The Samba-Bugzilla – Attachment 10660 Details for
Bug 11069
Backport remaining performance patches for vfs_glusterfs to 4.2/4.1
Home
|
New
|
Browse
|
Search
|
[?]
|
Reports
|
Requests
|
Help
|
New Account
|
Log In
[x]
|
Forgot Password
Login:
[x]
[patch]
new patchset for both 4.2 and 4.1 (including docfix)
v4-2-test.patch (text/plain), 32.98 KB, created by
Guenther Deschner
on 2015-01-23 23:44:51 UTC
(
hide
)
Description:
new patchset for both 4.2 and 4.1 (including docfix)
Filename:
MIME Type:
Creator:
Guenther Deschner
Created:
2015-01-23 23:44:51 UTC
Size:
32.98 KB
patch
obsolete
>From 852625d8c9d9bd79eacf4f2889675ddd809aee5a Mon Sep 17 00:00:00 2001 >From: Poornima G <pgurusid@redhat.com> >Date: Thu, 30 Oct 2014 17:14:51 +0530 >Subject: [PATCH 01/10] vfs_glusterfs: In vfs_gluster_sys_acl_get_file/fd, > reduce the number of getxattr calls. > >Signed-off-by: Poornima G <pgurusid@redhat.com> >Reviewed-by: Ira Cooper <ira@samba.org> >Reviewed-by: Guenther Deschner <gd@samba.org> >(cherry picked from commit a4fa9ca5a7a4c0b770079ab126f8172ff6d6851c) >--- > source3/modules/vfs_glusterfs.c | 42 +++++++++++++++++++++++++++++++---------- > 1 file changed, 32 insertions(+), 10 deletions(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index ba2d8e8..c47013b 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -1006,6 +1006,8 @@ static int vfs_gluster_set_offline(struct vfs_handle_struct *handle, > #define GLUSTER_ACL_HEADER_SIZE 4 > #define GLUSTER_ACL_ENTRY_SIZE 8 > >+#define GLUSTER_ACL_SIZE(n) (GLUSTER_ACL_HEADER_SIZE + (n * GLUSTER_ACL_ENTRY_SIZE)) >+ > static SMB_ACL_T gluster_to_smb_acl(const char *buf, size_t xattr_size, > TALLOC_CTX *mem_ctx) > { >@@ -1275,7 +1277,7 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_file(struct vfs_handle_struct *handle, > struct smb_acl_t *result; > char *buf; > const char *key; >- ssize_t ret; >+ ssize_t ret, size = GLUSTER_ACL_SIZE(20); > > switch (type) { > case SMB_ACL_TYPE_ACCESS: >@@ -1289,13 +1291,22 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_file(struct vfs_handle_struct *handle, > return NULL; > } > >- ret = glfs_getxattr(handle->data, path_p, key, 0, 0); >- if (ret <= 0) { >+ buf = alloca(size); >+ if (!buf) { > return NULL; > } > >- buf = alloca(ret); >- ret = glfs_getxattr(handle->data, path_p, key, buf, ret); >+ ret = glfs_getxattr(handle->data, path_p, key, buf, size); >+ if (ret == -1 && errno == ERANGE) { >+ ret = glfs_getxattr(handle->data, path_p, key, 0, 0); >+ if (ret > 0) { >+ buf = alloca(ret); >+ if (!buf) { >+ return NULL; >+ } >+ ret = glfs_getxattr(handle->data, path_p, key, buf, ret); >+ } >+ } > if (ret <= 0) { > return NULL; > } >@@ -1310,18 +1321,29 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_fd(struct vfs_handle_struct *handle, > TALLOC_CTX *mem_ctx) > { > struct smb_acl_t *result; >- int ret; >+ ssize_t ret, size = GLUSTER_ACL_SIZE(20); > char *buf; > glfs_fd_t *glfd; > > glfd = *(glfs_fd_t **)VFS_FETCH_FSP_EXTENSION(handle, fsp); >- ret = glfs_fgetxattr(glfd, "system.posix_acl_access", 0, 0); >- if (ret <= 0) { >+ >+ buf = alloca(size); >+ if (!buf) { > return NULL; > } > >- buf = alloca(ret); >- ret = glfs_fgetxattr(glfd, "system.posix_acl_access", buf, ret); >+ ret = glfs_fgetxattr(glfd, "system.posix_acl_access", buf, size); >+ if (ret == -1 && errno == ERANGE) { >+ ret = glfs_fgetxattr(glfd, "system.posix_acl_access", 0, 0); >+ if (ret > 0) { >+ buf = alloca(ret); >+ if (!buf) { >+ return NULL; >+ } >+ ret = glfs_fgetxattr(glfd, "system.posix_acl_access", >+ buf, ret); >+ } >+ } > if (ret <= 0) { > return NULL; > } >-- >1.9.3 > > >From 151865c39295155dd0b25c86dc57dbfee50af8a4 Mon Sep 17 00:00:00 2001 >From: raghavendra talur <raghavendra.talur@gmail.com> >Date: Wed, 10 Dec 2014 12:26:34 +0530 >Subject: [PATCH 02/10] vfs_glusterfs: Remember the connect path too for > reopening. > >As Samba allows us to share subdirs, lets re-use preopened >connections to glusterfs only if they are for same volume >AND same connectpath. > >Signed-off-by: raghavendra talur <raghavendra.talur@gmail.com> >Reviewed-by: Guenther Deschner <gd@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> >(cherry picked from commit 2fb4317352a165d0dd4af78fcf0a576fa246eab2) >--- > source3/modules/vfs_glusterfs.c | 20 +++++++++++++++----- > 1 file changed, 15 insertions(+), 5 deletions(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index c47013b..2dd9d71 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -78,13 +78,14 @@ static void smb_stat_ex_from_stat(struct stat_ex *dst, const struct stat *src) > > static struct glfs_preopened { > char *volume; >+ char *connectpath; > glfs_t *fs; > int ref; > struct glfs_preopened *next, *prev; > } *glfs_preopened; > > >-static int glfs_set_preopened(const char *volume, glfs_t *fs) >+static int glfs_set_preopened(const char *volume, const char *connectpath, glfs_t *fs) > { > struct glfs_preopened *entry = NULL; > >@@ -101,6 +102,13 @@ static int glfs_set_preopened(const char *volume, glfs_t *fs) > return -1; > } > >+ entry->connectpath = talloc_strdup(entry, connectpath); >+ if (entry->connectpath == NULL) { >+ talloc_free(entry); >+ errno = ENOMEM; >+ return -1; >+ } >+ > entry->fs = fs; > entry->ref = 1; > >@@ -109,12 +117,14 @@ static int glfs_set_preopened(const char *volume, glfs_t *fs) > return 0; > } > >-static glfs_t *glfs_find_preopened(const char *volume) >+static glfs_t *glfs_find_preopened(const char *volume, const char *connectpath) > { > struct glfs_preopened *entry = NULL; > > for (entry = glfs_preopened; entry; entry = entry->next) { >- if (strcmp(entry->volume, volume) == 0) { >+ if (strcmp(entry->volume, volume) == 0 && >+ strcmp(entry->connectpath, connectpath) == 0) >+ { > entry->ref++; > return entry->fs; > } >@@ -176,7 +186,7 @@ static int vfs_gluster_connect(struct vfs_handle_struct *handle, > volume = service; > } > >- fs = glfs_find_preopened(volume); >+ fs = glfs_find_preopened(volume, handle->conn->connectpath); > if (fs) { > goto done; > } >@@ -214,7 +224,7 @@ static int vfs_gluster_connect(struct vfs_handle_struct *handle, > goto done; > } > >- ret = glfs_set_preopened(volume, fs); >+ ret = glfs_set_preopened(volume, handle->conn->connectpath, fs); > if (ret < 0) { > DEBUG(0, ("%s: Failed to register volume (%s)\n", > volume, strerror(errno))); >-- >1.9.3 > > >From 30d03b5a232d1c37aaaf928d3f18c210a328c622 Mon Sep 17 00:00:00 2001 >From: raghavendra talur <raghavendra.talur@gmail.com> >Date: Wed, 10 Dec 2014 12:30:10 +0530 >Subject: [PATCH 03/10] vfs_glusterfs: Set connectpath as snapdir-entry-path. >MIME-Version: 1.0 >Content-Type: text/plain; charset=UTF-8 >Content-Transfer-Encoding: 8bit > >snapview client in gluster would require connectpath >in order add snapdir entry to readdir result. > >Signed-off-by: raghavendra talur <raghavendra.talur@gmail.com> >Reviewed-by: Guenther Deschner <gd@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> > >Autobuild-User(master): Günther Deschner <gd@samba.org> >Autobuild-Date(master): Wed Dec 10 16:38:27 CET 2014 on sn-devel-104 > >(cherry picked from commit a12ae815b01b372cbf8697d32d5d45d7b2d5f93d) >--- > source3/modules/vfs_glusterfs.c | 11 +++++++++++ > 1 file changed, 11 insertions(+) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 2dd9d71..c222164 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -210,6 +210,17 @@ static int vfs_gluster_connect(struct vfs_handle_struct *handle, > goto done; > } > >+ >+ ret = glfs_set_xlator_option(fs, "*-snapview-client", >+ "snapdir-entry-path", >+ handle->conn->connectpath); >+ if (ret < 0) { >+ DEBUG(0, ("%s: Failed to set xlator option:" >+ " snapdir-entry-path\n", volume)); >+ glfs_fini(fs); >+ return -1; >+ } >+ > ret = glfs_set_logging(fs, logfile, loglevel); > if (ret < 0) { > DEBUG(0, ("%s: Failed to set logfile %s loglevel %d\n", >-- >1.9.3 > > >From aefa9110b6d199eccfb599d5c281e1d4eb2597bf Mon Sep 17 00:00:00 2001 >From: Poornima G <pgurusid@redhat.com> >Date: Fri, 12 Dec 2014 14:11:49 +0100 >Subject: [PATCH 04/10] vfs_glusterfs: Change sys_get_acl_file/fd to return > ACLs corresponding to mode bits when there are no ACLs set. > >Signed-off-by: Poornima G <pgurusid@redhat.com> > >Reviewed-by: Guenther Deschner <gd@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> >(cherry picked from commit 26b3544251babdfcdf5ada338a4ed39ff18bc47a) >--- > source3/modules/vfs_glusterfs.c | 59 +++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 59 insertions(+) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index c222164..1b6304e 