journal: be more careful when keeping around mmaps we still need

4 files changed, 35 insertions, 60 deletions

diff --git a/src/journal/journal-file.c b/src/journal/journal-file.c
index f01f12496..79f759897 100644
--- a/src/journal/journal-file.c
+++ b/src/journal/journal-file.c
@@ -311,8 +311,6 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
         if (r != 0)
                 return -r;
 
-        mmap_cache_close_fd_range(f->mmap, f->fd, old_size);
-
         if (fstat(f->fd, &f->last_stat) < 0)
                 return -errno;
 
@@ -321,7 +319,7 @@ static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size)
         return 0;
 }
 
-static int journal_file_move_to(JournalFile *f, int context, uint64_t offset, uint64_t size, void **ret) {
+static int journal_file_move_to(JournalFile *f, int context, bool keep_always, uint64_t offset, uint64_t size, void **ret) {
         assert(f);
         assert(ret);
 
@@ -335,7 +333,7 @@ static int journal_file_move_to(JournalFile *f, int context, uint64_t offset, ui
                         return -EADDRNOTAVAIL;
         }
 
-        return mmap_cache_get(f->mmap, f->fd, f->prot, context, offset, size, ret);
+        return mmap_cache_get(f->mmap, f->fd, f->prot, context, keep_always, offset, size, &f->last_stat, ret);
 }
 
 static uint64_t minimum_header_size(Object *o) {
@@ -373,7 +371,7 @@ int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Objec
         /* One context for each type, plus one catch-all for the rest */
         context = type > 0 && type < _OBJECT_TYPE_MAX ? type : 0;
 
-        r = journal_file_move_to(f, context, offset, sizeof(ObjectHeader), &t);
+        r = journal_file_move_to(f, context, false, offset, sizeof(ObjectHeader), &t);
         if (r < 0)
                 return r;
 
@@ -393,7 +391,7 @@ int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Objec
                 return -EBADMSG;
 
         if (s > sizeof(ObjectHeader)) {
-                r = journal_file_move_to(f, o->object.type, offset, s, &t);
+                r = journal_file_move_to(f, o->object.type, false, offset, s, &t);
                 if (r < 0)
                         return r;
 
@@ -457,7 +455,7 @@ int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object *
         if (r < 0)
                 return r;
 
-        r = journal_file_move_to(f, type, p, size, &t);
+        r = journal_file_move_to(f, type, false, p, size, &t);
         if (r < 0)
                 return r;
 
@@ -544,6 +542,7 @@ static int journal_file_map_data_hash_table(JournalFile *f) {
 
         r = journal_file_move_to(f,
                                  OBJECT_DATA_HASH_TABLE,
+                                 true,
                                  p, s,
                                  &t);
         if (r < 0)
@@ -565,6 +564,7 @@ static int journal_file_map_field_hash_table(JournalFile *f) {
 
         r = journal_file_move_to(f,
                                  OBJECT_FIELD_HASH_TABLE,
+                                 true,
                                  p, s,
                                  &t);
         if (r < 0)
diff --git a/src/journal/journal-verify.c b/src/journal/journal-verify.c
index 90739a66c..9156fd5dc 100644
--- a/src/journal/journal-verify.c
+++ b/src/journal/journal-verify.c
@@ -249,7 +249,7 @@ static int contains_uint64(MMapCache *m, int fd, uint64_t n, uint64_t p) {
 
                 c = (a + b) / 2;
 
-                r = mmap_cache_get(m, fd, PROT_READ|PROT_WRITE, 0, c * sizeof(uint64_t), sizeof(uint64_t), (void **) &z);
+                r = mmap_cache_get(m, fd, PROT_READ|PROT_WRITE, 0, false, c * sizeof(uint64_t), sizeof(uint64_t), NULL, (void **) &z);
                 if (r < 0)
                         return r;
 
@@ -261,9 +261,8 @@ static int contains_uint64(MMapCache *m, int fd, uint64_t n, uint64_t p) {
 
                 if (p < *z)
                         b = c;
-                else {
+                else
                         a = c;
-                }
         }
 
         return 0;
diff --git a/src/journal/mmap-cache.c b/src/journal/mmap-cache.c
index 69efb20ad..2b45e375c 100644
--- a/src/journal/mmap-cache.c
+++ b/src/journal/mmap-cache.c
@@ -336,6 +336,7 @@ static int mmap_cache_make_room(MMapCache *m) {
                 Window *v;
 
                 v = m->windows + w;
+                assert(v->n_ref == 0);
 
                 if (v->ptr) {
                         mmap_cache_window_unmap(m, w);
@@ -354,8 +355,10 @@ static int mmap_cache_put(
                 unsigned fd_index,
                 int prot,
                 unsigned context,
+                bool keep_always,
                 uint64_t offset,
                 uint64_t size,
+                struct stat *st,
                 void **ret) {
 
         unsigned w;
@@ -387,6 +390,18 @@ static int mmap_cache_put(
                 wsize = WINDOW_SIZE;
         }
 
