Merge pull request #20257 from bluca/seqno

Use new diskseq block device property

src/basic/missing_loop.h

@@ -1,6 +1,7 @@
 /* SPDX-License-Identifier: LGPL-2.1-or-later */
 #pragma once
 
+#include <linux/fs.h>
 #include <linux/loop.h>
 
 #ifndef LOOP_CONFIGURE
@@ -13,3 +14,7 @@ struct loop_config {
 
 #define LOOP_CONFIGURE 0x4C0A
 #endif
+
+#ifndef BLKGETDISKSEQ
+#define BLKGETDISKSEQ _IOR(0x12,128,__u64)
+#endif

src/core/namespace.c

@@ -1905,6 +1905,7 @@ int setup_namespace(
                                 loop_device->fd,
                                 &verity,
                                 root_image_options,
+                                loop_device->diskseq,
                                 loop_device->uevent_seqnum_not_before,
                                 loop_device->timestamp_not_before,
                                 dissect_image_flags,

src/dissect/dissect.c

@@ -795,6 +795,7 @@ static int run(int argc, char *argv[]) {
                         arg_image,
                         &arg_verity_settings,
                         NULL,
+                        d->diskseq,
                         d->uevent_seqnum_not_before,
                         d->timestamp_not_before,
                         arg_flags,

src/gpt-auto-generator/gpt-auto-generator.c

@@ -699,6 +699,7 @@ static int enumerate_partitions(dev_t devnum) {
         r = dissect_image(
                         fd,
                         NULL, NULL,
+                        /* diskseq= */ 0,
                         UINT64_MAX,
                         USEC_INFINITY,
                         DISSECT_IMAGE_GPT_ONLY|

src/libsystemd/libsystemd.sym

@@ -762,3 +762,8 @@ global:
         sd_device_new_from_ifname;
         sd_device_new_from_ifindex;
 } LIBSYSTEMD_248;
+
+LIBSYSTEMD_250 {
+global:
+        sd_device_get_diskseq;
+} LIBSYSTEMD_249;

src/libsystemd/sd-device/device-internal.h

@@ -75,6 +75,8 @@ struct sd_device {
         uid_t devuid;
         gid_t devgid;
 
+        uint64_t diskseq; /* Block device sequence number, monothonically incremented by the kernel on create/attach */
+
         /* only set when device is passed through netlink */
         sd_device_action_t action;
         uint64_t seqnum;

src/libsystemd/sd-device/device-private.c

@@ -226,6 +226,28 @@ static int device_set_seqnum(sd_device *device, const char *str) {
         return 0;
 }
 
+static int device_set_diskseq(sd_device *device, const char *str) {
+        uint64_t diskseq;
+        int r;
+
+        assert(device);
+        assert(str);
+
+        r = safe_atou64(str, &diskseq);
+        if (r < 0)
+                return r;
+        if (diskseq == 0)
+                return -EINVAL;
+
+        r = device_add_property_internal(device, "DISKSEQ", str);
+        if (r < 0)
+                return r;
+
+        device->diskseq = diskseq;
+
+        return 0;
+}
+
 static int device_amend(sd_device *device, const char *key, const char *value) {
         int r;
 
@@ -292,6 +314,10 @@ static int device_amend(sd_device *device, const char *key, const char *value) {
                 r = device_set_seqnum(device, value);
                 if (r < 0)
                         return log_device_debug_errno(device, r, "sd-device: Failed to set SEQNUM to '%s': %m", value);
+        } else if (streq(key, "DISKSEQ")) {
+                r = device_set_diskseq(device, value);
+                if (r < 0)
+                        return log_device_debug_errno(device, r, "sd-device: Failed to set DISKSEQ to '%s': %m", value);
         } else if (streq(key, "DEVLINKS")) {
                 for (const char *p = value;;) {
                         _cleanup_free_ char *word = NULL;

src/libsystemd/sd-device/sd-device.c

@@ -1140,6 +1140,18 @@ _public_ int sd_device_get_seqnum(sd_device *device, uint64_t *ret) {
         return 0;
 }
 
