chiark / gitweb /
log_error() if inotify_add_watch() fails
[elogind.git] / src / readahead / readahead-common.c
index 4e8e63697541f51107ab5be2c1b93337935fe36b..1599ac15a476b628bc27b86a0176a446bf84bd86 100644 (file)
@@ -32,6 +32,7 @@
 #include "log.h"
 #include "readahead-common.h"
 #include "util.h"
+#include "missing.h"
 
 int file_verify(int fd, const char *fn, off_t file_size_max, struct stat *st) {
         assert(fd >= 0);
@@ -62,23 +63,74 @@ int fs_on_ssd(const char *p) {
         struct udev_device *udev_device = NULL, *look_at = NULL;
         bool b = false;
         const char *devtype, *rotational, *model, *id;
+        int r;
 
         assert(p);
 
         if (stat(p, &st) < 0)
                 return -errno;
 
-        if (major(st.st_dev) == 0)
+        if (major(st.st_dev) == 0) {
+                _cleanup_fclose_ FILE *f = NULL;
+                int mount_id;
+                struct file_handle *h;
+
+                /* Might be btrfs, which exposes "ssd" as mount flag if it is on ssd.
+                 *
+                 * We first determine the mount ID here, if we can,
+                 * and then lookup the mount ID in mountinfo to find
+                 * the mount options. */
+
+                h = alloca(MAX_HANDLE_SZ);
+                h->handle_bytes = MAX_HANDLE_SZ;
+                r = name_to_handle_at(AT_FDCWD, p, h, &mount_id, AT_SYMLINK_FOLLOW);
+                if (r < 0)
+                        return false;
+
+                f = fopen("/proc/self/mountinfo", "re");
+                if (!f)
+                        return false;
+
+                for (;;) {
+                        char line[LINE_MAX], *e;
+                        _cleanup_free_ char *opts = NULL;
+                        int mid;
+
+                        if (!fgets(line, sizeof(line), f))
+                                return false;
+
+                        truncate_nl(line);
+
+                        if (sscanf(line, "%i", &mid) != 1)
+                                continue;
+
+                        if (mid != mount_id)
+                                continue;
+
+                        e = strstr(line, " - ");
+                        if (!e)
+                                continue;
+
+                        if (sscanf(e+3, "%*s %*s %ms", &opts) != 1)
+                                continue;
+
+                        if (streq(opts, "ssd") || startswith(opts, "ssd,") || endswith(opts, ",ssd") || strstr(opts, ",ssd,"))
+                                return true;
+                }
+
                 return false;
+        }
 
-        if (!(udev = udev_new()))
+        udev = udev_new();
+        if (!udev)
                 return -ENOMEM;
 
-        if (!(udev_device = udev_device_new_from_devnum(udev, 'b', st.st_dev)))
+        udev_device = udev_device_new_from_devnum(udev, 'b', st.st_dev);
+        if (!udev_device)
                 goto finish;
 
-        if ((devtype = udev_device_get_property_value(udev_device, "DEVTYPE")) &&
-            streq(devtype, "partition"))
+        devtype = udev_device_get_property_value(udev_device, "DEVTYPE");
+        if (devtype && streq(devtype, "partition"))
                 look_at = udev_device_get_parent(udev_device);
         else
                 look_at = udev_device;
@@ -87,21 +139,26 @@ int fs_on_ssd(const char *p) {
                 goto finish;
 
         /* First, try high-level property */
-        if ((id = udev_device_get_property_value(look_at, "ID_SSD"))) {
+        id = udev_device_get_property_value(look_at, "ID_SSD");
+        if (id) {
                 b = streq(id, "1");
                 goto finish;
         }
 
         /* Second, try kernel attribute */
-        if ((rotational = udev_device_get_sysattr_value(look_at, "queue/rotational")))
-                if ((b = streq(rotational, "0")))
-                        goto finish;
+        rotational = udev_device_get_sysattr_value(look_at, "queue/rotational");
+        if (rotational) {
+                b = streq(rotational, "0");
+                goto finish;
+        }
 
         /* Finally, fallback to heuristics */
-        if (!(look_at = udev_device_get_parent(look_at)))
+        look_at = udev_device_get_parent(look_at);
+        if (!look_at)
                 goto finish;
 
-        if ((model = udev_device_get_sysattr_value(look_at, "model")))
+        model = udev_device_get_sysattr_value(look_at, "model");
+        if (model)
                 b = !!strstr(model, "SSD");
 
 finish:
@@ -170,7 +227,7 @@ int open_inotify(void) {
         mkdir("/run/systemd/readahead", 0755);
 
         if (inotify_add_watch(fd, "/run/systemd/readahead", IN_CREATE) < 0) {
-                log_error("Failed to watch /run/systemd/readahead: %m");
+                log_error("Failed to add watch on /run/systemd/readahead: %m");
                 close_nointr_nofail(fd);
                 return -errno;
         }
@@ -208,9 +265,13 @@ finish:
         return m;
 }
 
-#define BUMP_REQUEST_NR (16*1024)
+/* We use 20K instead of the more human digestable 16K here. Why?
+   Simply so that it is more unlikely that users end up picking this
+   value too so that we can recognize better whether the user changed
+   the value while we had it temporarily bumped. */
+#define BUMP_REQUEST_NR (20*1024)
 
-int bump_request_nr(const char *p) {
+int block_bump_request_nr(const char *p) {
         struct stat st;
         uint64_t u;
         char *ap = NULL, *line = NULL;
@@ -267,3 +328,83 @@ finish:
 
         return r;
 }
+
+int block_get_readahead(const char *p, uint64_t *bytes) {
+        struct stat st;
+        char *ap = NULL, *line = NULL;
+        int r;
+        dev_t d;
+        uint64_t u;
+
+        assert(p);
+        assert(bytes);
+
+        if (stat(p, &st) < 0)
+                return -errno;
+
+        if (major(st.st_dev) == 0)
+                return 0;
+
+        d = st.st_dev;
+        block_get_whole_disk(d, &d);
+
+        if (asprintf(&ap, "/sys/dev/block/%u:%u/bdi/read_ahead_kb", major(d), minor(d)) < 0) {
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        r = read_one_line_file(ap, &line);
+        if (r < 0)
+                goto finish;
+
+        r = safe_atou64(line, &u);
+        if (r < 0)
+                goto finish;
+
+        *bytes = u * 1024ULL;
+
+finish:
+        free(ap);
+        free(line);
+
+        return r;
+}
+
+int block_set_readahead(const char *p, uint64_t bytes) {
+        struct stat st;
+        char *ap = NULL, *line = NULL;
+        int r;
+        dev_t d;
+
+        assert(p);
+        assert(bytes);
+
+        if (stat(p, &st) < 0)
+                return -errno;
+
+        if (major(st.st_dev) == 0)
+                return 0;
+
+        d = st.st_dev;
+        block_get_whole_disk(d, &d);
+
+        if (asprintf(&ap, "/sys/dev/block/%u:%u/bdi/read_ahead_kb", major(d), minor(d)) < 0) {
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        if (asprintf(&line, "%llu", (unsigned long long) bytes / 1024ULL) < 0) {
+                r = -ENOMEM;
+                goto finish;
+        }
+
+        r = write_one_line_file(ap, line);
+        if (r < 0)
+                goto finish;
+
+finish:
+        free(ap);
+        free(line);
+
+        return r;
+}