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> 12c9af9fb6SYehuda Sadeh #include <linux/pagevec.h> 13355da1ebSSage Weil 14355da1ebSSage Weil #include "super.h" 153d14c5d2SYehuda Sadeh #include "mds_client.h" 163d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 17355da1ebSSage Weil 18355da1ebSSage Weil /* 19355da1ebSSage Weil * Ceph inode operations 20355da1ebSSage Weil * 21355da1ebSSage Weil * Implement basic inode helpers (get, alloc) and inode ops (getattr, 22355da1ebSSage Weil * setattr, etc.), xattr helpers, and helpers for assimilating 23355da1ebSSage Weil * metadata returned by the MDS into our cache. 24355da1ebSSage Weil * 25355da1ebSSage Weil * Also define helpers for doing asynchronous writeback, invalidation, 26355da1ebSSage Weil * and truncation for the benefit of those who can't afford to block 27355da1ebSSage Weil * (typically because they are in the message handler path). 28355da1ebSSage Weil */ 29355da1ebSSage Weil 30355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops; 31355da1ebSSage Weil 323c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work); 333c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work); 343c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work); 35355da1ebSSage Weil 36355da1ebSSage Weil /* 37355da1ebSSage Weil * find or create an inode, given the ceph ino number 38355da1ebSSage Weil */ 39355da1ebSSage Weil struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 40355da1ebSSage Weil { 41355da1ebSSage Weil struct inode *inode; 42355da1ebSSage Weil ino_t t = ceph_vino_to_ino(vino); 43355da1ebSSage Weil 44355da1ebSSage Weil inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 45355da1ebSSage Weil if (inode == NULL) 46355da1ebSSage Weil return ERR_PTR(-ENOMEM); 47355da1ebSSage Weil if (inode->i_state & I_NEW) { 48355da1ebSSage Weil dout("get_inode created new inode %p %llx.%llx ino %llx\n", 49355da1ebSSage Weil inode, ceph_vinop(inode), (u64)inode->i_ino); 50355da1ebSSage Weil unlock_new_inode(inode); 51355da1ebSSage Weil } 52355da1ebSSage Weil 53355da1ebSSage Weil dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 54355da1ebSSage Weil vino.snap, inode); 55355da1ebSSage Weil return inode; 56355da1ebSSage Weil } 57355da1ebSSage Weil 58355da1ebSSage Weil /* 59355da1ebSSage Weil * get/constuct snapdir inode for a given directory 60355da1ebSSage Weil */ 61355da1ebSSage Weil struct inode *ceph_get_snapdir(struct inode *parent) 62355da1ebSSage Weil { 63355da1ebSSage Weil struct ceph_vino vino = { 64355da1ebSSage Weil .ino = ceph_ino(parent), 65355da1ebSSage Weil .snap = CEPH_SNAPDIR, 66355da1ebSSage Weil }; 67355da1ebSSage Weil struct inode *inode = ceph_get_inode(parent->i_sb, vino); 68b377ff13SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 69355da1ebSSage Weil 70355da1ebSSage Weil BUG_ON(!S_ISDIR(parent->i_mode)); 71355da1ebSSage Weil if (IS_ERR(inode)) 727e34bc52SJulia Lawall return inode; 73355da1ebSSage Weil inode->i_mode = parent->i_mode; 74355da1ebSSage Weil inode->i_uid = parent->i_uid; 75355da1ebSSage Weil inode->i_gid = parent->i_gid; 76355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 77355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 78b377ff13SSage Weil ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 79b377ff13SSage Weil ci->i_rbytes = 0; 80355da1ebSSage Weil return inode; 81355da1ebSSage Weil } 82355da1ebSSage Weil 83355da1ebSSage Weil const struct inode_operations ceph_file_iops = { 84355da1ebSSage Weil .permission = ceph_permission, 85355da1ebSSage Weil .setattr = ceph_setattr, 86355da1ebSSage Weil .getattr = ceph_getattr, 87355da1ebSSage Weil .setxattr = ceph_setxattr, 88355da1ebSSage Weil .getxattr = ceph_getxattr, 89355da1ebSSage Weil .listxattr = ceph_listxattr, 90355da1ebSSage Weil .removexattr = ceph_removexattr, 91355da1ebSSage Weil }; 92355da1ebSSage Weil 93355da1ebSSage Weil 94355da1ebSSage Weil /* 95355da1ebSSage Weil * We use a 'frag tree' to keep track of the MDS's directory fragments 96355da1ebSSage Weil * for a given inode (usually there is just a single fragment). We 97355da1ebSSage Weil * need to know when a child frag is delegated to a new MDS, or when 98355da1ebSSage Weil * it is flagged as replicated, so we can direct our requests 99355da1ebSSage Weil * accordingly. 100355da1ebSSage Weil */ 101355da1ebSSage Weil 102355da1ebSSage Weil /* 103355da1ebSSage Weil * find/create a frag in the tree 104355da1ebSSage Weil */ 105355da1ebSSage Weil static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 106355da1ebSSage Weil u32 f) 107355da1ebSSage Weil { 108355da1ebSSage Weil struct rb_node **p; 109355da1ebSSage Weil struct rb_node *parent = NULL; 110355da1ebSSage Weil struct ceph_inode_frag *frag; 111355da1ebSSage Weil int c; 112355da1ebSSage Weil 113355da1ebSSage Weil p = &ci->i_fragtree.rb_node; 114355da1ebSSage Weil while (*p) { 115355da1ebSSage Weil parent = *p; 116355da1ebSSage Weil frag = rb_entry(parent, struct ceph_inode_frag, node); 117355da1ebSSage Weil c = ceph_frag_compare(f, frag->frag); 118355da1ebSSage Weil if (c < 0) 119355da1ebSSage Weil p = &(*p)->rb_left; 120355da1ebSSage Weil else if (c > 0) 121355da1ebSSage Weil p = &(*p)->rb_right; 122355da1ebSSage Weil else 123355da1ebSSage Weil return frag; 124355da1ebSSage Weil } 125355da1ebSSage Weil 126355da1ebSSage Weil frag = kmalloc(sizeof(*frag), GFP_NOFS); 127355da1ebSSage Weil if (!frag) { 128355da1ebSSage Weil pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 129355da1ebSSage Weil "frag %x\n", &ci->vfs_inode, 130355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 131355da1ebSSage Weil return ERR_PTR(-ENOMEM); 132355da1ebSSage Weil } 133355da1ebSSage Weil frag->frag = f; 134355da1ebSSage Weil frag->split_by = 0; 135355da1ebSSage Weil frag->mds = -1; 136355da1ebSSage Weil frag->ndist = 0; 137355da1ebSSage Weil 138355da1ebSSage Weil rb_link_node(&frag->node, parent, p); 139355da1ebSSage Weil rb_insert_color(&frag->node, &ci->i_fragtree); 140355da1ebSSage Weil 141355da1ebSSage Weil dout("get_or_create_frag added %llx.%llx frag %x\n", 142355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 143355da1ebSSage Weil return frag; 144355da1ebSSage Weil } 145355da1ebSSage Weil 146355da1ebSSage Weil /* 147355da1ebSSage Weil * find a specific frag @f 148355da1ebSSage Weil */ 149355da1ebSSage Weil struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 150355da1ebSSage Weil { 151355da1ebSSage Weil struct rb_node *n = ci->i_fragtree.rb_node; 152355da1ebSSage Weil 153355da1ebSSage Weil while (n) { 154355da1ebSSage Weil struct ceph_inode_frag *frag = 155355da1ebSSage Weil rb_entry(n, struct ceph_inode_frag, node); 156355da1ebSSage Weil int c = ceph_frag_compare(f, frag->frag); 157355da1ebSSage Weil if (c < 0) 158355da1ebSSage Weil n = n->rb_left; 159355da1ebSSage Weil else if (c > 0) 160355da1ebSSage Weil n = n->rb_right; 161355da1ebSSage Weil else 162355da1ebSSage Weil return frag; 163355da1ebSSage Weil } 164355da1ebSSage Weil return NULL; 165355da1ebSSage Weil } 166355da1ebSSage Weil 167355da1ebSSage Weil /* 168355da1ebSSage Weil * Choose frag containing the given value @v. If @pfrag is 169355da1ebSSage Weil * specified, copy the frag delegation info to the caller if 170355da1ebSSage Weil * it is present. 171355da1ebSSage Weil */ 172355da1ebSSage Weil u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 173355da1ebSSage Weil struct ceph_inode_frag *pfrag, 174355da1ebSSage Weil int *found) 175355da1ebSSage Weil { 176355da1ebSSage Weil u32 t = ceph_frag_make(0, 0); 177355da1ebSSage Weil struct ceph_inode_frag *frag; 178355da1ebSSage Weil unsigned nway, i; 179355da1ebSSage Weil u32 n; 180355da1ebSSage Weil 181355da1ebSSage Weil if (found) 182355da1ebSSage Weil *found = 0; 183355da1ebSSage Weil 184355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 185355da1ebSSage Weil while (1) { 186355da1ebSSage Weil WARN_ON(!ceph_frag_contains_value(t, v)); 187355da1ebSSage Weil frag = __ceph_find_frag(ci, t); 188355da1ebSSage Weil if (!frag) 189355da1ebSSage Weil break; /* t is a leaf */ 190355da1ebSSage Weil if (frag->split_by == 0) { 191355da1ebSSage Weil if (pfrag) 192355da1ebSSage Weil memcpy(pfrag, frag, sizeof(*pfrag)); 193355da1ebSSage Weil if (found) 194355da1ebSSage Weil *found = 1; 195355da1ebSSage Weil break; 196355da1ebSSage Weil } 197355da1ebSSage Weil 198355da1ebSSage Weil /* choose child */ 199355da1ebSSage Weil nway = 1 << frag->split_by; 200355da1ebSSage Weil dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 201355da1ebSSage Weil frag->split_by, nway); 202355da1ebSSage Weil for (i = 0; i < nway; i++) { 203355da1ebSSage Weil n = ceph_frag_make_child(t, frag->split_by, i); 204355da1ebSSage Weil if (ceph_frag_contains_value(n, v)) { 205355da1ebSSage Weil t = n; 206355da1ebSSage Weil break; 207355da1ebSSage Weil } 208355da1ebSSage Weil } 209355da1ebSSage Weil BUG_ON(i == nway); 210355da1ebSSage Weil } 211355da1ebSSage Weil dout("choose_frag(%x) = %x\n", v, t); 212355da1ebSSage Weil 213355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 214355da1ebSSage Weil return t; 215355da1ebSSage Weil } 216355da1ebSSage Weil 217355da1ebSSage Weil /* 218355da1ebSSage Weil * Process dirfrag (delegation) info from the mds. Include leaf 219355da1ebSSage Weil * fragment in tree ONLY if ndist > 0. Otherwise, only 220355da1ebSSage Weil * branches/splits are included in i_fragtree) 221355da1ebSSage Weil */ 222355da1ebSSage Weil static int ceph_fill_dirfrag(struct inode *inode, 223355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo) 224355da1ebSSage Weil { 225355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 226355da1ebSSage Weil struct ceph_inode_frag *frag; 227355da1ebSSage Weil u32 id = le32_to_cpu(dirinfo->frag); 228355da1ebSSage Weil int mds = le32_to_cpu(dirinfo->auth); 229355da1ebSSage Weil int ndist = le32_to_cpu(dirinfo->ndist); 230355da1ebSSage Weil int i; 231355da1ebSSage Weil int err = 0; 232355da1ebSSage Weil 233355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 234355da1ebSSage Weil if (ndist == 0) { 235355da1ebSSage Weil /* no delegation info needed. */ 236355da1ebSSage Weil frag = __ceph_find_frag(ci, id); 237355da1ebSSage Weil if (!frag) 238355da1ebSSage Weil goto out; 239355da1ebSSage Weil if (frag->split_by == 0) { 240355da1ebSSage Weil /* tree leaf, remove */ 241355da1ebSSage Weil dout("fill_dirfrag removed %llx.%llx frag %x" 242355da1ebSSage Weil " (no ref)\n", ceph_vinop(inode), id); 243355da1ebSSage Weil rb_erase(&frag->node, &ci->i_fragtree); 244355da1ebSSage Weil kfree(frag); 245355da1ebSSage Weil } else { 246355da1ebSSage Weil /* tree branch, keep and clear */ 247355da1ebSSage Weil dout("fill_dirfrag cleared %llx.