]> git.ipfire.org Git - thirdparty/systemd.git/blobdiff - src/journal/journal-file.c
util-lib: split our string related calls from util.[ch] into its own file string...
[thirdparty/systemd.git] / src / journal / journal-file.c
index fec54f31b9888730d6efb933f1a5532f92e93cc8..f0962faeb5150d54dfc7df79a37862e7e6b31d19 100644 (file)
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 ***/
 
-#include <sys/mman.h>
 #include <errno.h>
-#include <sys/uio.h>
-#include <unistd.h>
-#include <sys/statvfs.h>
 #include <fcntl.h>
+#include <linux/fs.h>
 #include <stddef.h>
-#include <sys/xattr.h>
+#include <sys/mman.h>
+#include <sys/statvfs.h>
+#include <sys/uio.h>
+#include <unistd.h>
 
-#include "journal-def.h"
-#include "journal-file.h"
+#include "btrfs-util.h"
+#include "compress.h"
 #include "journal-authenticate.h"
+#include "journal-def.h"
 #include "lookup3.h"
-#include "compress.h"
-#include "fsprg.h"
+#include "random-util.h"
+#include "string-util.h"
+#include "journal-file.h"
 
 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
@@ -48,6 +50,9 @@
 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL)           /* 1 MiB */
 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL)   /* 4 GiB */
 
+/* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
+#define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL)                 /* 1 MiB */
+
 /* This is the upper bound if we deduce max_size from max_use */
 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL)        /* 128 MiB */
 
@@ -59,6 +64,9 @@
  * size */
 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL)                    /* 1 MB */
 
+/* This is the default maximum number of journal files to keep around. */
+#define DEFAULT_N_MAX_FILES (100)
+
 /* n_data was the first entry we added after the initial file format design */
 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
 
 /* How much to increase the journal file size at once each time we allocate something new. */
 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL)              /* 8MB */
 
+/* Reread fstat() of the file for detecting deletions at least this often */
+#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
+
+/* The mmap context to use for the header we pick as one above the last defined typed */
+#define CONTEXT_HEADER _OBJECT_TYPE_MAX
+
 static int journal_file_set_online(JournalFile *f) {
         assert(f);
 
@@ -77,6 +91,9 @@ static int journal_file_set_online(JournalFile *f) {
         if (!(f->fd >= 0 && f->header))
                 return -EINVAL;
 
+        if (mmap_cache_got_sigbus(f->mmap, f->fd))
+                return -EIO;
+
         switch(f->header->state) {
                 case STATE_ONLINE:
                         return 0;
@@ -105,14 +122,20 @@ int journal_file_set_offline(JournalFile *f) {
 
         fsync(f->fd);
 
+        if (mmap_cache_got_sigbus(f->mmap, f->fd))
+                return -EIO;
+
         f->header->state = STATE_OFFLINE;
 
+        if (mmap_cache_got_sigbus(f->mmap, f->fd))
+                return -EIO;
+
         fsync(f->fd);
 
         return 0;
 }
 
-void journal_file_close(JournalFile *f) {
+JournalFile* journal_file_close(JournalFile *f) {
         assert(f);
 
 #ifdef HAVE_GCRYPT
@@ -121,14 +144,22 @@ void journal_file_close(JournalFile *f) {
                 journal_file_append_tag(f);
 #endif
 
-        /* Sync everything to disk, before we mark the file offline */
+        journal_file_set_offline(f);
+
         if (f->mmap && f->fd >= 0)
                 mmap_cache_close_fd(f->mmap, f->fd);
 
-        journal_file_set_offline(f);
+        if (f->fd >= 0 && f->defrag_on_close) {
 
