A file lives exactly as long as something holds it

Deleting a post, editing media out, replacing an avatar or a header, and removing a whole root deleted no file: every
one stayed on disk and publicly served from /media/files with a year-long immutable cache, its row orphaned. Now every
upload is a row, profile pictures too (Kind avatar or header, ProfileOfAvatarId), and each of those acts trashes what
it held, as does an upload never posted for a day and a dropped scheduled post. A trashed row is marked in one
conditional update (an upload attached meanwhile is left alone), its files move into media-trash, beside the media
root and outside what /media/files serves, and the janitor deletes them a day later. Nothing is deleted for looking
unused.

Along the way: a profile picture that isn't an image, or can't be read, answers 422 instead of being silently ignored
with a 200; a removed root's scheduled posts are dropped, so nothing of it publishes later; media rows get indexes
(they had none), and the janitor's first pass comes five minutes after boot instead of an hour.

`PrivaPub admin media audit [--fix]` compares the disk with the database. With --fix (as www-data) it gives the
pictures personas show today a row, and trashes media of deleted posts or personas, rows whose files are missing, and
files nothing holds: the leftovers of every deletion until now. MediaLifecycleTests covers each act, that the trash is
never served, and the audit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw

CLAUDE.md

@@ -297,19 +297,30 @@ group www-data and reaches the private mongod; `sudo -u www-data` works too.
 1. **No upload keeps its metadata.** Images are re-encoded by libvips with `keep=none`; audio and video are remuxed with
    `-map_metadata -1`. `MediaProcessingTests` checks EXIF and XMP are gone.
 2. Files live under `Media:Root` (`/var/lib/privapub/media`), never in the published directory; the proxy cache is the
-   sibling `media-proxy`, which `/media/files` does not serve.
+   sibling `media-proxy` and the trash the sibling `media-trash`, neither of which `/media/files` serves.
-3. **A client never contacts a remote server for media:** every remote URL the API returns goes through
+3. **A file lives exactly as long as something holds it.** Every upload is a `MediaAttachment` row, profile pictures
+   too (`Kind` avatar or header, `ProfileOfAvatarId`). Deleting a post, an edit leaving media out, a replaced picture,
+   a dropped scheduled post, a removed root, and an upload never posted for a day each trash theirs
+   (`IMediaService.Trash`):
+   - the row gets `TrashedAt` in one conditional update, so a row attached meanwhile is left alone;
+   - its files move into `media-trash` at once, so `/media/files` stops serving them;
+   - `MediaJanitor` deletes them a day later (`TrashGrace`).
+
+   Nothing is deleted because it looks unused. `PrivaPub admin media audit [--fix]` compares disk and database: with
+   `--fix` (as www-data) it gives pictures shown from before their rows a row, and trashes media of deleted posts or
+   personas, rows whose files are missing, and files nothing holds.
+4. **A client never contacts a remote server for media:** every remote URL the API returns goes through
    `IMediaProxy.Wrap`, an HMAC-signed `/media/proxy/` URL fetched by `IFederationHttp.GetMedia`.
-4. **The proxy serves three ways:**
+5. **The proxy serves three ways:**
    - **Cached:** a file already cached is served from disk, ranges included.
    - **Downloaded:** a request without a `Range` is downloaded whole, up to `Media:MaxProxiedBytes`, then cached.
    - **Streamed:** a ranged request, or anything too big to cache, is streamed from the origin with the range passed on,
      and never cached. That is how remote video plays.
 
    nginx has a `/media/proxy/` location with `proxy_buffering off` and a 600 s read timeout for those streams.
-5. **A focal point is two finite numbers** within -1..1 (`FocalPoint.Parse`); anything else is ignored. A stored NaN made
+6. **A focal point is two finite numbers** within -1..1 (`FocalPoint.Parse`); anything else is ignored. A stored NaN made
    every status, timeline and Note holding its post fail to serialise; migration `_016` removed the ones stored before.
-6. **Remote video and audio become one playable attachment** in the Mastodon API (`MastodonMapper.Playable`): the best MP4
+7. **Remote video and audio become one playable attachment** in the Mastodon API (`MastodonMapper.Playable`): the best MP4
    up to 720p that carries both sound and picture, including PeerTube's fragmented files inside an HLS entry. HLS
    playlists themselves are not rewritten.
 

PrivaPub.Tests/Federation/JobHandlerTests.cs

@@ -489,9 +489,17 @@ namespace PrivaPub.Tests.Federation
 
 				await new MediaJanitor(media, options, NullLogger<MediaJanitor>.Instance).Sweep(token);
 
-				Assert.False(await DB.Default.Find<MediaAttachment>().Match(m => m.ID == stale.ID).ExecuteAnyAsync(token));
+				// never posted for a day: trashed, its files out of what /media/files serves at once
+				Assert.NotNull((await DB.Default.Find<MediaAttachment>().OneAsync(stale.ID, token)).TrashedAt);
 				Assert.False(File.Exists(Path.Combine(root, stale.FilePath)));
 				Assert.False(File.Exists(Path.Combine(root, stale.PreviewPath)));
+				Assert.True(File.Exists(Path.Combine(media.TrashRoot, stale.FilePath)));
+				// and once its grace has passed, deleted with its row
+				await DB.Default.Update<MediaAttachment>().MatchID(stale.ID).Modify(m => m.TrashedAt, DateTime.UtcNow - MediaJanitor.TrashGrace - TimeSpan.FromMinutes(1)).ExecuteAsync(token);
+				await new MediaJanitor(media, options, NullLogger<MediaJanitor>.Instance).Sweep(token);
+				Assert.False(await DB.Default.Find<MediaAttachment>().Match(m => m.ID == stale.ID).ExecuteAnyAsync(token));
+				Assert.False(File.Exists(Path.Combine(media.TrashRoot, stale.FilePath)));
+				Assert.False(File.Exists(Path.Combine(media.TrashRoot, stale.PreviewPath)));
 				Assert.True(await DB.Default.Find<MediaAttachment>().Match(m => m.ID == fresh.ID).ExecuteAnyAsync(token));
 				Assert.True(await DB.Default.Find<MediaAttachment>().Match(m => m.ID == attached.ID).ExecuteAnyAsync(token));
 				Assert.True(File.Exists(Path.Combine(root, fresh.FilePath)));