+        if (st) {
+                /* Memory maps that are larger then the files
+                   underneath have undefined behaviour. Hence, clamp
+                   things to the file size if we know it */
+
+                if (woffset >= (uint64_t) st->st_size)
+                        return -EADDRNOTAVAIL;
+
+                if (woffset + wsize > (uint64_t) st->st_size)
+                        wsize = PAGE_ALIGN(st->st_size - woffset);
+        }
+
         for (;;) {
                 d = mmap(NULL, wsize, prot, MAP_SHARED, fd, woffset);
                 if (d != MAP_FAILED)
@@ -413,8 +428,13 @@ static int mmap_cache_put(
         v->offset = woffset;
         v->size = wsize;
 
-        v->n_ref = 0;
-        mmap_cache_window_add_lru(m, w);
+        if (keep_always)
+                v->n_ref = 1;
+        else {
+                v->n_ref = 0;
+                mmap_cache_window_add_lru(m, w);
+        }
+
         mmap_cache_fd_add(m, fd_index, w);
         mmap_cache_context_set(m, context, w);
 
@@ -581,8 +601,10 @@ int mmap_cache_get(
                 int fd,
                 int prot,
                 unsigned context,
+                bool keep_always,
                 uint64_t offset,
                 uint64_t size,
+                struct stat *st,
                 void **ret) {
 
         unsigned fd_index;
@@ -638,7 +660,7 @@ int mmap_cache_get(
                 return r;
 
         /* Not found? Then, let's add it */
-        return mmap_cache_put(m, fd, fd_index, prot, context, offset, size, ret);
+        return mmap_cache_put(m, fd, fd_index, prot, context, keep_always, offset, size, st, ret);
 }
 
 void mmap_cache_close_fd(MMapCache *m, int fd) {
@@ -679,51 +701,6 @@ void mmap_cache_close_fd(MMapCache *m, int fd) {
         m->n_fds --;
 }
 
-void mmap_cache_close_fd_range(MMapCache *m, int fd, uint64_t p) {
-        unsigned fd_index, c, w;
-        int r;
-
-        assert(m);
-        assert(fd > 0);
-
-        /* This drops all windows that include space right of the
-         * specified offset. This is useful to ensure that after the
-         * file size is extended we drop our mappings of the end and
-         * create it anew, since otherwise it is undefined whether
-         * mapping will continue to work as intended. */
-
-        r = mmap_cache_peek_fd_index(m, fd, &fd_index);
-        if (r <= 0)
-                return;
-
-        for (c = 0; c < m->contexts_max; c++) {
-                w = m->by_context[c];
-
-                if (w != (unsigned) -1 && m->windows[w].fd == fd)
-                        mmap_cache_context_unset(m, c);
-        }
-
-        w = m->by_fd[fd_index].windows;
-        while (w != (unsigned) -1) {
-                Window *v;
-
-                v = m->windows + w;
-                assert(v->fd == fd);
-                assert(v->by_fd_next == (unsigned) -1 ||
-                       m->windows[v->by_fd_next].fd == fd);
-
-                if (v->offset + v->size > p) {
-
-                        mmap_cache_window_unmap(m, w);
-                        mmap_cache_fd_remove(m, fd_index, w);
-                        v->fd = -1;
-
-                        w = m->by_fd[fd_index].windows;
-                } else
-                        w = v->by_fd_next;
-        }
-}
-
 void mmap_cache_close_context(MMapCache *m, unsigned context) {
         mmap_cache_context_unset(m, context);
 }
diff --git a/src/journal/mmap-cache.h b/src/journal/mmap-cache.h
index aae454435..98a1a731f 100644
--- a/src/journal/mmap-cache.h
+++ b/src/journal/mmap-cache.h
@@ -29,7 +29,6 @@ MMapCache* mmap_cache_new(void);
 MMapCache* mmap_cache_ref(MMapCache *m);
 MMapCache* mmap_cache_unref(MMapCache *m);
 
-int mmap_cache_get(MMapCache *m, int fd, int prot, unsigned context, uint64_t offset, uint64_t size, void **ret);
+int mmap_cache_get(MMapCache *m, int fd, int prot, unsigned context, bool keep_always, uint64_t offset, uint64_t size, struct stat *st, void **ret);
 void mmap_cache_close_fd(MMapCache *m, int fd);
-void mmap_cache_close_fd_range(MMapCache *m, int fd, uint64_t range);
 void mmap_cache_close_context(MMapCache *m, unsigned context);