-        if (f->header)
-                munmap(f->header, PAGE_ALIGN(sizeof(Header)));
+                /* Be friendly to btrfs: turn COW back on again now,
+                 * and defragment the file. We won't write to the file
+                 * ever again, hence remove all fragmentation, and
+                 * reenable all the good bits COW usually provides
+                 * (such as data checksumming). */
+
+                (void) chattr_fd(f->fd, 0, FS_NOCOW_FL);
+                (void) btrfs_defrag_fd(f->fd);
+        }
 
         safe_close(f->fd);
         free(f->path);
@@ -145,7 +176,7 @@ void journal_file_close(JournalFile *f) {
 #ifdef HAVE_GCRYPT
         if (f->fss_file)
                 munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
-        else if (f->fsprg_state)
+        else
                 free(f->fsprg_state);
 
         free(f->fsprg_seed);
@@ -155,6 +186,7 @@ void journal_file_close(JournalFile *f) {
 #endif
 
         free(f);
+        return NULL;
 }
 
 static int journal_file_init_header(JournalFile *f, JournalFile *template) {
@@ -195,8 +227,8 @@ static int journal_file_init_header(JournalFile *f, JournalFile *template) {
 }
 
 static int journal_file_refresh_header(JournalFile *f) {
-        int r;
         sd_id128_t boot_id;
+        int r;
 
         assert(f);
 
@@ -213,12 +245,12 @@ static int journal_file_refresh_header(JournalFile *f) {
 
         f->header->boot_id = boot_id;
 
-        journal_file_set_online(f);
+        r = journal_file_set_online(f);
 
         /* Sync the online state to disk */
         fsync(f->fd);
 
-        return 0;
+        return r;
 }
 
 static int journal_file_verify_header(JournalFile *f) {
@@ -299,7 +331,7 @@ static int journal_file_verify_header(JournalFile *f) {
                 } else if (state == STATE_ARCHIVED)
                         return -ESHUTDOWN;
                 else if (state != STATE_OFFLINE) {
-                        log_debug("Journal file %s has unknown state %u.", f->path, state);
+                        log_debug("Journal file %s has unknown state %i.", f->path, state);
                         return -EBUSY;
                 }
         }
@@ -312,6 +344,22 @@ static int journal_file_verify_header(JournalFile *f) {
         return 0;
 }
 
+static int journal_file_fstat(JournalFile *f) {
+        assert(f);
+        assert(f->fd >= 0);
+
+        if (fstat(f->fd, &f->last_stat) < 0)
+                return -errno;
+
+        f->last_stat_usec = now(CLOCK_MONOTONIC);
+
+        /* Refuse appending to files that are already deleted */
+        if (f->last_stat.st_nlink <= 0)
+                return -EIDRM;
+
+        return 0;
+}
+
 static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
         uint64_t old_size, new_size;
         int r;
@@ -322,6 +370,9 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
          * for sure, since we always call posix_fallocate()
          * ourselves */
 
+        if (mmap_cache_got_sigbus(f->mmap, f->fd))
+                return -EIO;
+
         old_size =
                 le64toh(f->header->header_size) +
                 le64toh(f->header->arena_size);
@@ -330,8 +381,21 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
         if (new_size < le64toh(f->header->header_size))
                 new_size = le64toh(f->header->header_size);
 
-        if (new_size <= old_size)
-                return 0;
+        if (new_size <= old_size) {
+
+                /* We already pre-allocated enough space, but before
+                 * we write to it, let's check with fstat() if the
+                 * file got deleted, in order make sure we don't throw
+                 * away the data immediately. Don't check fstat() for
+                 * all writes though, but only once ever 10s. */
+
+                if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
+                        return 0;
+
+                return journal_file_fstat(f);
+        }
+
+        /* Allocate more space. */
 
         if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
                 return -E2BIG;
@@ -342,12 +406,7 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
                 if (fstatvfs(f->fd, &svfs) >= 0) {
                         uint64_t available;
 
-                        available = svfs.f_bfree * svfs.f_bsize;
-
-                        if (available >= f->metrics.keep_free)
-                                available -= f->metrics.keep_free;
-                        else
-                                available = 0;
+                        available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
 
                         if (new_size - old_size > available)
                                 return -E2BIG;
@@ -366,21 +425,21 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
         if (r != 0)
                 return -r;
 
-        if (fstat(f->fd, &f->last_stat) < 0)
-                return -errno;
-
         f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
 
-        return 0;
+        return journal_file_fstat(f);
 }
 
 static unsigned type_to_context(ObjectType type) {
         /* One context for each type, plus one catch-all for the rest */
         assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
+        assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
         return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
 }
 
 static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
+        int r;
+
         assert(f);
         assert(ret);
 