PrivaPub.Tests/Http/MediaLifecycleTests.cs

@@ -0,0 +1,153 @@
+using MongoDB.Entities;
+
+using PrivaPub.Domain.Media;
+using PrivaPub.Models.Media;
+using PrivaPub.Services;
+using PrivaPub.Tests.Support;
+using PrivaPub.Tests.Support.Host;
+
+using System.Net;
+
+namespace PrivaPub.Tests.Http
+{
+	// A file lives exactly as long as something holds it: a deleted post, an edit leaving media out, a replaced picture and
+	// a removed root each trash theirs, which stops /media/files serving them at once; the janitor deletes them later.
+	[Trait("Category", "Integration")]
+	public sealed class MediaLifecycleTests : IAsyncLifetime
+	{
+		PrivaPubHost _host;
+
+		public async ValueTask InitializeAsync()
+		{
+			Assert.SkipUnless(MongoFixture.Enabled, MongoFixture.Skip);
+			_host = await PrivaPubHost.Shared();
+		}
+
+		public ValueTask DisposeAsync() => ValueTask.CompletedTask;
+
+		static CancellationToken Token => TestContext.Current.CancellationToken;
+
+		async Task<string> Upload(Mastodon account, string name)
+		{
+			var form = MastodonHelpers.Multipart(("file", MastodonHelpers.JpegWithMetadata(64, 48), "image/jpeg", name));
+			return (await account.Client.Exchange(new HttpRequestMessage(HttpMethod.Post, "/api/v2/media") { Content = form })).Ok().Body.Text("id");
+		}
+
+		async Task<HttpStatusCode> Served(string url) =>
+			(await _host.Client().GetAsync(new Uri(url).PathAndQuery, Token)).StatusCode;
+
+		[Fact]
+		public async Task A_deleted_posts_media_stop_being_served()
+		{
+			var alice = await _host.Mastodon("alice");
+			var id = await Upload(alice, "a.jpg");
+			var status = (await alice.Client.Post("/api/v1/statuses", ("status", "with a picture"), ("media_ids[]", id))).Ok();
+			var url = status.Body["media_attachments"]![0]!["url"]!.GetValue<string>();
+			var preview = status.Body["media_attachments"]![0]!["preview_url"]!.GetValue<string>();
+			Assert.Equal(HttpStatusCode.OK, await Served(url));
+
+			(await alice.Client.Delete($"/api/v1/statuses/{status.Body.Text("id")}")).Ok();
+
+			Assert.Equal(HttpStatusCode.NotFound, await Served(url));
+			Assert.Equal(HttpStatusCode.NotFound, await Served(preview));
+			Assert.Equal("post deleted", (await DB.Default.Find<MediaAttachment>().OneAsync(id, Token)).TrashReason);
+			// the file waits in the trash, beside what /media/files serves
+			var relative = url[(url.IndexOf("/media/files/", StringComparison.Ordinal) + "/media/files/".Length)..];
+			Assert.True(File.Exists(Path.Combine(_host.Get<IMediaService>().TrashRoot, relative)));
+			Assert.Equal(HttpStatusCode.NotFound, await Served(url.Replace("/media/files/", "/media/files/.trash/")));
+		}
+
+		[Fact]
+		public async Task An_edit_that_leaves_media_out_trashes_them()
+		{
+			var alice = await _host.Mastodon("alice");
+			var kept = await Upload(alice, "kept.jpg");
+			var dropped = await Upload(alice, "dropped.jpg");
+			var status = (await alice.Client.Post("/api/v1/statuses", ("status", "two pictures"), ("media_ids[]", kept), ("media_ids[]", dropped))).Ok();
+			var urls = status.Body["media_attachments"]!.AsArray().ToDictionary(m => m!["id"]!.GetValue<string>(), m => m!["url"]!.GetValue<string>());
+
+			(await alice.Client.Put($"/api/v1/statuses/{status.Body.Text("id")}", ("status", "one picture"), ("media_ids[]", kept))).Ok();
+
+			Assert.Equal(HttpStatusCode.OK, await Served(urls[kept]));
+			Assert.Equal(HttpStatusCode.NotFound, await Served(urls[dropped]));
+			Assert.Null((await DB.Default.Find<MediaAttachment>().OneAsync(kept, Token)).TrashedAt);
+			Assert.Equal("edited out", (await DB.Default.Find<MediaAttachment>().OneAsync(dropped, Token)).TrashReason);
+		}
+
+		[Fact]
+		public async Task A_replaced_avatar_is_trashed_and_a_picture_that_is_not_one_is_refused()
+		{
+			var alice = await _host.Mastodon("alice");
+			async Task<ApiAnswer> Picture(byte[] bytes, string type, string name) =>
+				await alice.Client.Exchange(new HttpRequestMessage(HttpMethod.Patch, "/api/v1/accounts/update_credentials")
+				{
+					Content = MastodonHelpers.Multipart(("avatar", bytes, type, name))