%llx frag %x" 248355da1ebSSage Weil " referral\n", ceph_vinop(inode), id); 249355da1ebSSage Weil frag->mds = -1; 250355da1ebSSage Weil frag->ndist = 0; 251355da1ebSSage Weil } 252355da1ebSSage Weil goto out; 253355da1ebSSage Weil } 254355da1ebSSage Weil 255355da1ebSSage Weil 256355da1ebSSage Weil /* find/add this frag to store mds delegation info */ 257355da1ebSSage Weil frag = __get_or_create_frag(ci, id); 258355da1ebSSage Weil if (IS_ERR(frag)) { 259355da1ebSSage Weil /* this is not the end of the world; we can continue 260355da1ebSSage Weil with bad/inaccurate delegation info */ 261355da1ebSSage Weil pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 262355da1ebSSage Weil ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 263355da1ebSSage Weil err = -ENOMEM; 264355da1ebSSage Weil goto out; 265355da1ebSSage Weil } 266355da1ebSSage Weil 267355da1ebSSage Weil frag->mds = mds; 268355da1ebSSage Weil frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 269355da1ebSSage Weil for (i = 0; i < frag->ndist; i++) 270355da1ebSSage Weil frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 271355da1ebSSage Weil dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 272355da1ebSSage Weil ceph_vinop(inode), frag->frag, frag->ndist); 273355da1ebSSage Weil 274355da1ebSSage Weil out: 275355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 276355da1ebSSage Weil return err; 277355da1ebSSage Weil } 278355da1ebSSage Weil 279355da1ebSSage Weil 280355da1ebSSage Weil /* 281355da1ebSSage Weil * initialize a newly allocated inode. 282355da1ebSSage Weil */ 283355da1ebSSage Weil struct inode *ceph_alloc_inode(struct super_block *sb) 284355da1ebSSage Weil { 285355da1ebSSage Weil struct ceph_inode_info *ci; 286355da1ebSSage Weil int i; 287355da1ebSSage Weil 288355da1ebSSage Weil ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 289355da1ebSSage Weil if (!ci) 290355da1ebSSage Weil return NULL; 291355da1ebSSage Weil 292355da1ebSSage Weil dout("alloc_inode %p\n", &ci->vfs_inode); 293355da1ebSSage Weil 294355da1ebSSage Weil ci->i_version = 0; 295355da1ebSSage Weil ci->i_time_warp_seq = 0; 296355da1ebSSage Weil ci->i_ceph_flags = 0; 297355da1ebSSage Weil ci->i_release_count = 0; 298355da1ebSSage Weil ci->i_symlink = NULL; 299355da1ebSSage Weil 3006c0f3af7SSage Weil memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 3016c0f3af7SSage Weil 302355da1ebSSage Weil ci->i_fragtree = RB_ROOT; 303355da1ebSSage Weil mutex_init(&ci->i_fragtree_mutex); 304355da1ebSSage Weil 305355da1ebSSage Weil ci->i_xattrs.blob = NULL; 306355da1ebSSage Weil ci->i_xattrs.prealloc_blob = NULL; 307355da1ebSSage Weil ci->i_xattrs.dirty = false; 308355da1ebSSage Weil ci->i_xattrs.index = RB_ROOT; 309355da1ebSSage Weil ci->i_xattrs.count = 0; 310355da1ebSSage Weil ci->i_xattrs.names_size = 0; 311355da1ebSSage Weil ci->i_xattrs.vals_size = 0; 312355da1ebSSage Weil ci->i_xattrs.version = 0; 313355da1ebSSage Weil ci->i_xattrs.index_version = 0; 314355da1ebSSage Weil 315355da1ebSSage Weil ci->i_caps = RB_ROOT; 316355da1ebSSage Weil ci->i_auth_cap = NULL; 317355da1ebSSage Weil ci->i_dirty_caps = 0; 318355da1ebSSage Weil ci->i_flushing_caps = 0; 319355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_dirty_item); 320355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_flushing_item); 321355da1ebSSage Weil ci->i_cap_flush_seq = 0; 322355da1ebSSage Weil ci->i_cap_flush_last_tid = 0; 323355da1ebSSage Weil memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 324355da1ebSSage Weil init_waitqueue_head(&ci->i_cap_wq); 325355da1ebSSage Weil ci->i_hold_caps_min = 0; 326355da1ebSSage Weil ci->i_hold_caps_max = 0; 327355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_delay_list); 328355da1ebSSage Weil ci->i_cap_exporting_mds = 0; 329355da1ebSSage Weil ci->i_cap_exporting_mseq = 0; 330355da1ebSSage Weil ci->i_cap_exporting_issued = 0; 331355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_snaps); 332355da1ebSSage Weil ci->i_head_snapc = NULL; 333355da1ebSSage Weil ci->i_snap_caps = 0; 334355da1ebSSage Weil 335355da1ebSSage Weil for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 336355da1ebSSage Weil ci->i_nr_by_mode[i] = 0; 337355da1ebSSage Weil 338355da1ebSSage Weil ci->i_truncate_seq = 0; 339355da1ebSSage Weil ci->i_truncate_size = 0; 340355da1ebSSage Weil ci->i_truncate_pending = 0; 341355da1ebSSage Weil 342355da1ebSSage Weil ci->i_max_size = 0; 343355da1ebSSage Weil ci->i_reported_size = 0; 344355da1ebSSage Weil ci->i_wanted_max_size = 0; 345355da1ebSSage Weil ci->i_requested_max_size = 0; 346355da1ebSSage Weil 347355da1ebSSage Weil ci->i_pin_ref = 0; 348355da1ebSSage Weil ci->i_rd_ref = 0; 349355da1ebSSage Weil ci->i_rdcache_ref = 0; 350355da1ebSSage Weil ci->i_wr_ref = 0; 351355da1ebSSage Weil ci->i_wrbuffer_ref = 0; 352355da1ebSSage Weil ci->i_wrbuffer_ref_head = 0; 353355da1ebSSage Weil ci->i_shared_gen = 0; 354355da1ebSSage Weil ci->i_rdcache_gen = 0; 355355da1ebSSage Weil ci->i_rdcache_revoking = 0; 356355da1ebSSage Weil 357355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_writes); 358355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_dirops); 359355da1ebSSage Weil spin_lock_init(&ci->i_unsafe_lock); 360355da1ebSSage Weil 361355da1ebSSage Weil ci->i_snap_realm = NULL; 362355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_realm_item); 363355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_flush_item); 364355da1ebSSage Weil 3653c6f6b79SSage Weil INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 3663c6f6b79SSage Weil INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 367355da1ebSSage Weil 368355da1ebSSage Weil INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 369355da1ebSSage Weil 370355da1ebSSage Weil return &ci->vfs_inode; 371355da1ebSSage Weil } 372355da1ebSSage Weil 373fa0d7e3dSNick Piggin static void ceph_i_callback(struct rcu_head *head) 374fa0d7e3dSNick Piggin { 375fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 376fa0d7e3dSNick Piggin struct ceph_inode_info *ci = ceph_inode(inode); 377fa0d7e3dSNick Piggin 378fa0d7e3dSNick Piggin INIT_LIST_HEAD(&inode->i_dentry); 379fa0d7e3dSNick Piggin kmem_cache_free(ceph_inode_cachep, ci); 380fa0d7e3dSNick Piggin } 381fa0d7e3dSNick Piggin 382355da1ebSSage Weil void ceph_destroy_inode(struct inode *inode) 383355da1ebSSage Weil { 384355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 385355da1ebSSage Weil struct ceph_inode_frag *frag; 386355da1ebSSage Weil struct rb_node *n; 387355da1ebSSage Weil 388355da1ebSSage Weil dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 389355da1ebSSage Weil 390355da1ebSSage Weil ceph_queue_caps_release(inode); 391355da1ebSSage Weil 3928b218b8aSSage Weil /* 3938b218b8aSSage Weil * we may still have a snap_realm reference if there are stray 3948b218b8aSSage Weil * caps in i_cap_exporting_issued or i_snap_caps. 3958b218b8aSSage Weil */ 3968b218b8aSSage Weil if (ci->i_snap_realm) { 3978b218b8aSSage Weil struct ceph_mds_client *mdsc = 3983d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 3998b218b8aSSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 4008b218b8aSSage Weil 4018b218b8aSSage Weil dout(" dropping residual ref to snap realm %p\n", realm); 4028b218b8aSSage Weil spin_lock(&realm->inodes_with_caps_lock); 4038b218b8aSSage Weil list_del_init(&ci->i_snap_realm_item); 4048b218b8aSSage Weil spin_unlock(&realm->inodes_with_caps_lock); 4058b218b8aSSage Weil ceph_put_snap_realm(mdsc, realm); 4068b218b8aSSage Weil } 4078b218b8aSSage Weil 408355da1ebSSage Weil kfree(ci->i_symlink); 409355da1ebSSage Weil while ((n = rb_first(&ci->i_fragtree)) != NULL) { 410355da1ebSSage Weil frag = rb_entry(n, struct ceph_inode_frag, node); 411355da1ebSSage Weil rb_erase(n, &ci->i_fragtree); 412355da1ebSSage Weil kfree(frag); 413355da1ebSSage Weil } 414355da1ebSSage Weil 415355da1ebSSage Weil __ceph_destroy_xattrs(ci); 416b6c1d5b8SSage Weil if (ci->i_xattrs.blob) 417355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 418b6c1d5b8SSage Weil if (ci->i_xattrs.prealloc_blob) 419355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.prealloc_blob); 420355da1ebSSage Weil 421fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, ceph_i_callback); 422355da1ebSSage Weil } 423355da1ebSSage Weil 424355da1ebSSage Weil 425355da1ebSSage Weil /* 426355da1ebSSage Weil * Helpers to fill in size, ctime, mtime, and atime. We have to be 427355da1ebSSage Weil * careful because either the client or MDS may have more up to date 428355da1ebSSage Weil * info, depending on which capabilities are held, and whether 429355da1ebSSage Weil * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 430355da1ebSSage Weil * and size are monotonically increasing, except when utimes() or 431355da1ebSSage Weil * truncate() increments the corresponding _seq values.) 432355da1ebSSage Weil */ 433355da1ebSSage Weil int ceph_fill_file_size(struct inode *inode, int issued, 434355da1ebSSage Weil u32 truncate_seq, u64 truncate_size, u64 size) 435355da1ebSSage Weil { 436355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 437355da1ebSSage Weil int queue_trunc = 0; 438355da1ebSSage Weil 439355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 440355da1ebSSage Weil (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 441355da1ebSSage Weil dout("size %lld -> %llu\n", inode->i_size, size); 442355da1ebSSage Weil inode->i_size = size; 443355da1ebSSage Weil inode->i_blocks = (size + (1<<9) - 1) >> 9; 444355da1ebSSage Weil ci->i_reported_size = size; 445355da1ebSSage Weil if (truncate_seq != ci->i_truncate_seq) { 446355da1ebSSage Weil dout("truncate_seq %u -> %u\n", 447355da1ebSSage Weil ci->i_truncate_seq, truncate_seq); 448355da1ebSSage Weil ci->i_truncate_seq = truncate_seq; 4493d497d85SYehuda Sadeh /* 4503d497d85SYehuda Sadeh * If we hold relevant caps, or in the case where we're 4513d497d85SYehuda Sadeh * not the only client referencing this file and we 4523d497d85SYehuda Sadeh * don't hold those caps, then we need to check whether 4533d497d85SYehuda Sadeh * the file is either opened or mmaped 4543d497d85SYehuda Sadeh */ 4553d497d85SYehuda Sadeh if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 456355da1ebSSage Weil CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 4572962507cSSage Weil CEPH_CAP_FILE_EXCL| 4582962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) || 4593d497d85SYehuda Sadeh mapping_mapped(inode->i_mapping) || 4603d497d85SYehuda Sadeh __ceph_caps_file_wanted(ci)) { 461355da1ebSSage Weil ci->i_truncate_pending++; 462355da1ebSSage Weil queue_trunc = 1; 463355da1ebSSage Weil } 464355da1ebSSage Weil } 465355da1ebSSage Weil } 466355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 467355da1ebSSage Weil ci->i_truncate_size != truncate_size) { 468355da1ebSSage Weil dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 469355da1ebSSage Weil truncate_size); 470355da1ebSSage Weil ci->i_truncate_size = truncate_size; 471355da1ebSSage Weil } 472355da1ebSSage Weil return queue_trunc; 473355da1ebSSage