Whamcloud - gitweb
LU-10698 obdclass: allow specifying complex jobids
[fs/lustre-release.git] / lustre / obdclass / linux / linux-module.c
index e0a54c5..464d288 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * 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
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -45,7 +41,6 @@
 #include <linux/module.h>
 #include <linux/errno.h>
 #include <linux/kernel.h>
-#include <linux/major.h>
 #include <linux/sched.h>
 #include <linux/lp.h>
 #include <linux/slab.h>
 #include <asm/uaccess.h>
 #include <linux/miscdevice.h>
 #include <linux/seq_file.h>
+#include <linux/kobject.h>
 
 #include <libcfs/libcfs.h>
 #include <obd_support.h>
 #include <obd_class.h>
-#include <lnet/lnetctl.h>
 #include <lprocfs_status.h>
-#include <lustre_ioctl.h>
-#include <lustre_ver.h>
+#include <uapi/linux/lnet/lnetctl.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
+#include <uapi/linux/lustre/lustre_ver.h>
+
+#define OBD_MAX_IOCTL_BUFFER   8192
+
+static int obd_ioctl_is_invalid(struct obd_ioctl_data *data)
+{
+       if (data->ioc_len > BIT(30)) {
+               CERROR("OBD ioctl: ioc_len larger than 1<<30\n");
+               return 1;
+       }
+
+       if (data->ioc_inllen1 > BIT(30)) {
+               CERROR("OBD ioctl: ioc_inllen1 larger than 1<<30\n");
+               return 1;
+       }
+
+       if (data->ioc_inllen2 > BIT(30)) {
+               CERROR("OBD ioctl: ioc_inllen2 larger than 1<<30\n");
+               return 1;
+       }
+
+       if (data->ioc_inllen3 > BIT(30)) {
+               CERROR("OBD ioctl: ioc_inllen3 larger than 1<<30\n");
+               return 1;
+       }
+
+       if (data->ioc_inllen4 > BIT(30)) {
+               CERROR("OBD ioctl: ioc_inllen4 larger than 1<<30\n");
+               return 1;
+       }
+
+       if (data->ioc_inlbuf1 && data->ioc_inllen1 == 0) {
+               CERROR("OBD ioctl: inlbuf1 pointer but 0 length\n");
+               return 1;
+       }
+
+       if (data->ioc_inlbuf2 && data->ioc_inllen2 == 0) {
+               CERROR("OBD ioctl: inlbuf2 pointer but 0 length\n");
+               return 1;
+       }
 
-int proc_version;
+       if (data->ioc_inlbuf3 && data->ioc_inllen3 == 0) {
+               CERROR("OBD ioctl: inlbuf3 pointer but 0 length\n");
+               return 1;
+       }
+
+       if (data->ioc_inlbuf4 && data->ioc_inllen4 == 0) {
+               CERROR("OBD ioctl: inlbuf4 pointer but 0 length\n");
+               return 1;
+       }
+
+       if (data->ioc_pbuf1 && data->ioc_plen1 == 0) {
+               CERROR("OBD ioctl: pbuf1 pointer but 0 length\n");
+               return 1;
+       }
+
+       if (data->ioc_pbuf2 && data->ioc_plen2 == 0) {
+               CERROR("OBD ioctl: pbuf2 pointer but 0 length\n");
+               return 1;
+       }
+
+       if (!data->ioc_pbuf1 && data->ioc_plen1 != 0) {
+               CERROR("OBD ioctl: plen1 set but NULL pointer\n");
+               return 1;
+       }
+
+       if (!data->ioc_pbuf2 && data->ioc_plen2 != 0) {
+               CERROR("OBD ioctl: plen2 set but NULL pointer\n");
+               return 1;
+       }
+
+       if (obd_ioctl_packlen(data) > data->ioc_len) {
+               CERROR("OBD ioctl: packlen exceeds ioc_len (%d > %d)\n",
+                      obd_ioctl_packlen(data), data->ioc_len);
+               return 1;
+       }
+
+       return 0;
+}
 
 /* buffer MUST be at least the size of obd_ioctl_hdr */
 int obd_ioctl_getdata(char **buf, int *len, void __user *arg)
@@ -150,15 +222,6 @@ int obd_ioctl_getdata(char **buf, int *len, void __user *arg)
 }
 EXPORT_SYMBOL(obd_ioctl_getdata);
 
-int obd_ioctl_popdata(void __user *arg, void *data, int len)
-{
-       int err;
-       ENTRY;
-
-       err = copy_to_user(arg, data, len) ? -EFAULT : 0;
-       RETURN(err);
-}
-
 /*  opening /dev/obd */
 static int obd_class_open(struct inode * inode, struct file * file)
 {
@@ -205,50 +268,153 @@ static struct file_operations obd_psdev_fops = {
 