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -1029,6 +1029,39 @@ static int vfs_gluster_set_offline(struct vfs_handle_struct *handle, > > #define GLUSTER_ACL_SIZE(n) (GLUSTER_ACL_HEADER_SIZE + (n * GLUSTER_ACL_ENTRY_SIZE)) > >+static SMB_ACL_T mode_to_smb_acls(const struct stat *mode, TALLOC_CTX *mem_ctx) >+{ >+ struct smb_acl_t *result; >+ int count; >+ >+ count = 3; >+ result = sys_acl_init(mem_ctx); >+ if (!result) { >+ errno = ENOMEM; >+ return NULL; >+ } >+ >+ result->acl = talloc_array(result, struct smb_acl_entry, count); >+ if (!result->acl) { >+ errno = ENOMEM; >+ talloc_free(result); >+ return NULL; >+ } >+ >+ result->count = count; >+ >+ result->acl[0].a_type = SMB_ACL_USER_OBJ; >+ result->acl[0].a_perm = (mode->st_mode & S_IRWXU) >> 6;; >+ >+ result->acl[1].a_type = SMB_ACL_GROUP_OBJ; >+ result->acl[1].a_perm = (mode->st_mode & S_IRWXG) >> 3;; >+ >+ result->acl[2].a_type = SMB_ACL_OTHER; >+ result->acl[2].a_perm = mode->st_mode & S_IRWXO;; >+ >+ return result; >+} >+ > static SMB_ACL_T gluster_to_smb_acl(const char *buf, size_t xattr_size, > TALLOC_CTX *mem_ctx) > { >@@ -1296,6 +1329,7 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_file(struct vfs_handle_struct *handle, > TALLOC_CTX *mem_ctx) > { > struct smb_acl_t *result; >+ struct stat st; > char *buf; > const char *key; > ssize_t ret, size = GLUSTER_ACL_SIZE(20); >@@ -1328,6 +1362,18 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_file(struct vfs_handle_struct *handle, > ret = glfs_getxattr(handle->data, path_p, key, buf, ret); > } > } >+ >+ /* retrieving the ACL from the xattr has finally failed, do a >+ * mode-to-acl mapping */ >+ >+ if (ret == -1 && errno == ENODATA) { >+ ret = glfs_stat(handle->data, path_p, &st); >+ if (ret == 0) { >+ result = mode_to_smb_acls(&st, mem_ctx); >+ return result; >+ } >+ } >+ > if (ret <= 0) { > return NULL; > } >@@ -1342,6 +1388,7 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_fd(struct vfs_handle_struct *handle, > TALLOC_CTX *mem_ctx) > { > struct smb_acl_t *result; >+ struct stat st; > ssize_t ret, size = GLUSTER_ACL_SIZE(20); > char *buf; > glfs_fd_t *glfd; >@@ -1365,6 +1412,18 @@ static SMB_ACL_T vfs_gluster_sys_acl_get_fd(struct vfs_handle_struct *handle, > buf, ret); > } > } >+ >+ /* retrieving the ACL from the xattr has finally failed, do a >+ * mode-to-acl mapping */ >+ >+ if (ret == -1 && errno == ENODATA) { >+ ret = glfs_fstat(glfd, &st); >+ if (ret == 0) { >+ result = mode_to_smb_acls(&st, mem_ctx); >+ return result; >+ } >+ } >+ > if (ret <= 0) { > return NULL; > } >-- >1.9.3 > > >From 490b11807d1ad94cbecbce69a178f02b241711f2 Mon Sep 17 00:00:00 2001 >From: Poornima G <pgurusid@redhat.com> >Date: Thu, 11 Dec 2014 07:35:10 +0530 >Subject: [PATCH 05/10] vfs_glusterfs: Implement AIO support >MIME-Version: 1.0 >Content-Type: text/plain; charset=UTF-8 >Content-Transfer-Encoding: 8bit > >Signed-off-by: Poornima G <pgurusid@redhat.com> > >Reviewed-by: Guenther Deschner <gd@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> > >Autobuild-User(master): Günther Deschner <gd@samba.org> >Autobuild-Date(master): Wed Dec 17 16:35:37 CET 2014 on sn-devel-104 > >(cherry picked from commit 051b81aac81c7ccb234da221e9f8272b70b265ce) >--- > source3/modules/vfs_glusterfs.c | 254 ++++++++++++++++++++++++++++++++++++++-- > source3/wscript | 1 + > 2 files changed, 243 insertions(+), 12 deletions(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 1b6304e..68aea40 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -26,8 +26,6 @@ > * @brief Samba VFS module for glusterfs > * > * @todo >- * - AIO support\n >- * See, for example \c vfs_aio_linux.c in the \c sourc3/modules directory > * - sendfile/recvfile support > * > * A Samba VFS module for GlusterFS, based on Gluster's libgfapi. >@@ -42,9 +40,24 @@ > #include <stdio.h> > #include "api/glfs.h" > #include "lib/util/dlinklist.h" >+#include "lib/util/tevent_unix.h" >+#ifdef HAVE_SYS_EVENTFD_H >+#include <sys/eventfd.h> >+#endif >+#include <pthread.h> >+#include "smbd/globals.h" > > #define DEFAULT_VOLFILE_SERVER "localhost" > >+#ifdef HAVE_EVENTFD >+static pthread_mutex_t lock_req_list = PTHREAD_MUTEX_INITIALIZER; >+static int event_fd = -1; >+static struct tevent_fd *aio_read_event = NULL; >+static struct tevent_req **req_producer_list = NULL; >+static struct tevent_req **req_consumer_list = NULL; >+static uint64_t req_counter = 0; >+#endif >+ > /** > * Helper to convert struct stat to struct stat_ex. > */ >@@ -482,20 +495,174 @@ static ssize_t vfs_gluster_pread(struct vfs_handle_struct *handle, > return glfs_pread(*(glfs_fd_t **)VFS_FETCH_FSP_EXTENSION(handle, fsp), data, n, offset, 0); > } > >+struct glusterfs_aio_state { >+ ssize_t ret; >+ int err; >+}; >+ >+/* >+ * This function is the callback that will be called on glusterfs >+ * threads once the async IO submitted is complete. To notify >+ * Samba of the completion we use eventfd mechanism. >+ */ >+static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) >+{ >+#if HAVE_EVENTFD >+ struct tevent_req *req = NULL; >+ struct glusterfs_aio_state *state = NULL; >+ int i, sts = 0; >+ uint64_t u = 1; >+ >+ req = talloc_get_type_abort(data, struct tevent_req); >+ state = tevent_req_data(req, struct glusterfs_aio_state); >+ >+ if (ret < 0) { >+ state->ret = -1; >+ state->err = errno; >+ } else { >+ state->ret = ret; >+ state->err = 0; >+ } >+ >+ /* >+ * Store the reqs that needs to be completed by calling >+ * tevent_req_done(). tevent_req_done() cannot be called >+ * here, as it is not designed to be executed in the >+ * multithread environment, tevent_req_done() should be >+ * executed from the smbd main thread. >+ */ >+ pthread_mutex_lock (&lock_req_list); >+ { >+ for (i = 0 ; i < aio_pending_size ; i++) { >+ if(!req_producer_list[i]) { >+ req_producer_list[i] = req; >+ req_counter = req_counter + 1; >+ break; >+ } >+ } >+ } >+ pthread_mutex_unlock (&lock_req_list); >+ >+ /* >+ * For a bunch of fops notify only once >+ */ >+ if (req_counter == 1) { >+ sts = write (event_fd, &u, sizeof(uint64_t)); >+ if (sts < 0 && errno == EAGAIN) >+ DEBUG(0,("\nWRITE: reached max value")); >+ } >+ return; >+#endif >+} >+ >+#ifdef HAVE_EVENTFD >+static void aio_tevent_fd_done(struct tevent_context *event_ctx, >+ struct tevent_fd *fde, >+ uint16 flags, void *data) >+{ >+ struct tevent_req *req = NULL; >+ struct tevent_req **temp = NULL; >+ int i = 0, sts = 0; >+ uint64_t u = 0; >+ >+ sts = read (event_fd, &u, sizeof(uint64_t)); >+ if (sts < 0 && errno == EAGAIN) >+ DEBUG(0,("\nREAD: eventfd read failed (%s)",strerror(errno))); >+ >+ pthread_mutex_lock (&lock_req_list); >+ { >+ temp = req_producer_list; >+ req_producer_list = req_consumer_list; >+ req_consumer_list = temp; >+ req_counter = 0; >+ } >+ pthread_mutex_unlock (&lock_req_list); >+ >+ for (i = 0 ; i < aio_pending_size ; i++) { >+ req = req_consumer_list[i]; >+ if (req) { >+ tevent_req_done(req); >+ req_consumer_list[i] = 0; >+ } >+ } >+ return; >+} >+#endif >+ >+static bool init_gluster_aio(struct vfs_handle_struct *handle) >+{ >+#ifdef HAVE_EVENTFD >+ if (event_fd != -1) { >+ /* >+ * Already initialized. >+ */ >+ return true; >+ } >+ >+ event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); >+ if (event_fd == -1) { >+ goto fail; >+ } >+ >+ aio_read_event = tevent_add_fd(handle->conn->sconn->ev_ctx, >+ NULL, >+ event_fd, >+ TEVENT_FD_READ, >+ aio_tevent_fd_done, >+ NULL); >+ if (aio_read_event == NULL) { >+ goto fail; >+ } >+ >+ req_producer_list = talloc_zero_array(NULL, struct tevent_req *, >+ aio_pending_size); >+ req_consumer_list = talloc_zero_array(NULL, struct tevent_req *, >+ aio_pending_size); >+ >+ return true; >+fail: >+ TALLOC_FREE(aio_read_event); >+ if (event_fd != -1) { >+ close(event_fd); >+ event_fd = -1; >+ } >+#endif >+ return false; >+} >+ > static struct tevent_req *vfs_gluster_pread_send(struct vfs_handle_struct > *handle, TALLOC_CTX *mem_ctx, > struct tevent_context *ev, > files_struct *fsp, void *data, > size_t n, off_t offset) > { >+ struct tevent_req *req = NULL; >+ struct glusterfs_aio_state *state = NULL; >+ int ret = 0; >+ >+#ifndef HAVE_EVENTFD > errno = ENOTSUP; > return NULL; >-} >+#endif > >-static ssize_t vfs_gluster_pread_recv(struct tevent_req *req, int *err) >-{ >- errno = ENOTSUP; >- return -1; >+ req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); >+ if (req == NULL) { >+ return NULL; >+ } >+ >+ if (!init_gluster_aio(handle)) { >+ tevent_req_error(req, EIO); >+ return tevent_req_post(req, ev); >+ } >+ ret = glfs_pread_async(*(glfs_fd_t **)VFS_FETCH_FSP_EXTENSION(handle, >+ fsp), data, n, offset, 