diff options
Diffstat (limited to 'source4/ntvfs/posix')
-rw-r--r-- | source4/ntvfs/posix/config.mk | 1 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_lock.c | 14 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_open.c | 326 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_oplock.c | 239 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_setfileinfo.c | 116 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_unlink.c | 94 | ||||
-rw-r--r-- | source4/ntvfs/posix/pvfs_write.c | 5 | ||||
-rw-r--r-- | source4/ntvfs/posix/vfs_posix.c | 4 | ||||
-rw-r--r-- | source4/ntvfs/posix/vfs_posix.h | 22 |
9 files changed, 729 insertions, 92 deletions
diff --git a/source4/ntvfs/posix/config.mk b/source4/ntvfs/posix/config.mk index 6879940337..88048c2af7 100644 --- a/source4/ntvfs/posix/config.mk +++ b/source4/ntvfs/posix/config.mk @@ -53,6 +53,7 @@ OBJ_FILES = \ pvfs_resolve.o \ pvfs_shortname.o \ pvfs_lock.o \ + pvfs_oplock.o \ pvfs_wait.o \ pvfs_seek.o \ pvfs_ioctl.o \ diff --git a/source4/ntvfs/posix/pvfs_lock.c b/source4/ntvfs/posix/pvfs_lock.c index df85b2b775..baa92880f1 100644 --- a/source4/ntvfs/posix/pvfs_lock.c +++ b/source4/ntvfs/posix/pvfs_lock.c @@ -294,6 +294,10 @@ NTSTATUS pvfs_lock(struct ntvfs_module_context *ntvfs, return ntvfs_map_lock(ntvfs, req, lck); } + if (lck->lockx.in.mode & LOCKING_ANDX_OPLOCK_RELEASE) { + return pvfs_oplock_release(ntvfs, req, lck); + } + f = pvfs_find_fd(pvfs, req, lck->lockx.in.file.ntvfs); if (!f) { return NT_STATUS_INVALID_HANDLE; @@ -303,6 +307,9 @@ NTSTATUS pvfs_lock(struct ntvfs_module_context *ntvfs, return NT_STATUS_FILE_IS_A_DIRECTORY; } + status = pvfs_break_level2_oplocks(f); + NT_STATUS_NOT_OK_RETURN(status); + if (lck->lockx.in.timeout != 0 && (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) { pending = talloc(f, struct pvfs_pending_lock); @@ -338,13 +345,6 @@ NTSTATUS pvfs_lock(struct ntvfs_module_context *ntvfs, return NT_STATUS_DOS(ERRDOS, ERRnoatomiclocks); } - if (lck->lockx.in.mode & LOCKING_ANDX_OPLOCK_RELEASE) { - DEBUG(0,("received unexpected oplock break\n")); - talloc_free(pending); - return NT_STATUS_NOT_IMPLEMENTED; - } - - /* the unlocks happen first */ locks = lck->lockx.in.locks; diff --git a/source4/ntvfs/posix/pvfs_open.c b/source4/ntvfs/posix/pvfs_open.c index 3a8b17ab16..12b70c00fd 100644 --- a/source4/ntvfs/posix/pvfs_open.c +++ b/source4/ntvfs/posix/pvfs_open.c @@ -268,6 +268,7 @@ static NTSTATUS pvfs_open_directory(struct pvfs_state *pvfs, f->handle->seek_offset = 0; f->handle->position = 0; f->handle->mode = 0; + f->handle->oplock = NULL; f->handle->sticky_write_time = false; f->handle->open_completed = false; @@ -296,9 +297,10 @@ static NTSTATUS pvfs_open_directory(struct pvfs_state *pvfs, } /* see if we are allowed to open at the same time as existing opens */ - status = odb_open_file(lck, f->handle, f->handle->name->stream_id, + status = odb_open_file(lck, f->handle, name->full_name, name->stream_id, share_access, access_mask, del_on_close, - name->full_name, OPLOCK_NONE, NULL); + io->generic.in.open_disposition, + false, OPLOCK_NONE, NULL); if (!NT_STATUS_IS_OK(status)) { talloc_free(lck); @@ -349,9 +351,10 @@ static NTSTATUS pvfs_open_directory(struct pvfs_state *pvfs, return NT_STATUS_INTERNAL_DB_CORRUPTION; } - status = odb_open_file(lck, f->handle, f->handle->name->stream_id, + status = odb_open_file(lck, f->handle, name->full_name, name->stream_id, share_access, access_mask, del_on_close, - name->full_name, OPLOCK_NONE, NULL); + io->generic.in.open_disposition, + false, OPLOCK_NONE, NULL); if (!NT_STATUS_IS_OK(status)) { goto cleanup_delete; @@ -669,9 +672,10 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs, oplock_level = OPLOCK_EXCLUSIVE; } - status = odb_open_file(lck, f->handle, name->stream_id, + status = odb_open_file(lck, f->handle, name->full_name, name->stream_id, share_access, access_mask, del_on_close, - name->full_name, oplock_level, &oplock_granted); + io->generic.in.open_disposition, + false, oplock_level, &oplock_granted); talloc_free(lck); if (!NT_STATUS_IS_OK(status)) { /* bad news, we must have hit a race - we don't delete the file @@ -681,9 +685,6 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs, return status; } - if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) { - oplock_granted = OPLOCK_BATCH; - } f->ntvfs = h; f->pvfs = pvfs; @@ -702,12 +703,23 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs, f->handle->seek_offset = 0; f->handle->position = 0; f->handle->mode = 0; + f->handle->oplock = NULL; f->handle->have_opendb_entry = true; f->handle->sticky_write_time = false; f->handle->open_completed = false; DLIST_ADD(pvfs->files.list, f); + if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) { + oplock_granted = OPLOCK_BATCH; + } else if (oplock_granted != OPLOCK_NONE) { + status = pvfs_setup_oplock(f, oplock_granted); + if (!NT_STATUS_IS_OK(status)) { + talloc_free(lck); + return status; + } + } + /* setup a destructor to avoid file descriptor leaks on abnormal termination */ talloc_set_destructor(f, pvfs_fnum_destructor); @@ -748,20 +760,25 @@ cleanup_delete: return status; } - /* - state of a pending open retry + state of a pending retry */ -struct pvfs_open_retry { +struct pvfs_odb_retry { struct ntvfs_module_context *ntvfs; struct ntvfs_request *req; - union smb_open *io; - struct pvfs_wait *wait_handle; DATA_BLOB odb_locking_key; + void *io; + void *private_data; + void (*callback)(struct pvfs_odb_retry *r, + struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + void *io, + void *private_data, + enum pvfs_wait_notice reason); }; -/* destroy a pending open request */ -static int pvfs_retry_destructor(struct pvfs_open_retry *r) +/* destroy a pending request */ +static int pvfs_odb_retry_destructor(struct pvfs_odb_retry *r) { struct pvfs_state *pvfs = r->ntvfs->private_data; if (r->odb_locking_key.data) { @@ -775,15 +792,92 @@ static int pvfs_retry_destructor(struct pvfs_open_retry *r) return 0; } +static void pvfs_odb_retry_callback(void *_r, enum pvfs_wait_notice reason) +{ + struct pvfs_odb_retry *r = talloc_get_type(_r, struct pvfs_odb_retry); + + if (reason == PVFS_WAIT_EVENT) { + /* + * The pending odb entry is already removed. + * We use a null locking key to indicate this + * to the destructor. + */ + data_blob_free(&r->odb_locking_key); + } + + r->callback(r, r->ntvfs, r->req, r->io, r->private_data, reason); +} + +/* + setup for a retry of a request that was rejected + by odb_open_file() or odb_can_open() +*/ +NTSTATUS pvfs_odb_retry_setup(struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + struct odb_lock *lck, + struct timeval end_time, + void *io, + void *private_data, + void (*callback)(struct pvfs_odb_retry *r, + struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + void *io, + void *private_data, + enum pvfs_wait_notice reason)) +{ + struct pvfs_state *pvfs = ntvfs->private_data; + struct pvfs_odb_retry *r; + struct pvfs_wait *wait_handle; + NTSTATUS status; + + r = talloc(req, struct pvfs_odb_retry); + NT_STATUS_HAVE_NO_MEMORY(r); + + r->ntvfs = ntvfs; + r->req = req; + r->io = io; + r->private_data = private_data; + r->callback = callback; + r->odb_locking_key = odb_get_key(r, lck); + if (r->odb_locking_key.data == NULL) { + return NT_STATUS_NO_MEMORY; + } + + /* setup a pending lock */ + status = odb_open_file_pending(lck, r); + if (!NT_STATUS_IS_OK(status)) { + return status; + } + + talloc_free(lck); + + talloc_set_destructor(r, pvfs_odb_retry_destructor); + + wait_handle = pvfs_wait_message(pvfs, req, + MSG_PVFS_RETRY_OPEN, end_time, + pvfs_odb_retry_callback, r); + if (wait_handle == NULL) { + return NT_STATUS_NO_MEMORY; + } + + talloc_steal(r, wait_handle); + + talloc_steal(pvfs, r); + + return NT_STATUS_OK; +} + /* - retry an open + retry an open after a sharing violation */ -static void pvfs_open_retry(void *private, enum pvfs_wait_notice reason) +static void pvfs_retry_open_sharing(struct pvfs_odb_retry *r, + struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + void *_io, + void *private_data, + enum pvfs_wait_notice reason) { - struct pvfs_open_retry *r = private; - struct ntvfs_module_context *ntvfs = r->ntvfs; - struct ntvfs_request *req = r->req; - union smb_open *io = r->io; + union smb_open *io = talloc_get_type(_io, union smb_open); NTSTATUS status; /* w2k3 ignores SMBntcancel for outstanding open requests. It's probably @@ -792,8 +886,6 @@ static void pvfs_open_retry(void *private, enum pvfs_wait_notice reason) return; } - talloc_free(r->wait_handle); - if (reason == PVFS_WAIT_TIMEOUT) { /* if it timed out, then give the failure immediately */ @@ -803,9 +895,6 @@ static void pvfs_open_retry(void *private, enum pvfs_wait_notice reason) return; } - /* the pending odb entry is already removed. We use a null locking - key to indicate this */ - data_blob_free(&r->odb_locking_key); talloc_free(r); /* try the open again, which could trigger another retry setup @@ -919,10 +1008,10 @@ static NTSTATUS pvfs_open_setup_retry(struct ntvfs_module_context *ntvfs, struct ntvfs_request *req, union smb_open *io, struct pvfs_file *f, - struct odb_lock *lck) + struct odb_lock *lck, + NTSTATUS parent_status) { struct pvfs_state *pvfs = ntvfs->private_data; - struct pvfs_open_retry *r; NTSTATUS status; struct timeval end_time; @@ -936,40 +1025,21 @@ static NTSTATUS pvfs_open_setup_retry(struct ntvfs_module_context *ntvfs, } } - r = talloc(req, struct pvfs_open_retry); - if (r == NULL) { - return NT_STATUS_NO_MEMORY; - } - - r->ntvfs = ntvfs; - r->req = req; - r->io = io; - r->odb_locking_key = data_blob_talloc(r, - f->handle->odb_locking_key.data, - f->handle->odb_locking_key.length); - - end_time = timeval_add(&req->statistics.request_time, 0, pvfs->sharing_violation_delay); - - /* setup a pending lock */ - status = odb_open_file_pending(lck, r); - if (!NT_STATUS_IS_OK(status)) { - return status; - } - - talloc_free(lck); + /* the retry should allocate a new file handle */ talloc_free(f); - talloc_set_destructor(r, pvfs_retry_destructor); - - r->wait_handle = pvfs_wait_message(pvfs, req, MSG_PVFS_RETRY_OPEN, end_time, - pvfs_open_retry, r); - if (r->wait_handle == NULL) { - return NT_STATUS_NO_MEMORY; + if (NT_STATUS_EQUAL(parent_status, NT_STATUS_SHARING_VIOLATION)) { + end_time = timeval_add(&req->statistics.request_time, + 0, pvfs->sharing_violation_delay); + } else if (NT_STATUS_EQUAL(parent_status, NT_STATUS_OPLOCK_NOT_GRANTED)) { + end_time = timeval_add(&req->statistics.request_time, + pvfs->oplock_break_timeout, 0); + } else { + return NT_STATUS_INTERNAL_ERROR; } - talloc_steal(pvfs, r); - - return NT_STATUS_OK; + return pvfs_odb_retry_setup(ntvfs, req, lck, end_time, io, NULL, + pvfs_retry_open_sharing); } /* @@ -1133,6 +1203,7 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs, f->handle->seek_offset = 0; f->handle->position = 0; f->handle->mode = 0; + f->handle->oplock = NULL; f->handle->have_opendb_entry = false; f->handle->sticky_write_time = false; f->handle->open_completed = false; @@ -1186,15 +1257,21 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs, } /* see if we are allowed to open at the same time as existing opens */ - status = odb_open_file(lck, f->handle, f->handle->name->stream_id, + status = odb_open_file(lck, f->handle, name->full_name, name->stream_id, share_access, access_mask, del_on_close, - name->full_name, oplock_level, &oplock_granted); + io->generic.in.open_disposition, + false, oplock_level, &oplock_granted); - /* on a sharing violation we need to retry when the file is closed by - the other user, or after 1 second */ - if (NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) && + /* + * on a sharing violation we need to retry when the file is closed by + * the other user, or after 1 second + * on a non granted oplock we need to retry when the file is closed by + * the other user, or after 30 seconds + */ + if ((NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) || + NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) && (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) { - return pvfs_open_setup_retry(ntvfs, req, io, f, lck); + return pvfs_open_setup_retry(ntvfs, req, io, f, lck, status); } if (!NT_STATUS_IS_OK(status)) { @@ -1204,6 +1281,12 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs, if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) { oplock_granted = OPLOCK_BATCH; + } else if (oplock_granted != OPLOCK_NONE) { + status = pvfs_setup_oplock(f, oplock_granted); + if (!NT_STATUS_IS_OK(status)) { + talloc_free(lck); + return status; + } } f->handle->have_opendb_entry = true; @@ -1424,6 +1507,9 @@ NTSTATUS pvfs_can_delete(struct pvfs_state *pvfs, NTSTATUS status; DATA_BLOB key; struct odb_lock *lck; + uint32_t share_access; + uint32_t access_mask; + bool delete_on_close; status = pvfs_locking_key(name, name, &key); if (!NT_STATUS_IS_OK(status)) { @@ -1436,15 +1522,18 @@ NTSTATUS pvfs_can_delete(struct pvfs_state *pvfs, return