+				});
+			var first = (await Picture(MastodonHelpers.JpegWithMetadata(300, 300), "image/jpeg", "one.jpg")).Ok().Body.Text("avatar");
+			var second = (await Picture(MastodonHelpers.JpegWithMetadata(320, 300), "image/jpeg", "two.jpg")).Ok().Body.Text("avatar");
+
+			Assert.NotEqual(first, second);
+			Assert.Equal(HttpStatusCode.NotFound, await Served(first));
+			Assert.Equal(HttpStatusCode.OK, await Served(second));
+			var row = await DB.Default.Find<MediaAttachment>().Match(m => m.ProfileOfAvatarId == alice.Persona.Id && m.Kind == "avatar" && m.TrashedAt == null).ExecuteSingleAsync(Token);
+			Assert.EndsWith(row.FilePath, second);
+
+			// not an image: 422, and the avatar stays as it was
+			var refused = await Picture("<svg xmlns='http://www.w3.org/2000/svg'/>"u8.ToArray(), "image/svg+xml", "x.svg");
+			Assert.Equal(HttpStatusCode.UnprocessableEntity, refused.Status);
+			Assert.Equal(second, (await alice.Client.Get("/api/v1/accounts/verify_credentials")).Ok().Body.Text("avatar"));
+		}
+
+		[Fact]
+		public async Task A_removed_roots_media_and_pictures_are_trashed_and_its_scheduled_posts_dropped()
+		{
+			var gone = await _host.Mastodon($"gone{Guid.NewGuid():N}"[..12]);
+			var posted = await Upload(gone, "posted.jpg");
+			var url = (await gone.Client.Post("/api/v1/statuses", ("status", "soon gone"), ("media_ids[]", posted))).Ok()
+				.Body["media_attachments"]![0]!["url"]!.GetValue<string>();
+			var held = await Upload(gone, "held.jpg");
+			(await gone.Client.Post("/api/v1/statuses", ("status", "later"), ("media_ids[]", held),
+				("scheduled_at", DateTime.UtcNow.AddHours(2).ToString("O")))).Ok();
+			var avatar = (await gone.Client.Exchange(new HttpRequestMessage(HttpMethod.Patch, "/api/v1/accounts/update_credentials")
+			{
+				Content = MastodonHelpers.Multipart(("avatar", MastodonHelpers.JpegWithMetadata(200, 200), "image/jpeg", "me.jpg"))
+			})).Ok().Body.Text("avatar");
+
+			Assert.True(await _host.Get<IRootRemoval>().Remove(gone.Persona.Root.Id, Token));
+
+			Assert.Equal(HttpStatusCode.NotFound, await Served(url));
+			Assert.Equal(HttpStatusCode.NotFound, await Served(avatar));
+			Assert.False(await DB.Default.Find<MediaAttachment>().Match(m => m.OwnerAvatarId == gone.Persona.Id && m.TrashedAt == null).ExecuteAnyAsync(Token));
+			Assert.False(await DB.Default.Find<Models.Social.ScheduledStatus>().Match(s => s.AvatarId == gone.Persona.Id).ExecuteAnyAsync(Token));
+		}
+
+		[Fact]
+		public async Task The_audit_adopts_todays_pictures_and_trashes_what_nothing_holds()
+		{
+			var media = _host.Get<IMediaService>();
+			var bob = await _host.Mastodon($"bob{Guid.NewGuid():N}"[..12]);
+			// a picture shown from before pictures had rows, and a leftover nothing holds
+			var picture = $"2020/01/{Guid.NewGuid():N}.jpg";
+			var leftover = $"2020/01/{Guid.NewGuid():N}.jpg";
+			foreach (var relative in new[] { picture, leftover })
+			{
+				Directory.CreateDirectory(Path.GetDirectoryName(Path.Combine(media.Root, relative))!);
+				await File.WriteAllBytesAsync(Path.Combine(media.Root, relative), new byte[] { 1, 2, 3 }, Token);
+			}
+			await DB.Default.Update<Models.User.Avatar>().MatchID(bob.Persona.Id).Modify(a => a.PictureURL, media.Url(picture)).ExecuteAsync(Token);
+
+			var looked = await MediaAudit.Run(media, fix: false, Token);
+			Assert.True(looked.Adopted >= 1);
+			Assert.True(looked.Unheld >= 1);
+			Assert.True(File.Exists(Path.Combine(media.Root, leftover)));
+
+			await MediaAudit.Run(media, fix: true, Token);
+
+			Assert.True(await DB.Default.Find<MediaAttachment>().Match(m => m.ProfileOfAvatarId == bob.Persona.Id && m.FilePath == picture && m.TrashedAt == null).ExecuteAnyAsync(Token));
+			Assert.Equal(HttpStatusCode.OK, await Served(media.Url(picture)));
+			Assert.Equal(HttpStatusCode.NotFound, await Served(media.Url(leftover)));
+			Assert.True(File.Exists(Path.Combine(media.TrashRoot, leftover)));
+			Assert.True(await DB.Default.Find<MediaAttachment>().Match(m => m.FilePath == leftover && m.TrashedAt != null).ExecuteAnyAsync(Token));
+		}
+	}
+}

PrivaPub/Api/Mastodon/Controllers/AccountsController.cs

@@ -51,13 +51,28 @@ namespace PrivaPub.Api.Mastodon.Controllers
 		public async Task<IActionResult> UpdateCredentials([FromServices] IMediaService media, CancellationToken token)
 		{
 			var avatar = await _dbEntities.Avatars.MatchID(Me.Id).ExecuteFirstAsync(token);
+			// a new picture is a row of its own; the one it replaces goes to the trash once the profile is saved
+			var replaced = new List<Models.Media.MediaAttachment>();
 			if (Request.HasFormContentType)
 			{
 				var form = await Request.ReadFormAsync(token);
-				if (form.Files["avatar"] is { } picture && await media.ProfileImage(picture, 400, 400, token) is { } pictureUrl)
+				foreach (var (field, kind, width, height) in new[] { ("avatar", "avatar", 400, 400), ("header", "header", 1500, 500) })
-					avatar.PictureURL = pictureUrl;
+				{
-				if (form.Files["header"] is { } header && await media.ProfileImage(header, 1500, 500, token) is { } headerUrl)
+					if (form.Files[field] is not { } file)
-					avatar.ThumbnailURL = headerUrl;
+						continue;
+					var outcome = await media.ProfileImage(avatar.ID, kind, file, width, height, token);
+					if (!outcome.Ok)
+					{
+						foreach (var made in replaced)
+							await media.Trash(m => m.ID == made.ID, "profile not saved", token);
+						return Error(outcome.Status, outcome.Error);
+					}
+					replaced.Add(outcome.Attachment);
+					if (kind == "avatar")
+						avatar.PictureURL = media.Url(outcome.Attachment.FilePath);
+					else
+						avatar.ThumbnailURL = media.Url(outcome.Attachment.FilePath);
+				}
 			}
 			if (Params.Has("display_name"))
 				avatar.Name = Params.Get("display_name")?.Trim();
@@ -98,6 +113,8 @@ namespace PrivaPub.Api.Mastodon.Controllers
 				avatar.Fields = fields;
 			avatar.UpdatedAt = DateTime.UtcNow;
 			await DB.Default.SaveAsync(avatar, token);
+			foreach (var made in replaced)
+				await media.Trash(m => m.ProfileOfAvatarId == avatar.ID && m.Kind == made.Kind && m.ID != made.ID, "replaced", token);
 
 			var actor = _localActors.FromAvatar(avatar);
 			await _outbox.PublishProfile(actor, token);

PrivaPub/Api/Mastodon/Controllers/MediaController.cs

@@ -42,14 +42,14 @@ namespace PrivaPub.Api.Mastodon.Controllers
 		[HttpGet("/api/v1/media/{id}"), Scope("write:media")]
 		public async Task<IActionResult> Get(string id, CancellationToken token)
 		{
-			var attachment = await DB.Default.Find<MediaAttachment>().Match(m => m.ID == id && m.OwnerAvatarId == MyId).ExecuteFirstAsync(token);
+			var attachment = await DB.Default.Find<MediaAttachment>().Match(m => m.ID == id && m.OwnerAvatarId == MyId && m.TrashedAt == null && m.ProfileOfAvatarId == null).ExecuteFirstAsync(token);
 			return attachment == default ? NotFoundError() : Json(View(attachment));
 		}
 
 		[HttpPut("/api/v1/media/{id}"), Scope("write:media")]
 		public async Task<IActionResult> Update(string id, CancellationToken token)
 		{
-			var attachment = await DB.Default.Find<MediaAttachment>().Match(m => m.ID == id && m.OwnerAvatarId == MyId).ExecuteFirstAsync(token);
+			var attachment = await DB.Default.Find<MediaAttachment>().Match(m => m.ID == id && m.OwnerAvatarId == MyId && m.TrashedAt == null && m.ProfileOfAvatarId == null).ExecuteFirstAsync(token);
 			if (attachment == default)
 				return NotFoundError();
 			if (Params.Has("description"))

PrivaPub/Api/Mastodon/Controllers/ScheduledStatusesController.cs

@@ -96,7 +96,8 @@ namespace PrivaPub.Api.Mastodon.Controllers
 			if (await Mine(id, token) is not { } scheduled)
 				return NotFoundError();
 			await DB.Default.DeleteAsync<ScheduledStatus>(scheduled.ID);
-			await ScheduledStatuses.Release(scheduled.ID, token);
+			// its media were the scheduled post's alone, never posted
+			await _media.Trash(m => m.ScheduledStatusId == scheduled.ID, "scheduled post dropped", token);
 			return Json(new { });
 		}
 	}

PrivaPub/Api/Mastodon/Controllers/StatusesController.cs

@@ -111,7 +111,8 @@ namespace PrivaPub.Api.Mastodon.Controllers
 				return Error(StatusCodes.Status422UnprocessableEntity, "Validation failed: Text can't be blank");
 			var mediaIds = asked.MediaIds.Distinct().ToList();
 			if (mediaIds.Count > 4 || mediaIds.Count > 0 && await DB.Default.CountAsync<Models.Media.MediaAttachment>(
-					m => mediaIds.Contains(m.ID) && m.OwnerAvatarId == MyId && m.PostId == null && m.ScheduledStatusId == null, token) != mediaIds.Count)
+					m => mediaIds.Contains(m.ID) && m.OwnerAvatarId == MyId && m.PostId == null && m.ScheduledStatusId == null
+					&& m.TrashedAt == null && m.ProfileOfAvatarId == null, token) != mediaIds.Count)
 				return Error(StatusCodes.Status422UnprocessableEntity, "Validation failed: Media attachments are not yours, already posted or too many");
 
 			var scheduled = new ScheduledStatus { AvatarId = MyId, ScheduledAt = at, Params = asked };

