diff options
Diffstat (limited to 'fs/ceph/snap.c')
| -rw-r--r-- | fs/ceph/snap.c | 70 |
1 files changed, 55 insertions, 15 deletions
diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c index 54b14de..d5df940 100644 --- a/fs/ceph/snap.c +++ b/fs/ceph/snap.c @@ -288,6 +288,9 @@ static int cmpu64_rev(const void *a, const void *b) return 0; } + +static struct ceph_snap_context *empty_snapc; + /* * build the snap context for a given realm. */ @@ -329,9 +332,15 @@ static int build_snap_context(struct ceph_snap_realm *realm) return 0; } + if (num == 0 && realm->seq == empty_snapc->seq) { + ceph_get_snap_context(empty_snapc); + snapc = empty_snapc; + goto done; + } + /* alloc new snap context */ err = -ENOMEM; - if (num > ULONG_MAX / sizeof(u64) - sizeof(*snapc)) + if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64)) goto fail; snapc = kzalloc(sizeof(*snapc) + num*sizeof(u64), GFP_NOFS); if (!snapc) @@ -364,6 +373,7 @@ static int build_snap_context(struct ceph_snap_realm *realm) dout("build_snap_context %llx %p: %p seq %lld (%d snaps)\n", realm->ino, realm, snapc, snapc->seq, snapc->num_snaps); +done: if (realm->cached_context) ceph_put_snap_context(realm->cached_context); realm->cached_context = snapc; @@ -446,9 +456,18 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci) return; } - spin_lock(&inode->i_lock); + spin_lock(&ci->i_ceph_lock); used = __ceph_caps_used(ci); dirty = __ceph_caps_dirty(ci); + + /* + * If there is a write in progress, treat that as a dirty Fw, + * even though it hasn't completed yet; by the time we finish + * up this capsnap it will be. + */ + if (used & CEPH_CAP_FILE_WR) + dirty |= CEPH_CAP_FILE_WR; + if (__ceph_have_pending_cap_snap(ci)) { /* there is no point in queuing multiple "pending" cap_snaps, as no new writes are allowed to start when pending, so any @@ -456,13 +475,22 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci) cap_snap. lucky us. */ dout("queue_cap_snap %p already pending\n", inode); kfree(capsnap); - } else if (ci->i_wrbuffer_ref_head || (used & CEPH_CAP_FILE_WR) || - (dirty & (CEPH_CAP_AUTH_EXCL|CEPH_CAP_XATTR_EXCL| - CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR))) { + } else if (ci->i_snap_realm->cached_context == empty_snapc) { + dout("queue_cap_snap %p empty snapc\n", inode); + kfree(capsnap); + } else if (dirty & (CEPH_CAP_AUTH_EXCL|CEPH_CAP_XATTR_EXCL| + CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR)) { struct ceph_snap_context *snapc = ci->i_head_snapc; - dout("queue_cap_snap %p cap_snap %p queuing under %p\n", inode, - capsnap, snapc); + /* + * if we are a sync write, we may need to go to the snaprealm + * to get the current snapc. + */ + if (!snapc) + snapc = ci->i_snap_realm->cached_context; + + dout("queue_cap_snap %p cap_snap %p queuing under %p %s\n", + inode, capsnap, snapc, ceph_cap_string(dirty)); ihold(inode); atomic_set(&capsnap->nref, 1); @@ -513,7 +541,7 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci) kfree(capsnap); } - spin_unlock(&inode->i_lock); + spin_unlock(&ci->i_ceph_lock); } /* @@ -522,7 +550,7 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci) * * If capsnap can now be flushed, add to snap_flush list, and return 1. * - * Caller must hold i_lock. + * Caller must hold i_ceph_lock. */ int __ceph_finish_cap_snap(struct ceph_inode_info *ci, struct ceph_cap_snap *capsnap) @@ -724,9 +752,9 @@ static void flush_snaps(struct ceph_mds_client *mdsc) inode = &ci->vfs_inode; ihold(inode); spin_unlock(&mdsc->snap_flush_lock); - spin_lock(&inode->i_lock); + spin_lock(&ci->i_ceph_lock); __ceph_flush_snaps(ci, &session, 0); - spin_unlock(&inode->i_lock); + spin_unlock(&ci->i_ceph_lock); iput(inode); spin_lock(&mdsc->snap_flush_lock); } @@ -832,7 +860,7 @@ void ceph_handle_snap(struct ceph_mds_client *mdsc, continue; ci = ceph_inode(inode); - spin_lock(&inode->i_lock); + spin_lock(&ci->i_ceph_lock); if (!ci->i_snap_realm) goto skip_inode; /* @@ -861,7 +889,7 @@ void ceph_handle_snap(struct ceph_mds_client *mdsc, oldrealm = ci->i_snap_realm; ci->i_snap_realm = realm; spin_unlock(&realm->inodes_with_caps_lock); - spin_unlock(&inode->i_lock); + spin_unlock(&ci->i_ceph_lock); ceph_get_snap_realm(mdsc, realm); ceph_put_snap_realm(mdsc, oldrealm); @@ -870,7 +898,7 @@ void ceph_handle_snap(struct ceph_mds_client *mdsc, continue; skip_inode: - spin_unlock(&inode->i_lock); + spin_unlock(&ci->i_ceph_lock); iput(inode); } @@ -912,5 +940,17 @@ out: return; } +int __init ceph_snap_init(void) +{ + empty_snapc = kzalloc(sizeof(struct ceph_snap_context), GFP_NOFS); + if (!empty_snapc) + return -ENOMEM; + atomic_set(&empty_snapc->nref, 1); + empty_snapc->seq = 1; + return 0; +} - +void ceph_snap_exit(void) +{ + ceph_put_snap_context(empty_snapc); +} |