NT_STATUS_INTERNAL_DB_CORRUPTION; } - status = odb_can_open(lck, - NTCREATEX_SHARE_ACCESS_READ | - NTCREATEX_SHARE_ACCESS_WRITE | - NTCREATEX_SHARE_ACCESS_DELETE, - NTCREATEX_OPTIONS_DELETE_ON_CLOSE, - SEC_STD_DELETE); + share_access = NTCREATEX_SHARE_ACCESS_READ | + NTCREATEX_SHARE_ACCESS_WRITE | + NTCREATEX_SHARE_ACCESS_DELETE; + access_mask = SEC_STD_DELETE; + delete_on_close = true; + + status = odb_can_open(lck, name->stream_id, + share_access, access_mask, delete_on_close, + 0, false); if (NT_STATUS_IS_OK(status)) { - status = pvfs_access_check_simple(pvfs, req, name, SEC_STD_DELETE); + status = pvfs_access_check_simple(pvfs, req, name, access_mask); } /* @@ -1483,6 +1572,9 @@ NTSTATUS pvfs_can_rename(struct pvfs_state *pvfs, NTSTATUS status; DATA_BLOB key; struct odb_lock *lck; + uint32_t share_access; + uint32_t access_mask; + bool delete_on_close; status = pvfs_locking_key(name, name, &key); if (!NT_STATUS_IS_OK(status)) { @@ -1495,11 +1587,76 @@ NTSTATUS pvfs_can_rename(struct pvfs_state *pvfs, return NT_STATUS_INTERNAL_DB_CORRUPTION; } - status = odb_can_open(lck, - NTCREATEX_SHARE_ACCESS_READ | - NTCREATEX_SHARE_ACCESS_WRITE, - 0, - SEC_STD_DELETE); + share_access = NTCREATEX_SHARE_ACCESS_READ | + NTCREATEX_SHARE_ACCESS_WRITE; + access_mask = SEC_STD_DELETE; + delete_on_close = false; + + status = odb_can_open(lck, name->stream_id, + share_access, access_mask, delete_on_close, + 0, false); + + /* + * if it's a sharing violation or we got no oplock + * only keep the lock if the caller requested access + * to the lock + */ + if (NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) || + NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) { + if (lckp) { + *lckp = lck; + } else { + talloc_free(lck); + } + } else if (!NT_STATUS_IS_OK(status)) { + talloc_free(lck); + if (lckp) { + *lckp = NULL; + } + } else if (lckp) { + *lckp = lck; + } + + return status; +} + +/* + determine if the file size of a file can be changed, + or if it is prevented by an already open file +*/ +NTSTATUS pvfs_can_update_file_size(struct pvfs_state *pvfs, + struct ntvfs_request *req, + struct pvfs_filename *name, + struct odb_lock **lckp) +{ + NTSTATUS status; + DATA_BLOB key; + struct odb_lock *lck; + uint32_t share_access; + uint32_t access_mask; + bool break_to_none; + bool delete_on_close; + + status = pvfs_locking_key(name, name, &key); + if (!NT_STATUS_IS_OK(status)) { + return NT_STATUS_NO_MEMORY; + } + + lck = odb_lock(req, pvfs->odb_context, &key); + if (lck == NULL) { + DEBUG(0,("Unable to lock opendb for can_stat\n")); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + /* TODO: this may needs some more flags */ + share_access = NTCREATEX_SHARE_ACCESS_WRITE; + access_mask = 0; + delete_on_close = false; + break_to_none = true; + + status = odb_can_open(lck, name->stream_id, + share_access, access_mask, delete_on_close, + 0, break_to_none); /* * if it's a sharing violation or we got no oplock @@ -1536,6 +1693,9 @@ NTSTATUS pvfs_can_stat(struct pvfs_state *pvfs, NTSTATUS status; DATA_BLOB key; struct odb_lock *lck; + uint32_t share_access; + uint32_t access_mask; + bool delete_on_close; status = pvfs_locking_key(name, name, &key); if (!NT_STATUS_IS_OK(status)) { @@ -1548,10 +1708,14 @@ NTSTATUS pvfs_can_stat(struct pvfs_state *pvfs, return NT_STATUS_INTERNAL_DB_CORRUPTION; } - status = odb_can_open(lck, - NTCREATEX_SHARE_ACCESS_READ | - NTCREATEX_SHARE_ACCESS_WRITE, - 0, 0); + share_access = NTCREATEX_SHARE_ACCESS_READ | + NTCREATEX_SHARE_ACCESS_WRITE; + access_mask = 0; + delete_on_close = false; + + status = odb_can_open(lck, name->stream_id, + share_access, access_mask, delete_on_close, + 0, false); if (!NT_STATUS_IS_OK(status)) { talloc_free(lck); diff --git a/source4/ntvfs/posix/pvfs_oplock.c b/source4/ntvfs/posix/pvfs_oplock.c new file mode 100644 index 0000000000..cf30ddbc59 --- /dev/null +++ b/source4/ntvfs/posix/pvfs_oplock.c @@ -0,0 +1,239 @@ +/* + Unix SMB/CIFS implementation. + + POSIX NTVFS backend - oplock handling + + 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/>. +*/ + +#include "includes.h" +#include "lib/messaging/messaging.h" +#include "lib/messaging/irpc.h" +#include "vfs_posix.h" + + +struct pvfs_oplock { + struct pvfs_file_handle *handle; + struct pvfs_file *file; + uint32_t level; + struct messaging_context *msg_ctx; +}; + +/* + receive