 /* modules setup */
 struct miscdevice obd_psdev = {
-        .minor = OBD_DEV_MINOR,
-        .name  = OBD_DEV_NAME,
-        .fops  = &obd_psdev_fops,
+       .minor  = MISC_DYNAMIC_MINOR,
+       .name   = OBD_DEV_NAME,
+       .fops   = &obd_psdev_fops,
 };
 
+struct static_lustre_uintvalue_attr {
+       struct {
+               struct attribute attr;
+               ssize_t (*show)(struct kobject *kobj, struct attribute *attr,
+                               char *buf);
+               ssize_t (*store)(struct kobject *kobj, struct attribute *attr,
+                                const char *buf, size_t len);
+       } u;
+       int *value;
+};
 
-#ifdef CONFIG_PROC_FS
-static int obd_proc_version_seq_show(struct seq_file *m, void *v)
+static ssize_t static_uintvalue_show(struct kobject *kobj,
+                                    struct attribute *attr,
+                                    char *buf)
 {
-       seq_printf(m, "lustre: %s\nkernel: %s\nbuild:  %s\n",
-                  LUSTRE_VERSION_STRING, "patchless_client",
-                  LUSTRE_VERSION_STRING);
-       return 0;
+       struct static_lustre_uintvalue_attr *lattr = (void *)attr;
+
+       return sprintf(buf, "%d\n", *lattr->value);
+}
+
+static ssize_t static_uintvalue_store(struct kobject *kobj,
+                                     struct attribute *attr,
+                                     const char *buffer, size_t count)
+{
+       struct static_lustre_uintvalue_attr *lattr = (void *)attr;
+       unsigned int val;
+       int rc;
+
+       rc = kstrtouint(buffer, 10, &val);
+       if (rc)
+               return rc;
+
+       *lattr->value = val;
+
+       return count;
+}
+
+#define LUSTRE_STATIC_UINT_ATTR(name, value)                           \
+static struct static_lustre_uintvalue_attr lustre_sattr_##name =       \
+       { __ATTR(name, 0644, static_uintvalue_show,                     \
+                static_uintvalue_store), value }
+
+LUSTRE_STATIC_UINT_ATTR(timeout, &obd_timeout);
+LUSTRE_STATIC_UINT_ATTR(debug_peer_on_timeout, &obd_debug_peer_on_timeout);
+LUSTRE_STATIC_UINT_ATTR(dump_on_timeout, &obd_dump_on_timeout);
+LUSTRE_STATIC_UINT_ATTR(dump_on_eviction, &obd_dump_on_eviction);
+LUSTRE_STATIC_UINT_ATTR(at_min, &at_min);
+LUSTRE_STATIC_UINT_ATTR(at_max, &at_max);
+LUSTRE_STATIC_UINT_ATTR(at_extra, &at_extra);
+LUSTRE_STATIC_UINT_ATTR(at_early_margin, &at_early_margin);
+LUSTRE_STATIC_UINT_ATTR(at_history, &at_history);
+
+#ifdef HAVE_SERVER_SUPPORT
+LUSTRE_STATIC_UINT_ATTR(ldlm_timeout, &ldlm_timeout);
+LUSTRE_STATIC_UINT_ATTR(bulk_timeout, &bulk_timeout);
+#endif
+
+static ssize_t memused_show(struct kobject *kobj, struct attribute *attr,
+                           char *buf)
+{
+       return sprintf(buf, "%llu\n", obd_memory_sum());
+}
+LUSTRE_RO_ATTR(memused);
+
+static ssize_t memused_max_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
+{
+       return sprintf(buf, "%llu\n", obd_memory_max());
+}
+LUSTRE_RO_ATTR(memused_max);
+
+static ssize_t max_dirty_mb_show(struct kobject *kobj, struct attribute *attr,
+                                char *buf)
+{
+       return sprintf(buf, "%lu\n",
+                      obd_max_dirty_pages / (1 << (20 - PAGE_SHIFT)));
+}
+
+static ssize_t max_dirty_mb_store(struct kobject *kobj, struct attribute *attr,
+                                 const char *buffer, size_t count)
+{
+       unsigned long val;
+       int rc;
+
+       rc = kstrtoul(buffer, 10, &val);
+       if (rc)
+               return rc;
+
+       val *= 1 << (20 - PAGE_SHIFT); /* convert to pages */
+
+       if (val > ((totalram_pages / 10) * 9)) {
+               /* Somebody wants to assign too much memory to dirty pages */
+               return -EINVAL;
+       }
+
+       if (val < 4 << (20 - PAGE_SHIFT)) {
+               /* Less than 4 Mb for dirty cache is also bad */
+               return -EINVAL;
+       }
+
+       obd_max_dirty_pages = val;
+
+       return count;
+}
+LUSTRE_RW_ATTR(max_dirty_mb);
+
+static ssize_t version_show(struct kobject *kobj, struct attribute *attr,
+                           char *buf)
+{
+       return sprintf(buf, "%s\n", LUSTRE_VERSION_STRING);
 }
-LPROC_SEQ_FOPS_RO(obd_proc_version);
 
