Whamcloud - gitweb
Revert "LU-147 avoid 8k obd device amount limit"
[fs/lustre-release.git] / lustre / obdclass / linux / linux-module.c
index c043218..d893f16 100644 (file)
@@ -16,8 +16,8 @@
  * in the LICENSE file that accompanied this code).
  *
  * You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see [sun.com URL with a
- * copy of GPLv2].
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  * CA 95054 USA or visit www.sun.com if you need additional information or
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -81,6 +81,7 @@
 #include <libcfs/libcfs.h>
 #include <obd_support.h>
 #include <obd_class.h>
+#include <lnet/lnetctl.h>
 #include <lprocfs_status.h>
 #include <lustre_ver.h>
 #include <lustre/lustre_build_version.h>
@@ -98,7 +99,7 @@ int obd_ioctl_getdata(char **buf, int *len, void *arg)
         int offset = 0;
         ENTRY;
 
-        err = copy_from_user(&hdr, (void *)arg, sizeof(hdr));
+        err = cfs_copy_from_user(&hdr, (void *)arg, sizeof(hdr));
         if ( err )
                 RETURN(err);
 
@@ -119,9 +120,11 @@ int obd_ioctl_getdata(char **buf, int *len, void *arg)
                 RETURN(-EINVAL);
         }
 
-        /* XXX allocate this more intelligently, using kmalloc when
-         * appropriate */
-        OBD_VMALLOC(*buf, hdr.ioc_len);
+        /* When there are lots of processes calling vmalloc on multi-core
+         * system, the high lock contention will hurt performance badly,
+         * obdfilter-survey is an example, which relies on ioctl. So we'd
+         * better avoid vmalloc on ioctl path. LU-66 */
+        OBD_ALLOC_LARGE(*buf, hdr.ioc_len);
         if (*buf == NULL) {
                 CERROR("Cannot allocate control buffer of len %d\n",
                        hdr.ioc_len);
@@ -130,31 +133,31 @@ int obd_ioctl_getdata(char **buf, int *len, void *arg)
         *len = hdr.ioc_len;
         data = (struct obd_ioctl_data *)*buf;
 
-        err = copy_from_user(*buf, (void *)arg, hdr.ioc_len);
+        err = cfs_copy_from_user(*buf, (void *)arg, hdr.ioc_len);
         if ( err ) {
-                OBD_VFREE(*buf, hdr.ioc_len);
+                OBD_FREE_LARGE(*buf, hdr.ioc_len);
                 RETURN(err);
         }
 
         if (obd_ioctl_is_invalid(data)) {
                 CERROR("ioctl not correctly formatted\n");
-                OBD_VFREE(*buf, hdr.ioc_len);
+                OBD_FREE_LARGE(*buf, hdr.ioc_len);
                 RETURN(-EINVAL);
         }
 
         if (data->ioc_inllen1) {
                 data->ioc_inlbuf1 = &data->ioc_bulk[0];
-                offset += size_round(data->ioc_inllen1);
+                offset += cfs_size_round(data->ioc_inllen1);
         }
 
         if (data->ioc_inllen2) {
                 data->ioc_inlbuf2 = &data->ioc_bulk[0] + offset;
-                offset += size_round(data->ioc_inllen2);
+                offset += cfs_size_round(data->ioc_inllen2);
         }
 
         if (data->ioc_inllen3) {
                 data->ioc_inlbuf3 = &data->ioc_bulk[0] + offset;
-                offset += size_round(data->ioc_inllen3);
+                offset += cfs_size_round(data->ioc_inllen3);
         }
 
         if (data->ioc_inllen4) {
@@ -169,7 +172,7 @@ int obd_ioctl_popdata(void *arg, void *data, int len)
 {
         int err;
 
-        err = copy_to_user(arg, data, len);
+        err = cfs_copy_to_user(arg, data, len);
         if (err)
                 err = -EFAULT;
         return err;
@@ -178,41 +181,43 @@ int obd_ioctl_popdata(void *arg, void *data, int len)
 EXPORT_SYMBOL(obd_ioctl_getdata);
 EXPORT_SYMBOL(obd_ioctl_popdata);
 
-#define OBD_MINOR 241
-extern struct cfs_psdev_ops          obd_psdev_ops;
-
 /*  opening /dev/obd */
 static int obd_class_open(struct inode * inode, struct file * file)
 {
-        if (obd_psdev_ops.p_open != NULL)
-                return obd_psdev_ops.p_open(0, NULL);
-        return -EPERM;
+        ENTRY;
+
+        PORTAL_MODULE_USE;
+        RETURN(0);
 }
 
 /*  closing /dev/obd */
 static int obd_class_release(struct inode * inode, struct file * file)
 {
-        if (obd_psdev_ops.p_close != NULL)
-                return obd_psdev_ops.p_close(0, NULL);
-        return -EPERM;
+        ENTRY;
+
+        PORTAL_MODULE_UNUSE;
+        RETURN(0);
 }
 
 /* to control /dev/obd */
+#ifdef HAVE_UNLOCKED_IOCTL
+static long obd_class_ioctl(struct file *filp, unsigned int cmd,
+                            unsigned long arg)
+#else
 static int obd_class_ioctl(struct inode *inode, struct file *filp,
                            unsigned int cmd, unsigned long arg)
+#endif
 {
         int err = 0;
         ENTRY;
 