oplock breaks and forward them to the client +*/ +static void pvfs_oplock_break(struct pvfs_oplock *opl, uint8_t level) +{ + NTSTATUS status; + struct pvfs_file *f = opl->file; + struct pvfs_file_handle *h = opl->handle; + struct pvfs_state *pvfs = h->pvfs; + + DEBUG(10,("pvfs_oplock_break: sending oplock break level %d for '%s' %p\n", + level, h->name->original_name, h)); + status = ntvfs_send_oplock_break(pvfs->ntvfs, f->ntvfs, level); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0,("pvfs_oplock_break: sending oplock break failed: %s\n", + nt_errstr(status))); + } +} + +static void pvfs_oplock_break_dispatch(struct messaging_context *msg, + void *private_data, uint32_t msg_type, + struct server_id src, DATA_BLOB *data) +{ + struct pvfs_oplock *opl = talloc_get_type(private_data, + struct pvfs_oplock); + struct opendb_oplock_break opb; + + ZERO_STRUCT(opb); + + /* we need to check that this one is for us. See + messaging_send_ptr() for the other side of this. + */ + if (data->length == sizeof(struct opendb_oplock_break)) { + struct opendb_oplock_break *p; + p = (struct opendb_oplock_break *)data->data; + opb = *p; + } else { + DEBUG(0,("%s: ignore oplock break with length[%u]\n", + __location__, data->length)); + return; + } + if (opb.file_handle != opl->handle) { + return; + } + + /* + * maybe we should use ntvfs_setup_async() + */ + pvfs_oplock_break(opl, opb.level); +} + +static int pvfs_oplock_destructor(struct pvfs_oplock *opl) +{ + messaging_deregister(opl->msg_ctx, MSG_NTVFS_OPLOCK_BREAK, opl); + return 0; +} + +NTSTATUS pvfs_setup_oplock(struct pvfs_file *f, uint32_t oplock_granted) +{ + NTSTATUS status; + struct pvfs_oplock *opl; + uint32_t level = OPLOCK_NONE; + + f->handle->oplock = NULL; + + switch (oplock_granted) { + case EXCLUSIVE_OPLOCK_RETURN: + level = OPLOCK_EXCLUSIVE; + break; + case BATCH_OPLOCK_RETURN: + level = OPLOCK_BATCH; + break; + case LEVEL_II_OPLOCK_RETURN: + level = OPLOCK_LEVEL_II; + break; + } + + if (level == OPLOCK_NONE) { + return NT_STATUS_OK; + } + + opl = talloc(f->handle, struct pvfs_oplock); + NT_STATUS_HAVE_NO_MEMORY(opl); + + opl->handle = f->handle; + opl->file = f; + opl->level = level; + opl->msg_ctx = f->pvfs->ntvfs->ctx->msg_ctx; + + status = messaging_register(opl->msg_ctx, + opl, + MSG_NTVFS_OPLOCK_BREAK, + pvfs_oplock_break_dispatch); + NT_STATUS_NOT_OK_RETURN(status); + + /* destructor */ + talloc_set_destructor(opl, pvfs_oplock_destructor); + + f->handle->oplock = opl; + + return NT_STATUS_OK; +} + +NTSTATUS pvfs_oplock_release(struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, union smb_lock *lck) +{ + struct pvfs_state *pvfs = ntvfs->private_data; + struct pvfs_file *f; + struct pvfs_file_handle *h; + struct odb_lock *olck; + uint8_t oplock_break; + NTSTATUS status; + + f = pvfs_find_fd(pvfs, req, lck->lockx.in.file.ntvfs); + if (!f) { + return NT_STATUS_INVALID_HANDLE; + } + + h = f->handle; + + if (h->fd == -1) { + return NT_STATUS_FILE_IS_A_DIRECTORY; + } + + if (!h->have_opendb_entry) { + return NT_STATUS_FOOBAR; + } + + if (!h->oplock) { + return NT_STATUS_FOOBAR; + } + + olck = odb_lock(h, h->pvfs->odb_context, &h->odb_locking_key); + if (olck == NULL) { + DEBUG(0,("Unable to lock opendb for oplock update\n")); + return NT_STATUS_FOOBAR; + } + + oplock_break = (lck->lockx.in.mode >> 8) & 0xFF; + if (oplock_break == OPLOCK_BREAK_TO_NONE) { + h->oplock->level = OPLOCK_NONE; + } else if (oplock_break == OPLOCK_BREAK_TO_LEVEL_II) { + h->oplock->level = OPLOCK_LEVEL_II; + } else { + /* fallback to level II in case of a invalid value */ + DEBUG(1,("unexpected oplock break level[0x%02X]\n", oplock_break)); + h->oplock->level = OPLOCK_LEVEL_II; + } + status = odb_update_oplock(olck, h, h->oplock->level); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0,("Unable to update oplock level for '%s' - %s\n", + h->name->full_name, nt_errstr(status))); + talloc_free(olck); + return status; + } + + talloc_free(olck); + + /* after a break to none, we no longer have an oplock attached */ + if (h->oplock->level == OPLOCK_NONE) { + talloc_free(h->oplock); + h->oplock = NULL; + } + + return NT_STATUS_OK; +} + +NTSTATUS pvfs_break_level2_oplocks(struct pvfs_file *f) +{ + struct pvfs_file_handle *h = f->handle; + struct odb_lock *olck; + NTSTATUS status; + + if (h->oplock && h->oplock->level == OPLOCK_EXCLUSIVE) { + return NT_STATUS_OK; + } + + olck = odb_lock(h, h->pvfs->odb_context, &h->odb_locking_key); + if (olck == NULL) { + DEBUG(0,("Unable to lock opendb for oplock update\n")); + return NT_STATUS_FOOBAR; + } + + if (h->oplock && h->oplock->level == OPLOCK_BATCH) { + status = odb_update_oplock(olck, h, OPLOCK_LEVEL_II); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0,("Unable to update oplock level for '%s' - %s\n", + h->name->full_name, nt_errstr(status))); + talloc_free(olck); + return status; + } + } + + status = odb_break_oplocks(olck); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0,("Unable to break level2 oplocks to none for '%s' - %s\n", + h->name->full_name, nt_errstr(status))); + talloc_free(olck); + return status; + } + + talloc_free(olck); + + return NT_STATUS_OK; +} diff --git a/source4/ntvfs/posix/pvfs_setfileinfo.c b/source4/ntvfs/posix/pvfs_setfileinfo.c index fbbb8c2d4b..da1e31e639 100644 --- a/source4/ntvfs/posix/pvfs_setfileinfo.c +++ b/source4/ntvfs/posix/pvfs_setfileinfo.c @@ -152,6 +152,9 @@ static NTSTATUS pvfs_setfileinfo_rename(struct pvfs_state *pvfs, } status = pvfs_can_delete(pvfs, req, name2, NULL); + if (NT_STATUS_EQUAL(status, NT_STATUS_DELETE_PENDING)) { + return NT_STATUS_ACCESS_DENIED; + } if (NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION)) { return NT_STATUS_ACCESS_DENIED; } @@ -352,6 +355,9 @@ NTSTATUS pvfs_setfileinfo(struct ntvfs_module_context *ntvfs, case RAW_SFILEINFO_ALLOCATION_INFO: case RAW_SFILEINFO_ALLOCATION_INFORMATION: + status = pvfs_break_level2_oplocks(f); + NT_STATUS_NOT_OK_RETURN(status); + newstats.dos.alloc_size = info->allocation_info.in.alloc_size; if (newstats.dos.alloc_size < newstats.st.st_size) { newstats.st.st_size = newstats.dos.alloc_size; @@ -362,6 +368,9 @@ NTSTATUS pvfs_setfileinfo(struct ntvfs_module_context *ntvfs, case RAW_SFILEINFO_END_OF_FILE_INFO: case RAW_SFILEINFO_END_OF_FILE_INFORMATION: + status = pvfs_break_level2_oplocks(f); + NT_STATUS_NOT_OK_RETURN(status); + newstats.st.st_size = info->end_of_file_info.in.size; break; @@ -463,6 +472,84 @@ NTSTATUS pvfs_setfileinfo(struct ntvfs_module_context *ntvfs, return pvfs_dosattrib_save(pvfs, h->name, h->fd); } +/* + retry an open after a sharing violation +*/ +static void pvfs_retry_setpathinfo(struct pvfs_odb_retry *r, + struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + void *_info, + void *private_data, + enum pvfs_wait_notice reason) +{ + union smb_setfileinfo *info = talloc_get_type(_info, + union smb_setfileinfo); + NTSTATUS status = NT_STATUS_INTERNAL_ERROR; + + talloc_free(r); + + switch (reason) { + case PVFS_WAIT_CANCEL: +/*TODO*/ + status = NT_STATUS_CANCELLED; + break; + case PVFS_WAIT_TIMEOUT: + /* if it timed out, then give the failure + immediately */ +/*TODO*/ + status = NT_STATUS_SHARING_VIOLATION; + break; + case PVFS_WAIT_EVENT: + + /* try the open again, which could trigger another retry setup + if it wants to, so we have to unmark the async flag so we + will know if it does a second async reply */ + req->async_states->state &= ~NTVFS_ASYNC_STATE_ASYNC; + + status = pvfs_setpathinfo(ntvfs, req, info); + if (req->async_states->state & NTVFS_ASYNC_STATE_ASYNC) { + /* the 2nd try also replied async, so we don't send + the reply yet */ + return; + } + + /* re-mark it async, just in case someone up the chain does + paranoid checking */ + req->async_states->state |= NTVFS_ASYNC_STATE_ASYNC; + break; + } + + /* send the reply up the chain */ + req->async_states->status = status; + req->async_states->send_fn(req); +} + +/* + setup for a unlink retry after a sharing violation + or a non granted oplock +*/ +static NTSTATUS pvfs_setpathinfo_setup_retry(struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + union smb_setfileinfo *info, + struct odb_lock *lck, + NTSTATUS status) +{ + struct pvfs_state *pvfs = ntvfs->private_data; + struct timeval end_time; + + if (NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION)) { + end_time = timeval_add(&req->statistics.request_time, + 0, pvfs->sharing_violation_delay); + } else if (NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) { + end_time = timeval_add(&req->statistics.request_time, + pvfs->oplock_break_timeout, 0); + } else { + return NT_STATUS_INTERNAL_ERROR; + } + + return