0, aio_glusterfs_done, >+ req); >+ if (ret < 0) { >+ tevent_req_error(req, -ret); >+ return tevent_req_post(req, ev); >+ } >+ >+ return req; > } > > static ssize_t vfs_gluster_write(struct vfs_handle_struct *handle, >@@ -518,14 +685,53 @@ static struct tevent_req *vfs_gluster_pwrite_send(struct vfs_handle_struct > const void *data, size_t n, > off_t offset) > { >+ struct tevent_req *req = NULL; >+ struct glusterfs_aio_state *state = NULL; >+ int ret = 0; >+ >+#ifndef HAVE_EVENTFD > errno = ENOTSUP; > return NULL; >+#endif >+ >+ req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); >+ if (req == NULL) { >+ return NULL; >+ } >+ if (!init_gluster_aio(handle)) { >+ tevent_req_error(req, EIO); >+ return tevent_req_post(req, ev); >+ } >+ ret = glfs_pwrite_async(*(glfs_fd_t **)VFS_FETCH_FSP_EXTENSION(handle, >+ fsp), data, n, offset, 0, aio_glusterfs_done, >+ req); >+ if (ret < 0) { >+ tevent_req_error(req, -ret); >+ return tevent_req_post(req, ev); >+ } >+ return req; > } > >-static ssize_t vfs_gluster_pwrite_recv(struct tevent_req *req, int *err) >+static ssize_t vfs_gluster_recv(struct tevent_req *req, int *err) > { >+ struct glusterfs_aio_state *state = NULL; >+ >+#ifndef HAVE_EVENTFD > errno = ENOTSUP; > return -1; >+#endif >+ state = tevent_req_data(req, struct glusterfs_aio_state); >+ if (state == NULL) { >+ return -1; >+ } >+ >+ if (tevent_req_is_unix_error(req, err)) { >+ return -1; >+ } >+ if (state->ret == -1) { >+ *err = state->err; >+ } >+ return state->ret; > } > > static off_t vfs_gluster_lseek(struct vfs_handle_struct *handle, >@@ -570,14 +776,38 @@ static struct tevent_req *vfs_gluster_fsync_send(struct vfs_handle_struct > struct tevent_context *ev, > files_struct *fsp) > { >+ struct tevent_req *req = NULL; >+ struct glusterfs_aio_state *state = NULL; >+ int ret = 0; >+ >+#ifndef HAVE_EVENTFD > errno = ENOTSUP; > return NULL; >+#endif >+ >+ req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); >+ if (req == NULL) { >+ return NULL; >+ } >+ if (!init_gluster_aio(handle)) { >+ tevent_req_error(req, EIO); >+ return tevent_req_post(req, ev); >+ } >+ ret = glfs_fsync_async(*(glfs_fd_t **)VFS_FETCH_FSP_EXTENSION(handle, >+ fsp), aio_glusterfs_done, req); >+ if (ret < 0) { >+ tevent_req_error(req, -ret); >+ return tevent_req_post(req, ev); >+ } >+ return req; > } > > static int vfs_gluster_fsync_recv(struct tevent_req *req, int *err) > { >- errno = ENOTSUP; >- return -1; >+ /* >+ * Use implicit conversion ssize_t->int >+ */ >+ return vfs_gluster_recv(req, err); > } > > static int vfs_gluster_stat(struct vfs_handle_struct *handle, >@@ -1530,11 +1760,11 @@ static struct vfs_fn_pointers glusterfs_fns = { > .read_fn = vfs_gluster_read, > .pread_fn = vfs_gluster_pread, > .pread_send_fn = vfs_gluster_pread_send, >- .pread_recv_fn = vfs_gluster_pread_recv, >+ .pread_recv_fn = vfs_gluster_recv, > .write_fn = vfs_gluster_write, > .pwrite_fn = vfs_gluster_pwrite, > .pwrite_send_fn = vfs_gluster_pwrite_send, >- .pwrite_recv_fn = vfs_gluster_pwrite_recv, >+ .pwrite_recv_fn = vfs_gluster_recv, > .lseek_fn = vfs_gluster_lseek, > .sendfile_fn = vfs_gluster_sendfile, > .recvfile_fn = vfs_gluster_recvfile, >diff --git a/source3/wscript b/source3/wscript >index bf025fc..b317a32 100644 >--- a/source3/wscript >+++ b/source3/wscript >@@ -534,6 +534,7 @@ return acl_get_perm_np(permset_d, perm); > conf.DEFINE('HAVE_NO_AIO', '1') > > if host_os.rfind('linux') > -1: >+ conf.CHECK_FUNCS('eventfd') > conf.CHECK_FUNCS_IN('io_submit', 'aio') > conf.CHECK_CODE(''' > struct io_event ioev; >-- >1.9.3 > > >From 566a61511938892ee0515fb66ff42ea05c9ad11c Mon Sep 17 00:00:00 2001 >From: raghavendra talur <raghavendra.talur@gmail.com> >Date: Wed, 10 Dec 2014 21:15:03 +0530 >Subject: [PATCH 06/10] vfs/glusterfs: Change xattr key to match gluster key. > >Gluster changed the key used to do get_real_filename >in 3.5. This changes Samba's glusterfs vfs-module >to use the same key. > >This will render SMB_VFS_GET_REALFILENAME() working >again under 3.5 and will remove support for that functionality >for Gluster 3.4 and older. > >Signed-off-by: raghavendra talur <raghavendra.talur@gmail.com> >Reviewed-by: Ira Cooper <ira@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> >(cherry picked from commit af9ec9fea5a730023cdee6e236f9585e3a18b0e6) >--- > source3/modules/vfs_glusterfs.