-        if (current->fsuid != 0)
+        /* Allow non-root access for OBD_IOC_PING_TARGET - used by lfs check */
+        if (!cfs_capable(CFS_CAP_SYS_ADMIN) && (cmd != OBD_IOC_PING_TARGET))
                 RETURN(err = -EACCES);
         if ((cmd & 0xffffff00) == ((int)'T') << 8) /* ignore all tty ioctls */
                 RETURN(err = -ENOTTY);
 
-        if (obd_psdev_ops.p_ioctl != NULL)
-                err = obd_psdev_ops.p_ioctl(NULL, cmd, (void *)arg);
-        else
-                err = -EPERM;
+        err = class_handle_ioctl(cmd, (unsigned long)arg);
 
         RETURN(err);
 }
@@ -220,15 +225,19 @@ static int obd_class_ioctl(struct inode *inode, struct file *filp,
 /* declare character device */
 static struct file_operations obd_psdev_fops = {
         .owner   = THIS_MODULE,
+#if HAVE_UNLOCKED_IOCTL
+        .unlocked_ioctl = obd_class_ioctl, /* unlocked_ioctl */
+#else
         .ioctl   = obd_class_ioctl,     /* ioctl */
+#endif
         .open    = obd_class_open,      /* open */
         .release = obd_class_release,   /* release */
 };
 
 /* modules setup */
 cfs_psdev_t obd_psdev = {
-        .minor = OBD_MINOR,
-        .name  = "obd_psdev",
+        .minor = OBD_DEV_MINOR,
+        .name  = OBD_DEV_NAME,
         .fops  = &obd_psdev_fops,
 };
 
@@ -239,14 +248,9 @@ int obd_proc_read_version(char *page, char **start, off_t off, int count,
                           int *eof, void *data)
 {
         *eof = 1;
-#ifdef HAVE_VFS_INTENT_PATCHES
-        return snprintf(page, count, "lustre: %s\nkernel: %u\nbuild:  %s\n",
-                        LUSTRE_VERSION_STRING, LUSTRE_KERNEL_VERSION,
-                        BUILD_VERSION);
-#else
         return snprintf(page, count, "lustre: %s\nkernel: %s\nbuild:  %s\n",
-                        LUSTRE_VERSION_STRING, "patchless", BUILD_VERSION);
-#endif
+                        LUSTRE_VERSION_STRING, "patchless_client",
+                        BUILD_VERSION);
 }
 
 int obd_proc_read_pinger(char *page, char **start, off_t off, int count,
@@ -286,7 +290,7 @@ static int obd_proc_read_health(char *page, char **start, off_t off,
         if (libcfs_catastrophe)
                 rc += snprintf(page + rc, count - rc, "LBUG\n");
 
-        spin_lock(&obd_dev_lock);
+        cfs_spin_lock(&obd_dev_lock);
         for (i = 0; i < class_devno_max(); i++) {
                 struct obd_device *obd;
 
@@ -298,18 +302,18 @@ static int obd_proc_read_health(char *page, char **start, off_t off,
                 if (obd->obd_stopping)
                         continue;
 
-                class_incref(obd);
-                spin_unlock(&obd_dev_lock);
+                class_incref(obd, __FUNCTION__, cfs_current());
+                cfs_spin_unlock(&obd_dev_lock);
 
                 if (obd_health_check(obd)) {
                         rc += snprintf(page + rc, count - rc,
                                        "device %s reported unhealthy\n",
                                        obd->obd_name);
                 }
-                class_decref(obd);
-                spin_lock(&obd_dev_lock);
+                class_decref(obd, __FUNCTION__, cfs_current());
+                cfs_spin_lock(&obd_dev_lock);
         }
-        spin_unlock(&obd_dev_lock);
+        cfs_spin_unlock(&obd_dev_lock);
 
         if (rc == 0)
                 return snprintf(page, count, "healthy\n");
@@ -345,7 +349,7 @@ static void obd_device_list_seq_stop(struct seq_file *p, void *v)
 }
 
 static void *obd_device_list_seq_next(struct seq_file *p, void *v, loff_t *pos)
-{ 
+{
         ++*pos;
         if (*pos >= class_devno_max())
                 return NULL;
@@ -377,7 +381,7 @@ static int obd_device_list_seq_show(struct seq_file *p, void *v)
         return seq_printf(p, "%3d %s %s %s %s %d\n",
                           (int)index, status, obd->obd_type->typ_name,
                           obd->obd_name, obd->obd_uuid.uuid,
-                          atomic_read(&obd->obd_refcount));
+                          cfs_atomic_read(&obd->obd_refcount));
 }
 
 struct seq_operations obd_device_list_sops = {
@@ -418,7 +422,7 @@ int class_procfs_init(void)
         ENTRY;
 
         obd_sysctl_init();
-        proc_lustre_root = lprocfs_register("lustre", proc_root_fs,
+        proc_lustre_root = lprocfs_register("fs/lustre", NULL,
                                             lprocfs_base, NULL);
         rc = lprocfs_seq_create(proc_lustre_root, "devices", 0444,
                                 &obd_device_list_fops, NULL);