+_public_ int sd_device_get_diskseq(sd_device *device, uint64_t *ret) {
+        assert_return(device, -EINVAL);
+
+        if (device->diskseq == 0)
+                return -ENOENT;
+
+        if (ret)
+                *ret = device->diskseq;
+
+        return 0;
+}
+
 static bool is_valid_tag(const char *tag) {
         assert(tag);
 

src/nspawn/nspawn.c

@@ -5692,6 +5692,7 @@ static int run(int argc, char *argv[]) {
                                 arg_image,
                                 &arg_verity_settings,
                                 NULL,
+                                loop->diskseq,
                                 loop->uevent_seqnum_not_before,
                                 loop->timestamp_not_before,
                                 dissect_image_flags,

src/portable/portable.c

@@ -393,6 +393,7 @@ static int portable_extract_by_path(
                 r = dissect_image(
                                 d->fd,
                                 NULL, NULL,
+                                d->diskseq,
                                 d->uevent_seqnum_not_before,
                                 d->timestamp_not_before,
                                 DISSECT_IMAGE_READ_ONLY |

src/shared/discover-image.c

@@ -1203,6 +1203,7 @@ int image_read_metadata(Image *i) {
                 r = dissect_image(
                                 d->fd,
                                 NULL, NULL,
+                                d->diskseq,
                                 d->uevent_seqnum_not_before,
                                 d->timestamp_not_before,
                                 DISSECT_IMAGE_GENERIC_ROOT |

src/shared/dissect-image.c

@@ -284,6 +284,7 @@ struct wait_data {
         sd_device *parent_device;
         blkid_partition blkidp;
         sd_device *found;
+        uint64_t diskseq;
         uint64_t uevent_seqnum_not_before;
         usec_t timestamp_not_before;
         DissectImageFlags flags;
@@ -302,7 +303,25 @@ static int device_monitor_handler(sd_device_monitor *monitor, sd_device *device,
         if (device_for_action(device, SD_DEVICE_REMOVE))
                 return 0;
 
-        if (w->uevent_seqnum_not_before != UINT64_MAX) {
+        if (w->diskseq != 0) {
+                uint64_t diskseq;
+
+                /* If w->diskseq is non-zero, then we must have a disk seqnum */
+                r = sd_device_get_diskseq(device, &diskseq);
+                if (r < 0) {
+                        log_debug_errno(r, "Dropping event because it has no diskseq, but waiting for %" PRIu64, w->diskseq);
+                        return 0;
+                }
+                if (diskseq < w->diskseq) {
+                        log_debug("Dropping event because diskseq too old (%" PRIu64 " < %" PRIu64 ")",
+                                  diskseq, w->diskseq);
+                        return 0;
+                }
+                if (diskseq > w->diskseq) {
+                        r = -EBUSY;
+                        goto finish; /* Newer than what we were expecting, so we missed it, stop waiting */
+                }
+        } else if (w->uevent_seqnum_not_before != UINT64_MAX) {
                 uint64_t seqnum;
 
                 r = sd_device_get_seqnum(device, &seqnum);
@@ -378,6 +397,7 @@ static int wait_for_partition_device(
                 sd_device *parent,
                 blkid_partition pp,
                 usec_t deadline,
+                uint64_t diskseq,
                 uint64_t uevent_seqnum_not_before,
                 usec_t timestamp_not_before,
                 DissectImageFlags flags,
@@ -423,6 +443,7 @@ static int wait_for_partition_device(
         _cleanup_(wait_data_done) struct wait_data w = {
                 .parent_device = parent,
                 .blkidp = pp,
+                .diskseq = diskseq,
                 .uevent_seqnum_not_before = uevent_seqnum_not_before,
                 .timestamp_not_before = timestamp_not_before,
                 .flags = flags,
@@ -450,16 +471,21 @@ static int wait_for_partition_device(
                         return r;
         }
 
