Utility function called to see if a file region is locked.****************************************************************************/BOOL is_locked(files_struct *fsp,connection_struct *conn, SMB_BIG_UINT count,SMB_BIG_UINT offset, enum brl_type lock_type){ int snum = SNUM(conn); int strict_locking = lp_strict_locking(snum); BOOL ret; if (count == 0) return(False); if (!lp_locking(snum) || !strict_locking) return(False); if (strict_locking == Auto) { if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && (lock_type == READ_LOCK || lock_type == WRITE_LOCK)) { DEBUG(10,("is_locked: optimisation - exclusive oplock on file %s\n", fsp->fsp_name )); ret = 0; } else if ((fsp->oplock_type == LEVEL_II_OPLOCK) && (lock_type == READ_LOCK)) { DEBUG(10,("is_locked: optimisation - level II oplock on file %s\n", fsp->fsp_name )); ret = 0; } else { ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum, global_smbpid, procid_self(), conn->cnum, offset, count, lock_type); } } else { ret = !brl_locktest(fsp->dev, fsp->inode, fsp->fnum, global_smbpid, procid_self(), conn->cnum, offset, count, lock_type); } DEBUG(10,("is_locked: brl start=%.0f len=%.0f %s for file %s\n", (double)offset, (double)count, ret ? "locked" : "unlocked", fsp->fsp_name )); /* * There is no lock held by an SMB daemon, check to * see if there is a POSIX lock from a UNIX or NFS process. */ if(!ret && lp_posix_locking(snum)) { ret = is_posix_locked(fsp, offset, count, lock_type); DEBUG(10,("is_locked: posix start=%.0f len=%.0f %s for file %s\n", (double)offset, (double)count, ret ? "locked" : "unlocked", fsp->fsp_name )); } return ret;}/**************************************************************************** Utility function called by locking requests.****************************************************************************/static NTSTATUS do_lock(files_struct *fsp,connection_struct *conn, uint16 lock_pid, SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx){ NTSTATUS status = NT_STATUS_LOCK_NOT_GRANTED; if (!lp_locking(SNUM(conn))) return NT_STATUS_OK; /* NOTE! 0 byte long ranges ARE allowed and should be stored */ DEBUG(10,("do_lock: lock type %s start=%.0f len=%.0f requested for file %s\n", lock_type_name(lock_type), (double)offset, (double)count, fsp->fsp_name )); if (OPEN_FSP(fsp) && fsp->can_lock && (fsp->conn == conn)) { status = brl_lock(fsp->dev, fsp->inode, fsp->fnum, lock_pid, procid_self(), conn->cnum, offset, count, lock_type, my_lock_ctx); if (NT_STATUS_IS_OK(status) && lp_posix_locking(SNUM(conn))) { /* * Try and get a POSIX lock on this range. * Note that this is ok if it is a read lock * overlapping on a different fd. JRA. */ if (!set_posix_lock(fsp, offset, count, lock_type)) { if (errno == EACCES || errno == EAGAIN) status = NT_STATUS_FILE_LOCK_CONFLICT; else status = map_nt_error_from_unix(errno); /* * We failed to map - we must now remove the brl * lock entry. */ (void)brl_unlock(fsp->dev, fsp->inode, fsp->fnum, lock_pid, procid_self(), conn->cnum, offset, count, False, NULL, NULL); } } } return status;}/**************************************************************************** Utility function called by locking requests. This is *DISGUSTING*. It also appears to be "What Windows Does" (tm). Andrew, ever wonder why Windows 2000 is so slow on the locking tests...... ? This is the reason. Much though I hate it, we need this. JRA.****************************************************************************/NTSTATUS do_lock_spin(files_struct *fsp,connection_struct *conn, uint16 lock_pid, SMB_BIG_UINT count,SMB_BIG_UINT offset,enum brl_type lock_type, BOOL *my_lock_ctx){ int j, maxj = lp_lock_spin_count(); int sleeptime = lp_lock_sleep_time(); NTSTATUS status, ret; if (maxj <= 0) maxj = 1; ret = NT_STATUS_OK; /* to keep dumb compilers happy */ for (j = 0; j < maxj; j++) { status = do_lock(fsp, conn, lock_pid, count, offset, lock_type, my_lock_ctx); if (!NT_STATUS_EQUAL(status, NT_STATUS_LOCK_NOT_GRANTED) && !NT_STATUS_EQUAL(status, NT_STATUS_FILE_LOCK_CONFLICT)) { return status; } /* if we do fail then return the first error code we got */ if (j == 0) { ret = status; /* Don't spin if we blocked ourselves. */ if (*my_lock_ctx) return ret; } if (sleeptime) sys_usleep(sleeptime); } return ret;}/* Struct passed to brl_unlock. */struct posix_unlock_data_struct { files_struct *fsp; SMB_BIG_UINT offset; SMB_BIG_UINT count;};/**************************************************************************** Function passed to brl_unlock to allow POSIX unlock to be done first.