Weil } 474355da1ebSSage Weil 475355da1ebSSage Weil void ceph_fill_file_time(struct inode *inode, int issued, 476355da1ebSSage Weil u64 time_warp_seq, struct timespec *ctime, 477355da1ebSSage Weil struct timespec *mtime, struct timespec *atime) 478355da1ebSSage Weil { 479355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 480355da1ebSSage Weil int warn = 0; 481355da1ebSSage Weil 482355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_EXCL| 483355da1ebSSage Weil CEPH_CAP_FILE_WR| 484d8672d64SSage Weil CEPH_CAP_FILE_BUFFER| 485d8672d64SSage Weil CEPH_CAP_AUTH_EXCL| 486d8672d64SSage Weil CEPH_CAP_XATTR_EXCL)) { 487355da1ebSSage Weil if (timespec_compare(ctime, &inode->i_ctime) > 0) { 488355da1ebSSage Weil dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 489355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 490355da1ebSSage Weil ctime->tv_sec, ctime->tv_nsec); 491355da1ebSSage Weil inode->i_ctime = *ctime; 492355da1ebSSage Weil } 493355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 494355da1ebSSage Weil /* the MDS did a utimes() */ 495355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld " 496355da1ebSSage Weil "tw %d -> %d\n", 497355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 498355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec, 499355da1ebSSage Weil ci->i_time_warp_seq, (int)time_warp_seq); 500355da1ebSSage Weil 501355da1ebSSage Weil inode->i_mtime = *mtime; 502355da1ebSSage Weil inode->i_atime = *atime; 503355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 504355da1ebSSage Weil } else if (time_warp_seq == ci->i_time_warp_seq) { 505355da1ebSSage Weil /* nobody did utimes(); take the max */ 506355da1ebSSage Weil if (timespec_compare(mtime, &inode->i_mtime) > 0) { 507355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 508355da1ebSSage Weil inode->i_mtime.tv_sec, 509355da1ebSSage Weil inode->i_mtime.tv_nsec, 510355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec); 511355da1ebSSage Weil inode->i_mtime = *mtime; 512355da1ebSSage Weil } 513355da1ebSSage Weil if (timespec_compare(atime, &inode->i_atime) > 0) { 514355da1ebSSage Weil dout("atime %ld.%09ld -> %ld.%09ld inc\n", 515355da1ebSSage Weil inode->i_atime.tv_sec, 516355da1ebSSage Weil inode->i_atime.tv_nsec, 517355da1ebSSage Weil atime->tv_sec, atime->tv_nsec); 518355da1ebSSage Weil inode->i_atime = *atime; 519355da1ebSSage Weil } 520355da1ebSSage Weil } else if (issued & CEPH_CAP_FILE_EXCL) { 521355da1ebSSage Weil /* we did a utimes(); ignore mds values */ 522355da1ebSSage Weil } else { 523355da1ebSSage Weil warn = 1; 524355da1ebSSage Weil } 525355da1ebSSage Weil } else { 526d8672d64SSage Weil /* we have no write|excl caps; whatever the MDS says is true */ 527355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 528355da1ebSSage Weil inode->i_ctime = *ctime; 529355da1ebSSage Weil inode->i_mtime = *mtime; 530355da1ebSSage Weil inode->i_atime = *atime; 531355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 532355da1ebSSage Weil } else { 533355da1ebSSage Weil warn = 1; 534355da1ebSSage Weil } 535355da1ebSSage Weil } 536355da1ebSSage Weil if (warn) /* time_warp_seq shouldn't go backwards */ 537355da1ebSSage Weil dout("%p mds time_warp_seq %llu < %u\n", 538355da1ebSSage Weil inode, time_warp_seq, ci->i_time_warp_seq); 539355da1ebSSage Weil } 540355da1ebSSage Weil 541355da1ebSSage Weil /* 542355da1ebSSage Weil * Populate an inode based on info from mds. May be called on new or 543355da1ebSSage Weil * existing inodes. 544355da1ebSSage Weil */ 545355da1ebSSage Weil static int fill_inode(struct inode *inode, 546355da1ebSSage Weil struct ceph_mds_reply_info_in *iinfo, 547355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo, 548355da1ebSSage Weil struct ceph_mds_session *session, 549355da1ebSSage Weil unsigned long ttl_from, int cap_fmode, 550355da1ebSSage Weil struct ceph_cap_reservation *caps_reservation) 551355da1ebSSage Weil { 552355da1ebSSage Weil struct ceph_mds_reply_inode *info = iinfo->in; 553355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 554355da1ebSSage Weil int i; 555355da1ebSSage Weil int issued, implemented; 556355da1ebSSage Weil struct timespec mtime, atime, ctime; 557355da1ebSSage Weil u32 nsplits; 558355da1ebSSage Weil struct ceph_buffer *xattr_blob = NULL; 559355da1ebSSage Weil int err = 0; 560355da1ebSSage Weil int queue_trunc = 0; 561355da1ebSSage Weil 562355da1ebSSage Weil dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 563355da1ebSSage Weil inode, ceph_vinop(inode), le64_to_cpu(info->version), 564355da1ebSSage Weil ci->i_version); 565355da1ebSSage Weil 566355da1ebSSage Weil /* 567355da1ebSSage Weil * prealloc xattr data, if it looks like we'll need it. only 568355da1ebSSage Weil * if len > 4 (meaning there are actually xattrs; the first 4 569355da1ebSSage Weil * bytes are the xattr count). 570355da1ebSSage Weil */ 571355da1ebSSage Weil if (iinfo->xattr_len > 4) { 572b6c1d5b8SSage Weil xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 573355da1ebSSage Weil if (!xattr_blob) 574355da1ebSSage Weil pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 575355da1ebSSage Weil iinfo->xattr_len); 576355da1ebSSage Weil } 577355da1ebSSage Weil 578355da1ebSSage Weil spin_lock(&inode->i_lock); 579355da1ebSSage Weil 580355da1ebSSage Weil /* 581355da1ebSSage Weil * provided version will be odd if inode value is projected, 5828bd59e01SSage Weil * even if stable. skip the update if we have newer stable 5838bd59e01SSage Weil * info (ours>=theirs, e.g. due to racing mds replies), unless 5848bd59e01SSage Weil * we are getting projected (unstable) info (in which case the 5858bd59e01SSage Weil * version is odd, and we want ours>theirs). 5868bd59e01SSage Weil * us them 5878bd59e01SSage Weil * 2 2 skip 5888bd59e01SSage Weil * 3 2 skip 5898bd59e01SSage Weil * 3 3 update 590355da1ebSSage Weil */ 591355da1ebSSage Weil if (le64_to_cpu(info->version) > 0 && 5928bd59e01SSage Weil (ci->i_version & ~1) >= le64_to_cpu(info->version)) 593355da1ebSSage Weil goto no_change; 594355da1ebSSage Weil 595355da1ebSSage Weil issued = __ceph_caps_issued(ci, &implemented); 596355da1ebSSage Weil issued |= implemented | __ceph_caps_dirty(ci); 597355da1ebSSage Weil 598355da1ebSSage Weil /* update inode */ 599355da1ebSSage Weil ci->i_version = le64_to_cpu(info->version); 600355da1ebSSage Weil inode->i_version++; 601355da1ebSSage Weil inode->i_rdev = le32_to_cpu(info->rdev); 602355da1ebSSage Weil 603355da1ebSSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 604355da1ebSSage Weil inode->i_mode = le32_to_cpu(info->mode); 605355da1ebSSage Weil inode->i_uid = le32_to_cpu(info->uid); 606355da1ebSSage Weil inode->i_gid = le32_to_cpu(info->gid); 607355da1ebSSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 608355da1ebSSage Weil inode->i_uid, inode->i_gid); 609355da1ebSSage Weil } 610355da1ebSSage Weil 611355da1ebSSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 612355da1ebSSage Weil inode->i_nlink = le32_to_cpu(info->nlink); 613355da1ebSSage Weil 614355da1ebSSage Weil /* be careful with mtime, atime, size */ 615355da1ebSSage Weil ceph_decode_timespec(&atime, &info->atime); 616355da1ebSSage Weil ceph_decode_timespec(&mtime, &info->mtime); 617355da1ebSSage Weil ceph_decode_timespec(&ctime, &info->ctime); 618355da1ebSSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 619355da1ebSSage Weil le32_to_cpu(info->truncate_seq), 620355da1ebSSage Weil le64_to_cpu(info->truncate_size), 621355da1ebSSage Weil le64_to_cpu(info->size)); 622355da1ebSSage Weil ceph_fill_file_time(inode, issued, 623355da1ebSSage Weil le32_to_cpu(info->time_warp_seq), 624355da1ebSSage Weil &ctime, &mtime, &atime); 625355da1ebSSage Weil 626912a9b03SSage Weil /* only update max_size on auth cap */ 627912a9b03SSage Weil if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 628912a9b03SSage Weil ci->i_max_size != le64_to_cpu(info->max_size)) { 629912a9b03SSage Weil dout("max_size %lld -> %llu\n", ci->i_max_size, 630912a9b03SSage Weil le64_to_cpu(info->max_size)); 631355da1ebSSage Weil ci->i_max_size = le64_to_cpu(info->max_size); 632912a9b03SSage Weil } 633912a9b03SSage Weil 634355da1ebSSage Weil ci->i_layout = info->layout; 635355da1ebSSage Weil inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 636355da1ebSSage Weil 637355da1ebSSage Weil /* xattrs */ 638355da1ebSSage Weil /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 639355da1ebSSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 640355da1ebSSage Weil le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 641355da1ebSSage Weil if (ci->i_xattrs.blob) 642355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 643355da1ebSSage Weil ci->i_xattrs.blob = xattr_blob; 644355da1ebSSage Weil if (xattr_blob) 645355da1ebSSage Weil memcpy(ci->i_xattrs.blob->vec.iov_base, 646355da1ebSSage Weil iinfo->xattr_data, iinfo->xattr_len); 647355da1ebSSage Weil ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 648a6424e48SSage Weil xattr_blob = NULL; 649355da1ebSSage Weil } 650355da1ebSSage Weil 651355da1ebSSage Weil inode->i_mapping->a_ops = &ceph_aops; 652355da1ebSSage Weil inode->i_mapping->backing_dev_info = 653640ef79dSCheng Renquan &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 654355da1ebSSage Weil 655355da1ebSSage Weil switch (inode->i_mode & S_IFMT) { 656355da1ebSSage Weil case S_IFIFO: 657355da1ebSSage Weil case S_IFBLK: 658355da1ebSSage Weil case S_IFCHR: 659355da1ebSSage Weil case S_IFSOCK: 660355da1ebSSage Weil init_special_inode(inode, inode->i_mode, inode->i_rdev); 661355da1ebSSage Weil inode->i_op = &ceph_file_iops; 662355da1ebSSage Weil break; 663355da1ebSSage Weil case S_IFREG: 664355da1ebSSage Weil inode->i_op = &ceph_file_iops; 665355da1ebSSage Weil inode->i_fop = &ceph_file_fops; 666355da1ebSSage Weil break; 667355da1ebSSage Weil case S_IFLNK: 668355da1ebSSage Weil inode->i_op = &ceph_symlink_iops; 669355da1ebSSage Weil if (!ci->i_symlink) { 670355da1ebSSage Weil int symlen = iinfo->symlink_len; 671355da1ebSSage Weil char *sym; 672355da1ebSSage Weil 673355da1ebSSage Weil BUG_ON(symlen != inode->i_size); 674355da1ebSSage Weil spin_unlock(&inode->i_lock); 675355da1ebSSage Weil 676355da1ebSSage Weil err = -ENOMEM; 677355da1ebSSage Weil sym = kmalloc(symlen+1, GFP_NOFS); 678355da1ebSSage Weil if (!sym) 679355da1ebSSage Weil goto out; 680355da1ebSSage Weil memcpy(sym, iinfo->symlink, symlen); 681355da1ebSSage Weil sym[symlen] = 0; 682355da1ebSSage Weil 683355da1ebSSage Weil spin_lock(&inode->i_lock); 684355da1ebSSage Weil if (!ci->i_symlink) 685355da1ebSSage Weil ci->i_symlink = sym; 686355da1ebSSage Weil else 687355da1ebSSage Weil kfree(sym); /* lost a race */ 688355da1ebSSage Weil } 689355da1ebSSage Weil break; 690355da1ebSSage Weil case S_IFDIR: 691355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 692355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 