PrivaPub/Domain/Media/MediaAudit.cs

@@ -0,0 +1,101 @@
+using MongoDB.Entities;
+
+using PrivaPub.Models.Media;
+using PrivaPub.Models.User;
+
+using PostEntity = PrivaPub.Models.Post.Post;
+
+namespace PrivaPub.Domain.Media
+{
+	// What the media directory holds against what the database says holds it, and, with fix, the two made to agree:
+	// - a persona's current avatar or header with no row (they had none before) becomes a row of its own;
+	// - media of a deleted post or of a deleted persona go to the trash;
+	// - rows whose files are gone go to the trash;
+	// - files nothing holds (deleted posts' media and replaced pictures, from before media were trashed) get a trashed row
+	//   each.
+	// Trashed files leave what /media/files serves at once, and the janitor deletes them after its grace. Run it without
+	// fix first; with fix it must run as the user that owns the media (www-data).
+	public static class MediaAudit
+	{
+		public sealed record Report(int Files, int Held, int Adopted, int OfDeleted, int MissingFiles, int Unheld, IReadOnlyList<string> Examples, bool Fixed);
+
+		public static async Task<Report> Run(IMediaService media, bool fix, CancellationToken token)
+		{
+			var files = Served(media.Root);
+			var rows = await DB.Default.Find<MediaAttachment>().Match(m => m.TrashedAt == null).ExecuteAsync(token);
+			var held = rows.SelectMany(r => new[] { r.FilePath, r.PreviewPath }).Where(p => !string.IsNullOrEmpty(p)).ToHashSet();
+			var examples = new List<string>();
+
+			// pictures personas show today, kept as rows from now on
+			var prefix = media.Url(string.Empty);
+			var adopted = 0;
+			foreach (var avatar in await DB.Default.Find<Avatar>().Match(a => !a.DeletionAt.HasValue && (a.PictureURL != null || a.ThumbnailURL != null)).ExecuteAsync(token))
+				foreach (var (url, kind) in new[] { (avatar.PictureURL, "avatar"), (avatar.ThumbnailURL, "header") })
+				{
+					if (url?.StartsWith(prefix, StringComparison.Ordinal) != true)
+						continue;
+					var relative = url[prefix.Length..];
+					if (held.Contains(relative) || !files.Contains(relative))
+						continue;
+					adopted++;
+					held.Add(relative);
+					if (fix)
+						await DB.Default.SaveAsync(new MediaAttachment
+						{
+							OwnerAvatarId = avatar.ID,
+							ProfileOfAvatarId = avatar.ID,
+							Kind = kind,
+							ContentType = "image/jpeg",
+							FilePath = relative,
+							Size = new FileInfo(Path.Combine(media.Root, relative)).Length,
+							AttachedAt = DateTime.UtcNow
+						}, token);
+				}
+
+			// held by something that is gone: a deleted post, a deleted persona
+			var postIds = rows.Where(r => r.PostId != null).Select(r => r.PostId).Distinct().ToList();
+			var livePosts = (await DB.Default.Find<PostEntity>().Match(p => postIds.Contains(p.ID) && !p.DeletedAt.HasValue).Project(p => p.Include(x => x.ID)).ExecuteAsync(token))
+				.Select(p => p.ID).ToHashSet();
+			var ownerIds = rows.Select(r => r.OwnerAvatarId).Where(id => id != null).Distinct().ToList();
+			var liveOwners = (await DB.Default.Find<Avatar>().Match(a => ownerIds.Contains(a.ID) && !a.DeletionAt.HasValue).Project(a => a.Include(x => x.ID)).ExecuteAsync(token))
+				.Select(a => a.ID).ToHashSet();
+			var ofDeleted = rows.Where(r => (r.PostId != null && !livePosts.Contains(r.PostId)) || (r.OwnerAvatarId != null && !liveOwners.Contains(r.OwnerAvatarId))).ToList();
+			var missing = rows.Except(ofDeleted).Where(r => !string.IsNullOrEmpty(r.FilePath) && !files.Contains(r.FilePath)).ToList();
+			if (fix)
+				foreach (var row in ofDeleted.Concat(missing))
+					await media.Trash(m => m.ID == row.ID, ofDeleted.Contains(row) ? "audit: its holder is deleted" : "audit: its file is missing", token);
+			foreach (var row in ofDeleted)
+				held.Remove(row.FilePath);
+
+			// files nothing holds
+			var unheld = files.Where(f => !held.Contains(f)).OrderBy(f => f, StringComparer.Ordinal).ToList();
+			examples.AddRange(unheld.Take(10));
+			if (fix)
+				foreach (var relative in unheld)
+				{
+					var orphan = new MediaAttachment
+					{
+						Kind = "orphan",
+						FilePath = relative,
+						Size = new FileInfo(Path.Combine(media.Root, relative)).Length,
+						TrashedAt = DateTime.UtcNow,
+						TrashReason = "audit: nothing holds it"
+					};
+					await DB.Default.SaveAsync(orphan, token);
+					media.Hide(orphan);
+				}
+
+			return new Report(files.Count, rows.Count, adopted, ofDeleted.Count, missing.Count, unheld.Count, examples, fix);
+		}
+
+		// every file /media/files serves: everything under the root
+		static HashSet<string> Served(string root)
+		{
+			if (!Directory.Exists(root))
+				return new HashSet<string>();
+			return Directory.EnumerateFiles(root, "*", SearchOption.AllDirectories)
+				.Select(f => Path.GetRelativePath(root, f).Replace('\\', '/'))
+				.ToHashSet();
+		}
+	}
+}

PrivaPub/Domain/Media/MediaProxy.cs

@@ -123,8 +123,11 @@ namespace PrivaPub.Domain.Media
 
 	public class MediaJanitor : BackgroundService
 	{
+		static readonly TimeSpan FirstPass = TimeSpan.FromMinutes(5);
 		static readonly TimeSpan Interval = TimeSpan.FromHours(1);
 		static readonly TimeSpan UnattachedLifetime = TimeSpan.FromDays(1);
+		// a trashed file waits this long before it is deleted, already out of what /media/files serves
+		public static readonly TimeSpan TrashGrace = TimeSpan.FromDays(1);
 
 		readonly IMediaService _media;
 		readonly IOptionsMonitor<MediaOptions> _options;
@@ -139,11 +142,13 @@ namespace PrivaPub.Domain.Media
 
 		protected override async Task ExecuteAsync(CancellationToken stoppingToken)
 		{
+			var wait = FirstPass;
 			while (!stoppingToken.IsCancellationRequested)
 			{
 				try
 				{
-					await Task.Delay(Interval, stoppingToken);
+					await Task.Delay(wait, stoppingToken);
+					wait = Interval;
 					await Sweep(stoppingToken);
 				}
 				catch (OperationCanceledException) when (stoppingToken.IsCancellationRequested)
@@ -157,12 +162,33 @@ namespace PrivaPub.Domain.Media
 			}
 		}
 