@@ -392,8 +451,11 @@ static int journal_file_move_to(JournalFile *f, ObjectType type, bool keep_alway
                 /* Hmm, out of range? Let's refresh the fstat() data
                  * first, before we trust that check. */
 
-                if (fstat(f->fd, &f->last_stat) < 0 ||
-                    offset + size > (uint64_t) f->last_stat.st_size)
+                r = journal_file_fstat(f);
+                if (r < 0)
+                        return r;
+
+                if (offset + size > (uint64_t) f->last_stat.st_size)
                         return -EADDRNOTAVAIL;
         }
 
@@ -545,10 +607,10 @@ static int journal_file_setup_data_hash_table(JournalFile *f) {
 
         assert(f);
 
-        /* We estimate that we need 1 hash table entry per 768 of
-           journal file and we want to make sure we never get beyond
-           75% fill level. Calculate the hash table size for the
-           maximum file size based on these metrics. */
+        /* We estimate that we need 1 hash table entry per 768 bytes
+           of journal file and we want to make sure we never get
+           beyond 75% fill level. Calculate the hash table size for
+           the maximum file size based on these metrics. */
 
         s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
         if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
@@ -597,13 +659,16 @@ static int journal_file_setup_field_hash_table(JournalFile *f) {
         return 0;
 }
 
-static int journal_file_map_data_hash_table(JournalFile *f) {
+int journal_file_map_data_hash_table(JournalFile *f) {
         uint64_t s, p;
         void *t;
         int r;
 
         assert(f);
 
+        if (f->data_hash_table)
+                return 0;
+
         p = le64toh(f->header->data_hash_table_offset);
         s = le64toh(f->header->data_hash_table_size);
 
@@ -619,13 +684,16 @@ static int journal_file_map_data_hash_table(JournalFile *f) {
         return 0;
 }
 
-static int journal_file_map_field_hash_table(JournalFile *f) {
+int journal_file_map_field_hash_table(JournalFile *f) {
         uint64_t s, p;
         void *t;
         int r;
 
         assert(f);
 
+        if (f->field_hash_table)
+                return 0;
+
         p = le64toh(f->header->field_hash_table_offset);
         s = le64toh(f->header->field_hash_table_size);
 
@@ -647,7 +715,7 @@ static int journal_file_link_field(
                 uint64_t offset,
                 uint64_t hash) {
 
-        uint64_t p, h;
+        uint64_t p, h, m;
         int r;
 
         assert(f);
@@ -657,11 +725,14 @@ static int journal_file_link_field(
         if (o->object.type != OBJECT_FIELD)
                 return -EINVAL;
 
-        /* This might alter the window we are looking at */
+        m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
+        if (m <= 0)
+                return -EBADMSG;
 
+        /* This might alter the window we are looking at */
         o->field.next_hash_offset = o->field.head_data_offset = 0;
 
-        h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
+        h = hash % m;
         p = le64toh(f->field_hash_table[h].tail_hash_offset);
         if (p == 0)
                 f->field_hash_table[h].head_hash_offset = htole64(offset);
@@ -687,7 +758,7 @@ static int journal_file_link_data(
                 uint64_t offset,
                 uint64_t hash) {
 
-        uint64_t p, h;
+        uint64_t p, h, m;
         int r;
 
         assert(f);
@@ -697,13 +768,16 @@ static int journal_file_link_data(
         if (o->object.type != OBJECT_DATA)
                 return -EINVAL;
 
-        /* This might alter the window we are looking at */
+        m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
+        if (m <= 0)
+                return -EBADMSG;
 
+        /* This might alter the window we are looking at */
         o->data.next_hash_offset = o->data.next_field_offset = 0;
         o->data.entry_offset = o->data.entry_array_offset = 0;
         o->data.n_entries = 0;
 
-        h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+        h = hash % m;
         p = le64toh(f->data_hash_table[h].tail_hash_offset);
         if (p == 0)
                 /* Only entry in the hash table is easy */
@@ -732,18 +806,28 @@ int journal_file_find_field_object_with_hash(
                 const void *field, uint64_t size, uint64_t hash,
                 Object **ret, uint64_t *offset) {
 