693355da1ebSSage Weil 69414303d20SSage Weil ci->i_dir_layout = iinfo->dir_layout; 69514303d20SSage Weil 696355da1ebSSage Weil ci->i_files = le64_to_cpu(info->files); 697355da1ebSSage Weil ci->i_subdirs = le64_to_cpu(info->subdirs); 698355da1ebSSage Weil ci->i_rbytes = le64_to_cpu(info->rbytes); 699355da1ebSSage Weil ci->i_rfiles = le64_to_cpu(info->rfiles); 700355da1ebSSage Weil ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 701355da1ebSSage Weil ceph_decode_timespec(&ci->i_rctime, &info->rctime); 702355da1ebSSage Weil 703355da1ebSSage Weil /* set dir completion flag? */ 704355da1ebSSage Weil if (ci->i_files == 0 && ci->i_subdirs == 0 && 705355da1ebSSage Weil ceph_snap(inode) == CEPH_NOSNAP && 7061b7facc4SSage Weil (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 70712451491SSage Weil (issued & CEPH_CAP_FILE_EXCL) == 0 && 7081b7facc4SSage Weil (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 709355da1ebSSage Weil dout(" marking %p complete (empty)\n", inode); 710b545cc15SSage Weil /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ 711355da1ebSSage Weil ci->i_max_offset = 2; 712355da1ebSSage Weil } 713355da1ebSSage Weil break; 714355da1ebSSage Weil default: 715355da1ebSSage Weil pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 716355da1ebSSage Weil ceph_vinop(inode), inode->i_mode); 717355da1ebSSage Weil } 718355da1ebSSage Weil 719355da1ebSSage Weil no_change: 720355da1ebSSage Weil spin_unlock(&inode->i_lock); 721355da1ebSSage Weil 722355da1ebSSage Weil /* queue truncate if we saw i_size decrease */ 723355da1ebSSage Weil if (queue_trunc) 7243c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 725355da1ebSSage Weil 726355da1ebSSage Weil /* populate frag tree */ 727355da1ebSSage Weil /* FIXME: move me up, if/when version reflects fragtree changes */ 728355da1ebSSage Weil nsplits = le32_to_cpu(info->fragtree.nsplits); 729355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 730355da1ebSSage Weil for (i = 0; i < nsplits; i++) { 731355da1ebSSage Weil u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 732355da1ebSSage Weil struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 733355da1ebSSage Weil 734355da1ebSSage Weil if (IS_ERR(frag)) 735355da1ebSSage Weil continue; 736355da1ebSSage Weil frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 737355da1ebSSage Weil dout(" frag %x split by %d\n", frag->frag, frag->split_by); 738355da1ebSSage Weil } 739355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 740355da1ebSSage Weil 741355da1ebSSage Weil /* were we issued a capability? */ 742355da1ebSSage Weil if (info->cap.caps) { 743355da1ebSSage Weil if (ceph_snap(inode) == CEPH_NOSNAP) { 744355da1ebSSage Weil ceph_add_cap(inode, session, 745355da1ebSSage Weil le64_to_cpu(info->cap.cap_id), 746355da1ebSSage Weil cap_fmode, 747355da1ebSSage Weil le32_to_cpu(info->cap.caps), 748355da1ebSSage Weil le32_to_cpu(info->cap.wanted), 749355da1ebSSage Weil le32_to_cpu(info->cap.seq), 750355da1ebSSage Weil le32_to_cpu(info->cap.mseq), 751355da1ebSSage Weil le64_to_cpu(info->cap.realm), 752355da1ebSSage Weil info->cap.flags, 753355da1ebSSage Weil caps_reservation); 754355da1ebSSage Weil } else { 755355da1ebSSage Weil spin_lock(&inode->i_lock); 756355da1ebSSage Weil dout(" %p got snap_caps %s\n", inode, 757355da1ebSSage Weil ceph_cap_string(le32_to_cpu(info->cap.caps))); 758355da1ebSSage Weil ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 759355da1ebSSage Weil if (cap_fmode >= 0) 760355da1ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 761355da1ebSSage Weil spin_unlock(&inode->i_lock); 762355da1ebSSage Weil } 76304d000ebSSage Weil } else if (cap_fmode >= 0) { 76404d000ebSSage Weil pr_warning("mds issued no caps on %llx.%llx\n", 76504d000ebSSage Weil ceph_vinop(inode)); 76604d000ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 767355da1ebSSage Weil } 768355da1ebSSage Weil 769355da1ebSSage Weil /* update delegation info? */ 770355da1ebSSage Weil if (dirinfo) 771355da1ebSSage Weil ceph_fill_dirfrag(inode, dirinfo); 772355da1ebSSage Weil 773355da1ebSSage Weil err = 0; 774355da1ebSSage Weil 775355da1ebSSage Weil out: 776b6c1d5b8SSage Weil if (xattr_blob) 777355da1ebSSage Weil ceph_buffer_put(xattr_blob); 778355da1ebSSage Weil return err; 779355da1ebSSage Weil } 780355da1ebSSage Weil 781355da1ebSSage Weil /* 782355da1ebSSage Weil * caller should hold session s_mutex. 783355da1ebSSage Weil */ 784355da1ebSSage Weil static void update_dentry_lease(struct dentry *dentry, 785355da1ebSSage Weil struct ceph_mds_reply_lease *lease, 786355da1ebSSage Weil struct ceph_mds_session *session, 787355da1ebSSage Weil unsigned long from_time) 788355da1ebSSage Weil { 789355da1ebSSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 790355da1ebSSage Weil long unsigned duration = le32_to_cpu(lease->duration_ms); 791355da1ebSSage Weil long unsigned ttl = from_time + (duration * HZ) / 1000; 792355da1ebSSage Weil long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 793355da1ebSSage Weil struct inode *dir; 794355da1ebSSage Weil 795355da1ebSSage Weil /* only track leases on regular dentries */ 796355da1ebSSage Weil if (dentry->d_op != &ceph_dentry_ops) 797355da1ebSSage Weil return; 798355da1ebSSage Weil 799355da1ebSSage Weil spin_lock(&dentry->d_lock); 800355da1ebSSage Weil dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 801355da1ebSSage Weil dentry, le16_to_cpu(lease->mask), duration, ttl); 802355da1ebSSage Weil 803355da1ebSSage Weil /* make lease_rdcache_gen match directory */ 804355da1ebSSage Weil dir = dentry->d_parent->d_inode; 805355da1ebSSage Weil di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 806355da1ebSSage Weil 807355da1ebSSage Weil if (lease->mask == 0) 808355da1ebSSage Weil goto out_unlock; 809355da1ebSSage Weil 810355da1ebSSage Weil if (di->lease_gen == session->s_cap_gen && 811355da1ebSSage Weil time_before(ttl, dentry->d_time)) 812355da1ebSSage Weil goto out_unlock; /* we already have a newer lease. */ 813355da1ebSSage Weil 814355da1ebSSage Weil if (di->lease_session && di->lease_session != session) 815355da1ebSSage Weil goto out_unlock; 816355da1ebSSage Weil 817355da1ebSSage Weil ceph_dentry_lru_touch(dentry); 818355da1ebSSage Weil 819355da1ebSSage Weil if (!di->lease_session) 820355da1ebSSage Weil di->lease_session = ceph_get_mds_session(session); 821355da1ebSSage Weil di->lease_gen = session->s_cap_gen; 822355da1ebSSage Weil di->lease_seq = le32_to_cpu(lease->seq); 823355da1ebSSage Weil di->lease_renew_after = half_ttl; 824355da1ebSSage Weil di->lease_renew_from = 0; 825355da1ebSSage Weil dentry->d_time = ttl; 826355da1ebSSage Weil out_unlock: 827355da1ebSSage Weil spin_unlock(&dentry->d_lock); 828355da1ebSSage Weil return; 829355da1ebSSage Weil } 830355da1ebSSage Weil 831355da1ebSSage Weil /* 8324baa75efSYehuda Sadeh * Set dentry's directory position based on the current dir's max, and 8334baa75efSYehuda Sadeh * order it in d_subdirs, so that dcache_readdir behaves. 8344baa75efSYehuda Sadeh */ 8354baa75efSYehuda Sadeh static void ceph_set_dentry_offset(struct dentry *dn) 8364baa75efSYehuda Sadeh { 8374baa75efSYehuda Sadeh struct dentry *dir = dn->d_parent; 8384baa75efSYehuda Sadeh struct inode *inode = dn->d_parent->d_inode; 8394baa75efSYehuda Sadeh struct ceph_dentry_info *di; 8404baa75efSYehuda Sadeh 8414baa75efSYehuda Sadeh BUG_ON(!inode); 8424baa75efSYehuda Sadeh 8434baa75efSYehuda Sadeh di = ceph_dentry(dn); 8444baa75efSYehuda Sadeh 8454baa75efSYehuda Sadeh spin_lock(&inode->i_lock); 846e8a74987SSage Weil if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 847e8a74987SSage Weil spin_unlock(&inode->i_lock); 848e8a74987SSage Weil return; 849e8a74987SSage Weil } 8504baa75efSYehuda Sadeh di->offset = ceph_inode(inode)->i_max_offset++; 8514baa75efSYehuda Sadeh spin_unlock(&inode->i_lock); 8524baa75efSYehuda Sadeh 8532fd6b7f5SNick Piggin spin_lock(&dir->d_lock); 8542fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 85513a4214cSHenry C Chang list_move(&dn->d_u.d_child, &dir->d_subdirs); 8564baa75efSYehuda Sadeh dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 8574baa75efSYehuda Sadeh dn->d_u.d_child.prev, dn->d_u.d_child.next); 8584baa75efSYehuda Sadeh spin_unlock(&dn->d_lock); 8592fd6b7f5SNick Piggin spin_unlock(&dir->d_lock); 8604baa75efSYehuda Sadeh } 8614baa75efSYehuda Sadeh 8624baa75efSYehuda Sadeh /* 8631cd3935bSSage Weil * splice a dentry to an inode. 8641cd3935bSSage Weil * caller must hold directory i_mutex for this to be safe. 8651cd3935bSSage Weil * 8661cd3935bSSage Weil * we will only rehash the resulting dentry if @prehash is 8671cd3935bSSage Weil * true; @prehash will be set to false (for the benefit of 8681cd3935bSSage Weil * the caller) if we fail. 8691cd3935bSSage Weil */ 8701cd3935bSSage Weil static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 871467c5251SSage Weil bool *prehash, bool set_offset) 8721cd3935bSSage Weil { 8731cd3935bSSage Weil struct dentry *realdn; 8741cd3935bSSage Weil 8751cd3935bSSage Weil BUG_ON(dn->d_inode); 8761cd3935bSSage Weil 8771cd3935bSSage Weil /* dn must be unhashed */ 8781cd3935bSSage Weil if (!d_unhashed(dn)) 8791cd3935bSSage Weil d_drop(dn); 8801cd3935bSSage Weil realdn = d_materialise_unique(dn, in); 8811cd3935bSSage Weil if (IS_ERR(realdn)) { 882d69ed05aSSage Weil pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 883d69ed05aSSage Weil PTR_ERR(realdn), dn, in, ceph_vinop(in)); 8841cd3935bSSage Weil if (prehash) 8851cd3935bSSage Weil *prehash = false; /* don't rehash on error */ 8861cd3935bSSage Weil dn = realdn; /* note realdn contains the error */ 8871cd3935bSSage Weil goto out; 8881cd3935bSSage Weil } else if (realdn) { 8891cd3935bSSage Weil dout("dn %p (%d) spliced with %p (%d) " 8901cd3935bSSage Weil "inode %p ino %llx.%llx\n", 891b7ab39f6SNick Piggin dn, dn->d_count, 892b7ab39f6SNick Piggin realdn, realdn->d_count, 8931cd3935bSSage Weil realdn->d_inode, ceph_vinop(realdn->d_inode)); 8941cd3935bSSage Weil dput(dn); 8951cd3935bSSage Weil dn = realdn; 8961cd3935bSSage Weil } else { 8971cd3935bSSage Weil BUG_ON(!ceph_dentry(dn)); 8981cd3935bSSage Weil dout("dn %p attached to %p ino %llx.%llx\n", 8991cd3935bSSage Weil dn, dn->d_inode, ceph_vinop(dn->d_inode)); 9001cd3935bSSage Weil } 9011cd3935bSSage Weil if ((!prehash || *prehash) && d_unhashed(dn)) 9021cd3935bSSage Weil d_rehash(dn); 903467c5251SSage Weil if (set_offset) 9041cd3935bSSage Weil ceph_set_dentry_offset(dn); 9051cd3935bSSage Weil out: 9061cd3935bSSage Weil return dn; 9071cd3935bSSage Weil } 9081cd3935bSSage Weil 9091cd3935bSSage Weil /* 910355da1ebSSage Weil * Incorporate results into the local cache. This is either just 911355da1ebSSage Weil * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 912355da1ebSSage Weil * after a lookup). 913355da1ebSSage Weil * 914355da1ebSSage Weil * A reply may contain 915355da1ebSSage Weil * a directory inode along with a dentry. 916355da1ebSSage Weil * and/or a target inode 917355da1ebSSage Weil * 918355da1ebSSage Weil * Called with snap_rwsem (read). 919355da1ebSSage Weil */ 920355da1ebSSage Weil int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 921355da1ebSSage Weil struct ceph_mds_session *session) 922355da1ebSSage Weil { 923355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 924355da1ebSSage Weil struct inode *in = NULL; 925355da1ebSSage Weil struct ceph_mds_reply_inode *ininfo; 926355da1ebSSage Weil struct ceph_vino vino; 9273d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 928355da1ebSSage Weil int i = 0; 929355da1ebSSage Weil int err = 0; 930355da1ebSSage Weil 931355da1ebSSage Weil dout("fill_trace %p is_dentry %d is_target %d\n", req, 932355da1ebSSage Weil rinfo->head->is_dentry, rinfo->head->is_target); 933355da1ebSSage Weil 934355da1ebSSage Weil #if 0 935355da1ebSSage Weil /* 936355da1ebSSage Weil * Debugging hook: 937355da1ebSSage Weil * 938355da1ebSSage Weil * If we resend completed ops to a recovering mds, we get no 939355da1ebSSage Weil * trace. Since that is very rare, pretend this is the case 940355da1ebSSage Weil * to ensure the 'no trace' handlers in the callers behave. 941355da1ebSSage Weil * 942355da1ebSSage Weil * Fill in inodes unconditionally to avoid breaking cap 943355da1ebSSage Weil * invariants. 944355da1ebSSage Weil */ 945355da1ebSSage Weil if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 946355da1ebSSage Weil pr_info("fill_trace faking empty trace on %lld %s\n", 947355da1ebSSage Weil req->r_tid, ceph_mds_op_name(rinfo->head->op)); 948355da1ebSSage Weil if (rinfo->head->is_dentry) { 949355da1ebSSage Weil rinfo->head->is_dentry = 0; 950355da1ebSSage Weil err = fill_inode(req->r_locked_dir, 951355da1ebSSage Weil &rinfo->diri, rinfo->dirfrag, 952355da1ebSSage Weil session, req->r_request_started, -1); 953355da1ebSSage Weil } 954355da1ebSSage Weil if (rinfo->head->is_target) { 955355da1ebSSage Weil rinfo->head->is_target = 0; 956355da1ebSSage Weil ininfo = rinfo->targeti.in; 957355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 958355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 959355da1ebSSage Weil in = ceph_get_inode(sb, vino); 960355da1ebSSage Weil err = fill_inode(in, &rinfo->targeti, NULL, 961355da1ebSSage Weil session, req->r_request_started, 962355da1ebSSage Weil req->r_fmode); 963355da1ebSSage Weil iput(in); 964355da1ebSSage Weil } 965355da1ebSSage Weil } 966355da1ebSSage Weil #endif 967355da1ebSSage Weil 968355da1ebSSage Weil if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 969355da1ebSSage Weil dout("fill_trace reply is empty!\n"); 970167c9e35SSage Weil if (rinfo->head->result == 0 && req->r_locked_dir) 971167c9e35SSage Weil ceph_invalidate_dir_request(req); 972355da1ebSSage Weil return 0; 973355da1ebSSage Weil } 974355da1ebSSage Weil 975355da1ebSSage Weil if (rinfo->head->is_dentry) { 9765b1daecdSSage Weil struct inode *dir = req->r_locked_dir; 9775b1daecdSSage Weil 9785b1daecdSSage Weil err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 9795b1daecdSSage Weil session, req->r_request_started, -1, 9805b1daecdSSage Weil &req->r_caps_reservation); 9815b1daecdSSage Weil if (err < 0) 9825b1daecdSSage Weil return err; 9835b1daecdSSage Weil } 9845b1daecdSSage Weil 9859358c6d4SSage Weil /* 9869358c6d4SSage Weil * ignore null lease/binding on snapdir ENOENT, or else we 9879358c6d4SSage Weil * will have trouble splicing in the virtual snapdir later 9889358c6d4SSage Weil */ 9899358c6d4SSage Weil if (rinfo->head->is_dentry && !req->r_aborted && 9909358c6d4SSage Weil (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 9913d14c5d2SYehuda Sadeh fsc->mount_options->snapdir_name, 9929358c6d4SSage Weil req->r_dentry->d_name.len))) { 993355da1ebSSage Weil /* 994355da1ebSSage Weil * lookup link rename : null -> possibly existing inode 995355da1ebSSage Weil * mknod symlink mkdir : null -> new inode 996355da1ebSSage Weil * unlink : linked -> null 997355da1ebSSage Weil */ 998355da1ebSSage Weil struct inode *dir = req->r_locked_dir; 999355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1000355da1ebSSage Weil bool have_dir_cap, have_lease; 1001355da1ebSSage Weil 1002355da1ebSSage Weil BUG_ON(!dn); 1003355da1ebSSage Weil BUG_ON(!dir); 1004355da1ebSSage Weil BUG_ON(dn->d_parent->d_inode != dir); 1005355da1ebSSage Weil BUG_ON(ceph_ino(dir) != 1006355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->ino)); 1007355da1ebSSage Weil BUG_ON(ceph_snap(dir) != 1008355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->snapid)); 1009355da1ebSSage Weil 1010355da1ebSSage Weil /* do we have a lease on the whole dir? */ 1011355da1ebSSage Weil have_dir_cap = 1012355da1ebSSage Weil (le32_to_cpu(rinfo->diri.in->cap.caps) & 1013355da1ebSSage Weil CEPH_CAP_FILE_SHARED); 1014355da1ebSSage Weil 1015355da1ebSSage Weil /* do we have a dn lease? */ 1016355da1ebSSage Weil have_lease = have_dir_cap || 1017355da1ebSSage Weil (le16_to_cpu(rinfo->dlease->mask) & 1018355da1ebSSage Weil CEPH_LOCK_DN); 1019355da1ebSSage Weil 1020355da1ebSSage Weil if (!have_lease) 1021355da1ebSSage Weil dout("fill_trace no dentry lease or dir cap\n"); 1022355da1ebSSage Weil 1023355da1ebSSage Weil /* rename? */ 1024355da1ebSSage Weil if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1025355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1026355da1ebSSage Weil req->r_old_dentry, 1027355da1ebSSage Weil req->r_old_dentry->d_name.len, 1028355da1ebSSage Weil req->r_old_dentry->d_name.name, 1029355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 1030355da1ebSSage Weil dout("fill_trace doing d_move %p -> %p\n", 1031355da1ebSSage Weil req->r_old_dentry, dn); 1032c10f5e12SSage Weil 1033c10f5e12SSage Weil /* d_move screws up d_subdirs order */ 1034c10f5e12SSage Weil ceph_i_clear(dir, CEPH_I_COMPLETE); 1035c10f5e12SSage Weil 1036355da1ebSSage Weil d_move(req->r_old_dentry, dn); 1037355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1038355da1ebSSage Weil req->r_old_dentry, 1039355da1ebSSage Weil req->r_old_dentry->d_name.len, 1040355da1ebSSage Weil req->r_old_dentry->d_name.name, 1041355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 104281a6cf2dSSage Weil 1043c4a29f26SSage Weil /* ensure target dentry is invalidated, despite 1044c4a29f26SSage Weil rehashing bug in vfs_rename_dir */ 104581a6cf2dSSage Weil ceph_invalidate_dentry_lease(dn); 104681a6cf2dSSage Weil 1047355da1ebSSage Weil /* take overwritten dentry's readdir offset */ 10481cd3935bSSage Weil dout("dn %p gets %p offset %lld (old offset %lld)\n", 10491cd3935bSSage Weil req->r_old_dentry, dn, ceph_dentry(dn)->offset, 10501cd3935bSSage Weil ceph_dentry(req->r_old_dentry)->offset); 1051355da1ebSSage Weil ceph_dentry(req->r_old_dentry)->offset = 1052355da1ebSSage Weil ceph_dentry(dn)->offset; 105381a6cf2dSSage Weil 1054355da1ebSSage Weil dn = req->r_old_dentry; /* use old_dentry */ 1055355da1ebSSage Weil in = dn->d_inode; 1056355da1ebSSage Weil } 1057355da1ebSSage Weil 1058355da1ebSSage Weil /* null dentry? */ 1059355da1ebSSage Weil if (!rinfo->head->is_target) { 1060355da1ebSSage Weil dout("fill_trace null dentry\n"); 1061355da1ebSSage Weil if (dn->d_inode) { 1062355da1ebSSage Weil dout("d_delete %p\n", dn); 1063355da1ebSSage Weil d_delete(dn); 1064355da1ebSSage Weil } else { 1065355da1ebSSage Weil dout("d_instantiate %p NULL\n", dn); 1066355da1ebSSage Weil d_instantiate(dn, NULL); 1067355da1ebSSage Weil if (have_lease && d_unhashed(dn)) 1068355da1ebSSage Weil d_rehash(dn); 1069355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, 1070355da1ebSSage Weil session, 1071355da1ebSSage Weil req->r_request_started); 1072355da1ebSSage Weil } 1073355da1ebSSage Weil goto done; 1074355da1ebSSage Weil } 1075355da1ebSSage Weil 1076355da1ebSSage Weil /* attach proper inode */ 1077355da1ebSSage Weil ininfo = rinfo->targeti.in; 1078355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1079355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1080d8b16b3dSSage Weil in = dn->d_inode; 1081d8b16b3dSSage Weil if (!in) { 1082355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1083355da1ebSSage Weil if (IS_ERR(in)) { 1084355da1ebSSage Weil pr_err("fill_trace bad get_inode " 1085355da1ebSSage Weil "%llx.%llx\n", vino.ino, vino.snap); 1086355da1ebSSage Weil err = PTR_ERR(in); 1087355da1ebSSage Weil d_delete(dn); 1088355da1ebSSage Weil goto done; 1089355da1ebSSage Weil } 1090467c5251SSage Weil dn = splice_dentry(dn, in, &have_lease, true); 1091355da1ebSSage Weil if (IS_ERR(dn)) { 1092355da1ebSSage Weil err = PTR_ERR(dn); 1093355da1ebSSage Weil goto done; 1094355da1ebSSage Weil } 1095355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 1096355da1ebSSage Weil igrab(in); 1097355da1ebSSage Weil } else if (ceph_ino(in) == vino.ino && 1098355da1ebSSage Weil ceph_snap(in) == vino.snap) { 1099355da1ebSSage Weil igrab(in); 1100355da1ebSSage Weil } else { 1101355da1ebSSage Weil dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1102355da1ebSSage Weil dn, in, ceph_ino(in), ceph_snap(in), 1103355da1ebSSage Weil vino.ino, vino.snap); 1104355da1ebSSage Weil have_lease = false; 1105355da1ebSSage Weil in = NULL; 1106355da1ebSSage Weil } 1107355da1ebSSage Weil 1108355da1ebSSage Weil if (have_lease) 1109355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, session, 1110355da1ebSSage Weil req->r_request_started); 1111355da1ebSSage Weil dout(" final dn %p\n", dn); 1112355da1ebSSage Weil i++; 1113355da1ebSSage Weil } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1114355da1ebSSage Weil req->r_op == CEPH_MDS_OP_MKSNAP) { 1115355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1116355da1ebSSage Weil 1117355da1ebSSage Weil /* fill out a snapdir LOOKUPSNAP dentry */ 1118355da1ebSSage Weil BUG_ON(!dn); 1119355da1ebSSage Weil BUG_ON(!req->r_locked_dir); 1120355da1ebSSage Weil BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1121355da1ebSSage Weil ininfo = rinfo->targeti.in; 1122355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1123355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1124355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1125355da1ebSSage Weil if (IS_ERR(in)) { 1126355da1ebSSage Weil pr_err("fill_inode get_inode badness %llx.%llx\n", 1127355da1ebSSage Weil vino.ino, vino.snap); 1128355da1ebSSage Weil err = PTR_ERR(in); 1129355da1ebSSage Weil d_delete(dn); 1130355da1ebSSage Weil goto done; 1131355da1ebSSage Weil } 1132355da1ebSSage Weil dout(" linking snapped dir %p to dn %p\n", in, dn); 1133467c5251SSage Weil dn = splice_dentry(dn, in, NULL, true); 1134355da1ebSSage Weil if (IS_ERR(dn)) { 1135355da1ebSSage Weil err = PTR_ERR(dn); 1136355da1ebSSage Weil goto done; 1137355da1ebSSage Weil } 1138355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 1139355da1ebSSage Weil igrab(in); 1140355da1ebSSage Weil rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1141355da1ebSSage Weil } 1142355da1ebSSage Weil 1143355da1ebSSage Weil if (rinfo->head->is_target) { 1144355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1145355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1146355da1ebSSage Weil 1147355da1ebSSage Weil if (in == NULL || ceph_ino(in) != vino.ino || 1148355da1ebSSage Weil ceph_snap(in) != vino.snap) { 1149355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1150355da1ebSSage Weil if (IS_ERR(in)) { 1151355da1ebSSage Weil err = PTR_ERR(in); 1152355da1ebSSage Weil goto done; 1153355da1ebSSage Weil } 1154355da1ebSSage Weil } 1155355da1ebSSage Weil req->r_target_inode = in; 1156355da1ebSSage Weil 1157355da1ebSSage Weil err = fill_inode(in, 1158355da1ebSSage Weil &rinfo->targeti, NULL, 1159355da1ebSSage Weil session, req->r_request_started, 1160355da1ebSSage Weil (le32_to_cpu(rinfo->head->result) == 0) ? 1161355da1ebSSage Weil req->r_fmode : -1, 1162355da1ebSSage Weil &req->r_caps_reservation); 1163355da1ebSSage Weil if (err < 0) { 1164355da1ebSSage Weil pr_err("fill_inode badness %p %llx.%llx\n", 1165355da1ebSSage Weil in, ceph_vinop(in)); 1166355da1ebSSage Weil goto done; 1167355da1ebSSage Weil } 1168355da1ebSSage Weil } 1169355da1ebSSage Weil 1170355da1ebSSage Weil done: 1171355da1ebSSage Weil dout("fill_trace done err=%d\n", err); 1172355da1ebSSage Weil return err; 1173355da1ebSSage Weil } 1174355da1ebSSage Weil 1175355da1ebSSage Weil /* 1176355da1ebSSage Weil * Prepopulate our cache with readdir results, leases, etc. 1177355da1ebSSage Weil */ 1178355da1ebSSage Weil int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1179355da1ebSSage Weil struct ceph_mds_session *session) 1180355da1ebSSage Weil { 1181355da1ebSSage Weil struct dentry *parent = req->r_dentry; 1182355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1183355da1ebSSage Weil struct qstr dname; 1184355da1ebSSage Weil struct dentry *dn; 1185355da1ebSSage Weil struct inode *in; 1186355da1ebSSage Weil int err = 0, i; 1187355da1ebSSage Weil struct inode *snapdir = NULL; 1188355da1ebSSage Weil struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1189355da1ebSSage Weil u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1190355da1ebSSage Weil struct ceph_dentry_info *di; 1191355da1ebSSage Weil 1192355da1ebSSage Weil if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1193355da1ebSSage Weil snapdir = ceph_get_snapdir(parent->d_inode); 1194355da1ebSSage Weil parent = d_find_alias(snapdir); 1195355da1ebSSage Weil dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1196355da1ebSSage Weil rinfo->dir_nr, parent); 1197355da1ebSSage Weil } else { 1198355da1ebSSage Weil dout("readdir_prepopulate %d items under dn %p\n", 1199355da1ebSSage Weil rinfo->dir_nr, parent); 1200355da1ebSSage Weil if (rinfo->dir_dir) 1201355da1ebSSage Weil ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1202355da1ebSSage Weil } 1203355da1ebSSage Weil 1204355da1ebSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 1205355da1ebSSage Weil struct ceph_vino vino; 1206355da1ebSSage Weil 1207355da1ebSSage Weil dname.name = rinfo->dir_dname[i]; 1208355da1ebSSage Weil dname.len = rinfo->dir_dname_len[i]; 1209355da1ebSSage Weil dname.hash = full_name_hash(dname.name, dname.len); 1210355da1ebSSage Weil 1211355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1212355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1213355da1ebSSage Weil 1214355da1ebSSage Weil retry_lookup: 1215355da1ebSSage Weil dn = d_lookup(parent, &dname); 1216355da1ebSSage Weil dout("d_lookup on parent=%p name=%.*s got %p\n", 1217355da1ebSSage Weil parent, dname.len, dname.name, dn); 1218355da1ebSSage Weil 1219355da1ebSSage Weil if (!dn) { 1220355da1ebSSage Weil dn = d_alloc(parent, &dname); 1221355da1ebSSage Weil dout("d_alloc %p '%.*s' = %p\n", parent, 1222355da1ebSSage Weil dname.len, dname.name, dn); 1223355da1ebSSage Weil if (dn == NULL) { 1224355da1ebSSage Weil dout("d_alloc badness\n"); 1225355da1ebSSage Weil err = -ENOMEM; 1226355da1ebSSage Weil goto out; 1227355da1ebSSage Weil } 1228355da1ebSSage Weil err = ceph_init_dentry(dn); 12298c696737SSage Weil if (err < 0) { 12308c696737SSage Weil dput(dn); 1231355da1ebSSage Weil goto out; 12328c696737SSage Weil } 1233355da1ebSSage Weil } else if (dn->d_inode && 1234355da1ebSSage Weil (ceph_ino(dn->d_inode) != vino.ino || 1235355da1ebSSage Weil ceph_snap(dn->d_inode) != vino.snap)) { 1236355da1ebSSage Weil dout(" dn %p points to wrong inode %p\n", 1237355da1ebSSage Weil dn, dn->d_inode); 1238355da1ebSSage Weil d_delete(dn); 1239355da1ebSSage Weil dput(dn); 1240355da1ebSSage Weil goto retry_lookup; 1241355da1ebSSage Weil } else { 1242355da1ebSSage Weil /* reorder parent's d_subdirs */ 12432fd6b7f5SNick Piggin spin_lock(&parent->d_lock); 12442fd6b7f5SNick Piggin spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1245355da1ebSSage Weil list_move(&dn->d_u.d_child, &parent->d_subdirs); 1246355da1ebSSage Weil spin_unlock(&dn->d_lock); 12472fd6b7f5SNick Piggin spin_unlock(&parent->d_lock); 1248355da1ebSSage Weil } 1249355da1ebSSage Weil 1250355da1ebSSage Weil di = dn->d_fsdata; 1251355da1ebSSage Weil di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1252355da1ebSSage Weil 1253355da1ebSSage Weil /* inode */ 1254355da1ebSSage Weil if (dn->d_inode) { 1255355da1ebSSage Weil in = dn->d_inode; 1256355da1ebSSage Weil } else { 1257355da1ebSSage Weil in = ceph_get_inode(parent->d_sb, vino); 1258ac1f12efSDan Carpenter if (IS_ERR(in)) { 1259355da1ebSSage Weil dout("new_inode badness\n"); 1260355da1ebSSage Weil d_delete(dn); 1261355da1ebSSage Weil dput(dn); 1262ac1f12efSDan Carpenter err = PTR_ERR(in); 1263355da1ebSSage Weil goto out; 1264355da1ebSSage Weil } 1265467c5251SSage Weil dn = splice_dentry(dn, in, NULL, false); 1266d69ed05aSSage Weil if (IS_ERR(dn)) 1267d69ed05aSSage Weil dn = NULL; 1268355da1ebSSage Weil } 1269355da1ebSSage Weil 1270355da1ebSSage Weil if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1271355da1ebSSage Weil req->r_request_started, -1, 1272355da1ebSSage Weil &req->r_caps_reservation) < 0) { 1273355da1ebSSage Weil pr_err("fill_inode badness on %p\n", in); 1274d69ed05aSSage Weil goto next_item; 1275355da1ebSSage Weil } 1276d69ed05aSSage Weil if (dn) 1277355da1ebSSage Weil update_dentry_lease(dn, rinfo->dir_dlease[i], 1278d69ed05aSSage Weil req->r_session, 1279d69ed05aSSage Weil req->r_request_started); 1280d69ed05aSSage Weil next_item: 1281d69ed05aSSage Weil if (dn) 1282355da1ebSSage Weil dput(dn); 1283355da1ebSSage Weil } 1284355da1ebSSage Weil req->r_did_prepopulate = true; 1285355da1ebSSage Weil 1286355da1ebSSage Weil out: 1287355da1ebSSage Weil if (snapdir) { 1288355da1ebSSage Weil iput(snapdir); 1289355da1ebSSage Weil dput(parent); 1290355da1ebSSage Weil } 1291355da1ebSSage Weil dout("readdir_prepopulate done\n"); 1292355da1ebSSage Weil return err; 1293355da1ebSSage Weil } 1294355da1ebSSage Weil 1295355da1ebSSage Weil int ceph_inode_set_size(struct inode *inode, loff_t size) 1296355da1ebSSage Weil { 1297355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1298355da1ebSSage Weil int ret = 0; 1299355da1ebSSage Weil 1300355da1ebSSage Weil spin_lock(&inode->i_lock); 1301355da1ebSSage Weil dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1302355da1ebSSage Weil inode->i_size = size; 1303355da1ebSSage Weil inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1304355da1ebSSage Weil 1305355da1ebSSage Weil /* tell the MDS if we are approaching max_size */ 1306355da1ebSSage Weil if ((size << 1) >= ci->i_max_size && 1307355da1ebSSage Weil (ci->i_reported_size << 1) < ci->i_max_size) 1308355da1ebSSage Weil ret = 1; 1309355da1ebSSage Weil 1310355da1ebSSage Weil spin_unlock(&inode->i_lock); 1311355da1ebSSage Weil return ret; 1312355da1ebSSage Weil } 1313355da1ebSSage Weil 1314355da1ebSSage Weil /* 1315355da1ebSSage Weil * Write back inode data in a worker thread. (This can't be done 1316355da1ebSSage Weil * in the message handler context.) 1317355da1ebSSage Weil */ 13183c6f6b79SSage Weil void ceph_queue_writeback(struct inode *inode) 13193c6f6b79SSage Weil { 13203c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->wb_wq, 13213c6f6b79SSage Weil &ceph_inode(inode)->i_wb_work)) { 13222c27c9a5SSage Weil dout("ceph_queue_writeback %p\n", inode); 13233c6f6b79SSage Weil igrab(inode); 13243c6f6b79SSage Weil } else { 13252c27c9a5SSage Weil dout("ceph_queue_writeback %p failed\n", inode); 13263c6f6b79SSage Weil } 13273c6f6b79SSage Weil } 13283c6f6b79SSage Weil 13293c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work) 1330355da1ebSSage Weil { 1331355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1332355da1ebSSage Weil i_wb_work); 1333355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1334355da1ebSSage Weil 1335355da1ebSSage Weil dout("writeback %p\n", inode); 1336355da1ebSSage Weil filemap_fdatawrite(&inode->i_data); 1337355da1ebSSage Weil iput(inode); 1338355da1ebSSage Weil } 1339355da1ebSSage Weil 1340355da1ebSSage Weil /* 13413c6f6b79SSage Weil * queue an async invalidation 13423c6f6b79SSage Weil */ 13433c6f6b79SSage Weil void ceph_queue_invalidate(struct inode *inode) 13443c6f6b79SSage Weil { 13453c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 13463c6f6b79SSage Weil &ceph_inode(inode)->i_pg_inv_work)) { 13473c6f6b79SSage Weil dout("ceph_queue_invalidate %p\n", inode); 13483c6f6b79SSage Weil igrab(inode); 13493c6f6b79SSage Weil } else { 13503c6f6b79SSage Weil dout("ceph_queue_invalidate %p failed\n", inode); 13513c6f6b79SSage Weil } 13523c6f6b79SSage Weil } 13533c6f6b79SSage Weil 13543c6f6b79SSage Weil /* 1355c9af9fb6SYehuda Sadeh * invalidate any pages that are not dirty or under writeback. this 1356c9af9fb6SYehuda Sadeh * includes pages that are clean and mapped. 1357c9af9fb6SYehuda Sadeh */ 1358c9af9fb6SYehuda Sadeh static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1359c9af9fb6SYehuda Sadeh { 1360c9af9fb6SYehuda Sadeh struct pagevec pvec; 1361c9af9fb6SYehuda Sadeh pgoff_t next = 0; 1362c9af9fb6SYehuda Sadeh int i; 1363c9af9fb6SYehuda Sadeh 1364c9af9fb6SYehuda Sadeh pagevec_init(&pvec, 0); 1365c9af9fb6SYehuda Sadeh while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1366c9af9fb6SYehuda Sadeh for (i = 0; i < pagevec_count(&pvec); i++) { 1367c9af9fb6SYehuda Sadeh struct page *page = pvec.pages[i]; 1368c9af9fb6SYehuda Sadeh pgoff_t index; 1369c9af9fb6SYehuda Sadeh int skip_page = 1370c9af9fb6SYehuda Sadeh (PageDirty(page) || PageWriteback(page)); 1371c9af9fb6SYehuda Sadeh 1372c9af9fb6SYehuda Sadeh if (!skip_page) 1373c9af9fb6SYehuda Sadeh skip_page = !trylock_page(page); 1374c9af9fb6SYehuda Sadeh 1375c9af9fb6SYehuda Sadeh /* 1376c9af9fb6SYehuda Sadeh * We really shouldn't be looking at the ->index of an 1377c9af9fb6SYehuda Sadeh * unlocked page. But we're not allowed to lock these 1378c9af9fb6SYehuda Sadeh * pages. So we rely upon nobody altering the ->index 1379c9af9fb6SYehuda Sadeh * of this (pinned-by-us) page. 1380c9af9fb6SYehuda Sadeh */ 1381c9af9fb6SYehuda Sadeh index = page->index; 1382c9af9fb6SYehuda Sadeh if (index > next) 1383c9af9fb6SYehuda Sadeh next = index; 1384c9af9fb6SYehuda Sadeh next++; 1385c9af9fb6SYehuda Sadeh 1386c9af9fb6SYehuda Sadeh if (skip_page) 1387c9af9fb6SYehuda Sadeh continue; 1388c9af9fb6SYehuda Sadeh 1389c9af9fb6SYehuda Sadeh generic_error_remove_page(mapping, page); 1390c9af9fb6SYehuda Sadeh unlock_page(page); 1391c9af9fb6SYehuda Sadeh } 1392c9af9fb6SYehuda Sadeh pagevec_release(&pvec); 1393c9af9fb6SYehuda Sadeh cond_resched(); 1394c9af9fb6SYehuda Sadeh } 1395c9af9fb6SYehuda Sadeh } 1396c9af9fb6SYehuda Sadeh 1397c9af9fb6SYehuda Sadeh /* 1398355da1ebSSage Weil * Invalidate inode pages in a worker thread. (This can't be done 1399355da1ebSSage Weil * in the message handler context.) 1400355da1ebSSage Weil */ 14013c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work) 1402355da1ebSSage Weil { 1403355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1404355da1ebSSage Weil i_pg_inv_work); 1405355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1406355da1ebSSage Weil u32 orig_gen; 1407355da1ebSSage Weil int check = 0; 1408355da1ebSSage Weil 1409355da1ebSSage Weil spin_lock(&inode->i_lock); 1410355da1ebSSage Weil dout("invalidate_pages %p gen %d revoking %d\n", inode, 1411355da1ebSSage Weil ci->i_rdcache_gen, ci->i_rdcache_revoking); 1412cd045cb4SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1413355da1ebSSage Weil /* nevermind! */ 1414355da1ebSSage Weil spin_unlock(&inode->i_lock); 1415355da1ebSSage Weil goto out; 1416355da1ebSSage Weil } 1417355da1ebSSage Weil orig_gen = ci->i_rdcache_gen; 1418355da1ebSSage Weil spin_unlock(&inode->i_lock); 1419355da1ebSSage Weil 1420c9af9fb6SYehuda Sadeh ceph_invalidate_nondirty_pages(inode->i_mapping); 1421355da1ebSSage Weil 1422355da1ebSSage Weil spin_lock(&inode->i_lock); 1423cd045cb4SSage Weil if (orig_gen == ci->i_rdcache_gen && 1424cd045cb4SSage Weil orig_gen == ci->i_rdcache_revoking) { 1425355da1ebSSage Weil dout("invalidate_pages %p gen %d successful\n", inode, 1426355da1ebSSage Weil ci->i_rdcache_gen); 1427cd045cb4SSage Weil ci->i_rdcache_revoking--; 1428355da1ebSSage Weil check = 1; 1429355da1ebSSage Weil } else { 1430cd045cb4SSage Weil dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1431cd045cb4SSage Weil inode, orig_gen, ci->i_rdcache_gen, 1432cd045cb4SSage Weil ci->i_rdcache_revoking); 1433355da1ebSSage Weil } 1434355da1ebSSage Weil spin_unlock(&inode->i_lock); 1435355da1ebSSage Weil 1436355da1ebSSage Weil if (check) 1437355da1ebSSage Weil ceph_check_caps(ci, 0, NULL); 1438355da1ebSSage Weil out: 1439355da1ebSSage Weil iput(inode); 1440355da1ebSSage Weil } 1441355da1ebSSage Weil 1442355da1ebSSage Weil 1443355da1ebSSage Weil /* 1444355da1ebSSage Weil * called by trunc_wq; take i_mutex ourselves 1445355da1ebSSage Weil * 1446355da1ebSSage Weil * We also truncate in a separate thread as well. 1447355da1ebSSage Weil */ 14483c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work) 1449355da1ebSSage Weil { 1450355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1451355da1ebSSage Weil i_vmtruncate_work); 1452355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1453355da1ebSSage Weil 1454355da1ebSSage Weil dout("vmtruncate_work %p\n", inode); 1455355da1ebSSage Weil mutex_lock(&inode->i_mutex); 1456355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1457355da1ebSSage Weil mutex_unlock(&inode->i_mutex); 1458355da1ebSSage Weil iput(inode); 1459355da1ebSSage Weil } 1460355da1ebSSage Weil 1461355da1ebSSage Weil /* 14623c6f6b79SSage Weil * Queue an async vmtruncate. If we fail to queue work, we will handle 14633c6f6b79SSage Weil * the truncation the next time we call __ceph_do_pending_vmtruncate. 14643c6f6b79SSage Weil */ 14653c6f6b79SSage Weil void ceph_queue_vmtruncate(struct inode *inode) 14663c6f6b79SSage Weil { 14673c6f6b79SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14683c6f6b79SSage Weil 1469640ef79dSCheng Renquan if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 14703c6f6b79SSage Weil &ci->i_vmtruncate_work)) { 14713c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p\n", inode); 14723c6f6b79SSage Weil igrab(inode); 14733c6f6b79SSage Weil } else { 14743c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 14753c6f6b79SSage Weil inode, ci->i_truncate_pending); 14763c6f6b79SSage Weil } 14773c6f6b79SSage Weil } 14783c6f6b79SSage Weil 14793c6f6b79SSage Weil /* 1480355da1ebSSage Weil * called with i_mutex held. 1481355da1ebSSage Weil * 1482355da1ebSSage Weil * Make sure any pending truncation is applied before doing anything 1483355da1ebSSage Weil * that may depend on it. 1484355da1ebSSage Weil */ 1485355da1ebSSage Weil void __ceph_do_pending_vmtruncate(struct inode *inode) 1486355da1ebSSage Weil { 1487355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1488355da1ebSSage Weil u64 to; 1489355da1ebSSage Weil int wrbuffer_refs, wake = 0; 1490355da1ebSSage Weil 1491355da1ebSSage Weil retry: 1492355da1ebSSage Weil spin_lock(&inode->i_lock); 1493355da1ebSSage Weil if (ci->i_truncate_pending == 0) { 1494355da1ebSSage Weil dout("__do_pending_vmtruncate %p none pending\n", inode); 1495355da1ebSSage Weil spin_unlock(&inode->i_lock); 1496355da1ebSSage Weil return; 1497355da1ebSSage Weil } 1498355da1ebSSage Weil 1499355da1ebSSage Weil /* 1500355da1ebSSage Weil * make sure any dirty snapped pages are flushed before we 1501355da1ebSSage Weil * possibly truncate them.. so write AND block! 1502355da1ebSSage Weil */ 1503355da1ebSSage Weil if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1504355da1ebSSage Weil dout("__do_pending_vmtruncate %p flushing snaps first\n", 1505355da1ebSSage Weil inode); 1506355da1ebSSage Weil spin_unlock(&inode->i_lock); 1507355da1ebSSage Weil filemap_write_and_wait_range(&inode->i_data, 0, 1508355da1ebSSage Weil inode->i_sb->s_maxbytes); 1509355da1ebSSage Weil goto retry; 1510355da1ebSSage Weil } 1511355da1ebSSage Weil 1512355da1ebSSage Weil to = ci->i_truncate_size; 1513355da1ebSSage Weil wrbuffer_refs = ci->i_wrbuffer_ref; 1514355da1ebSSage Weil dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1515355da1ebSSage Weil ci->i_truncate_pending, to); 1516355da1ebSSage Weil spin_unlock(&inode->i_lock); 1517355da1ebSSage Weil 1518355da1ebSSage Weil truncate_inode_pages(inode->i_mapping, to); 1519355da1ebSSage Weil 1520355da1ebSSage Weil spin_lock(&inode->i_lock); 1521355da1ebSSage Weil ci->i_truncate_pending--; 1522355da1ebSSage Weil if (ci->i_truncate_pending == 0) 1523355da1ebSSage Weil wake = 1; 1524355da1ebSSage Weil spin_unlock(&inode->i_lock); 1525355da1ebSSage Weil 1526355da1ebSSage Weil if (wrbuffer_refs == 0) 1527355da1ebSSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1528355da1ebSSage Weil if (wake) 152903066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1530355da1ebSSage Weil } 1531355da1ebSSage Weil 1532355da1ebSSage Weil 1533355da1ebSSage Weil /* 1534355da1ebSSage Weil * symlinks 1535355da1ebSSage Weil */ 1536355da1ebSSage Weil static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1537355da1ebSSage Weil { 1538355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1539355da1ebSSage Weil nd_set_link(nd, ci->i_symlink); 1540355da1ebSSage Weil return NULL; 1541355da1ebSSage Weil } 1542355da1ebSSage Weil 1543355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops = { 1544355da1ebSSage Weil .readlink = generic_readlink, 1545355da1ebSSage Weil .follow_link = ceph_sym_follow_link, 1546355da1ebSSage Weil }; 1547355da1ebSSage Weil 1548355da1ebSSage Weil /* 1549355da1ebSSage Weil * setattr 1550355da1ebSSage Weil */ 1551355da1ebSSage Weil int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1552355da1ebSSage Weil { 1553355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1554355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1555355da1ebSSage Weil struct inode *parent_inode = dentry->d_parent->d_inode; 1556355da1ebSSage Weil const unsigned int ia_valid = attr->ia_valid; 1557355da1ebSSage Weil struct ceph_mds_request *req; 15583d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1559355da1ebSSage Weil int issued; 1560355da1ebSSage Weil int release = 0, dirtied = 0; 1561355da1ebSSage Weil int mask = 0; 1562355da1ebSSage Weil int err = 0; 1563355da1ebSSage Weil 1564355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1565355da1ebSSage Weil return -EROFS; 1566355da1ebSSage Weil 1567355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1568355da1ebSSage Weil 1569355da1ebSSage Weil err = inode_change_ok(inode, attr); 1570355da1ebSSage Weil if (err != 0) 1571355da1ebSSage Weil return err; 1572355da1ebSSage Weil 1573355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1574355da1ebSSage Weil USE_AUTH_MDS); 1575355da1ebSSage Weil if (IS_ERR(req)) 1576355da1ebSSage Weil return PTR_ERR(req); 1577355da1ebSSage Weil 1578355da1ebSSage Weil spin_lock(&inode->i_lock); 1579355da1ebSSage Weil issued = __ceph_caps_issued(ci, NULL); 1580355da1ebSSage Weil dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1581355da1ebSSage Weil 1582355da1ebSSage Weil if (ia_valid & ATTR_UID) { 1583355da1ebSSage Weil dout("setattr %p uid %d -> %d\n", inode, 1584355da1ebSSage Weil inode->i_uid, attr->ia_uid); 1585355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1586355da1ebSSage Weil inode->i_uid = attr->ia_uid; 1587355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1588355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1589355da1ebSSage Weil attr->ia_uid != inode->i_uid) { 1590355da1ebSSage Weil req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1591355da1ebSSage Weil mask |= CEPH_SETATTR_UID; 1592355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1593355da1ebSSage Weil } 1594355da1ebSSage Weil } 1595355da1ebSSage Weil if (ia_valid & ATTR_GID) { 1596355da1ebSSage Weil dout("setattr %p gid %d -> %d\n", inode, 1597355da1ebSSage Weil inode->i_gid, attr->ia_gid); 1598355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1599355da1ebSSage Weil inode->i_gid = attr->ia_gid; 1600355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1601355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1602355da1ebSSage Weil attr->ia_gid != inode->i_gid) { 1603355da1ebSSage Weil req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1604355da1ebSSage Weil mask |= CEPH_SETATTR_GID; 1605355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1606355da1ebSSage Weil } 1607355da1ebSSage Weil } 1608355da1ebSSage Weil if (ia_valid & ATTR_MODE) { 1609355da1ebSSage Weil dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1610355da1ebSSage Weil attr->ia_mode); 1611355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1612355da1ebSSage Weil inode->i_mode = attr->ia_mode; 1613355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1614355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1615355da1ebSSage Weil attr->ia_mode != inode->i_mode) { 1616355da1ebSSage Weil req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1617355da1ebSSage Weil mask |= CEPH_SETATTR_MODE; 1618355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1619355da1ebSSage Weil } 1620355da1ebSSage Weil } 1621355da1ebSSage Weil 1622355da1ebSSage Weil if (ia_valid & ATTR_ATIME) { 1623355da1ebSSage Weil dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1624355da1ebSSage Weil inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1625355da1ebSSage Weil attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1626355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1627355da1ebSSage Weil ci->i_time_warp_seq++; 1628355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1629355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1630355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1631355da1ebSSage Weil timespec_compare(&inode->i_atime, 1632355da1ebSSage Weil &attr->ia_atime) < 0) { 1633355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1634355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1635355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1636355da1ebSSage Weil !