-		//one pass: uploads left unattached for a day go (unless a scheduled post waits for them), then the proxy cache is trimmed to its size, oldest first
+		// one pass:
+		// - uploads never posted for a day (and not waiting for a scheduled post, nor a profile picture) go to the trash;
+		// - trashed files still served (a crash between the mark and the move) are moved out;
+		// - trashed files past their grace are deleted, with their rows;
+		// - the proxy cache is trimmed to its size, oldest first
 		public async Task Sweep(CancellationToken token)
 		{
 			var cutoff = DateTime.UtcNow - UnattachedLifetime;
-			foreach (var stale in await DB.Default.Find<MediaAttachment>().Match(m => m.PostId == null && m.ScheduledStatusId == null && m.CreatedAt < cutoff).Limit(500).ExecuteAsync(token))
+			var unattached = await _media.Trash(m => m.PostId == null && m.ScheduledStatusId == null && m.ProfileOfAvatarId == null && m.CreatedAt < cutoff,
-				await _media.Delete(stale);
+				"never posted", token);
+
+			var trashed = await DB.Default.Find<MediaAttachment>().Match(m => m.TrashedAt != null).Limit(2000).ExecuteAsync(token);
+			var purgeBefore = DateTime.UtcNow - TrashGrace;
+			var purged = 0;
+			foreach (var row in trashed)
+			{
+				if (row.TrashedAt < purgeBefore)
+				{
+					await _media.Purge(row, token);
+					purged++;
+				}
+				else
+					_media.Hide(row);
+			}
+			if (unattached > 0 || purged > 0)
+				_logger.LogInformation("{Service}: {Unattached} uploads never posted trashed, {Purged} trashed files deleted",
+					nameof(MediaJanitor), unattached, purged);
 			TrimProxyCache();
 		}
 

PrivaPub/Domain/Media/MediaService.cs

@@ -27,9 +27,20 @@ namespace PrivaPub.Domain.Media
 		string Root { get; }
 		string ProxyRoot { get; }
 		string Url(string relativePath);
+		/// <summary>Where trashed files wait for the janitor: the sibling of Root, on the same disk (a move is a rename) and
+		/// outside what /media/files serves. (Not a dot-prefixed folder inside Root: the file provider only hides a file whose
+		/// own name starts with a dot.)</summary>
+		string TrashRoot { get; }
 		Task<MediaOutcome> Upload(LocalActor owner, IFormFile file, string description, string focus, CancellationToken token);
-		Task<string> ProfileImage(IFormFile file, int width, int height, CancellationToken token);
+		/// <summary>A persona's new avatar or header ("avatar" or "header"), cropped to its size, as a row of its own.</summary>
-		Task Delete(MediaAttachment attachment);
+		Task<MediaOutcome> ProfileImage(string avatarId, string kind, IFormFile file, int width, int height, CancellationToken token);
+		/// <summary>Trashes the media <paramref name="which"/> matches (and that isn't trashed yet): each row is marked in one
+		/// conditional update, so a row attached meanwhile is left alone, and its files stop being served at once.</summary>
+		Task<long> Trash(System.Linq.Expressions.Expression<Func<MediaAttachment, bool>> which, string reason, CancellationToken token);
+		/// <summary>Moves a trashed row's files out of what is served, if they are still there (a crash between the mark and the move).</summary>
+		void Hide(MediaAttachment trashed);
+		/// <summary>Deletes a trashed row's files and the row.</summary>
+		Task Purge(MediaAttachment trashed, CancellationToken token);
 	}
 
 	public class MediaService : IMediaService
@@ -62,6 +73,8 @@ namespace PrivaPub.Domain.Media
 
 		public string ProxyRoot => Root.TrimEnd(Path.DirectorySeparatorChar) + "-proxy";
 
+		public string TrashRoot => Root.TrimEnd(Path.DirectorySeparatorChar) + "-trash";
+
 		public string Url(string relativePath) => relativePath == default ? default : $"{_localActors.BaseAddress}/media/files/{relativePath.Replace('\\', '/')}";
 
 		public async Task<MediaOutcome> Upload(LocalActor owner, IFormFile file, string description, string focus, CancellationToken token)
@@ -118,34 +131,86 @@ namespace PrivaPub.Domain.Media
 			return new MediaOutcome(attachment);
 		}
 