-        uint64_t p, osize, h;
+        uint64_t p, osize, h, m;
         int r;
 
         assert(f);
         assert(field && size > 0);
 
+        /* If the field hash table is empty, we can't find anything */
+        if (le64toh(f->header->field_hash_table_size) <= 0)
+                return 0;
+
+        /* Map the field hash table, if it isn't mapped yet. */
+        r = journal_file_map_field_hash_table(f);
+        if (r < 0)
+                return r;
+
         osize = offsetof(Object, field.payload) + size;
 
-        if (f->header->field_hash_table_size == 0)
+        m = le64toh(f->header->field_hash_table_size) / sizeof(HashItem);
+        if (m <= 0)
                 return -EBADMSG;
 
-        h = hash % (le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
+        h = hash % m;
         p = le64toh(f->field_hash_table[h].head_hash_offset);
 
         while (p > 0) {
@@ -793,18 +877,28 @@ int journal_file_find_data_object_with_hash(
                 const void *data, uint64_t size, uint64_t hash,
                 Object **ret, uint64_t *offset) {
 
-        uint64_t p, osize, h;
+        uint64_t p, osize, h, m;
         int r;
 
         assert(f);
         assert(data || size == 0);
 
+        /* If there's no data hash table, then there's no entry. */
+        if (le64toh(f->header->data_hash_table_size) <= 0)
+                return 0;
+
+        /* Map the data hash table, if it isn't mapped yet. */
+        r = journal_file_map_data_hash_table(f);
+        if (r < 0)
+                return r;
+
         osize = offsetof(Object, data.payload) + size;
 
-        if (f->header->data_hash_table_size == 0)
+        m = le64toh(f->header->data_hash_table_size) / sizeof(HashItem);
+        if (m <= 0)
                 return -EBADMSG;
 
-        h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
+        h = hash % m;
         p = le64toh(f->data_hash_table[h].head_hash_offset);
 
         while (p > 0) {
@@ -820,7 +914,7 @@ int journal_file_find_data_object_with_hash(
                 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
                         uint64_t l;
-                        size_t rsize;
+                        size_t rsize = 0;
 
                         l = le64toh(o->object.size);
                         if (l <= offsetof(Object, data.payload))
@@ -985,7 +1079,7 @@ static int journal_file_append_data(
 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
         if (f->compress_xz &&
             size >= COMPRESSION_SIZE_THRESHOLD) {
-                size_t rsize;
+                size_t rsize = 0;
 
                 compression = compress_blob(data, size, o->data.payload, &rsize);
 
@@ -1358,6 +1452,14 @@ int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const st
 
         r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
 
+        /* If the memory mapping triggered a SIGBUS then we return an
+         * IO error and ignore the error code passed down to us, since
+         * it is very likely just an effect of a nullified replacement
+         * mapping page */
+
+        if (mmap_cache_got_sigbus(f->mmap, f->fd))
+                r = -EIO;
+
         journal_file_post_change(f);
 
         return r;
@@ -1713,7 +1815,6 @@ found:
         return 1;
 }
 
-
 static int generic_array_bisect_plus_one(
                 JournalFile *f,
                 uint64_t extra,
@@ -1888,11 +1989,12 @@ static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
                 return TEST_RIGHT;
 }
 
-static inline int find_data_object_by_boot_id(
+static int find_data_object_by_boot_id(
                 JournalFile *f,
                 sd_id128_t boot_id,
                 Object **o,
                 uint64_t *b) {
+
         char t[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
 
         sd_id128_to_string(boot_id, t + 9);
@@ -1938,8 +2040,7 @@ void journal_file_reset_location(JournalFile *f) {
         f->current_xor_hash = 0;
 }
 
-void journal_file_save_location(JournalFile *f, direction_t direction, Object *o, uint64_t offset) {
-        f->last_direction = direction;
+void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
         f->location_type = LOCATION_SEEK;
         f->current_offset = offset;
         f->current_seqnum = le64toh(o->entry.seqnum);
@@ -2342,7 +2443,7 @@ void journal_file_dump(JournalFile *f) {
                         break;
 
                 default:
-                        printf("Type: unknown (%u)\n", o->object.type);
+                        printf("Type: unknown (%i)\n", o->object.type);
                         break;
                 }
 