****************************************************************************/static void posix_unlock(void *pre_data){ struct posix_unlock_data_struct *pdata = (struct posix_unlock_data_struct *)pre_data; if (lp_posix_locking(SNUM(pdata->fsp->conn))) release_posix_lock(pdata->fsp, pdata->offset, pdata->count);}/**************************************************************************** Utility function called by unlocking requests.****************************************************************************/NTSTATUS do_unlock(files_struct *fsp,connection_struct *conn, uint16 lock_pid, SMB_BIG_UINT count,SMB_BIG_UINT offset){ BOOL ok = False; struct posix_unlock_data_struct posix_data; if (!lp_locking(SNUM(conn))) return NT_STATUS_OK; if (!OPEN_FSP(fsp) || !fsp->can_lock || (fsp->conn != conn)) { return NT_STATUS_INVALID_HANDLE; } DEBUG(10,("do_unlock: unlock start=%.0f len=%.0f requested for file %s\n", (double)offset, (double)count, fsp->fsp_name )); /* * Remove the existing lock record from the tdb lockdb * before looking at POSIX locks. If this record doesn't * match then don't bother looking to remove POSIX locks. */ posix_data.fsp = fsp; posix_data.offset = offset; posix_data.count = count; ok = brl_unlock(fsp->dev, fsp->inode, fsp->fnum, lock_pid, procid_self(), conn->cnum, offset, count, False, posix_unlock, (void *)&posix_data); if (!ok) { DEBUG(10,("do_unlock: returning ERRlock.\n" )); return NT_STATUS_RANGE_NOT_LOCKED; } return NT_STATUS_OK;}/**************************************************************************** Remove any locks on this fd. Called from file_close().****************************************************************************/void locking_close_file(files_struct *fsp){ struct process_id pid = procid_self(); if (!lp_locking(SNUM(fsp->conn))) return; /* * Just release all the brl locks, no need to release individually. */ brl_close(fsp->dev, fsp->inode, pid, fsp->conn->cnum, fsp->fnum); if(lp_posix_locking(SNUM(fsp->conn))) { /* * Release all the POSIX locks. */ posix_locking_close_file(fsp); }}/**************************************************************************** Initialise the locking functions.****************************************************************************/static int open_read_only;BOOL locking_init(int read_only){ brl_init(read_only); if (tdb) return True; tdb = tdb_open_log(lock_path("locking.tdb"), SMB_OPEN_DATABASE_TDB_HASH_SIZE, TDB_DEFAULT|(read_only?0x0:TDB_CLEAR_IF_FIRST), read_only?O_RDONLY:O_RDWR|O_CREAT, 0644); if (!tdb) { DEBUG(0,("ERROR: Failed to initialise locking database\n")); return False; } if (!posix_locking_init(read_only)) return False; open_read_only = read_only; return True;}/******************************************************************* Deinitialize the share_mode management.******************************************************************/BOOL locking_end(void){ BOOL ret = True; brl_shutdown(open_read_only); if (tdb) { if (tdb_close(tdb) != 0) ret = False; } return ret;}/******************************************************************* Form a static locking key for a dev/inode pair.******************************************************************//* key and data records in the tdb locking database */struct locking_key { SMB_DEV_T dev; SMB_INO_T ino;};/******************************************************************* Form a static locking key for a dev/inode pair.