linux/fs/dlm/user.c

926 lines
22 KiB
C
Raw Normal View History

/*
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
* Copyright (C) 2006-2007 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU General Public License v.2.
*/
#include <linux/miscdevice.h>
#include <linux/init.h>
#include <linux/wait.h>
#include <linux/module.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/poll.h>
#include <linux/signal.h>
#include <linux/spinlock.h>
#include <linux/dlm.h>
#include <linux/dlm_device.h>
#include "dlm_internal.h"
#include "lockspace.h"
#include "lock.h"
#include "lvb_table.h"
#include "user.h"
static const char name_prefix[] = "dlm";
static const struct file_operations device_fops;
#ifdef CONFIG_COMPAT
struct dlm_lock_params32 {
__u8 mode;
__u8 namelen;
__u16 unused;
__u32 flags;
__u32 lkid;
__u32 parent;
__u64 xid;
__u64 timeout;
__u32 castparam;
__u32 castaddr;
__u32 bastparam;
__u32 bastaddr;
__u32 lksb;
char lvb[DLM_USER_LVB_LEN];
char name[0];
};
struct dlm_write_request32 {
__u32 version[3];
__u8 cmd;
__u8 is64bit;
__u8 unused[2];
union {
struct dlm_lock_params32 lock;
struct dlm_lspace_params lspace;
struct dlm_purge_params purge;
} i;
};
struct dlm_lksb32 {
__u32 sb_status;
__u32 sb_lkid;
__u8 sb_flags;
__u32 sb_lvbptr;
};
struct dlm_lock_result32 {
__u32 version[3];
__u32 length;
__u32 user_astaddr;
__u32 user_astparam;
__u32 user_lksb;
struct dlm_lksb32 lksb;
__u8 bast_mode;
__u8 unused[3];
/* Offsets may be zero if no data is present */
__u32 lvb_offset;
};
static void compat_input(struct dlm_write_request *kb,
struct dlm_write_request32 *kb32,
int max_namelen)
{
kb->version[0] = kb32->version[0];
kb->version[1] = kb32->version[1];
kb->version[2] = kb32->version[2];
kb->cmd = kb32->cmd;
kb->is64bit = kb32->is64bit;
if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
kb->i.lspace.flags = kb32->i.lspace.flags;
kb->i.lspace.minor = kb32->i.lspace.minor;
strcpy(kb->i.lspace.name, kb32->i.lspace.name);
} else if (kb->cmd == DLM_USER_PURGE) {
kb->i.purge.nodeid = kb32->i.purge.nodeid;
kb->i.purge.pid = kb32->i.purge.pid;
} else {
kb->i.lock.mode = kb32->i.lock.mode;
kb->i.lock.namelen = kb32->i.lock.namelen;
kb->i.lock.flags = kb32->i.lock.flags;
kb->i.lock.lkid = kb32->i.lock.lkid;
kb->i.lock.parent = kb32->i.lock.parent;
kb->i.lock.xid = kb32->i.lock.xid;
kb->i.lock.timeout = kb32->i.lock.timeout;
kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
if (kb->i.lock.namelen <= max_namelen)
memcpy(kb->i.lock.name, kb32->i.lock.name,
kb->i.lock.namelen);
else
kb->i.lock.namelen = max_namelen;
}
}
static void compat_output(struct dlm_lock_result *res,
struct dlm_lock_result32 *res32)
{
res32->version[0] = res->version[0];
res32->version[1] = res->version[1];
res32->version[2] = res->version[2];
res32->user_astaddr = (__u32)(long)res->user_astaddr;
res32->user_astparam = (__u32)(long)res->user_astparam;
res32->user_lksb = (__u32)(long)res->user_lksb;
res32->bast_mode = res->bast_mode;
res32->lvb_offset = res->lvb_offset;
res32->length = res->length;
res32->lksb.sb_status = res->lksb.sb_status;
res32->lksb.sb_flags = res->lksb.sb_flags;
res32->lksb.sb_lkid = res->lksb.sb_lkid;
res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
}
#endif
/* Figure out if this lock is at the end of its life and no longer
available for the application to use. The lkb still exists until
the final ast is read. A lock becomes EOL in three situations:
1. a noqueue request fails with EAGAIN
2. an unlock completes with EUNLOCK
3. a cancel of a waiting request completes with ECANCEL/EDEADLK
An EOL lock needs to be removed from the process's list of locks.
And we can't allow any new operation on an EOL lock. This is
not related to the lifetime of the lkb struct which is managed
entirely by refcount. */
static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
{
switch (sb_status) {
case -DLM_EUNLOCK:
return 1;
case -DLM_ECANCEL:
case -ETIMEDOUT:
case -EDEADLK:
if (lkb->lkb_grmode == DLM_LOCK_IV)
return 1;
break;
case -EAGAIN:
if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
return 1;
break;
}
return 0;
}
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
/* we could possibly check if the cancel of an orphan has resulted in the lkb
being removed and then remove that lkb from the orphans list and free it */
void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
{
struct dlm_ls *ls;
struct dlm_user_args *ua;
struct dlm_user_proc *proc;
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
int eol = 0, ast_type;
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
return;
ls = lkb->lkb_resource->res_ls;
mutex_lock(&ls->ls_clear_proc_locks);
/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
lkb->ua so we can't try to use it. This second check is necessary
for cases where a completion ast is received for an operation that
began before clear_proc_locks did its cancel/unlock. */
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
goto out;
DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb););
ua = (struct dlm_user_args *)lkb->lkb_astparam;
proc = ua->proc;
if (type == AST_BAST && ua->bastaddr == NULL)
goto out;
spin_lock(&proc->asts_spin);
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
ast_type = lkb->lkb_ast_type;
lkb->lkb_ast_type |= type;
if (!ast_type) {
kref_get(&lkb->lkb_ref);
list_add_tail(&lkb->lkb_astqueue, &proc->asts);
wake_up_interruptible(&proc->wait);
}
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (type == AST_COMP && (ast_type & AST_COMP))
log_debug(ls, "ast overlap %x status %x %x",
lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (eol) {
lkb->lkb_ast_type &= ~AST_BAST;
lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
}
/* We want to copy the lvb to userspace when the completion
ast is read if the status is 0, the lock has an lvb and
lvb_ops says we should. We could probably have set_lvb_lock()
set update_user_lvb instead and not need old_mode */
if ((lkb->lkb_ast_type & AST_COMP) &&
(lkb->lkb_lksb->sb_status == 0) &&
lkb->lkb_lksb->sb_lvbptr &&
dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
ua->update_user_lvb = 1;
else
ua->update_user_lvb = 0;
spin_unlock(&proc->asts_spin);
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (eol) {
spin_lock(&proc->locks_spin);
[DLM] overlapping cancel and unlock Full cancel and force-unlock support. In the past, cancel and force-unlock wouldn't work if there was another operation in progress on the lock. Now, both cancel and unlock-force can overlap an operation on a lock, meaning there may be 2 or 3 operations in progress on a lock in parallel. This support is important not only because cancel and force-unlock are explicit operations that an app can use, but both are used implicitly when a process exits while holding locks. Summary of changes: - add-to and remove-from waiters functions were rewritten to handle situations with more than one remote operation outstanding on a lock - validate_unlock_args detects when an overlapping cancel/unlock-force can be sent and when it needs to be delayed until a request/lookup reply is received - processing request/lookup replies detects when cancel/unlock-force occured during the op, and carries out the delayed cancel/unlock-force - manipulation of the "waiters" (remote operation) state of a lock moved under the standard rsb mutex that protects all the other lock state - the two recovery routines related to locks on the waiters list changed according to the way lkb's are now locked before accessing waiters state - waiters recovery detects when lkb's being recovered have overlapping cancel/unlock-force, and may not recover such locks - revert_lock (cancel) returns a value to distinguish cases where it did nothing vs cases where it actually did a cancel; the cancel completion ast should only be done when cancel did something - orphaned locks put on new list so they can be found later for purging - cancel must be called on a lock when making it an orphan - flag user locks (ENDOFLIFE) at the end of their useful life (to the application) so we can return an error for any further cancel/unlock-force - we weren't setting COMP/BAST ast flags if one was already set, so we'd lose either a completion or blocking ast - clear an unread bast on a lock that's become unlocked Signed-off-by: David Teigland <teigland@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-03-28 16:56:46 +02:00
if (!list_empty(&lkb->lkb_ownqueue)) {
list_del_init(&lkb->lkb_ownqueue);
dlm_put_lkb(lkb);
}
spin_unlock(&proc->locks_spin);
}
out:
mutex_unlock(&ls->ls_clear_proc_locks);
}
static int device_user_lock(struct dlm_user_proc *proc,
struct dlm_lock_params *params)
{
struct dlm_ls *ls;
struct dlm_user_args *ua;
int error = -ENOMEM;
ls = dlm_find_lockspace_local(proc->lockspace);
if (!ls)
return -ENOENT;
if (!params->castaddr || !params->lksb) {
error = -EINVAL;
goto out;
}
ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
if (!ua)
goto out;
ua->proc = proc;
ua->user_lksb = params->lksb;
ua->castparam = params->castparam;
ua->castaddr = params->castaddr;
ua->bastparam = params->bastparam;
ua->bastaddr = params->bastaddr;
ua->xid = params->xid;
if (params->flags & DLM_LKF_CONVERT)
error = dlm_user_convert(ls, ua,
params->mode, params->flags,
params->lkid, params->lvb,
(unsigned long) params->timeout);
else {
error = dlm_user_request(ls, ua,
params->mode, params->flags,
params->name, params->namelen,
(unsigned long) params->timeout);
if (!error)
error = ua->lksb.sb_lkid;
}
out:
dlm_put_lockspace(ls);
return error;
}
static int device_user_unlock(struct dlm_user_proc *proc,
struct dlm_lock_params *params)
{
struct dlm_ls *ls;
struct dlm_user_args *ua;
int error = -ENOMEM;
ls = dlm_find_lockspace_local(proc->lockspace);
if (!ls)
return -ENOENT;
ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
if (!ua)
goto out;
ua->proc = proc;
ua->user_lksb = params->lksb;
ua->castparam = params->castparam;
ua->castaddr = params->castaddr;
if (params->flags & DLM_LKF_CANCEL)
error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
else
error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
params->lvb);
out:
dlm_put_lockspace(ls);
return error;
}
static int device_user_deadlock(struct dlm_user_proc *proc,
struct dlm_lock_params *params)
{
struct dlm_ls *ls;
int error;
ls = dlm_find_lockspace_local(proc->lockspace);
if (!ls)
return -ENOENT;
error = dlm_user_deadlock(ls, params->flags, params->lkid);
dlm_put_lockspace(ls);
return error;
}
static int create_misc_device(struct dlm_ls *ls, char *name)
{
int error, len;
error = -ENOMEM;
len = strlen(name) + strlen(name_prefix) + 2;
ls->ls_device.name = kzalloc(len, GFP_KERNEL);
if (!ls->ls_device.name)
goto fail;
snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
name);
ls->ls_device.fops = &device_fops;
ls->ls_device.minor = MISC_DYNAMIC_MINOR;
error = misc_register(&ls->ls_device);
if (error) {
kfree(ls->ls_device.name);
}
fail:
return error;
}
static int device_user_purge(struct dlm_user_proc *proc,
struct dlm_purge_params *params)
{
struct dlm_ls *ls;
int error;
ls = dlm_find_lockspace_local(proc->lockspace);
if (!ls)
return -ENOENT;
error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
dlm_put_lockspace(ls);
return error;
}
static int device_create_lockspace(struct dlm_lspace_params *params)
{
dlm_lockspace_t *lockspace;
struct dlm_ls *ls;
int error;
if (!capable(CAP_SYS_ADMIN))
return -EPERM;
error = dlm_new_lockspace(params->name, strlen(params->name),
&lockspace, params->flags, DLM_USER_LVB_LEN);
if (error)
return error;
ls = dlm_find_lockspace_local(lockspace);
if (!ls)
return -ENOENT;
error = create_misc_device(ls, params->name);
dlm_put_lockspace(ls);
if (error)
dlm_release_lockspace(lockspace, 0);
else
error = ls->ls_device.minor;
return error;
}
static int device_remove_lockspace(struct dlm_lspace_params *params)
{
dlm_lockspace_t *lockspace;
struct dlm_ls *ls;
int error, force = 0;
if (!capable(CAP_SYS_ADMIN))
return -EPERM;
ls = dlm_find_lockspace_device(params->minor);
if (!ls)
return -ENOENT;
/* Deregister the misc device first, so we don't have
* a device that's not attached to a lockspace. If
* dlm_release_lockspace fails then we can recreate it
*/
error = misc_deregister(&ls->ls_device);
if (error) {
dlm_put_lockspace(ls);
goto out;
}
kfree(ls->ls_device.name);
if (params->flags & DLM_USER_LSFLG_FORCEFREE)
force = 2;
lockspace = ls->ls_local_handle;
/* dlm_release_lockspace waits for references to go to zero,
so all processes will need to close their device for the ls
before the release will procede */
dlm_put_lockspace(ls);
error = dlm_release_lockspace(lockspace, force);
if (error)
create_misc_device(ls, ls->ls_name);
out:
return error;
}
/* Check the user's version matches ours */
static int check_version(struct dlm_write_request *req)
{
if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
(req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
"user (%d.%d.%d) kernel (%d.%d.%d)\n",
current->comm,
task_pid_nr(current),
req->version[0],
req->version[1],
req->version[2],
DLM_DEVICE_VERSION_MAJOR,
DLM_DEVICE_VERSION_MINOR,
DLM_DEVICE_VERSION_PATCH);
return -EINVAL;
}
return 0;
}
/*
* device_write
*
* device_user_lock
* dlm_user_request -> request_lock
* dlm_user_convert -> convert_lock
*
* device_user_unlock
* dlm_user_unlock -> unlock_lock
* dlm_user_cancel -> cancel_lock
*
* device_create_lockspace
* dlm_new_lockspace
*
* device_remove_lockspace
* dlm_release_lockspace
*/
/* a write to a lockspace device is a lock or unlock request, a write
to the control device is to create/remove a lockspace */
static ssize_t device_write(struct file *file, const char __user *buf,
size_t count, loff_t *ppos)
{
struct dlm_user_proc *proc = file->private_data;
struct dlm_write_request *kbuf;
sigset_t tmpsig, allsigs;
int error;
#ifdef CONFIG_COMPAT
if (count < sizeof(struct dlm_write_request32))
#else
if (count < sizeof(struct dlm_write_request))
#endif
return -EINVAL;
kbuf = kmalloc(count, GFP_KERNEL);
if (!kbuf)
return -ENOMEM;
if (copy_from_user(kbuf, buf, count)) {
error = -EFAULT;
goto out_free;
}
if (check_version(kbuf)) {
error = -EBADE;
goto out_free;
}
#ifdef CONFIG_COMPAT
if (!kbuf->is64bit) {
struct dlm_write_request32 *k32buf;
k32buf = (struct dlm_write_request32 *)kbuf;
kbuf = kmalloc(count + (sizeof(struct dlm_write_request) -
sizeof(struct dlm_write_request32)), GFP_KERNEL);
if (!kbuf)
return -ENOMEM;
if (proc)
set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
compat_input(kbuf, k32buf,
count - sizeof(struct dlm_write_request32));
kfree(k32buf);
}
#endif
/* do we really need this? can a write happen after a close? */
if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
return -EINVAL;
sigfillset(&allsigs);
sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
error = -EINVAL;
switch (kbuf->cmd)
{
case DLM_USER_LOCK:
if (!proc) {
log_print("no locking on control device");
goto out_sig;
}
error = device_user_lock(proc, &kbuf->i.lock);
break;
case DLM_USER_UNLOCK:
if (!proc) {
log_print("no locking on control device");
goto out_sig;
}
error = device_user_unlock(proc, &kbuf->i.lock);
break;
case DLM_USER_DEADLOCK:
if (!proc) {
log_print("no locking on control device");
goto out_sig;
}
error = device_user_deadlock(proc, &kbuf->i.lock);
break;
case DLM_USER_CREATE_LOCKSPACE:
if (proc) {
log_print("create/remove only on control device");
goto out_sig;
}
error = device_create_lockspace(&kbuf->i.lspace);
break;
case DLM_USER_REMOVE_LOCKSPACE:
if (proc) {
log_print("create/remove only on control device");
goto out_sig;
}
error = device_remove_lockspace(&kbuf->i.lspace);
break;
case DLM_USER_PURGE:
if (!proc) {
log_print("no locking on control device");
goto out_sig;
}
error = device_user_purge(proc, &kbuf->i.purge);
break;
default:
log_print("Unknown command passed to DLM device : %d\n",
kbuf->cmd);
}
out_sig:
sigprocmask(SIG_SETMASK, &tmpsig, NULL);
recalc_sigpending();
out_free:
kfree(kbuf);
return error;
}
/* Every process that opens the lockspace device has its own "proc" structure
hanging off the open file that's used to keep track of locks owned by the
process and asts that need to be delivered to the process. */
static int device_open(struct inode *inode, struct file *file)
{
struct dlm_user_proc *proc;
struct dlm_ls *ls;
ls = dlm_find_lockspace_device(iminor(inode));
if (!ls)
return -ENOENT;
proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
if (!proc) {
dlm_put_lockspace(ls);
return -ENOMEM;
}
proc->lockspace = ls->ls_local_handle;
INIT_LIST_HEAD(&proc->asts);
INIT_LIST_HEAD(&proc->locks);
INIT_LIST_HEAD(&proc->unlocking);
spin_lock_init(&proc->asts_spin);
spin_lock_init(&proc->locks_spin);
init_waitqueue_head(&proc->wait);
file->private_data = proc;
return 0;
}
static int device_close(struct inode *inode, struct file *file)
{
struct dlm_user_proc *proc = file->private_data;
struct dlm_ls *ls;
sigset_t tmpsig, allsigs;
ls = dlm_find_lockspace_local(proc->lockspace);
if (!ls)
return -ENOENT;
sigfillset(&allsigs);
sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
dlm_clear_proc_locks(ls, proc);
/* at this point no more lkb's should exist for this lockspace,
so there's no chance of dlm_user_add_ast() being called and
looking for lkb->ua->proc */
kfree(proc);
file->private_data = NULL;
dlm_put_lockspace(ls);
dlm_put_lockspace(ls); /* for the find in device_open() */
/* FIXME: AUTOFREE: if this ls is no longer used do
device_remove_lockspace() */
sigprocmask(SIG_SETMASK, &tmpsig, NULL);
recalc_sigpending();
return 0;
}
static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
int bmode, char __user *buf, size_t count)
{
#ifdef CONFIG_COMPAT
struct dlm_lock_result32 result32;
#endif
struct dlm_lock_result result;
void *resultptr;
int error=0;
int len;
int struct_len;
memset(&result, 0, sizeof(struct dlm_lock_result));
result.version[0] = DLM_DEVICE_VERSION_MAJOR;
result.version[1] = DLM_DEVICE_VERSION_MINOR;
result.version[2] = DLM_DEVICE_VERSION_PATCH;
memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
result.user_lksb = ua->user_lksb;
/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
in a conversion unless the conversion is successful. See code
in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
notes that a new blocking AST address and parameter are set even if
the conversion fails, so maybe we should just do that. */
if (type == AST_BAST) {
result.user_astaddr = ua->bastaddr;
result.user_astparam = ua->bastparam;
result.bast_mode = bmode;
} else {
result.user_astaddr = ua->castaddr;
result.user_astparam = ua->castparam;
}
#ifdef CONFIG_COMPAT
if (compat)
len = sizeof(struct dlm_lock_result32);
else
#endif
len = sizeof(struct dlm_lock_result);
struct_len = len;
/* copy lvb to userspace if there is one, it's been updated, and
the user buffer has space for it */
if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
count >= len + DLM_USER_LVB_LEN) {
if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
DLM_USER_LVB_LEN)) {
error = -EFAULT;
goto out;
}
result.lvb_offset = len;
len += DLM_USER_LVB_LEN;
}
result.length = len;
resultptr = &result;
#ifdef CONFIG_COMPAT
if (compat) {
compat_output(&result, &result32);
resultptr = &result32;
}
#endif
if (copy_to_user(buf, resultptr, struct_len))
error = -EFAULT;
else
error = len;
out:
return error;
}
static int copy_version_to_user(char __user *buf, size_t count)
{
struct dlm_device_version ver;
memset(&ver, 0, sizeof(struct dlm_device_version));
ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
ver.version[1] = DLM_DEVICE_VERSION_MINOR;
ver.version[2] = DLM_DEVICE_VERSION_PATCH;
if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
return -EFAULT;
return sizeof(struct dlm_device_version);
}
/* a read returns a single ast described in a struct dlm_lock_result */
static ssize_t device_read(struct file *file, char __user *buf, size_t count,
loff_t *ppos)
{
struct dlm_user_proc *proc = file->private_data;
struct dlm_lkb *lkb;
struct dlm_user_args *ua;
DECLARE_WAITQUEUE(wait, current);
int error, type=0, bmode=0, removed = 0;
if (count == sizeof(struct dlm_device_version)) {
error = copy_version_to_user(buf, count);
return error;
}
if (!proc) {
log_print("non-version read from control device %zu", count);
return -EINVAL;
}
#ifdef CONFIG_COMPAT
if (count < sizeof(struct dlm_lock_result32))
#else
if (count < sizeof(struct dlm_lock_result))
#endif
return -EINVAL;
/* do we really need this? can a read happen after a close? */
if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
return -EINVAL;
spin_lock(&proc->asts_spin);
if (list_empty(&proc->asts)) {
if (file->f_flags & O_NONBLOCK) {
spin_unlock(&proc->asts_spin);
return -EAGAIN;
}
add_wait_queue(&proc->wait, &wait);
repeat:
set_current_state(TASK_INTERRUPTIBLE);
if (list_empty(&proc->asts) && !signal_pending(current)) {
spin_unlock(&proc->asts_spin);
schedule();
spin_lock(&proc->asts_spin);
goto repeat;
}
set_current_state(TASK_RUNNING);
remove_wait_queue(&proc->wait, &wait);
if (signal_pending(current)) {
spin_unlock(&proc->asts_spin);
return -ERESTARTSYS;
}
}
/* there may be both completion and blocking asts to return for
the lkb, don't remove lkb from asts list unless no asts remain */
lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
if (lkb->lkb_ast_type & AST_COMP) {
lkb->lkb_ast_type &= ~AST_COMP;
type = AST_COMP;
} else if (lkb->lkb_ast_type & AST_BAST) {
lkb->lkb_ast_type &= ~AST_BAST;
type = AST_BAST;
bmode = lkb->lkb_bastmode;
}
if (!lkb->lkb_ast_type) {
list_del(&lkb->lkb_astqueue);
removed = 1;
}
spin_unlock(&proc->asts_spin);
ua = (struct dlm_user_args *)lkb->lkb_astparam;
error = copy_result_to_user(ua,
test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
type, bmode, buf, count);
/* removes reference for the proc->asts lists added by
dlm_user_add_ast() and may result in the lkb being freed */
if (removed)
dlm_put_lkb(lkb);
return error;
}
static unsigned int device_poll(struct file *file, poll_table *wait)
{
struct dlm_user_proc *proc = file->private_data;
poll_wait(file, &proc->wait, wait);
spin_lock(&proc->asts_spin);
if (!list_empty(&proc->asts)) {
spin_unlock(&proc->asts_spin);
return POLLIN | POLLRDNORM;
}
spin_unlock(&proc->asts_spin);
return 0;
}
static int ctl_device_open(struct inode *inode, struct file *file)
{
file->private_data = NULL;
return 0;
}
static int ctl_device_close(struct inode *inode, struct file *file)
{
return 0;
}
static const struct file_operations device_fops = {
.open = device_open,
.release = device_close,
.read = device_read,
.write = device_write,
.poll = device_poll,
.owner = THIS_MODULE,
};
static const struct file_operations ctl_device_fops = {
.open = ctl_device_open,
.release = ctl_device_close,
.read = device_read,
.write = device_write,
.owner = THIS_MODULE,
};
static struct miscdevice ctl_device = {
.name = "dlm-control",
.fops = &ctl_device_fops,
.minor = MISC_DYNAMIC_MINOR,
};
int dlm_user_init(void)
{
int error;
error = misc_register(&ctl_device);
if (error)
log_print("misc_register failed for control device");
return error;
}
void dlm_user_exit(void)
{
misc_deregister(&ctl_device);
}