From 15f519e9f883b316d86e2bb6b767a023aafd9d83 Mon Sep 17 00:00:00 2001 From: Alex Markuze Date: Tue, 12 Aug 2025 09:57:38 +0000 Subject: [PATCH 1/5] ceph: fix race condition validating r_parent before applying state Add validation to ensure the cached parent directory inode matches the directory info in MDS replies. This prevents client-side race conditions where concurrent operations (e.g. rename) cause r_parent to become stale between request initiation and reply processing, which could lead to applying state changes to incorrect directory inodes. [ idryomov: folded a kerneldoc fixup and a follow-up fix from Alex to move CEPH_CAP_PIN reference when r_parent is updated: When the parent directory lock is not held, req->r_parent can become stale and is updated to point to the correct inode. However, the associated CEPH_CAP_PIN reference was not being adjusted. The CEPH_CAP_PIN is a reference on an inode that is tracked for accounting purposes. Moving this pin is important to keep the accounting balanced. When the pin was not moved from the old parent to the new one, it created two problems: The reference on the old, stale parent was never released, causing a reference leak. A reference for the new parent was never acquired, creating the risk of a reference underflow later in ceph_mdsc_release_request(). This patch corrects the logic by releasing the pin from the old parent and acquiring it for the new parent when r_parent is switched. This ensures reference accounting stays balanced. ] Cc: stable@vger.kernel.org Signed-off-by: Alex Markuze Reviewed-by: Viacheslav Dubeyko Signed-off-by: Ilya Dryomov --- fs/ceph/debugfs.c | 14 ++-- fs/ceph/dir.c | 17 ++--- fs/ceph/file.c | 24 +++--- fs/ceph/inode.c | 7 +- fs/ceph/mds_client.c | 172 ++++++++++++++++++++++++++----------------- fs/ceph/mds_client.h | 18 ++++- 6 files changed, 145 insertions(+), 107 deletions(-) diff --git a/fs/ceph/debugfs.c b/fs/ceph/debugfs.c index fdd404fc8112..f3fe786b4143 100644 --- a/fs/ceph/debugfs.c +++ b/fs/ceph/debugfs.c @@ -55,8 +55,6 @@ static int mdsc_show(struct seq_file *s, void *p) struct ceph_mds_client *mdsc = fsc->mdsc; struct ceph_mds_request *req; struct rb_node *rp; - int pathlen = 0; - u64 pathbase; char *path; mutex_lock(&mdsc->mutex); @@ -81,8 +79,8 @@ static int mdsc_show(struct seq_file *s, void *p) if (req->r_inode) { seq_printf(s, " #%llx", ceph_ino(req->r_inode)); } else if (req->r_dentry) { - path = ceph_mdsc_build_path(mdsc, req->r_dentry, &pathlen, - &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, req->r_dentry, &path_info, 0); if (IS_ERR(path)) path = NULL; spin_lock(&req->r_dentry->d_lock); @@ -91,7 +89,7 @@ static int mdsc_show(struct seq_file *s, void *p) req->r_dentry, path ? path : ""); spin_unlock(&req->r_dentry->d_lock); - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); } else if (req->r_path1) { seq_printf(s, " #%llx/%s", req->r_ino1.ino, req->r_path1); @@ -100,8 +98,8 @@ static int mdsc_show(struct seq_file *s, void *p) } if (req->r_old_dentry) { - path = ceph_mdsc_build_path(mdsc, req->r_old_dentry, &pathlen, - &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, req->r_old_dentry, &path_info, 0); if (IS_ERR(path)) path = NULL; spin_lock(&req->r_old_dentry->d_lock); @@ -111,7 +109,7 @@ static int mdsc_show(struct seq_file *s, void *p) req->r_old_dentry, path ? path : ""); spin_unlock(&req->r_old_dentry->d_lock); - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); } else if (req->r_path2 && req->r_op != CEPH_MDS_OP_SYMLINK) { if (req->r_ino2.ino) seq_printf(s, " #%llx/%s", req->r_ino2.ino, diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c index 8478e7e75df6..32973c62c1a2 100644 --- a/fs/ceph/dir.c +++ b/fs/ceph/dir.c @@ -1271,10 +1271,8 @@ static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc, /* If op failed, mark everyone involved for errors */ if (result) { - int pathlen = 0; - u64 base = 0; - char *path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, - &base, 0); + struct ceph_path_info path_info = {0}; + char *path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0); /* mark error on parent + clear complete */ mapping_set_error(req->r_parent->i_mapping, result); @@ -1288,8 +1286,8 @@ static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc, mapping_set_error(req->r_old_inode->i_mapping, result); pr_warn_client(cl, "failure path=(%llx)%s result=%d!\n", - base, IS_ERR(path) ? "<>" : path, result); - ceph_mdsc_free_path(path, pathlen); + path_info.vino.ino, IS_ERR(path) ? "<>" : path, result); + ceph_mdsc_free_path_info(&path_info); } out: iput(req->r_old_inode); @@ -1347,8 +1345,6 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry) int err = -EROFS; int op; char *path; - int pathlen; - u64 pathbase; if (ceph_snap(dir) == CEPH_SNAPDIR) { /* rmdir .snap/foo is RMSNAP */ @@ -1367,14 +1363,15 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry) if (!dn) { try_async = false; } else { - path = ceph_mdsc_build_path(mdsc, dn, &pathlen, &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, dn, &path_info, 0); if (IS_ERR(path)) { try_async = false; err = 0; } else { err = ceph_mds_check_access(mdsc, path, MAY_WRITE); } - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); dput(dn); /* For none EACCES cases will let the MDS do the mds auth check */ diff --git a/fs/ceph/file.c b/fs/ceph/file.c index c02f100f8552..978acd3d4b32 100644 --- a/fs/ceph/file.c +++ b/fs/ceph/file.c @@ -368,8 +368,6 @@ int ceph_open(struct inode *inode, struct file *file) int flags, fmode, wanted; struct dentry *dentry; char *path; - int pathlen; - u64 pathbase; bool do_sync = false; int mask = MAY_READ; @@ -399,14 +397,15 @@ int ceph_open(struct inode *inode, struct file *file) if (!dentry) { do_sync = true; } else { - path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0); if (IS_ERR(path)) { do_sync = true; err = 0; } else { err = ceph_mds_check_access(mdsc, path, mask); } - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); dput(dentry); /* For none EACCES cases will let the MDS do the mds auth check */ @@ -614,15 +613,13 @@ static void ceph_async_create_cb(struct ceph_mds_client *mdsc, mapping_set_error(req->r_parent->i_mapping, result); if (result) { - int pathlen = 0; - u64 base = 0; - char *path = ceph_mdsc_build_path(mdsc, req->r_dentry, &pathlen, - &base, 0); + struct ceph_path_info path_info = {0}; + char *path = ceph_mdsc_build_path(mdsc, req->r_dentry, &path_info, 0); pr_warn_client(cl, "async create failure path=(%llx)%s result=%d!\n", - base, IS_ERR(path) ? "<>" : path, result); - ceph_mdsc_free_path(path, pathlen); + path_info.vino.ino, IS_ERR(path) ? "<>" : path, result); + ceph_mdsc_free_path_info(&path_info); ceph_dir_clear_complete(req->r_parent); if (!d_unhashed(dentry)) @@ -791,8 +788,6 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry, int mask; int err; char *path; - int pathlen; - u64 pathbase; doutc(cl, "%p %llx.%llx dentry %p '%pd' %s flags %d mode 0%o\n", dir, ceph_vinop(dir), dentry, dentry, @@ -814,7 +809,8 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry, if (!dn) { try_async = false; } else { - path = ceph_mdsc_build_path(mdsc, dn, &pathlen, &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, dn, &path_info, 0); if (IS_ERR(path)) { try_async = false; err = 0; @@ -826,7 +822,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry, mask |= MAY_WRITE; err = ceph_mds_check_access(mdsc, path, mask); } - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); dput(dn); /* For none EACCES cases will let the MDS do the mds auth check */ diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c index fc543075b827..8ac89ce6435c 100644 --- a/fs/ceph/inode.c +++ b/fs/ceph/inode.c @@ -2487,22 +2487,21 @@ int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode, int truncate_retry = 20; /* The RMW will take around 50ms */ struct dentry *dentry; char *path; - int pathlen; - u64 pathbase; bool do_sync = false; dentry = d_find_alias(inode); if (!dentry) { do_sync = true; } else { - path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, 0); + struct ceph_path_info path_info; + path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0); if (IS_ERR(path)) { do_sync = true; err = 0; } else { err = ceph_mds_check_access(mdsc, path, MAY_WRITE); } - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); dput(dentry); /* For none EACCES cases will let the MDS do the mds auth check */ diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 0f497c39ff82..3bc72b47fe4d 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -2681,8 +2681,7 @@ static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen) * ceph_mdsc_build_path - build a path string to a given dentry * @mdsc: mds client * @dentry: dentry to which path should be built - * @plen: returned length of string - * @pbase: returned base inode number + * @path_info: output path, length, base ino+snap, and freepath ownership flag * @for_wire: is this path going to be sent to the MDS? * * Build a string that represents the path to the dentry. This is mostly called @@ -2700,7 +2699,7 @@ static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen) * foo/.snap/bar -> foo//bar */ char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc, struct dentry *dentry, - int *plen, u64 *pbase, int for_wire) + struct ceph_path_info *path_info, int for_wire) { struct ceph_client *cl = mdsc->fsc->client; struct dentry *cur; @@ -2810,16 +2809,28 @@ retry: return ERR_PTR(-ENAMETOOLONG); } - *pbase = base; - *plen = PATH_MAX - 1 - pos; + /* Initialize the output structure */ + memset(path_info, 0, sizeof(*path_info)); + + path_info->vino.ino = base; + path_info->pathlen = PATH_MAX - 1 - pos; + path_info->path = path + pos; + path_info->freepath = true; + + /* Set snap from dentry if available */ + if (d_inode(dentry)) + path_info->vino.snap = ceph_snap(d_inode(dentry)); + else + path_info->vino.snap = CEPH_NOSNAP; + doutc(cl, "on %p %d built %llx '%.*s'\n", dentry, d_count(dentry), - base, *plen, path + pos); + base, PATH_MAX - 1 - pos, path + pos); return path + pos; } static int build_dentry_path(struct ceph_mds_client *mdsc, struct dentry *dentry, - struct inode *dir, const char **ppath, int *ppathlen, - u64 *pino, bool *pfreepath, bool parent_locked) + struct inode *dir, struct ceph_path_info *path_info, + bool parent_locked) { char *path; @@ -2828,41 +2839,47 @@ static int build_dentry_path(struct ceph_mds_client *mdsc, struct dentry *dentry dir = d_inode_rcu(dentry->d_parent); if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP && !IS_ENCRYPTED(dir)) { - *pino = ceph_ino(dir); + path_info->vino.ino = ceph_ino(dir); + path_info->vino.snap = ceph_snap(dir); rcu_read_unlock(); - *ppath = dentry->d_name.name; - *ppathlen = dentry->d_name.len; + path_info->path = dentry->d_name.name; + path_info->pathlen = dentry->d_name.len; + path_info->freepath = false; return 0; } rcu_read_unlock(); - path = ceph_mdsc_build_path(mdsc, dentry, ppathlen, pino, 1); + path = ceph_mdsc_build_path(mdsc, dentry, path_info, 1); if (IS_ERR(path)) return PTR_ERR(path); - *ppath = path; - *pfreepath = true; + /* + * ceph_mdsc_build_path already fills path_info, including snap handling. + */ return 0; } -static int build_inode_path(struct inode *inode, - const char **ppath, int *ppathlen, u64 *pino, - bool *pfreepath) +static int build_inode_path(struct inode *inode, struct ceph_path_info *path_info) { struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); struct dentry *dentry; char *path; if (ceph_snap(inode) == CEPH_NOSNAP) { - *pino = ceph_ino(inode); - *ppathlen = 0; + path_info->vino.ino = ceph_ino(inode); + path_info->vino.snap = ceph_snap(inode); + path_info->pathlen = 0; + path_info->freepath = false; return 0; } dentry = d_find_alias(inode); - path = ceph_mdsc_build_path(mdsc, dentry, ppathlen, pino, 1); + path = ceph_mdsc_build_path(mdsc, dentry, path_info, 1); dput(dentry); if (IS_ERR(path)) return PTR_ERR(path); - *ppath = path; - *pfreepath = true; + /* + * ceph_mdsc_build_path already fills path_info, including snap from dentry. + * Override with inode's snap since that's what this function is for. + */ + path_info->vino.snap = ceph_snap(inode); return 0; } @@ -2872,26 +2889,32 @@ static int build_inode_path(struct inode *inode, */ static int set_request_path_attr(struct ceph_mds_client *mdsc, struct inode *rinode, struct dentry *rdentry, struct inode *rdiri, - const char *rpath, u64 rino, const char **ppath, - int *pathlen, u64 *ino, bool *freepath, + const char *rpath, u64 rino, + struct ceph_path_info *path_info, bool parent_locked) { struct ceph_client *cl = mdsc->fsc->client; int r = 0; + /* Initialize the output structure */ + memset(path_info, 0, sizeof(*path_info)); + if (rinode) { - r = build_inode_path(rinode, ppath, pathlen, ino, freepath); + r = build_inode_path(rinode, path_info); doutc(cl, " inode %p %llx.%llx\n", rinode, ceph_ino(rinode), ceph_snap(rinode)); } else if (rdentry) { - r = build_dentry_path(mdsc, rdentry, rdiri, ppath, pathlen, ino, - freepath, parent_locked); - doutc(cl, " dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen, *ppath); + r = build_dentry_path(mdsc, rdentry, rdiri, path_info, parent_locked); + doutc(cl, " dentry %p %llx/%.*s\n", rdentry, path_info->vino.ino, + path_info->pathlen, path_info->path); } else if (rpath || rino) { - *ino = rino; - *ppath = rpath; - *pathlen = rpath ? strlen(rpath) : 0; - doutc(cl, " path %.*s\n", *pathlen, rpath); + path_info->vino.ino = rino; + path_info->vino.snap = CEPH_NOSNAP; + path_info->path = rpath; + path_info->pathlen = rpath ? strlen(rpath) : 0; + path_info->freepath = false; + + doutc(cl, " path %.*s\n", path_info->pathlen, rpath); } return r; @@ -2968,11 +2991,8 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, struct ceph_client *cl = mdsc->fsc->client; struct ceph_msg *msg; struct ceph_mds_request_head_legacy *lhead; - const char *path1 = NULL; - const char *path2 = NULL; - u64 ino1 = 0, ino2 = 0; - int pathlen1 = 0, pathlen2 = 0; - bool freepath1 = false, freepath2 = false; + struct ceph_path_info path_info1 = {0}; + struct ceph_path_info path_info2 = {0}; struct dentry *old_dentry = NULL; int len; u16 releases; @@ -2982,25 +3002,49 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, u16 request_head_version = mds_supported_head_version(session); kuid_t caller_fsuid = req->r_cred->fsuid; kgid_t caller_fsgid = req->r_cred->fsgid; + bool parent_locked = test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry, - req->r_parent, req->r_path1, req->r_ino1.ino, - &path1, &pathlen1, &ino1, &freepath1, - test_bit(CEPH_MDS_R_PARENT_LOCKED, - &req->r_req_flags)); + req->r_parent, req->r_path1, req->r_ino1.ino, + &path_info1, parent_locked); if (ret < 0) { msg = ERR_PTR(ret); goto out; } + /* + * When the parent directory's i_rwsem is *not* locked, req->r_parent may + * have become stale (e.g. after a concurrent rename) between the time the + * dentry was looked up and now. If we detect that the stored r_parent + * does not match the inode number we just encoded for the request, switch + * to the correct inode so that the MDS receives a valid parent reference. + */ + if (!parent_locked && req->r_parent && path_info1.vino.ino && + ceph_ino(req->r_parent) != path_info1.vino.ino) { + struct inode *old_parent = req->r_parent; + struct inode *correct_dir = ceph_get_inode(mdsc->fsc->sb, path_info1.vino, NULL); + if (!IS_ERR(correct_dir)) { + WARN_ONCE(1, "ceph: r_parent mismatch (had %llx wanted %llx) - updating\n", + ceph_ino(old_parent), path_info1.vino.ino); + /* + * Transfer CEPH_CAP_PIN from the old parent to the new one. + * The pin was taken earlier in ceph_mdsc_submit_request(). + */ + ceph_put_cap_refs(ceph_inode(old_parent), CEPH_CAP_PIN); + iput(old_parent); + req->r_parent = correct_dir; + ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN); + } + } + /* If r_old_dentry is set, then assume that its parent is locked */ if (req->r_old_dentry && !(req->r_old_dentry->d_flags & DCACHE_DISCONNECTED)) old_dentry = req->r_old_dentry; ret = set_request_path_attr(mdsc, NULL, old_dentry, - req->r_old_dentry_dir, - req->r_path2, req->r_ino2.ino, - &path2, &pathlen2, &ino2, &freepath2, true); + req->r_old_dentry_dir, + req->r_path2, req->r_ino2.ino, + &path_info2, true); if (ret < 0) { msg = ERR_PTR(ret); goto out_free1; @@ -3031,7 +3075,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, /* filepaths */ len += 2 * (1 + sizeof(u32) + sizeof(u64)); - len += pathlen1 + pathlen2; + len += path_info1.pathlen + path_info2.pathlen; /* cap releases */ len += sizeof(struct ceph_mds_request_release) * @@ -3039,9 +3083,9 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, !!req->r_old_inode_drop + !!req->r_old_dentry_drop); if (req->r_dentry_drop) - len += pathlen1; + len += path_info1.pathlen; if (req->r_old_dentry_drop) - len += pathlen2; + len += path_info2.pathlen; /* MClientRequest tail */ @@ -3154,8 +3198,8 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, lhead->ino = cpu_to_le64(req->r_deleg_ino); lhead->args = req->r_args; - ceph_encode_filepath(&p, end, ino1, path1); - ceph_encode_filepath(&p, end, ino2, path2); + ceph_encode_filepath(&p, end, path_info1.vino.ino, path_info1.path); + ceph_encode_filepath(&p, end, path_info2.vino.ino, path_info2.path); /* make note of release offset, in case we need to replay */ req->r_request_release_offset = p - msg->front.iov_base; @@ -3218,11 +3262,9 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session, msg->hdr.data_off = cpu_to_le16(0); out_free2: - if (freepath2) - ceph_mdsc_free_path((char *)path2, pathlen2); + ceph_mdsc_free_path_info(&path_info2); out_free1: - if (freepath1) - ceph_mdsc_free_path((char *)path1, pathlen1); + ceph_mdsc_free_path_info(&path_info1); out: return msg; out_err: @@ -4579,24 +4621,20 @@ static int reconnect_caps_cb(struct inode *inode, int mds, void *arg) struct ceph_pagelist *pagelist = recon_state->pagelist; struct dentry *dentry; struct ceph_cap *cap; - char *path; - int pathlen = 0, err; - u64 pathbase; + struct ceph_path_info path_info = {0}; + int err; u64 snap_follows; dentry = d_find_primary(inode); if (dentry) { /* set pathbase to parent dir when msg_version >= 2 */ - path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, + char *path = ceph_mdsc_build_path(mdsc, dentry, &path_info, recon_state->msg_version >= 2); dput(dentry); if (IS_ERR(path)) { err = PTR_ERR(path); goto out_err; } - } else { - path = NULL; - pathbase = 0; } spin_lock(&ci->i_ceph_lock); @@ -4629,7 +4667,7 @@ static int reconnect_caps_cb(struct inode *inode, int mds, void *arg) rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); rec.v2.issued = cpu_to_le32(cap->issued); rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); - rec.v2.pathbase = cpu_to_le64(pathbase); + rec.v2.pathbase = cpu_to_le64(path_info.vino.ino); rec.v2.flock_len = (__force __le32) ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1); } else { @@ -4644,7 +4682,7 @@ static int reconnect_caps_cb(struct inode *inode, int mds, void *arg) ts = inode_get_atime(inode); ceph_encode_timespec64(&rec.v1.atime, &ts); rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); - rec.v1.pathbase = cpu_to_le64(pathbase); + rec.v1.pathbase = cpu_to_le64(path_info.vino.ino); } if (list_empty(&ci->i_cap_snaps)) { @@ -4706,7 +4744,7 @@ encode_again: sizeof(struct ceph_filelock); rec.v2.flock_len = cpu_to_le32(struct_len); - struct_len += sizeof(u32) + pathlen + sizeof(rec.v2); + struct_len += sizeof(u32) + path_info.pathlen + sizeof(rec.v2); if (struct_v >= 2) struct_len += sizeof(u64); /* snap_follows */ @@ -4730,7 +4768,7 @@ encode_again: ceph_pagelist_encode_8(pagelist, 1); ceph_pagelist_encode_32(pagelist, struct_len); } - ceph_pagelist_encode_string(pagelist, path, pathlen); + ceph_pagelist_encode_string(pagelist, (char *)path_info.path, path_info.pathlen); ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2)); ceph_locks_to_pagelist(flocks, pagelist, num_fcntl_locks, num_flock_locks); @@ -4741,17 +4779,17 @@ out_freeflocks: } else { err = ceph_pagelist_reserve(pagelist, sizeof(u64) + sizeof(u32) + - pathlen + sizeof(rec.v1)); + path_info.pathlen + sizeof(rec.v1)); if (err) goto out_err; ceph_pagelist_encode_64(pagelist, ceph_ino(inode)); - ceph_pagelist_encode_string(pagelist, path, pathlen); + ceph_pagelist_encode_string(pagelist, (char *)path_info.path, path_info.pathlen); ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1)); } out_err: - ceph_mdsc_free_path(path, pathlen); + ceph_mdsc_free_path_info(&path_info); if (!err) recon_state->nr_caps++; return err; diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h index 3e2a6fa7c19a..0428a5eaf28c 100644 --- a/fs/ceph/mds_client.h +++ b/fs/ceph/mds_client.h @@ -617,14 +617,24 @@ extern int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath, extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc); -static inline void ceph_mdsc_free_path(char *path, int len) +/* + * Structure to group path-related output parameters for build_*_path functions + */ +struct ceph_path_info { + const char *path; + int pathlen; + struct ceph_vino vino; + bool freepath; +}; + +static inline void ceph_mdsc_free_path_info(const struct ceph_path_info *path_info) { - if (!IS_ERR_OR_NULL(path)) - __putname(path - (PATH_MAX - 1 - len)); + if (path_info && path_info->freepath && !IS_ERR_OR_NULL(path_info->path)) + __putname((char *)path_info->path - (PATH_MAX - 1 - path_info->pathlen)); } extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc, - struct dentry *dentry, int *plen, u64 *base, + struct dentry *dentry, struct ceph_path_info *path_info, int for_wire); extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry); From bec324f33d1ed346394b2eee25bf6dbf3511f727 Mon Sep 17 00:00:00 2001 From: Alex Markuze Date: Tue, 12 Aug 2025 09:57:39 +0000 Subject: [PATCH 2/5] ceph: fix race condition where r_parent becomes stale before sending message When the parent directory's i_rwsem is not locked, req->r_parent may become stale due to concurrent operations (e.g. rename) between dentry lookup and message creation. Validate that r_parent matches the encoded parent inode and update to the correct inode if a mismatch is detected. [ idryomov: folded a follow-up fix from Alex to drop extra reference from ceph_get_reply_dir() in ceph_fill_trace(): ceph_get_reply_dir() may return a different, referenced inode when r_parent is stale and the parent directory lock is not held. ceph_fill_trace() used that inode but failed to drop the reference when it differed from req->r_parent, leaking an inode reference. Keep the directory inode in a local variable and iput() it at function end if it does not match req->r_parent. ] Cc: stable@vger.kernel.org Signed-off-by: Alex Markuze Reviewed-by: Viacheslav Dubeyko Signed-off-by: Ilya Dryomov --- fs/ceph/inode.c | 81 +++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 69 insertions(+), 12 deletions(-) diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c index 8ac89ce6435c..f67025465de0 100644 --- a/fs/ceph/inode.c +++ b/fs/ceph/inode.c @@ -55,6 +55,52 @@ static int ceph_set_ino_cb(struct inode *inode, void *data) return 0; } +/* + * Check if the parent inode matches the vino from directory reply info + */ +static inline bool ceph_vino_matches_parent(struct inode *parent, + struct ceph_vino vino) +{ + return ceph_ino(parent) == vino.ino && ceph_snap(parent) == vino.snap; +} + +/* + * Validate that the directory inode referenced by @req->r_parent matches the + * inode number and snapshot id contained in the reply's directory record. If + * they do not match – which can theoretically happen if the parent dentry was + * moved between the time the request was issued and the reply arrived – fall + * back to looking up the correct inode in the inode cache. + * + * A reference is *always* returned. Callers that receive a different inode + * than the original @parent are responsible for dropping the extra reference + * once the reply has been processed. + */ +static struct inode *ceph_get_reply_dir(struct super_block *sb, + struct inode *parent, + struct ceph_mds_reply_info_parsed *rinfo) +{ + struct ceph_vino vino; + + if (unlikely(!rinfo->diri.in)) + return parent; /* nothing to compare against */ + + /* If we didn't have a cached parent inode to begin with, just bail out. */ + if (!parent) + return NULL; + + vino.ino = le64_to_cpu(rinfo->diri.in->ino); + vino.snap = le64_to_cpu(rinfo->diri.in->snapid); + + if (likely(ceph_vino_matches_parent(parent, vino))) + return parent; /* matches – use the original reference */ + + /* Mismatch – this should be rare. Emit a WARN and obtain the correct inode. */ + WARN_ONCE(1, "ceph: reply dir mismatch (parent valid %llx.%llx reply %llx.%llx)\n", + ceph_ino(parent), ceph_snap(parent), vino.ino, vino.snap); + + return ceph_get_inode(sb, vino, NULL); +} + /** * ceph_new_inode - allocate a new inode in advance of an expected create * @dir: parent directory for new inode @@ -1523,6 +1569,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) struct ceph_vino tvino, dvino; struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb); struct ceph_client *cl = fsc->client; + struct inode *parent_dir = NULL; int err = 0; doutc(cl, "%p is_dentry %d is_target %d\n", req, @@ -1536,10 +1583,17 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) } if (rinfo->head->is_dentry) { - struct inode *dir = req->r_parent; - - if (dir) { - err = ceph_fill_inode(dir, NULL, &rinfo->diri, + /* + * r_parent may be stale, in cases when R_PARENT_LOCKED is not set, + * so we need to get the correct inode + */ + parent_dir = ceph_get_reply_dir(sb, req->r_parent, rinfo); + if (unlikely(IS_ERR(parent_dir))) { + err = PTR_ERR(parent_dir); + goto done; + } + if (parent_dir) { + err = ceph_fill_inode(parent_dir, NULL, &rinfo->diri, rinfo->dirfrag, session, -1, &req->r_caps_reservation); if (err < 0) @@ -1548,14 +1602,14 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) WARN_ON_ONCE(1); } - if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && + if (parent_dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { bool is_nokey = false; struct qstr dname; struct dentry *dn, *parent; struct fscrypt_str oname = FSTR_INIT(NULL, 0); - struct ceph_fname fname = { .dir = dir, + struct ceph_fname fname = { .dir = parent_dir, .name = rinfo->dname, .ctext = rinfo->altname, .name_len = rinfo->dname_len, @@ -1564,10 +1618,10 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) BUG_ON(!rinfo->head->is_target); BUG_ON(req->r_dentry); - parent = d_find_any_alias(dir); + parent = d_find_any_alias(parent_dir); BUG_ON(!parent); - err = ceph_fname_alloc_buffer(dir, &oname); + err = ceph_fname_alloc_buffer(parent_dir, &oname); if (err < 0) { dput(parent); goto done; @@ -1576,7 +1630,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey); if (err < 0) { dput(parent); - ceph_fname_free_buffer(dir, &oname); + ceph_fname_free_buffer(parent_dir, &oname); goto done; } dname.name = oname.name; @@ -1595,7 +1649,7 @@ retry_lookup: dname.len, dname.name, dn); if (!dn) { dput(parent); - ceph_fname_free_buffer(dir, &oname); + ceph_fname_free_buffer(parent_dir, &oname); err = -ENOMEM; goto done; } @@ -1610,12 +1664,12 @@ retry_lookup: ceph_snap(d_inode(dn)) != tvino.snap)) { doutc(cl, " dn %p points to wrong inode %p\n", dn, d_inode(dn)); - ceph_dir_clear_ordered(dir); + ceph_dir_clear_ordered(parent_dir); d_delete(dn); dput(dn); goto retry_lookup; } - ceph_fname_free_buffer(dir, &oname); + ceph_fname_free_buffer(parent_dir, &oname); req->r_dentry = dn; dput(parent); @@ -1794,6 +1848,9 @@ retry_lookup: &dvino, ptvino); } done: + /* Drop extra ref from ceph_get_reply_dir() if it returned a new inode */ + if (unlikely(!IS_ERR_OR_NULL(parent_dir) && parent_dir != req->r_parent)) + iput(parent_dir); doutc(cl, "done err=%d\n", err); return err; } From cce7c15faaac79b532a07ed6ab8332280ad83762 Mon Sep 17 00:00:00 2001 From: Max Kellermann Date: Wed, 27 Aug 2025 20:17:08 +0200 Subject: [PATCH 3/5] ceph: always call ceph_shift_unused_folios_left() The function ceph_process_folio_batch() sets folio_batch entries to NULL, which is an illegal state. Before folio_batch_release() crashes due to this API violation, the function ceph_shift_unused_folios_left() is supposed to remove those NULLs from the array. However, since commit ce80b76dd327 ("ceph: introduce ceph_process_folio_batch() method"), this shifting doesn't happen anymore because the "for" loop got moved to ceph_process_folio_batch(), and now the `i` variable that remains in ceph_writepages_start() doesn't get incremented anymore, making the shifting effectively unreachable much of the time. Later, commit 1551ec61dc55 ("ceph: introduce ceph_submit_write() method") added more preconditions for doing the shift, replacing the `i` check (with something that is still just as broken): - if ceph_process_folio_batch() fails, shifting never happens - if ceph_move_dirty_page_in_page_array() was never called (because ceph_process_folio_batch() has returned early for some of various reasons), shifting never happens - if `processed_in_fbatch` is zero (because ceph_process_folio_batch() has returned early for some of the reasons mentioned above or because ceph_move_dirty_page_in_page_array() has failed), shifting never happens Since those two commits, any problem in ceph_process_folio_batch() could crash the kernel, e.g. this way: BUG: kernel NULL pointer dereference, address: 0000000000000034 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: Oops: 0002 [#1] SMP NOPTI CPU: 172 UID: 0 PID: 2342707 Comm: kworker/u778:8 Not tainted 6.15.10-cm4all1-es #714 NONE Hardware name: Dell Inc. PowerEdge R7615/0G9DHV, BIOS 1.6.10 12/08/2023 Workqueue: writeback wb_workfn (flush-ceph-1) RIP: 0010:folios_put_refs+0x85/0x140 Code: 83 c5 01 39 e8 7e 76 48 63 c5 49 8b 5c c4 08 b8 01 00 00 00 4d 85 ed 74 05 41 8b 44 ad 00 48 8b 15 b0 > RSP: 0018:ffffb880af8db778 EFLAGS: 00010207 RAX: 0000000000000001 RBX: 0000000000000000 RCX: 0000000000000003 RDX: ffffe377cc3b0000 RSI: 0000000000000000 RDI: ffffb880af8db8c0 RBP: 0000000000000000 R08: 000000000000007d R09: 000000000102b86f R10: 0000000000000001 R11: 00000000000000ac R12: ffffb880af8db8c0 R13: 0000000000000000 R14: 0000000000000000 R15: ffff9bd262c97000 FS: 0000000000000000(0000) GS:ffff9c8efc303000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000000034 CR3: 0000000160958004 CR4: 0000000000770ef0 PKRU: 55555554 Call Trace: ceph_writepages_start+0xeb9/0x1410 The crash can be reproduced easily by changing the ceph_check_page_before_write() return value to `-E2BIG`. (Interestingly, the crash happens only if `huge_zero_folio` has already been allocated; without `huge_zero_folio`, is_huge_zero_folio(NULL) returns true and folios_put_refs() skips NULL entries instead of dereferencing them. That makes reproducing the bug somewhat unreliable. See https://lore.kernel.org/20250826231626.218675-1-max.kellermann@ionos.com for a discussion of this detail.) My suggestion is to move the ceph_shift_unused_folios_left() to right after ceph_process_folio_batch() to ensure it always gets called to fix up the illegal folio_batch state. Cc: stable@vger.kernel.org Fixes: ce80b76dd327 ("ceph: introduce ceph_process_folio_batch() method") Link: https://lore.kernel.org/ceph-devel/aK4v548CId5GIKG1@swift.blarg.de/ Signed-off-by: Max Kellermann Reviewed-by: Viacheslav Dubeyko Signed-off-by: Ilya Dryomov --- fs/ceph/addr.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c index 8b202d789e93..8bc66b45dade 100644 --- a/fs/ceph/addr.c +++ b/fs/ceph/addr.c @@ -1687,6 +1687,7 @@ get_more_pages: process_folio_batch: rc = ceph_process_folio_batch(mapping, wbc, &ceph_wbc); + ceph_shift_unused_folios_left(&ceph_wbc.fbatch); if (rc) goto release_folios; @@ -1695,8 +1696,6 @@ process_folio_batch: goto release_folios; if (ceph_wbc.processed_in_fbatch) { - ceph_shift_unused_folios_left(&ceph_wbc.fbatch); - if (folio_batch_count(&ceph_wbc.fbatch) == 0 && ceph_wbc.locked_pages < ceph_wbc.max_pages) { doutc(cl, "reached end fbatch, trying for more\n"); From 249e0a47cdb46bb9eae65511c569044bd8698d7d Mon Sep 17 00:00:00 2001 From: Max Kellermann Date: Thu, 28 Aug 2025 13:15:52 +0200 Subject: [PATCH 4/5] ceph: fix crash after fscrypt_encrypt_pagecache_blocks() error The function move_dirty_folio_in_page_array() was created by commit ce80b76dd327 ("ceph: introduce ceph_process_folio_batch() method") by moving code from ceph_writepages_start() to this function. This new function is supposed to return an error code which is checked by the caller (now ceph_process_folio_batch()), and on error, the caller invokes redirty_page_for_writepage() and then breaks from the loop. However, the refactoring commit has gone wrong, and it by accident, it always returns 0 (= success) because it first NULLs the pointer and then returns PTR_ERR(NULL) which is always 0. This means errors are silently ignored, leaving NULL entries in the page array, which may later crash the kernel. The simple solution is to call PTR_ERR() before clearing the pointer. Cc: stable@vger.kernel.org Fixes: ce80b76dd327 ("ceph: introduce ceph_process_folio_batch() method") Link: https://lore.kernel.org/ceph-devel/aK4v548CId5GIKG1@swift.blarg.de/ Signed-off-by: Max Kellermann Reviewed-by: Ilya Dryomov Signed-off-by: Ilya Dryomov --- fs/ceph/addr.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c index 8bc66b45dade..322ed268f14a 100644 --- a/fs/ceph/addr.c +++ b/fs/ceph/addr.c @@ -1264,7 +1264,9 @@ static inline int move_dirty_folio_in_page_array(struct address_space *mapping, 0, gfp_flags); if (IS_ERR(pages[index])) { - if (PTR_ERR(pages[index]) == -EINVAL) { + int err = PTR_ERR(pages[index]); + + if (err == -EINVAL) { pr_err_client(cl, "inode->i_blkbits=%hhu\n", inode->i_blkbits); } @@ -1273,7 +1275,7 @@ static inline int move_dirty_folio_in_page_array(struct address_space *mapping, BUG_ON(ceph_wbc->locked_pages == 0); pages[index] = NULL; - return PTR_ERR(pages[index]); + return err; } } else { pages[index] = &folio->page; From cdbc9836c7afadad68f374791738f118263c5371 Mon Sep 17 00:00:00 2001 From: Ilya Dryomov Date: Thu, 3 Jul 2025 12:10:50 +0200 Subject: [PATCH 5/5] libceph: fix invalid accesses to ceph_connection_v1_info There is a place where generic code in messenger.c is reading and another place where it is writing to con->v1 union member without checking that the union member is active (i.e. msgr1 is in use). On 64-bit systems, con->v1.auth_retry overlaps with con->v2.out_iter, so such a read is almost guaranteed to return a bogus value instead of 0 when msgr2 is in use. This ends up being fairly benign because the side effect is just the invalidation of the authorizer and successive fetching of new tickets. con->v1.connect_seq overlaps with con->v2.conn_bufs and the fact that it's being written to can cause more serious consequences, but luckily it's not something that happens often. Cc: stable@vger.kernel.org Fixes: cd1a677cad99 ("libceph, ceph: implement msgr2.1 protocol (crc and secure modes)") Signed-off-by: Ilya Dryomov Reviewed-by: Viacheslav Dubeyko --- net/ceph/messenger.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c index d1b5705dc0c6..9f6d860411cb 100644 --- a/net/ceph/messenger.c +++ b/net/ceph/messenger.c @@ -1524,7 +1524,7 @@ static void con_fault_finish(struct ceph_connection *con) * in case we faulted due to authentication, invalidate our * current tickets so that we can get new ones. */ - if (con->v1.auth_retry) { + if (!ceph_msgr2(from_msgr(con->msgr)) && con->v1.auth_retry) { dout("auth_retry %d, invalidating\n", con->v1.auth_retry); if (con->ops->invalidate_authorizer) con->ops->invalidate_authorizer(con); @@ -1714,9 +1714,10 @@ static void clear_standby(struct ceph_connection *con) { /* come back from STANDBY? */ if (con->state == CEPH_CON_S_STANDBY) { - dout("clear_standby %p and ++connect_seq\n", con); + dout("clear_standby %p\n", con); con->state = CEPH_CON_S_PREOPEN; - con->v1.connect_seq++; + if (!ceph_msgr2(from_msgr(con->msgr))) + con->v1.connect_seq++; WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)); WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)); }