******************************************************************/static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode){ static struct locking_key key; TDB_DATA kbuf; memset(&key, '\0', sizeof(key)); key.dev = dev; key.ino = inode; kbuf.dptr = (char *)&key; kbuf.dsize = sizeof(key); return kbuf;}/******************************************************************* Print out a share mode.********************************************************************/char *share_mode_str(int num, struct share_mode_entry *e){ static pstring share_str; slprintf(share_str, sizeof(share_str)-1, "share_mode_entry[%d]: %s " "pid = %s, share_access = 0x%x, private_options = 0x%x, " "access_mask = 0x%x, mid = 0x%x, type= 0x%x, file_id = %lu, " "dev = 0x%x, inode = %.0f", num, e->op_type == UNUSED_SHARE_MODE_ENTRY ? "UNUSED" : "", procid_str_static(&e->pid), e->share_access, e->private_options, e->access_mask, e->op_mid, e->op_type, e->share_file_id, (unsigned int)e->dev, (double)e->inode ); return share_str;}/******************************************************************* Print out a share mode table.********************************************************************/static void print_share_mode_table(struct locking_data *data){ int num_share_modes = data->u.s.num_share_mode_entries; struct share_mode_entry *shares = (struct share_mode_entry *)(data + 1); int i; for (i = 0; i < num_share_modes; i++) { struct share_mode_entry entry; memcpy(&entry, &shares[i], sizeof(struct share_mode_entry)); DEBUG(10,("print_share_mode_table: %s\n", share_mode_str(i, &entry))); }}/******************************************************************* Get all share mode entries for a dev/inode pair.********************************************************************/static BOOL parse_share_modes(TDB_DATA dbuf, struct share_mode_lock *lck){ struct locking_data *data; int i; if (dbuf.dsize < sizeof(struct locking_data)) { DEBUG(0, ("parse_share_modes: buffer too short\n")); return False; } data = (struct locking_data *)dbuf.dptr; lck->delete_on_close = data->u.s.delete_on_close; lck->num_share_modes = data->u.s.num_share_mode_entries; DEBUG(10, ("parse_share_modes: delete_on_close: %d, " "num_share_modes: %d\n", lck->delete_on_close, lck->num_share_modes)); if ((lck->num_share_modes < 0) || (lck->num_share_modes > 1000000)) { DEBUG(0, ("invalid number of share modes: %d\n", lck->num_share_modes)); return False; } lck->share_modes = NULL; if (lck->num_share_modes != 0) { if (dbuf.dsize < (sizeof(struct locking_data) + (lck->num_share_modes * sizeof(struct share_mode_entry)))) { DEBUG(0, ("parse_share_modes: buffer too short\n")); return False; } lck->share_modes = talloc_memdup(lck, dbuf.dptr+sizeof(*data), lck->num_share_modes * sizeof(struct share_mode_entry)); if (lck->share_modes == NULL) { DEBUG(0, ("talloc failed\n")); return False; } } /* Save off the associated service path and filename. */ lck->servicepath = talloc_strdup(lck, dbuf.dptr + sizeof(*data) + (lck->num_share_modes * sizeof(struct share_mode_entry))); lck->filename = talloc_strdup(lck, dbuf.dptr + sizeof(*data) + (lck->num_share_modes * sizeof(struct share_mode_entry)) + strlen(lck->servicepath) + 1 ); /* * Ensure that each entry has a real process attached. */ for (i = 0; i < lck->num_share_modes; i++) { struct share_mode_entry *entry_p = &lck->share_modes[i]; DEBUG(10,("parse_share_modes: %s\n", share_mode_str(i, entry_p) )); if (!process_exists(entry_p->pid)) { DEBUG(10,("parse_share_modes: deleted %s\n", share_mode_str(i, entry_p) )); entry_p->op_type = UNUSED_SHARE_MODE_ENTRY; lck->modified = True; } } return True;}static TDB_DATA unparse_share_modes(struct share_mode_lock *lck){ TDB_DATA result; int num_valid = 0; int i; struct locking_data *data; ssize_t offset; ssize_t sp_len; result.dptr = NULL; result.dsize = 0; for (i=0; i<lck->num_share_modes; i++) { if (!is_unused_share_mode_entry(&lck->share_modes[i])) { num_valid += 1; } }
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -