diff options
Diffstat (limited to '')
| -rw-r--r-- | source4/ntvfs/common/brlock.c | 136 | ||||
| -rw-r--r-- | source4/ntvfs/common/brlock.h | 55 | ||||
| -rw-r--r-- | source4/ntvfs/common/brlock_tdb.c | 775 | ||||
| -rw-r--r-- | source4/ntvfs/common/init.c | 34 | ||||
| -rw-r--r-- | source4/ntvfs/common/notify.c | 687 | ||||
| -rw-r--r-- | source4/ntvfs/common/ntvfs_common.h | 32 | ||||
| -rw-r--r-- | source4/ntvfs/common/opendb.c | 200 | ||||
| -rw-r--r-- | source4/ntvfs/common/opendb.h | 59 | ||||
| -rw-r--r-- | source4/ntvfs/common/opendb_tdb.c | 886 | ||||
| -rw-r--r-- | source4/ntvfs/common/wscript_build | 9 |
10 files changed, 2873 insertions, 0 deletions
diff --git a/source4/ntvfs/common/brlock.c b/source4/ntvfs/common/brlock.c new file mode 100644 index 0000000..d99cc4e --- /dev/null +++ b/source4/ntvfs/common/brlock.c @@ -0,0 +1,136 @@ +/* + Unix SMB/CIFS implementation. + + generic byte range locking code + + Copyright (C) Andrew Tridgell 1992-2004 + Copyright (C) Jeremy Allison 1992-2000 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* This module implements a tdb based byte range locking service, + replacing the fcntl() based byte range locking previously + used. This allows us to provide the same semantics as NT */ + +#include "includes.h" +#include "system/filesys.h" +#include "messaging/messaging.h" +#include "lib/messaging/irpc.h" +#include "libcli/libcli.h" +#include "cluster/cluster.h" +#include "ntvfs/common/ntvfs_common.h" + +static const struct brlock_ops *ops; + +/* + set the brl backend ops +*/ +void brlock_set_ops(const struct brlock_ops *new_ops) +{ + ops = new_ops; +} + +/* + Open up the brlock database. Close it down using talloc_free(). We + need the imessaging_ctx to allow for pending lock notifications. +*/ +struct brl_context *brlock_init(TALLOC_CTX *mem_ctx, struct server_id server, + struct loadparm_context *lp_ctx, + struct imessaging_context *imessaging_ctx) +{ + if (ops == NULL) { + brl_tdb_init_ops(); + } + return ops->brl_init(mem_ctx, server, lp_ctx, imessaging_ctx); +} + +struct brl_handle *brlock_create_handle(TALLOC_CTX *mem_ctx, struct ntvfs_handle *ntvfs, DATA_BLOB *file_key) +{ + return ops->brl_create_handle(mem_ctx, ntvfs, file_key); +} + +/* + Lock a range of bytes. The lock_type can be a PENDING_*_LOCK, in + which case a real lock is first tried, and if that fails then a + pending lock is created. When the pending lock is triggered (by + someone else closing an overlapping lock range) a messaging + notification is sent, identified by the notify_ptr +*/ +NTSTATUS brlock_lock(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size, + enum brl_type lock_type, + void *notify_ptr) +{ + return ops->brl_lock(brl, brlh, smbpid, start, size, lock_type, notify_ptr); +} + + +/* + Unlock a range of bytes. +*/ +NTSTATUS brlock_unlock(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size) +{ + return ops->brl_unlock(brl, brlh, smbpid, start, size); +} + +/* + remove a pending lock. This is called when the caller has either + given up trying to establish a lock or when they have succeeded in + getting it. In either case they no longer need to be notified. +*/ +NTSTATUS brlock_remove_pending(struct brl_context *brl, + struct brl_handle *brlh, + void *notify_ptr) +{ + return ops->brl_remove_pending(brl, brlh, notify_ptr); +} + + +/* + Test if we are allowed to perform IO on a region of an open file +*/ +NTSTATUS brlock_locktest(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size, + enum brl_type lock_type) +{ + return ops->brl_locktest(brl, brlh, smbpid, start, size, lock_type); +} + + +/* + Remove any locks associated with a open file. +*/ +NTSTATUS brlock_close(struct brl_context *brl, + struct brl_handle *brlh) +{ + return ops->brl_close(brl, brlh); +} + +/* + Get a number of locks associated with a open file. +*/ +NTSTATUS brlock_count(struct brl_context *brl, + struct brl_handle *brlh, + int *count) +{ + return ops->brl_count(brl, brlh, count); +} diff --git a/source4/ntvfs/common/brlock.h b/source4/ntvfs/common/brlock.h new file mode 100644 index 0000000..650136b --- /dev/null +++ b/source4/ntvfs/common/brlock.h @@ -0,0 +1,55 @@ +/* + Unix SMB/CIFS implementation. + + generic byte range locking code - common include + + Copyright (C) Andrew Tridgell 2006 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include "libcli/libcli.h" + +struct brlock_ops { + struct brl_context *(*brl_init)(TALLOC_CTX *, struct server_id , + struct loadparm_context *lp_ctx, + struct imessaging_context *); + struct brl_handle *(*brl_create_handle)(TALLOC_CTX *, struct ntvfs_handle *, DATA_BLOB *); + NTSTATUS (*brl_lock)(struct brl_context *, + struct brl_handle *, + uint32_t , + uint64_t , uint64_t , + enum brl_type , + void *); + NTSTATUS (*brl_unlock)(struct brl_context *, + struct brl_handle *, + uint32_t , + uint64_t , uint64_t ); + NTSTATUS (*brl_remove_pending)(struct brl_context *, + struct brl_handle *, + void *); + NTSTATUS (*brl_locktest)(struct brl_context *, + struct brl_handle *, + uint32_t , + uint64_t , uint64_t , + enum brl_type ); + NTSTATUS (*brl_close)(struct brl_context *, + struct brl_handle *); + NTSTATUS (*brl_count)(struct brl_context *, + struct brl_handle *, + int *count); +}; + +void brlock_set_ops(const struct brlock_ops *new_ops); +void brl_tdb_init_ops(void); diff --git a/source4/ntvfs/common/brlock_tdb.c b/source4/ntvfs/common/brlock_tdb.c new file mode 100644 index 0000000..77a864a --- /dev/null +++ b/source4/ntvfs/common/brlock_tdb.c @@ -0,0 +1,775 @@ +/* + Unix SMB/CIFS implementation. + + generic byte range locking code - tdb backend + + Copyright (C) Andrew Tridgell 1992-2006 + Copyright (C) Jeremy Allison 1992-2000 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* This module implements a tdb based byte range locking service, + replacing the fcntl() based byte range locking previously + used. This allows us to provide the same semantics as NT */ + +#include "includes.h" +#include "system/filesys.h" +#include "messaging/messaging.h" +#include "lib/messaging/irpc.h" +#include "libcli/libcli.h" +#include "cluster/cluster.h" +#include "ntvfs/common/brlock.h" +#include "ntvfs/ntvfs.h" +#include "param/param.h" +#include "dbwrap/dbwrap.h" + +/* + in this module a "DATA_BLOB *file_key" is a blob that uniquely identifies + a file. For a local posix filesystem this will usually be a combination + of the device and inode numbers of the file, but it can be anything + that uniquely idetifies a file for locking purposes, as long + as it is applied consistently. +*/ + +/* this struct is typicaly attached to tcon */ +struct brl_context { + struct db_context *db; + struct server_id server; + struct imessaging_context *imessaging_ctx; +}; + +/* + the lock context contains the elements that define whether one + lock is the same as another lock +*/ +struct lock_context { + struct server_id server; + uint32_t smbpid; + struct brl_context *ctx; +}; + +/* The data in brlock records is an unsorted linear array of these + records. It is unnecessary to store the count as tdb provides the + size of the record */ +struct lock_struct { + struct lock_context context; + struct ntvfs_handle *ntvfs; + uint64_t start; + uint64_t size; + enum brl_type lock_type; + void *notify_ptr; +}; + +/* this struct is attached to on oprn file handle */ +struct brl_handle { + DATA_BLOB key; + struct ntvfs_handle *ntvfs; + struct lock_struct last_lock; +}; + +/* see if we have wrapped locks, which are no longer allowed (windows + * changed this in win7 */ +static bool brl_invalid_lock_range(uint64_t start, uint64_t size) +{ + return (size > 1 && (start + size < start)); +} + +/* + Open up the brlock.tdb database. Close it down using + talloc_free(). We need the imessaging_ctx to allow for + pending lock notifications. +*/ +static struct brl_context *brl_tdb_init(TALLOC_CTX *mem_ctx, struct server_id server, + struct loadparm_context *lp_ctx, + struct imessaging_context *imessaging_ctx) +{ + struct brl_context *brl; + + brl = talloc(mem_ctx, struct brl_context); + if (brl == NULL) { + return NULL; + } + + brl->db = cluster_db_tmp_open(brl, lp_ctx, "brlock", TDB_DEFAULT); + if (brl->db == NULL) { + talloc_free(brl); + return NULL; + } + + brl->server = server; + brl->imessaging_ctx = imessaging_ctx; + + return brl; +} + +static struct brl_handle *brl_tdb_create_handle(TALLOC_CTX *mem_ctx, struct ntvfs_handle *ntvfs, + DATA_BLOB *file_key) +{ + struct brl_handle *brlh; + + brlh = talloc(mem_ctx, struct brl_handle); + if (brlh == NULL) { + return NULL; + } + + brlh->key = *file_key; + brlh->ntvfs = ntvfs; + ZERO_STRUCT(brlh->last_lock); + + return brlh; +} + +/* + see if two locking contexts are equal +*/ +static bool brl_tdb_same_context(struct lock_context *ctx1, struct lock_context *ctx2) +{ + return (cluster_id_equal(&ctx1->server, &ctx2->server) && + ctx1->smbpid == ctx2->smbpid && + ctx1->ctx == ctx2->ctx); +} + +/* + see if lck1 and lck2 overlap + + lck1 is the existing lock. lck2 is the new lock we are + looking at adding +*/ +static bool brl_tdb_overlap(struct lock_struct *lck1, + struct lock_struct *lck2) +{ + /* this extra check is not redundant - it copes with locks + that go beyond the end of 64 bit file space */ + if (lck1->size != 0 && + lck1->start == lck2->start && + lck1->size == lck2->size) { + return true; + } + + if (lck1->start >= (lck2->start+lck2->size) || + lck2->start >= (lck1->start+lck1->size)) { + return false; + } + + /* we have a conflict. Now check to see if lck1 really still + * exists, which involves checking if the process still + * exists. We leave this test to last as its the most + * expensive test, especially when we are clustered */ + /* TODO: need to do this via a server_id_exists() call, which + * hasn't been written yet. When clustered this will need to + * call into ctdb */ + + return true; +} + +/* + See if lock2 can be added when lock1 is in place. +*/ +static bool brl_tdb_conflict(struct lock_struct *lck1, + struct lock_struct *lck2) +{ + /* pending locks don't conflict with anything */ + if (lck1->lock_type >= PENDING_READ_LOCK || + lck2->lock_type >= PENDING_READ_LOCK) { + return false; + } + + if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) { + return false; + } + + if (brl_tdb_same_context(&lck1->context, &lck2->context) && + lck2->lock_type == READ_LOCK && lck1->ntvfs == lck2->ntvfs) { + return false; + } + + return brl_tdb_overlap(lck1, lck2); +} + + +/* + Check to see if this lock conflicts, but ignore our own locks on the + same fnum only. +*/ +static bool brl_tdb_conflict_other(struct lock_struct *lck1, struct lock_struct *lck2) +{ + /* pending locks don't conflict with anything */ + if (lck1->lock_type >= PENDING_READ_LOCK || + lck2->lock_type >= PENDING_READ_LOCK) { + return false; + } + + if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) + return false; + + /* + * note that incoming write calls conflict with existing READ + * locks even if the context is the same. JRA. See LOCKTEST7 + * in smbtorture. + */ + if (brl_tdb_same_context(&lck1->context, &lck2->context) && + lck1->ntvfs == lck2->ntvfs && + (lck2->lock_type == READ_LOCK || lck1->lock_type == WRITE_LOCK)) { + return false; + } + + return brl_tdb_overlap(lck1, lck2); +} + + +/* + amazingly enough, w2k3 "remembers" whether the last lock failure + is the same as this one and changes its error code. I wonder if any + app depends on this? +*/ +static NTSTATUS brl_tdb_lock_failed(struct brl_handle *brlh, struct lock_struct *lock) +{ + /* + * this function is only called for non pending lock! + */ + + /* in SMB2 mode always return NT_STATUS_LOCK_NOT_GRANTED! */ + if (lock->ntvfs->ctx->protocol >= PROTOCOL_SMB2_02) { + return NT_STATUS_LOCK_NOT_GRANTED; + } + + /* + * if the notify_ptr is non NULL, + * it means that we're at the end of a pending lock + * and the real lock is requested after the timout went by + * In this case we need to remember the last_lock and always + * give FILE_LOCK_CONFLICT + */ + if (lock->notify_ptr) { + brlh->last_lock = *lock; + return NT_STATUS_FILE_LOCK_CONFLICT; + } + + /* + * 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 (lock->start >= 0xEF000000 && (lock->start >> 63) == 0) { + brlh->last_lock = *lock; + return NT_STATUS_FILE_LOCK_CONFLICT; + } + + /* + * if the current lock matches the last failed lock on the file handle + * and starts at the same offset, then FILE_LOCK_CONFLICT should be returned + */ + if (cluster_id_equal(&lock->context.server, &brlh->last_lock.context.server) && + lock->context.ctx == brlh->last_lock.context.ctx && + lock->ntvfs == brlh->last_lock.ntvfs && + lock->start == brlh->last_lock.start) { + return NT_STATUS_FILE_LOCK_CONFLICT; + } + + brlh->last_lock = *lock; + return NT_STATUS_LOCK_NOT_GRANTED; +} + +/* + Lock a range of bytes. The lock_type can be a PENDING_*_LOCK, in + which case a real lock is first tried, and if that fails then a + pending lock is created. When the pending lock is triggered (by + someone else closing an overlapping lock range) a messaging + notification is sent, identified by the notify_ptr +*/ +static NTSTATUS brl_tdb_lock(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size, + enum brl_type lock_type, + void *notify_ptr) +{ + TDB_DATA kbuf, dbuf; + int count=0, i; + struct lock_struct lock, *locks=NULL; + NTSTATUS status; + struct db_record *locked; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + + if (brl_invalid_lock_range(start, size)) { + return NT_STATUS_INVALID_LOCK_RANGE; + } + + locked = dbwrap_fetch_locked(brl->db, brl, kbuf); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + /* if this is a pending lock, then with the chainlock held we + try to get the real lock. If we succeed then we don't need + to make it pending. This prevents a possible race condition + where the pending lock gets created after the lock that is + preventing the real lock gets removed */ + if (lock_type >= PENDING_READ_LOCK) { + enum brl_type rw = (lock_type==PENDING_READ_LOCK? READ_LOCK : WRITE_LOCK); + + /* here we need to force that the last_lock isn't overwritten */ + lock = brlh->last_lock; + status = brl_tdb_lock(brl, brlh, smbpid, start, size, rw, NULL); + brlh->last_lock = lock; + + if (NT_STATUS_IS_OK(status)) { + talloc_free(locked); + return NT_STATUS_OK; + } + } + + dbuf = dbwrap_record_get_value(locked); + + lock.context.smbpid = smbpid; + lock.context.server = brl->server; + lock.context.ctx = brl; + lock.ntvfs = brlh->ntvfs; + lock.context.ctx = brl; + lock.start = start; + lock.size = size; + lock.lock_type = lock_type; + lock.notify_ptr = notify_ptr; + + if (dbuf.dptr) { + /* there are existing locks - make sure they don't conflict */ + locks = (struct lock_struct *)dbuf.dptr; + count = dbuf.dsize / sizeof(*locks); + for (i=0; i<count; i++) { + if (brl_tdb_conflict(&locks[i], &lock)) { + status = brl_tdb_lock_failed(brlh, &lock); + goto fail; + } + } + } + + /* no conflicts - add it to the list of locks */ + /* FIXME: a dbwrap_record_append() would help here! */ + locks = talloc_array(locked, struct lock_struct, count+1); + if (!locks) { + status = NT_STATUS_NO_MEMORY; + goto fail; + } + if (dbuf.dsize > 0) { + memcpy(locks, dbuf.dptr, dbuf.dsize); + } + locks[count] = lock; + + dbuf.dptr = (unsigned char *)locks; + dbuf.dsize += sizeof(lock); + + status = dbwrap_record_store(locked, dbuf, TDB_REPLACE); + if (!NT_STATUS_IS_OK(status)) { + goto fail; + } + + talloc_free(locked); + + /* the caller needs to know if the real lock was granted. If + we have reached here then it must be a pending lock that + was granted, so tell them the lock failed */ + if (lock_type >= PENDING_READ_LOCK) { + return NT_STATUS_LOCK_NOT_GRANTED; + } + + return NT_STATUS_OK; + + fail: + talloc_free(locked); + return status; +} + + +/* + we are removing a lock that might be holding up a pending lock. Scan for pending + locks that cover this range and if we find any then notify the server that it should + retry the lock +*/ +static void brl_tdb_notify_unlock(struct brl_context *brl, + struct lock_struct *locks, int count, + struct lock_struct *removed_lock) +{ + int i, last_notice; + + /* the last_notice logic is to prevent stampeding on a lock + range. It prevents us sending hundreds of notifies on the + same range of bytes. It doesn't prevent all possible + stampedes, but it does prevent the most common problem */ + last_notice = -1; + + for (i=0;i<count;i++) { + if (locks[i].lock_type >= PENDING_READ_LOCK && + brl_tdb_overlap(&locks[i], removed_lock)) { + if (last_notice != -1 && brl_tdb_overlap(&locks[i], &locks[last_notice])) { + continue; + } + if (locks[i].lock_type == PENDING_WRITE_LOCK) { + last_notice = i; + } + imessaging_send_ptr(brl->imessaging_ctx, locks[i].context.server, + MSG_BRL_RETRY, locks[i].notify_ptr); + } + } +} + + +/* + send notifications for all pending locks - the file is being closed by this + user +*/ +static void brl_tdb_notify_all(struct brl_context *brl, + struct lock_struct *locks, int count) +{ + int i; + for (i=0;i<count;i++) { + if (locks->lock_type >= PENDING_READ_LOCK) { + brl_tdb_notify_unlock(brl, locks, count, &locks[i]); + } + } +} + + + +/* + Unlock a range of bytes. +*/ +static NTSTATUS brl_tdb_unlock(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size) +{ + TDB_DATA kbuf, dbuf; + int count, i; + struct lock_struct *locks, *lock; + struct lock_context context; + struct db_record *locked; + NTSTATUS status; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + + if (brl_invalid_lock_range(start, size)) { + return NT_STATUS_INVALID_LOCK_RANGE; + } + + locked = dbwrap_fetch_locked(brl->db, brl, kbuf); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + dbuf = dbwrap_record_get_value(locked); + + context.smbpid = smbpid; + context.server = brl->server; + context.ctx = brl; + + /* there are existing locks - find a match */ + locks = (struct lock_struct *)dbuf.dptr; + count = dbuf.dsize / sizeof(*locks); + + for (i=0; i<count; i++) { + lock = &locks[i]; + if (brl_tdb_same_context(&lock->context, &context) && + lock->ntvfs == brlh->ntvfs && + lock->start == start && + lock->size == size && + lock->lock_type == WRITE_LOCK) { + break; + } + } + if (i < count) goto found; + + for (i=0; i<count; i++) { + lock = &locks[i]; + if (brl_tdb_same_context(&lock->context, &context) && + lock->ntvfs == brlh->ntvfs && + lock->start == start && + lock->size == size && + lock->lock_type < PENDING_READ_LOCK) { + break; + } + } + +found: + if (i < count) { + /* found it - delete it */ + if (count == 1) { + status = dbwrap_record_delete(locked); + if (!NT_STATUS_IS_OK(status)) { + goto fail; + } + } else { + struct lock_struct removed_lock = *lock; + if (i < count-1) { + memmove(&locks[i], &locks[i+1], + sizeof(*locks)*((count-1) - i)); + } + count--; + + /* send notifications for any relevant pending locks */ + brl_tdb_notify_unlock(brl, locks, count, &removed_lock); + + dbuf.dsize = count * sizeof(*locks); + + status = dbwrap_record_store(locked, dbuf, TDB_REPLACE); + if (!NT_STATUS_IS_OK(status)) { + goto fail; + } + } + + talloc_free(locked); + return NT_STATUS_OK; + } + + /* we didn't find it */ + status = NT_STATUS_RANGE_NOT_LOCKED; + + fail: + talloc_free(locked); + return status; +} + + +/* + remove a pending lock. This is called when the caller has either + given up trying to establish a lock or when they have succeeded in + getting it. In either case they no longer need to be notified. +*/ +static NTSTATUS brl_tdb_remove_pending(struct brl_context *brl, + struct brl_handle *brlh, + void *notify_ptr) +{ + TDB_DATA kbuf, dbuf; + int count, i; + struct lock_struct *locks; + NTSTATUS status; + struct db_record *locked; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + + locked = dbwrap_fetch_locked(brl->db, brl, kbuf); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + dbuf = dbwrap_record_get_value(locked); + if (!dbuf.dptr) { + talloc_free(locked); + return NT_STATUS_RANGE_NOT_LOCKED; + } + + /* there are existing locks - find a match */ + locks = (struct lock_struct *)dbuf.dptr; + count = dbuf.dsize / sizeof(*locks); + + for (i=0; i<count; i++) { + struct lock_struct *lock = &locks[i]; + + if (lock->lock_type >= PENDING_READ_LOCK && + lock->notify_ptr == notify_ptr && + cluster_id_equal(&lock->context.server, &brl->server)) { + /* found it - delete it */ + if (count == 1) { + status = dbwrap_record_delete(locked); + if (!NT_STATUS_IS_OK(status)) { + goto fail; + } + } else { + if (i < count-1) { + memmove(&locks[i], &locks[i+1], + sizeof(*locks)*((count-1) - i)); + } + count--; + dbuf.dsize = count * sizeof(*locks); + status = dbwrap_record_store(locked, dbuf, + TDB_REPLACE); + if (!NT_STATUS_IS_OK(status)) { + goto fail; + } + } + + talloc_free(locked); + return NT_STATUS_OK; + } + } + + /* we didn't find it */ + status = NT_STATUS_RANGE_NOT_LOCKED; + + fail: + talloc_free(locked); + return status; +} + + +/* + Test if we are allowed to perform IO on a region of an open file +*/ +static NTSTATUS brl_tdb_locktest(struct brl_context *brl, + struct brl_handle *brlh, + uint32_t smbpid, + uint64_t start, uint64_t size, + enum brl_type lock_type) +{ + TDB_DATA kbuf, dbuf; + int count, i; + struct lock_struct lock, *locks; + NTSTATUS status; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + + if (brl_invalid_lock_range(start, size)) { + return NT_STATUS_INVALID_LOCK_RANGE; + } + + status = dbwrap_fetch(brl->db, brl, kbuf, &dbuf); + if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + return NT_STATUS_OK; + } else if (!NT_STATUS_IS_OK(status)) { + return status; + } + + lock.context.smbpid = smbpid; + lock.context.server = brl->server; + lock.context.ctx = brl; + lock.ntvfs = brlh->ntvfs; + lock.start = start; + lock.size = size; + lock.lock_type = lock_type; + + /* there are existing locks - make sure they don't conflict */ + locks = (struct lock_struct *)dbuf.dptr; + count = dbuf.dsize / sizeof(*locks); + + for (i=0; i<count; i++) { + if (brl_tdb_conflict_other(&locks[i], &lock)) { + talloc_free(dbuf.dptr); + return NT_STATUS_FILE_LOCK_CONFLICT; + } + } + + talloc_free(dbuf.dptr); + return NT_STATUS_OK; +} + + +/* + Remove any locks associated with a open file. +*/ +static NTSTATUS brl_tdb_close(struct brl_context *brl, + struct brl_handle *brlh) +{ + TDB_DATA kbuf, dbuf; + int count, i, dcount=0; + struct lock_struct *locks; + struct db_record *locked; + NTSTATUS status; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + + locked = dbwrap_fetch_locked(brl->db, brl, kbuf); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + dbuf = dbwrap_record_get_value(locked); + if (!dbuf.dptr) { + talloc_free(locked); + return NT_STATUS_OK; + } + + /* there are existing locks - remove any for this fnum */ + locks = (struct lock_struct *)dbuf.dptr; + count = dbuf.dsize / sizeof(*locks); + + for (i=0; i<count; i++) { + struct lock_struct *lock = &locks[i]; + + if (lock->context.ctx == brl && + cluster_id_equal(&lock->context.server, &brl->server) && + lock->ntvfs == brlh->ntvfs) { + /* found it - delete it */ + if (count > 1 && i < count-1) { + memmove(&locks[i], &locks[i+1], + sizeof(*locks)*((count-1) - i)); + } + count--; + i--; + dcount++; + } + } + + status = NT_STATUS_OK; + + if (count == 0) { + status = dbwrap_record_delete(locked); + } else if (dcount != 0) { + /* tell all pending lock holders for this file that + they have a chance now. This is a bit indiscriminant, + but works OK */ + brl_tdb_notify_all(brl, locks, count); + + dbuf.dsize = count * sizeof(*locks); + + status = dbwrap_record_store(locked, dbuf, TDB_REPLACE); + } + talloc_free(locked); + + return status; +} + +static NTSTATUS brl_tdb_count(struct brl_context *brl, struct brl_handle *brlh, + int *count) +{ + TDB_DATA kbuf, dbuf; + NTSTATUS status; + + kbuf.dptr = brlh->key.data; + kbuf.dsize = brlh->key.length; + *count = 0; + + status = dbwrap_fetch(brl->db, brl, kbuf, &dbuf); + if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) { + return NT_STATUS_OK; + } else if (!NT_STATUS_IS_OK(status)) { + return status; + } + *count = dbuf.dsize / sizeof(struct lock_struct); + + talloc_free(dbuf.dptr); + + return NT_STATUS_OK; +} + +static const struct brlock_ops brlock_tdb_ops = { + .brl_init = brl_tdb_init, + .brl_create_handle = brl_tdb_create_handle, + .brl_lock = brl_tdb_lock, + .brl_unlock = brl_tdb_unlock, + .brl_remove_pending = brl_tdb_remove_pending, + .brl_locktest = brl_tdb_locktest, + .brl_close = brl_tdb_close, + .brl_count = brl_tdb_count +}; + + +void brl_tdb_init_ops(void) +{ + brlock_set_ops(&brlock_tdb_ops); +} diff --git a/source4/ntvfs/common/init.c b/source4/ntvfs/common/init.c new file mode 100644 index 0000000..f8f8e27 --- /dev/null +++ b/source4/ntvfs/common/init.c @@ -0,0 +1,34 @@ +/* + Unix SMB/CIFS implementation. + + Copyright (C) Stefan Metzmacher 2006 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* + this is the change notify database. It implements mechanisms for + storing current change notify waiters in a tdb, and checking if a + given event matches any of the stored notify waiiters. +*/ + +#include "includes.h" +#include "ntvfs/sysdep/sys_notify.h" + +NTSTATUS ntvfs_common_init(void); + +NTSTATUS ntvfs_common_init(void) +{ + return sys_notify_init(); +} diff --git a/source4/ntvfs/common/notify.c b/source4/ntvfs/common/notify.c new file mode 100644 index 0000000..a434fab --- /dev/null +++ b/source4/ntvfs/common/notify.c @@ -0,0 +1,687 @@ +/* + Unix SMB/CIFS implementation. + + Copyright (C) Andrew Tridgell 2006 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* + this is the change notify database. It implements mechanisms for + storing current change notify waiters in a tdb, and checking if a + given event matches any of the stored notify waiiters. +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "messaging/messaging.h" +#include "lib/messaging/irpc.h" +#include "librpc/gen_ndr/ndr_notify.h" +#include "../lib/util/dlinklist.h" +#include "ntvfs/common/ntvfs_common.h" +#include "ntvfs/sysdep/sys_notify.h" +#include "cluster/cluster.h" +#include "param/param.h" +#include "lib/util/tsort.h" +#include "lib/dbwrap/dbwrap.h" +#include "../lib/util/util_tdb.h" + +struct notify_context { + struct db_context *db; + struct server_id server; + struct imessaging_context *imessaging_ctx; + struct notify_list *list; + struct notify_array *array; + int64_t seqnum; + struct sys_notify_context *sys_notify_ctx; +}; + + +struct notify_list { + struct notify_list *next, *prev; + void *private_data; + void (*callback)(void *, const struct notify_event *); + void *sys_notify_handle; + int depth; +}; + +#define NOTIFY_KEY "notify array" + +#define NOTIFY_ENABLE "notify:enable" +#define NOTIFY_ENABLE_DEFAULT true + +static NTSTATUS notify_remove_all(struct notify_context *notify); +static void notify_handler(struct imessaging_context *msg_ctx, + void *private_data, + uint32_t msg_type, + struct server_id server_id, + size_t num_fds, + int *fds, + DATA_BLOB *data); + +/* + destroy the notify context +*/ +static int notify_destructor(struct notify_context *notify) +{ + imessaging_deregister(notify->imessaging_ctx, MSG_PVFS_NOTIFY, notify); + notify_remove_all(notify); + return 0; +} + +/* + Open up the notify.tdb database. You should close it down using + talloc_free(). We need the imessaging_ctx