journal: fix space reservation limit enforcement

5 files changed, 118 insertions, 29 deletions

diff --git a/src/journal/journal-def.h b/src/journal/journal-def.h
index 1a63ca1f8d..5f026ee0f6 100644
--- a/src/journal/journal-def.h
+++ b/src/journal/journal-def.h
@@ -135,9 +135,6 @@ _packed_ struct Header {
         sd_id128_t seqnum_id;
         uint64_t arena_offset;
         uint64_t arena_size;
-        uint64_t arena_max_size;  /* obsolete */
-        uint64_t arena_min_size;  /* obsolete */
-        uint64_t arena_keep_free; /* obsolete */
         uint64_t data_hash_table_offset;     /* for looking up data objects */
         uint64_t data_hash_table_size;
         uint64_t field_hash_table_offset;     /* for looking up field objects */
diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
index 8a864cb913..8f9b61bc2f 100644
--- a/src/journal/journal-file.c
+++ b/src/journal/journal-file.c
@@ -31,12 +31,6 @@
 #include "journal-file.h"
 #include "lookup3.h"
 
-#define DEFAULT_ARENA_MAX_SIZE (16ULL*1024ULL*1024ULL*1024ULL)
-#define DEFAULT_ARENA_MIN_SIZE (256ULL*1024ULL)
-#define DEFAULT_ARENA_KEEP_FREE (1ULL*1024ULL*1024ULL)
-
-#define DEFAULT_MAX_USE (16ULL*1024ULL*1024ULL*16ULL)
-
 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
 #define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
 
@@ -76,9 +70,6 @@ static int journal_file_init_header(JournalFile *f, JournalFile *template) {
         zero(h);
         memcpy(h.signature, signature, 8);
         h.arena_offset = htole64(ALIGN64(sizeof(h)));
-        h.arena_max_size = htole64(DEFAULT_ARENA_MAX_SIZE);
-        h.arena_min_size = htole64(DEFAULT_ARENA_MIN_SIZE);
-        h.arena_keep_free = htole64(DEFAULT_ARENA_KEEP_FREE);
 
         r = sd_id128_randomize(&h.file_id);
         if (r < 0)
@@ -161,16 +152,10 @@ static int journal_file_verify_header(JournalFile *f) {
 }
 
 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
-        uint64_t asize;
         uint64_t old_size, new_size;
 
         assert(f);
 
-        if (offset < le64toh(f->header->arena_offset))
-                return -EINVAL;
-
-        new_size = PAGE_ALIGN(offset + size);
-
         /* We assume that this file is not sparse, and we know that
          * for sure, since we always call posix_fallocate()
          * ourselves */
@@ -179,12 +164,19 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
                 le64toh(f->header->arena_offset) +
                 le64toh(f->header->arena_size);
 
-        if (old_size >= new_size)
+        new_size = PAGE_ALIGN(offset + size);
+        if (new_size < le64toh(f->header->arena_offset))
+                new_size = le64toh(f->header->arena_offset);
+
+        if (new_size <= old_size)
                 return 0;
 
-        asize = new_size - le64toh(f->header->arena_offset);
+        if (f->metrics.max_size > 0 &&
+            new_size > f->metrics.max_size)
+                return -E2BIG;
 
-        if (asize > le64toh(f->header->arena_min_size)) {
+        if (new_size > f->metrics.min_size &&
+            f->metrics.keep_free > 0) {
                 struct statvfs svfs;
 
                 if (fstatvfs(f->fd, &svfs) >= 0) {
@@ -192,8 +184,8 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
 
                         available = svfs.f_bfree * svfs.f_bsize;
 
-                        if (available >= f->header->arena_keep_free)
-                                available -= f->header->arena_keep_free;
+                        if (available >= f->metrics.keep_free)
+                                available -= f->metrics.keep_free;
                         else
                                 available = 0;
 
@@ -202,16 +194,16 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
                 }
         }
 
-        if (asize > le64toh(f->header->arena_max_size))
-                return -E2BIG;
-
+        /* Note that the glibc fallocate() fallback is very
+           inefficient, hence we try to minimize the allocation area
+           as we can. */
         if (posix_fallocate(f->fd, old_size, new_size - old_size) < 0)
                 return -errno;
 
         if (fstat(f->fd, &f->last_stat) < 0)
                 return -errno;
 
-        f->header->arena_size = htole64(asize);
+        f->header->arena_size = new_size - htole64(f->header->arena_offset);
 
         return 0;
 }
@@ -576,6 +568,9 @@ int journal_file_find_data_object_with_hash(
 
         osize = offsetof(Object, data.payload) + size;
 
+        if (f->header->data_hash_table_size == 0)
+                return -EBADMSG;
+
         h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
         p = le64toh(f->data_hash_table[h].head_hash_offset);
 
@@ -816,7 +811,7 @@ static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
         if (r < 0)
                 return r;
 
-        log_error("%s %lu", f->path, (unsigned long) f->header->n_entries);
+        log_error("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries);
 
         if (f->header->head_entry_realtime == 0)
                 f->header->head_entry_realtime = o->entry.realtime;
@@ -887,6 +882,8 @@ static void journal_file_post_change(JournalFile *f) {
          * trigger IN_MODIFY by truncating the journal file to its
          * current size which triggers IN_MODIFY. */
 
+        __sync_synchronize();
+
         if (ftruncate(f->fd, f->last_stat.st_size) < 0)
                 log_error("Failed to to truncate file to its own size: %m");
 }
@@ -1626,6 +1623,10 @@ int journal_file_open(
         f->writable = (flags & O_ACCMODE) != O_RDONLY;
         f->prot = prot_from_flags(flags);
 