pvfs_odb_retry_setup(ntvfs, req, lck, end_time, info, NULL, + pvfs_retry_setpathinfo); +} /* set info on a pathname @@ -477,6 +564,7 @@ NTSTATUS pvfs_setpathinfo(struct ntvfs_module_context *ntvfs, struct utimbuf unix_times; uint32_t access_needed; uint32_t change_mask = 0; + struct odb_lock *lck = NULL; /* resolve the cifs name to a posix name */ status = pvfs_resolve_name(pvfs, req, info->generic.in.file.path, @@ -551,6 +639,20 @@ NTSTATUS pvfs_setpathinfo(struct ntvfs_module_context *ntvfs, case RAW_SFILEINFO_ALLOCATION_INFO: case RAW_SFILEINFO_ALLOCATION_INFORMATION: + status = pvfs_can_update_file_size(pvfs, req, name, &lck); + /* + * on a sharing violation we need to retry when the file is closed by + * the other user, or after 1 second + * on a non granted oplock we need to retry when the file is closed by + * the other user, or after 30 seconds + */ + if ((NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) || + NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) && + (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) { + return pvfs_setpathinfo_setup_retry(pvfs->ntvfs, req, info, lck, status); + } + NT_STATUS_NOT_OK_RETURN(status); + if (info->allocation_info.in.alloc_size > newstats.dos.alloc_size) { /* strange. Increasing the allocation size via setpathinfo should be silently ignored */ @@ -566,6 +668,20 @@ NTSTATUS pvfs_setpathinfo(struct ntvfs_module_context *ntvfs, case RAW_SFILEINFO_END_OF_FILE_INFO: case RAW_SFILEINFO_END_OF_FILE_INFORMATION: + status = pvfs_can_update_file_size(pvfs, req, name, &lck); + /* + * on a sharing violation we need to retry when the file is closed by + * the other user, or after 1 second + * on a non granted oplock we need to retry when the file is closed by + * the other user, or after 30 seconds + */ + if ((NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) || + NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) && + (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) { + return pvfs_setpathinfo_setup_retry(pvfs->ntvfs, req, info, lck, status); + } + NT_STATUS_NOT_OK_RETURN(status); + newstats.st.st_size = info->end_of_file_info.in.size; break; diff --git a/source4/ntvfs/posix/pvfs_unlink.c b/source4/ntvfs/posix/pvfs_unlink.c index bda4014bee..7a2d964b9d 100644 --- a/source4/ntvfs/posix/pvfs_unlink.c +++ b/source4/ntvfs/posix/pvfs_unlink.c @@ -23,6 +23,84 @@ #include "vfs_posix.h" #include "system/dir.h" +/* + retry an open after a sharing violation +*/ +static void pvfs_retry_unlink(struct pvfs_odb_retry *r, + struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + void *_io, + void *private_data, + enum pvfs_wait_notice reason) +{ + union smb_unlink *io = talloc_get_type(_io, union smb_unlink); + NTSTATUS status = NT_STATUS_INTERNAL_ERROR; + + talloc_free(r); + + switch (reason) { + case PVFS_WAIT_CANCEL: +/*TODO*/ + status = NT_STATUS_CANCELLED; + break; + case PVFS_WAIT_TIMEOUT: + /* if it timed out, then give the failure + immediately */ +/*TODO*/ + status = NT_STATUS_SHARING_VIOLATION; + break; + case PVFS_WAIT_EVENT: + + /* try the open again, which could trigger another retry setup + if it wants to, so we have to unmark the async flag so we + will know if it does a second async reply */ + req->async_states->state &= ~NTVFS_ASYNC_STATE_ASYNC; + + status = pvfs_unlink(ntvfs, req, io); + if (req->async_states->state & NTVFS_ASYNC_STATE_ASYNC) { + /* the 2nd try also replied async, so we don't send + the reply yet */ + return; + } + + /* re-mark it async, just in case someone up the chain does + paranoid checking */ + req->async_states->state |= NTVFS_ASYNC_STATE_ASYNC; + break; + } + + /* send the reply up the chain */ + req->async_states->status = status; + req->async_states->send_fn(req); +} + +/* + setup for a unlink retry after a sharing violation + or a non granted oplock +*/ +static NTSTATUS pvfs_unlink_setup_retry(struct ntvfs_module_context *ntvfs, + struct ntvfs_request *req, + union smb_unlink *io, + struct odb_lock *lck, + NTSTATUS status) +{ + struct pvfs_state *pvfs = ntvfs->private_data; + struct timeval end_time; + + if (NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION)) { + end_time = timeval_add(&req->statistics.request_time, + 0, pvfs->sharing_violation_delay); + } else if (NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) { + end_time = timeval_add(&req->statistics.request_time, + pvfs->oplock_break_timeout, 0); + } else { + return NT_STATUS_INTERNAL_ERROR; + } + + return pvfs_odb_retry_setup(ntvfs, req, lck, end_time, io, NULL, + pvfs_retry_unlink); +} + /* unlink a file @@ -67,6 +145,7 @@ static NTSTATUS pvfs_unlink_one(struct pvfs_state *pvfs, struct pvfs_filename *name) { NTSTATUS status; + struct odb_lock *lck = NULL; /* make sure its matches the given attributes */ status = pvfs_match_attrib(pvfs, name, @@ -75,7 +154,20 @@ static NTSTATUS pvfs_unlink_one(struct pvfs_state *pvfs, return status; } - status = pvfs_can_delete(pvfs, req, name, NULL); + status = pvfs_can_delete(pvfs, req, name, &lck); + + /* + * on a sharing violation we need to retry when the file is closed by + * the other user, or after 1 second + * on a non granted oplock we need to retry when the file is closed by + * the other user, or after 30 seconds + */ + if ((NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION) || + NT_STATUS_EQUAL(status, NT_STATUS_OPLOCK_NOT_GRANTED)) && + (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) { + return pvfs_unlink_setup_retry(pvfs->ntvfs, req, unl, lck, status); + } + if (!NT_STATUS_IS_OK(status)) { return status; } diff --git a/source4/ntvfs/posix/pvfs_write.c b/source4/ntvfs/posix/pvfs_write.c index 5a11f01952..dda8c83407 100644 --- a/source4/ntvfs/posix/pvfs_write.c +++ b/source4/ntvfs/posix/pvfs_write.c @@ -57,7 +57,10 @@ NTSTATUS pvfs_write(struct ntvfs_module_context *ntvfs, wr->writex.in.count, WRITE_LOCK); NT_STATUS_NOT_OK_RETURN(status); - + + status = pvfs_break_level2_oplocks(f); + NT_STATUS_NOT_OK_RETURN(status); + if (f->handle->name->stream_name) { ret = pvfs_stream_write(pvfs, f->handle, diff --git a/source4/ntvfs/posix/vfs_posix.c b/source4/ntvfs/posix/vfs_posix.c index e058e6f546..ca874d1db1 100644 --- a/source4/ntvfs/posix/vfs_posix.c +++ b/source4/ntvfs/posix/vfs_posix.c @@ -91,6 +91,10 @@ static void pvfs_setup_options(struct pvfs_state *pvfs) PVFS_SHARE_DELAY, PVFS_SHARE_DELAY_DEFAULT); + pvfs->oplock_break_timeout = share_int_option(scfg, + PVFS_OPLOCK_TIMEOUT, + PVFS_OPLOCK_TIMEOUT_DEFAULT); + pvfs->share_name = talloc_strdup(pvfs, scfg->name); pvfs->fs_attribs = diff --git a/source4/ntvfs/posix/vfs_posix.h b/source4/ntvfs/posix/vfs_posix.h index 84c456dcfe..4d22a91714 100644 --- a/source4/ntvfs/posix/vfs_posix.h +++ b/source4/ntvfs/posix/vfs_posix.h @@ -29,6 +29,7 @@ #include "dsdb/samdb/samdb.h" struct pvfs_wait; +struct pvfs_oplock; /* this is the private structure for the posix vfs backend. It is used to hold per-connection (per tree connect) state information */ @@ -51,9 +52,12 @@ struct pvfs_state { ntcancel */ struct pvfs_wait *wait_list; - /* the sharing violation timeout */ + /* the sharing violation timeout (nsecs) */ uint_t sharing_violation_delay; + /* the oplock break timeout (secs) */ + uint_t oplock_break_timeout; + /* filesystem attributes (see FS_ATTR_*) */ uint32_t fs_attribs; @@ -154,6 +158,13 @@ struct pvfs_file_handle { bool have_opendb_entry; + /* + * we need to wait for oplock break requests from other processes, + * and we need to remember the pvfs_file so we can correctly + * forward the oplock break to the client + */ + struct pvfs_oplock *oplock; + /* we need this hook back to our parent for lock destruction */ struct pvfs_state *pvfs; @@ -230,10 +241,16 @@ struct pvfs_dir; /* types of notification for pvfs wait events */ enum pvfs_wait_notice {PVFS_WAIT_EVENT, PVFS_WAIT_TIMEOUT, PVFS_WAIT_CANCEL}; +/* + state of a pending retry +*/ +struct pvfs_odb_retry; + #define PVFS_EADB "posix:eadb" #define PVFS_XATTR "posix:xattr" #define PVFS_FAKE_OPLOCKS "posix:fakeoplocks" #define PVFS_SHARE_DELAY "posix:sharedelay" +#define PVFS_OPLOCK_TIMEOUT "posix:oplocktimeout" #define PVFS_ALLOCATION_ROUNDING "posix:allocationrounding" #define PVFS_SEARCH_INACTIVITY "posix:searchinactivity" #define PVFS_ACL "posix:acl" @@ -241,7 +258,8 @@ enum pvfs_wait_notice {PVFS_WAIT_EVENT, PVFS_WAIT_TIMEOUT, PVFS_WAIT_CANCEL}; #define PVFS_XATTR_DEFAULT true #define PVFS_FAKE_OPLOCKS_DEFAULT false -#define PVFS_SHARE_DELAY_DEFAULT 1000000 +#define PVFS_SHARE_DELAY_DEFAULT 1000000 /* nsecs */ +#define PVFS_OPLOCK_TIMEOUT_DEFAULT 30 /* secs */ #define PVFS_ALLOCATION_ROUNDING_DEFAULT 512 #define PVFS_SEARCH_INACTIVITY_DEFAULT 300 |