to allow for notifications + via internal messages +*/ +struct notify_context *notify_init(TALLOC_CTX *mem_ctx, struct server_id server, + struct imessaging_context *imessaging_ctx, + struct loadparm_context *lp_ctx, + struct tevent_context *ev, + struct share_config *scfg) +{ + struct notify_context *notify; + + if (share_bool_option(scfg, NOTIFY_ENABLE, NOTIFY_ENABLE_DEFAULT) != true) { + return NULL; + } + + if (ev == NULL) { + return NULL; + } + + notify = talloc(mem_ctx, struct notify_context); + if (notify == NULL) { + return NULL; + } + + notify->db = cluster_db_tmp_open(notify, lp_ctx, "notify", TDB_SEQNUM); + if (notify->db == NULL) { + talloc_free(notify); + return NULL; + } + + notify->server = server; + notify->imessaging_ctx = imessaging_ctx; + notify->list = NULL; + notify->array = NULL; + notify->seqnum = dbwrap_get_seqnum(notify->db); + + talloc_set_destructor(notify, notify_destructor); + + /* register with the messaging subsystem for the notify + message type */ + imessaging_register(notify->imessaging_ctx, notify, + MSG_PVFS_NOTIFY, notify_handler); + + notify->sys_notify_ctx = sys_notify_context_create(scfg, notify, ev); + + return notify; +} + + +/* + lock the notify db +*/ +static struct db_record *notify_lock(struct notify_context *notify) +{ + TDB_DATA key = string_term_tdb_data(NOTIFY_KEY); + + return dbwrap_fetch_locked(notify->db, notify, key); +} + +static void notify_unlock(struct db_record *lock) +{ + talloc_free(lock); +} + +/* + load the notify array +*/ +static NTSTATUS notify_load(struct notify_context *notify) +{ + TDB_DATA dbuf; + DATA_BLOB blob; + enum ndr_err_code ndr_err; + int seqnum; + NTSTATUS status; + + seqnum = dbwrap_get_seqnum(notify->db); + + if (seqnum == notify->seqnum && notify->array != NULL) { + return NT_STATUS_OK; + } + + notify->seqnum = seqnum; + + talloc_free(notify->array); + notify->array = talloc_zero(notify, struct notify_array); + NT_STATUS_HAVE_NO_MEMORY(notify->array); + + status = dbwrap_fetch_bystring(notify->db, notify, NOTIFY_KEY, &dbuf); + if (!NT_STATUS_IS_OK(status)) { + return NT_STATUS_OK; + } + + blob.data = dbuf.dptr; + blob.length = dbuf.dsize; + + ndr_err = ndr_pull_struct_blob(&blob, notify->array, notify->array, + (ndr_pull_flags_fn_t)ndr_pull_notify_array); + talloc_free(dbuf.dptr); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + return ndr_map_error2ntstatus(ndr_err); + } + + return NT_STATUS_OK; +} + +/* + compare notify entries for sorting +*/ +static int notify_compare(const void *p1, const void *p2) +{ + const struct notify_entry *e1 = p1, *e2 = p2; + return strcmp(e1->path, e2->path); +} + +/* + save the notify array +*/ +static NTSTATUS notify_save(struct notify_context *notify) +{ + TDB_DATA dbuf; + DATA_BLOB blob; + enum ndr_err_code ndr_err; + TALLOC_CTX *tmp_ctx; + NTSTATUS status; + + /* if possible, remove some depth arrays */ + while (notify->array->num_depths > 0 && + notify->array->depth[notify->array->num_depths-1].num_entries == 0) { + notify->array->num_depths--; + } + + /* we might just be able to delete the record */ + if (notify->array->num_depths == 0) { + return dbwrap_delete_bystring(notify->db, NOTIFY_KEY); + } + + tmp_ctx = talloc_new(notify); + NT_STATUS_HAVE_NO_MEMORY(tmp_ctx); + + ndr_err = ndr_push_struct_blob(&blob, tmp_ctx, notify->array, + (ndr_push_flags_fn_t)ndr_push_notify_array); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + talloc_free(tmp_ctx); + return ndr_map_error2ntstatus(ndr_err); + } + + dbuf.dptr = blob.data; + dbuf.dsize = blob.length; + + status = dbwrap_store_bystring(notify->db, NOTIFY_KEY, dbuf, + TDB_REPLACE); + talloc_free(tmp_ctx); + return status; +} + + +/* + handle incoming notify messages +*/ +static void notify_handler(struct imessaging_context *msg_ctx, + void *private_data, + uint32_t msg_type, + struct server_id server_id, + size_t num_fds, + int *fds, + DATA_BLOB *data) +{ + struct notify_context *notify = talloc_get_type(private_data, struct notify_context); + enum ndr_err_code ndr_err; + struct notify_event ev; + TALLOC_CTX *tmp_ctx = talloc_new(notify); + struct notify_list *listel; + + if (num_fds != 0) { + DBG_WARNING("Received %zu fds, ignoring message\n", num_fds); + return; + } + + if (tmp_ctx == NULL) { + return; + } + + ndr_err = ndr_pull_struct_blob(data, tmp_ctx, &ev, + (ndr_pull_flags_fn_t)ndr_pull_notify_event); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + talloc_free(tmp_ctx); + return; + } + + for (listel=notify->list;listel;listel=listel->next) { + if (listel->private_data == ev.private_data) { + listel->callback(listel->private_data, &ev); + break; + } + } + + talloc_free(tmp_ctx); +} + +/* + callback from sys_notify telling us about changes from the OS +*/ +static void sys_notify_callback(struct sys_notify_context *ctx, + void *ptr, struct notify_event *ev) +{ + struct notify_list *listel = talloc_get_type(ptr, struct notify_list); + ev->private_data = listel; + listel->callback(listel->private_data, ev); +} + +/* + add an entry to the notify array +*/ +static NTSTATUS notify_add_array(struct notify_context *notify, struct notify_entry *e, + void *private_data, int depth) +{ + int i; + struct notify_depth *d; + struct notify_entry *ee; + + /* possibly expand the depths array */ + if (depth >= notify->array->num_depths) { + d = talloc_realloc(notify->array, notify->array->depth, + struct notify_depth, depth+1); + NT_STATUS_HAVE_NO_MEMORY(d); + for (i=notify->array->num_depths;i<=depth;i++) { + ZERO_STRUCT(d[i]); + } + notify->array->depth = d; + notify->array->num_depths = depth+1; + } + d = ¬ify->array->depth[depth]; + + /* expand the entries array */ + ee = talloc_realloc(notify->array->depth, d->entries, struct notify_entry, + d->num_entries+1); + NT_STATUS_HAVE_NO_MEMORY(ee); + d->entries = ee; + + d->entries[d->num_entries] = *e; + d->entries[d->num_entries].private_data = private_data; + d->entries[d->num_entries].server = notify->server; + d->entries[d->num_entries].path_len = strlen(e->path); + d->num_entries++; + + d->max_mask |= e->filter; + d->max_mask_subdir |= e->subdir_filter; + + TYPESAFE_QSORT(d->entries, d->num_entries, notify_compare); + + /* recalculate the maximum masks */ + d->max_mask = 0; + d->max_mask_subdir = 0; + + for (i=0;i<d->num_entries;i++) { + d->max_mask |= d->entries[i].filter; + d->max_mask_subdir |= d->entries[i].subdir_filter; + } + + return notify_save(notify); +} + +/* + add a notify watch. This is called when a notify is first setup on a open + directory handle. +*/ +NTSTATUS notify_add(struct notify_context *notify, struct notify_entry *e0, + void (*callback)(void *, const struct notify_event *), + void *private_data) +{ + struct notify_entry e = *e0; + NTSTATUS status; + char *tmp_path = NULL; + struct notify_list *listel; + size_t len; + int depth; + struct db_record *locked; + + /* see if change notify is enabled at all */ + if (notify == NULL) { + return NT_STATUS_NOT_IMPLEMENTED; + } + + locked = notify_lock(notify); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + status = notify_load(notify); + if (!NT_STATUS_IS_OK(status)) { + goto done; + } + + /* cope with /. on the end of the path */ + len = strlen(e.path); + if (len > 1 && e.path[len-1] == '.' && e.path[len-2] == '/') { + tmp_path = talloc_strndup(notify, e.path, len-2); + if (tmp_path == NULL) { + status = NT_STATUS_NO_MEMORY; + goto done; + } + e.path = tmp_path; + } + + depth = count_chars(e.path, '/'); + + listel = talloc_zero(notify, struct notify_list); + if (listel == NULL) { + status = NT_STATUS_NO_MEMORY; + goto done; + } + + listel->private_data = private_data; + listel->callback = callback; + listel->depth = depth; + DLIST_ADD(notify->list, listel); + + /* ignore failures from sys_notify */ + if (notify->sys_notify_ctx != NULL) { + /* + this call will modify e.filter and e.subdir_filter + to remove bits handled by the backend + */ + status = sys_notify_watch(notify->sys_notify_ctx, &e, + sys_notify_callback, listel, + &listel->sys_notify_handle); + if (NT_STATUS_IS_OK(status)) { + talloc_steal(listel, listel->sys_notify_handle); + } + } + + /* if the system notify handler couldn't handle some of the + filter bits, or couldn't handle a request for recursion + then we need to install it in the array used for the + intra-samba notify handling */ + if (e.filter != 0 || e.subdir_filter != 0) { + status = notify_add_array(notify, &e, private_data, depth); + } + +done: + notify_unlock(locked); + talloc_free(tmp_path); + + return status; +} + +/* + remove a notify watch. Called when the directory handle is closed +*/ +NTSTATUS notify_remove(struct notify_context *notify, void *private_data) +{ + NTSTATUS status; + struct notify_list *listel; + int i, depth; + struct notify_depth *d; + struct db_record *locked; + + /* see if change notify is enabled at all */ + if (notify == NULL) { + return NT_STATUS_NOT_IMPLEMENTED; + } + + for (listel=notify->list;listel;listel=listel->next) { + if (listel->private_data == private_data) { + DLIST_REMOVE(notify->list, listel); + break; + } + } + if (listel == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + depth = listel->depth; + + talloc_free(listel); + + locked = notify_lock(notify); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + status = notify_load(notify); + if (!NT_STATUS_IS_OK(status)) { + notify_unlock(locked); + return status; + } + + if (depth >= notify->array->num_depths) { + notify_unlock(locked); + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + /* we only have to search at the depth of this element */ + d = ¬ify->array->depth[depth]; + + for (i=0;i<d->num_entries;i++) { + if (private_data == d->entries[i].private_data && + cluster_id_equal(¬ify->server, &d->entries[i].server)) { + break; + } + } + if (i == d->num_entries) { + notify_unlock(locked); + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + if (i < d->num_entries-1) { + memmove(&d->entries[i], &d->entries[i+1], + sizeof(d->entries[i])*(d->num_entries-(i+1))); + } + d->num_entries--; + + status = notify_save(notify); + + notify_unlock(locked); + + return status; +} + +/* + remove all notify watches for this messaging server +*/ +static NTSTATUS notify_remove_all(struct notify_context *notify) +{ + NTSTATUS status; + int i, depth, del_count=0; + struct db_record *locked; + + if (notify->list == NULL) { + return NT_STATUS_OK; + } + + locked = notify_lock(notify); + if (!locked) { + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + status = notify_load(notify); + if (!NT_STATUS_IS_OK(status)) { + notify_unlock(locked); + return status; + } + + /* we have to search for all entries across all depths, looking for matches + for our server id */ + for (depth=0;depth<notify->array->num_depths;depth++) { + struct notify_depth *d = ¬ify->array->depth[depth]; + for (i=0;i<d->num_entries;i++) { + if (cluster_id_equal(¬ify->server, &d->entries[i].server)) { + if (i < d->num_entries-1) { + memmove(&d->entries[i], &d->entries[i+1], + sizeof(d->entries[i])*(d->num_entries-(i+1))); + } + i--; + d->num_entries--; + del_count++; + } + } + } + + if (del_count > 0) { + status = notify_save(notify); + } + + notify_unlock(locked); + + return status; +} + + +/* + send a notify message to another messaging server +*/ +static void notify_send(struct notify_context *notify, struct notify_entry *e, + const char *path, uint32_t action) +{ + struct notify_event ev; + DATA_BLOB data; + NTSTATUS status; + enum ndr_err_code ndr_err; + TALLOC_CTX *tmp_ctx; + + ev.action = action; + ev.dir = discard_const_p(char, ""); + ev.path = path; + ev.private_data = e->private_data; + + tmp_ctx = talloc_new(notify); + + ndr_err = ndr_push_struct_blob(&data, tmp_ctx, &ev, (ndr_push_flags_fn_t)ndr_push_notify_event); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + talloc_free(tmp_ctx); + return; + } + + status = imessaging_send(notify->imessaging_ctx, e->server, + MSG_PVFS_NOTIFY, &data); + if (!NT_STATUS_IS_OK(status)) { + talloc_free(tmp_ctx); + return; + } + + talloc_free(tmp_ctx); +} + + +/* + trigger a notify message for anyone waiting on a matching event + + This function is called a lot, and needs to be very fast. The unusual data structure + and traversal is designed to be fast in the average case, even for large numbers of + notifies +*/ +void notify_trigger(struct notify_context *notify, + uint32_t action, uint32_t filter, const char *path) +{ + NTSTATUS status; + int depth; + const char *p, *next_p; + + /* see if change notify is enabled at all */ + if (notify == NULL) { + return; + } + + status = notify_load(notify); + if (!NT_STATUS_IS_OK(status)) { + return; + } + + /* loop along the given path, working with each directory depth separately */ + for (depth=0,p=path; + p && depth < notify->array->num_depths; + p=next_p,depth++) { + int p_len = p - path; + int min_i, max_i, i; + struct notify_depth *d = ¬ify->array->depth[depth]; + next_p = strchr(p+1, '/'); + + /* see if there are any entries at this depth */ + if (d->num_entries == 0) continue; + + /* try to skip based on the maximum mask. If next_p is + NULL then we know it will be a 'this directory' + match, otherwise it must be a subdir match */ + if (next_p != NULL) { + if (0 == (filter & d->max_mask_subdir)) { + continue; + } + } else { + if (0 == (filter & d->max_mask)) { + continue; + } + } + + /* we know there is an entry here worth looking + for. Use a bisection search to find the first entry + with a matching path */ + min_i = 0; + max_i = d->num_entries-1; + + while (min_i < max_i) { + struct notify_entry *e; + int cmp; + i = (min_i+max_i)/2; + e = &d->entries[i]; + cmp = strncmp(path, e->path, p_len); + if (cmp == 0) { + if (p_len == e->path_len) { + max_i = i; + } else { + max_i = i-1; + } + } else if (cmp < 0) { + max_i = i-1; + } else { + min_i = i+1; + } + } + + if (min_i != max_i) { + /* none match */ + continue; + } + + /* we now know that the entries start at min_i */ + for (i=min_i;i<d->num_entries;i++) { + struct notify_entry *e = &d->entries[i]; + if (p_len != e->path_len || + strncmp(path, e->path, p_len) != 0) break; + if (next_p != NULL) { + if (0 == (filter & e->subdir_filter)) { + continue; + } + } else { + if (0 == (filter & e->filter)) { + continue; + } + } + notify_send(notify, e, path + e->path_len + 1, action); + } + } +} diff --git a/source4/ntvfs/common/ntvfs_common.h b/source4/ntvfs/common/ntvfs_common.h new file mode 100644 index 0000000..37dd553 --- /dev/null +++ b/source4/ntvfs/common/ntvfs_common.h @@ -0,0 +1,32 @@ +/* + Unix SMB/CIFS implementation. + + Copyright (C) Stefan Metzmacher 2007 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#ifndef _NTVFS_COMMON_H_ +#define _NTVFS_COMMON_H_ + +#include "ntvfs/ntvfs.h" + +struct notify_event; +struct notify_entry; + +#include "ntvfs/common/brlock.h" +#include "ntvfs/common/opendb.h" +#include "ntvfs/common/proto.h" + +#endif /* _NTVFS_COMMON_H_ */ diff --git a/source4/ntvfs/common/opendb.c b/source4/ntvfs/common/opendb.c new file mode 100644 index 0000000..29081ef --- /dev/null +++ b/source4/ntvfs/common/opendb.c @@ -0,0 +1,200 @@ +/* + Unix SMB/CIFS implementation. + + Copyright (C) Andrew Tridgell 2004 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* + this is the open files database. It implements shared storage of + what files are open between server instances, and implements the rules + of shared access to files. + + The caller needs to provide a file_key, which specifies what file + they are talking about. This needs to be a unique key across all + filesystems, and is usually implemented in terms of a device/inode + pair. + + Before any operations can be performed the caller needs to establish + a lock on the record associated with file_key. That is done by + calling odb_lock(). The caller releases this lock by calling + talloc_free() on the returned handle. + + All other operations on a record are done by passing the odb_lock() + handle back to this module. The handle contains internal + information about what file_key is being operated on. +*/ + +#include "includes.h" +#include "ntvfs/ntvfs.h" +#include "ntvfs/common/ntvfs_common.h" +#include "cluster/cluster.h" +#include "param/param.h" + +static const struct opendb_ops *ops; + +/* + set the odb backend ops +*/ +void odb_set_ops(const struct opendb_ops *new_ops) +{ + ops = new_ops; +} + +/* + Open up the openfiles.tdb database. Close it down using + talloc_free(). We need the imessaging_ctx to allow for pending open + notifications. +*/ +struct odb_context *odb_init(TALLOC_CTX *mem_ctx, + struct ntvfs_context *ntvfs_ctx) +{ + if (ops == NULL) { + odb_tdb_init_ops(); + } + return ops->odb_init(mem_ctx, ntvfs_ctx); +} + +/* + get a lock on a entry in the odb. This call returns a lock handle, + which the caller should unlock using talloc_free(). +*/ +struct odb_lock *odb_lock(TALLOC_CTX *mem_ctx, + struct odb_context *odb, DATA_BLOB *file_key) +{ + return ops->odb_lock(mem_ctx, odb, file_key); +} + +DATA_BLOB odb_get_key(TALLOC_CTX *mem_ctx, struct odb_lock *lck) +{ + return ops->odb_get_key(mem_ctx, lck); +} + +/* + register an open file in the open files database. + The share_access rules are implemented by odb_can_open() + and it's needed to call odb_can_open() before + odb_open_file() otherwise NT_STATUS_INTERNAL_ERROR is returned + + Note that the path is only used by the delete on close logic, not + for comparing with other filenames +*/ +NTSTATUS odb_open_file(struct odb_lock *lck, + void *file_handle, const char *path, + int *fd, NTTIME open_write_time, + bool allow_level_II_oplock, + uint32_t oplock_level, uint32_t *oplock_granted) +{ + return ops->odb_open_file(lck, file_handle, path, + fd, open_write_time, + allow_level_II_oplock, + oplock_level, oplock_granted); +} + + +/* + register a pending open file in the open files database +*/ +NTSTATUS odb_open_file_pending(struct odb_lock *lck, void *private_data) +{ + return ops->odb_open_file_pending(lck, private_data); +} + + +/* + remove a opendb entry +*/ +NTSTATUS odb_close_file(struct odb_lock *lck, void *file_handle, + const char **delete_path) +{ + return ops->odb_close_file(lck, file_handle, delete_path); +} + + +/* + remove a pending opendb entry +*/ +NTSTATUS