+        f->metrics.max_size = DEFAULT_MAX_SIZE;
+        f->metrics.min_size = DEFAULT_MIN_SIZE;
+        f->metrics.keep_free = DEFAULT_KEEP_FREE;
+
         f->path = strdup(fname);
         if (!f->path) {
                 r = -ENOMEM;
diff --git a/src/journal/journal-file.h b/src/journal/journal-file.h
index 664f917bb0..20712b5dcc 100644
--- a/src/journal/journal-file.h
+++ b/src/journal/journal-file.h
@@ -28,6 +28,11 @@
 #include "util.h"
 #include "sd-id128.h"
 
+#define DEFAULT_MAX_SIZE (1024ULL*128ULL)
+#define DEFAULT_MIN_SIZE (256ULL*1024ULL)
+#define DEFAULT_KEEP_FREE (1ULL*1024ULL*1024ULL)
+#define DEFAULT_MAX_USE (16ULL*1024ULL*1024ULL*16ULL)
+
 typedef struct Window {
         void *ptr;
         uint64_t offset;
@@ -45,6 +50,12 @@ enum {
         _WINDOW_MAX
 };
 
+typedef struct JournalMetrics {
+        uint64_t max_size;
+        uint64_t min_size;
+        uint64_t keep_free;
+} JournalMetrics;
+
 typedef struct JournalFile {
         int fd;
         char *path;
@@ -62,6 +73,8 @@ typedef struct JournalFile {
         Window windows[_WINDOW_MAX];
 
         uint64_t current_offset;
+
+        JournalMetrics metrics;
 } JournalFile;
 
 typedef enum direction {
diff --git a/src/journal/journald.c b/src/journal/journald.c
index c457d2786b..37f8f16754 100644
--- a/src/journal/journald.c
+++ b/src/journal/journald.c
@@ -54,6 +54,9 @@ typedef struct Server {
 
         char *buffer;
         size_t buffer_size;
+
+        JournalMetrics metrics;
+        uint64_t max_use;
 } Server;
 
 static void fix_perms(JournalFile *f, uid_t uid) {
@@ -153,6 +156,66 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         return f;
 }
 
+static void server_vacuum(Server *s) {
+        Iterator i;
+        void *k;
+        char *p;
+        char ids[33];
+        sd_id128_t machine;
+        int r;
+        JournalFile *f;
+
+        log_info("Rotating...");
+
+        if (s->runtime_journal) {
+                r = journal_file_rotate(&s->runtime_journal);
+                if (r < 0)
+                        log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
+        }
+
+        if (s->system_journal) {
+                r = journal_file_rotate(&s->system_journal);
+                if (r < 0)
+                        log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
+        }
+
+        HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
+                r = journal_file_rotate(&f);
+                if (r < 0)
+                        log_error("Failed to rotate %s: %s", f->path, strerror(-r));
+                else
+                        hashmap_replace(s->user_journals, k, f);
+        }
+
+        log_info("Vacuuming...");
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0) {
+                log_error("Failed to get machine ID: %s", strerror(-r));
+                return;
+        }
+
+        if (asprintf(&p, "/var/log/journal/%s", sd_id128_to_string(machine, ids)) < 0) {
+                log_error("Out of memory.");
+                return;
+        }
+
+        r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
+        if (r < 0 && r != -ENOENT)
+                log_error("Failed to vacuum %s: %s", p, strerror(-r));
+        free(p);
+
+        if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
+                log_error("Out of memory.");
+                return;
+        }
+
+        r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
+        if (r < 0 && r != -ENOENT)
+                log_error("Failed to vacuum %s: %s", p, strerror(-r));
+        free(p);
+}
+
 static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigned m, struct ucred *ucred, struct timeval *tv) {
         char *pid = NULL, *uid = NULL, *gid = NULL,
                 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
@@ -166,6 +229,7 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
         char *t;
         uid_t loginuid = 0, realuid = 0;
         JournalFile *f;
+        bool vacuumed = false;
 
         assert(s);
         assert(iovec || n == 0);
@@ -262,12 +326,23 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
 
         assert(n <= m);
 
+retry:
         f = find_journal(s, realuid == 0 ? 0 : loginuid);
         if (!f)
                 log_warning("Dropping message, as we can't find a place to store the data.");
         else {
                 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
 
+                if (r == -E2BIG && !vacuumed) {
+                        log_info("Allocation limit reached.");
+
+                        server_vacuum(s);
+                        vacuumed = true;
+
+                        log_info("Retrying write.");
+                        goto retry;
+                }
+
                 if (r < 0)
                         log_error("Failed to write entry, ignoring: %s", strerror(-r));
         }
@@ -715,6 +790,10 @@ static int server_init(Server *s) {
 
         zero(*s);
         s->syslog_fd = s->native_fd = s->signal_fd = -1;
+        s->metrics.max_size = DEFAULT_MAX_SIZE;
+        s->metrics.min_size = DEFAULT_MIN_SIZE;
+        s->metrics.keep_free = DEFAULT_KEEP_FREE;
+        s->max_use = DEFAULT_MAX_USE;
 
         s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
         if (s->epoll_fd < 0) {
diff --git a/src/journal/sd-journal.h b/src/journal/sd-journal.h
index 33e4b78855..ee9813f28c 100644
--- a/src/journal/sd-journal.h
+++ b/src/journal/sd-journal.h
@@ -38,7 +38,6 @@
  *   - accelerate looking for "all hostnames" and suchlike.
  *   - throttling
  *   - cryptographic hash
- *   - fix space reservation logic
  *   - compression
  */