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> 124db658eaSLinus Torvalds #include <linux/posix_acl.h> 13355da1ebSSage Weil 14355da1ebSSage Weil #include "super.h" 153d14c5d2SYehuda Sadeh #include "mds_client.h" 1699ccbd22SMilosz Tanski #include "cache.h" 173d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 18355da1ebSSage Weil 19355da1ebSSage Weil /* 20355da1ebSSage Weil * Ceph inode operations 21355da1ebSSage Weil * 22355da1ebSSage Weil * Implement basic inode helpers (get, alloc) and inode ops (getattr, 23355da1ebSSage Weil * setattr, etc.), xattr helpers, and helpers for assimilating 24355da1ebSSage Weil * metadata returned by the MDS into our cache. 25355da1ebSSage Weil * 26355da1ebSSage Weil * Also define helpers for doing asynchronous writeback, invalidation, 27355da1ebSSage Weil * and truncation for the benefit of those who can't afford to block 28355da1ebSSage Weil * (typically because they are in the message handler path). 29355da1ebSSage Weil */ 30355da1ebSSage Weil 31355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops; 32355da1ebSSage Weil 333c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work); 343c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work); 353c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work); 36355da1ebSSage Weil 37355da1ebSSage Weil /* 38355da1ebSSage Weil * find or create an inode, given the ceph ino number 39355da1ebSSage Weil */ 40ad1fee96SYehuda Sadeh static int ceph_set_ino_cb(struct inode *inode, void *data) 41ad1fee96SYehuda Sadeh { 42ad1fee96SYehuda Sadeh ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 43ad1fee96SYehuda Sadeh inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 44ad1fee96SYehuda Sadeh return 0; 45ad1fee96SYehuda Sadeh } 46ad1fee96SYehuda Sadeh 47355da1ebSSage Weil struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 48355da1ebSSage Weil { 49355da1ebSSage Weil struct inode *inode; 50355da1ebSSage Weil ino_t t = ceph_vino_to_ino(vino); 51355da1ebSSage Weil 52355da1ebSSage Weil inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 53355da1ebSSage Weil if (inode == NULL) 54355da1ebSSage Weil return ERR_PTR(-ENOMEM); 55355da1ebSSage Weil if (inode->i_state & I_NEW) { 56355da1ebSSage Weil dout("get_inode created new inode %p %llx.%llx ino %llx\n", 57355da1ebSSage Weil inode, ceph_vinop(inode), (u64)inode->i_ino); 58355da1ebSSage Weil unlock_new_inode(inode); 59355da1ebSSage Weil } 60355da1ebSSage Weil 61355da1ebSSage Weil dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 62355da1ebSSage Weil vino.snap, inode); 63355da1ebSSage Weil return inode; 64355da1ebSSage Weil } 65355da1ebSSage Weil 66355da1ebSSage Weil /* 67355da1ebSSage Weil * get/constuct snapdir inode for a given directory 68355da1ebSSage Weil */ 69355da1ebSSage Weil struct inode *ceph_get_snapdir(struct inode *parent) 70355da1ebSSage Weil { 71355da1ebSSage Weil struct ceph_vino vino = { 72355da1ebSSage Weil .ino = ceph_ino(parent), 73355da1ebSSage Weil .snap = CEPH_SNAPDIR, 74355da1ebSSage Weil }; 75355da1ebSSage Weil struct inode *inode = ceph_get_inode(parent->i_sb, vino); 76b377ff13SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 77355da1ebSSage Weil 78355da1ebSSage Weil BUG_ON(!S_ISDIR(parent->i_mode)); 79355da1ebSSage Weil if (IS_ERR(inode)) 807e34bc52SJulia Lawall return inode; 81355da1ebSSage Weil inode->i_mode = parent->i_mode; 82355da1ebSSage Weil inode->i_uid = parent->i_uid; 83355da1ebSSage Weil inode->i_gid = parent->i_gid; 84355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 85355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 86b377ff13SSage Weil ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 87b377ff13SSage Weil ci->i_rbytes = 0; 88355da1ebSSage Weil return inode; 89355da1ebSSage Weil } 90355da1ebSSage Weil 91355da1ebSSage Weil const struct inode_operations ceph_file_iops = { 92355da1ebSSage Weil .permission = ceph_permission, 93355da1ebSSage Weil .setattr = ceph_setattr, 94355da1ebSSage Weil .getattr = ceph_getattr, 95355da1ebSSage Weil .setxattr = ceph_setxattr, 96355da1ebSSage Weil .getxattr = ceph_getxattr, 97355da1ebSSage Weil .listxattr = ceph_listxattr, 98355da1ebSSage Weil .removexattr = ceph_removexattr, 997221fe4cSGuangliang Zhao .get_acl = ceph_get_acl, 10072466d0bSSage Weil .set_acl = ceph_set_acl, 101355da1ebSSage Weil }; 102355da1ebSSage Weil 103355da1ebSSage Weil 104355da1ebSSage Weil /* 105355da1ebSSage Weil * We use a 'frag tree' to keep track of the MDS's directory fragments 106355da1ebSSage Weil * for a given inode (usually there is just a single fragment). We 107355da1ebSSage Weil * need to know when a child frag is delegated to a new MDS, or when 108355da1ebSSage Weil * it is flagged as replicated, so we can direct our requests 109355da1ebSSage Weil * accordingly. 110355da1ebSSage Weil */ 111355da1ebSSage Weil 112355da1ebSSage Weil /* 113355da1ebSSage Weil * find/create a frag in the tree 114355da1ebSSage Weil */ 115355da1ebSSage Weil static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 116355da1ebSSage Weil u32 f) 117355da1ebSSage Weil { 118355da1ebSSage Weil struct rb_node **p; 119355da1ebSSage Weil struct rb_node *parent = NULL; 120355da1ebSSage Weil struct ceph_inode_frag *frag; 121355da1ebSSage Weil int c; 122355da1ebSSage Weil 123355da1ebSSage Weil p = &ci->i_fragtree.rb_node; 124355da1ebSSage Weil while (*p) { 125355da1ebSSage Weil parent = *p; 126355da1ebSSage Weil frag = rb_entry(parent, struct ceph_inode_frag, node); 127355da1ebSSage Weil c = ceph_frag_compare(f, frag->frag); 128355da1ebSSage Weil if (c < 0) 129355da1ebSSage Weil p = &(*p)->rb_left; 130355da1ebSSage Weil else if (c > 0) 131355da1ebSSage Weil p = &(*p)->rb_right; 132355da1ebSSage Weil else 133355da1ebSSage Weil return frag; 134355da1ebSSage Weil } 135355da1ebSSage Weil 136355da1ebSSage Weil frag = kmalloc(sizeof(*frag), GFP_NOFS); 137355da1ebSSage Weil if (!frag) { 138355da1ebSSage Weil pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 139355da1ebSSage Weil "frag %x\n", &ci->vfs_inode, 140355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 141355da1ebSSage Weil return ERR_PTR(-ENOMEM); 142355da1ebSSage Weil } 143355da1ebSSage Weil frag->frag = f; 144355da1ebSSage Weil frag->split_by = 0; 145355da1ebSSage Weil frag->mds = -1; 146355da1ebSSage Weil frag->ndist = 0; 147355da1ebSSage Weil 148355da1ebSSage Weil rb_link_node(&frag->node, parent, p); 149355da1ebSSage Weil rb_insert_color(&frag->node, &ci->i_fragtree); 150355da1ebSSage Weil 151355da1ebSSage Weil dout("get_or_create_frag added %llx.%llx frag %x\n", 152355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 153355da1ebSSage Weil return frag; 154355da1ebSSage Weil } 155355da1ebSSage Weil 156355da1ebSSage Weil /* 157355da1ebSSage Weil * find a specific frag @f 158355da1ebSSage Weil */ 159355da1ebSSage Weil struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 160355da1ebSSage Weil { 161355da1ebSSage Weil struct rb_node *n = ci->i_fragtree.rb_node; 162355da1ebSSage Weil 163355da1ebSSage Weil while (n) { 164355da1ebSSage Weil struct ceph_inode_frag *frag = 165355da1ebSSage Weil rb_entry(n, struct ceph_inode_frag, node); 166355da1ebSSage Weil int c = ceph_frag_compare(f, frag->frag); 167355da1ebSSage Weil if (c < 0) 168355da1ebSSage Weil n = n->rb_left; 169355da1ebSSage Weil else if (c > 0) 170355da1ebSSage Weil n = n->rb_right; 171355da1ebSSage Weil else 172355da1ebSSage Weil return frag; 173355da1ebSSage Weil } 174355da1ebSSage Weil return NULL; 175355da1ebSSage Weil } 176355da1ebSSage Weil 177355da1ebSSage Weil /* 178355da1ebSSage Weil * Choose frag containing the given value @v. If @pfrag is 179355da1ebSSage Weil * specified, copy the frag delegation info to the caller if 180355da1ebSSage Weil * it is present. 181355da1ebSSage Weil */ 182355da1ebSSage Weil u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 183355da1ebSSage Weil struct ceph_inode_frag *pfrag, 184355da1ebSSage Weil int *found) 185355da1ebSSage Weil { 186355da1ebSSage Weil u32 t = ceph_frag_make(0, 0); 187355da1ebSSage Weil struct ceph_inode_frag *frag; 188355da1ebSSage Weil unsigned nway, i; 189355da1ebSSage Weil u32 n; 190355da1ebSSage Weil 191355da1ebSSage Weil if (found) 192355da1ebSSage Weil *found = 0; 193355da1ebSSage Weil 194355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 195355da1ebSSage Weil while (1) { 196355da1ebSSage Weil WARN_ON(!ceph_frag_contains_value(t, v)); 197355da1ebSSage Weil frag = __ceph_find_frag(ci, t); 198355da1ebSSage Weil if (!frag) 199355da1ebSSage Weil break; /* t is a leaf */ 200355da1ebSSage Weil if (frag->split_by == 0) { 201355da1ebSSage Weil if (pfrag) 202355da1ebSSage Weil memcpy(pfrag, frag, sizeof(*pfrag)); 203355da1ebSSage Weil if (found) 204355da1ebSSage Weil *found = 1; 205355da1ebSSage Weil break; 206355da1ebSSage Weil } 207355da1ebSSage Weil 208355da1ebSSage Weil /* choose child */ 209355da1ebSSage Weil nway = 1 << frag->split_by; 210355da1ebSSage Weil dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 211355da1ebSSage Weil frag->split_by, nway); 212355da1ebSSage Weil for (i = 0; i < nway; i++) { 213355da1ebSSage Weil n = ceph_frag_make_child(t, frag->split_by, i); 214355da1ebSSage Weil if (ceph_frag_contains_value(n, v)) { 215355da1ebSSage Weil t = n; 216355da1ebSSage Weil break; 217355da1ebSSage Weil } 218355da1ebSSage Weil } 219355da1ebSSage Weil BUG_ON(i == nway); 220355da1ebSSage Weil } 221355da1ebSSage Weil dout("choose_frag(%x) = %x\n", v, t); 222355da1ebSSage Weil 223355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 224355da1ebSSage Weil return t; 225355da1ebSSage Weil } 226355da1ebSSage Weil 227355da1ebSSage Weil /* 228355da1ebSSage Weil * Process dirfrag (delegation) info from the mds. Include leaf 229355da1ebSSage Weil * fragment in tree ONLY if ndist > 0. Otherwise, only 230355da1ebSSage Weil * branches/splits are included in i_fragtree) 231355da1ebSSage Weil */ 232355da1ebSSage Weil static int ceph_fill_dirfrag(struct inode *inode, 233355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo) 234355da1ebSSage Weil { 235355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 236355da1ebSSage Weil struct ceph_inode_frag *frag; 237355da1ebSSage Weil u32 id = le32_to_cpu(dirinfo->frag); 238355da1ebSSage Weil int mds = le32_to_cpu(dirinfo->auth); 239355da1ebSSage Weil int ndist = le32_to_cpu(dirinfo->ndist); 240355da1ebSSage Weil int i; 241355da1ebSSage Weil int err = 0; 242355da1ebSSage Weil 243355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 244355da1ebSSage Weil if (ndist == 0) { 245355da1ebSSage Weil /* no delegation info needed. */ 246355da1ebSSage Weil frag = __ceph_find_frag(ci, id); 247355da1ebSSage Weil if (!frag) 248355da1ebSSage Weil goto out; 249355da1ebSSage Weil if (frag->split_by == 0) { 250355da1ebSSage Weil /* tree leaf, remove */ 251355da1ebSSage Weil dout("fill_dirfrag removed %llx.%llx frag %x" 252355da1ebSSage Weil " (no ref)\n", ceph_vinop(inode), id); 253355da1ebSSage Weil rb_erase(&frag->node, &ci->i_fragtree); 254355da1ebSSage Weil kfree(frag); 255355da1ebSSage Weil } else { 256355da1ebSSage Weil /* tree branch, keep and clear */ 257355da1ebSSage Weil dout("fill_dirfrag cleared %llx.%llx frag %x" 258355da1ebSSage Weil " referral\n", ceph_vinop(inode), id); 259355da1ebSSage Weil frag->mds = -1; 260355da1ebSSage Weil frag->ndist = 0; 261355da1ebSSage Weil } 262355da1ebSSage Weil goto out; 263355da1ebSSage Weil } 264355da1ebSSage Weil 265355da1ebSSage Weil 266355da1ebSSage Weil /* find/add this frag to store mds delegation info */ 267355da1ebSSage Weil frag = __get_or_create_frag(ci, id); 268355da1ebSSage Weil if (IS_ERR(frag)) { 269355da1ebSSage Weil /* this is not the end of the world; we can continue 270355da1ebSSage Weil with bad/inaccurate delegation info */ 271355da1ebSSage Weil pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 272355da1ebSSage Weil ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 273355da1ebSSage Weil err = -ENOMEM; 274355da1ebSSage Weil goto out; 275355da1ebSSage Weil } 276355da1ebSSage Weil 277355da1ebSSage Weil frag->mds = mds; 278355da1ebSSage Weil frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 279355da1ebSSage Weil for (i = 0; i < frag->ndist; i++) 280355da1ebSSage Weil frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 281355da1ebSSage Weil dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 282355da1ebSSage Weil ceph_vinop(inode), frag->frag, frag->ndist); 283355da1ebSSage Weil 284355da1ebSSage Weil out: 285355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 286355da1ebSSage Weil return err; 287355da1ebSSage Weil } 288355da1ebSSage Weil 289355da1ebSSage Weil 290355da1ebSSage Weil /* 291355da1ebSSage Weil * initialize a newly allocated inode. 292355da1ebSSage Weil */ 293355da1ebSSage Weil struct inode *ceph_alloc_inode(struct super_block *sb) 294355da1ebSSage Weil { 295355da1ebSSage Weil struct ceph_inode_info *ci; 296355da1ebSSage Weil int i; 297355da1ebSSage Weil 298355da1ebSSage Weil ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 299355da1ebSSage Weil if (!ci) 300355da1ebSSage Weil return NULL; 301355da1ebSSage Weil 302355da1ebSSage Weil dout("alloc_inode %p\n", &ci->vfs_inode); 303355da1ebSSage Weil 304be655596SSage Weil spin_lock_init(&ci->i_ceph_lock); 305be655596SSage Weil 306355da1ebSSage Weil ci->i_version = 0; 307355da1ebSSage Weil ci->i_time_warp_seq = 0; 308355da1ebSSage Weil ci->i_ceph_flags = 0; 3092f276c51SYan, Zheng atomic_set(&ci->i_release_count, 1); 3102f276c51SYan, Zheng atomic_set(&ci->i_complete_count, 0); 311355da1ebSSage Weil ci->i_symlink = NULL; 312355da1ebSSage Weil 3136c0f3af7SSage Weil memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 3146c0f3af7SSage Weil 315355da1ebSSage Weil ci->i_fragtree = RB_ROOT; 316355da1ebSSage Weil mutex_init(&ci->i_fragtree_mutex); 317355da1ebSSage Weil 318355da1ebSSage Weil ci->i_xattrs.blob = NULL; 319355da1ebSSage Weil ci->i_xattrs.prealloc_blob = NULL; 320355da1ebSSage Weil ci->i_xattrs.dirty = false; 321355da1ebSSage Weil ci->i_xattrs.index = RB_ROOT; 322355da1ebSSage Weil ci->i_xattrs.count = 0; 323355da1ebSSage Weil ci->i_xattrs.names_size = 0; 324355da1ebSSage Weil ci->i_xattrs.vals_size = 0; 325355da1ebSSage Weil ci->i_xattrs.version = 0; 326355da1ebSSage Weil ci->i_xattrs.index_version = 0; 327355da1ebSSage Weil 328355da1ebSSage Weil ci->i_caps = RB_ROOT; 329355da1ebSSage Weil ci->i_auth_cap = NULL; 330355da1ebSSage Weil ci->i_dirty_caps = 0; 331355da1ebSSage Weil ci->i_flushing_caps = 0; 332355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_dirty_item); 333355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_flushing_item); 334355da1ebSSage Weil ci->i_cap_flush_seq = 0; 335355da1ebSSage Weil ci->i_cap_flush_last_tid = 0; 336355da1ebSSage Weil memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 337355da1ebSSage Weil init_waitqueue_head(&ci->i_cap_wq); 338355da1ebSSage Weil ci->i_hold_caps_min = 0; 339355da1ebSSage Weil ci->i_hold_caps_max = 0; 340355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_delay_list); 341355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_snaps); 342355da1ebSSage Weil ci->i_head_snapc = NULL; 343355da1ebSSage Weil ci->i_snap_caps = 0; 34411df2dfbSYan, Zheng ci->i_cap_exporting_issued = 0; 345355da1ebSSage Weil 346355da1ebSSage Weil for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 347355da1ebSSage Weil ci->i_nr_by_mode[i] = 0; 348355da1ebSSage Weil 349b0d7c223SYan, Zheng mutex_init(&ci->i_truncate_mutex); 350355da1ebSSage Weil ci->i_truncate_seq = 0; 351355da1ebSSage Weil ci->i_truncate_size = 0; 352355da1ebSSage Weil ci->i_truncate_pending = 0; 353355da1ebSSage Weil 354355da1ebSSage Weil ci->i_max_size = 0; 355355da1ebSSage Weil ci->i_reported_size = 0; 356355da1ebSSage Weil ci->i_wanted_max_size = 0; 357355da1ebSSage Weil ci->i_requested_max_size = 0; 358355da1ebSSage Weil 359355da1ebSSage Weil ci->i_pin_ref = 0; 360355da1ebSSage Weil ci->i_rd_ref = 0; 361355da1ebSSage Weil ci->i_rdcache_ref = 0; 362355da1ebSSage Weil ci->i_wr_ref = 0; 363d3d0720dSHenry C Chang ci->i_wb_ref = 0; 364355da1ebSSage Weil ci->i_wrbuffer_ref = 0; 365355da1ebSSage Weil ci->i_wrbuffer_ref_head = 0; 366355da1ebSSage Weil ci->i_shared_gen = 0; 367355da1ebSSage Weil ci->i_rdcache_gen = 0; 368355da1ebSSage Weil ci->i_rdcache_revoking = 0; 369355da1ebSSage Weil 370355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_writes); 371355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_dirops); 372355da1ebSSage Weil spin_lock_init(&ci->i_unsafe_lock); 373355da1ebSSage Weil 374355da1ebSSage Weil ci->i_snap_realm = NULL; 375355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_realm_item); 376355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_flush_item); 377355da1ebSSage Weil 3783c6f6b79SSage Weil INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 3793c6f6b79SSage Weil INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 380355da1ebSSage Weil 381355da1ebSSage Weil INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 382355da1ebSSage Weil 38399ccbd22SMilosz Tanski ceph_fscache_inode_init(ci); 38499ccbd22SMilosz Tanski 385355da1ebSSage Weil return &ci->vfs_inode; 386355da1ebSSage Weil } 387355da1ebSSage Weil 388fa0d7e3dSNick Piggin static void ceph_i_callback(struct rcu_head *head) 389fa0d7e3dSNick Piggin { 390fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 391fa0d7e3dSNick Piggin struct ceph_inode_info *ci = ceph_inode(inode); 392fa0d7e3dSNick Piggin 393fa0d7e3dSNick Piggin kmem_cache_free(ceph_inode_cachep, ci); 394fa0d7e3dSNick Piggin } 395fa0d7e3dSNick Piggin 396355da1ebSSage Weil void ceph_destroy_inode(struct inode *inode) 397355da1ebSSage Weil { 398355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 399355da1ebSSage Weil struct ceph_inode_frag *frag; 400355da1ebSSage Weil struct rb_node *n; 401355da1ebSSage Weil 402355da1ebSSage Weil dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 403355da1ebSSage Weil 40499ccbd22SMilosz Tanski ceph_fscache_unregister_inode_cookie(ci); 40599ccbd22SMilosz Tanski 406355da1ebSSage Weil ceph_queue_caps_release(inode); 407355da1ebSSage Weil 4088b218b8aSSage Weil /* 4098b218b8aSSage Weil * we may still have a snap_realm reference if there are stray 4108b218b8aSSage Weil * caps in i_cap_exporting_issued or i_snap_caps. 4118b218b8aSSage Weil */ 4128b218b8aSSage Weil if (ci->i_snap_realm) { 4138b218b8aSSage Weil struct ceph_mds_client *mdsc = 4143d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 4158b218b8aSSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 4168b218b8aSSage Weil 4178b218b8aSSage Weil dout(" dropping residual ref to snap realm %p\n", realm); 4188b218b8aSSage Weil spin_lock(&realm->inodes_with_caps_lock); 4198b218b8aSSage Weil list_del_init(&ci->i_snap_realm_item); 4208b218b8aSSage Weil spin_unlock(&realm->inodes_with_caps_lock); 4218b218b8aSSage Weil ceph_put_snap_realm(mdsc, realm); 4228b218b8aSSage Weil } 4238b218b8aSSage Weil 424355da1ebSSage Weil kfree(ci->i_symlink); 425355da1ebSSage Weil while ((n = rb_first(&ci->i_fragtree)) != NULL) { 426355da1ebSSage Weil frag = rb_entry(n, struct ceph_inode_frag, node); 427355da1ebSSage Weil rb_erase(n, &ci->i_fragtree); 428355da1ebSSage Weil kfree(frag); 429355da1ebSSage Weil } 430355da1ebSSage Weil 431355da1ebSSage Weil __ceph_destroy_xattrs(ci); 432b6c1d5b8SSage Weil if (ci->i_xattrs.blob) 433355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 434b6c1d5b8SSage Weil if (ci->i_xattrs.prealloc_blob) 435355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.prealloc_blob); 436355da1ebSSage Weil 437fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, ceph_i_callback); 438355da1ebSSage Weil } 439355da1ebSSage Weil 4409f12bd11SYan, Zheng int ceph_drop_inode(struct inode *inode) 4419f12bd11SYan, Zheng { 4429f12bd11SYan, Zheng /* 4439f12bd11SYan, Zheng * Positve dentry and corresponding inode are always accompanied 4449f12bd11SYan, Zheng * in MDS reply. So no need to keep inode in the cache after 4459f12bd11SYan, Zheng * dropping all its aliases. 4469f12bd11SYan, Zheng */ 4479f12bd11SYan, Zheng return 1; 4489f12bd11SYan, Zheng } 4499f12bd11SYan, Zheng 450355da1ebSSage Weil /* 451355da1ebSSage Weil * Helpers to fill in size, ctime, mtime, and atime. We have to be 452355da1ebSSage Weil * careful because either the client or MDS may have more up to date 453355da1ebSSage Weil * info, depending on which capabilities are held, and whether 454355da1ebSSage Weil * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 455355da1ebSSage Weil * and size are monotonically increasing, except when utimes() or 456355da1ebSSage Weil * truncate() increments the corresponding _seq values.) 457355da1ebSSage Weil */ 458355da1ebSSage Weil int ceph_fill_file_size(struct inode *inode, int issued, 459355da1ebSSage Weil u32 truncate_seq, u64 truncate_size, u64 size) 460355da1ebSSage Weil { 461355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 462355da1ebSSage Weil int queue_trunc = 0; 463355da1ebSSage Weil 464355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 465355da1ebSSage Weil (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 466355da1ebSSage Weil dout("size %lld -> %llu\n", inode->i_size, size); 467355da1ebSSage Weil inode->i_size = size; 468355da1ebSSage Weil inode->i_blocks = (size + (1<<9) - 1) >> 9; 469355da1ebSSage Weil ci->i_reported_size = size; 470355da1ebSSage Weil if (truncate_seq != ci->i_truncate_seq) { 471355da1ebSSage Weil dout("truncate_seq %u -> %u\n", 472355da1ebSSage Weil ci->i_truncate_seq, truncate_seq); 473355da1ebSSage Weil ci->i_truncate_seq = truncate_seq; 474b0d7c223SYan, Zheng 475b0d7c223SYan, Zheng /* the MDS should have revoked these caps */ 476b0d7c223SYan, Zheng WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 477b0d7c223SYan, Zheng CEPH_CAP_FILE_RD | 478b0d7c223SYan, Zheng CEPH_CAP_FILE_WR | 479b0d7c223SYan, Zheng CEPH_CAP_FILE_LAZYIO)); 4803d497d85SYehuda Sadeh /* 4813d497d85SYehuda Sadeh * If we hold relevant caps, or in the case where we're 4823d497d85SYehuda Sadeh * not the only client referencing this file and we 4833d497d85SYehuda Sadeh * don't hold those caps, then we need to check whether 4843d497d85SYehuda Sadeh * the file is either opened or mmaped 4853d497d85SYehuda Sadeh */ 486b0d7c223SYan, Zheng if ((issued & (CEPH_CAP_FILE_CACHE| 487b0d7c223SYan, Zheng CEPH_CAP_FILE_BUFFER)) || 4883d497d85SYehuda Sadeh mapping_mapped(inode->i_mapping) || 4893d497d85SYehuda Sadeh __ceph_caps_file_wanted(ci)) { 490355da1ebSSage Weil ci->i_truncate_pending++; 491355da1ebSSage Weil queue_trunc = 1; 492355da1ebSSage Weil } 493355da1ebSSage Weil } 494355da1ebSSage Weil } 495355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 496355da1ebSSage Weil ci->i_truncate_size != truncate_size) { 497355da1ebSSage Weil dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 498355da1ebSSage Weil truncate_size); 499355da1ebSSage Weil ci->i_truncate_size = truncate_size; 500355da1ebSSage Weil } 50199ccbd22SMilosz Tanski 50299ccbd22SMilosz Tanski if (queue_trunc) 50399ccbd22SMilosz Tanski ceph_fscache_invalidate(inode); 50499ccbd22SMilosz Tanski 505355da1ebSSage Weil return queue_trunc; 506355da1ebSSage Weil } 507355da1ebSSage Weil 508355da1ebSSage Weil void ceph_fill_file_time(struct inode *inode, int issued, 509355da1ebSSage Weil u64 time_warp_seq, struct timespec *ctime, 510355da1ebSSage Weil struct timespec *mtime, struct timespec *atime) 511355da1ebSSage Weil { 512355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 513355da1ebSSage Weil int warn = 0; 514355da1ebSSage Weil 515355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_EXCL| 516355da1ebSSage Weil CEPH_CAP_FILE_WR| 517d8672d64SSage Weil CEPH_CAP_FILE_BUFFER| 518d8672d64SSage Weil CEPH_CAP_AUTH_EXCL| 519d8672d64SSage Weil CEPH_CAP_XATTR_EXCL)) { 520355da1ebSSage Weil if (timespec_compare(ctime, &inode->i_ctime) > 0) { 521355da1ebSSage Weil dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 522355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 523355da1ebSSage Weil ctime->tv_sec, ctime->tv_nsec); 524355da1ebSSage Weil inode->i_ctime = *ctime; 525355da1ebSSage Weil } 526355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 527355da1ebSSage Weil /* the MDS did a utimes() */ 528355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld " 529355da1ebSSage Weil "tw %d -> %d\n", 530355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 531355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec, 532355da1ebSSage Weil ci->i_time_warp_seq, (int)time_warp_seq); 533355da1ebSSage Weil 534355da1ebSSage Weil inode->i_mtime = *mtime; 535355da1ebSSage Weil inode->i_atime = *atime; 536355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 537355da1ebSSage Weil } else if (time_warp_seq == ci->i_time_warp_seq) { 538355da1ebSSage Weil /* nobody did utimes(); take the max */ 539355da1ebSSage Weil if (timespec_compare(mtime, &inode->i_mtime) > 0) { 540355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 541355da1ebSSage Weil inode->i_mtime.tv_sec, 542355da1ebSSage Weil inode->i_mtime.tv_nsec, 543355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec); 544355da1ebSSage Weil inode->i_mtime = *mtime; 545355da1ebSSage Weil } 546355da1ebSSage Weil if (timespec_compare(atime, &inode->i_atime) > 0) { 547355da1ebSSage Weil dout("atime %ld.%09ld -> %ld.%09ld inc\n", 548355da1ebSSage Weil inode->i_atime.tv_sec, 549355da1ebSSage Weil inode->i_atime.tv_nsec, 550355da1ebSSage Weil atime->tv_sec, atime->tv_nsec); 551355da1ebSSage Weil inode->i_atime = *atime; 552355da1ebSSage Weil } 553355da1ebSSage Weil } else if (issued & CEPH_CAP_FILE_EXCL) { 554355da1ebSSage Weil /* we did a utimes(); ignore mds values */ 555355da1ebSSage Weil } else { 556355da1ebSSage Weil warn = 1; 557355da1ebSSage Weil } 558355da1ebSSage Weil } else { 559d8672d64SSage Weil /* we have no write|excl caps; whatever the MDS says is true */ 560355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 561355da1ebSSage Weil inode->i_ctime = *ctime; 562355da1ebSSage Weil inode->i_mtime = *mtime; 563355da1ebSSage Weil inode->i_atime = *atime; 564355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 565355da1ebSSage Weil } else { 566355da1ebSSage Weil warn = 1; 567355da1ebSSage Weil } 568355da1ebSSage Weil } 569355da1ebSSage Weil if (warn) /* time_warp_seq shouldn't go backwards */ 570355da1ebSSage Weil dout("%p mds time_warp_seq %llu < %u\n", 571355da1ebSSage Weil inode, time_warp_seq, ci->i_time_warp_seq); 572355da1ebSSage Weil } 573355da1ebSSage Weil 574355da1ebSSage Weil /* 575355da1ebSSage Weil * Populate an inode based on info from mds. May be called on new or 576355da1ebSSage Weil * existing inodes. 577355da1ebSSage Weil */ 578355da1ebSSage Weil static int fill_inode(struct inode *inode, 579355da1ebSSage Weil struct ceph_mds_reply_info_in *iinfo, 580355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo, 581355da1ebSSage Weil struct ceph_mds_session *session, 582355da1ebSSage Weil unsigned long ttl_from, int cap_fmode, 583355da1ebSSage Weil struct ceph_cap_reservation *caps_reservation) 584355da1ebSSage Weil { 585355da1ebSSage Weil struct ceph_mds_reply_inode *info = iinfo->in; 586355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 587355da1ebSSage Weil int i; 588dfabbed6SSage Weil int issued = 0, implemented; 589355da1ebSSage Weil struct timespec mtime, atime, ctime; 590355da1ebSSage Weil u32 nsplits; 59153e879a4SYan, Zheng struct ceph_inode_frag *frag; 59253e879a4SYan, Zheng struct rb_node *rb_node; 593355da1ebSSage Weil struct ceph_buffer *xattr_blob = NULL; 594355da1ebSSage Weil int err = 0; 595355da1ebSSage Weil int queue_trunc = 0; 596355da1ebSSage Weil 597355da1ebSSage Weil dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 598355da1ebSSage Weil inode, ceph_vinop(inode), le64_to_cpu(info->version), 599355da1ebSSage Weil ci->i_version); 600355da1ebSSage Weil 601355da1ebSSage Weil /* 602355da1ebSSage Weil * prealloc xattr data, if it looks like we'll need it. only 603355da1ebSSage Weil * if len > 4 (meaning there are actually xattrs; the first 4 604355da1ebSSage Weil * bytes are the xattr count). 605355da1ebSSage Weil */ 606355da1ebSSage Weil if (iinfo->xattr_len > 4) { 607b6c1d5b8SSage Weil xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 608355da1ebSSage Weil if (!xattr_blob) 609355da1ebSSage Weil pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 610355da1ebSSage Weil iinfo->xattr_len); 611355da1ebSSage Weil } 612355da1ebSSage Weil 613be655596SSage Weil spin_lock(&ci->i_ceph_lock); 614355da1ebSSage Weil 615355da1ebSSage Weil /* 616355da1ebSSage Weil * provided version will be odd if inode value is projected, 6178bd59e01SSage Weil * even if stable. skip the update if we have newer stable 6188bd59e01SSage Weil * info (ours>=theirs, e.g. due to racing mds replies), unless 6198bd59e01SSage Weil * we are getting projected (unstable) info (in which case the 6208bd59e01SSage Weil * version is odd, and we want ours>theirs). 6218bd59e01SSage Weil * us them 6228bd59e01SSage Weil * 2 2 skip 6238bd59e01SSage Weil * 3 2 skip 6248bd59e01SSage Weil * 3 3 update 625355da1ebSSage Weil */ 626355da1ebSSage Weil if (le64_to_cpu(info->version) > 0 && 6278bd59e01SSage Weil (ci->i_version & ~1) >= le64_to_cpu(info->version)) 628355da1ebSSage Weil goto no_change; 629355da1ebSSage Weil 630355da1ebSSage Weil issued = __ceph_caps_issued(ci, &implemented); 631355da1ebSSage Weil issued |= implemented | __ceph_caps_dirty(ci); 632355da1ebSSage Weil 633355da1ebSSage Weil /* update inode */ 634355da1ebSSage Weil ci->i_version = le64_to_cpu(info->version); 635355da1ebSSage Weil inode->i_version++; 636355da1ebSSage Weil inode->i_rdev = le32_to_cpu(info->rdev); 637355da1ebSSage Weil 638355da1ebSSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 639355da1ebSSage Weil inode->i_mode = le32_to_cpu(info->mode); 640ab871b90SEric W. Biederman inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 641ab871b90SEric W. Biederman inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 642355da1ebSSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 643bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 644bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid)); 645355da1ebSSage Weil } 646355da1ebSSage Weil 647355da1ebSSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 648bfe86848SMiklos Szeredi set_nlink(inode, le32_to_cpu(info->nlink)); 649355da1ebSSage Weil 650355da1ebSSage Weil /* be careful with mtime, atime, size */ 651355da1ebSSage Weil ceph_decode_timespec(&atime, &info->atime); 652355da1ebSSage Weil ceph_decode_timespec(&mtime, &info->mtime); 653355da1ebSSage Weil ceph_decode_timespec(&ctime, &info->ctime); 654355da1ebSSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 655355da1ebSSage Weil le32_to_cpu(info->truncate_seq), 656355da1ebSSage Weil le64_to_cpu(info->truncate_size), 657355da1ebSSage Weil le64_to_cpu(info->size)); 658355da1ebSSage Weil ceph_fill_file_time(inode, issued, 659355da1ebSSage Weil le32_to_cpu(info->time_warp_seq), 660355da1ebSSage Weil &ctime, &mtime, &atime); 661355da1ebSSage Weil 662355da1ebSSage Weil ci->i_layout = info->layout; 663355da1ebSSage Weil inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 664355da1ebSSage Weil 665355da1ebSSage Weil /* xattrs */ 666355da1ebSSage Weil /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 667355da1ebSSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 668355da1ebSSage Weil le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 669355da1ebSSage Weil if (ci->i_xattrs.blob) 670355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 671355da1ebSSage Weil ci->i_xattrs.blob = xattr_blob; 672355da1ebSSage Weil if (xattr_blob) 673355da1ebSSage Weil memcpy(ci->i_xattrs.blob->vec.iov_base, 674355da1ebSSage Weil iinfo->xattr_data, iinfo->xattr_len); 675355da1ebSSage Weil ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 6767221fe4cSGuangliang Zhao ceph_forget_all_cached_acls(inode); 677a6424e48SSage Weil xattr_blob = NULL; 678355da1ebSSage Weil } 679355da1ebSSage Weil 680355da1ebSSage Weil inode->i_mapping->a_ops = &ceph_aops; 681355da1ebSSage Weil inode->i_mapping->backing_dev_info = 682640ef79dSCheng Renquan &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 683355da1ebSSage Weil 684355da1ebSSage Weil switch (inode->i_mode & S_IFMT) { 685355da1ebSSage Weil case S_IFIFO: 686355da1ebSSage Weil case S_IFBLK: 687355da1ebSSage Weil case S_IFCHR: 688355da1ebSSage Weil case S_IFSOCK: 689355da1ebSSage Weil init_special_inode(inode, inode->i_mode, inode->i_rdev); 690355da1ebSSage Weil inode->i_op = &ceph_file_iops; 691355da1ebSSage Weil break; 692355da1ebSSage Weil case S_IFREG: 693355da1ebSSage Weil inode->i_op = &ceph_file_iops; 694355da1ebSSage Weil inode->i_fop = &ceph_file_fops; 695355da1ebSSage Weil break; 696355da1ebSSage Weil case S_IFLNK: 697355da1ebSSage Weil inode->i_op = &ceph_symlink_iops; 698355da1ebSSage Weil if (!ci->i_symlink) { 699810339ecSXi Wang u32 symlen = iinfo->symlink_len; 700355da1ebSSage Weil char *sym; 701355da1ebSSage Weil 702be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 703355da1ebSSage Weil 704810339ecSXi Wang err = -EINVAL; 705810339ecSXi Wang if (WARN_ON(symlen != inode->i_size)) 706810339ecSXi Wang goto out; 707810339ecSXi Wang 708355da1ebSSage Weil err = -ENOMEM; 709810339ecSXi Wang sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 710355da1ebSSage Weil if (!sym) 711355da1ebSSage Weil goto out; 712355da1ebSSage Weil 713be655596SSage Weil spin_lock(&ci->i_ceph_lock); 714355da1ebSSage Weil if (!ci->i_symlink) 715355da1ebSSage Weil ci->i_symlink = sym; 716355da1ebSSage Weil else 717355da1ebSSage Weil kfree(sym); /* lost a race */ 718355da1ebSSage Weil } 719355da1ebSSage Weil break; 720355da1ebSSage Weil case S_IFDIR: 721355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 722355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 723355da1ebSSage Weil 72414303d20SSage Weil ci->i_dir_layout = iinfo->dir_layout; 72514303d20SSage Weil 726355da1ebSSage Weil ci->i_files = le64_to_cpu(info->files); 727355da1ebSSage Weil ci->i_subdirs = le64_to_cpu(info->subdirs); 728355da1ebSSage Weil ci->i_rbytes = le64_to_cpu(info->rbytes); 729355da1ebSSage Weil ci->i_rfiles = le64_to_cpu(info->rfiles); 730355da1ebSSage Weil ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 731355da1ebSSage Weil ceph_decode_timespec(&ci->i_rctime, &info->rctime); 732355da1ebSSage Weil break; 733355da1ebSSage Weil default: 734355da1ebSSage Weil pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 735355da1ebSSage Weil ceph_vinop(inode), inode->i_mode); 736355da1ebSSage Weil } 737355da1ebSSage Weil 738a8673d61SYan, Zheng /* set dir completion flag? */ 739a8673d61SYan, Zheng if (S_ISDIR(inode->i_mode) && 740a8673d61SYan, Zheng ci->i_files == 0 && ci->i_subdirs == 0 && 741a8673d61SYan, Zheng ceph_snap(inode) == CEPH_NOSNAP && 742a8673d61SYan, Zheng (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 743a8673d61SYan, Zheng (issued & CEPH_CAP_FILE_EXCL) == 0 && 7442f276c51SYan, Zheng !__ceph_dir_is_complete(ci)) { 745a8673d61SYan, Zheng dout(" marking %p complete (empty)\n", inode); 7462f276c51SYan, Zheng __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count)); 747a8673d61SYan, Zheng ci->i_max_offset = 2; 748a8673d61SYan, Zheng } 749355da1ebSSage Weil no_change: 7508c93cd61SYan, Zheng /* only update max_size on auth cap */ 7518c93cd61SYan, Zheng if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 7528c93cd61SYan, Zheng ci->i_max_size != le64_to_cpu(info->max_size)) { 7538c93cd61SYan, Zheng dout("max_size %lld -> %llu\n", ci->i_max_size, 7548c93cd61SYan, Zheng le64_to_cpu(info->max_size)); 7558c93cd61SYan, Zheng ci->i_max_size = le64_to_cpu(info->max_size); 7568c93cd61SYan, Zheng } 7578c93cd61SYan, Zheng 758be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 759355da1ebSSage Weil 760355da1ebSSage Weil /* queue truncate if we saw i_size decrease */ 761355da1ebSSage Weil if (queue_trunc) 7623c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 763355da1ebSSage Weil 764355da1ebSSage Weil /* populate frag tree */ 765355da1ebSSage Weil /* FIXME: move me up, if/when version reflects fragtree changes */ 766355da1ebSSage Weil nsplits = le32_to_cpu(info->fragtree.nsplits); 767355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 76853e879a4SYan, Zheng rb_node = rb_first(&ci->i_fragtree); 769355da1ebSSage Weil for (i = 0; i < nsplits; i++) { 770355da1ebSSage Weil u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 77153e879a4SYan, Zheng frag = NULL; 77253e879a4SYan, Zheng while (rb_node) { 77353e879a4SYan, Zheng frag = rb_entry(rb_node, struct ceph_inode_frag, node); 77453e879a4SYan, Zheng if (ceph_frag_compare(frag->frag, id) >= 0) { 77553e879a4SYan, Zheng if (frag->frag != id) 77653e879a4SYan, Zheng frag = NULL; 77753e879a4SYan, Zheng else 77853e879a4SYan, Zheng rb_node = rb_next(rb_node); 77953e879a4SYan, Zheng break; 78053e879a4SYan, Zheng } 78153e879a4SYan, Zheng rb_node = rb_next(rb_node); 78253e879a4SYan, Zheng rb_erase(&frag->node, &ci->i_fragtree); 78353e879a4SYan, Zheng kfree(frag); 78453e879a4SYan, Zheng frag = NULL; 78553e879a4SYan, Zheng } 78653e879a4SYan, Zheng if (!frag) { 78753e879a4SYan, Zheng frag = __get_or_create_frag(ci, id); 788355da1ebSSage Weil if (IS_ERR(frag)) 789355da1ebSSage Weil continue; 79053e879a4SYan, Zheng } 791355da1ebSSage Weil frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 792355da1ebSSage Weil dout(" frag %x split by %d\n", frag->frag, frag->split_by); 793355da1ebSSage Weil } 79453e879a4SYan, Zheng while (rb_node) { 79553e879a4SYan, Zheng frag = rb_entry(rb_node, struct ceph_inode_frag, node); 79653e879a4SYan, Zheng rb_node = rb_next(rb_node); 79753e879a4SYan, Zheng rb_erase(&frag->node, &ci->i_fragtree); 79853e879a4SYan, Zheng kfree(frag); 79953e879a4SYan, Zheng } 800355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 801355da1ebSSage Weil 802355da1ebSSage Weil /* were we issued a capability? */ 803355da1ebSSage Weil if (info->cap.caps) { 804355da1ebSSage Weil if (ceph_snap(inode) == CEPH_NOSNAP) { 805355da1ebSSage Weil ceph_add_cap(inode, session, 806355da1ebSSage Weil le64_to_cpu(info->cap.cap_id), 807355da1ebSSage Weil cap_fmode, 808355da1ebSSage Weil le32_to_cpu(info->cap.caps), 809355da1ebSSage Weil le32_to_cpu(info->cap.wanted), 810355da1ebSSage Weil le32_to_cpu(info->cap.seq), 811355da1ebSSage Weil le32_to_cpu(info->cap.mseq), 812355da1ebSSage Weil le64_to_cpu(info->cap.realm), 813355da1ebSSage Weil info->cap.flags, 814355da1ebSSage Weil caps_reservation); 815355da1ebSSage Weil } else { 816be655596SSage Weil spin_lock(&ci->i_ceph_lock); 817355da1ebSSage Weil dout(" %p got snap_caps %s\n", inode, 818355da1ebSSage Weil ceph_cap_string(le32_to_cpu(info->cap.caps))); 819355da1ebSSage Weil ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 820355da1ebSSage Weil if (cap_fmode >= 0) 821355da1ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 822be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 823355da1ebSSage Weil } 82404d000ebSSage Weil } else if (cap_fmode >= 0) { 82504d000ebSSage Weil pr_warning("mds issued no caps on %llx.%llx\n", 82604d000ebSSage Weil ceph_vinop(inode)); 82704d000ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 828355da1ebSSage Weil } 829355da1ebSSage Weil 830355da1ebSSage Weil /* update delegation info? */ 831355da1ebSSage Weil if (dirinfo) 832355da1ebSSage Weil ceph_fill_dirfrag(inode, dirinfo); 833355da1ebSSage Weil 834355da1ebSSage Weil err = 0; 835355da1ebSSage Weil 836355da1ebSSage Weil out: 837b6c1d5b8SSage Weil if (xattr_blob) 838355da1ebSSage Weil ceph_buffer_put(xattr_blob); 839355da1ebSSage Weil return err; 840355da1ebSSage Weil } 841355da1ebSSage Weil 842355da1ebSSage Weil /* 843355da1ebSSage Weil * caller should hold session s_mutex. 844355da1ebSSage Weil */ 845355da1ebSSage Weil static void update_dentry_lease(struct dentry *dentry, 846355da1ebSSage Weil struct ceph_mds_reply_lease *lease, 847355da1ebSSage Weil struct ceph_mds_session *session, 848355da1ebSSage Weil unsigned long from_time) 849355da1ebSSage Weil { 850355da1ebSSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 851355da1ebSSage Weil long unsigned duration = le32_to_cpu(lease->duration_ms); 852355da1ebSSage Weil long unsigned ttl = from_time + (duration * HZ) / 1000; 853355da1ebSSage Weil long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 854355da1ebSSage Weil struct inode *dir; 855355da1ebSSage Weil 856355da1ebSSage Weil /* only track leases on regular dentries */ 857355da1ebSSage Weil if (dentry->d_op != &ceph_dentry_ops) 858355da1ebSSage Weil return; 859355da1ebSSage Weil 860355da1ebSSage Weil spin_lock(&dentry->d_lock); 8612f90b852SSage Weil dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 8622f90b852SSage Weil dentry, duration, ttl); 863355da1ebSSage Weil 864355da1ebSSage Weil /* make lease_rdcache_gen match directory */ 865355da1ebSSage Weil dir = dentry->d_parent->d_inode; 866355da1ebSSage Weil di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 867355da1ebSSage Weil 8682f90b852SSage Weil if (duration == 0) 869355da1ebSSage Weil goto out_unlock; 870355da1ebSSage Weil 871355da1ebSSage Weil if (di->lease_gen == session->s_cap_gen && 872355da1ebSSage Weil time_before(ttl, dentry->d_time)) 873355da1ebSSage Weil goto out_unlock; /* we already have a newer lease. */ 874355da1ebSSage Weil 875355da1ebSSage Weil if (di->lease_session && di->lease_session != session) 876355da1ebSSage Weil goto out_unlock; 877355da1ebSSage Weil 878355da1ebSSage Weil ceph_dentry_lru_touch(dentry); 879355da1ebSSage Weil 880355da1ebSSage Weil if (!di->lease_session) 881355da1ebSSage Weil di->lease_session = ceph_get_mds_session(session); 882355da1ebSSage Weil di->lease_gen = session->s_cap_gen; 883355da1ebSSage Weil di->lease_seq = le32_to_cpu(lease->seq); 884355da1ebSSage Weil di->lease_renew_after = half_ttl; 885355da1ebSSage Weil di->lease_renew_from = 0; 886355da1ebSSage Weil dentry->d_time = ttl; 887355da1ebSSage Weil out_unlock: 888355da1ebSSage Weil spin_unlock(&dentry->d_lock); 889355da1ebSSage Weil return; 890355da1ebSSage Weil } 891355da1ebSSage Weil 892355da1ebSSage Weil /* 8934baa75efSYehuda Sadeh * Set dentry's directory position based on the current dir's max, and 8944baa75efSYehuda Sadeh * order it in d_subdirs, so that dcache_readdir behaves. 8954f177264SSage Weil * 8964f177264SSage Weil * Always called under directory's i_mutex. 8974baa75efSYehuda Sadeh */ 8984baa75efSYehuda Sadeh static void ceph_set_dentry_offset(struct dentry *dn) 8994baa75efSYehuda Sadeh { 9004baa75efSYehuda Sadeh struct dentry *dir = dn->d_parent; 9014f177264SSage Weil struct inode *inode = dir->d_inode; 902b8cd952bSYehuda Sadeh struct ceph_inode_info *ci; 9034baa75efSYehuda Sadeh struct ceph_dentry_info *di; 9044baa75efSYehuda Sadeh 9054baa75efSYehuda Sadeh BUG_ON(!inode); 9064baa75efSYehuda Sadeh 907b8cd952bSYehuda Sadeh ci = ceph_inode(inode); 9084baa75efSYehuda Sadeh di = ceph_dentry(dn); 9094baa75efSYehuda Sadeh 910be655596SSage Weil spin_lock(&ci->i_ceph_lock); 9112f276c51SYan, Zheng if (!__ceph_dir_is_complete(ci)) { 912be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 913e8a74987SSage Weil return; 914e8a74987SSage Weil } 9154baa75efSYehuda Sadeh di->offset = ceph_inode(inode)->i_max_offset++; 916be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 9174baa75efSYehuda Sadeh 9182fd6b7f5SNick Piggin spin_lock(&dir->d_lock); 9192fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 92013a4214cSHenry C Chang list_move(&dn->d_u.d_child, &dir->d_subdirs); 9214baa75efSYehuda Sadeh dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 9224baa75efSYehuda Sadeh dn->d_u.d_child.prev, dn->d_u.d_child.next); 9234baa75efSYehuda Sadeh spin_unlock(&dn->d_lock); 9242fd6b7f5SNick Piggin spin_unlock(&dir->d_lock); 9254baa75efSYehuda Sadeh } 9264baa75efSYehuda Sadeh 9274baa75efSYehuda Sadeh /* 9281cd3935bSSage Weil * splice a dentry to an inode. 9291cd3935bSSage Weil * caller must hold directory i_mutex for this to be safe. 9301cd3935bSSage Weil * 9311cd3935bSSage Weil * we will only rehash the resulting dentry if @prehash is 9321cd3935bSSage Weil * true; @prehash will be set to false (for the benefit of 9331cd3935bSSage Weil * the caller) if we fail. 9341cd3935bSSage Weil */ 9351cd3935bSSage Weil static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 936467c5251SSage Weil bool *prehash, bool set_offset) 9371cd3935bSSage Weil { 9381cd3935bSSage Weil struct dentry *realdn; 9391cd3935bSSage Weil 9401cd3935bSSage Weil BUG_ON(dn->d_inode); 9411cd3935bSSage Weil 9421cd3935bSSage Weil /* dn must be unhashed */ 9431cd3935bSSage Weil if (!d_unhashed(dn)) 9441cd3935bSSage Weil d_drop(dn); 9451cd3935bSSage Weil realdn = d_materialise_unique(dn, in); 9461cd3935bSSage Weil if (IS_ERR(realdn)) { 947d69ed05aSSage Weil pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 948d69ed05aSSage Weil PTR_ERR(realdn), dn, in, ceph_vinop(in)); 9491cd3935bSSage Weil if (prehash) 9501cd3935bSSage Weil *prehash = false; /* don't rehash on error */ 9511cd3935bSSage Weil dn = realdn; /* note realdn contains the error */ 9521cd3935bSSage Weil goto out; 9531cd3935bSSage Weil } else if (realdn) { 9541cd3935bSSage Weil dout("dn %p (%d) spliced with %p (%d) " 9551cd3935bSSage Weil "inode %p ino %llx.%llx\n", 95684d08fa8SAl Viro dn, d_count(dn), 95784d08fa8SAl Viro realdn, d_count(realdn), 9581cd3935bSSage Weil realdn->d_inode, ceph_vinop(realdn->d_inode)); 9591cd3935bSSage Weil dput(dn); 9601cd3935bSSage Weil dn = realdn; 9611cd3935bSSage Weil } else { 9621cd3935bSSage Weil BUG_ON(!ceph_dentry(dn)); 9631cd3935bSSage Weil dout("dn %p attached to %p ino %llx.%llx\n", 9641cd3935bSSage Weil dn, dn->d_inode, ceph_vinop(dn->d_inode)); 9651cd3935bSSage Weil } 9661cd3935bSSage Weil if ((!prehash || *prehash) && d_unhashed(dn)) 9671cd3935bSSage Weil d_rehash(dn); 968467c5251SSage Weil if (set_offset) 9691cd3935bSSage Weil ceph_set_dentry_offset(dn); 9701cd3935bSSage Weil out: 9711cd3935bSSage Weil return dn; 9721cd3935bSSage Weil } 9731cd3935bSSage Weil 9741cd3935bSSage Weil /* 975355da1ebSSage Weil * Incorporate results into the local cache. This is either just 976355da1ebSSage Weil * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 977355da1ebSSage Weil * after a lookup). 978355da1ebSSage Weil * 979355da1ebSSage Weil * A reply may contain 980355da1ebSSage Weil * a directory inode along with a dentry. 981355da1ebSSage Weil * and/or a target inode 982355da1ebSSage Weil * 983355da1ebSSage Weil * Called with snap_rwsem (read). 984355da1ebSSage Weil */ 985355da1ebSSage Weil int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 986355da1ebSSage Weil struct ceph_mds_session *session) 987355da1ebSSage Weil { 988355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 989355da1ebSSage Weil struct inode *in = NULL; 990355da1ebSSage Weil struct ceph_mds_reply_inode *ininfo; 991355da1ebSSage Weil struct ceph_vino vino; 9923d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 993355da1ebSSage Weil int err = 0; 994355da1ebSSage Weil 995355da1ebSSage Weil dout("fill_trace %p is_dentry %d is_target %d\n", req, 996355da1ebSSage Weil rinfo->head->is_dentry, rinfo->head->is_target); 997355da1ebSSage Weil 998355da1ebSSage Weil #if 0 999355da1ebSSage Weil /* 1000355da1ebSSage Weil * Debugging hook: 1001355da1ebSSage Weil * 1002355da1ebSSage Weil * If we resend completed ops to a recovering mds, we get no 1003355da1ebSSage Weil * trace. Since that is very rare, pretend this is the case 1004355da1ebSSage Weil * to ensure the 'no trace' handlers in the callers behave. 1005355da1ebSSage Weil * 1006355da1ebSSage Weil * Fill in inodes unconditionally to avoid breaking cap 1007355da1ebSSage Weil * invariants. 1008355da1ebSSage Weil */ 1009355da1ebSSage Weil if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 1010355da1ebSSage Weil pr_info("fill_trace faking empty trace on %lld %s\n", 1011355da1ebSSage Weil req->r_tid, ceph_mds_op_name(rinfo->head->op)); 1012355da1ebSSage Weil if (rinfo->head->is_dentry) { 1013355da1ebSSage Weil rinfo->head->is_dentry = 0; 1014355da1ebSSage Weil err = fill_inode(req->r_locked_dir, 1015355da1ebSSage Weil &rinfo->diri, rinfo->dirfrag, 1016355da1ebSSage Weil session, req->r_request_started, -1); 1017355da1ebSSage Weil } 1018355da1ebSSage Weil if (rinfo->head->is_target) { 1019355da1ebSSage Weil rinfo->head->is_target = 0; 1020355da1ebSSage Weil ininfo = rinfo->targeti.in; 1021355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1022355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1023355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1024355da1ebSSage Weil err = fill_inode(in, &rinfo->targeti, NULL, 1025355da1ebSSage Weil session, req->r_request_started, 1026355da1ebSSage Weil req->r_fmode); 1027355da1ebSSage Weil iput(in); 1028355da1ebSSage Weil } 1029355da1ebSSage Weil } 1030355da1ebSSage Weil #endif 1031355da1ebSSage Weil 1032355da1ebSSage Weil if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1033355da1ebSSage Weil dout("fill_trace reply is empty!\n"); 1034167c9e35SSage Weil if (rinfo->head->result == 0 && req->r_locked_dir) 1035167c9e35SSage Weil ceph_invalidate_dir_request(req); 1036355da1ebSSage Weil return 0; 1037355da1ebSSage Weil } 1038355da1ebSSage Weil 1039355da1ebSSage Weil if (rinfo->head->is_dentry) { 10405b1daecdSSage Weil struct inode *dir = req->r_locked_dir; 10415b1daecdSSage Weil 10426c5e50faSSage Weil if (dir) { 10435b1daecdSSage Weil err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 10445b1daecdSSage Weil session, req->r_request_started, -1, 10455b1daecdSSage Weil &req->r_caps_reservation); 10465b1daecdSSage Weil if (err < 0) 104719913b4eSYan, Zheng goto done; 10486c5e50faSSage Weil } else { 10496c5e50faSSage Weil WARN_ON_ONCE(1); 10506c5e50faSSage Weil } 105119913b4eSYan, Zheng 105219913b4eSYan, Zheng if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { 105319913b4eSYan, Zheng struct qstr dname; 105419913b4eSYan, Zheng struct dentry *dn, *parent; 105519913b4eSYan, Zheng 105619913b4eSYan, Zheng BUG_ON(!rinfo->head->is_target); 105719913b4eSYan, Zheng BUG_ON(req->r_dentry); 105819913b4eSYan, Zheng 105919913b4eSYan, Zheng parent = d_find_any_alias(dir); 106019913b4eSYan, Zheng BUG_ON(!parent); 106119913b4eSYan, Zheng 106219913b4eSYan, Zheng dname.name = rinfo->dname; 106319913b4eSYan, Zheng dname.len = rinfo->dname_len; 106419913b4eSYan, Zheng dname.hash = full_name_hash(dname.name, dname.len); 106519913b4eSYan, Zheng vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 106619913b4eSYan, Zheng vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 106719913b4eSYan, Zheng retry_lookup: 106819913b4eSYan, Zheng dn = d_lookup(parent, &dname); 106919913b4eSYan, Zheng dout("d_lookup on parent=%p name=%.*s got %p\n", 107019913b4eSYan, Zheng parent, dname.len, dname.name, dn); 107119913b4eSYan, Zheng 107219913b4eSYan, Zheng if (!dn) { 107319913b4eSYan, Zheng dn = d_alloc(parent, &dname); 107419913b4eSYan, Zheng dout("d_alloc %p '%.*s' = %p\n", parent, 107519913b4eSYan, Zheng dname.len, dname.name, dn); 107619913b4eSYan, Zheng if (dn == NULL) { 107719913b4eSYan, Zheng dput(parent); 107819913b4eSYan, Zheng err = -ENOMEM; 107919913b4eSYan, Zheng goto done; 108019913b4eSYan, Zheng } 108119913b4eSYan, Zheng err = ceph_init_dentry(dn); 108219913b4eSYan, Zheng if (err < 0) { 108319913b4eSYan, Zheng dput(dn); 108419913b4eSYan, Zheng dput(parent); 108519913b4eSYan, Zheng goto done; 108619913b4eSYan, Zheng } 108719913b4eSYan, Zheng } else if (dn->d_inode && 108819913b4eSYan, Zheng (ceph_ino(dn->d_inode) != vino.ino || 108919913b4eSYan, Zheng ceph_snap(dn->d_inode) != vino.snap)) { 109019913b4eSYan, Zheng dout(" dn %p points to wrong inode %p\n", 109119913b4eSYan, Zheng dn, dn->d_inode); 109219913b4eSYan, Zheng d_delete(dn); 109319913b4eSYan, Zheng dput(dn); 109419913b4eSYan, Zheng goto retry_lookup; 109519913b4eSYan, Zheng } 109619913b4eSYan, Zheng 109719913b4eSYan, Zheng req->r_dentry = dn; 109819913b4eSYan, Zheng dput(parent); 109919913b4eSYan, Zheng } 11005b1daecdSSage Weil } 11015b1daecdSSage Weil 110286b58d13SYan, Zheng if (rinfo->head->is_target) { 110386b58d13SYan, Zheng vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 110486b58d13SYan, Zheng vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 110586b58d13SYan, Zheng 110686b58d13SYan, Zheng in = ceph_get_inode(sb, vino); 110786b58d13SYan, Zheng if (IS_ERR(in)) { 110886b58d13SYan, Zheng err = PTR_ERR(in); 110986b58d13SYan, Zheng goto done; 111086b58d13SYan, Zheng } 111186b58d13SYan, Zheng req->r_target_inode = in; 111286b58d13SYan, Zheng 111386b58d13SYan, Zheng err = fill_inode(in, &rinfo->targeti, NULL, 111486b58d13SYan, Zheng session, req->r_request_started, 111586b58d13SYan, Zheng (le32_to_cpu(rinfo->head->result) == 0) ? 111686b58d13SYan, Zheng req->r_fmode : -1, 111786b58d13SYan, Zheng &req->r_caps_reservation); 111886b58d13SYan, Zheng if (err < 0) { 111986b58d13SYan, Zheng pr_err("fill_inode badness %p %llx.%llx\n", 112086b58d13SYan, Zheng in, ceph_vinop(in)); 112186b58d13SYan, Zheng goto done; 112286b58d13SYan, Zheng } 112386b58d13SYan, Zheng } 112486b58d13SYan, Zheng 11259358c6d4SSage Weil /* 11269358c6d4SSage Weil * ignore null lease/binding on snapdir ENOENT, or else we 11279358c6d4SSage Weil * will have trouble splicing in the virtual snapdir later 11289358c6d4SSage Weil */ 11299358c6d4SSage Weil if (rinfo->head->is_dentry && !req->r_aborted && 11306c5e50faSSage Weil req->r_locked_dir && 11319358c6d4SSage Weil (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 11323d14c5d2SYehuda Sadeh fsc->mount_options->snapdir_name, 11339358c6d4SSage Weil req->r_dentry->d_name.len))) { 1134355da1ebSSage Weil /* 1135355da1ebSSage Weil * lookup link rename : null -> possibly existing inode 1136355da1ebSSage Weil * mknod symlink mkdir : null -> new inode 1137355da1ebSSage Weil * unlink : linked -> null 1138355da1ebSSage Weil */ 1139355da1ebSSage Weil struct inode *dir = req->r_locked_dir; 1140355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1141355da1ebSSage Weil bool have_dir_cap, have_lease; 1142355da1ebSSage Weil 1143355da1ebSSage Weil BUG_ON(!dn); 1144355da1ebSSage Weil BUG_ON(!dir); 1145355da1ebSSage Weil BUG_ON(dn->d_parent->d_inode != dir); 1146355da1ebSSage Weil BUG_ON(ceph_ino(dir) != 1147355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->ino)); 1148355da1ebSSage Weil BUG_ON(ceph_snap(dir) != 1149355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->snapid)); 1150355da1ebSSage Weil 1151355da1ebSSage Weil /* do we have a lease on the whole dir? */ 1152355da1ebSSage Weil have_dir_cap = 1153355da1ebSSage Weil (le32_to_cpu(rinfo->diri.in->cap.caps) & 1154355da1ebSSage Weil CEPH_CAP_FILE_SHARED); 1155355da1ebSSage Weil 1156355da1ebSSage Weil /* do we have a dn lease? */ 1157355da1ebSSage Weil have_lease = have_dir_cap || 11582f90b852SSage Weil le32_to_cpu(rinfo->dlease->duration_ms); 1159355da1ebSSage Weil if (!have_lease) 1160355da1ebSSage Weil dout("fill_trace no dentry lease or dir cap\n"); 1161355da1ebSSage Weil 1162355da1ebSSage Weil /* rename? */ 1163355da1ebSSage Weil if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1164355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1165355da1ebSSage Weil req->r_old_dentry, 1166355da1ebSSage Weil req->r_old_dentry->d_name.len, 1167355da1ebSSage Weil req->r_old_dentry->d_name.name, 1168355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 1169355da1ebSSage Weil dout("fill_trace doing d_move %p -> %p\n", 1170355da1ebSSage Weil req->r_old_dentry, dn); 1171c10f5e12SSage Weil 1172355da1ebSSage Weil d_move(req->r_old_dentry, dn); 1173355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1174355da1ebSSage Weil req->r_old_dentry, 1175355da1ebSSage Weil req->r_old_dentry->d_name.len, 1176355da1ebSSage Weil req->r_old_dentry->d_name.name, 1177355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 117881a6cf2dSSage Weil 1179c4a29f26SSage Weil /* ensure target dentry is invalidated, despite 1180c4a29f26SSage Weil rehashing bug in vfs_rename_dir */ 118181a6cf2dSSage Weil ceph_invalidate_dentry_lease(dn); 118281a6cf2dSSage Weil 118309adc80cSSage Weil /* 118409adc80cSSage Weil * d_move() puts the renamed dentry at the end of 118509adc80cSSage Weil * d_subdirs. We need to assign it an appropriate 11862f276c51SYan, Zheng * directory offset so we can behave when dir is 11872f276c51SYan, Zheng * complete. 118809adc80cSSage Weil */ 118909adc80cSSage Weil ceph_set_dentry_offset(req->r_old_dentry); 119009adc80cSSage Weil dout("dn %p gets new offset %lld\n", req->r_old_dentry, 11911cd3935bSSage Weil ceph_dentry(req->r_old_dentry)->offset); 119281a6cf2dSSage Weil 1193355da1ebSSage Weil dn = req->r_old_dentry; /* use old_dentry */ 1194355da1ebSSage Weil } 1195355da1ebSSage Weil 1196355da1ebSSage Weil /* null dentry? */ 1197355da1ebSSage Weil if (!rinfo->head->is_target) { 1198355da1ebSSage Weil dout("fill_trace null dentry\n"); 1199355da1ebSSage Weil if (dn->d_inode) { 1200355da1ebSSage Weil dout("d_delete %p\n", dn); 1201355da1ebSSage Weil d_delete(dn); 1202355da1ebSSage Weil } else { 1203355da1ebSSage Weil dout("d_instantiate %p NULL\n", dn); 1204355da1ebSSage Weil d_instantiate(dn, NULL); 1205355da1ebSSage Weil if (have_lease && d_unhashed(dn)) 1206355da1ebSSage Weil d_rehash(dn); 1207355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, 1208355da1ebSSage Weil session, 1209355da1ebSSage Weil req->r_request_started); 1210355da1ebSSage Weil } 1211355da1ebSSage Weil goto done; 1212355da1ebSSage Weil } 1213355da1ebSSage Weil 1214355da1ebSSage Weil /* attach proper inode */ 121586b58d13SYan, Zheng if (!dn->d_inode) { 121686b58d13SYan, Zheng ihold(in); 1217467c5251SSage Weil dn = splice_dentry(dn, in, &have_lease, true); 1218355da1ebSSage Weil if (IS_ERR(dn)) { 1219355da1ebSSage Weil err = PTR_ERR(dn); 1220355da1ebSSage Weil goto done; 1221355da1ebSSage Weil } 1222355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 122386b58d13SYan, Zheng } else if (dn->d_inode && dn->d_inode != in) { 1224355da1ebSSage Weil dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 122586b58d13SYan, Zheng dn, dn->d_inode, ceph_vinop(dn->d_inode), 122686b58d13SYan, Zheng ceph_vinop(in)); 1227355da1ebSSage Weil have_lease = false; 1228355da1ebSSage Weil } 1229355da1ebSSage Weil 1230355da1ebSSage Weil if (have_lease) 1231355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, session, 1232355da1ebSSage Weil req->r_request_started); 1233355da1ebSSage Weil dout(" final dn %p\n", dn); 123486b58d13SYan, Zheng } else if (!req->r_aborted && 123586b58d13SYan, Zheng (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 123686b58d13SYan, Zheng req->r_op == CEPH_MDS_OP_MKSNAP)) { 1237355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1238355da1ebSSage Weil 1239355da1ebSSage Weil /* fill out a snapdir LOOKUPSNAP dentry */ 1240355da1ebSSage Weil BUG_ON(!dn); 1241355da1ebSSage Weil BUG_ON(!req->r_locked_dir); 1242355da1ebSSage Weil BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1243355da1ebSSage Weil ininfo = rinfo->targeti.in; 1244355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1245355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1246355da1ebSSage Weil dout(" linking snapped dir %p to dn %p\n", in, dn); 124786b58d13SYan, Zheng ihold(in); 1248467c5251SSage Weil dn = splice_dentry(dn, in, NULL, true); 1249355da1ebSSage Weil if (IS_ERR(dn)) { 1250355da1ebSSage Weil err = PTR_ERR(dn); 1251355da1ebSSage Weil goto done; 1252355da1ebSSage Weil } 1253355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 1254355da1ebSSage Weil } 1255355da1ebSSage Weil done: 1256355da1ebSSage Weil dout("fill_trace done err=%d\n", err); 1257355da1ebSSage Weil return err; 1258355da1ebSSage Weil } 1259355da1ebSSage Weil 1260355da1ebSSage Weil /* 1261355da1ebSSage Weil * Prepopulate our cache with readdir results, leases, etc. 1262355da1ebSSage Weil */ 126379f9f99aSSage Weil static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 126479f9f99aSSage Weil struct ceph_mds_session *session) 126579f9f99aSSage Weil { 126679f9f99aSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 126779f9f99aSSage Weil int i, err = 0; 126879f9f99aSSage Weil 126979f9f99aSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 127079f9f99aSSage Weil struct ceph_vino vino; 127179f9f99aSSage Weil struct inode *in; 127279f9f99aSSage Weil int rc; 127379f9f99aSSage Weil 127479f9f99aSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 127579f9f99aSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 127679f9f99aSSage Weil 127779f9f99aSSage Weil in = ceph_get_inode(req->r_dentry->d_sb, vino); 127879f9f99aSSage Weil if (IS_ERR(in)) { 127979f9f99aSSage Weil err = PTR_ERR(in); 128079f9f99aSSage Weil dout("new_inode badness got %d\n", err); 128179f9f99aSSage Weil continue; 128279f9f99aSSage Weil } 128379f9f99aSSage Weil rc = fill_inode(in, &rinfo->dir_in[i], NULL, session, 128479f9f99aSSage Weil req->r_request_started, -1, 128579f9f99aSSage Weil &req->r_caps_reservation); 128679f9f99aSSage Weil if (rc < 0) { 128779f9f99aSSage Weil pr_err("fill_inode badness on %p got %d\n", in, rc); 128879f9f99aSSage Weil err = rc; 128979f9f99aSSage Weil continue; 129079f9f99aSSage Weil } 129179f9f99aSSage Weil } 129279f9f99aSSage Weil 129379f9f99aSSage Weil return err; 129479f9f99aSSage Weil } 129579f9f99aSSage Weil 1296355da1ebSSage Weil int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1297355da1ebSSage Weil struct ceph_mds_session *session) 1298355da1ebSSage Weil { 1299355da1ebSSage Weil struct dentry *parent = req->r_dentry; 1300355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1301355da1ebSSage Weil struct qstr dname; 1302355da1ebSSage Weil struct dentry *dn; 1303355da1ebSSage Weil struct inode *in; 130486b58d13SYan, Zheng int err = 0, ret, i; 1305355da1ebSSage Weil struct inode *snapdir = NULL; 1306355da1ebSSage Weil struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1307355da1ebSSage Weil struct ceph_dentry_info *di; 130881c6aea5SYan, Zheng u64 r_readdir_offset = req->r_readdir_offset; 130981c6aea5SYan, Zheng u32 frag = le32_to_cpu(rhead->args.readdir.frag); 131081c6aea5SYan, Zheng 131181c6aea5SYan, Zheng if (rinfo->dir_dir && 131281c6aea5SYan, Zheng le32_to_cpu(rinfo->dir_dir->frag) != frag) { 131381c6aea5SYan, Zheng dout("readdir_prepopulate got new frag %x -> %x\n", 131481c6aea5SYan, Zheng frag, le32_to_cpu(rinfo->dir_dir->frag)); 131581c6aea5SYan, Zheng frag = le32_to_cpu(rinfo->dir_dir->frag); 131681c6aea5SYan, Zheng if (ceph_frag_is_leftmost(frag)) 131781c6aea5SYan, Zheng r_readdir_offset = 2; 131881c6aea5SYan, Zheng else 131981c6aea5SYan, Zheng r_readdir_offset = 0; 132081c6aea5SYan, Zheng } 1321355da1ebSSage Weil 132279f9f99aSSage Weil if (req->r_aborted) 132379f9f99aSSage Weil return readdir_prepopulate_inodes_only(req, session); 132479f9f99aSSage Weil 1325355da1ebSSage Weil if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1326355da1ebSSage Weil snapdir = ceph_get_snapdir(parent->d_inode); 1327355da1ebSSage Weil parent = d_find_alias(snapdir); 1328355da1ebSSage Weil dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1329355da1ebSSage Weil rinfo->dir_nr, parent); 1330355da1ebSSage Weil } else { 1331355da1ebSSage Weil dout("readdir_prepopulate %d items under dn %p\n", 1332355da1ebSSage Weil rinfo->dir_nr, parent); 1333355da1ebSSage Weil if (rinfo->dir_dir) 1334355da1ebSSage Weil ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1335355da1ebSSage Weil } 1336355da1ebSSage Weil 133786b58d13SYan, Zheng /* FIXME: release caps/leases if error occurs */ 1338355da1ebSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 1339355da1ebSSage Weil struct ceph_vino vino; 1340355da1ebSSage Weil 1341355da1ebSSage Weil dname.name = rinfo->dir_dname[i]; 1342355da1ebSSage Weil dname.len = rinfo->dir_dname_len[i]; 1343355da1ebSSage Weil dname.hash = full_name_hash(dname.name, dname.len); 1344355da1ebSSage Weil 1345355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1346355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1347355da1ebSSage Weil 1348355da1ebSSage Weil retry_lookup: 1349355da1ebSSage Weil dn = d_lookup(parent, &dname); 1350355da1ebSSage Weil dout("d_lookup on parent=%p name=%.*s got %p\n", 1351355da1ebSSage Weil parent, dname.len, dname.name, dn); 1352355da1ebSSage Weil 1353355da1ebSSage Weil if (!dn) { 1354355da1ebSSage Weil dn = d_alloc(parent, &dname); 1355355da1ebSSage Weil dout("d_alloc %p '%.*s' = %p\n", parent, 1356355da1ebSSage Weil dname.len, dname.name, dn); 1357355da1ebSSage Weil if (dn == NULL) { 1358355da1ebSSage Weil dout("d_alloc badness\n"); 1359355da1ebSSage Weil err = -ENOMEM; 1360355da1ebSSage Weil goto out; 1361355da1ebSSage Weil } 136286b58d13SYan, Zheng ret = ceph_init_dentry(dn); 136386b58d13SYan, Zheng if (ret < 0) { 13648c696737SSage Weil dput(dn); 136586b58d13SYan, Zheng err = ret; 1366355da1ebSSage Weil goto out; 13678c696737SSage Weil } 1368355da1ebSSage Weil } else if (dn->d_inode && 1369355da1ebSSage Weil (ceph_ino(dn->d_inode) != vino.ino || 1370355da1ebSSage Weil ceph_snap(dn->d_inode) != vino.snap)) { 1371355da1ebSSage Weil dout(" dn %p points to wrong inode %p\n", 1372355da1ebSSage Weil dn, dn->d_inode); 1373355da1ebSSage Weil d_delete(dn); 1374355da1ebSSage Weil dput(dn); 1375355da1ebSSage Weil goto retry_lookup; 1376355da1ebSSage Weil } else { 1377355da1ebSSage Weil /* reorder parent's d_subdirs */ 13782fd6b7f5SNick Piggin spin_lock(&parent->d_lock); 13792fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1380355da1ebSSage Weil list_move(&dn->d_u.d_child, &parent->d_subdirs); 1381355da1ebSSage Weil spin_unlock(&dn->d_lock); 13822fd6b7f5SNick Piggin spin_unlock(&parent->d_lock); 1383355da1ebSSage Weil } 1384355da1ebSSage Weil 1385355da1ebSSage Weil /* inode */ 1386355da1ebSSage Weil if (dn->d_inode) { 1387355da1ebSSage Weil in = dn->d_inode; 1388355da1ebSSage Weil } else { 1389355da1ebSSage Weil in = ceph_get_inode(parent->d_sb, vino); 1390ac1f12efSDan Carpenter if (IS_ERR(in)) { 1391355da1ebSSage Weil dout("new_inode badness\n"); 13922744c171SAl Viro d_drop(dn); 1393355da1ebSSage Weil dput(dn); 1394ac1f12efSDan Carpenter err = PTR_ERR(in); 1395355da1ebSSage Weil goto out; 1396355da1ebSSage Weil } 1397355da1ebSSage Weil } 1398355da1ebSSage Weil 1399355da1ebSSage Weil if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1400355da1ebSSage Weil req->r_request_started, -1, 1401355da1ebSSage Weil &req->r_caps_reservation) < 0) { 1402355da1ebSSage Weil pr_err("fill_inode badness on %p\n", in); 140386b58d13SYan, Zheng if (!dn->d_inode) 140486b58d13SYan, Zheng iput(in); 140586b58d13SYan, Zheng d_drop(dn); 1406d69ed05aSSage Weil goto next_item; 1407355da1ebSSage Weil } 140886b58d13SYan, Zheng 140986b58d13SYan, Zheng if (!dn->d_inode) { 141086b58d13SYan, Zheng dn = splice_dentry(dn, in, NULL, false); 141186b58d13SYan, Zheng if (IS_ERR(dn)) { 141286b58d13SYan, Zheng err = PTR_ERR(dn); 141386b58d13SYan, Zheng dn = NULL; 141486b58d13SYan, Zheng goto next_item; 141586b58d13SYan, Zheng } 141686b58d13SYan, Zheng } 141786b58d13SYan, Zheng 141886b58d13SYan, Zheng di = dn->d_fsdata; 141986b58d13SYan, Zheng di->offset = ceph_make_fpos(frag, i + r_readdir_offset); 142086b58d13SYan, Zheng 1421355da1ebSSage Weil update_dentry_lease(dn, rinfo->dir_dlease[i], 1422d69ed05aSSage Weil req->r_session, 1423d69ed05aSSage Weil req->r_request_started); 1424d69ed05aSSage Weil next_item: 1425d69ed05aSSage Weil if (dn) 1426355da1ebSSage Weil dput(dn); 1427355da1ebSSage Weil } 142886b58d13SYan, Zheng if (err == 0) 1429355da1ebSSage Weil req->r_did_prepopulate = true; 1430355da1ebSSage Weil 1431355da1ebSSage Weil out: 1432355da1ebSSage Weil if (snapdir) { 1433355da1ebSSage Weil iput(snapdir); 1434355da1ebSSage Weil dput(parent); 1435355da1ebSSage Weil } 1436355da1ebSSage Weil dout("readdir_prepopulate done\n"); 1437355da1ebSSage Weil return err; 1438355da1ebSSage Weil } 1439355da1ebSSage Weil 1440355da1ebSSage Weil int ceph_inode_set_size(struct inode *inode, loff_t size) 1441355da1ebSSage Weil { 1442355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1443355da1ebSSage Weil int ret = 0; 1444355da1ebSSage Weil 1445be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1446355da1ebSSage Weil dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1447355da1ebSSage Weil inode->i_size = size; 1448355da1ebSSage Weil inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1449355da1ebSSage Weil 1450355da1ebSSage Weil /* tell the MDS if we are approaching max_size */ 1451355da1ebSSage Weil if ((size << 1) >= ci->i_max_size && 1452355da1ebSSage Weil (ci->i_reported_size << 1) < ci->i_max_size) 1453355da1ebSSage Weil ret = 1; 1454355da1ebSSage Weil 1455be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1456355da1ebSSage Weil return ret; 1457355da1ebSSage Weil } 1458355da1ebSSage Weil 1459355da1ebSSage Weil /* 1460355da1ebSSage Weil * Write back inode data in a worker thread. (This can't be done 1461355da1ebSSage Weil * in the message handler context.) 1462355da1ebSSage Weil */ 14633c6f6b79SSage Weil void ceph_queue_writeback(struct inode *inode) 14643c6f6b79SSage Weil { 146515a2015fSSage Weil ihold(inode); 14663c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->wb_wq, 14673c6f6b79SSage Weil &ceph_inode(inode)->i_wb_work)) { 14682c27c9a5SSage Weil dout("ceph_queue_writeback %p\n", inode); 14693c6f6b79SSage Weil } else { 14702c27c9a5SSage Weil dout("ceph_queue_writeback %p failed\n", inode); 147115a2015fSSage Weil iput(inode); 14723c6f6b79SSage Weil } 14733c6f6b79SSage Weil } 14743c6f6b79SSage Weil 14753c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work) 1476355da1ebSSage Weil { 1477355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1478355da1ebSSage Weil i_wb_work); 1479355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1480355da1ebSSage Weil 1481355da1ebSSage Weil dout("writeback %p\n", inode); 1482355da1ebSSage Weil filemap_fdatawrite(&inode->i_data); 1483355da1ebSSage Weil iput(inode); 1484355da1ebSSage Weil } 1485355da1ebSSage Weil 1486355da1ebSSage Weil /* 14873c6f6b79SSage Weil * queue an async invalidation 14883c6f6b79SSage Weil */ 14893c6f6b79SSage Weil void ceph_queue_invalidate(struct inode *inode) 14903c6f6b79SSage Weil { 149115a2015fSSage Weil ihold(inode); 14923c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 14933c6f6b79SSage Weil &ceph_inode(inode)->i_pg_inv_work)) { 14943c6f6b79SSage Weil dout("ceph_queue_invalidate %p\n", inode); 14953c6f6b79SSage Weil } else { 14963c6f6b79SSage Weil dout("ceph_queue_invalidate %p failed\n", inode); 149715a2015fSSage Weil iput(inode); 14983c6f6b79SSage Weil } 14993c6f6b79SSage Weil } 15003c6f6b79SSage Weil 15013c6f6b79SSage Weil /* 1502355da1ebSSage Weil * Invalidate inode pages in a worker thread. (This can't be done 1503355da1ebSSage Weil * in the message handler context.) 1504355da1ebSSage Weil */ 15053c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work) 1506355da1ebSSage Weil { 1507355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1508355da1ebSSage Weil i_pg_inv_work); 1509355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1510355da1ebSSage Weil u32 orig_gen; 1511355da1ebSSage Weil int check = 0; 1512355da1ebSSage Weil 1513b0d7c223SYan, Zheng mutex_lock(&ci->i_truncate_mutex); 1514be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1515355da1ebSSage Weil dout("invalidate_pages %p gen %d revoking %d\n", inode, 1516355da1ebSSage Weil ci->i_rdcache_gen, ci->i_rdcache_revoking); 1517cd045cb4SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 15189563f88cSYan, Zheng if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 15199563f88cSYan, Zheng check = 1; 1520be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1521b0d7c223SYan, Zheng mutex_unlock(&ci->i_truncate_mutex); 1522355da1ebSSage Weil goto out; 1523355da1ebSSage Weil } 1524355da1ebSSage Weil orig_gen = ci->i_rdcache_gen; 1525be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1526355da1ebSSage Weil 1527b0d7c223SYan, Zheng truncate_inode_pages(inode->i_mapping, 0); 1528355da1ebSSage Weil 1529be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1530cd045cb4SSage Weil if (orig_gen == ci->i_rdcache_gen && 1531cd045cb4SSage Weil orig_gen == ci->i_rdcache_revoking) { 1532355da1ebSSage Weil dout("invalidate_pages %p gen %d successful\n", inode, 1533355da1ebSSage Weil ci->i_rdcache_gen); 1534cd045cb4SSage Weil ci->i_rdcache_revoking--; 1535355da1ebSSage Weil check = 1; 1536355da1ebSSage Weil } else { 1537cd045cb4SSage Weil dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1538cd045cb4SSage Weil inode, orig_gen, ci->i_rdcache_gen, 1539cd045cb4SSage Weil ci->i_rdcache_revoking); 15409563f88cSYan, Zheng if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 15419563f88cSYan, Zheng check = 1; 1542355da1ebSSage Weil } 1543be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1544b0d7c223SYan, Zheng mutex_unlock(&ci->i_truncate_mutex); 15459563f88cSYan, Zheng out: 1546355da1ebSSage Weil if (check) 1547355da1ebSSage Weil ceph_check_caps(ci, 0, NULL); 1548355da1ebSSage Weil iput(inode); 1549355da1ebSSage Weil } 1550355da1ebSSage Weil 1551355da1ebSSage Weil 1552355da1ebSSage Weil /* 15533f99969fSYan, Zheng * called by trunc_wq; 1554355da1ebSSage Weil * 1555355da1ebSSage Weil * We also truncate in a separate thread as well. 1556355da1ebSSage Weil */ 15573c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work) 1558355da1ebSSage Weil { 1559355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1560355da1ebSSage Weil i_vmtruncate_work); 1561355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1562355da1ebSSage Weil 1563355da1ebSSage Weil dout("vmtruncate_work %p\n", inode); 1564b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 1565355da1ebSSage Weil iput(inode); 1566355da1ebSSage Weil } 1567355da1ebSSage Weil 1568355da1ebSSage Weil /* 15693c6f6b79SSage Weil * Queue an async vmtruncate. If we fail to queue work, we will handle 15703c6f6b79SSage Weil * the truncation the next time we call __ceph_do_pending_vmtruncate. 15713c6f6b79SSage Weil */ 15723c6f6b79SSage Weil void ceph_queue_vmtruncate(struct inode *inode) 15733c6f6b79SSage Weil { 15743c6f6b79SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 15753c6f6b79SSage Weil 157615a2015fSSage Weil ihold(inode); 157799ccbd22SMilosz Tanski 1578640ef79dSCheng Renquan if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 15793c6f6b79SSage Weil &ci->i_vmtruncate_work)) { 15803c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p\n", inode); 15813c6f6b79SSage Weil } else { 15823c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 15833c6f6b79SSage Weil inode, ci->i_truncate_pending); 158415a2015fSSage Weil iput(inode); 15853c6f6b79SSage Weil } 15863c6f6b79SSage Weil } 15873c6f6b79SSage Weil 15883c6f6b79SSage Weil /* 1589355da1ebSSage Weil * Make sure any pending truncation is applied before doing anything 1590355da1ebSSage Weil * that may depend on it. 1591355da1ebSSage Weil */ 1592b415bf4fSYan, Zheng void __ceph_do_pending_vmtruncate(struct inode *inode) 1593355da1ebSSage Weil { 1594355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1595355da1ebSSage Weil u64 to; 1596a85f50b6SYan, Zheng int wrbuffer_refs, finish = 0; 1597355da1ebSSage Weil 1598b0d7c223SYan, Zheng mutex_lock(&ci->i_truncate_mutex); 1599355da1ebSSage Weil retry: 1600be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1601355da1ebSSage Weil if (ci->i_truncate_pending == 0) { 1602355da1ebSSage Weil dout("__do_pending_vmtruncate %p none pending\n", inode); 1603be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1604b0d7c223SYan, Zheng mutex_unlock(&ci->i_truncate_mutex); 1605355da1ebSSage Weil return; 1606355da1ebSSage Weil } 1607355da1ebSSage Weil 1608355da1ebSSage Weil /* 1609355da1ebSSage Weil * make sure any dirty snapped pages are flushed before we 1610355da1ebSSage Weil * possibly truncate them.. so write AND block! 1611355da1ebSSage Weil */ 1612355da1ebSSage Weil if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1613355da1ebSSage Weil dout("__do_pending_vmtruncate %p flushing snaps first\n", 1614355da1ebSSage Weil inode); 1615be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1616355da1ebSSage Weil filemap_write_and_wait_range(&inode->i_data, 0, 1617355da1ebSSage Weil inode->i_sb->s_maxbytes); 1618355da1ebSSage Weil goto retry; 1619355da1ebSSage Weil } 1620355da1ebSSage Weil 1621b0d7c223SYan, Zheng /* there should be no reader or writer */ 1622b0d7c223SYan, Zheng WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1623b0d7c223SYan, Zheng 1624355da1ebSSage Weil to = ci->i_truncate_size; 1625355da1ebSSage Weil wrbuffer_refs = ci->i_wrbuffer_ref; 1626355da1ebSSage Weil dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1627355da1ebSSage Weil ci->i_truncate_pending, to); 1628be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1629355da1ebSSage Weil 1630355da1ebSSage Weil truncate_inode_pages(inode->i_mapping, to); 1631355da1ebSSage Weil 1632be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1633a85f50b6SYan, Zheng if (to == ci->i_truncate_size) { 1634a85f50b6SYan, Zheng ci->i_truncate_pending = 0; 1635a85f50b6SYan, Zheng finish = 1; 1636a85f50b6SYan, Zheng } 1637be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1638a85f50b6SYan, Zheng if (!finish) 1639a85f50b6SYan, Zheng goto retry; 1640355da1ebSSage Weil 1641b0d7c223SYan, Zheng mutex_unlock(&ci->i_truncate_mutex); 1642b0d7c223SYan, Zheng 1643355da1ebSSage Weil if (wrbuffer_refs == 0) 1644355da1ebSSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1645a85f50b6SYan, Zheng 164603066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1647355da1ebSSage Weil } 1648355da1ebSSage Weil 1649355da1ebSSage Weil /* 1650355da1ebSSage Weil * symlinks 1651355da1ebSSage Weil */ 1652355da1ebSSage Weil static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1653355da1ebSSage Weil { 1654355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1655355da1ebSSage Weil nd_set_link(nd, ci->i_symlink); 1656355da1ebSSage Weil return NULL; 1657355da1ebSSage Weil } 1658355da1ebSSage Weil 1659355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops = { 1660355da1ebSSage Weil .readlink = generic_readlink, 1661355da1ebSSage Weil .follow_link = ceph_sym_follow_link, 16620b932672SYan, Zheng .setattr = ceph_setattr, 16630b932672SYan, Zheng .getattr = ceph_getattr, 16640b932672SYan, Zheng .setxattr = ceph_setxattr, 16650b932672SYan, Zheng .getxattr = ceph_getxattr, 16660b932672SYan, Zheng .listxattr = ceph_listxattr, 16670b932672SYan, Zheng .removexattr = ceph_removexattr, 16687221fe4cSGuangliang Zhao .get_acl = ceph_get_acl, 166972466d0bSSage Weil .set_acl = ceph_set_acl, 1670355da1ebSSage Weil }; 1671355da1ebSSage Weil 1672355da1ebSSage Weil /* 1673355da1ebSSage Weil * setattr 1674355da1ebSSage Weil */ 1675355da1ebSSage Weil int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1676355da1ebSSage Weil { 1677355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1678355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1679355da1ebSSage Weil const unsigned int ia_valid = attr->ia_valid; 1680355da1ebSSage Weil struct ceph_mds_request *req; 16813d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1682355da1ebSSage Weil int issued; 1683355da1ebSSage Weil int release = 0, dirtied = 0; 1684355da1ebSSage Weil int mask = 0; 1685355da1ebSSage Weil int err = 0; 1686fca65b4aSSage Weil int inode_dirty_flags = 0; 1687355da1ebSSage Weil 1688355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1689355da1ebSSage Weil return -EROFS; 1690355da1ebSSage Weil 1691355da1ebSSage Weil err = inode_change_ok(inode, attr); 1692355da1ebSSage Weil if (err != 0) 1693355da1ebSSage Weil return err; 1694355da1ebSSage Weil 1695355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1696355da1ebSSage Weil USE_AUTH_MDS); 1697355da1ebSSage Weil if (IS_ERR(req)) 1698355da1ebSSage Weil return PTR_ERR(req); 1699355da1ebSSage Weil 1700be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1701355da1ebSSage Weil issued = __ceph_caps_issued(ci, NULL); 1702355da1ebSSage Weil dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1703355da1ebSSage Weil 1704355da1ebSSage Weil if (ia_valid & ATTR_UID) { 1705355da1ebSSage Weil dout("setattr %p uid %d -> %d\n", inode, 1706bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 1707bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, attr->ia_uid)); 1708355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1709355da1ebSSage Weil inode->i_uid = attr->ia_uid; 1710355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1711355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1712ab871b90SEric W. Biederman !uid_eq(attr->ia_uid, inode->i_uid)) { 1713ab871b90SEric W. Biederman req->r_args.setattr.uid = cpu_to_le32( 1714ab871b90SEric W. Biederman from_kuid(&init_user_ns, attr->ia_uid)); 1715355da1ebSSage Weil mask |= CEPH_SETATTR_UID; 1716355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1717355da1ebSSage Weil } 1718355da1ebSSage Weil } 1719355da1ebSSage Weil if (ia_valid & ATTR_GID) { 1720355da1ebSSage Weil dout("setattr %p gid %d -> %d\n", inode, 1721bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid), 1722bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, attr->ia_gid)); 1723355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1724355da1ebSSage Weil inode->i_gid = attr->ia_gid; 1725355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1726355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1727ab871b90SEric W. Biederman !gid_eq(attr->ia_gid, inode->i_gid)) { 1728ab871b90SEric W. Biederman req->r_args.setattr.gid = cpu_to_le32( 1729ab871b90SEric W. Biederman from_kgid(&init_user_ns, attr->ia_gid)); 1730355da1ebSSage Weil mask |= CEPH_SETATTR_GID; 1731355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1732355da1ebSSage Weil } 1733355da1ebSSage Weil } 1734355da1ebSSage Weil if (ia_valid & ATTR_MODE) { 1735355da1ebSSage Weil dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1736355da1ebSSage Weil attr->ia_mode); 1737355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1738355da1ebSSage Weil inode->i_mode = attr->ia_mode; 1739355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1740355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1741355da1ebSSage Weil attr->ia_mode != inode->i_mode) { 17427221fe4cSGuangliang Zhao inode->i_mode = attr->ia_mode; 1743355da1ebSSage Weil req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1744355da1ebSSage Weil mask |= CEPH_SETATTR_MODE; 1745355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1746355da1ebSSage Weil } 1747355da1ebSSage Weil } 1748355da1ebSSage Weil 1749355da1ebSSage Weil if (ia_valid & ATTR_ATIME) { 1750355da1ebSSage Weil dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1751355da1ebSSage Weil inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1752355da1ebSSage Weil attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1753355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1754355da1ebSSage Weil ci->i_time_warp_seq++; 1755355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1756355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1757355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1758355da1ebSSage Weil timespec_compare(&inode->i_atime, 1759355da1ebSSage Weil &attr->ia_atime) < 0) { 1760355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1761355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1762355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1763355da1ebSSage Weil !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1764355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.atime, 1765355da1ebSSage Weil &attr->ia_atime); 1766355da1ebSSage Weil mask |= CEPH_SETATTR_ATIME; 1767355da1ebSSage Weil release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1768355da1ebSSage Weil CEPH_CAP_FILE_WR; 1769355da1ebSSage Weil } 1770355da1ebSSage Weil } 1771355da1ebSSage Weil if (ia_valid & ATTR_MTIME) { 1772355da1ebSSage Weil dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1773355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1774355da1ebSSage Weil attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1775355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1776355da1ebSSage Weil ci->i_time_warp_seq++; 1777355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1778355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1779355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1780355da1ebSSage Weil timespec_compare(&inode->i_mtime, 1781355da1ebSSage Weil &attr->ia_mtime) < 0) { 1782355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1783355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1784355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1785355da1ebSSage Weil !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1786355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.mtime, 1787355da1ebSSage Weil &attr->ia_mtime); 1788355da1ebSSage Weil mask |= CEPH_SETATTR_MTIME; 1789355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1790355da1ebSSage Weil CEPH_CAP_FILE_WR; 1791355da1ebSSage Weil } 1792355da1ebSSage Weil } 1793355da1ebSSage Weil if (ia_valid & ATTR_SIZE) { 1794355da1ebSSage Weil dout("setattr %p size %lld -> %lld\n", inode, 1795355da1ebSSage Weil inode->i_size, attr->ia_size); 1796355da1ebSSage Weil if (attr->ia_size > inode->i_sb->s_maxbytes) { 1797355da1ebSSage Weil err = -EINVAL; 1798355da1ebSSage Weil goto out; 1799355da1ebSSage Weil } 1800355da1ebSSage Weil if ((issued & CEPH_CAP_FILE_EXCL) && 1801355da1ebSSage Weil attr->ia_size > inode->i_size) { 1802355da1ebSSage Weil inode->i_size = attr->ia_size; 1803355da1ebSSage Weil inode->i_blocks = 1804355da1ebSSage Weil (attr->ia_size + (1 << 9) - 1) >> 9; 1805355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1806355da1ebSSage Weil ci->i_reported_size = attr->ia_size; 1807355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1808355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1809355da1ebSSage Weil attr->ia_size != inode->i_size) { 1810355da1ebSSage Weil req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1811355da1ebSSage Weil req->r_args.setattr.old_size = 1812355da1ebSSage Weil cpu_to_le64(inode->i_size); 1813355da1ebSSage Weil mask |= CEPH_SETATTR_SIZE; 1814355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1815355da1ebSSage Weil CEPH_CAP_FILE_WR; 1816355da1ebSSage Weil } 1817355da1ebSSage Weil } 1818355da1ebSSage Weil 1819355da1ebSSage Weil /* these do nothing */ 1820355da1ebSSage Weil if (ia_valid & ATTR_CTIME) { 1821355da1ebSSage Weil bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1822355da1ebSSage Weil ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1823355da1ebSSage Weil dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1824355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1825355da1ebSSage Weil attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1826355da1ebSSage Weil only ? "ctime only" : "ignored"); 1827355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1828355da1ebSSage Weil if (only) { 1829355da1ebSSage Weil /* 1830355da1ebSSage Weil * if kernel wants to dirty ctime but nothing else, 1831355da1ebSSage Weil * we need to choose a cap to dirty under, or do 1832355da1ebSSage Weil * a almost-no-op setattr 1833355da1ebSSage Weil */ 1834355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) 1835355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1836355da1ebSSage Weil else if (issued & CEPH_CAP_FILE_EXCL) 1837355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1838355da1ebSSage Weil else if (issued & CEPH_CAP_XATTR_EXCL) 1839355da1ebSSage Weil dirtied |= CEPH_CAP_XATTR_EXCL; 1840355da1ebSSage Weil else 1841355da1ebSSage Weil mask |= CEPH_SETATTR_CTIME; 1842355da1ebSSage Weil } 1843355da1ebSSage Weil } 1844355da1ebSSage Weil if (ia_valid & ATTR_FILE) 1845355da1ebSSage Weil dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1846355da1ebSSage Weil 1847355da1ebSSage Weil if (dirtied) { 1848fca65b4aSSage Weil inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1849355da1ebSSage Weil inode->i_ctime = CURRENT_TIME; 1850355da1ebSSage Weil } 1851355da1ebSSage Weil 1852355da1ebSSage Weil release &= issued; 1853be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1854355da1ebSSage Weil 1855fca65b4aSSage Weil if (inode_dirty_flags) 1856fca65b4aSSage Weil __mark_inode_dirty(inode, inode_dirty_flags); 1857fca65b4aSSage Weil 18587221fe4cSGuangliang Zhao if (ia_valid & ATTR_MODE) { 18594db658eaSLinus Torvalds err = posix_acl_chmod(inode, attr->ia_mode); 18607221fe4cSGuangliang Zhao if (err) 18617221fe4cSGuangliang Zhao goto out_put; 18627221fe4cSGuangliang Zhao } 18637221fe4cSGuangliang Zhao 1864355da1ebSSage Weil if (mask) { 186570b666c3SSage Weil req->r_inode = inode; 186670b666c3SSage Weil ihold(inode); 1867355da1ebSSage Weil req->r_inode_drop = release; 1868355da1ebSSage Weil req->r_args.setattr.mask = cpu_to_le32(mask); 1869355da1ebSSage Weil req->r_num_caps = 1; 1870752c8bdcSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1871355da1ebSSage Weil } 1872355da1ebSSage Weil dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1873355da1ebSSage Weil ceph_cap_string(dirtied), mask); 1874355da1ebSSage Weil 1875355da1ebSSage Weil ceph_mdsc_put_request(req); 1876b0d7c223SYan, Zheng if (mask & CEPH_SETATTR_SIZE) 1877b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 1878355da1ebSSage Weil return err; 1879355da1ebSSage Weil out: 1880be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 18817221fe4cSGuangliang Zhao out_put: 1882355da1ebSSage Weil ceph_mdsc_put_request(req); 1883355da1ebSSage Weil return err; 1884355da1ebSSage Weil } 1885355da1ebSSage Weil 1886355da1ebSSage Weil /* 1887355da1ebSSage Weil * Verify that we have a lease on the given mask. If not, 1888355da1ebSSage Weil * do a getattr against an mds. 1889355da1ebSSage Weil */ 1890355da1ebSSage Weil int ceph_do_getattr(struct inode *inode, int mask) 1891355da1ebSSage Weil { 18923d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 18933d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1894355da1ebSSage Weil struct ceph_mds_request *req; 1895355da1ebSSage Weil int err; 1896355da1ebSSage Weil 1897355da1ebSSage Weil if (ceph_snap(inode) == CEPH_SNAPDIR) { 1898355da1ebSSage Weil dout("do_getattr inode %p SNAPDIR\n", inode); 1899355da1ebSSage Weil return 0; 1900355da1ebSSage Weil } 1901355da1ebSSage Weil 1902b7495fc2SSage Weil dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1903355da1ebSSage Weil if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1904355da1ebSSage Weil return 0; 1905355da1ebSSage Weil 1906355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1907355da1ebSSage Weil if (IS_ERR(req)) 1908355da1ebSSage Weil return PTR_ERR(req); 190970b666c3SSage Weil req->r_inode = inode; 191070b666c3SSage Weil ihold(inode); 1911355da1ebSSage Weil req->r_num_caps = 1; 1912355da1ebSSage Weil req->r_args.getattr.mask = cpu_to_le32(mask); 1913355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1914355da1ebSSage Weil ceph_mdsc_put_request(req); 1915355da1ebSSage Weil dout("do_getattr result=%d\n", err); 1916355da1ebSSage Weil return err; 1917355da1ebSSage Weil } 1918355da1ebSSage Weil 1919355da1ebSSage Weil 1920355da1ebSSage Weil /* 1921355da1ebSSage Weil * Check inode permissions. We verify we have a valid value for 1922355da1ebSSage Weil * the AUTH cap, then call the generic handler. 1923355da1ebSSage Weil */ 192410556cb2SAl Viro int ceph_permission(struct inode *inode, int mask) 1925355da1ebSSage Weil { 1926b74c79e9SNick Piggin int err; 1927b74c79e9SNick Piggin 192810556cb2SAl Viro if (mask & MAY_NOT_BLOCK) 1929b74c79e9SNick Piggin return -ECHILD; 1930b74c79e9SNick Piggin 1931b74c79e9SNick Piggin err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1932355da1ebSSage Weil 1933355da1ebSSage Weil if (!err) 19342830ba7fSAl Viro err = generic_permission(inode, mask); 1935355da1ebSSage Weil return err; 1936355da1ebSSage Weil } 1937355da1ebSSage Weil 1938355da1ebSSage Weil /* 1939355da1ebSSage Weil * Get all attributes. Hopefully somedata we'll have a statlite() 1940355da1ebSSage Weil * and can limit the fields we require to be accurate. 1941355da1ebSSage Weil */ 1942355da1ebSSage Weil int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1943355da1ebSSage Weil struct kstat *stat) 1944355da1ebSSage Weil { 1945355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1946232d4b01SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1947355da1ebSSage Weil int err; 1948355da1ebSSage Weil 1949355da1ebSSage Weil err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1950355da1ebSSage Weil if (!err) { 1951355da1ebSSage Weil generic_fillattr(inode, stat); 1952ad1fee96SYehuda Sadeh stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1953355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1954355da1ebSSage Weil stat->dev = ceph_snap(inode); 1955355da1ebSSage Weil else 1956355da1ebSSage Weil stat->dev = 0; 1957232d4b01SSage Weil if (S_ISDIR(inode->i_mode)) { 19581c1266bbSYehuda Sadeh if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 19591c1266bbSYehuda Sadeh RBYTES)) 1960232d4b01SSage Weil stat->size = ci->i_rbytes; 19611c1266bbSYehuda Sadeh else 19621c1266bbSYehuda Sadeh stat->size = ci->i_files + ci->i_subdirs; 1963232d4b01SSage Weil stat->blocks = 0; 1964355da1ebSSage Weil stat->blksize = 65536; 1965355da1ebSSage Weil } 1966232d4b01SSage Weil } 1967355da1ebSSage Weil return err; 1968355da1ebSSage Weil } 1969