-		public async Task<string> ProfileImage(IFormFile file, int width, int height, CancellationToken token)
+		public async Task<MediaOutcome> ProfileImage(string avatarId, string kind, IFormFile file, int width, int height, CancellationToken token)
 		{
-			if (file == default || file.Length == 0 || file.Length > _options.CurrentValue.MaxImageBytes)
+			var contentType = file?.ContentType?.Split(';')[0].Trim().ToLowerInvariant();
-				return default;
+			if (file == default || file.Length == 0)
+				return MediaOutcome.Fail(StatusCodes.Status422UnprocessableEntity, "Validation failed: File can't be blank");
+			if (!ImageTypes.Contains(contentType))
+				return MediaOutcome.Fail(StatusCodes.Status422UnprocessableEntity, "Validation failed: File type is not supported");
+			if (file.Length > _options.CurrentValue.MaxImageBytes)
+				return MediaOutcome.Fail(StatusCodes.Status422UnprocessableEntity, "Validation failed: File is too big");
 			await using var stream = file.OpenReadStream();
 			using var buffer = new MemoryStream();
 			await stream.CopyToAsync(buffer, token);
+			byte[] bytes;
+			int outWidth, outHeight;
 			try
 			{
 				using var image = Image.ThumbnailBuffer(buffer.ToArray(), width, height: height, crop: Enums.Interesting.Centre, size: Enums.Size.Down);
 				using var flat = Flatten(image);
-				return Url(await Save(flat.WriteToBuffer(".jpg[Q=85,keep=none]"), "jpg", token));
+				bytes = flat.WriteToBuffer(".jpg[Q=85,keep=none]");
+				(outWidth, outHeight) = (flat.Width, flat.Height);
 			}
-			catch (VipsException)
+			catch (VipsException ex)
 			{
-				return default;
+				_logger.LogInformation("Refused a profile picture: {Error}", ex.Message);
+				return MediaOutcome.Fail(StatusCodes.Status422UnprocessableEntity, "Validation failed: The file is not a readable image");
+			}
+			var attachment = new MediaAttachment
+			{
+				OwnerAvatarId = avatarId,
+				ProfileOfAvatarId = avatarId,
+				Kind = kind,
+				ContentType = "image/jpeg",
+				FilePath = await Save(bytes, "jpg", token),
+				Size = bytes.Length,
+				Width = outWidth,
+				Height = outHeight,
+				AttachedAt = DateTime.UtcNow
+			};
+			await DB.Default.SaveAsync(attachment, token);
+			return new MediaOutcome(attachment);
+		}
+
+		public async Task<long> Trash(System.Linq.Expressions.Expression<Func<MediaAttachment, bool>> which, string reason, CancellationToken token)
+		{
+			var trashed = 0L;
+			foreach (var candidate in await DB.Default.Find<MediaAttachment>().Match(which).Match(m => m.TrashedAt == null).ExecuteAsync(token))
+			{
+				var marked = await DB.Default.Update<MediaAttachment>()
+					.Match(which).Match(m => m.ID == candidate.ID && m.TrashedAt == null)
+					.Modify(m => m.TrashedAt, DateTime.UtcNow)
+					.Modify(m => m.TrashReason, reason)
+					.ExecuteAsync(token);
+				if (marked.ModifiedCount == 0)
+					continue;
+				Hide(candidate);
+				trashed++;
+			}
+			return trashed;
+		}
+
+		public void Hide(MediaAttachment trashed)
+		{
+			foreach (var relative in new[] { trashed.FilePath, trashed.PreviewPath }.Where(p => !string.IsNullOrEmpty(p)))
+			{
+				var served = Path.Combine(Root, relative);
+				if (!File.Exists(served))
+					continue;
+				var hidden = Path.Combine(TrashRoot, relative);
+				Directory.CreateDirectory(Path.GetDirectoryName(hidden)!);
+				File.Move(served, hidden, overwrite: true);
 			}
 		}
 
-		public Task Delete(MediaAttachment attachment)
+		public async Task Purge(MediaAttachment trashed, CancellationToken token)
 		{
-			foreach (var path in new[] { attachment.FilePath, attachment.PreviewPath }.Where(p => p != default))
+			foreach (var relative in new[] { trashed.FilePath, trashed.PreviewPath }.Where(p => !string.IsNullOrEmpty(p)))
-			{
+				foreach (var full in new[] { Path.Combine(TrashRoot, relative), Path.Combine(Root, relative) })
-				var full = Path.Combine(Root, path);
+					if (File.Exists(full))
-				if (File.Exists(full))
+						File.Delete(full);
-					File.Delete(full);
+			await DB.Default.DeleteAsync<MediaAttachment>(m => m.ID == trashed.ID && m.TrashedAt != null);
-			}
-			return DB.Default.DeleteAsync<MediaAttachment>(attachment.ID);
 		}
 
 		public static ProcessedImage ProcessImage(byte[] input, int maxSide, int previewSide, bool animated)

PrivaPub/Domain/Statuses/StatusService.cs

@@ -320,6 +320,9 @@ namespace PrivaPub.Domain.Statuses
 			{
 				post.Media = media.Select(ToPostMedia).ToList();
 				await Attach(media, post.ID, token);
+				// what the edit left out is no longer held by anything
+				var kept = media.Select(m => m.ID).ToList();
+				await _media.Trash(m => m.PostId == post.ID && !kept.Contains(m.ID), "edited out", token);
 			}
 			post.Title = draft.Title == default ? post.Title : Clean(draft.Title);
 			post.SpoilerText = Clean(draft.SpoilerText);
@@ -380,6 +383,7 @@ namespace PrivaPub.Domain.Statuses
 				.Modify(p => p.Revisions, new List<PostRevision>())
 				.ExecuteAsync(token);
 			await Fanout.Deleting(post, token);
+			await _media.Trash(m => m.PostId == post.ID, "post deleted", token);
 			await DB.Default.DeleteAsync<TimelineEntry>(e => e.PostId == post.ID || e.ReblogOfPostId == post.ID);
 			// boosts of it end with it, as on Mastodon, so no count keeps them
 			await DB.Default.Update<PostEntity>().Match(p => p.ReblogOfPostId == post.ID && !p.DeletedAt.HasValue)
@@ -692,7 +696,8 @@ namespace PrivaPub.Domain.Statuses
 				return default;
 			var wanted = ids.Distinct().ToList();
 			var found = await DB.Default.Find<MediaAttachment>()
-				.Match(m => wanted.Contains(m.ID) && m.OwnerAvatarId == author.Id && (m.PostId == null || m.PostId == postId))
+				.Match(m => wanted.Contains(m.ID) && m.OwnerAvatarId == author.Id && (m.PostId == null || m.PostId == postId)
+					&& m.TrashedAt == null && m.ProfileOfAvatarId == null)
 				.ExecuteAsync(token);
 			return found.Count == wanted.Count ? wanted.Select(id => found.First(m => m.ID == id)).ToList() : default;
 		}

PrivaPub/Infrastructure/Cli/AdminCommands.cs

@@ -22,6 +22,8 @@ namespace PrivaPub.Infrastructure.Cli
 			usage: PrivaPub admin promote|demote <root>
 			       PrivaPub admin create-root <login> [--admin]   the password is read from standard input
 			       PrivaPub admin smoke <persona>                 prints "<login> <password>" for the deploy's signed-in check
+			       PrivaPub admin media audit [--fix]             media files against what holds them; --fix (as www-data) trashes
+			                                                      what nothing holds and gives today's pictures their rows
 			""";
 
 		public static async Task<int> Run(string[] args, IServiceProvider services, TextReader input = default, TextWriter output = default)
@@ -36,12 +38,29 @@ namespace PrivaPub.Infrastructure.Cli
 					return await CreateRoot(services, login, input.ReadLine(), flags.Contains("--admin"), output);
 				case ["smoke", var persona]:
 					return await Smoke(services, persona, output);
+				case ["media", "audit", .. var flags] when flags.All(f => f == "--fix"):
+					return await AuditMedia(services, flags.Contains("--fix"), output);
 				default:
 					Console.Error.WriteLine(Usage);
 					return 2;
 			}
 		}
 
+		static async Task<int> AuditMedia(IServiceProvider services, bool fix, TextWriter output)
+		{
+			var report = await Domain.Media.MediaAudit.Run(services.GetRequiredService<Domain.Media.IMediaService>(), fix, CancellationToken.None);
+			output.WriteLine($"{report.Files} files served, {report.Held} rows holding media");
+			output.WriteLine($"{report.Adopted} pictures personas show without a row{(fix ? ": given one" : string.Empty)}");
+			output.WriteLine($"{report.OfDeleted} media of deleted posts or personas{(fix ? ": trashed" : string.Empty)}");
+			output.WriteLine($"{report.MissingFiles} rows whose files are missing{(fix ? ": trashed" : string.Empty)}");
+			output.WriteLine($"{report.Unheld} files nothing holds{(fix ? ": trashed" : string.Empty)}");
+			foreach (var example in report.Examples)
+				output.WriteLine($"  {example}");
+			if (!fix && report.Adopted + report.OfDeleted + report.MissingFiles + report.Unheld > 0)
+				output.WriteLine("run again with --fix, as www-data, to apply this; trashed files are deleted after a day");
+			return 0;
+		}
+
 		static async Task<int> Promote(bool promote, string userName, TextWriter output)
 		{
 			userName = userName.ToLowerInvariant();

PrivaPub/Infrastructure/Data/Indexes.cs

@@ -32,6 +32,13 @@ namespace PrivaPub.Infrastructure.Data
 			await Plain<Post>(token, p => p.AnsweringToPostId);
 			await DB.Default.Index<Post>().Key(p => p.Geo, KeyType.Geo2DSphere).CreateAsync(token);
 
+			// media: a post's, a persona's, what a scheduled post holds, the janitor's never-posted uploads and its trash
+			await Plain<Models.Media.MediaAttachment>(token, m => m.PostId, m => m.ScheduledStatusId, m => m.CreatedAt);
+			await Plain<Models.Media.MediaAttachment>(token, m => m.OwnerAvatarId);
+			await Plain<Models.Media.MediaAttachment>(token, m => m.ScheduledStatusId);
+			await Plain<Models.Media.MediaAttachment>(token, m => m.ProfileOfAvatarId);
+			await Plain<Models.Media.MediaAttachment>(token, m => m.TrashedAt);
+
 			await Unique<DmPost>(p => p.ObjectURI, Builders<DmPost>.Filter.Gt(p => p.ObjectURI, ""), token);
 			await Plain<DmPost>(token, p => p.GroupId, p => p.ID);
 

PrivaPub/Models/Media/MediaAttachment.cs

@@ -19,6 +19,11 @@ namespace PrivaPub.Models.Media
 		public DateTime CreatedAt { get; set; } = DateTime.UtcNow;
 		public DateTime? AttachedAt { get; set; }
 		public string ScheduledStatusId { get; set; }//kept for a scheduled post until it is published or dropped
+		public string ProfileOfAvatarId { get; set; }//a persona's avatar or header (Kind "avatar" or "header"), never a post's
+		// set once nothing holds it any more (its post deleted, edited out, replaced, its root removed, never posted): its
+		// files move out of what /media/files serves at once, and the janitor deletes them after a grace
+		public DateTime? TrashedAt { get; set; }
+		public string TrashReason { get; set; }
 	}
 
 	public class MediaSecret : Entity

PrivaPub/Services/RootRemoval.cs

@@ -7,6 +7,7 @@ using PrivaPub.Models.Federation;
 using PrivaPub.Models.Social;
 using PrivaPub.Models.User;
 using PrivaPub.StaticServices;
+using PrivaPub.Domain.Media;
 
 using System.Text.Json.Nodes;
 
@@ -30,9 +31,11 @@ namespace PrivaPub.Services
 		readonly ILocalActorService _localActors;
 		readonly IDeliveryService _delivery;
 		readonly IRootSessions _sessions;
+		readonly IMediaService _media;
 
-		public RootRemoval(DbEntities dbEntities, ILocalActorService localActors, IDeliveryService delivery, IRootSessions sessions)
+		public RootRemoval(DbEntities dbEntities, ILocalActorService localActors, IDeliveryService delivery, IRootSessions sessions, IMediaService media)
 		{
+			_media = media;
 			_dbEntities = dbEntities;
 			_localActors = localActors;
 			_delivery = delivery;
@@ -66,6 +69,10 @@ namespace PrivaPub.Services
 				await DB.Default.DeleteAsync<Models.Social.PersonaList>(l => l.AvatarId == avatar.ID);
 				await DB.Default.DeleteAsync<Models.Social.PersonaFilter>(f => f.AvatarId == avatar.ID);
 				await DB.Default.DeleteAsync<Models.Social.FollowedTag>(t => t.AvatarId == avatar.ID);
+				// nothing of it publishes later, and none of its files stays served: its posts' media, its pictures and
+				// what its scheduled posts held
+				await DB.Default.DeleteAsync<Models.Social.ScheduledStatus>(s => s.AvatarId == avatar.ID);
+				await _media.Trash(m => m.OwnerAvatarId == avatar.ID, "root removed", token);
 			}
 
 			await DB.Default.Update<RootUser>().MatchID(root.ID)