@@ -2444,7 +2545,42 @@ void journal_file_print_header(JournalFile *f) {
                        le64toh(f->header->n_entry_arrays));
 
         if (fstat(f->fd, &st) >= 0)
-                printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (off_t) st.st_blocks * 512ULL));
+                printf("Disk usage: %s\n", format_bytes(bytes, sizeof(bytes), (uint64_t) st.st_blocks * 512ULL));
+}
+
+static int journal_file_warn_btrfs(JournalFile *f) {
+        unsigned attrs;
+        int r;
+
+        assert(f);
+
+        /* Before we write anything, check if the COW logic is turned
+         * off on btrfs. Given our write pattern that is quite
+         * unfriendly to COW file systems this should greatly improve
+         * performance on COW file systems, such as btrfs, at the
+         * expense of data integrity features (which shouldn't be too
+         * bad, given that we do our own checksumming). */
+
+        r = btrfs_is_filesystem(f->fd);
+        if (r < 0)
+                return log_warning_errno(r, "Failed to determine if journal is on btrfs: %m");
+        if (!r)
+                return 0;
+
+        r = read_attr_fd(f->fd, &attrs);
+        if (r < 0)
+                return log_warning_errno(r, "Failed to read file attributes: %m");
+
+        if (attrs & FS_NOCOW_FL) {
+                log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
+                return 0;
+        }
+
+        log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
+                   "This is likely to slow down journal access substantially, please consider turning "
+                   "off the copy-on-write file attribute on the journal directory, using chattr +C.", f->path);
+
+        return 1;
 }
 
 int journal_file_open(
@@ -2458,9 +2594,10 @@ int journal_file_open(
                 JournalFile *template,
                 JournalFile **ret) {
 
+        bool newly_created = false;
         JournalFile *f;
+        void *h;
         int r;
-        bool newly_created = false;
 
         assert(fname);
         assert(ret);
@@ -2520,13 +2657,13 @@ int journal_file_open(
                 goto fail;
         }
 
-        if (fstat(f->fd, &f->last_stat) < 0) {
-                r = -errno;
+        r = journal_file_fstat(f);
+        if (r < 0)
                 goto fail;
-        }
 
         if (f->last_stat.st_size == 0 && f->writable) {
-                uint64_t crtime;
+
+                (void) journal_file_warn_btrfs(f);
 
                 /* Let's attach the creation time to the journal file,
                  * so that the vacuuming code knows the age of this
@@ -2538,8 +2675,7 @@ int journal_file_open(
                  * attributes are not supported we'll just skip this,
                  * and rely solely on mtime/atime/ctime of the file. */
 
-                crtime = htole64((uint64_t) now(CLOCK_REALTIME));
-                fsetxattr(f->fd, "user.crtime_usec", &crtime, sizeof(crtime), XATTR_CREATE);
+                fd_setcrtime(f->fd, 0);
 
 #ifdef HAVE_GCRYPT
                 /* Try to load the FSPRG state, and if we can't, then
@@ -2555,10 +2691,9 @@ int journal_file_open(
                 if (r < 0)
                         goto fail;
 
-                if (fstat(f->fd, &f->last_stat) < 0) {
-                        r = -errno;
+                r = journal_file_fstat(f);
+                if (r < 0)
                         goto fail;
-                }
 
                 newly_created = true;
         }
@@ -2568,12 +2703,11 @@ int journal_file_open(
                 goto fail;
         }
 
-        f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
-        if (f->header == MAP_FAILED) {
-                f->header = NULL;
-                r = -errno;
+        r = mmap_cache_get(f->mmap, f->fd, f->prot, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
+        if (r < 0)
                 goto fail;
-        }
+
+        f->header = h;
 
         if (!newly_created) {
                 r = journal_file_verify_header(f);
@@ -2623,18 +2757,18 @@ int journal_file_open(
 #endif
         }
 
-        r = journal_file_map_field_hash_table(f);
-        if (r < 0)
-                goto fail;
-
-        r = journal_file_map_data_hash_table(f);
-        if (r < 0)
+        if (mmap_cache_got_sigbus(f->mmap, f->fd)) {
+                r = -EIO;
                 goto fail;
+        }
 