odb_remove_pending(struct odb_lock *lck, void *private_data) +{ + return ops->odb_remove_pending(lck, private_data); +} + + +/* + rename the path in a open file +*/ +NTSTATUS odb_rename(struct odb_lock *lck, const char *path) +{ + return ops->odb_rename(lck, path); +} + +/* + get back the path of an open file +*/ +NTSTATUS odb_get_path(struct odb_lock *lck, const char **path) +{ + return ops->odb_get_path(lck, path); +} + +/* + update delete on close flag on an open file +*/ +NTSTATUS odb_set_delete_on_close(struct odb_lock *lck, bool del_on_close) +{ + return ops->odb_set_delete_on_close(lck, del_on_close); +} + +/* + update the write time on an open file +*/ +NTSTATUS odb_set_write_time(struct odb_lock *lck, + NTTIME write_time, bool force) +{ + return ops->odb_set_write_time(lck, write_time, force); +} + +/* + return the current value of the delete_on_close bit, + and the current write time. +*/ +NTSTATUS odb_get_file_infos(struct odb_context *odb, DATA_BLOB *key, + bool *del_on_close, NTTIME *write_time) +{ + return ops->odb_get_file_infos(odb, key, del_on_close, write_time); +} + +/* + determine if a file can be opened with the given share_access, + create_options and access_mask +*/ +NTSTATUS odb_can_open(struct odb_lock *lck, + uint32_t stream_id, uint32_t share_access, + uint32_t access_mask, bool delete_on_close, + uint32_t open_disposition, bool break_to_none) +{ + return ops->odb_can_open(lck, stream_id, share_access, access_mask, + delete_on_close, open_disposition, break_to_none); +} + +NTSTATUS odb_update_oplock(struct odb_lock *lck, void *file_handle, + uint32_t oplock_level) +{ + return ops->odb_update_oplock(lck, file_handle, oplock_level); +} + +NTSTATUS odb_break_oplocks(struct odb_lock *lck) +{ + return ops->odb_break_oplocks(lck); +} diff --git a/source4/ntvfs/common/opendb.h b/source4/ntvfs/common/opendb.h new file mode 100644 index 0000000..1bfc6aa --- /dev/null +++ b/source4/ntvfs/common/opendb.h @@ -0,0 +1,59 @@ +/* + Unix SMB/CIFS implementation. + + open database code - common include + + Copyright (C) Andrew Tridgell 2007 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +struct opendb_ops { + struct odb_context *(*odb_init)(TALLOC_CTX *mem_ctx, + struct ntvfs_context *ntvfs_ctx); + struct odb_lock *(*odb_lock)(TALLOC_CTX *mem_ctx, + struct odb_context *odb, DATA_BLOB *file_key); + DATA_BLOB (*odb_get_key)(TALLOC_CTX *mem_ctx, struct odb_lock *lck); + NTSTATUS (*odb_open_file)(struct odb_lock *lck, + void *file_handle, const char *path, + int *fd, NTTIME open_write_time, + bool allow_level_II_oplock, + uint32_t oplock_level, uint32_t *oplock_granted); + NTSTATUS (*odb_open_file_pending)(struct odb_lock *lck, void *private_data); + NTSTATUS (*odb_close_file)(struct odb_lock *lck, void *file_handle, + const char **delete_path); + NTSTATUS (*odb_remove_pending)(struct odb_lock *lck, void *private_data); + NTSTATUS (*odb_rename)(struct odb_lock *lck, const char *path); + NTSTATUS (*odb_get_path)(struct odb_lock *lck, const char **path); + NTSTATUS (*odb_set_delete_on_close)(struct odb_lock *lck, bool del_on_close); + NTSTATUS (*odb_set_write_time)(struct odb_lock *lck, + NTTIME write_time, bool force); + NTSTATUS (*odb_get_file_infos)(struct odb_context *odb, DATA_BLOB *key, + bool *del_on_close, NTTIME *write_time); + NTSTATUS (*odb_can_open)(struct odb_lock *lck, + uint32_t stream_id, uint32_t share_access, + uint32_t access_mask, bool delete_on_close, + uint32_t open_disposition, bool break_to_none); + NTSTATUS (*odb_update_oplock)(struct odb_lock *lck, void *file_handle, + uint32_t oplock_level); + NTSTATUS (*odb_break_oplocks)(struct odb_lock *lck); +}; + +struct opendb_oplock_break { + void *file_handle; + uint8_t level; +}; + +void odb_set_ops(const struct opendb_ops *new_ops); +void odb_tdb_init_ops(void); diff --git a/source4/ntvfs/common/opendb_tdb.c b/source4/ntvfs/common/opendb_tdb.c new file mode 100644 index 0000000..9fe0b26 --- /dev/null +++ b/source4/ntvfs/common/opendb_tdb.c @@ -0,0 +1,886 @@ +/* + Unix SMB/CIFS implementation. + + Copyright (C) Andrew Tridgell 2004 + Copyright (C) Stefan Metzmacher 2008 + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +/* + this is the open files database, tdb backend. It implements shared + storage of what files are open between server instances, and + implements the rules of shared access to files. + + The caller needs to provide a file_key, which specifies what file + they are talking about. This needs to be a unique key across all + filesystems, and is usually implemented in terms of a device/inode + pair. + + Before any operations can be performed the caller needs to establish + a lock on the record associated with file_key. That is done by + calling odb_lock(). The caller releases this lock by calling + talloc_free() on the returned handle. + + All other operations on a record are done by passing the odb_lock() + handle back to this module. The handle contains internal + information about what file_key is being operated on. +*/ + +#include "includes.h" +#include "system/filesys.h" +#include "lib/dbwrap/dbwrap.h" +#include "messaging/messaging.h" +#include "lib/messaging/irpc.h" +#include "librpc/gen_ndr/ndr_opendb.h" +#include "ntvfs/ntvfs.h" +#include "ntvfs/common/ntvfs_common.h" +#include "cluster/cluster.h" +#include "param/param.h" +#include "ntvfs/sysdep/sys_lease.h" + +struct odb_context { + struct db_context *db; + struct ntvfs_context *ntvfs_ctx; + bool oplocks; + struct sys_lease_context *lease_ctx; +}; + +/* + an odb lock handle. You must obtain one of these using odb_lock() before doing + any other operations. +*/ +struct odb_lock { + struct odb_context *odb; + struct db_record *locked; + + struct opendb_file file; + + struct { + struct opendb_entry *e; + bool attrs_only; + } can_open; +}; + +static NTSTATUS odb_oplock_break_send(struct imessaging_context *msg_ctx, + struct opendb_entry *e, + uint8_t level); + +/* + Open up the openfiles.tdb database. Close it down using + talloc_free(). We need the imessaging_ctx to allow for pending open + notifications. +*/ +static struct odb_context *odb_tdb_init(TALLOC_CTX *mem_ctx, + struct ntvfs_context *ntvfs_ctx) +{ + struct odb_context *odb; + + odb = talloc(mem_ctx, struct odb_context); + if (odb == NULL) { + return NULL; + } + + odb->db = cluster_db_tmp_open(odb, ntvfs_ctx->lp_ctx, + "openfiles", TDB_DEFAULT); + if (odb->db == NULL) { + talloc_free(odb); + return NULL; + } + + odb->ntvfs_ctx = ntvfs_ctx; + + odb->oplocks = share_bool_option(ntvfs_ctx->config, SHARE_OPLOCKS, SHARE_OPLOCKS_DEFAULT); + + odb->lease_ctx = sys_lease_context_create(ntvfs_ctx->config, odb, + ntvfs_ctx->event_ctx, + ntvfs_ctx->msg_ctx, + odb_oplock_break_send); + + return odb; +} + +static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file); + +/* + get a lock on a entry in the odb. This call returns a lock handle, + which the caller should unlock using talloc_free(). +*/ +static struct odb_lock *odb_tdb_lock(TALLOC_CTX *mem_ctx, + struct odb_context *odb, DATA_BLOB *file_key) +{ + struct odb_lock *lck; + NTSTATUS status; + TDB_DATA key; + + lck = talloc(mem_ctx, struct odb_lock); + if (lck == NULL) { + return NULL; + } + + lck->odb = talloc_reference(lck, odb); + key.dptr = talloc_memdup(lck, file_key->data, file_key->length); + key.dsize = file_key->length; + if (key.dptr == NULL) { + talloc_free(lck); + return NULL; + } + + lck->locked = dbwrap_fetch_locked(odb->db, lck, key); + if (!lck->locked) { + talloc_free(lck); + return NULL; + } + + ZERO_STRUCT(lck->can_open); + + status = odb_pull_record(lck, &lck->file); + if (NT_STATUS_EQUAL(status, NT_STATUS_OBJECT_NAME_NOT_FOUND)) { + /* initialise a blank structure */ + ZERO_STRUCT(lck->file); + } else if (!NT_STATUS_IS_OK(status)) { + talloc_free(lck); + return NULL; + } + + return lck; +} + +static DATA_BLOB odb_tdb_get_key(TALLOC_CTX *mem_ctx, struct odb_lock *lck) +{ + TDB_DATA key = dbwrap_record_get_key(lck->locked); + return data_blob_talloc(mem_ctx, key.dptr, key.dsize); +} + + +/* + determine if two odb_entry structures conflict + + return NT_STATUS_OK on no conflict +*/ +static NTSTATUS share_conflict(struct opendb_entry *e1, + uint32_t stream_id, + uint32_t share_access, + uint32_t access_mask) +{ + /* if either open involves no read.write or delete access then + it can't conflict */ + if (!(e1->access_mask & (SEC_FILE_WRITE_DATA | + SEC_FILE_APPEND_DATA | + SEC_FILE_READ_DATA | + SEC_FILE_EXECUTE | + SEC_STD_DELETE))) { + return NT_STATUS_OK; + } + if (!(access_mask & (SEC_FILE_WRITE_DATA | + SEC_FILE_APPEND_DATA | + SEC_FILE_READ_DATA | + SEC_FILE_EXECUTE | + SEC_STD_DELETE))) { + return NT_STATUS_OK; + } + + /* data IO access masks. This is skipped if the two open handles + are on different streams (as in that case the masks don't + interact) */ + if (e1->stream_id != stream_id) { + return NT_STATUS_OK; + } + +#define CHECK_MASK(am, right, sa, share) \ + if (((am) & (right)) && !((sa) & (share))) return NT_STATUS_SHARING_VIOLATION + + CHECK_MASK(e1->access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA, + share_access, NTCREATEX_SHARE_ACCESS_WRITE); + CHECK_MASK(access_mask, SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA, + e1->share_access, NTCREATEX_SHARE_ACCESS_WRITE); + + CHECK_MASK(e1->access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE, + share_access, NTCREATEX_SHARE_ACCESS_READ); + CHECK_MASK(access_mask, SEC_FILE_READ_DATA | SEC_FILE_EXECUTE, + e1->share_access, NTCREATEX_SHARE_ACCESS_READ); + + CHECK_MASK(e1->access_mask, SEC_STD_DELETE, + share_access, NTCREATEX_SHARE_ACCESS_DELETE); + CHECK_MASK(access_mask, SEC_STD_DELETE, + e1->share_access, NTCREATEX_SHARE_ACCESS_DELETE); +#undef CHECK_MASK + return NT_STATUS_OK; +} + +/* + pull a record, translating from the db format to the opendb_file structure defined + in opendb.idl +*/ +static NTSTATUS odb_pull_record(struct odb_lock *lck, struct opendb_file *file) +{ + TDB_DATA dbuf; + DATA_BLOB blob; + enum ndr_err_code ndr_err; + + dbuf = dbwrap_record_get_value(lck->locked); + if (!dbuf.dptr) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + blob.data = dbuf.dptr; + blob.length = dbuf.dsize; + + ndr_err = ndr_pull_struct_blob(&blob, lck, file, (ndr_pull_flags_fn_t)ndr_pull_opendb_file); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + return ndr_map_error2ntstatus(ndr_err); + } + + return NT_STATUS_OK; +} + +/* + push a record, translating from the opendb_file structure defined in opendb.idl +*/ +static NTSTATUS odb_push_record(struct odb_lock *lck, struct opendb_file *file) +{ + TDB_DATA dbuf; + DATA_BLOB blob; + enum ndr_err_code ndr_err; + NTSTATUS status; + + if (file->num_entries == 0) { + return dbwrap_record_delete(lck->locked); + } + + ndr_err = ndr_push_struct_blob(&blob, lck, file, (ndr_push_flags_fn_t)ndr_push_opendb_file); + if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) { + return ndr_map_error2ntstatus(ndr_err); + } + + dbuf.dptr = blob.data; + dbuf.dsize = blob.length; + + status = dbwrap_record_store(lck->locked, dbuf, TDB_REPLACE); + data_blob_free(&blob); + return status; +} + +/* + send an oplock break to a client +*/ +static NTSTATUS odb_oplock_break_send(struct imessaging_context *msg_ctx, + struct opendb_entry *e, + uint8_t level) +{ + NTSTATUS status; + struct opendb_oplock_break op_break; + DATA_BLOB blob; + + ZERO_STRUCT(op_break); + + /* tell the server handling this open file about the need to send the client + a break */ + op_break.file_handle = e->file_handle; + op_break.level = level; + + blob = data_blob_const(&op_break, sizeof(op_break)); + + status = imessaging_send(msg_ctx, e->server, + MSG_NTVFS_OPLOCK_BREAK, &blob); + NT_STATUS_NOT_OK_RETURN(status); + + return NT_STATUS_OK; +} + +static bool access_attributes_only(uint32_t access_mask, + uint32_t open_disposition, + bool break_to_none) +{ + switch (open_disposition) { + case NTCREATEX_DISP_SUPERSEDE: + case NTCREATEX_DISP_OVERWRITE_IF: + case NTCREATEX_DISP_OVERWRITE: + return false; + default: + break; + } + + if (break_to_none) { + return false; + } + +#define CHECK_MASK(m,g) ((m) && (((m) & ~(g))==0) && (((m) & (g)) != 0)) + return CHECK_MASK(access_mask, + SEC_STD_SYNCHRONIZE | + SEC_FILE_READ_ATTRIBUTE | + SEC_FILE_WRITE_ATTRIBUTE); +#undef CHECK_MASK +} + +static NTSTATUS odb_tdb_open_can_internal(struct odb_context *odb, + const struct opendb_file *file, + uint32_t stream_id, uint32_t share_access, + uint32_t access_mask, bool delete_on_close, + uint32_t open_disposition, bool break_to_none, + bool *_attrs_only) +{ + NTSTATUS status; + uint32_t i; + bool attrs_only = false; + + /* see if anyone has an oplock, which we need to break */ + for (i=0;i<file->num_entries;i++) { + if (file->entries[i].oplock_level == OPLOCK_BATCH) { + bool oplock_return = OPLOCK_BREAK_TO_LEVEL_II; + /* if this is an attribute only access + * it doesn't conflict with a BACTCH oplock + * but we'll not grant the oplock below + */ + attrs_only = access_attributes_only(access_mask, + open_disposition, + break_to_none); + if (attrs_only) { + break; + } + /* a batch oplock caches close calls, which + means the client application might have + already closed the file. We have to allow + this close to propagate by sending a oplock + break request and suspending this call + until the break is acknowledged or the file + is closed */ + if (break_to_none || + !file->entries[i].allow_level_II_oplock) { + oplock_return = OPLOCK_BREAK_TO_NONE; + } + odb_oplock_break_send(odb->ntvfs_ctx->msg_ctx, + &file->entries[i], + oplock_return); + return NT_STATUS_OPLOCK_NOT_GRANTED; + } + } + + if (file->delete_on_close) { + /* while delete on close is set, no new opens are allowed */ + return NT_STATUS_DELETE_PENDING; + } + + if (file->num_entries != 0 && delete_on_close) { + return NT_STATUS_SHARING_VIOLATION; + } + + /* check for sharing violations */ + for (i=0;i<file->num_entries;i++) { + status = share_conflict(&file->entries[i], stream_id, + share_access, access_mask); + NT_STATUS_NOT_OK_RETURN(status); + } + + /* we now know the open could succeed, but we need to check + for any exclusive oplocks. We can't grant a second open + till these are broken. Note that we check for batch oplocks + before checking for sharing violations, and check for + exclusive oplocks afterwards. */ + for (i=0;i<file->num_entries;i++) { + if (file->entries[i].oplock_level == OPLOCK_EXCLUSIVE) { + bool oplock_return = OPLOCK_BREAK_TO_LEVEL_II; + /* if this is an attribute only access + * it doesn't conflict with an EXCLUSIVE oplock + * but we'll not grant the oplock below + */ + attrs_only = access_attributes_only(access_mask, + open_disposition, + break_to_none); + if (attrs_only) { + break; + } + /* + * send an oplock break to the holder of the + * oplock and tell caller to retry later + */ + if (break_to_none || + !file->entries[i].allow_level_II_oplock) { + oplock_return = OPLOCK_BREAK_TO_NONE; + } + odb_oplock_break_send(odb->ntvfs_ctx->msg_ctx, + &file->entries[i], + oplock_return); + return NT_STATUS_OPLOCK_NOT_GRANTED; + } + } + + if (_attrs_only) { + *_attrs_only = attrs_only; + } + return NT_STATUS_OK; +} + +/* + register an open file in the open files database. + The share_access rules are implemented by odb_can_open() + and it's needed to call odb_can_open() before + odb_open_file() otherwise NT_STATUS_INTERNAL_ERROR is returned + + Note that the path is only used by the delete on close logic, not + for comparing with other filenames +*/ +static NTSTATUS odb_tdb_open_file(struct odb_lock *lck, + void *file_handle, const char *path, + int *fd, NTTIME open_write_time, + bool allow_level_II_oplock, + uint32_t oplock_level, uint32_t *oplock_granted) +{ + struct odb_context *odb = lck->odb; + + if (!lck->can_open.e) { + return NT_STATUS_INTERNAL_ERROR; + } + + if (odb->oplocks == false) { + oplock_level = OPLOCK_NONE; + } + + if (!oplock_granted) { + oplock_level = OPLOCK_NONE; + } + + if (lck->file.path == NULL) { + lck->file.path = talloc_strdup(lck, path); + NT_STATUS_HAVE_NO_MEMORY(lck->file.path); + } + + if (lck->file.open_write_time == 0) { + lck->file.open_write_time = open_write_time; + } + + /* + possibly grant an exclusive, batch or level2 oplock + */ + if (lck->can_open.attrs_only) { + oplock_level = OPLOCK_NONE; + } else if (oplock_level == OPLOCK_EXCLUSIVE) { + if (lck->file.num_entries == 0) { + oplock_level = OPLOCK_EXCLUSIVE; + } else if (allow_level_II_oplock) { + oplock_level = OPLOCK_LEVEL_II; + } else { + oplock_level = OPLOCK_NONE; + } + } else if (oplock_level == OPLOCK_BATCH) { + if (lck->file.num_entries == 0) { + oplock_level = OPLOCK_BATCH; + } else if (allow_level_II_oplock) { + oplock_level = OPLOCK_LEVEL_II; + } else { + oplock_level = OPLOCK_NONE; + } + } else if (oplock_level == OPLOCK_LEVEL_II) { + oplock_level = OPLOCK_LEVEL_II; + } else { + oplock_level = OPLOCK_NONE; + } + + lck->can_open.e->file_handle = file_handle; + lck->can_open.e->fd = fd; + lck->can_open.e->allow_level_II_oplock = allow_level_II_oplock; + lck->can_open.e->oplock_level = oplock_level; + + if (odb->lease_ctx && fd) { + NTSTATUS status; + status = sys_lease_setup(odb->lease_ctx, lck->can_open.e); + NT_STATUS_NOT_OK_RETURN(status); + } + + if (oplock_granted) { + if (lck->can_open.e->oplock_level == OPLOCK_EXCLUSIVE) { + *oplock_granted = EXCLUSIVE_OPLOCK_RETURN; + } else if (lck->can_open.e->oplock_level == OPLOCK_BATCH) { + *oplock_granted = BATCH_OPLOCK_RETURN; + } else if (lck->can_open.e->oplock_level == OPLOCK_LEVEL_II) { + *oplock_granted = LEVEL_II_OPLOCK_RETURN; + } else { + *oplock_granted = NO_OPLOCK_RETURN; + } + } + + /* it doesn't conflict, so add it to the end */ + lck->file.entries = talloc_realloc(lck, lck->file.entries, + struct opendb_entry, + lck->file.num_entries+1); + NT_STATUS_HAVE_NO_MEMORY(lck->file.entries); + + lck->file.entries[lck->file.num_entries] = *lck->can_open.e; + lck->file.num_entries++; + + talloc_free(lck->can_open.e); + lck->can_open.e = NULL; + + return odb_push_record(lck, &lck->file); +} + + +/* + register a pending open file in the open files database +*/ +static NTSTATUS odb_tdb_open_file_pending(struct odb_lock *lck, void *private_data) +{ + struct odb_context *odb = lck->odb; + + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + lck->file.pending = talloc_realloc(lck, lck->file.pending, + struct opendb_pending, + lck->file.num_pending+1); + NT_STATUS_HAVE_NO_MEMORY(lck->file.pending); + + lck->file.pending[lck->file.num_pending].server = odb->ntvfs_ctx->server_id; + lck->file.pending[lck->file.num_pending].notify_ptr = private_data; + + lck->file.num_pending++; + + return odb_push_record(lck, &lck->file); +} + + +/* + remove a opendb entry +*/ +static NTSTATUS odb_tdb_close_file(struct odb_lock *lck, void *file_handle, + const char **_delete_path) +{ + struct odb_context *odb = lck->odb; + const char *delete_path = NULL; + int i; + + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + /* find the entry, and delete it */ + for (i=0;i<lck->file.num_entries;i++) { + if (file_handle == lck->file.entries[i].file_handle && + cluster_id_equal(&odb->ntvfs_ctx->server_id, &lck->file.entries[i].server)) { + if (lck->file.entries[i].delete_on_close) { + lck->file.delete_on_close = true; + } + if (odb->lease_ctx && lck->file.entries[i].fd) { + NTSTATUS status; + status = sys_lease_remove(odb->lease_ctx, &lck->file.entries[i]); + NT_STATUS_NOT_OK_RETURN(status); + } + if (i < lck->file.num_entries-1) { + memmove(lck->file.entries+i, lck->file.entries+i+1, + (lck->file.num_entries - (i+1)) * + sizeof(struct opendb_entry)); + } + break; + } + } + + if (i == lck->file.num_entries) { + return NT_STATUS_UNSUCCESSFUL; + } + + /* send any pending notifications, removing them once sent */ + for (i=0;i<lck->file.num_pending;i++) { + imessaging_send_ptr(odb->ntvfs_ctx->msg_ctx, + lck->file.pending[i].server, + MSG_PVFS_RETRY_OPEN, + lck->file.pending[i].notify_ptr); + } + lck->file.num_pending = 0; + + lck->file.num_entries--; + + if (lck->file.num_entries == 0 && lck->file.delete_on_close) { + delete_path = lck->file.path; + } + + if (_delete_path) { + *_delete_path = delete_path; + } + + return odb_push_record(lck, &lck->file); +} + +/* + update the oplock level of the client +*/ +static NTSTATUS odb_tdb_update_oplock(struct odb_lock *lck, void *file_handle, + uint32_t oplock_level) +{ + struct odb_context *odb = lck->odb; + int i; + + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + /* find the entry, and update it */ + for (i=0;i<lck->file.num_entries;i++) { + if (file_handle == lck->file.entries[i].file_handle && + cluster_id_equal(&odb->ntvfs_ctx->server_id, &lck->file.entries[i].server)) { + lck->file.entries[i].oplock_level = oplock_level; + + if (odb->lease_ctx && lck->file.entries[i].fd) { + NTSTATUS status; + status = sys_lease_update(odb->lease_ctx, &lck->file.entries[i]); + NT_STATUS_NOT_OK_RETURN(status); + } + + break; + } + } + + if (i == lck->file.num_entries) { + return NT_STATUS_UNSUCCESSFUL; + } + + /* send any pending notifications, removing them once sent */ + for (i=0;i<lck->file.num_pending;i++) { + imessaging_send_ptr(odb->ntvfs_ctx->msg_ctx, + lck->file.pending[i].server, + MSG_PVFS_RETRY_OPEN, + lck->file.pending[i].notify_ptr); + } + lck->file.num_pending = 0; + + return odb_push_record(lck, &lck->file); +} + +/* + send oplocks breaks to none to all level2 holders +*/ +static NTSTATUS odb_tdb_break_oplocks(struct odb_lock *lck) +{ + struct odb_context *odb = lck->odb; + int i; + bool modified = false; + + /* see if anyone has an oplock, which we need to break */ + for (i=0;i<lck->file.num_entries;i++) { + if (lck->file.entries[i].oplock_level == OPLOCK_LEVEL_II) { + /* + * there could be multiple level2 oplocks + * and we just send a break to none to all of them + * without waiting for a release + */ + odb_oplock_break_send(odb->ntvfs_ctx->msg_ctx, + &lck->file.entries[i], + OPLOCK_BREAK_TO_NONE); + lck->file.entries[i].oplock_level = OPLOCK_NONE; + modified = true; + } + } + + if (modified) { + return odb_push_record(lck, &lck->file); + } + return NT_STATUS_OK; +} + +/* + remove a pending opendb entry +*/ +static NTSTATUS odb_tdb_remove_pending(struct odb_lock *lck, void *private_data) +{ + struct odb_context *odb = lck->odb; + int i; + + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + /* find the entry, and delete it */ + for (i=0;i<lck->file.num_pending;i++) { + if (private_data == lck->file.pending[i].notify_ptr && + cluster_id_equal(&odb->ntvfs_ctx->server_id, &lck->file.pending[i].server)) { + if (i < lck->file.num_pending-1) { + memmove(lck->file.pending+i, lck->file.pending+i+1, + (lck->file.num_pending - (i+1)) * + sizeof(struct opendb_pending)); + } + break; + } + } + + if (i == lck->file.num_pending) { + return NT_STATUS_UNSUCCESSFUL; + } + + lck->file.num_pending--; + + return odb_push_record(lck, &lck->file); +} + + +/* + rename the path in a open file +*/ +static NTSTATUS odb_tdb_rename(struct odb_lock *lck, const char *path) +{ + if (lck->file.path == NULL) { + /* not having the record at all is OK */ + return NT_STATUS_OK; + } + + lck->file.path = talloc_strdup(lck, path); + NT_STATUS_HAVE_NO_MEMORY(lck->file.path); + + return odb_push_record(lck, &lck->file); +} + +/* + get the path of an open file +*/ +static NTSTATUS odb_tdb_get_path(struct odb_lock *lck, const char **path) +{ + *path = NULL; + + /* we don't ignore NT_STATUS_OBJECT_NAME_NOT_FOUND here */ + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + *path = lck->file.path; + + return NT_STATUS_OK; +} + +/* + update delete on close flag on an open file +*/ +static NTSTATUS odb_tdb_set_delete_on_close(struct odb_lock *lck, bool del_on_close) +{ + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + lck->file.delete_on_close = del_on_close; + + return odb_push_record(lck, &lck->file); +} + +/* + update the write time on an open file +*/ +static NTSTATUS odb_tdb_set_write_time(struct odb_lock *lck, + NTTIME write_time, bool force) +{ + if (lck->file.path == NULL) { + return NT_STATUS_OBJECT_NAME_NOT_FOUND; + } + + if (lck->file.changed_write_time != 0 && !force) { + return NT_STATUS_OK; + } + + lck->file.changed_write_time = write_time; + + return odb_push_record(lck, &lck->file); +} + +/* + return the current value of the delete_on_close bit, and how many + people still have the file open +*/ +static NTSTATUS odb_tdb_get_file_infos(struct odb_context *odb, DATA_BLOB *key, + bool *del_on_close, NTTIME *write_time) +{ + struct odb_lock *lck; + + if (del_on_close) { + *del_on_close = false; + } + if (write_time) { + *write_time = 0; + } + + lck = odb_lock(odb, odb, key); + NT_STATUS_HAVE_NO_MEMORY(lck); + + if (del_on_close) { + *del_on_close = lck->file.delete_on_close; + } + if (write_time) { + if (lck->file.changed_write_time == 0) { + *write_time = lck->file.open_write_time; + } else { + *write_time = lck->file.changed_write_time; + } + } + + talloc_free(lck); + + return NT_STATUS_OK; +} + + +/* + determine if a file can be opened with the given share_access, + create_options and access_mask +*/ +static NTSTATUS odb_tdb_can_open(struct odb_lock *lck, + uint32_t stream_id, uint32_t share_access, + uint32_t access_mask, bool delete_on_close, + uint32_t open_disposition, bool break_to_none) +{ + struct odb_context *odb = lck->odb; + NTSTATUS status; + + status = odb_tdb_open_can_internal(odb, &lck->file, stream_id, + share_access, access_mask, + delete_on_close, open_disposition, + break_to_none, &lck->can_open.attrs_only); + NT_STATUS_NOT_OK_RETURN(status); + + lck->can_open.e = talloc(lck, struct opendb_entry); + NT_STATUS_HAVE_NO_MEMORY(lck->can_open.e); + + lck->can_open.e->server = odb->ntvfs_ctx->server_id; + lck->can_open.e->file_handle = NULL; + lck->can_open.e->fd = NULL; + lck->can_open.e->stream_id = stream_id; + lck->can_open.e->share_access = share_access; + lck->can_open.e->access_mask = access_mask; + lck->can_open.e->delete_on_close = delete_on_close; + lck->can_open.e->allow_level_II_oplock = false; + lck->can_open.e->oplock_level = OPLOCK_NONE; + + return NT_STATUS_OK; +} + + +static const struct opendb_ops opendb_tdb_ops = { + .odb_init = odb_tdb_init, + .odb_lock = odb_tdb_lock, + .odb_get_key = odb_tdb_get_key, + .odb_open_file = odb_tdb_open_file, + .odb_open_file_pending = odb_tdb_open_file_pending, + .odb_close_file = odb_tdb_close_file, + .odb_remove_pending = odb_tdb_remove_pending, + .odb_rename = odb_tdb_rename, + .odb_get_path = odb_tdb_get_path, + .odb_set_delete_on_close = odb_tdb_set_delete_on_close, + .odb_set_write_time = odb_tdb_set_write_time, + .odb_get_file_infos = odb_tdb_get_file_infos, + .odb_can_open = odb_tdb_can_open, + .odb_update_oplock = odb_tdb_update_oplock, + .odb_break_oplocks = odb_tdb_break_oplocks +}; + + +void odb_tdb_init_ops(void) +{ + sys_lease_init(); + odb_set_ops(&opendb_tdb_ops); +} diff --git a/source4/ntvfs/common/wscript_build b/source4/ntvfs/common/wscript_build new file mode 100644 index 0000000..b144472 --- /dev/null +++ b/source4/ntvfs/common/wscript_build @@ -0,0 +1,9 @@ +#!/usr/bin/env python + +bld.SAMBA_SUBSYSTEM('ntvfs_common', + source='init.c brlock.c brlock_tdb.c opendb.c opendb_tdb.c notify.c', + autoproto='proto.h', + deps='util_tdb tdb-wrap', + public_deps='NDR_OPENDB NDR_NOTIFY sys_notify sys_lease share' + ) + |