-static int obd_proc_pinger_seq_show(struct seq_file *m, void *v)
+static ssize_t pinger_show(struct kobject *kobj, struct attribute *attr,
+                          char *buf)
 {
-       seq_printf(m, "%s\n",
 #ifdef ENABLE_PINGER
-                  "on"
+       const char *state = "on";
 #else
-                  "off"
+       const char *state = "off";
 #endif
-                );
-       return 0;
+       return sprintf(buf, "%s\n", state);
 }
-LPROC_SEQ_FOPS_RO(obd_proc_pinger);
 
 /**
  * Check all obd devices health
  *
- * \param seq_file
- * \param data [in] unused
+ * \param kobj
+ * \param buf [in]
  *
  * \retval number of characters printed if healthy
  */
-static int obd_proc_health_seq_show(struct seq_file *m, void *data)
+static ssize_t
+health_check_show(struct kobject *kobj, struct attribute *attr, char *buf)
 {
        bool healthy = true;
+       size_t len = 0;
        int i;
 
-       if (libcfs_catastrophe)
-               seq_printf(m, "LBUG\n");
+       if (libcfs_catastrophe) {
+               len = sprintf(buf, "LBUG\n");
+               healthy = false;
+       }
 
        read_lock(&obd_dev_lock);
        for (i = 0; i < class_devno_max(); i++) {
@@ -266,8 +432,8 @@ static int obd_proc_health_seq_show(struct seq_file *m, void *data)
                read_unlock(&obd_dev_lock);
 
                if (obd_health_check(NULL, obd)) {
-                       seq_printf(m, "device %s reported unhealthy\n",
-                                  obd->obd_name);
+                       len = sprintf(buf, "device %s reported unhealthy\n",
+                                     obd->obd_name);
                        healthy = false;
                }
                class_decref(obd, __FUNCTION__, current);
@@ -276,32 +442,32 @@ static int obd_proc_health_seq_show(struct seq_file *m, void *data)
        read_unlock(&obd_dev_lock);
 
        if (healthy)
-               seq_puts(m, "healthy\n");
+               len = sprintf(buf, "healthy\n");
        else
-               seq_puts(m, "NOT HEALTHY\n");
-       return 0;
+               len = sprintf(buf, "NOT HEALTHY\n");
+
+       return len;
 }
-LPROC_SEQ_FOPS_RO(obd_proc_health);
 
-static int obd_proc_jobid_var_seq_show(struct seq_file *m, void *v)
+static ssize_t jobid_var_show(struct kobject *kobj, struct attribute *attr,
+                             char *buf)
 {
-       if (strlen(obd_jobid_var) != 0)
-               seq_printf(m, "%s\n", obd_jobid_var);
-       return 0;
+       int rc = 0;
+
+       if (strlen(obd_jobid_var))
+               rc = snprintf(buf, PAGE_SIZE, "%s\n", obd_jobid_var);
+       return rc;
 }
 
-static ssize_t
-obd_proc_jobid_var_seq_write(struct file *file, const char __user *buffer,
-                            size_t count, loff_t *off)
+static ssize_t jobid_var_store(struct kobject *kobj, struct attribute *attr,
+                              const char *buffer, size_t count)
 {
        if (!count || count > JOBSTATS_JOBID_VAR_MAX_LEN)
                return -EINVAL;
 
        memset(obd_jobid_var, 0, JOBSTATS_JOBID_VAR_MAX_LEN + 1);
 
-       /* This might leave the var invalid on error, which is probably fine.*/
-       if (copy_from_user(obd_jobid_var, buffer, count))
-               return -EFAULT;
+       memcpy(obd_jobid_var, buffer, count);
 
        /* Trim the trailing '\n' if any */
        if (obd_jobid_var[count - 1] == '\n')
@@ -309,61 +475,88 @@ obd_proc_jobid_var_seq_write(struct file *file, const char __user *buffer,
 
        return count;
 }
-LPROC_SEQ_FOPS(obd_proc_jobid_var);
 