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 68aea40..10c3a22 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -1110,7 +1110,7 @@ static int vfs_gluster_get_real_filename(struct vfs_handle_struct *handle, > } > > snprintf(key_buf, NAME_MAX + 64, >- "user.glusterfs.get_real_filename:%s", name); >+ "glusterfs.get_real_filename:%s", name); > > ret = glfs_getxattr(handle->data, path, key_buf, val_buf, NAME_MAX + 1); > if (ret == -1) { >-- >1.9.3 > > >From fa1921d85b5ba91970a02304e38e3e334af16331 Mon Sep 17 00:00:00 2001 >From: Ira Cooper <ira@samba.org> >Date: Mon, 19 Jan 2015 23:08:17 -0500 >Subject: [PATCH 07/10] vfs_glusterfs: Replace eventfd with pipes, for AIO use >MIME-Version: 1.0 >Content-Type: text/plain; charset=UTF-8 >Content-Transfer-Encoding: 8bit > >Pipes clean up the AIO implementation substantially, due to the fact >that they implement a natural ithread safe queue instead of us >creating our own queue. > >Signed-off-by: Ira Cooper <ira@samba.org> >Signed-off-by: Poornima G <pgurusid@redhat.com> >Reviewed-by: Günther Deschner <gd@samba.org> >Reviewed-by: Michael Adam <obnox@samba.org> > >Autobuild-User(master): Ira Cooper <ira@samba.org> >Autobuild-Date(master): Wed Jan 21 20:40:11 CET 2015 on sn-devel-104 > >(cherry picked from commit 4c3a3d9e6adc95d0f0e1f6030b2406613d9f9f53) >--- > source3/modules/vfs_glusterfs.c | 136 ++++++++++++---------------------------- > source3/wscript | 1 - > 2 files changed, 40 insertions(+), 97 deletions(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 10c3a22..1d3abd8 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -41,22 +41,14 @@ > #include "api/glfs.h" > #include "lib/util/dlinklist.h" > #include "lib/util/tevent_unix.h" >-#ifdef HAVE_SYS_EVENTFD_H >-#include <sys/eventfd.h> >-#endif >-#include <pthread.h> >+#include "lib/tevent/tevent_internal.h" > #include "smbd/globals.h" > > #define DEFAULT_VOLFILE_SERVER "localhost" > >-#ifdef HAVE_EVENTFD >-static pthread_mutex_t lock_req_list = PTHREAD_MUTEX_INITIALIZER; >-static int event_fd = -1; >+static int read_fd = -1; >+static int write_fd = -1; > static struct tevent_fd *aio_read_event = NULL; >-static struct tevent_req **req_producer_list = NULL; >-static struct tevent_req **req_consumer_list = NULL; >-static uint64_t req_counter = 0; >-#endif > > /** > * Helper to convert struct stat to struct stat_ex. >@@ -503,15 +495,13 @@ struct glusterfs_aio_state { > /* > * This function is the callback that will be called on glusterfs > * threads once the async IO submitted is complete. To notify >- * Samba of the completion we use eventfd mechanism. >+ * Samba of the completion we use a pipe pased queue. > */ > static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) > { >-#if HAVE_EVENTFD > struct tevent_req *req = NULL; > struct glusterfs_aio_state *state = NULL; >- int i, sts = 0; >- uint64_t u = 1; >+ int sts = 0; > > req = talloc_get_type_abort(data, struct tevent_req); > state = tevent_req_data(req, struct glusterfs_aio_state); >@@ -525,88 +515,65 @@ static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) > } > > /* >- * Store the reqs that needs to be completed by calling >- * tevent_req_done(). tevent_req_done() cannot be called >- * here, as it is not designed to be executed in the >- * multithread environment, tevent_req_done() should be >+ * Write the pointer to each req that needs to be completed >+ * by calling tevent_req_done(). tevent_req_done() cannot >+ * be called here, as it is not designed to be executed >+ * in the multithread environment, tevent_req_done() must be > * executed from the smbd main thread. > */ >- pthread_mutex_lock (&lock_req_list); >- { >- for (i = 0 ; i < aio_pending_size ; i++) { >- if(!req_producer_list[i]) { >- req_producer_list[i] = req; >- req_counter = req_counter + 1; >- break; >- } >- } >- } >- pthread_mutex_unlock (&lock_req_list); > >- /* >- * For a bunch of fops notify only once >- */ >- if (req_counter == 1) { >- sts = write (event_fd, &u, sizeof(uint64_t)); >- if (sts < 0 && errno == EAGAIN) >- DEBUG(0,("\nWRITE: reached max value")); >+ sts = sys_write (write_fd, &req, sizeof(struct tevent_req *)); >+ if (sts < 0) { >+ DEBUG(0,("\nWrite to pipe failed (%s)", strerror(errno))); > } >+ > return; >-#endif > } > >-#ifdef HAVE_EVENTFD >+/* >+ * Read each req off the pipe and process it. >+ */ > static void aio_tevent_fd_done(struct tevent_context *event_ctx, > struct tevent_fd *fde, > uint16 flags, void *data) > { > struct tevent_req *req = NULL; >- struct tevent_req **temp = NULL; >- int i = 0, sts = 0; >- uint64_t u = 0; >- >- sts = read (event_fd, &u, sizeof(uint64_t)); >- if (sts < 0 && errno == EAGAIN) >- DEBUG(0,("\nREAD: eventfd read failed (%s)",strerror(errno))); >- >- pthread_mutex_lock (&lock_req_list); >- { >- temp = req_producer_list; >- req_producer_list = req_consumer_list; >- req_consumer_list = temp; >- req_counter = 0; >- } >- pthread_mutex_unlock (&lock_req_list); >- >- for (i = 0 ; i < aio_pending_size ; i++) { >- req = req_consumer_list[i]; >- if (req) { >- tevent_req_done(req); >- req_consumer_list[i] = 0; >- } >+ int sts = 0; >+ >+ sts = sys_read(read_fd, &req, sizeof(struct tevent_req *)); >+ if (sts < 0) { >+ DEBUG(0,("\nRead from pipe failed (%s)", strerror(errno))); >+ } >+ >+ if (req) { >+ tevent_req_done(req); > } > return; > } >-#endif > > static bool init_gluster_aio(struct vfs_handle_struct *handle) > { >-#ifdef HAVE_EVENTFD >- if (event_fd != -1) { >+ int fds[2]; >+ int ret = -1; >+ >+ if (read_fd != -1) { > /* > * Already initialized. > */ > return true; > } > >- event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); >- if (event_fd == -1) { >+ ret = pipe(fds); >+ if (ret == -1) { > goto fail; > } > >+ read_fd = fds[0]; >+ write_fd = fds[1]; >+ > aio_read_event = tevent_add_fd(handle->conn->sconn->ev_ctx, > NULL, >- event_fd, >+ read_fd, > TEVENT_FD_READ, > aio_tevent_fd_done, > NULL); >@@ -614,19 +581,15 @@ static bool init_gluster_aio(struct vfs_handle_struct *handle) > goto fail; > } > >- req_producer_list = talloc_zero_array(NULL, struct tevent_req *, >- aio_pending_size); >- req_consumer_list = talloc_zero_array(NULL, struct tevent_req *, >- aio_pending_size); >- > return true; > fail: > TALLOC_FREE(aio_read_event); >- if (event_fd != -1) { >- close(event_fd); >- event_fd = -1; >+ if (read_fd != -1) { >+ close(read_fd); >+ close(write_fd); >+ read_fd = -1; >+ write_fd = -1; > } >-#endif > return false; > } > >@@ -640,11 +603,6 @@ static struct tevent_req *vfs_gluster_pread_send(struct vfs_handle_struct > struct glusterfs_aio_state *state = NULL; > int ret = 0; > >-#ifndef HAVE_EVENTFD >- errno = ENOTSUP; >- return NULL; >-#endif >- > req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); > if (req == NULL) { > return NULL; >@@ -689,11 +647,6 @@ static struct tevent_req *vfs_gluster_pwrite_send(struct vfs_handle_struct > struct glusterfs_aio_state *state = NULL; > int ret = 0; > >-#ifndef HAVE_EVENTFD >- errno = ENOTSUP; >- return NULL; >-#endif >- > req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); > if (req == NULL) { > return NULL; >@@ -716,10 +669,6 @@ static ssize_t vfs_gluster_recv(struct tevent_req *req, int *err) > { > struct glusterfs_aio_state *state = NULL; > >-#ifndef HAVE_EVENTFD >- errno = ENOTSUP; >- return -1; >-#endif > state = tevent_req_data(req, struct glusterfs_aio_state); > if (state == NULL) { > return -1; >@@ -780,11 +729,6 @@ static struct tevent_req *vfs_gluster_fsync_send(struct vfs_handle_struct > struct glusterfs_aio_state *state = NULL; > int ret = 0; > >-#ifndef HAVE_EVENTFD >- errno = ENOTSUP; >- return NULL; >-#endif >- > req = tevent_req_create(mem_ctx, &state, struct glusterfs_aio_state); > if (req == NULL) { > return NULL; >diff --git a/source3/wscript b/source3/wscript >index b317a32..bf025fc 100644 >--- a/source3/wscript >+++ b/source3/wscript >@@ -534,7 +534,6 @@ return acl_get_perm_np(permset_d, perm); > conf.DEFINE('HAVE_NO_AIO', '1') > > if host_os.rfind('linux') > -1: >- conf.CHECK_FUNCS('eventfd') > conf.CHECK_FUNCS_IN('io_submit', 'aio') > conf.CHECK_CODE(''' > struct io_event ioev; >-- >1.9.3 > > >From 187895823ce20191835640f6d159ecaf995fd143 Mon Sep 17 00:00:00 2001 >From: Michael Adam <obnox@samba.org> >Date: Wed, 21 Jan 2015 22:07:53 +0100 >Subject: [PATCH 08/10] vfs:glusterfs: whitespace fix. > >Signed-off-by: Michael Adam <obnox@samba.org> >Reviewed-by: Jeremy Allison <jra@samba.org> > >Autobuild-User(master): Jeremy Allison <jra@samba.org> >Autobuild-Date(master): Thu