         *ret = f;
         return 0;
 
 fail:
+        if (f->fd >= 0 && mmap_cache_got_sigbus(f->mmap, f->fd))
+                r = -EIO;
+
         journal_file_close(f);
 
         return r;
@@ -2666,12 +2800,20 @@ int journal_file_rotate(JournalFile **f, bool compress, bool seal) {
         if (r < 0)
                 return -ENOMEM;
 
+        /* Try to rename the file to the archived version. If the file
+         * already was deleted, we'll get ENOENT, let's ignore that
+         * case. */
         r = rename(old_file->path, p);
-        if (r < 0)
+        if (r < 0 && errno != ENOENT)
                 return -errno;
 
         old_file->header->state = STATE_ARCHIVED;
 
+        /* Currently, btrfs is not very good with out write patterns
+         * and fragments heavily. Let's defrag our journal files when
+         * we archive them */
+        old_file->defrag_on_close = true;
+
         r = journal_file_open(old_file->path, old_file->flags, old_file->mode, compress, seal, NULL, old_file->mmap, old_file, &new_file);
         journal_file_close(old_file);
 
@@ -2694,14 +2836,16 @@ int journal_file_open_reliably(
         size_t l;
         _cleanup_free_ char *p = NULL;
 
-        r = journal_file_open(fname, flags, mode, compress, seal,
-                              metrics, mmap_cache, template, ret);
-        if (r != -EBADMSG && /* corrupted */
-            r != -ENODATA && /* truncated */
-            r != -EHOSTDOWN && /* other machine */
-            r != -EPROTONOSUPPORT && /* incompatible feature */
-            r != -EBUSY && /* unclean shutdown */
-            r != -ESHUTDOWN /* already archived */)
+        r = journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret);
+        if (!IN_SET(r,
+                    -EBADMSG,           /* corrupted */
+                    -ENODATA,           /* truncated */
+                    -EHOSTDOWN,         /* other machine */
+                    -EPROTONOSUPPORT,   /* incompatible feature */
+                    -EBUSY,             /* unclean shutdown */
+                    -ESHUTDOWN,         /* already archived */
+                    -EIO,               /* IO error, including SIGBUS on mmap */
+                    -EIDRM              /* File has been deleted */))
                 return r;
 
         if ((flags & O_ACCMODE) == O_RDONLY)
@@ -2716,20 +2860,24 @@ int journal_file_open_reliably(
         /* The file is corrupted. Rotate it away and try it again (but only once) */
 
         l = strlen(fname);
-        if (asprintf(&p, "%.*s@%016llx-%016" PRIx64 ".journal~",
+        if (asprintf(&p, "%.*s@%016"PRIx64 "-%016"PRIx64 ".journal~",
                      (int) l - 8, fname,
-                     (unsigned long long) now(CLOCK_REALTIME),
+                     now(CLOCK_REALTIME),
                      random_u64()) < 0)
                 return -ENOMEM;
 
-        r = rename(fname, p);
-        if (r < 0)
+        if (rename(fname, p) < 0)
                 return -errno;
 
-        log_warning("File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
+        /* btrfs doesn't cope well with our write pattern and
+         * fragments heavily. Let's defrag all files we rotate */
 
-        return journal_file_open(fname, flags, mode, compress, seal,
-                                 metrics, mmap_cache, template, ret);
+        (void) chattr_path(p, false, FS_NOCOW_FL);
+        (void) btrfs_defrag(p);
+
+        log_warning_errno(r, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname);
+
+        return journal_file_open(fname, flags, mode, compress, seal, metrics, mmap_cache, template, ret);
 }
 
 int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
@@ -2780,7 +2928,7 @@ int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint6
 
                 if (o->object.flags & OBJECT_COMPRESSION_MASK) {
 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
-                        size_t rsize;
+                        size_t rsize = 0;
 
                         r = decompress_blob(o->object.flags & OBJECT_COMPRESSION_MASK,
                                             o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize, 0);
@@ -2808,19 +2956,43 @@ int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint6
                         return r;
         }
 