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1637355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.atime, 1638355da1ebSSage Weil &attr->ia_atime); 1639355da1ebSSage Weil mask |= CEPH_SETATTR_ATIME; 1640355da1ebSSage Weil release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1641355da1ebSSage Weil CEPH_CAP_FILE_WR; 1642355da1ebSSage Weil } 1643355da1ebSSage Weil } 1644355da1ebSSage Weil if (ia_valid & ATTR_MTIME) { 1645355da1ebSSage Weil dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1646355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1647355da1ebSSage Weil attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1648355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1649355da1ebSSage Weil ci->i_time_warp_seq++; 1650355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1651355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1652355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1653355da1ebSSage Weil timespec_compare(&inode->i_mtime, 1654355da1ebSSage Weil &attr->ia_mtime) < 0) { 1655355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1656355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1657355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1658355da1ebSSage Weil !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1659355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.mtime, 1660355da1ebSSage Weil &attr->ia_mtime); 1661355da1ebSSage Weil mask |= CEPH_SETATTR_MTIME; 1662355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1663355da1ebSSage Weil CEPH_CAP_FILE_WR; 1664355da1ebSSage Weil } 1665355da1ebSSage Weil } 1666355da1ebSSage Weil if (ia_valid & ATTR_SIZE) { 1667355da1ebSSage Weil dout("setattr %p size %lld -> %lld\n", inode, 1668355da1ebSSage Weil inode->i_size, attr->ia_size); 1669355da1ebSSage Weil if (attr->ia_size > inode->i_sb->s_maxbytes) { 1670355da1ebSSage Weil err = -EINVAL; 1671355da1ebSSage Weil goto out; 1672355da1ebSSage Weil } 1673355da1ebSSage Weil if ((issued & CEPH_CAP_FILE_EXCL) && 1674355da1ebSSage Weil attr->ia_size > inode->i_size) { 1675355da1ebSSage Weil inode->i_size = attr->ia_size; 1676355da1ebSSage Weil inode->i_blocks = 1677355da1ebSSage Weil (attr->ia_size + (1 << 9) - 1) >> 9; 1678355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1679355da1ebSSage Weil ci->i_reported_size = attr->ia_size; 1680355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1681355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1682355da1ebSSage Weil attr->ia_size != inode->i_size) { 1683355da1ebSSage Weil req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1684355da1ebSSage Weil req->r_args.setattr.old_size = 1685355da1ebSSage Weil cpu_to_le64(inode->i_size); 1686355da1ebSSage Weil mask |= CEPH_SETATTR_SIZE; 1687355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1688355da1ebSSage Weil CEPH_CAP_FILE_WR; 1689355da1ebSSage Weil } 1690355da1ebSSage Weil } 1691355da1ebSSage Weil 1692355da1ebSSage Weil /* these do nothing */ 1693355da1ebSSage Weil if (ia_valid & ATTR_CTIME) { 1694355da1ebSSage Weil bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1695355da1ebSSage Weil ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1696355da1ebSSage Weil dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1697355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1698355da1ebSSage Weil attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1699355da1ebSSage Weil only ? "ctime only" : "ignored"); 1700355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1701355da1ebSSage Weil if (only) { 1702355da1ebSSage Weil /* 1703355da1ebSSage Weil * if kernel wants to dirty ctime but nothing else, 1704355da1ebSSage Weil * we need to choose a cap to dirty under, or do 1705355da1ebSSage Weil * a almost-no-op setattr 1706355da1ebSSage Weil */ 1707355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) 1708355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1709355da1ebSSage Weil else if (issued & CEPH_CAP_FILE_EXCL) 1710355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1711355da1ebSSage Weil else if (issued & CEPH_CAP_XATTR_EXCL) 1712355da1ebSSage Weil dirtied |= CEPH_CAP_XATTR_EXCL; 1713355da1ebSSage Weil else 1714355da1ebSSage Weil mask |= CEPH_SETATTR_CTIME; 1715355da1ebSSage Weil } 1716355da1ebSSage Weil } 1717355da1ebSSage Weil if (ia_valid & ATTR_FILE) 1718355da1ebSSage Weil dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1719355da1ebSSage Weil 1720355da1ebSSage Weil if (dirtied) { 1721355da1ebSSage Weil __ceph_mark_dirty_caps(ci, dirtied); 1722355da1ebSSage Weil inode->i_ctime = CURRENT_TIME; 1723355da1ebSSage Weil } 1724355da1ebSSage Weil 1725355da1ebSSage Weil release &= issued; 1726355da1ebSSage Weil spin_unlock(&inode->i_lock); 1727355da1ebSSage Weil 1728355da1ebSSage Weil if (mask) { 1729355da1ebSSage Weil req->r_inode = igrab(inode); 1730355da1ebSSage Weil req->r_inode_drop = release; 1731355da1ebSSage Weil req->r_args.setattr.mask = cpu_to_le32(mask); 1732355da1ebSSage Weil req->r_num_caps = 1; 1733355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1734355da1ebSSage Weil } 1735355da1ebSSage Weil dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1736355da1ebSSage Weil ceph_cap_string(dirtied), mask); 1737355da1ebSSage Weil 1738355da1ebSSage Weil ceph_mdsc_put_request(req); 1739355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1740355da1ebSSage Weil return err; 1741355da1ebSSage Weil out: 1742355da1ebSSage Weil spin_unlock(&inode->i_lock); 1743355da1ebSSage Weil ceph_mdsc_put_request(req); 1744355da1ebSSage Weil return err; 1745355da1ebSSage Weil } 1746355da1ebSSage Weil 1747355da1ebSSage Weil /* 1748355da1ebSSage Weil * Verify that we have a lease on the given mask. If not, 1749355da1ebSSage Weil * do a getattr against an mds. 1750355da1ebSSage Weil */ 1751355da1ebSSage Weil int ceph_do_getattr(struct inode *inode, int mask) 1752355da1ebSSage Weil { 17533d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 17543d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1755355da1ebSSage Weil struct ceph_mds_request *req; 1756355da1ebSSage Weil int err; 1757355da1ebSSage Weil 1758355da1ebSSage Weil if (ceph_snap(inode) == CEPH_SNAPDIR) { 1759355da1ebSSage Weil dout("do_getattr inode %p SNAPDIR\n", inode); 1760355da1ebSSage Weil return 0; 1761355da1ebSSage Weil } 1762355da1ebSSage Weil 1763b7495fc2SSage Weil dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1764355da1ebSSage Weil if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1765355da1ebSSage Weil return 0; 1766355da1ebSSage Weil 1767355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1768355da1ebSSage Weil if (IS_ERR(req)) 1769355da1ebSSage Weil return PTR_ERR(req); 1770355da1ebSSage Weil req->r_inode = igrab(inode); 1771355da1ebSSage Weil req->r_num_caps = 1; 1772355da1ebSSage Weil req->r_args.getattr.mask = cpu_to_le32(mask); 1773355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1774355da1ebSSage Weil ceph_mdsc_put_request(req); 1775355da1ebSSage Weil dout("do_getattr result=%d\n", err); 1776355da1ebSSage Weil return err; 1777355da1ebSSage Weil } 1778355da1ebSSage Weil 1779355da1ebSSage Weil 1780355da1ebSSage Weil /* 1781355da1ebSSage Weil * Check inode permissions. We verify we have a valid value for 1782355da1ebSSage Weil * the AUTH cap, then call the generic handler. 1783355da1ebSSage Weil */ 1784b74c79e9SNick Piggin int ceph_permission(struct inode *inode, int mask, unsigned int flags) 1785355da1ebSSage Weil { 1786b74c79e9SNick Piggin int err; 1787b74c79e9SNick Piggin 1788b74c79e9SNick Piggin if (flags & IPERM_FLAG_RCU) 1789b74c79e9SNick Piggin return -ECHILD; 1790b74c79e9SNick Piggin 1791b74c79e9SNick Piggin err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1792355da1ebSSage Weil 1793355da1ebSSage Weil if (!err) 1794b74c79e9SNick Piggin err = generic_permission(inode, mask, flags, NULL); 1795355da1ebSSage Weil return err; 1796355da1ebSSage Weil } 1797355da1ebSSage Weil 1798355da1ebSSage Weil /* 1799355da1ebSSage Weil * Get all attributes. Hopefully somedata we'll have a statlite() 1800355da1ebSSage Weil * and can limit the fields we require to be accurate. 1801355da1ebSSage Weil */ 1802355da1ebSSage Weil int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1803355da1ebSSage Weil struct kstat *stat) 1804355da1ebSSage Weil { 1805355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1806232d4b01SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1807355da1ebSSage Weil int err; 1808355da1ebSSage Weil 1809355da1ebSSage Weil err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1810355da1ebSSage Weil if (!err) { 1811355da1ebSSage Weil generic_fillattr(inode, stat); 1812355da1ebSSage Weil stat->ino = inode->i_ino; 1813355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1814355da1ebSSage Weil stat->dev = ceph_snap(inode); 1815355da1ebSSage Weil else 1816355da1ebSSage Weil stat->dev = 0; 1817232d4b01SSage Weil if (S_ISDIR(inode->i_mode)) { 18181c1266bbSYehuda Sadeh if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 18191c1266bbSYehuda Sadeh RBYTES)) 1820232d4b01SSage Weil stat->size = ci->i_rbytes; 18211c1266bbSYehuda Sadeh else 18221c1266bbSYehuda Sadeh stat->size = ci->i_files + ci->i_subdirs; 1823232d4b01SSage Weil stat->blocks = 0; 1824355da1ebSSage Weil stat->blksize = 65536; 1825355da1ebSSage Weil } 1826232d4b01SSage Weil } 1827355da1ebSSage Weil return err; 1828355da1ebSSage Weil } 1829