-static int obd_proc_jobid_name_seq_show(struct seq_file *m, void *v)
+static ssize_t jobid_name_show(struct kobject *kobj, struct attribute *attr,
+                              char *buf)
 {
-       if (strlen(obd_jobid_node) != 0)
-               seq_printf(m, "%s\n", obd_jobid_node);
-       return 0;
+       int rc = 0;
+
+       if (strlen(obd_jobid_name))
+               rc = snprintf(buf, PAGE_SIZE, "%s\n", obd_jobid_name);
+       return rc;
 }
 
-static ssize_t obd_proc_jobid_name_seq_write(struct file *file,
-                                            const char __user *buffer,
-                                            size_t count, loff_t *off)
+static ssize_t jobid_name_store(struct kobject *kobj, struct attribute *attr,
+                               const char *buffer, size_t count)
 {
-       if (count == 0 || count > LUSTRE_JOBID_SIZE)
+       if (!count || count > LUSTRE_JOBID_SIZE)
                return -EINVAL;
 
+       if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) != 0 &&
+           !strchr(buffer, '%')) {
+               lustre_jobid_clear(buffer);
+               return count;
+       }
+
        /* clear previous value */
-       memset(obd_jobid_node, 0, LUSTRE_JOBID_SIZE);
+       memset(obd_jobid_name, 0, LUSTRE_JOBID_SIZE);
 
-       if (copy_from_user(obd_jobid_node, buffer, count))
-               return -EFAULT;
+       memcpy(obd_jobid_name, buffer, count);
 
        /* Trim the trailing '\n' if any */
-       if (obd_jobid_node[count - 1] == '\n') {
+       if (obd_jobid_name[count - 1] == '\n') {
                /* Don't echo just a newline */
                if (count == 1)
                        return -EINVAL;
-               obd_jobid_node[count - 1] = 0;
+               obd_jobid_name[count - 1] = 0;
        }
 
        return count;
 }
-LPROC_SEQ_FOPS(obd_proc_jobid_name);
 
+/* Root for /sys/kernel/debug/lustre */
+struct dentry *debugfs_lustre_root;
+EXPORT_SYMBOL_GPL(debugfs_lustre_root);
+
+#ifdef CONFIG_PROC_FS
 /* Root for /proc/fs/lustre */
 struct proc_dir_entry *proc_lustre_root = NULL;
 EXPORT_SYMBOL(proc_lustre_root);
-
-static struct lprocfs_vars lprocfs_base[] = {
-       { .name =       "version",
-         .fops =       &obd_proc_version_fops  },
-       { .name =       "pinger",
-         .fops =       &obd_proc_pinger_fops   },
-       { .name =       "health_check",
-         .fops =       &obd_proc_health_fops   },
-       { .name =       "jobid_var",
-         .fops =       &obd_proc_jobid_var_fops},
-       { .name =       "jobid_name",
-         .fops =       &obd_proc_jobid_name_fops},
-       { NULL }
-};
 #else
 #define lprocfs_base NULL
 #endif /* CONFIG_PROC_FS */
 
+LUSTRE_RO_ATTR(version);
+LUSTRE_RO_ATTR(pinger);
+LUSTRE_RO_ATTR(health_check);
+LUSTRE_RW_ATTR(jobid_var);
+LUSTRE_RW_ATTR(jobid_name);
+
+static struct attribute *lustre_attrs[] = {
+       &lustre_attr_version.attr,
+       &lustre_attr_pinger.attr,
+       &lustre_attr_health_check.attr,
+       &lustre_attr_jobid_name.attr,
+       &lustre_attr_jobid_var.attr,
+       &lustre_sattr_timeout.u.attr,
+       &lustre_attr_max_dirty_mb.attr,
+       &lustre_sattr_debug_peer_on_timeout.u.attr,
+       &lustre_sattr_dump_on_timeout.u.attr,
+       &lustre_sattr_dump_on_eviction.u.attr,
+       &lustre_sattr_at_min.u.attr,
+       &lustre_sattr_at_max.u.attr,
+       &lustre_sattr_at_extra.u.attr,
+       &lustre_sattr_at_early_margin.u.attr,
+       &lustre_sattr_at_history.u.attr,
+       &lustre_attr_memused_max.attr,
+       &lustre_attr_memused.attr,
+#ifdef HAVE_SERVER_SUPPORT
+       &lustre_sattr_ldlm_timeout.u.attr,
+       &lustre_sattr_bulk_timeout.u.attr,
+#endif
+       NULL,
+};
+
 static void *obd_device_list_seq_start(struct seq_file *p, loff_t *pos)
 {
         if (*pos >= class_devno_max())
@@ -429,7 +622,7 @@ static int obd_device_list_open(struct inode *inode, struct file *file)
                return rc;
 
        seq = file->private_data;
-       seq->private = PDE_DATA(inode);
+       seq->private = inode->i_private;
        return 0;
 }
 
@@ -441,44 +634,102 @@ static const struct file_operations obd_device_list_fops = {
         .release = seq_release,
 };
 
+struct kset *lustre_kset;
+EXPORT_SYMBOL_GPL(lustre_kset);
+
+static struct attribute_group lustre_attr_group = {
+       .attrs = lustre_attrs,
+};
+
+ssize_t class_set_global(const char *param)
+{
+       const char *value = strchr(param, '=') + 1;
+       size_t off = value - param - 1;
+       ssize_t count = -ENOENT;
+       int i;
+
+       for (i = 0; lustre_attrs[i]; i++) {
+               if (!strncmp(lustre_attrs[i]->name, param, off)) {
+                       count = lustre_attr_store(&lustre_kset->kobj,
+                                                 lustre_attrs[i], value,
+                                                 strlen(value));
+                       break;
+               }
+       }
+       return count;
+}
+
 int class_procfs_init(void)
 {
        struct proc_dir_entry *entry;
-       int rc;
+       struct dentry *file;
+       int rc = -ENOMEM;
        ENTRY;
 
-       obd_sysctl_init();
+       lustre_kset = kset_create_and_add("lustre", NULL, fs_kobj);
+       if (!lustre_kset)
+               goto out;
 
-       entry = lprocfs_register("fs/lustre", NULL, lprocfs_base, NULL);
-       if (IS_ERR(entry)) {
-               rc = PTR_ERR(entry);
-               CERROR("cannot create '/proc/fs/lustre': rc = %d\n", rc);
-               RETURN(rc);
+       /* Create the files associated with this kobject */
+       rc = sysfs_create_group(&lustre_kset->kobj, &lustre_attr_group);
+       if (rc) {
+               kset_unregister(lustre_kset);
+               goto out;
        }
 
-       proc_lustre_root = entry;
+       rc = jobid_cache_init();
+       if (rc) {
+               kset_unregister(lustre_kset);
+               goto out;
+       }
 
-       rc = lprocfs_seq_create(proc_lustre_root, "devices", 0444,
-                               &obd_device_list_fops, NULL);
-       if (rc < 0) {
-               CERROR("cannot create '/proc/fs/lustre/devices': rc = %d\n",
-                      rc);
-               GOTO(out_proc, rc);
+       debugfs_lustre_root = debugfs_create_dir("lustre", NULL);
+       if (IS_ERR_OR_NULL(debugfs_lustre_root)) {
+               rc = debugfs_lustre_root ? PTR_ERR(debugfs_lustre_root)
+                                        : -ENOMEM;
+               debugfs_lustre_root = NULL;
+               kset_unregister(lustre_kset);
+               goto out;
        }
 
-       RETURN(rc);
+       file = debugfs_create_file("devices", 0444, debugfs_lustre_root, NULL,
+                                  &obd_device_list_fops);
+       if (IS_ERR_OR_NULL(file)) {
+               rc = file ? PTR_ERR(file) : -ENOMEM;
+               debugfs_remove(debugfs_lustre_root);
+               kset_unregister(lustre_kset);
+               goto out;
+       }
 
-out_proc:
-       lprocfs_remove(&proc_lustre_root);
+       entry = lprocfs_register("fs/lustre", NULL, NULL, NULL);
+       if (IS_ERR(entry)) {
+               rc = PTR_ERR(entry);
+               CERROR("cannot create '/proc/fs/lustre': rc = %d\n", rc);
+               debugfs_remove_recursive(debugfs_lustre_root);
+               kset_unregister(lustre_kset);
+               goto out;
+       }
 
+       proc_lustre_root = entry;
+out:
        RETURN(rc);
 }
 
 int class_procfs_clean(void)
 {
-        ENTRY;
-        if (proc_lustre_root) {
-                lprocfs_remove(&proc_lustre_root);
-        }
-        RETURN(0);
+       ENTRY;
+
+       debugfs_remove_recursive(debugfs_lustre_root);
+
+       debugfs_lustre_root = NULL;
+       jobid_cache_fini();
+
+       if (proc_lustre_root)
+               lprocfs_remove(&proc_lustre_root);
+
+       sysfs_remove_group(&lustre_kset->kobj, &lustre_attr_group);
+
+       kset_unregister(lustre_kset);
+
+       RETURN(0);
 }