Jan 22 03:20:17 CET 2015 on sn-devel-104 > >(cherry picked from commit cde5d380e192487ae77734afcb3eb3d327aafb98) >--- > source3/modules/vfs_glusterfs.c | 6 +++--- > 1 file changed, 3 insertions(+), 3 deletions(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 1d3abd8..96f5559 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -522,9 +522,9 @@ static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) > * executed from the smbd main thread. > */ > >- sts = sys_write (write_fd, &req, sizeof(struct tevent_req *)); >- if (sts < 0) { >- DEBUG(0,("\nWrite to pipe failed (%s)", strerror(errno))); >+ sts = sys_write(write_fd, &req, sizeof(struct tevent_req *)); >+ if (sts < 0) { >+ DEBUG(0,("\nWrite to pipe failed (%s)", strerror(errno))); > } > > return; >-- >1.9.3 > > >From df7e1863257f8d58558d26607981d768e2c4785a Mon Sep 17 00:00:00 2001 >From: Volker Lendecke <vl@samba.org> >Date: Wed, 21 Jan 2015 20:49:24 +0100 >Subject: [PATCH 09/10] vfs: Fix a typo > >Signed-off-by: Volker Lendecke <vl@samba.org> >Reviewed-by: Andreas Schneider <asn@samba.org> > >Autobuild-User(master): Andreas Schneider <asn@cryptomilk.org> >Autobuild-Date(master): Thu Jan 22 13:14:38 CET 2015 on sn-devel-104 > >(cherry picked from commit db63e4887c81c3e458194004353d7ef92c67b393) >--- > source3/modules/vfs_glusterfs.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 96f5559..35a9c8d 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -495,7 +495,7 @@ struct glusterfs_aio_state { > /* > * This function is the callback that will be called on glusterfs > * threads once the async IO submitted is complete. To notify >- * Samba of the completion we use a pipe pased queue. >+ * Samba of the completion we use a pipe based queue. > */ > static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) > { >-- >1.9.3 > > >From 4b0d56ba068673ac0280ee5d4afd383aa795a85c Mon Sep 17 00:00:00 2001 >From: Ira Cooper <ira@samba.org> >Date: Thu, 22 Jan 2015 17:14:31 -0500 >Subject: [PATCH 10/10] vfs_glusterfs: Add comments to the pipe(2) code. > >The guarantees around read(2) and write(2) and pipes are critical >to understanding this code. Hopefully these comments will help. > >Signed-off-by: Ira Cooper <ira@samba.org> >Reviewed-by: Jeremy Allison <jra@samba.org> > >Autobuild-User(master): Jeremy Allison <jra@samba.org> >Autobuild-Date(master): Fri Jan 23 20:58:51 CET 2015 on sn-devel-104 > >(cherry picked from commit 3b2d8bdbb1851961536241d3aaaf6ac936964517) >--- > source3/modules/vfs_glusterfs.c | 17 +++++++++++++++++ > 1 file changed, 17 insertions(+) > >diff --git a/source3/modules/vfs_glusterfs.c b/source3/modules/vfs_glusterfs.c >index 35a9c8d..cfdd195 100644 >--- a/source3/modules/vfs_glusterfs.c >+++ b/source3/modules/vfs_glusterfs.c >@@ -520,6 +520,13 @@ static void aio_glusterfs_done(glfs_fd_t *fd, ssize_t ret, void *data) > * be called here, as it is not designed to be executed > * in the multithread environment, tevent_req_done() must be > * executed from the smbd main thread. >+ * >+ * write(2) on pipes with sizes under _POSIX_PIPE_BUF >+ * in size is atomic, without this, the use op pipes in this >+ * code would not work. >+ * >+ * sys_write is a thin enough wrapper around write(2) >+ * that we can trust it here. > */ > > sts = sys_write(write_fd, &req, sizeof(struct tevent_req *)); >@@ -540,6 +547,16 @@ static void aio_tevent_fd_done(struct tevent_context *event_ctx, > struct tevent_req *req = NULL; > int sts = 0; > >+ /* >+ * read(2) on pipes is atomic if the needed data is available >+ * in the pipe, per SUS and POSIX. Because we always write >+ * to the pipe in sizeof(struct tevent_req *) chunks, we can >+ * always read in those chunks, atomically. >+ * >+ * sys_read is a thin enough wrapper around read(2) that we >+ * can trust it here. >+ */ >+ > sts = sys_read(read_fd, &req, sizeof(struct tevent_req *)); > if (sts < 0) { > DEBUG(0,("\nRead from pipe failed (%s)", strerror(errno))); >-- >1.9.3 >
<b>You cannot view the attachment while viewing its details because your browser does not support IFRAMEs. <a href="attachment.cgi?id=10660">View the attachment on a separate page</a>.</b>
View Attachment As Raw
Flags:
obnox
:
review+
ira
:
review+
Actions:
View
Attachments on
bug 11069
:
10656
|
10657
| 10660