-        r = sd_event_add_time_relative(
-                        event, &retry_source,
-                        CLOCK_MONOTONIC, 500 * USEC_PER_MSEC, 0,
-                        retry_handler, &w);
-        if (r < 0)
-                return r;
+        /* If we don't have a disk sequence number then we cannot do exact matching,
+         * and we cannot know if we missed it or if it has not been sent yet, so set
+         * up additional retries to increase the chances of receiving the event. */
+        if (diskseq == 0) {
+                r = sd_event_add_time_relative(
+                                event, &retry_source,
+                                CLOCK_MONOTONIC, 500 * USEC_PER_MSEC, 0,
+                                retry_handler, &w);
+                if (r < 0)
+                        return r;
 
-        r = sd_event_source_set_exit_on_failure(retry_source, true);
-        if (r < 0)
-                return r;
+                r = sd_event_source_set_exit_on_failure(retry_source, true);
+                if (r < 0)
+                        return r;
+        }
 
         r = sd_event_loop(event);
         if (r < 0)
@@ -583,6 +609,7 @@ int dissect_image(
                 int fd,
                 const VeritySettings *verity,
                 const MountOptions *mount_options,
+                uint64_t diskseq,
                 uint64_t uevent_seqnum_not_before,
                 usec_t timestamp_not_before,
                 DissectImageFlags flags,
@@ -838,7 +865,7 @@ int dissect_image(
                 if (!pp)
                         return errno_or_else(EIO);
 
-                r = wait_for_partition_device(d, pp, deadline, uevent_seqnum_not_before, timestamp_not_before, flags, &q);
+                r = wait_for_partition_device(d, pp, deadline, diskseq, uevent_seqnum_not_before, timestamp_not_before, flags, &q);
                 if (r < 0)
                         return r;
 
@@ -2753,6 +2780,7 @@ int dissect_image_and_warn(
                 const char *name,
                 const VeritySettings *verity,
                 const MountOptions *mount_options,
+                uint64_t diskseq,
                 uint64_t uevent_seqnum_not_before,
                 usec_t timestamp_not_before,
                 DissectImageFlags flags,
@@ -2769,7 +2797,7 @@ int dissect_image_and_warn(
                 name = buffer;
         }
 
-        r = dissect_image(fd, verity, mount_options, uevent_seqnum_not_before, timestamp_not_before, flags, ret);
+        r = dissect_image(fd, verity, mount_options, diskseq, uevent_seqnum_not_before, timestamp_not_before, flags, ret);
         switch (r) {
 
         case -EOPNOTSUPP:
@@ -2877,7 +2905,7 @@ int mount_image_privately_interactively(
         if (r < 0)
                 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
 
-        r = dissect_image_and_warn(d->fd, image, &verity, NULL, d->uevent_seqnum_not_before, d->timestamp_not_before, flags, &dissected_image);
+        r = dissect_image_and_warn(d->fd, image, &verity, NULL, d->diskseq, d->uevent_seqnum_not_before, d->timestamp_not_before, flags, &dissected_image);
         if (r < 0)
                 return r;
 
@@ -2968,6 +2996,7 @@ int verity_dissect_and_mount(
                         loop_device->fd,
                         &verity,
                         options,
+                        loop_device->diskseq,
                         loop_device->uevent_seqnum_not_before,
                         loop_device->timestamp_not_before,
                         dissect_image_flags,
@@ -2978,6 +3007,7 @@ int verity_dissect_and_mount(
                                 loop_device->fd,
                                 &verity,
                                 options,
+                                loop_device->diskseq,
                                 loop_device->uevent_seqnum_not_before,
                                 loop_device->timestamp_not_before,
                                 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,

src/shared/dissect-image.h

@@ -162,8 +162,8 @@ DEFINE_TRIVIAL_CLEANUP_FUNC(MountOptions*, mount_options_free_all);
 const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator);
 
 int probe_filesystem(const char *node, char **ret_fstype);
-int dissect_image(int fd, const VeritySettings *verity, const MountOptions *mount_options, uint64_t uevent_seqnum_not_before, usec_t timestamp_not_before, DissectImageFlags flags, DissectedImage **ret);
-int dissect_image_and_warn(int fd, const char *name, const VeritySettings *verity, const MountOptions *mount_options, uint64_t uevent_seqnum_not_before, usec_t timestamp_not_before, DissectImageFlags flags, DissectedImage **ret);
+int dissect_image(int fd, const VeritySettings *verity, const MountOptions *mount_options, uint64_t diskseq, uint64_t uevent_seqnum_not_before, usec_t timestamp_not_before, DissectImageFlags flags, DissectedImage **ret);
+int dissect_image_and_warn(int fd, const char *name, const VeritySettings *verity, const MountOptions *mount_options, uint64_t diskseq, uint64_t uevent_seqnum_not_before, usec_t timestamp_not_before, DissectImageFlags flags, DissectedImage **ret);
 
 DissectedImage* dissected_image_unref(DissectedImage *m);
 DEFINE_TRIVIAL_CLEANUP_FUNC(DissectedImage*, dissected_image_unref);

src/shared/loop-util.c

@@ -127,6 +127,27 @@ static int device_has_block_children(sd_device *d) {
         return 0;
 }
 
+static int loop_get_diskseq(int fd, uint64_t *ret_diskseq) {
+        uint64_t diskseq;
+
+        assert(fd >= 0);
+        assert(ret_diskseq);
+
+        if (ioctl(fd, BLKGETDISKSEQ, &diskseq) < 0) {
+                /* Note that the kernel is weird: non-existing ioctls currently return EINVAL
+                 * rather than ENOTTY on loopback block devices. They should fix that in the kernel,
+                 * but in the meantime we accept both here. */
+                if (!ERRNO_IS_NOT_SUPPORTED(errno) && errno != EINVAL)
+                        return -errno;
+
+                return -EOPNOTSUPP;
+        }
+
+        *ret_diskseq = diskseq;
+
+        return 0;
+}
+
 static int loop_configure(
                 int fd,
                 int nr,
@@ -388,6 +409,7 @@ int loop_device_make(
 
                 if (offset == 0 && IN_SET(size, 0, UINT64_MAX)) {
                         _cleanup_close_ int copy = -1;
+                        uint64_t diskseq = 0;
 
                         /* If this is already a block device, store a copy of the fd as it is */
 
@@ -395,6 +417,10 @@ int loop_device_make(
                         if (copy < 0)
                                 return -errno;
 
+                        r = loop_get_diskseq(copy, &diskseq);
+                        if (r < 0 && r != -EOPNOTSUPP)
+                                return r;
+
                         d = new(LoopDevice, 1);
                         if (!d)
                                 return -ENOMEM;
@@ -404,6 +430,7 @@ int loop_device_make(
                                 .node = TAKE_PTR(loopdev),
                                 .relinquished = true, /* It's not allocated by us, don't destroy it when this object is freed */
                                 .devno = st.st_rdev,
+                                .diskseq = diskseq,
                                 .uevent_seqnum_not_before = UINT64_MAX,
                                 .timestamp_not_before = USEC_INFINITY,
                         };
@@ -482,6 +509,11 @@ int loop_device_make(
                 return -errno;
         assert(S_ISBLK(st.st_mode));
 
+        uint64_t diskseq = 0;
+        r = loop_get_diskseq(loop_with_fd, &diskseq);
+        if (r < 0 && r != -EOPNOTSUPP)
+                return r;
+
         d = new(LoopDevice, 1);
         if (!d)
                 return -ENOMEM;
@@ -490,6 +522,7 @@ int loop_device_make(
                 .node = TAKE_PTR(loopdev),
                 .nr = nr,
                 .devno = st.st_rdev,
+                .diskseq = diskseq,
                 .uevent_seqnum_not_before = seqnum,
                 .timestamp_not_before = timestamp,
         };

src/shared/loop-util.h

@@ -14,6 +14,7 @@ struct LoopDevice {
         dev_t devno;
         char *node;
         bool relinquished;
+        uint64_t diskseq; /* Block device sequence number, monothonically incremented by the kernel on create/attach, or 0 if we don't know */
         uint64_t uevent_seqnum_not_before; /* uevent sequm right before we attached the loopback device, or UINT64_MAX if we don't know */
         usec_t timestamp_not_before; /* CLOCK_MONOTONIC timestamp taken immediately before attaching the loopback device, or USEC_INFINITY if we don't know */
 };

src/sysext/sysext.c

@@ -532,6 +532,7 @@ static int merge_subprocess(Hashmap *images, const char *workspace) {
                                         img->path,
                                         &verity_settings,
                                         NULL,
+                                        d->diskseq,
                                         d->uevent_seqnum_not_before,
                                         d->timestamp_not_before,
                                         flags,

src/systemd/sd-device.h

@@ -80,6 +80,7 @@ int sd_device_get_sysname(sd_device *device, const char **ret);
 int sd_device_get_sysnum(sd_device *device, const char **ret);
 int sd_device_get_action(sd_device *device, sd_device_action_t *ret);
 int sd_device_get_seqnum(sd_device *device, uint64_t *ret);
+int sd_device_get_diskseq(sd_device *device, uint64_t *ret);
 
 int sd_device_get_is_initialized(sd_device *device);
 int sd_device_get_usec_initialized(sd_device *device, uint64_t *usec);

src/test/test-loop-block.c

@@ -51,7 +51,7 @@ static void* thread_func(void *ptr) {
 
                 log_notice("Acquired loop device %s, will mount on %s", loop->node, mounted);
 
-                r = dissect_image(loop->fd, NULL, NULL, loop->uevent_seqnum_not_before, loop->timestamp_not_before, DISSECT_IMAGE_READ_ONLY, &dissected);
+                r = dissect_image(loop->fd, NULL, NULL, loop->diskseq, loop->uevent_seqnum_not_before, loop->timestamp_not_before, DISSECT_IMAGE_READ_ONLY, &dissected);
                 if (r < 0)
                         log_error_errno(r, "Failed dissect loopback device %s: %m", loop->node);
                 assert_se(r >= 0);
@@ -188,7 +188,7 @@ int main(int argc, char *argv[]) {
         sfdisk = NULL;
 
         assert_se(loop_device_make(fd, O_RDWR, 0, UINT64_MAX, LO_FLAGS_PARTSCAN, &loop) >= 0);
-        assert_se(dissect_image(loop->fd, NULL, NULL, loop->uevent_seqnum_not_before, loop->timestamp_not_before, 0, &dissected) >= 0);
+        assert_se(dissect_image(loop->fd, NULL, NULL, loop->diskseq, loop->uevent_seqnum_not_before, loop->timestamp_not_before, 0, &dissected) >= 0);
 
         assert_se(dissected->partitions[PARTITION_ESP].found);
         assert_se(dissected->partitions[PARTITION_ESP].node);
@@ -212,7 +212,7 @@ int main(int argc, char *argv[]) {
         assert_se(make_filesystem(dissected->partitions[PARTITION_HOME].node, "ext4", "home", id, true) >= 0);
 
         dissected = dissected_image_unref(dissected);
-        assert_se(dissect_image(loop->fd, NULL, NULL, loop->uevent_seqnum_not_before, loop->timestamp_not_before, 0, &dissected) >= 0);
+        assert_se(dissect_image(loop->fd, NULL, NULL, loop->diskseq, loop->uevent_seqnum_not_before, loop->timestamp_not_before, 0, &dissected) >= 0);
 
         assert_se(mkdtemp_malloc(NULL, &mounted) >= 0);