13d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 2355da1ebSSage Weil 3355da1ebSSage Weil #include <linux/module.h> 4355da1ebSSage Weil #include <linux/fs.h> 5355da1ebSSage Weil #include <linux/slab.h> 6355da1ebSSage Weil #include <linux/string.h> 7355da1ebSSage Weil #include <linux/uaccess.h> 8355da1ebSSage Weil #include <linux/kernel.h> 9355da1ebSSage Weil #include <linux/namei.h> 10355da1ebSSage Weil #include <linux/writeback.h> 11355da1ebSSage Weil #include <linux/vmalloc.h> 12355da1ebSSage Weil 13355da1ebSSage Weil #include "super.h" 143d14c5d2SYehuda Sadeh #include "mds_client.h" 153d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 16355da1ebSSage Weil 17355da1ebSSage Weil /* 18355da1ebSSage Weil * Ceph inode operations 19355da1ebSSage Weil * 20355da1ebSSage Weil * Implement basic inode helpers (get, alloc) and inode ops (getattr, 21355da1ebSSage Weil * setattr, etc.), xattr helpers, and helpers for assimilating 22355da1ebSSage Weil * metadata returned by the MDS into our cache. 23355da1ebSSage Weil * 24355da1ebSSage Weil * Also define helpers for doing asynchronous writeback, invalidation, 25355da1ebSSage Weil * and truncation for the benefit of those who can't afford to block 26355da1ebSSage Weil * (typically because they are in the message handler path). 27355da1ebSSage Weil */ 28355da1ebSSage Weil 29355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops; 30355da1ebSSage Weil 313c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work); 323c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work); 333c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work); 34355da1ebSSage Weil 35355da1ebSSage Weil /* 36355da1ebSSage Weil * find or create an inode, given the ceph ino number 37355da1ebSSage Weil */ 38ad1fee96SYehuda Sadeh static int ceph_set_ino_cb(struct inode *inode, void *data) 39ad1fee96SYehuda Sadeh { 40ad1fee96SYehuda Sadeh ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 41ad1fee96SYehuda Sadeh inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 42ad1fee96SYehuda Sadeh return 0; 43ad1fee96SYehuda Sadeh } 44ad1fee96SYehuda Sadeh 45355da1ebSSage Weil struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 46355da1ebSSage Weil { 47355da1ebSSage Weil struct inode *inode; 48355da1ebSSage Weil ino_t t = ceph_vino_to_ino(vino); 49355da1ebSSage Weil 50355da1ebSSage Weil inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 51355da1ebSSage Weil if (inode == NULL) 52355da1ebSSage Weil return ERR_PTR(-ENOMEM); 53355da1ebSSage Weil if (inode->i_state & I_NEW) { 54355da1ebSSage Weil dout("get_inode created new inode %p %llx.%llx ino %llx\n", 55355da1ebSSage Weil inode, ceph_vinop(inode), (u64)inode->i_ino); 56355da1ebSSage Weil unlock_new_inode(inode); 57355da1ebSSage Weil } 58355da1ebSSage Weil 59355da1ebSSage Weil dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 60355da1ebSSage Weil vino.snap, inode); 61355da1ebSSage Weil return inode; 62355da1ebSSage Weil } 63355da1ebSSage Weil 64355da1ebSSage Weil /* 65355da1ebSSage Weil * get/constuct snapdir inode for a given directory 66355da1ebSSage Weil */ 67355da1ebSSage Weil struct inode *ceph_get_snapdir(struct inode *parent) 68355da1ebSSage Weil { 69355da1ebSSage Weil struct ceph_vino vino = { 70355da1ebSSage Weil .ino = ceph_ino(parent), 71355da1ebSSage Weil .snap = CEPH_SNAPDIR, 72355da1ebSSage Weil }; 73355da1ebSSage Weil struct inode *inode = ceph_get_inode(parent->i_sb, vino); 74b377ff13SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 75355da1ebSSage Weil 76355da1ebSSage Weil BUG_ON(!S_ISDIR(parent->i_mode)); 77355da1ebSSage Weil if (IS_ERR(inode)) 787e34bc52SJulia Lawall return inode; 79355da1ebSSage Weil inode->i_mode = parent->i_mode; 80355da1ebSSage Weil inode->i_uid = parent->i_uid; 81355da1ebSSage Weil inode->i_gid = parent->i_gid; 82355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 83355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 84b377ff13SSage Weil ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 85b377ff13SSage Weil ci->i_rbytes = 0; 86355da1ebSSage Weil return inode; 87355da1ebSSage Weil } 88355da1ebSSage Weil 89355da1ebSSage Weil const struct inode_operations ceph_file_iops = { 90355da1ebSSage Weil .permission = ceph_permission, 91355da1ebSSage Weil .setattr = ceph_setattr, 92355da1ebSSage Weil .getattr = ceph_getattr, 93355da1ebSSage Weil .setxattr = ceph_setxattr, 94355da1ebSSage Weil .getxattr = ceph_getxattr, 95355da1ebSSage Weil .listxattr = ceph_listxattr, 96355da1ebSSage Weil .removexattr = ceph_removexattr, 97355da1ebSSage Weil }; 98355da1ebSSage Weil 99355da1ebSSage Weil 100355da1ebSSage Weil /* 101355da1ebSSage Weil * We use a 'frag tree' to keep track of the MDS's directory fragments 102355da1ebSSage Weil * for a given inode (usually there is just a single fragment). We 103355da1ebSSage Weil * need to know when a child frag is delegated to a new MDS, or when 104355da1ebSSage Weil * it is flagged as replicated, so we can direct our requests 105355da1ebSSage Weil * accordingly. 106355da1ebSSage Weil */ 107355da1ebSSage Weil 108355da1ebSSage Weil /* 109355da1ebSSage Weil * find/create a frag in the tree 110355da1ebSSage Weil */ 111355da1ebSSage Weil static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 112355da1ebSSage Weil u32 f) 113355da1ebSSage Weil { 114355da1ebSSage Weil struct rb_node **p; 115355da1ebSSage Weil struct rb_node *parent = NULL; 116355da1ebSSage Weil struct ceph_inode_frag *frag; 117355da1ebSSage Weil int c; 118355da1ebSSage Weil 119355da1ebSSage Weil p = &ci->i_fragtree.rb_node; 120355da1ebSSage Weil while (*p) { 121355da1ebSSage Weil parent = *p; 122355da1ebSSage Weil frag = rb_entry(parent, struct ceph_inode_frag, node); 123355da1ebSSage Weil c = ceph_frag_compare(f, frag->frag); 124355da1ebSSage Weil if (c < 0) 125355da1ebSSage Weil p = &(*p)->rb_left; 126355da1ebSSage Weil else if (c > 0) 127355da1ebSSage Weil p = &(*p)->rb_right; 128355da1ebSSage Weil else 129355da1ebSSage Weil return frag; 130355da1ebSSage Weil } 131355da1ebSSage Weil 132355da1ebSSage Weil frag = kmalloc(sizeof(*frag), GFP_NOFS); 133355da1ebSSage Weil if (!frag) { 134355da1ebSSage Weil pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 135355da1ebSSage Weil "frag %x\n", &ci->vfs_inode, 136355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 137355da1ebSSage Weil return ERR_PTR(-ENOMEM); 138355da1ebSSage Weil } 139355da1ebSSage Weil frag->frag = f; 140355da1ebSSage Weil frag->split_by = 0; 141355da1ebSSage Weil frag->mds = -1; 142355da1ebSSage Weil frag->ndist = 0; 143355da1ebSSage Weil 144355da1ebSSage Weil rb_link_node(&frag->node, parent, p); 145355da1ebSSage Weil rb_insert_color(&frag->node, &ci->i_fragtree); 146355da1ebSSage Weil 147355da1ebSSage Weil dout("get_or_create_frag added %llx.%llx frag %x\n", 148355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 149355da1ebSSage Weil return frag; 150355da1ebSSage Weil } 151355da1ebSSage Weil 152355da1ebSSage Weil /* 153355da1ebSSage Weil * find a specific frag @f 154355da1ebSSage Weil */ 155355da1ebSSage Weil struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 156355da1ebSSage Weil { 157355da1ebSSage Weil struct rb_node *n = ci->i_fragtree.rb_node; 158355da1ebSSage Weil 159355da1ebSSage Weil while (n) { 160355da1ebSSage Weil struct ceph_inode_frag *frag = 161355da1ebSSage Weil rb_entry(n, struct ceph_inode_frag, node); 162355da1ebSSage Weil int c = ceph_frag_compare(f, frag->frag); 163355da1ebSSage Weil if (c < 0) 164355da1ebSSage Weil n = n->rb_left; 165355da1ebSSage Weil else if (c > 0) 166355da1ebSSage Weil n = n->rb_right; 167355da1ebSSage Weil else 168355da1ebSSage Weil return frag; 169355da1ebSSage Weil } 170355da1ebSSage Weil return NULL; 171355da1ebSSage Weil } 172355da1ebSSage Weil 173355da1ebSSage Weil /* 174355da1ebSSage Weil * Choose frag containing the given value @v. If @pfrag is 175355da1ebSSage Weil * specified, copy the frag delegation info to the caller if 176355da1ebSSage Weil * it is present. 177355da1ebSSage Weil */ 178355da1ebSSage Weil u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 179355da1ebSSage Weil struct ceph_inode_frag *pfrag, 180355da1ebSSage Weil int *found) 181355da1ebSSage Weil { 182355da1ebSSage Weil u32 t = ceph_frag_make(0, 0); 183355da1ebSSage Weil struct ceph_inode_frag *frag; 184355da1ebSSage Weil unsigned nway, i; 185355da1ebSSage Weil u32 n; 186355da1ebSSage Weil 187355da1ebSSage Weil if (found) 188355da1ebSSage Weil *found = 0; 189355da1ebSSage Weil 190355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 191355da1ebSSage Weil while (1) { 192355da1ebSSage Weil WARN_ON(!ceph_frag_contains_value(t, v)); 193355da1ebSSage Weil frag = __ceph_find_frag(ci, t); 194355da1ebSSage Weil if (!frag) 195355da1ebSSage Weil break; /* t is a leaf */ 196355da1ebSSage Weil if (frag->split_by == 0) { 197355da1ebSSage Weil if (pfrag) 198355da1ebSSage Weil memcpy(pfrag, frag, sizeof(*pfrag)); 199355da1ebSSage Weil if (found) 200355da1ebSSage Weil *found = 1; 201355da1ebSSage Weil break; 202355da1ebSSage Weil } 203355da1ebSSage Weil 204355da1ebSSage Weil /* choose child */ 205355da1ebSSage Weil nway = 1 << frag->split_by; 206355da1ebSSage Weil dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 207355da1ebSSage Weil frag->split_by, nway); 208355da1ebSSage Weil for (i = 0; i < nway; i++) { 209355da1ebSSage Weil n = ceph_frag_make_child(t, frag->split_by, i); 210355da1ebSSage Weil if (ceph_frag_contains_value(n, v)) { 211355da1ebSSage Weil t = n; 212355da1ebSSage Weil break; 213355da1ebSSage Weil } 214355da1ebSSage Weil } 215355da1ebSSage Weil BUG_ON(i == nway); 216355da1ebSSage Weil } 217355da1ebSSage Weil dout("choose_frag(%x) = %x\n", v, t); 218355da1ebSSage Weil 219355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 220355da1ebSSage Weil return t; 221355da1ebSSage Weil } 222355da1ebSSage Weil 223355da1ebSSage Weil /* 224355da1ebSSage Weil * Process dirfrag (delegation) info from the mds. Include leaf 225355da1ebSSage Weil * fragment in tree ONLY if ndist > 0. Otherwise, only 226355da1ebSSage Weil * branches/splits are included in i_fragtree) 227355da1ebSSage Weil */ 228355da1ebSSage Weil static int ceph_fill_dirfrag(struct inode *inode, 229355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo) 230355da1ebSSage Weil { 231355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 232355da1ebSSage Weil struct ceph_inode_frag *frag; 233355da1ebSSage Weil u32 id = le32_to_cpu(dirinfo->frag); 234355da1ebSSage Weil int mds = le32_to_cpu(dirinfo->auth); 235355da1ebSSage Weil int ndist = le32_to_cpu(dirinfo->ndist); 236355da1ebSSage Weil int i; 237355da1ebSSage Weil int err = 0; 238355da1ebSSage Weil 239355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 240355da1ebSSage Weil if (ndist == 0) { 241355da1ebSSage Weil /* no delegation info needed. */ 242355da1ebSSage Weil frag = __ceph_find_frag(ci, id); 243355da1ebSSage Weil if (!frag) 244355da1ebSSage Weil goto out; 245355da1ebSSage Weil if (frag->split_by == 0) { 246355da1ebSSage Weil /* tree leaf, remove */ 247355da1ebSSage Weil dout("fill_dirfrag removed %llx.%llx frag %x" 248355da1ebSSage Weil " (no ref)\n", ceph_vinop(inode), id); 249355da1ebSSage Weil rb_erase(&frag->node, &ci->i_fragtree); 250355da1ebSSage Weil kfree(frag); 251355da1ebSSage Weil } else { 252355da1ebSSage Weil /* tree branch, keep and clear */ 253355da1ebSSage Weil dout("fill_dirfrag cleared %llx.%llx frag %x" 254355da1ebSSage Weil " referral\n", ceph_vinop(inode), id); 255355da1ebSSage Weil frag->mds = -1; 256355da1ebSSage Weil frag->ndist = 0; 257355da1ebSSage Weil } 258355da1ebSSage Weil goto out; 259355da1ebSSage Weil } 260355da1ebSSage Weil 261355da1ebSSage Weil 262355da1ebSSage Weil /* find/add this frag to store mds delegation info */ 263355da1ebSSage Weil frag = __get_or_create_frag(ci, id); 264355da1ebSSage Weil if (IS_ERR(frag)) { 265355da1ebSSage Weil /* this is not the end of the world; we can continue 266355da1ebSSage Weil with bad/inaccurate delegation info */ 267355da1ebSSage Weil pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 268355da1ebSSage Weil ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 269355da1ebSSage Weil err = -ENOMEM; 270355da1ebSSage Weil goto out; 271355da1ebSSage Weil } 272355da1ebSSage Weil 273355da1ebSSage Weil frag->mds = mds; 274355da1ebSSage Weil frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 275355da1ebSSage Weil for (i = 0; i < frag->ndist; i++) 276355da1ebSSage Weil frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 277355da1ebSSage Weil dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 278355da1ebSSage Weil ceph_vinop(inode), frag->frag, frag->ndist); 279355da1ebSSage Weil 280355da1ebSSage Weil out: 281355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 282355da1ebSSage Weil return err; 283355da1ebSSage Weil } 284355da1ebSSage Weil 285355da1ebSSage Weil 286355da1ebSSage Weil /* 287355da1ebSSage Weil * initialize a newly allocated inode. 288355da1ebSSage Weil */ 289355da1ebSSage Weil struct inode *ceph_alloc_inode(struct super_block *sb) 290355da1ebSSage Weil { 291355da1ebSSage Weil struct ceph_inode_info *ci; 292355da1ebSSage Weil int i; 293355da1ebSSage Weil 294355da1ebSSage Weil ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 295355da1ebSSage Weil if (!ci) 296355da1ebSSage Weil return NULL; 297355da1ebSSage Weil 298355da1ebSSage Weil dout("alloc_inode %p\n", &ci->vfs_inode); 299355da1ebSSage Weil 300be655596SSage Weil spin_lock_init(&ci->i_ceph_lock); 301be655596SSage Weil 302355da1ebSSage Weil ci->i_version = 0; 303355da1ebSSage Weil ci->i_time_warp_seq = 0; 304355da1ebSSage Weil ci->i_ceph_flags = 0; 3052f276c51SYan, Zheng atomic_set(&ci->i_release_count, 1); 3062f276c51SYan, Zheng atomic_set(&ci->i_complete_count, 0); 307355da1ebSSage Weil ci->i_symlink = NULL; 308355da1ebSSage Weil 3096c0f3af7SSage Weil memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 3106c0f3af7SSage Weil 311355da1ebSSage Weil ci->i_fragtree = RB_ROOT; 312355da1ebSSage Weil mutex_init(&ci->i_fragtree_mutex); 313355da1ebSSage Weil 314355da1ebSSage Weil ci->i_xattrs.blob = NULL; 315355da1ebSSage Weil ci->i_xattrs.prealloc_blob = NULL; 316355da1ebSSage Weil ci->i_xattrs.dirty = false; 317355da1ebSSage Weil ci->i_xattrs.index = RB_ROOT; 318355da1ebSSage Weil ci->i_xattrs.count = 0; 319355da1ebSSage Weil ci->i_xattrs.names_size = 0; 320355da1ebSSage Weil ci->i_xattrs.vals_size = 0; 321355da1ebSSage Weil ci->i_xattrs.version = 0; 322355da1ebSSage Weil ci->i_xattrs.index_version = 0; 323355da1ebSSage Weil 324355da1ebSSage Weil ci->i_caps = RB_ROOT; 325355da1ebSSage Weil ci->i_auth_cap = NULL; 326355da1ebSSage Weil ci->i_dirty_caps = 0; 327355da1ebSSage Weil ci->i_flushing_caps = 0; 328355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_dirty_item); 329355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_flushing_item); 330355da1ebSSage Weil ci->i_cap_flush_seq = 0; 331355da1ebSSage Weil ci->i_cap_flush_last_tid = 0; 332355da1ebSSage Weil memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 333355da1ebSSage Weil init_waitqueue_head(&ci->i_cap_wq); 334355da1ebSSage Weil ci->i_hold_caps_min = 0; 335355da1ebSSage Weil ci->i_hold_caps_max = 0; 336355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_delay_list); 337355da1ebSSage Weil ci->i_cap_exporting_mds = 0; 338355da1ebSSage Weil ci->i_cap_exporting_mseq = 0; 339355da1ebSSage Weil ci->i_cap_exporting_issued = 0; 340355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_snaps); 341355da1ebSSage Weil ci->i_head_snapc = NULL; 342355da1ebSSage Weil ci->i_snap_caps = 0; 343355da1ebSSage Weil 344355da1ebSSage Weil for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 345355da1ebSSage Weil ci->i_nr_by_mode[i] = 0; 346355da1ebSSage Weil 347355da1ebSSage Weil ci->i_truncate_seq = 0; 348355da1ebSSage Weil ci->i_truncate_size = 0; 349355da1ebSSage Weil ci->i_truncate_pending = 0; 350355da1ebSSage Weil 351355da1ebSSage Weil ci->i_max_size = 0; 352355da1ebSSage Weil ci->i_reported_size = 0; 353355da1ebSSage Weil ci->i_wanted_max_size = 0; 354355da1ebSSage Weil ci->i_requested_max_size = 0; 355355da1ebSSage Weil 356355da1ebSSage Weil ci->i_pin_ref = 0; 357355da1ebSSage Weil ci->i_rd_ref = 0; 358355da1ebSSage Weil ci->i_rdcache_ref = 0; 359355da1ebSSage Weil ci->i_wr_ref = 0; 360d3d0720dSHenry C Chang ci->i_wb_ref = 0; 361355da1ebSSage Weil ci->i_wrbuffer_ref = 0; 362355da1ebSSage Weil ci->i_wrbuffer_ref_head = 0; 363355da1ebSSage Weil ci->i_shared_gen = 0; 364355da1ebSSage Weil ci->i_rdcache_gen = 0; 365355da1ebSSage Weil ci->i_rdcache_revoking = 0; 366355da1ebSSage Weil 367355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_writes); 368355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_dirops); 369355da1ebSSage Weil spin_lock_init(&ci->i_unsafe_lock); 370355da1ebSSage Weil 371355da1ebSSage Weil ci->i_snap_realm = NULL; 372355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_realm_item); 373355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_flush_item); 374355da1ebSSage Weil 3753c6f6b79SSage Weil INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 3763c6f6b79SSage Weil INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 377355da1ebSSage Weil 378355da1ebSSage Weil INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 379355da1ebSSage Weil 380355da1ebSSage Weil return &ci->vfs_inode; 381355da1ebSSage Weil } 382355da1ebSSage Weil 383fa0d7e3dSNick Piggin static void ceph_i_callback(struct rcu_head *head) 384fa0d7e3dSNick Piggin { 385fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 386fa0d7e3dSNick Piggin struct ceph_inode_info *ci = ceph_inode(inode); 387fa0d7e3dSNick Piggin 388fa0d7e3dSNick Piggin kmem_cache_free(ceph_inode_cachep, ci); 389fa0d7e3dSNick Piggin } 390fa0d7e3dSNick Piggin 391355da1ebSSage Weil void ceph_destroy_inode(struct inode *inode) 392355da1ebSSage Weil { 393355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 394355da1ebSSage Weil struct ceph_inode_frag *frag; 395355da1ebSSage Weil struct rb_node *n; 396355da1ebSSage Weil 397355da1ebSSage Weil dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 398355da1ebSSage Weil 399355da1ebSSage Weil ceph_queue_caps_release(inode); 400355da1ebSSage Weil 4018b218b8aSSage Weil /* 4028b218b8aSSage Weil * we may still have a snap_realm reference if there are stray 4038b218b8aSSage Weil * caps in i_cap_exporting_issued or i_snap_caps. 4048b218b8aSSage Weil */ 4058b218b8aSSage Weil if (ci->i_snap_realm) { 4068b218b8aSSage Weil struct ceph_mds_client *mdsc = 4073d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 4088b218b8aSSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 4098b218b8aSSage Weil 4108b218b8aSSage Weil dout(" dropping residual ref to snap realm %p\n", realm); 4118b218b8aSSage Weil spin_lock(&realm->inodes_with_caps_lock); 4128b218b8aSSage Weil list_del_init(&ci->i_snap_realm_item); 4138b218b8aSSage Weil spin_unlock(&realm->inodes_with_caps_lock); 4148b218b8aSSage Weil ceph_put_snap_realm(mdsc, realm); 4158b218b8aSSage Weil } 4168b218b8aSSage Weil 417355da1ebSSage Weil kfree(ci->i_symlink); 418355da1ebSSage Weil while ((n = rb_first(&ci->i_fragtree)) != NULL) { 419355da1ebSSage Weil frag = rb_entry(n, struct ceph_inode_frag, node); 420355da1ebSSage Weil rb_erase(n, &ci->i_fragtree); 421355da1ebSSage Weil kfree(frag); 422355da1ebSSage Weil } 423355da1ebSSage Weil 424355da1ebSSage Weil __ceph_destroy_xattrs(ci); 425b6c1d5b8SSage Weil if (ci->i_xattrs.blob) 426355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 427b6c1d5b8SSage Weil if (ci->i_xattrs.prealloc_blob) 428355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.prealloc_blob); 429355da1ebSSage Weil 430fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, ceph_i_callback); 431355da1ebSSage Weil } 432355da1ebSSage Weil 433355da1ebSSage Weil 434355da1ebSSage Weil /* 435355da1ebSSage Weil * Helpers to fill in size, ctime, mtime, and atime. We have to be 436355da1ebSSage Weil * careful because either the client or MDS may have more up to date 437355da1ebSSage Weil * info, depending on which capabilities are held, and whether 438355da1ebSSage Weil * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 439355da1ebSSage Weil * and size are monotonically increasing, except when utimes() or 440355da1ebSSage Weil * truncate() increments the corresponding _seq values.) 441355da1ebSSage Weil */ 442355da1ebSSage Weil int ceph_fill_file_size(struct inode *inode, int issued, 443355da1ebSSage Weil u32 truncate_seq, u64 truncate_size, u64 size) 444355da1ebSSage Weil { 445355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 446355da1ebSSage Weil int queue_trunc = 0; 447355da1ebSSage Weil 448355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 449355da1ebSSage Weil (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 450355da1ebSSage Weil dout("size %lld -> %llu\n", inode->i_size, size); 451355da1ebSSage Weil inode->i_size = size; 452355da1ebSSage Weil inode->i_blocks = (size + (1<<9) - 1) >> 9; 453355da1ebSSage Weil ci->i_reported_size = size; 454355da1ebSSage Weil if (truncate_seq != ci->i_truncate_seq) { 455355da1ebSSage Weil dout("truncate_seq %u -> %u\n", 456355da1ebSSage Weil ci->i_truncate_seq, truncate_seq); 457355da1ebSSage Weil ci->i_truncate_seq = truncate_seq; 4583d497d85SYehuda Sadeh /* 4593d497d85SYehuda Sadeh * If we hold relevant caps, or in the case where we're 4603d497d85SYehuda Sadeh * not the only client referencing this file and we 4613d497d85SYehuda Sadeh * don't hold those caps, then we need to check whether 4623d497d85SYehuda Sadeh * the file is either opened or mmaped 4633d497d85SYehuda Sadeh */ 4643d497d85SYehuda Sadeh if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 465355da1ebSSage Weil CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 4662962507cSSage Weil CEPH_CAP_FILE_EXCL| 4672962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) || 4683d497d85SYehuda Sadeh mapping_mapped(inode->i_mapping) || 4693d497d85SYehuda Sadeh __ceph_caps_file_wanted(ci)) { 470355da1ebSSage Weil ci->i_truncate_pending++; 471355da1ebSSage Weil queue_trunc = 1; 472355da1ebSSage Weil } 473355da1ebSSage Weil } 474355da1ebSSage Weil } 475355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 476355da1ebSSage Weil ci->i_truncate_size != truncate_size) { 477355da1ebSSage Weil dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 478355da1ebSSage Weil truncate_size); 479355da1ebSSage Weil ci->i_truncate_size = truncate_size; 480355da1ebSSage Weil } 481355da1ebSSage Weil return queue_trunc; 482355da1ebSSage Weil } 483355da1ebSSage Weil 484355da1ebSSage Weil void ceph_fill_file_time(struct inode *inode, int issued, 485355da1ebSSage Weil u64 time_warp_seq, struct timespec *ctime, 486355da1ebSSage Weil struct timespec *mtime, struct timespec *atime) 487355da1ebSSage Weil { 488355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 489355da1ebSSage Weil int warn = 0; 490355da1ebSSage Weil 491355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_EXCL| 492355da1ebSSage Weil CEPH_CAP_FILE_WR| 493d8672d64SSage Weil CEPH_CAP_FILE_BUFFER| 494d8672d64SSage Weil CEPH_CAP_AUTH_EXCL| 495d8672d64SSage Weil CEPH_CAP_XATTR_EXCL)) { 496355da1ebSSage Weil if (timespec_compare(ctime, &inode->i_ctime) > 0) { 497355da1ebSSage Weil dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 498355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 499355da1ebSSage Weil ctime->tv_sec, ctime->tv_nsec); 500355da1ebSSage Weil inode->i_ctime = *ctime; 501355da1ebSSage Weil } 502355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 503355da1ebSSage Weil /* the MDS did a utimes() */ 504355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld " 505355da1ebSSage Weil "tw %d -> %d\n", 506355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 507355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec, 508355da1ebSSage Weil ci->i_time_warp_seq, (int)time_warp_seq); 509355da1ebSSage Weil 510355da1ebSSage Weil inode->i_mtime = *mtime; 511355da1ebSSage Weil inode->i_atime = *atime; 512355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 513355da1ebSSage Weil } else if (time_warp_seq == ci->i_time_warp_seq) { 514355da1ebSSage Weil /* nobody did utimes(); take the max */ 515355da1ebSSage Weil if (timespec_compare(mtime, &inode->i_mtime) > 0) { 516355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 517355da1ebSSage Weil inode->i_mtime.tv_sec, 518355da1ebSSage Weil inode->i_mtime.tv_nsec, 519355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec); 520355da1ebSSage Weil inode->i_mtime = *mtime; 521355da1ebSSage Weil } 522355da1ebSSage Weil if (timespec_compare(atime, &inode->i_atime) > 0) { 523355da1ebSSage Weil dout("atime %ld.%09ld -> %ld.%09ld inc\n", 524355da1ebSSage Weil inode->i_atime.tv_sec, 525355da1ebSSage Weil inode->i_atime.tv_nsec, 526355da1ebSSage Weil atime->tv_sec, atime->tv_nsec); 527355da1ebSSage Weil inode->i_atime = *atime; 528355da1ebSSage Weil } 529355da1ebSSage Weil } else if (issued & CEPH_CAP_FILE_EXCL) { 530355da1ebSSage Weil /* we did a utimes(); ignore mds values */ 531355da1ebSSage Weil } else { 532355da1ebSSage Weil warn = 1; 533355da1ebSSage Weil } 534355da1ebSSage Weil } else { 535d8672d64SSage Weil /* we have no write|excl caps; whatever the MDS says is true */ 536355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 537355da1ebSSage Weil inode->i_ctime = *ctime; 538355da1ebSSage Weil inode->i_mtime = *mtime; 539355da1ebSSage Weil inode->i_atime = *atime; 540355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 541355da1ebSSage Weil } else { 542355da1ebSSage Weil warn = 1; 543355da1ebSSage Weil } 544355da1ebSSage Weil } 545355da1ebSSage Weil if (warn) /* time_warp_seq shouldn't go backwards */ 546355da1ebSSage Weil dout("%p mds time_warp_seq %llu < %u\n", 547355da1ebSSage Weil inode, time_warp_seq, ci->i_time_warp_seq); 548355da1ebSSage Weil } 549355da1ebSSage Weil 550355da1ebSSage Weil /* 551355da1ebSSage Weil * Populate an inode based on info from mds. May be called on new or 552355da1ebSSage Weil * existing inodes. 553355da1ebSSage Weil */ 554355da1ebSSage Weil static int fill_inode(struct inode *inode, 555355da1ebSSage Weil struct ceph_mds_reply_info_in *iinfo, 556355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo, 557355da1ebSSage Weil struct ceph_mds_session *session, 558355da1ebSSage Weil unsigned long ttl_from, int cap_fmode, 559355da1ebSSage Weil struct ceph_cap_reservation *caps_reservation) 560355da1ebSSage Weil { 561355da1ebSSage Weil struct ceph_mds_reply_inode *info = iinfo->in; 562355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 563355da1ebSSage Weil int i; 564dfabbed6SSage Weil int issued = 0, implemented; 565355da1ebSSage Weil struct timespec mtime, atime, ctime; 566355da1ebSSage Weil u32 nsplits; 567355da1ebSSage Weil struct ceph_buffer *xattr_blob = NULL; 568355da1ebSSage Weil int err = 0; 569355da1ebSSage Weil int queue_trunc = 0; 570355da1ebSSage Weil 571355da1ebSSage Weil dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 572355da1ebSSage Weil inode, ceph_vinop(inode), le64_to_cpu(info->version), 573355da1ebSSage Weil ci->i_version); 574355da1ebSSage Weil 575355da1ebSSage Weil /* 576355da1ebSSage Weil * prealloc xattr data, if it looks like we'll need it. only 577355da1ebSSage Weil * if len > 4 (meaning there are actually xattrs; the first 4 578355da1ebSSage Weil * bytes are the xattr count). 579355da1ebSSage Weil */ 580355da1ebSSage Weil if (iinfo->xattr_len > 4) { 581b6c1d5b8SSage Weil xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 582355da1ebSSage Weil if (!xattr_blob) 583355da1ebSSage Weil pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 584355da1ebSSage Weil iinfo->xattr_len); 585355da1ebSSage Weil } 586355da1ebSSage Weil 587be655596SSage Weil spin_lock(&ci->i_ceph_lock); 588355da1ebSSage Weil 589355da1ebSSage Weil /* 590355da1ebSSage Weil * provided version will be odd if inode value is projected, 5918bd59e01SSage Weil * even if stable. skip the update if we have newer stable 5928bd59e01SSage Weil * info (ours>=theirs, e.g. due to racing mds replies), unless 5938bd59e01SSage Weil * we are getting projected (unstable) info (in which case the 5948bd59e01SSage Weil * version is odd, and we want ours>theirs). 5958bd59e01SSage Weil * us them 5968bd59e01SSage Weil * 2 2 skip 5978bd59e01SSage Weil * 3 2 skip 5988bd59e01SSage Weil * 3 3 update 599355da1ebSSage Weil */ 600355da1ebSSage Weil if (le64_to_cpu(info->version) > 0 && 6018bd59e01SSage Weil (ci->i_version & ~1) >= le64_to_cpu(info->version)) 602355da1ebSSage Weil goto no_change; 603355da1ebSSage Weil 604355da1ebSSage Weil issued = __ceph_caps_issued(ci, &implemented); 605355da1ebSSage Weil issued |= implemented | __ceph_caps_dirty(ci); 606355da1ebSSage Weil 607355da1ebSSage Weil /* update inode */ 608355da1ebSSage Weil ci->i_version = le64_to_cpu(info->version); 609355da1ebSSage Weil inode->i_version++; 610355da1ebSSage Weil inode->i_rdev = le32_to_cpu(info->rdev); 611355da1ebSSage Weil 612355da1ebSSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 613355da1ebSSage Weil inode->i_mode = le32_to_cpu(info->mode); 614ab871b90SEric W. Biederman inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 615ab871b90SEric W. Biederman inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 616355da1ebSSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 617bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 618bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid)); 619355da1ebSSage Weil } 620355da1ebSSage Weil 621355da1ebSSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 622bfe86848SMiklos Szeredi set_nlink(inode, le32_to_cpu(info->nlink)); 623355da1ebSSage Weil 624355da1ebSSage Weil /* be careful with mtime, atime, size */ 625355da1ebSSage Weil ceph_decode_timespec(&atime, &info->atime); 626355da1ebSSage Weil ceph_decode_timespec(&mtime, &info->mtime); 627355da1ebSSage Weil ceph_decode_timespec(&ctime, &info->ctime); 628355da1ebSSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 629355da1ebSSage Weil le32_to_cpu(info->truncate_seq), 630355da1ebSSage Weil le64_to_cpu(info->truncate_size), 631355da1ebSSage Weil le64_to_cpu(info->size)); 632355da1ebSSage Weil ceph_fill_file_time(inode, issued, 633355da1ebSSage Weil le32_to_cpu(info->time_warp_seq), 634355da1ebSSage Weil &ctime, &mtime, &atime); 635355da1ebSSage Weil 636912a9b03SSage Weil /* only update max_size on auth cap */ 637912a9b03SSage Weil if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 638912a9b03SSage Weil ci->i_max_size != le64_to_cpu(info->max_size)) { 639912a9b03SSage Weil dout("max_size %lld -> %llu\n", ci->i_max_size, 640912a9b03SSage Weil le64_to_cpu(info->max_size)); 641355da1ebSSage Weil ci->i_max_size = le64_to_cpu(info->max_size); 642912a9b03SSage Weil } 643912a9b03SSage Weil 644355da1ebSSage Weil ci->i_layout = info->layout; 645355da1ebSSage Weil inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 646355da1ebSSage Weil 647355da1ebSSage Weil /* xattrs */ 648355da1ebSSage Weil /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 649355da1ebSSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 650355da1ebSSage Weil le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 651355da1ebSSage Weil if (ci->i_xattrs.blob) 652355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 653355da1ebSSage Weil ci->i_xattrs.blob = xattr_blob; 654355da1ebSSage Weil if (xattr_blob) 655355da1ebSSage Weil memcpy(ci->i_xattrs.blob->vec.iov_base, 656355da1ebSSage Weil iinfo->xattr_data, iinfo->xattr_len); 657355da1ebSSage Weil ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 658a6424e48SSage Weil xattr_blob = NULL; 659355da1ebSSage Weil } 660355da1ebSSage Weil 661355da1ebSSage Weil inode->i_mapping->a_ops = &ceph_aops; 662355da1ebSSage Weil inode->i_mapping->backing_dev_info = 663640ef79dSCheng Renquan &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 664355da1ebSSage Weil 665355da1ebSSage Weil switch (inode->i_mode & S_IFMT) { 666355da1ebSSage Weil case S_IFIFO: 667355da1ebSSage Weil case S_IFBLK: 668355da1ebSSage Weil case S_IFCHR: 669355da1ebSSage Weil case S_IFSOCK: 670355da1ebSSage Weil init_special_inode(inode, inode->i_mode, inode->i_rdev); 671355da1ebSSage Weil inode->i_op = &ceph_file_iops; 672355da1ebSSage Weil break; 673355da1ebSSage Weil case S_IFREG: 674355da1ebSSage Weil inode->i_op = &ceph_file_iops; 675355da1ebSSage Weil inode->i_fop = &ceph_file_fops; 676355da1ebSSage Weil break; 677355da1ebSSage Weil case S_IFLNK: 678355da1ebSSage Weil inode->i_op = &ceph_symlink_iops; 679355da1ebSSage Weil if (!ci->i_symlink) { 680810339ecSXi Wang u32 symlen = iinfo->symlink_len; 681355da1ebSSage Weil char *sym; 682355da1ebSSage Weil 683be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 684355da1ebSSage Weil 685810339ecSXi Wang err = -EINVAL; 686810339ecSXi Wang if (WARN_ON(symlen != inode->i_size)) 687810339ecSXi Wang goto out; 688810339ecSXi Wang 689355da1ebSSage Weil err = -ENOMEM; 690810339ecSXi Wang sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 691355da1ebSSage Weil if (!sym) 692355da1ebSSage Weil goto out; 693355da1ebSSage Weil 694be655596SSage Weil spin_lock(&ci->i_ceph_lock); 695355da1ebSSage Weil if (!ci->i_symlink) 696355da1ebSSage Weil ci->i_symlink = sym; 697355da1ebSSage Weil else 698355da1ebSSage Weil kfree(sym); /* lost a race */ 699355da1ebSSage Weil } 700355da1ebSSage Weil break; 701355da1ebSSage Weil case S_IFDIR: 702355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 703355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 704355da1ebSSage Weil 70514303d20SSage Weil ci->i_dir_layout = iinfo->dir_layout; 70614303d20SSage Weil 707355da1ebSSage Weil ci->i_files = le64_to_cpu(info->files); 708355da1ebSSage Weil ci->i_subdirs = le64_to_cpu(info->subdirs); 709355da1ebSSage Weil ci->i_rbytes = le64_to_cpu(info->rbytes); 710355da1ebSSage Weil ci->i_rfiles = le64_to_cpu(info->rfiles); 711355da1ebSSage Weil ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 712355da1ebSSage Weil ceph_decode_timespec(&ci->i_rctime, &info->rctime); 713355da1ebSSage Weil break; 714355da1ebSSage Weil default: 715355da1ebSSage Weil pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 716355da1ebSSage Weil ceph_vinop(inode), inode->i_mode); 717355da1ebSSage Weil } 718355da1ebSSage Weil 719a8673d61SYan, Zheng /* set dir completion flag? */ 720a8673d61SYan, Zheng if (S_ISDIR(inode->i_mode) && 721a8673d61SYan, Zheng ci->i_files == 0 && ci->i_subdirs == 0 && 722a8673d61SYan, Zheng ceph_snap(inode) == CEPH_NOSNAP && 723a8673d61SYan, Zheng (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 724a8673d61SYan, Zheng (issued & CEPH_CAP_FILE_EXCL) == 0 && 7252f276c51SYan, Zheng !__ceph_dir_is_complete(ci)) { 726a8673d61SYan, Zheng dout(" marking %p complete (empty)\n", inode); 7272f276c51SYan, Zheng __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count)); 728a8673d61SYan, Zheng ci->i_max_offset = 2; 729a8673d61SYan, Zheng } 730355da1ebSSage Weil no_change: 731be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 732355da1ebSSage Weil 733355da1ebSSage Weil /* queue truncate if we saw i_size decrease */ 734355da1ebSSage Weil if (queue_trunc) 7353c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 736355da1ebSSage Weil 737355da1ebSSage Weil /* populate frag tree */ 738355da1ebSSage Weil /* FIXME: move me up, if/when version reflects fragtree changes */ 739355da1ebSSage Weil nsplits = le32_to_cpu(info->fragtree.nsplits); 740355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 741355da1ebSSage Weil for (i = 0; i < nsplits; i++) { 742355da1ebSSage Weil u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 743355da1ebSSage Weil struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 744355da1ebSSage Weil 745355da1ebSSage Weil if (IS_ERR(frag)) 746355da1ebSSage Weil continue; 747355da1ebSSage Weil frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 748355da1ebSSage Weil dout(" frag %x split by %d\n", frag->frag, frag->split_by); 749355da1ebSSage Weil } 750355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 751355da1ebSSage Weil 752355da1ebSSage Weil /* were we issued a capability? */ 753355da1ebSSage Weil if (info->cap.caps) { 754355da1ebSSage Weil if (ceph_snap(inode) == CEPH_NOSNAP) { 755355da1ebSSage Weil ceph_add_cap(inode, session, 756355da1ebSSage Weil le64_to_cpu(info->cap.cap_id), 757355da1ebSSage Weil cap_fmode, 758355da1ebSSage Weil le32_to_cpu(info->cap.caps), 759355da1ebSSage Weil le32_to_cpu(info->cap.wanted), 760355da1ebSSage Weil le32_to_cpu(info->cap.seq), 761355da1ebSSage Weil le32_to_cpu(info->cap.mseq), 762355da1ebSSage Weil le64_to_cpu(info->cap.realm), 763355da1ebSSage Weil info->cap.flags, 764355da1ebSSage Weil caps_reservation); 765355da1ebSSage Weil } else { 766be655596SSage Weil spin_lock(&ci->i_ceph_lock); 767355da1ebSSage Weil dout(" %p got snap_caps %s\n", inode, 768355da1ebSSage Weil ceph_cap_string(le32_to_cpu(info->cap.caps))); 769355da1ebSSage Weil ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 770355da1ebSSage Weil if (cap_fmode >= 0) 771355da1ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 772be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 773355da1ebSSage Weil } 77404d000ebSSage Weil } else if (cap_fmode >= 0) { 77504d000ebSSage Weil pr_warning("mds issued no caps on %llx.%llx\n", 77604d000ebSSage Weil ceph_vinop(inode)); 77704d000ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 778355da1ebSSage Weil } 779355da1ebSSage Weil 780355da1ebSSage Weil /* update delegation info? */ 781355da1ebSSage Weil if (dirinfo) 782355da1ebSSage Weil ceph_fill_dirfrag(inode, dirinfo); 783355da1ebSSage Weil 784355da1ebSSage Weil err = 0; 785355da1ebSSage Weil 786355da1ebSSage Weil out: 787b6c1d5b8SSage Weil if (xattr_blob) 788355da1ebSSage Weil ceph_buffer_put(xattr_blob); 789355da1ebSSage Weil return err; 790355da1ebSSage Weil } 791355da1ebSSage Weil 792355da1ebSSage Weil /* 793355da1ebSSage Weil * caller should hold session s_mutex. 794355da1ebSSage Weil */ 795355da1ebSSage Weil static void update_dentry_lease(struct dentry *dentry, 796355da1ebSSage Weil struct ceph_mds_reply_lease *lease, 797355da1ebSSage Weil struct ceph_mds_session *session, 798355da1ebSSage Weil unsigned long from_time) 799355da1ebSSage Weil { 800355da1ebSSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 801355da1ebSSage Weil long unsigned duration = le32_to_cpu(lease->duration_ms); 802355da1ebSSage Weil long unsigned ttl = from_time + (duration * HZ) / 1000; 803355da1ebSSage Weil long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 804355da1ebSSage Weil struct inode *dir; 805355da1ebSSage Weil 806355da1ebSSage Weil /* only track leases on regular dentries */ 807355da1ebSSage Weil if (dentry->d_op != &ceph_dentry_ops) 808355da1ebSSage Weil return; 809355da1ebSSage Weil 810355da1ebSSage Weil spin_lock(&dentry->d_lock); 8112f90b852SSage Weil dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 8122f90b852SSage Weil dentry, duration, ttl); 813355da1ebSSage Weil 814355da1ebSSage Weil /* make lease_rdcache_gen match directory */ 815355da1ebSSage Weil dir = dentry->d_parent->d_inode; 816355da1ebSSage Weil di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 817355da1ebSSage Weil 8182f90b852SSage Weil if (duration == 0) 819355da1ebSSage Weil goto out_unlock; 820355da1ebSSage Weil 821355da1ebSSage Weil if (di->lease_gen == session->s_cap_gen && 822355da1ebSSage Weil time_before(ttl, dentry->d_time)) 823355da1ebSSage Weil goto out_unlock; /* we already have a newer lease. */ 824355da1ebSSage Weil 825355da1ebSSage Weil if (di->lease_session && di->lease_session != session) 826355da1ebSSage Weil goto out_unlock; 827355da1ebSSage Weil 828355da1ebSSage Weil ceph_dentry_lru_touch(dentry); 829355da1ebSSage Weil 830355da1ebSSage Weil if (!di->lease_session) 831355da1ebSSage Weil di->lease_session = ceph_get_mds_session(session); 832355da1ebSSage Weil di->lease_gen = session->s_cap_gen; 833355da1ebSSage Weil di->lease_seq = le32_to_cpu(lease->seq); 834355da1ebSSage Weil di->lease_renew_after = half_ttl; 835355da1ebSSage Weil di->lease_renew_from = 0; 836355da1ebSSage Weil dentry->d_time = ttl; 837355da1ebSSage Weil out_unlock: 838355da1ebSSage Weil spin_unlock(&dentry->d_lock); 839355da1ebSSage Weil return; 840355da1ebSSage Weil } 841355da1ebSSage Weil 842355da1ebSSage Weil /* 8434baa75efSYehuda Sadeh * Set dentry's directory position based on the current dir's max, and 8444baa75efSYehuda Sadeh * order it in d_subdirs, so that dcache_readdir behaves. 8454f177264SSage Weil * 8464f177264SSage Weil * Always called under directory's i_mutex. 8474baa75efSYehuda Sadeh */ 8484baa75efSYehuda Sadeh static void ceph_set_dentry_offset(struct dentry *dn) 8494baa75efSYehuda Sadeh { 8504baa75efSYehuda Sadeh struct dentry *dir = dn->d_parent; 8514f177264SSage Weil struct inode *inode = dir->d_inode; 852b8cd952bSYehuda Sadeh struct ceph_inode_info *ci; 8534baa75efSYehuda Sadeh struct ceph_dentry_info *di; 8544baa75efSYehuda Sadeh 8554baa75efSYehuda Sadeh BUG_ON(!inode); 8564baa75efSYehuda Sadeh 857b8cd952bSYehuda Sadeh ci = ceph_inode(inode); 8584baa75efSYehuda Sadeh di = ceph_dentry(dn); 8594baa75efSYehuda Sadeh 860be655596SSage Weil spin_lock(&ci->i_ceph_lock); 8612f276c51SYan, Zheng if (!__ceph_dir_is_complete(ci)) { 862be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 863e8a74987SSage Weil return; 864e8a74987SSage Weil } 8654baa75efSYehuda Sadeh di->offset = ceph_inode(inode)->i_max_offset++; 866be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 8674baa75efSYehuda Sadeh 8682fd6b7f5SNick Piggin spin_lock(&dir->d_lock); 8692fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 87013a4214cSHenry C Chang list_move(&dn->d_u.d_child, &dir->d_subdirs); 8714baa75efSYehuda Sadeh dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 8724baa75efSYehuda Sadeh dn->d_u.d_child.prev, dn->d_u.d_child.next); 8734baa75efSYehuda Sadeh spin_unlock(&dn->d_lock); 8742fd6b7f5SNick Piggin spin_unlock(&dir->d_lock); 8754baa75efSYehuda Sadeh } 8764baa75efSYehuda Sadeh 8774baa75efSYehuda Sadeh /* 8781cd3935bSSage Weil * splice a dentry to an inode. 8791cd3935bSSage Weil * caller must hold directory i_mutex for this to be safe. 8801cd3935bSSage Weil * 8811cd3935bSSage Weil * we will only rehash the resulting dentry if @prehash is 8821cd3935bSSage Weil * true; @prehash will be set to false (for the benefit of 8831cd3935bSSage Weil * the caller) if we fail. 8841cd3935bSSage Weil */ 8851cd3935bSSage Weil static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 886467c5251SSage Weil bool *prehash, bool set_offset) 8871cd3935bSSage Weil { 8881cd3935bSSage Weil struct dentry *realdn; 8891cd3935bSSage Weil 8901cd3935bSSage Weil BUG_ON(dn->d_inode); 8911cd3935bSSage Weil 8921cd3935bSSage Weil /* dn must be unhashed */ 8931cd3935bSSage Weil if (!d_unhashed(dn)) 8941cd3935bSSage Weil d_drop(dn); 8951cd3935bSSage Weil realdn = d_materialise_unique(dn, in); 8961cd3935bSSage Weil if (IS_ERR(realdn)) { 897d69ed05aSSage Weil pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 898d69ed05aSSage Weil PTR_ERR(realdn), dn, in, ceph_vinop(in)); 8991cd3935bSSage Weil if (prehash) 9001cd3935bSSage Weil *prehash = false; /* don't rehash on error */ 9011cd3935bSSage Weil dn = realdn; /* note realdn contains the error */ 9021cd3935bSSage Weil goto out; 9031cd3935bSSage Weil } else if (realdn) { 9041cd3935bSSage Weil dout("dn %p (%d) spliced with %p (%d) " 9051cd3935bSSage Weil "inode %p ino %llx.%llx\n", 906b7ab39f6SNick Piggin dn, dn->d_count, 907b7ab39f6SNick Piggin realdn, realdn->d_count, 9081cd3935bSSage Weil realdn->d_inode, ceph_vinop(realdn->d_inode)); 9091cd3935bSSage Weil dput(dn); 9101cd3935bSSage Weil dn = realdn; 9111cd3935bSSage Weil } else { 9121cd3935bSSage Weil BUG_ON(!ceph_dentry(dn)); 9131cd3935bSSage Weil dout("dn %p attached to %p ino %llx.%llx\n", 9141cd3935bSSage Weil dn, dn->d_inode, ceph_vinop(dn->d_inode)); 9151cd3935bSSage Weil } 9161cd3935bSSage Weil if ((!prehash || *prehash) && d_unhashed(dn)) 9171cd3935bSSage Weil d_rehash(dn); 918467c5251SSage Weil if (set_offset) 9191cd3935bSSage Weil ceph_set_dentry_offset(dn); 9201cd3935bSSage Weil out: 9211cd3935bSSage Weil return dn; 9221cd3935bSSage Weil } 9231cd3935bSSage Weil 9241cd3935bSSage Weil /* 925355da1ebSSage Weil * Incorporate results into the local cache. This is either just 926355da1ebSSage Weil * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 927355da1ebSSage Weil * after a lookup). 928355da1ebSSage Weil * 929355da1ebSSage Weil * A reply may contain 930355da1ebSSage Weil * a directory inode along with a dentry. 931355da1ebSSage Weil * and/or a target inode 932355da1ebSSage Weil * 933355da1ebSSage Weil * Called with snap_rwsem (read). 934355da1ebSSage Weil */ 935355da1ebSSage Weil int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 936355da1ebSSage Weil struct ceph_mds_session *session) 937355da1ebSSage Weil { 938355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 939355da1ebSSage Weil struct inode *in = NULL; 940355da1ebSSage Weil struct ceph_mds_reply_inode *ininfo; 941355da1ebSSage Weil struct ceph_vino vino; 9423d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 943355da1ebSSage Weil int i = 0; 944355da1ebSSage Weil int err = 0; 945355da1ebSSage Weil 946355da1ebSSage Weil dout("fill_trace %p is_dentry %d is_target %d\n", req, 947355da1ebSSage Weil rinfo->head->is_dentry, rinfo->head->is_target); 948355da1ebSSage Weil 949355da1ebSSage Weil #if 0 950355da1ebSSage Weil /* 951355da1ebSSage Weil * Debugging hook: 952355da1ebSSage Weil * 953355da1ebSSage Weil * If we resend completed ops to a recovering mds, we get no 954355da1ebSSage Weil * trace. Since that is very rare, pretend this is the case 955355da1ebSSage Weil * to ensure the 'no trace' handlers in the callers behave. 956355da1ebSSage Weil * 957355da1ebSSage Weil * Fill in inodes unconditionally to avoid breaking cap 958355da1ebSSage Weil * invariants. 959355da1ebSSage Weil */ 960355da1ebSSage Weil if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 961355da1ebSSage Weil pr_info("fill_trace faking empty trace on %lld %s\n", 962355da1ebSSage Weil req->r_tid, ceph_mds_op_name(rinfo->head->op)); 963355da1ebSSage Weil if (rinfo->head->is_dentry) { 964355da1ebSSage Weil rinfo->head->is_dentry = 0; 965355da1ebSSage Weil err = fill_inode(req->r_locked_dir, 966355da1ebSSage Weil &rinfo->diri, rinfo->dirfrag, 967355da1ebSSage Weil session, req->r_request_started, -1); 968355da1ebSSage Weil } 969355da1ebSSage Weil if (rinfo->head->is_target) { 970355da1ebSSage Weil rinfo->head->is_target = 0; 971355da1ebSSage Weil ininfo = rinfo->targeti.in; 972355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 973355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 974355da1ebSSage Weil in = ceph_get_inode(sb, vino); 975355da1ebSSage Weil err = fill_inode(in, &rinfo->targeti, NULL, 976355da1ebSSage Weil session, req->r_request_started, 977355da1ebSSage Weil req->r_fmode); 978355da1ebSSage Weil iput(in); 979355da1ebSSage Weil } 980355da1ebSSage Weil } 981355da1ebSSage Weil #endif 982355da1ebSSage Weil 983355da1ebSSage Weil if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 984355da1ebSSage Weil dout("fill_trace reply is empty!\n"); 985167c9e35SSage Weil if (rinfo->head->result == 0 && req->r_locked_dir) 986167c9e35SSage Weil ceph_invalidate_dir_request(req); 987355da1ebSSage Weil return 0; 988355da1ebSSage Weil } 989355da1ebSSage Weil 990355da1ebSSage Weil if (rinfo->head->is_dentry) { 9915b1daecdSSage Weil struct inode *dir = req->r_locked_dir; 9925b1daecdSSage Weil 9936c5e50faSSage Weil if (dir) { 9945b1daecdSSage Weil err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 9955b1daecdSSage Weil session, req->r_request_started, -1, 9965b1daecdSSage Weil &req->r_caps_reservation); 9975b1daecdSSage Weil if (err < 0) 9985b1daecdSSage Weil return err; 9996c5e50faSSage Weil } else { 10006c5e50faSSage Weil WARN_ON_ONCE(1); 10016c5e50faSSage Weil } 10025b1daecdSSage Weil } 10035b1daecdSSage Weil 10049358c6d4SSage Weil /* 10059358c6d4SSage Weil * ignore null lease/binding on snapdir ENOENT, or else we 10069358c6d4SSage Weil * will have trouble splicing in the virtual snapdir later 10079358c6d4SSage Weil */ 10089358c6d4SSage Weil if (rinfo->head->is_dentry && !req->r_aborted && 10096c5e50faSSage Weil req->r_locked_dir && 10109358c6d4SSage Weil (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 10113d14c5d2SYehuda Sadeh fsc->mount_options->snapdir_name, 10129358c6d4SSage Weil req->r_dentry->d_name.len))) { 1013355da1ebSSage Weil /* 1014355da1ebSSage Weil * lookup link rename : null -> possibly existing inode 1015355da1ebSSage Weil * mknod symlink mkdir : null -> new inode 1016355da1ebSSage Weil * unlink : linked -> null 1017355da1ebSSage Weil */ 1018355da1ebSSage Weil struct inode *dir = req->r_locked_dir; 1019355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1020355da1ebSSage Weil bool have_dir_cap, have_lease; 1021355da1ebSSage Weil 1022355da1ebSSage Weil BUG_ON(!dn); 1023355da1ebSSage Weil BUG_ON(!dir); 1024355da1ebSSage Weil BUG_ON(dn->d_parent->d_inode != dir); 1025355da1ebSSage Weil BUG_ON(ceph_ino(dir) != 1026355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->ino)); 1027355da1ebSSage Weil BUG_ON(ceph_snap(dir) != 1028355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->snapid)); 1029355da1ebSSage Weil 1030355da1ebSSage Weil /* do we have a lease on the whole dir? */ 1031355da1ebSSage Weil have_dir_cap = 1032355da1ebSSage Weil (le32_to_cpu(rinfo->diri.in->cap.caps) & 1033355da1ebSSage Weil CEPH_CAP_FILE_SHARED); 1034355da1ebSSage Weil 1035355da1ebSSage Weil /* do we have a dn lease? */ 1036355da1ebSSage Weil have_lease = have_dir_cap || 10372f90b852SSage Weil le32_to_cpu(rinfo->dlease->duration_ms); 1038355da1ebSSage Weil if (!have_lease) 1039355da1ebSSage Weil dout("fill_trace no dentry lease or dir cap\n"); 1040355da1ebSSage Weil 1041355da1ebSSage Weil /* rename? */ 1042355da1ebSSage Weil if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1043355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1044355da1ebSSage Weil req->r_old_dentry, 1045355da1ebSSage Weil req->r_old_dentry->d_name.len, 1046355da1ebSSage Weil req->r_old_dentry->d_name.name, 1047355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 1048355da1ebSSage Weil dout("fill_trace doing d_move %p -> %p\n", 1049355da1ebSSage Weil req->r_old_dentry, dn); 1050c10f5e12SSage Weil 1051355da1ebSSage Weil d_move(req->r_old_dentry, dn); 1052355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1053355da1ebSSage Weil req->r_old_dentry, 1054355da1ebSSage Weil req->r_old_dentry->d_name.len, 1055355da1ebSSage Weil req->r_old_dentry->d_name.name, 1056355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 105781a6cf2dSSage Weil 1058c4a29f26SSage Weil /* ensure target dentry is invalidated, despite 1059c4a29f26SSage Weil rehashing bug in vfs_rename_dir */ 106081a6cf2dSSage Weil ceph_invalidate_dentry_lease(dn); 106181a6cf2dSSage Weil 106209adc80cSSage Weil /* 106309adc80cSSage Weil * d_move() puts the renamed dentry at the end of 106409adc80cSSage Weil * d_subdirs. We need to assign it an appropriate 10652f276c51SYan, Zheng * directory offset so we can behave when dir is 10662f276c51SYan, Zheng * complete. 106709adc80cSSage Weil */ 106809adc80cSSage Weil ceph_set_dentry_offset(req->r_old_dentry); 106909adc80cSSage Weil dout("dn %p gets new offset %lld\n", req->r_old_dentry, 10701cd3935bSSage Weil ceph_dentry(req->r_old_dentry)->offset); 107181a6cf2dSSage Weil 1072355da1ebSSage Weil dn = req->r_old_dentry; /* use old_dentry */ 1073355da1ebSSage Weil in = dn->d_inode; 1074355da1ebSSage Weil } 1075355da1ebSSage Weil 1076355da1ebSSage Weil /* null dentry? */ 1077355da1ebSSage Weil if (!rinfo->head->is_target) { 1078355da1ebSSage Weil dout("fill_trace null dentry\n"); 1079355da1ebSSage Weil if (dn->d_inode) { 1080355da1ebSSage Weil dout("d_delete %p\n", dn); 1081355da1ebSSage Weil d_delete(dn); 1082355da1ebSSage Weil } else { 1083355da1ebSSage Weil dout("d_instantiate %p NULL\n", dn); 1084355da1ebSSage Weil d_instantiate(dn, NULL); 1085355da1ebSSage Weil if (have_lease && d_unhashed(dn)) 1086355da1ebSSage Weil d_rehash(dn); 1087355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, 1088355da1ebSSage Weil session, 1089355da1ebSSage Weil req->r_request_started); 1090355da1ebSSage Weil } 1091355da1ebSSage Weil goto done; 1092355da1ebSSage Weil } 1093355da1ebSSage Weil 1094355da1ebSSage Weil /* attach proper inode */ 1095355da1ebSSage Weil ininfo = rinfo->targeti.in; 1096355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1097355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1098d8b16b3dSSage Weil in = dn->d_inode; 1099d8b16b3dSSage Weil if (!in) { 1100355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1101355da1ebSSage Weil if (IS_ERR(in)) { 1102355da1ebSSage Weil pr_err("fill_trace bad get_inode " 1103355da1ebSSage Weil "%llx.%llx\n", vino.ino, vino.snap); 1104355da1ebSSage Weil err = PTR_ERR(in); 11052744c171SAl Viro d_drop(dn); 1106355da1ebSSage Weil goto done; 1107355da1ebSSage Weil } 1108467c5251SSage Weil dn = splice_dentry(dn, in, &have_lease, true); 1109355da1ebSSage Weil if (IS_ERR(dn)) { 1110355da1ebSSage Weil err = PTR_ERR(dn); 1111355da1ebSSage Weil goto done; 1112355da1ebSSage Weil } 1113355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 111470b666c3SSage Weil ihold(in); 1115355da1ebSSage Weil } else if (ceph_ino(in) == vino.ino && 1116355da1ebSSage Weil ceph_snap(in) == vino.snap) { 111770b666c3SSage Weil ihold(in); 1118355da1ebSSage Weil } else { 1119355da1ebSSage Weil dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1120355da1ebSSage Weil dn, in, ceph_ino(in), ceph_snap(in), 1121355da1ebSSage Weil vino.ino, vino.snap); 1122355da1ebSSage Weil have_lease = false; 1123355da1ebSSage Weil in = NULL; 1124355da1ebSSage Weil } 1125355da1ebSSage Weil 1126355da1ebSSage Weil if (have_lease) 1127355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, session, 1128355da1ebSSage Weil req->r_request_started); 1129355da1ebSSage Weil dout(" final dn %p\n", dn); 1130355da1ebSSage Weil i++; 113179f9f99aSSage Weil } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 113279f9f99aSSage Weil req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) { 1133355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1134355da1ebSSage Weil 1135355da1ebSSage Weil /* fill out a snapdir LOOKUPSNAP dentry */ 1136355da1ebSSage Weil BUG_ON(!dn); 1137355da1ebSSage Weil BUG_ON(!req->r_locked_dir); 1138355da1ebSSage Weil BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1139355da1ebSSage Weil ininfo = rinfo->targeti.in; 1140355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1141355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1142355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1143355da1ebSSage Weil if (IS_ERR(in)) { 1144355da1ebSSage Weil pr_err("fill_inode get_inode badness %llx.%llx\n", 1145355da1ebSSage Weil vino.ino, vino.snap); 1146355da1ebSSage Weil err = PTR_ERR(in); 1147355da1ebSSage Weil d_delete(dn); 1148355da1ebSSage Weil goto done; 1149355da1ebSSage Weil } 1150355da1ebSSage Weil dout(" linking snapped dir %p to dn %p\n", in, dn); 1151467c5251SSage Weil dn = splice_dentry(dn, in, NULL, true); 1152355da1ebSSage Weil if (IS_ERR(dn)) { 1153355da1ebSSage Weil err = PTR_ERR(dn); 1154355da1ebSSage Weil goto done; 1155355da1ebSSage Weil } 1156355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 115770b666c3SSage Weil ihold(in); 1158355da1ebSSage Weil rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1159355da1ebSSage Weil } 1160355da1ebSSage Weil 1161355da1ebSSage Weil if (rinfo->head->is_target) { 1162355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1163355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1164355da1ebSSage Weil 1165355da1ebSSage Weil if (in == NULL || ceph_ino(in) != vino.ino || 1166355da1ebSSage Weil ceph_snap(in) != vino.snap) { 1167355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1168355da1ebSSage Weil if (IS_ERR(in)) { 1169355da1ebSSage Weil err = PTR_ERR(in); 1170355da1ebSSage Weil goto done; 1171355da1ebSSage Weil } 1172355da1ebSSage Weil } 1173355da1ebSSage Weil req->r_target_inode = in; 1174355da1ebSSage Weil 1175355da1ebSSage Weil err = fill_inode(in, 1176355da1ebSSage Weil &rinfo->targeti, NULL, 1177355da1ebSSage Weil session, req->r_request_started, 1178355da1ebSSage Weil (le32_to_cpu(rinfo->head->result) == 0) ? 1179355da1ebSSage Weil req->r_fmode : -1, 1180355da1ebSSage Weil &req->r_caps_reservation); 1181355da1ebSSage Weil if (err < 0) { 1182355da1ebSSage Weil pr_err("fill_inode badness %p %llx.%llx\n", 1183355da1ebSSage Weil in, ceph_vinop(in)); 1184355da1ebSSage Weil goto done; 1185355da1ebSSage Weil } 1186355da1ebSSage Weil } 1187355da1ebSSage Weil 1188355da1ebSSage Weil done: 1189355da1ebSSage Weil dout("fill_trace done err=%d\n", err); 1190355da1ebSSage Weil return err; 1191355da1ebSSage Weil } 1192355da1ebSSage Weil 1193355da1ebSSage Weil /* 1194355da1ebSSage Weil * Prepopulate our cache with readdir results, leases, etc. 1195355da1ebSSage Weil */ 119679f9f99aSSage Weil static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 119779f9f99aSSage Weil struct ceph_mds_session *session) 119879f9f99aSSage Weil { 119979f9f99aSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 120079f9f99aSSage Weil int i, err = 0; 120179f9f99aSSage Weil 120279f9f99aSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 120379f9f99aSSage Weil struct ceph_vino vino; 120479f9f99aSSage Weil struct inode *in; 120579f9f99aSSage Weil int rc; 120679f9f99aSSage Weil 120779f9f99aSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 120879f9f99aSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 120979f9f99aSSage Weil 121079f9f99aSSage Weil in = ceph_get_inode(req->r_dentry->d_sb, vino); 121179f9f99aSSage Weil if (IS_ERR(in)) { 121279f9f99aSSage Weil err = PTR_ERR(in); 121379f9f99aSSage Weil dout("new_inode badness got %d\n", err); 121479f9f99aSSage Weil continue; 121579f9f99aSSage Weil } 121679f9f99aSSage Weil rc = fill_inode(in, &rinfo->dir_in[i], NULL, session, 121779f9f99aSSage Weil req->r_request_started, -1, 121879f9f99aSSage Weil &req->r_caps_reservation); 121979f9f99aSSage Weil if (rc < 0) { 122079f9f99aSSage Weil pr_err("fill_inode badness on %p got %d\n", in, rc); 122179f9f99aSSage Weil err = rc; 122279f9f99aSSage Weil continue; 122379f9f99aSSage Weil } 122479f9f99aSSage Weil } 122579f9f99aSSage Weil 122679f9f99aSSage Weil return err; 122779f9f99aSSage Weil } 122879f9f99aSSage Weil 1229355da1ebSSage Weil int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1230355da1ebSSage Weil struct ceph_mds_session *session) 1231355da1ebSSage Weil { 1232355da1ebSSage Weil struct dentry *parent = req->r_dentry; 1233355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1234355da1ebSSage Weil struct qstr dname; 1235355da1ebSSage Weil struct dentry *dn; 1236355da1ebSSage Weil struct inode *in; 1237355da1ebSSage Weil int err = 0, i; 1238355da1ebSSage Weil struct inode *snapdir = NULL; 1239355da1ebSSage Weil struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1240355da1ebSSage Weil u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1241355da1ebSSage Weil struct ceph_dentry_info *di; 1242355da1ebSSage Weil 124379f9f99aSSage Weil if (req->r_aborted) 124479f9f99aSSage Weil return readdir_prepopulate_inodes_only(req, session); 124579f9f99aSSage Weil 1246355da1ebSSage Weil if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1247355da1ebSSage Weil snapdir = ceph_get_snapdir(parent->d_inode); 1248355da1ebSSage Weil parent = d_find_alias(snapdir); 1249355da1ebSSage Weil dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1250355da1ebSSage Weil rinfo->dir_nr, parent); 1251355da1ebSSage Weil } else { 1252355da1ebSSage Weil dout("readdir_prepopulate %d items under dn %p\n", 1253355da1ebSSage Weil rinfo->dir_nr, parent); 1254355da1ebSSage Weil if (rinfo->dir_dir) 1255355da1ebSSage Weil ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1256355da1ebSSage Weil } 1257355da1ebSSage Weil 1258355da1ebSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 1259355da1ebSSage Weil struct ceph_vino vino; 1260355da1ebSSage Weil 1261355da1ebSSage Weil dname.name = rinfo->dir_dname[i]; 1262355da1ebSSage Weil dname.len = rinfo->dir_dname_len[i]; 1263355da1ebSSage Weil dname.hash = full_name_hash(dname.name, dname.len); 1264355da1ebSSage Weil 1265355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1266355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1267355da1ebSSage Weil 1268355da1ebSSage Weil retry_lookup: 1269355da1ebSSage Weil dn = d_lookup(parent, &dname); 1270355da1ebSSage Weil dout("d_lookup on parent=%p name=%.*s got %p\n", 1271355da1ebSSage Weil parent, dname.len, dname.name, dn); 1272355da1ebSSage Weil 1273355da1ebSSage Weil if (!dn) { 1274355da1ebSSage Weil dn = d_alloc(parent, &dname); 1275355da1ebSSage Weil dout("d_alloc %p '%.*s' = %p\n", parent, 1276355da1ebSSage Weil dname.len, dname.name, dn); 1277355da1ebSSage Weil if (dn == NULL) { 1278355da1ebSSage Weil dout("d_alloc badness\n"); 1279355da1ebSSage Weil err = -ENOMEM; 1280355da1ebSSage Weil goto out; 1281355da1ebSSage Weil } 1282355da1ebSSage Weil err = ceph_init_dentry(dn); 12838c696737SSage Weil if (err < 0) { 12848c696737SSage Weil dput(dn); 1285355da1ebSSage Weil goto out; 12868c696737SSage Weil } 1287355da1ebSSage Weil } else if (dn->d_inode && 1288355da1ebSSage Weil (ceph_ino(dn->d_inode) != vino.ino || 1289355da1ebSSage Weil ceph_snap(dn->d_inode) != vino.snap)) { 1290355da1ebSSage Weil dout(" dn %p points to wrong inode %p\n", 1291355da1ebSSage Weil dn, dn->d_inode); 1292355da1ebSSage Weil d_delete(dn); 1293355da1ebSSage Weil dput(dn); 1294355da1ebSSage Weil goto retry_lookup; 1295355da1ebSSage Weil } else { 1296355da1ebSSage Weil /* reorder parent's d_subdirs */ 12972fd6b7f5SNick Piggin spin_lock(&parent->d_lock); 12982fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1299355da1ebSSage Weil list_move(&dn->d_u.d_child, &parent->d_subdirs); 1300355da1ebSSage Weil spin_unlock(&dn->d_lock); 13012fd6b7f5SNick Piggin spin_unlock(&parent->d_lock); 1302355da1ebSSage Weil } 1303355da1ebSSage Weil 1304355da1ebSSage Weil di = dn->d_fsdata; 1305355da1ebSSage Weil di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1306355da1ebSSage Weil 1307355da1ebSSage Weil /* inode */ 1308355da1ebSSage Weil if (dn->d_inode) { 1309355da1ebSSage Weil in = dn->d_inode; 1310355da1ebSSage Weil } else { 1311355da1ebSSage Weil in = ceph_get_inode(parent->d_sb, vino); 1312ac1f12efSDan Carpenter if (IS_ERR(in)) { 1313355da1ebSSage Weil dout("new_inode badness\n"); 13142744c171SAl Viro d_drop(dn); 1315355da1ebSSage Weil dput(dn); 1316ac1f12efSDan Carpenter err = PTR_ERR(in); 1317355da1ebSSage Weil goto out; 1318355da1ebSSage Weil } 1319467c5251SSage Weil dn = splice_dentry(dn, in, NULL, false); 1320d69ed05aSSage Weil if (IS_ERR(dn)) 1321d69ed05aSSage Weil dn = NULL; 1322355da1ebSSage Weil } 1323355da1ebSSage Weil 1324355da1ebSSage Weil if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1325355da1ebSSage Weil req->r_request_started, -1, 1326355da1ebSSage Weil &req->r_caps_reservation) < 0) { 1327355da1ebSSage Weil pr_err("fill_inode badness on %p\n", in); 1328d69ed05aSSage Weil goto next_item; 1329355da1ebSSage Weil } 1330d69ed05aSSage Weil if (dn) 1331355da1ebSSage Weil update_dentry_lease(dn, rinfo->dir_dlease[i], 1332d69ed05aSSage Weil req->r_session, 1333d69ed05aSSage Weil req->r_request_started); 1334d69ed05aSSage Weil next_item: 1335d69ed05aSSage Weil if (dn) 1336355da1ebSSage Weil dput(dn); 1337355da1ebSSage Weil } 1338355da1ebSSage Weil req->r_did_prepopulate = true; 1339355da1ebSSage Weil 1340355da1ebSSage Weil out: 1341355da1ebSSage Weil if (snapdir) { 1342355da1ebSSage Weil iput(snapdir); 1343355da1ebSSage Weil dput(parent); 1344355da1ebSSage Weil } 1345355da1ebSSage Weil dout("readdir_prepopulate done\n"); 1346355da1ebSSage Weil return err; 1347355da1ebSSage Weil } 1348355da1ebSSage Weil 1349355da1ebSSage Weil int ceph_inode_set_size(struct inode *inode, loff_t size) 1350355da1ebSSage Weil { 1351355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1352355da1ebSSage Weil int ret = 0; 1353355da1ebSSage Weil 1354be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1355355da1ebSSage Weil dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1356355da1ebSSage Weil inode->i_size = size; 1357355da1ebSSage Weil inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1358355da1ebSSage Weil 1359355da1ebSSage Weil /* tell the MDS if we are approaching max_size */ 1360355da1ebSSage Weil if ((size << 1) >= ci->i_max_size && 1361355da1ebSSage Weil (ci->i_reported_size << 1) < ci->i_max_size) 1362355da1ebSSage Weil ret = 1; 1363355da1ebSSage Weil 1364be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1365355da1ebSSage Weil return ret; 1366355da1ebSSage Weil } 1367355da1ebSSage Weil 1368355da1ebSSage Weil /* 1369355da1ebSSage Weil * Write back inode data in a worker thread. (This can't be done 1370355da1ebSSage Weil * in the message handler context.) 1371355da1ebSSage Weil */ 13723c6f6b79SSage Weil void ceph_queue_writeback(struct inode *inode) 13733c6f6b79SSage Weil { 137415a2015fSSage Weil ihold(inode); 13753c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->wb_wq, 13763c6f6b79SSage Weil &ceph_inode(inode)->i_wb_work)) { 13772c27c9a5SSage Weil dout("ceph_queue_writeback %p\n", inode); 13783c6f6b79SSage Weil } else { 13792c27c9a5SSage Weil dout("ceph_queue_writeback %p failed\n", inode); 138015a2015fSSage Weil iput(inode); 13813c6f6b79SSage Weil } 13823c6f6b79SSage Weil } 13833c6f6b79SSage Weil 13843c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work) 1385355da1ebSSage Weil { 1386355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1387355da1ebSSage Weil i_wb_work); 1388355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1389355da1ebSSage Weil 1390355da1ebSSage Weil dout("writeback %p\n", inode); 1391355da1ebSSage Weil filemap_fdatawrite(&inode->i_data); 1392355da1ebSSage Weil iput(inode); 1393355da1ebSSage Weil } 1394355da1ebSSage Weil 1395355da1ebSSage Weil /* 13963c6f6b79SSage Weil * queue an async invalidation 13973c6f6b79SSage Weil */ 13983c6f6b79SSage Weil void ceph_queue_invalidate(struct inode *inode) 13993c6f6b79SSage Weil { 140015a2015fSSage Weil ihold(inode); 14013c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 14023c6f6b79SSage Weil &ceph_inode(inode)->i_pg_inv_work)) { 14033c6f6b79SSage Weil dout("ceph_queue_invalidate %p\n", inode); 14043c6f6b79SSage Weil } else { 14053c6f6b79SSage Weil dout("ceph_queue_invalidate %p failed\n", inode); 140615a2015fSSage Weil iput(inode); 14073c6f6b79SSage Weil } 14083c6f6b79SSage Weil } 14093c6f6b79SSage Weil 14103c6f6b79SSage Weil /* 1411355da1ebSSage Weil * Invalidate inode pages in a worker thread. (This can't be done 1412355da1ebSSage Weil * in the message handler context.) 1413355da1ebSSage Weil */ 14143c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work) 1415355da1ebSSage Weil { 1416355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1417355da1ebSSage Weil i_pg_inv_work); 1418355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1419355da1ebSSage Weil u32 orig_gen; 1420355da1ebSSage Weil int check = 0; 1421355da1ebSSage Weil 1422be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1423355da1ebSSage Weil dout("invalidate_pages %p gen %d revoking %d\n", inode, 1424355da1ebSSage Weil ci->i_rdcache_gen, ci->i_rdcache_revoking); 1425cd045cb4SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1426355da1ebSSage Weil /* nevermind! */ 1427be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1428355da1ebSSage Weil goto out; 1429355da1ebSSage Weil } 1430355da1ebSSage Weil orig_gen = ci->i_rdcache_gen; 1431be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1432355da1ebSSage Weil 143383eaea22SSage Weil truncate_inode_pages(&inode->i_data, 0); 1434355da1ebSSage Weil 1435be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1436cd045cb4SSage Weil if (orig_gen == ci->i_rdcache_gen && 1437cd045cb4SSage Weil orig_gen == ci->i_rdcache_revoking) { 1438355da1ebSSage Weil dout("invalidate_pages %p gen %d successful\n", inode, 1439355da1ebSSage Weil ci->i_rdcache_gen); 1440cd045cb4SSage Weil ci->i_rdcache_revoking--; 1441355da1ebSSage Weil check = 1; 1442355da1ebSSage Weil } else { 1443cd045cb4SSage Weil dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1444cd045cb4SSage Weil inode, orig_gen, ci->i_rdcache_gen, 1445cd045cb4SSage Weil ci->i_rdcache_revoking); 1446355da1ebSSage Weil } 1447be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1448355da1ebSSage Weil 1449355da1ebSSage Weil if (check) 1450355da1ebSSage Weil ceph_check_caps(ci, 0, NULL); 1451355da1ebSSage Weil out: 1452355da1ebSSage Weil iput(inode); 1453355da1ebSSage Weil } 1454355da1ebSSage Weil 1455355da1ebSSage Weil 1456355da1ebSSage Weil /* 14573f99969fSYan, Zheng * called by trunc_wq; 1458355da1ebSSage Weil * 1459355da1ebSSage Weil * We also truncate in a separate thread as well. 1460355da1ebSSage Weil */ 14613c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work) 1462355da1ebSSage Weil { 1463355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1464355da1ebSSage Weil i_vmtruncate_work); 1465355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1466355da1ebSSage Weil 1467355da1ebSSage Weil dout("vmtruncate_work %p\n", inode); 1468b415bf4fSYan, Zheng mutex_lock(&inode->i_mutex); 1469b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 1470b415bf4fSYan, Zheng mutex_unlock(&inode->i_mutex); 1471355da1ebSSage Weil iput(inode); 1472355da1ebSSage Weil } 1473355da1ebSSage Weil 1474355da1ebSSage Weil /* 14753c6f6b79SSage Weil * Queue an async vmtruncate. If we fail to queue work, we will handle 14763c6f6b79SSage Weil * the truncation the next time we call __ceph_do_pending_vmtruncate. 14773c6f6b79SSage Weil */ 14783c6f6b79SSage Weil void ceph_queue_vmtruncate(struct inode *inode) 14793c6f6b79SSage Weil { 14803c6f6b79SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14813c6f6b79SSage Weil 148215a2015fSSage Weil ihold(inode); 1483640ef79dSCheng Renquan if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 14843c6f6b79SSage Weil &ci->i_vmtruncate_work)) { 14853c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p\n", inode); 14863c6f6b79SSage Weil } else { 14873c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 14883c6f6b79SSage Weil inode, ci->i_truncate_pending); 148915a2015fSSage Weil iput(inode); 14903c6f6b79SSage Weil } 14913c6f6b79SSage Weil } 14923c6f6b79SSage Weil 14933c6f6b79SSage Weil /* 1494355da1ebSSage Weil * Make sure any pending truncation is applied before doing anything 1495355da1ebSSage Weil * that may depend on it. 1496355da1ebSSage Weil */ 1497b415bf4fSYan, Zheng void __ceph_do_pending_vmtruncate(struct inode *inode) 1498355da1ebSSage Weil { 1499355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1500355da1ebSSage Weil u64 to; 1501a85f50b6SYan, Zheng int wrbuffer_refs, finish = 0; 1502355da1ebSSage Weil 1503355da1ebSSage Weil retry: 1504be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1505355da1ebSSage Weil if (ci->i_truncate_pending == 0) { 1506355da1ebSSage Weil dout("__do_pending_vmtruncate %p none pending\n", inode); 1507be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1508355da1ebSSage Weil return; 1509355da1ebSSage Weil } 1510355da1ebSSage Weil 1511355da1ebSSage Weil /* 1512355da1ebSSage Weil * make sure any dirty snapped pages are flushed before we 1513355da1ebSSage Weil * possibly truncate them.. so write AND block! 1514355da1ebSSage Weil */ 1515355da1ebSSage Weil if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1516355da1ebSSage Weil dout("__do_pending_vmtruncate %p flushing snaps first\n", 1517355da1ebSSage Weil inode); 1518be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1519355da1ebSSage Weil filemap_write_and_wait_range(&inode->i_data, 0, 1520355da1ebSSage Weil inode->i_sb->s_maxbytes); 1521355da1ebSSage Weil goto retry; 1522355da1ebSSage Weil } 1523355da1ebSSage Weil 1524355da1ebSSage Weil to = ci->i_truncate_size; 1525355da1ebSSage Weil wrbuffer_refs = ci->i_wrbuffer_ref; 1526355da1ebSSage Weil dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1527355da1ebSSage Weil ci->i_truncate_pending, to); 1528be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1529355da1ebSSage Weil 1530355da1ebSSage Weil truncate_inode_pages(inode->i_mapping, to); 1531355da1ebSSage Weil 1532be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1533a85f50b6SYan, Zheng if (to == ci->i_truncate_size) { 1534a85f50b6SYan, Zheng ci->i_truncate_pending = 0; 1535a85f50b6SYan, Zheng finish = 1; 1536a85f50b6SYan, Zheng } 1537be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1538a85f50b6SYan, Zheng if (!finish) 1539a85f50b6SYan, Zheng goto retry; 1540355da1ebSSage Weil 1541355da1ebSSage Weil if (wrbuffer_refs == 0) 1542355da1ebSSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1543a85f50b6SYan, Zheng 154403066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1545355da1ebSSage Weil } 1546355da1ebSSage Weil 1547355da1ebSSage Weil 1548355da1ebSSage Weil /* 1549355da1ebSSage Weil * symlinks 1550355da1ebSSage Weil */ 1551355da1ebSSage Weil static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1552355da1ebSSage Weil { 1553355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1554355da1ebSSage Weil nd_set_link(nd, ci->i_symlink); 1555355da1ebSSage Weil return NULL; 1556355da1ebSSage Weil } 1557355da1ebSSage Weil 1558355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops = { 1559355da1ebSSage Weil .readlink = generic_readlink, 1560355da1ebSSage Weil .follow_link = ceph_sym_follow_link, 15610b932672SYan, Zheng .setattr = ceph_setattr, 15620b932672SYan, Zheng .getattr = ceph_getattr, 15630b932672SYan, Zheng .setxattr = ceph_setxattr, 15640b932672SYan, Zheng .getxattr = ceph_getxattr, 15650b932672SYan, Zheng .listxattr = ceph_listxattr, 15660b932672SYan, Zheng .removexattr = ceph_removexattr, 1567355da1ebSSage Weil }; 1568355da1ebSSage Weil 1569355da1ebSSage Weil /* 1570355da1ebSSage Weil * setattr 1571355da1ebSSage Weil */ 1572355da1ebSSage Weil int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1573355da1ebSSage Weil { 1574355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1575355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 15765f21c96dSSage Weil struct inode *parent_inode; 1577355da1ebSSage Weil const unsigned int ia_valid = attr->ia_valid; 1578355da1ebSSage Weil struct ceph_mds_request *req; 15793d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1580355da1ebSSage Weil int issued; 1581355da1ebSSage Weil int release = 0, dirtied = 0; 1582355da1ebSSage Weil int mask = 0; 1583355da1ebSSage Weil int err = 0; 1584fca65b4aSSage Weil int inode_dirty_flags = 0; 1585355da1ebSSage Weil 1586355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1587355da1ebSSage Weil return -EROFS; 1588355da1ebSSage Weil 1589b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 1590355da1ebSSage Weil 1591355da1ebSSage Weil err = inode_change_ok(inode, attr); 1592355da1ebSSage Weil if (err != 0) 1593355da1ebSSage Weil return err; 1594355da1ebSSage Weil 1595355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1596355da1ebSSage Weil USE_AUTH_MDS); 1597355da1ebSSage Weil if (IS_ERR(req)) 1598355da1ebSSage Weil return PTR_ERR(req); 1599355da1ebSSage Weil 1600be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1601355da1ebSSage Weil issued = __ceph_caps_issued(ci, NULL); 1602355da1ebSSage Weil dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1603355da1ebSSage Weil 1604355da1ebSSage Weil if (ia_valid & ATTR_UID) { 1605355da1ebSSage Weil dout("setattr %p uid %d -> %d\n", inode, 1606bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 1607bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, attr->ia_uid)); 1608355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1609355da1ebSSage Weil inode->i_uid = attr->ia_uid; 1610355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1611355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1612ab871b90SEric W. Biederman !uid_eq(attr->ia_uid, inode->i_uid)) { 1613ab871b90SEric W. Biederman req->r_args.setattr.uid = cpu_to_le32( 1614ab871b90SEric W. Biederman from_kuid(&init_user_ns, attr->ia_uid)); 1615355da1ebSSage Weil mask |= CEPH_SETATTR_UID; 1616355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1617355da1ebSSage Weil } 1618355da1ebSSage Weil } 1619355da1ebSSage Weil if (ia_valid & ATTR_GID) { 1620355da1ebSSage Weil dout("setattr %p gid %d -> %d\n", inode, 1621bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid), 1622bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, attr->ia_gid)); 1623355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1624355da1ebSSage Weil inode->i_gid = attr->ia_gid; 1625355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1626355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1627ab871b90SEric W. Biederman !gid_eq(attr->ia_gid, inode->i_gid)) { 1628ab871b90SEric W. Biederman req->r_args.setattr.gid = cpu_to_le32( 1629ab871b90SEric W. Biederman from_kgid(&init_user_ns, attr->ia_gid)); 1630355da1ebSSage Weil mask |= CEPH_SETATTR_GID; 1631355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1632355da1ebSSage Weil } 1633355da1ebSSage Weil } 1634355da1ebSSage Weil if (ia_valid & ATTR_MODE) { 1635355da1ebSSage Weil dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1636355da1ebSSage Weil attr->ia_mode); 1637355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1638355da1ebSSage Weil inode->i_mode = attr->ia_mode; 1639355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1640355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1641355da1ebSSage Weil attr->ia_mode != inode->i_mode) { 1642355da1ebSSage Weil req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1643355da1ebSSage Weil mask |= CEPH_SETATTR_MODE; 1644355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1645355da1ebSSage Weil } 1646355da1ebSSage Weil } 1647355da1ebSSage Weil 1648355da1ebSSage Weil if (ia_valid & ATTR_ATIME) { 1649355da1ebSSage Weil dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1650355da1ebSSage Weil inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1651355da1ebSSage Weil attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1652355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1653355da1ebSSage Weil ci->i_time_warp_seq++; 1654355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1655355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1656355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1657355da1ebSSage Weil timespec_compare(&inode->i_atime, 1658355da1ebSSage Weil &attr->ia_atime) < 0) { 1659355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1660355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1661355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1662355da1ebSSage Weil !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1663355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.atime, 1664355da1ebSSage Weil &attr->ia_atime); 1665355da1ebSSage Weil mask |= CEPH_SETATTR_ATIME; 1666355da1ebSSage Weil release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1667355da1ebSSage Weil CEPH_CAP_FILE_WR; 1668355da1ebSSage Weil } 1669355da1ebSSage Weil } 1670355da1ebSSage Weil if (ia_valid & ATTR_MTIME) { 1671355da1ebSSage Weil dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1672355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1673355da1ebSSage Weil attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1674355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1675355da1ebSSage Weil ci->i_time_warp_seq++; 1676355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1677355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1678355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1679355da1ebSSage Weil timespec_compare(&inode->i_mtime, 1680355da1ebSSage Weil &attr->ia_mtime) < 0) { 1681355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1682355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1683355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1684355da1ebSSage Weil !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1685355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.mtime, 1686355da1ebSSage Weil &attr->ia_mtime); 1687355da1ebSSage Weil mask |= CEPH_SETATTR_MTIME; 1688355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1689355da1ebSSage Weil CEPH_CAP_FILE_WR; 1690355da1ebSSage Weil } 1691355da1ebSSage Weil } 1692355da1ebSSage Weil if (ia_valid & ATTR_SIZE) { 1693355da1ebSSage Weil dout("setattr %p size %lld -> %lld\n", inode, 1694355da1ebSSage Weil inode->i_size, attr->ia_size); 1695355da1ebSSage Weil if (attr->ia_size > inode->i_sb->s_maxbytes) { 1696355da1ebSSage Weil err = -EINVAL; 1697355da1ebSSage Weil goto out; 1698355da1ebSSage Weil } 1699355da1ebSSage Weil if ((issued & CEPH_CAP_FILE_EXCL) && 1700355da1ebSSage Weil attr->ia_size > inode->i_size) { 1701355da1ebSSage Weil inode->i_size = attr->ia_size; 1702355da1ebSSage Weil inode->i_blocks = 1703355da1ebSSage Weil (attr->ia_size + (1 << 9) - 1) >> 9; 1704355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1705355da1ebSSage Weil ci->i_reported_size = attr->ia_size; 1706355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1707355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1708355da1ebSSage Weil attr->ia_size != inode->i_size) { 1709355da1ebSSage Weil req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1710355da1ebSSage Weil req->r_args.setattr.old_size = 1711355da1ebSSage Weil cpu_to_le64(inode->i_size); 1712355da1ebSSage Weil mask |= CEPH_SETATTR_SIZE; 1713355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1714355da1ebSSage Weil CEPH_CAP_FILE_WR; 1715355da1ebSSage Weil } 1716355da1ebSSage Weil } 1717355da1ebSSage Weil 1718355da1ebSSage Weil /* these do nothing */ 1719355da1ebSSage Weil if (ia_valid & ATTR_CTIME) { 1720355da1ebSSage Weil bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1721355da1ebSSage Weil ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1722355da1ebSSage Weil dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1723355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1724355da1ebSSage Weil attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1725355da1ebSSage Weil only ? "ctime only" : "ignored"); 1726355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1727355da1ebSSage Weil if (only) { 1728355da1ebSSage Weil /* 1729355da1ebSSage Weil * if kernel wants to dirty ctime but nothing else, 1730355da1ebSSage Weil * we need to choose a cap to dirty under, or do 1731355da1ebSSage Weil * a almost-no-op setattr 1732355da1ebSSage Weil */ 1733355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) 1734355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1735355da1ebSSage Weil else if (issued & CEPH_CAP_FILE_EXCL) 1736355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1737355da1ebSSage Weil else if (issued & CEPH_CAP_XATTR_EXCL) 1738355da1ebSSage Weil dirtied |= CEPH_CAP_XATTR_EXCL; 1739355da1ebSSage Weil else 1740355da1ebSSage Weil mask |= CEPH_SETATTR_CTIME; 1741355da1ebSSage Weil } 1742355da1ebSSage Weil } 1743355da1ebSSage Weil if (ia_valid & ATTR_FILE) 1744355da1ebSSage Weil dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1745355da1ebSSage Weil 1746355da1ebSSage Weil if (dirtied) { 1747fca65b4aSSage Weil inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1748355da1ebSSage Weil inode->i_ctime = CURRENT_TIME; 1749355da1ebSSage Weil } 1750355da1ebSSage Weil 1751355da1ebSSage Weil release &= issued; 1752be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1753355da1ebSSage Weil 1754fca65b4aSSage Weil if (inode_dirty_flags) 1755fca65b4aSSage Weil __mark_inode_dirty(inode, inode_dirty_flags); 1756fca65b4aSSage Weil 1757355da1ebSSage Weil if (mask) { 175870b666c3SSage Weil req->r_inode = inode; 175970b666c3SSage Weil ihold(inode); 1760355da1ebSSage Weil req->r_inode_drop = release; 1761355da1ebSSage Weil req->r_args.setattr.mask = cpu_to_le32(mask); 1762355da1ebSSage Weil req->r_num_caps = 1; 17635f21c96dSSage Weil parent_inode = ceph_get_dentry_parent_inode(dentry); 1764355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, parent_inode, req); 17655f21c96dSSage Weil iput(parent_inode); 1766355da1ebSSage Weil } 1767355da1ebSSage Weil dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1768355da1ebSSage Weil ceph_cap_string(dirtied), mask); 1769355da1ebSSage Weil 1770355da1ebSSage Weil ceph_mdsc_put_request(req); 1771b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 1772355da1ebSSage Weil return err; 1773355da1ebSSage Weil out: 1774be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1775355da1ebSSage Weil ceph_mdsc_put_request(req); 1776355da1ebSSage Weil return err; 1777355da1ebSSage Weil } 1778355da1ebSSage Weil 1779355da1ebSSage Weil /* 1780355da1ebSSage Weil * Verify that we have a lease on the given mask. If not, 1781355da1ebSSage Weil * do a getattr against an mds. 1782355da1ebSSage Weil */ 1783355da1ebSSage Weil int ceph_do_getattr(struct inode *inode, int mask) 1784355da1ebSSage Weil { 17853d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 17863d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1787355da1ebSSage Weil struct ceph_mds_request *req; 1788355da1ebSSage Weil int err; 1789355da1ebSSage Weil 1790355da1ebSSage Weil if (ceph_snap(inode) == CEPH_SNAPDIR) { 1791355da1ebSSage Weil dout("do_getattr inode %p SNAPDIR\n", inode); 1792355da1ebSSage Weil return 0; 1793355da1ebSSage Weil } 1794355da1ebSSage Weil 1795b7495fc2SSage Weil dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1796355da1ebSSage Weil if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1797355da1ebSSage Weil return 0; 1798355da1ebSSage Weil 1799355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1800355da1ebSSage Weil if (IS_ERR(req)) 1801355da1ebSSage Weil return PTR_ERR(req); 180270b666c3SSage Weil req->r_inode = inode; 180370b666c3SSage Weil ihold(inode); 1804355da1ebSSage Weil req->r_num_caps = 1; 1805355da1ebSSage Weil req->r_args.getattr.mask = cpu_to_le32(mask); 1806355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1807355da1ebSSage Weil ceph_mdsc_put_request(req); 1808355da1ebSSage Weil dout("do_getattr result=%d\n", err); 1809355da1ebSSage Weil return err; 1810355da1ebSSage Weil } 1811355da1ebSSage Weil 1812355da1ebSSage Weil 1813355da1ebSSage Weil /* 1814355da1ebSSage Weil * Check inode permissions. We verify we have a valid value for 1815355da1ebSSage Weil * the AUTH cap, then call the generic handler. 1816355da1ebSSage Weil */ 181710556cb2SAl Viro int ceph_permission(struct inode *inode, int mask) 1818355da1ebSSage Weil { 1819b74c79e9SNick Piggin int err; 1820b74c79e9SNick Piggin 182110556cb2SAl Viro if (mask & MAY_NOT_BLOCK) 1822b74c79e9SNick Piggin return -ECHILD; 1823b74c79e9SNick Piggin 1824b74c79e9SNick Piggin err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1825355da1ebSSage Weil 1826355da1ebSSage Weil if (!err) 18272830ba7fSAl Viro err = generic_permission(inode, mask); 1828355da1ebSSage Weil return err; 1829355da1ebSSage Weil } 1830355da1ebSSage Weil 1831355da1ebSSage Weil /* 1832355da1ebSSage Weil * Get all attributes. Hopefully somedata we'll have a statlite() 1833355da1ebSSage Weil * and can limit the fields we require to be accurate. 1834355da1ebSSage Weil */ 1835355da1ebSSage Weil int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1836355da1ebSSage Weil struct kstat *stat) 1837355da1ebSSage Weil { 1838355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1839232d4b01SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1840355da1ebSSage Weil int err; 1841355da1ebSSage Weil 1842355da1ebSSage Weil err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1843355da1ebSSage Weil if (!err) { 1844355da1ebSSage Weil generic_fillattr(inode, stat); 1845ad1fee96SYehuda Sadeh stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1846355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1847355da1ebSSage Weil stat->dev = ceph_snap(inode); 1848355da1ebSSage Weil else 1849355da1ebSSage Weil stat->dev = 0; 1850232d4b01SSage Weil if (S_ISDIR(inode->i_mode)) { 18511c1266bbSYehuda Sadeh if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 18521c1266bbSYehuda Sadeh RBYTES)) 1853232d4b01SSage Weil stat->size = ci->i_rbytes; 18541c1266bbSYehuda Sadeh else 18551c1266bbSYehuda Sadeh stat->size = ci->i_files + ci->i_subdirs; 1856232d4b01SSage Weil stat->blocks = 0; 1857355da1ebSSage Weil stat->blksize = 65536; 1858355da1ebSSage Weil } 1859232d4b01SSage Weil } 1860355da1ebSSage Weil return err; 1861355da1ebSSage Weil } 1862