-        return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
+        r = journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
+
+        if (mmap_cache_got_sigbus(to->mmap, to->fd))
+                return -EIO;
+
+        return r;
+}
+
+void journal_reset_metrics(JournalMetrics *m) {
+        assert(m);
+
+        /* Set everything to "pick automatic values". */
+
+        *m = (JournalMetrics) {
+                .min_use = (uint64_t) -1,
+                .max_use = (uint64_t) -1,
+                .min_size = (uint64_t) -1,
+                .max_size = (uint64_t) -1,
+                .keep_free = (uint64_t) -1,
+                .n_max_files = (uint64_t) -1,
+        };
 }
 
 void journal_default_metrics(JournalMetrics *m, int fd) {
-        uint64_t fs_size = 0;
+        char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX], e[FORMAT_BYTES_MAX];
         struct statvfs ss;
-        char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
+        uint64_t fs_size;
 
         assert(m);
         assert(fd >= 0);
 
         if (fstatvfs(fd, &ss) >= 0)
                 fs_size = ss.f_frsize * ss.f_blocks;
+        else {
+                log_debug_errno(errno, "Failed to detremine disk size: %m");
+                fs_size = 0;
+        }
 
         if (m->max_use == (uint64_t) -1) {
 
@@ -2837,10 +3009,16 @@ void journal_default_metrics(JournalMetrics *m, int fd) {
         } else {
                 m->max_use = PAGE_ALIGN(m->max_use);
 
-                if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
+                if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
                         m->max_use = JOURNAL_FILE_SIZE_MIN*2;
         }
 
+        if (m->min_use == (uint64_t) -1)
+                m->min_use = DEFAULT_MIN_USE;
+
+        if (m->min_use > m->max_use)
+                m->min_use = m->max_use;
+
         if (m->max_size == (uint64_t) -1) {
                 m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
 
@@ -2849,11 +3027,13 @@ void journal_default_metrics(JournalMetrics *m, int fd) {
         } else
                 m->max_size = PAGE_ALIGN(m->max_size);
 
-        if (m->max_size < JOURNAL_FILE_SIZE_MIN)
-                m->max_size = JOURNAL_FILE_SIZE_MIN;
+        if (m->max_size != 0) {
+                if (m->max_size < JOURNAL_FILE_SIZE_MIN)
+                        m->max_size = JOURNAL_FILE_SIZE_MIN;
 
-        if (m->max_size*2 > m->max_use)
-                m->max_use = m->max_size*2;
+                if (m->max_use != 0 && m->max_size*2 > m->max_use)
+                        m->max_use = m->max_size*2;
+        }
 
         if (m->min_size == (uint64_t) -1)
                 m->min_size = JOURNAL_FILE_SIZE_MIN;
@@ -2863,7 +3043,7 @@ void journal_default_metrics(JournalMetrics *m, int fd) {
                 if (m->min_size < JOURNAL_FILE_SIZE_MIN)
                         m->min_size = JOURNAL_FILE_SIZE_MIN;
 
-                if (m->min_size > m->max_size)
+                if (m->max_size != 0 && m->min_size > m->max_size)
                         m->max_size = m->min_size;
         }
 
@@ -2879,11 +3059,16 @@ void journal_default_metrics(JournalMetrics *m, int fd) {
                         m->keep_free = DEFAULT_KEEP_FREE;
         }
 
-        log_debug("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
-                  format_bytes(a, sizeof(a), m->max_use),
-                  format_bytes(b, sizeof(b), m->max_size),
-                  format_bytes(c, sizeof(c), m->min_size),
-                  format_bytes(d, sizeof(d), m->keep_free));
+        if (m->n_max_files == (uint64_t) -1)
+                m->n_max_files = DEFAULT_N_MAX_FILES;
+
+        log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
+                  format_bytes(a, sizeof(a), m->min_use),
+                  format_bytes(b, sizeof(b), m->max_use),
+                  format_bytes(c, sizeof(c), m->max_size),
+                  format_bytes(d, sizeof(d), m->min_size),
+                  format_bytes(e, sizeof(e), m->keep_free),
+                  m->n_max_files);
 }
 
 int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {