1355da1ebSSage Weil #include "ceph_debug.h" 2355da1ebSSage Weil 3355da1ebSSage Weil #include <linux/module.h> 4355da1ebSSage Weil #include <linux/fs.h> 5355da1ebSSage Weil #include <linux/smp_lock.h> 6355da1ebSSage Weil #include <linux/slab.h> 7355da1ebSSage Weil #include <linux/string.h> 8355da1ebSSage Weil #include <linux/uaccess.h> 9355da1ebSSage Weil #include <linux/kernel.h> 10355da1ebSSage Weil #include <linux/namei.h> 11355da1ebSSage Weil #include <linux/writeback.h> 12355da1ebSSage Weil #include <linux/vmalloc.h> 13355da1ebSSage Weil 14355da1ebSSage Weil #include "super.h" 15355da1ebSSage Weil #include "decode.h" 16355da1ebSSage Weil 17355da1ebSSage Weil /* 18355da1ebSSage Weil * Ceph inode operations 19355da1ebSSage Weil * 20355da1ebSSage Weil * Implement basic inode helpers (get, alloc) and inode ops (getattr, 21355da1ebSSage Weil * setattr, etc.), xattr helpers, and helpers for assimilating 22355da1ebSSage Weil * metadata returned by the MDS into our cache. 23355da1ebSSage Weil * 24355da1ebSSage Weil * Also define helpers for doing asynchronous writeback, invalidation, 25355da1ebSSage Weil * and truncation for the benefit of those who can't afford to block 26355da1ebSSage Weil * (typically because they are in the message handler path). 27355da1ebSSage Weil */ 28355da1ebSSage Weil 29355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops; 30355da1ebSSage Weil 31355da1ebSSage Weil static void ceph_inode_invalidate_pages(struct work_struct *work); 32355da1ebSSage Weil 33355da1ebSSage Weil /* 34355da1ebSSage Weil * find or create an inode, given the ceph ino number 35355da1ebSSage Weil */ 36355da1ebSSage Weil struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 37355da1ebSSage Weil { 38355da1ebSSage Weil struct inode *inode; 39355da1ebSSage Weil ino_t t = ceph_vino_to_ino(vino); 40355da1ebSSage Weil 41355da1ebSSage Weil inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 42355da1ebSSage Weil if (inode == NULL) 43355da1ebSSage Weil return ERR_PTR(-ENOMEM); 44355da1ebSSage Weil if (inode->i_state & I_NEW) { 45355da1ebSSage Weil dout("get_inode created new inode %p %llx.%llx ino %llx\n", 46355da1ebSSage Weil inode, ceph_vinop(inode), (u64)inode->i_ino); 47355da1ebSSage Weil unlock_new_inode(inode); 48355da1ebSSage Weil } 49355da1ebSSage Weil 50355da1ebSSage Weil dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 51355da1ebSSage Weil vino.snap, inode); 52355da1ebSSage Weil return inode; 53355da1ebSSage Weil } 54355da1ebSSage Weil 55355da1ebSSage Weil /* 56355da1ebSSage Weil * get/constuct snapdir inode for a given directory 57355da1ebSSage Weil */ 58355da1ebSSage Weil struct inode *ceph_get_snapdir(struct inode *parent) 59355da1ebSSage Weil { 60355da1ebSSage Weil struct ceph_vino vino = { 61355da1ebSSage Weil .ino = ceph_ino(parent), 62355da1ebSSage Weil .snap = CEPH_SNAPDIR, 63355da1ebSSage Weil }; 64355da1ebSSage Weil struct inode *inode = ceph_get_inode(parent->i_sb, vino); 65b377ff13SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 66355da1ebSSage Weil 67355da1ebSSage Weil BUG_ON(!S_ISDIR(parent->i_mode)); 68355da1ebSSage Weil if (IS_ERR(inode)) 69355da1ebSSage Weil return ERR_PTR(PTR_ERR(inode)); 70355da1ebSSage Weil inode->i_mode = parent->i_mode; 71355da1ebSSage Weil inode->i_uid = parent->i_uid; 72355da1ebSSage Weil inode->i_gid = parent->i_gid; 73355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 74355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 75b377ff13SSage Weil ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 76b377ff13SSage Weil ci->i_rbytes = 0; 77355da1ebSSage Weil return inode; 78355da1ebSSage Weil } 79355da1ebSSage Weil 80355da1ebSSage Weil const struct inode_operations ceph_file_iops = { 81355da1ebSSage Weil .permission = ceph_permission, 82355da1ebSSage Weil .setattr = ceph_setattr, 83355da1ebSSage Weil .getattr = ceph_getattr, 84355da1ebSSage Weil .setxattr = ceph_setxattr, 85355da1ebSSage Weil .getxattr = ceph_getxattr, 86355da1ebSSage Weil .listxattr = ceph_listxattr, 87355da1ebSSage Weil .removexattr = ceph_removexattr, 88355da1ebSSage Weil }; 89355da1ebSSage Weil 90355da1ebSSage Weil 91355da1ebSSage Weil /* 92355da1ebSSage Weil * We use a 'frag tree' to keep track of the MDS's directory fragments 93355da1ebSSage Weil * for a given inode (usually there is just a single fragment). We 94355da1ebSSage Weil * need to know when a child frag is delegated to a new MDS, or when 95355da1ebSSage Weil * it is flagged as replicated, so we can direct our requests 96355da1ebSSage Weil * accordingly. 97355da1ebSSage Weil */ 98355da1ebSSage Weil 99355da1ebSSage Weil /* 100355da1ebSSage Weil * find/create a frag in the tree 101355da1ebSSage Weil */ 102355da1ebSSage Weil static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 103355da1ebSSage Weil u32 f) 104355da1ebSSage Weil { 105355da1ebSSage Weil struct rb_node **p; 106355da1ebSSage Weil struct rb_node *parent = NULL; 107355da1ebSSage Weil struct ceph_inode_frag *frag; 108355da1ebSSage Weil int c; 109355da1ebSSage Weil 110355da1ebSSage Weil p = &ci->i_fragtree.rb_node; 111355da1ebSSage Weil while (*p) { 112355da1ebSSage Weil parent = *p; 113355da1ebSSage Weil frag = rb_entry(parent, struct ceph_inode_frag, node); 114355da1ebSSage Weil c = ceph_frag_compare(f, frag->frag); 115355da1ebSSage Weil if (c < 0) 116355da1ebSSage Weil p = &(*p)->rb_left; 117355da1ebSSage Weil else if (c > 0) 118355da1ebSSage Weil p = &(*p)->rb_right; 119355da1ebSSage Weil else 120355da1ebSSage Weil return frag; 121355da1ebSSage Weil } 122355da1ebSSage Weil 123355da1ebSSage Weil frag = kmalloc(sizeof(*frag), GFP_NOFS); 124355da1ebSSage Weil if (!frag) { 125355da1ebSSage Weil pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 126355da1ebSSage Weil "frag %x\n", &ci->vfs_inode, 127355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 128355da1ebSSage Weil return ERR_PTR(-ENOMEM); 129355da1ebSSage Weil } 130355da1ebSSage Weil frag->frag = f; 131355da1ebSSage Weil frag->split_by = 0; 132355da1ebSSage Weil frag->mds = -1; 133355da1ebSSage Weil frag->ndist = 0; 134355da1ebSSage Weil 135355da1ebSSage Weil rb_link_node(&frag->node, parent, p); 136355da1ebSSage Weil rb_insert_color(&frag->node, &ci->i_fragtree); 137355da1ebSSage Weil 138355da1ebSSage Weil dout("get_or_create_frag added %llx.%llx frag %x\n", 139355da1ebSSage Weil ceph_vinop(&ci->vfs_inode), f); 140355da1ebSSage Weil return frag; 141355da1ebSSage Weil } 142355da1ebSSage Weil 143355da1ebSSage Weil /* 144355da1ebSSage Weil * find a specific frag @f 145355da1ebSSage Weil */ 146355da1ebSSage Weil struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 147355da1ebSSage Weil { 148355da1ebSSage Weil struct rb_node *n = ci->i_fragtree.rb_node; 149355da1ebSSage Weil 150355da1ebSSage Weil while (n) { 151355da1ebSSage Weil struct ceph_inode_frag *frag = 152355da1ebSSage Weil rb_entry(n, struct ceph_inode_frag, node); 153355da1ebSSage Weil int c = ceph_frag_compare(f, frag->frag); 154355da1ebSSage Weil if (c < 0) 155355da1ebSSage Weil n = n->rb_left; 156355da1ebSSage Weil else if (c > 0) 157355da1ebSSage Weil n = n->rb_right; 158355da1ebSSage Weil else 159355da1ebSSage Weil return frag; 160355da1ebSSage Weil } 161355da1ebSSage Weil return NULL; 162355da1ebSSage Weil } 163355da1ebSSage Weil 164355da1ebSSage Weil /* 165355da1ebSSage Weil * Choose frag containing the given value @v. If @pfrag is 166355da1ebSSage Weil * specified, copy the frag delegation info to the caller if 167355da1ebSSage Weil * it is present. 168355da1ebSSage Weil */ 169355da1ebSSage Weil u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 170355da1ebSSage Weil struct ceph_inode_frag *pfrag, 171355da1ebSSage Weil int *found) 172355da1ebSSage Weil { 173355da1ebSSage Weil u32 t = ceph_frag_make(0, 0); 174355da1ebSSage Weil struct ceph_inode_frag *frag; 175355da1ebSSage Weil unsigned nway, i; 176355da1ebSSage Weil u32 n; 177355da1ebSSage Weil 178355da1ebSSage Weil if (found) 179355da1ebSSage Weil *found = 0; 180355da1ebSSage Weil 181355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 182355da1ebSSage Weil while (1) { 183355da1ebSSage Weil WARN_ON(!ceph_frag_contains_value(t, v)); 184355da1ebSSage Weil frag = __ceph_find_frag(ci, t); 185355da1ebSSage Weil if (!frag) 186355da1ebSSage Weil break; /* t is a leaf */ 187355da1ebSSage Weil if (frag->split_by == 0) { 188355da1ebSSage Weil if (pfrag) 189355da1ebSSage Weil memcpy(pfrag, frag, sizeof(*pfrag)); 190355da1ebSSage Weil if (found) 191355da1ebSSage Weil *found = 1; 192355da1ebSSage Weil break; 193355da1ebSSage Weil } 194355da1ebSSage Weil 195355da1ebSSage Weil /* choose child */ 196355da1ebSSage Weil nway = 1 << frag->split_by; 197355da1ebSSage Weil dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 198355da1ebSSage Weil frag->split_by, nway); 199355da1ebSSage Weil for (i = 0; i < nway; i++) { 200355da1ebSSage Weil n = ceph_frag_make_child(t, frag->split_by, i); 201355da1ebSSage Weil if (ceph_frag_contains_value(n, v)) { 202355da1ebSSage Weil t = n; 203355da1ebSSage Weil break; 204355da1ebSSage Weil } 205355da1ebSSage Weil } 206355da1ebSSage Weil BUG_ON(i == nway); 207355da1ebSSage Weil } 208355da1ebSSage Weil dout("choose_frag(%x) = %x\n", v, t); 209355da1ebSSage Weil 210355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 211355da1ebSSage Weil return t; 212355da1ebSSage Weil } 213355da1ebSSage Weil 214355da1ebSSage Weil /* 215355da1ebSSage Weil * Process dirfrag (delegation) info from the mds. Include leaf 216355da1ebSSage Weil * fragment in tree ONLY if ndist > 0. Otherwise, only 217355da1ebSSage Weil * branches/splits are included in i_fragtree) 218355da1ebSSage Weil */ 219355da1ebSSage Weil static int ceph_fill_dirfrag(struct inode *inode, 220355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo) 221355da1ebSSage Weil { 222355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 223355da1ebSSage Weil struct ceph_inode_frag *frag; 224355da1ebSSage Weil u32 id = le32_to_cpu(dirinfo->frag); 225355da1ebSSage Weil int mds = le32_to_cpu(dirinfo->auth); 226355da1ebSSage Weil int ndist = le32_to_cpu(dirinfo->ndist); 227355da1ebSSage Weil int i; 228355da1ebSSage Weil int err = 0; 229355da1ebSSage Weil 230355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 231355da1ebSSage Weil if (ndist == 0) { 232355da1ebSSage Weil /* no delegation info needed. */ 233355da1ebSSage Weil frag = __ceph_find_frag(ci, id); 234355da1ebSSage Weil if (!frag) 235355da1ebSSage Weil goto out; 236355da1ebSSage Weil if (frag->split_by == 0) { 237355da1ebSSage Weil /* tree leaf, remove */ 238355da1ebSSage Weil dout("fill_dirfrag removed %llx.%llx frag %x" 239355da1ebSSage Weil " (no ref)\n", ceph_vinop(inode), id); 240355da1ebSSage Weil rb_erase(&frag->node, &ci->i_fragtree); 241355da1ebSSage Weil kfree(frag); 242355da1ebSSage Weil } else { 243355da1ebSSage Weil /* tree branch, keep and clear */ 244355da1ebSSage Weil dout("fill_dirfrag cleared %llx.%llx frag %x" 245355da1ebSSage Weil " referral\n", ceph_vinop(inode), id); 246355da1ebSSage Weil frag->mds = -1; 247355da1ebSSage Weil frag->ndist = 0; 248355da1ebSSage Weil } 249355da1ebSSage Weil goto out; 250355da1ebSSage Weil } 251355da1ebSSage Weil 252355da1ebSSage Weil 253355da1ebSSage Weil /* find/add this frag to store mds delegation info */ 254355da1ebSSage Weil frag = __get_or_create_frag(ci, id); 255355da1ebSSage Weil if (IS_ERR(frag)) { 256355da1ebSSage Weil /* this is not the end of the world; we can continue 257355da1ebSSage Weil with bad/inaccurate delegation info */ 258355da1ebSSage Weil pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 259355da1ebSSage Weil ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 260355da1ebSSage Weil err = -ENOMEM; 261355da1ebSSage Weil goto out; 262355da1ebSSage Weil } 263355da1ebSSage Weil 264355da1ebSSage Weil frag->mds = mds; 265355da1ebSSage Weil frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 266355da1ebSSage Weil for (i = 0; i < frag->ndist; i++) 267355da1ebSSage Weil frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 268355da1ebSSage Weil dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 269355da1ebSSage Weil ceph_vinop(inode), frag->frag, frag->ndist); 270355da1ebSSage Weil 271355da1ebSSage Weil out: 272355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 273355da1ebSSage Weil return err; 274355da1ebSSage Weil } 275355da1ebSSage Weil 276355da1ebSSage Weil 277355da1ebSSage Weil /* 278355da1ebSSage Weil * initialize a newly allocated inode. 279355da1ebSSage Weil */ 280355da1ebSSage Weil struct inode *ceph_alloc_inode(struct super_block *sb) 281355da1ebSSage Weil { 282355da1ebSSage Weil struct ceph_inode_info *ci; 283355da1ebSSage Weil int i; 284355da1ebSSage Weil 285355da1ebSSage Weil ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 286355da1ebSSage Weil if (!ci) 287355da1ebSSage Weil return NULL; 288355da1ebSSage Weil 289355da1ebSSage Weil dout("alloc_inode %p\n", &ci->vfs_inode); 290355da1ebSSage Weil 291355da1ebSSage Weil ci->i_version = 0; 292355da1ebSSage Weil ci->i_time_warp_seq = 0; 293355da1ebSSage Weil ci->i_ceph_flags = 0; 294355da1ebSSage Weil ci->i_release_count = 0; 295355da1ebSSage Weil ci->i_symlink = NULL; 296355da1ebSSage Weil 297355da1ebSSage Weil ci->i_fragtree = RB_ROOT; 298355da1ebSSage Weil mutex_init(&ci->i_fragtree_mutex); 299355da1ebSSage Weil 300355da1ebSSage Weil ci->i_xattrs.blob = NULL; 301355da1ebSSage Weil ci->i_xattrs.prealloc_blob = NULL; 302355da1ebSSage Weil ci->i_xattrs.dirty = false; 303355da1ebSSage Weil ci->i_xattrs.index = RB_ROOT; 304355da1ebSSage Weil ci->i_xattrs.count = 0; 305355da1ebSSage Weil ci->i_xattrs.names_size = 0; 306355da1ebSSage Weil ci->i_xattrs.vals_size = 0; 307355da1ebSSage Weil ci->i_xattrs.version = 0; 308355da1ebSSage Weil ci->i_xattrs.index_version = 0; 309355da1ebSSage Weil 310355da1ebSSage Weil ci->i_caps = RB_ROOT; 311355da1ebSSage Weil ci->i_auth_cap = NULL; 312355da1ebSSage Weil ci->i_dirty_caps = 0; 313355da1ebSSage Weil ci->i_flushing_caps = 0; 314355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_dirty_item); 315355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_flushing_item); 316355da1ebSSage Weil ci->i_cap_flush_seq = 0; 317355da1ebSSage Weil ci->i_cap_flush_last_tid = 0; 318355da1ebSSage Weil memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 319355da1ebSSage Weil init_waitqueue_head(&ci->i_cap_wq); 320355da1ebSSage Weil ci->i_hold_caps_min = 0; 321355da1ebSSage Weil ci->i_hold_caps_max = 0; 322355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_delay_list); 323355da1ebSSage Weil ci->i_cap_exporting_mds = 0; 324355da1ebSSage Weil ci->i_cap_exporting_mseq = 0; 325355da1ebSSage Weil ci->i_cap_exporting_issued = 0; 326355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_snaps); 327355da1ebSSage Weil ci->i_head_snapc = NULL; 328355da1ebSSage Weil ci->i_snap_caps = 0; 329355da1ebSSage Weil 330355da1ebSSage Weil for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 331355da1ebSSage Weil ci->i_nr_by_mode[i] = 0; 332355da1ebSSage Weil 333355da1ebSSage Weil ci->i_truncate_seq = 0; 334355da1ebSSage Weil ci->i_truncate_size = 0; 335355da1ebSSage Weil ci->i_truncate_pending = 0; 336355da1ebSSage Weil 337355da1ebSSage Weil ci->i_max_size = 0; 338355da1ebSSage Weil ci->i_reported_size = 0; 339355da1ebSSage Weil ci->i_wanted_max_size = 0; 340355da1ebSSage Weil ci->i_requested_max_size = 0; 341355da1ebSSage Weil 342355da1ebSSage Weil ci->i_pin_ref = 0; 343355da1ebSSage Weil ci->i_rd_ref = 0; 344355da1ebSSage Weil ci->i_rdcache_ref = 0; 345355da1ebSSage Weil ci->i_wr_ref = 0; 346355da1ebSSage Weil ci->i_wrbuffer_ref = 0; 347355da1ebSSage Weil ci->i_wrbuffer_ref_head = 0; 348355da1ebSSage Weil ci->i_shared_gen = 0; 349355da1ebSSage Weil ci->i_rdcache_gen = 0; 350355da1ebSSage Weil ci->i_rdcache_revoking = 0; 351355da1ebSSage Weil 352355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_writes); 353355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_dirops); 354355da1ebSSage Weil spin_lock_init(&ci->i_unsafe_lock); 355355da1ebSSage Weil 356355da1ebSSage Weil ci->i_snap_realm = NULL; 357355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_realm_item); 358355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_flush_item); 359355da1ebSSage Weil 360355da1ebSSage Weil INIT_WORK(&ci->i_wb_work, ceph_inode_writeback); 361355da1ebSSage Weil INIT_WORK(&ci->i_pg_inv_work, ceph_inode_invalidate_pages); 362355da1ebSSage Weil 363355da1ebSSage Weil INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 364355da1ebSSage Weil 365355da1ebSSage Weil return &ci->vfs_inode; 366355da1ebSSage Weil } 367355da1ebSSage Weil 368355da1ebSSage Weil void ceph_destroy_inode(struct inode *inode) 369355da1ebSSage Weil { 370355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 371355da1ebSSage Weil struct ceph_inode_frag *frag; 372355da1ebSSage Weil struct rb_node *n; 373355da1ebSSage Weil 374355da1ebSSage Weil dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 375355da1ebSSage Weil 376355da1ebSSage Weil ceph_queue_caps_release(inode); 377355da1ebSSage Weil 378355da1ebSSage Weil kfree(ci->i_symlink); 379355da1ebSSage Weil while ((n = rb_first(&ci->i_fragtree)) != NULL) { 380355da1ebSSage Weil frag = rb_entry(n, struct ceph_inode_frag, node); 381355da1ebSSage Weil rb_erase(n, &ci->i_fragtree); 382355da1ebSSage Weil kfree(frag); 383355da1ebSSage Weil } 384355da1ebSSage Weil 385355da1ebSSage Weil __ceph_destroy_xattrs(ci); 386b6c1d5b8SSage Weil if (ci->i_xattrs.blob) 387355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 388b6c1d5b8SSage Weil if (ci->i_xattrs.prealloc_blob) 389355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.prealloc_blob); 390355da1ebSSage Weil 391355da1ebSSage Weil kmem_cache_free(ceph_inode_cachep, ci); 392355da1ebSSage Weil } 393355da1ebSSage Weil 394355da1ebSSage Weil 395355da1ebSSage Weil /* 396355da1ebSSage Weil * Helpers to fill in size, ctime, mtime, and atime. We have to be 397355da1ebSSage Weil * careful because either the client or MDS may have more up to date 398355da1ebSSage Weil * info, depending on which capabilities are held, and whether 399355da1ebSSage Weil * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 400355da1ebSSage Weil * and size are monotonically increasing, except when utimes() or 401355da1ebSSage Weil * truncate() increments the corresponding _seq values.) 402355da1ebSSage Weil */ 403355da1ebSSage Weil int ceph_fill_file_size(struct inode *inode, int issued, 404355da1ebSSage Weil u32 truncate_seq, u64 truncate_size, u64 size) 405355da1ebSSage Weil { 406355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 407355da1ebSSage Weil int queue_trunc = 0; 408355da1ebSSage Weil 409355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 410355da1ebSSage Weil (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 411355da1ebSSage Weil dout("size %lld -> %llu\n", inode->i_size, size); 412355da1ebSSage Weil inode->i_size = size; 413355da1ebSSage Weil inode->i_blocks = (size + (1<<9) - 1) >> 9; 414355da1ebSSage Weil ci->i_reported_size = size; 415355da1ebSSage Weil if (truncate_seq != ci->i_truncate_seq) { 416355da1ebSSage Weil dout("truncate_seq %u -> %u\n", 417355da1ebSSage Weil ci->i_truncate_seq, truncate_seq); 418355da1ebSSage Weil ci->i_truncate_seq = truncate_seq; 419355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 420355da1ebSSage Weil CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 421355da1ebSSage Weil CEPH_CAP_FILE_EXCL)) { 422355da1ebSSage Weil ci->i_truncate_pending++; 423355da1ebSSage Weil queue_trunc = 1; 424355da1ebSSage Weil } 425355da1ebSSage Weil } 426355da1ebSSage Weil } 427355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 428355da1ebSSage Weil ci->i_truncate_size != truncate_size) { 429355da1ebSSage Weil dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 430355da1ebSSage Weil truncate_size); 431355da1ebSSage Weil ci->i_truncate_size = truncate_size; 432355da1ebSSage Weil } 433355da1ebSSage Weil return queue_trunc; 434355da1ebSSage Weil } 435355da1ebSSage Weil 436355da1ebSSage Weil void ceph_fill_file_time(struct inode *inode, int issued, 437355da1ebSSage Weil u64 time_warp_seq, struct timespec *ctime, 438355da1ebSSage Weil struct timespec *mtime, struct timespec *atime) 439355da1ebSSage Weil { 440355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 441355da1ebSSage Weil int warn = 0; 442355da1ebSSage Weil 443355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_EXCL| 444355da1ebSSage Weil CEPH_CAP_FILE_WR| 445355da1ebSSage Weil CEPH_CAP_FILE_BUFFER)) { 446355da1ebSSage Weil if (timespec_compare(ctime, &inode->i_ctime) > 0) { 447355da1ebSSage Weil dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 448355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 449355da1ebSSage Weil ctime->tv_sec, ctime->tv_nsec); 450355da1ebSSage Weil inode->i_ctime = *ctime; 451355da1ebSSage Weil } 452355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 453355da1ebSSage Weil /* the MDS did a utimes() */ 454355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld " 455355da1ebSSage Weil "tw %d -> %d\n", 456355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 457355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec, 458355da1ebSSage Weil ci->i_time_warp_seq, (int)time_warp_seq); 459355da1ebSSage Weil 460355da1ebSSage Weil inode->i_mtime = *mtime; 461355da1ebSSage Weil inode->i_atime = *atime; 462355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 463355da1ebSSage Weil } else if (time_warp_seq == ci->i_time_warp_seq) { 464355da1ebSSage Weil /* nobody did utimes(); take the max */ 465355da1ebSSage Weil if (timespec_compare(mtime, &inode->i_mtime) > 0) { 466355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 467355da1ebSSage Weil inode->i_mtime.tv_sec, 468355da1ebSSage Weil inode->i_mtime.tv_nsec, 469355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec); 470355da1ebSSage Weil inode->i_mtime = *mtime; 471355da1ebSSage Weil } 472355da1ebSSage Weil if (timespec_compare(atime, &inode->i_atime) > 0) { 473355da1ebSSage Weil dout("atime %ld.%09ld -> %ld.%09ld inc\n", 474355da1ebSSage Weil inode->i_atime.tv_sec, 475355da1ebSSage Weil inode->i_atime.tv_nsec, 476355da1ebSSage Weil atime->tv_sec, atime->tv_nsec); 477355da1ebSSage Weil inode->i_atime = *atime; 478355da1ebSSage Weil } 479355da1ebSSage Weil } else if (issued & CEPH_CAP_FILE_EXCL) { 480355da1ebSSage Weil /* we did a utimes(); ignore mds values */ 481355da1ebSSage Weil } else { 482355da1ebSSage Weil warn = 1; 483355da1ebSSage Weil } 484355da1ebSSage Weil } else { 485355da1ebSSage Weil /* we have no write caps; whatever the MDS says is true */ 486355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 487355da1ebSSage Weil inode->i_ctime = *ctime; 488355da1ebSSage Weil inode->i_mtime = *mtime; 489355da1ebSSage Weil inode->i_atime = *atime; 490355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 491355da1ebSSage Weil } else { 492355da1ebSSage Weil warn = 1; 493355da1ebSSage Weil } 494355da1ebSSage Weil } 495355da1ebSSage Weil if (warn) /* time_warp_seq shouldn't go backwards */ 496355da1ebSSage Weil dout("%p mds time_warp_seq %llu < %u\n", 497355da1ebSSage Weil inode, time_warp_seq, ci->i_time_warp_seq); 498355da1ebSSage Weil } 499355da1ebSSage Weil 500355da1ebSSage Weil /* 501355da1ebSSage Weil * Populate an inode based on info from mds. May be called on new or 502355da1ebSSage Weil * existing inodes. 503355da1ebSSage Weil */ 504355da1ebSSage Weil static int fill_inode(struct inode *inode, 505355da1ebSSage Weil struct ceph_mds_reply_info_in *iinfo, 506355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo, 507355da1ebSSage Weil struct ceph_mds_session *session, 508355da1ebSSage Weil unsigned long ttl_from, int cap_fmode, 509355da1ebSSage Weil struct ceph_cap_reservation *caps_reservation) 510355da1ebSSage Weil { 511355da1ebSSage Weil struct ceph_mds_reply_inode *info = iinfo->in; 512355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 513355da1ebSSage Weil int i; 514355da1ebSSage Weil int issued, implemented; 515355da1ebSSage Weil struct timespec mtime, atime, ctime; 516355da1ebSSage Weil u32 nsplits; 517355da1ebSSage Weil struct ceph_buffer *xattr_blob = NULL; 518355da1ebSSage Weil int err = 0; 519355da1ebSSage Weil int queue_trunc = 0; 520355da1ebSSage Weil 521355da1ebSSage Weil dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 522355da1ebSSage Weil inode, ceph_vinop(inode), le64_to_cpu(info->version), 523355da1ebSSage Weil ci->i_version); 524355da1ebSSage Weil 525355da1ebSSage Weil /* 526355da1ebSSage Weil * prealloc xattr data, if it looks like we'll need it. only 527355da1ebSSage Weil * if len > 4 (meaning there are actually xattrs; the first 4 528355da1ebSSage Weil * bytes are the xattr count). 529355da1ebSSage Weil */ 530355da1ebSSage Weil if (iinfo->xattr_len > 4) { 531b6c1d5b8SSage Weil xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 532355da1ebSSage Weil if (!xattr_blob) 533355da1ebSSage Weil pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 534355da1ebSSage Weil iinfo->xattr_len); 535355da1ebSSage Weil } 536355da1ebSSage Weil 537355da1ebSSage Weil spin_lock(&inode->i_lock); 538355da1ebSSage Weil 539355da1ebSSage Weil /* 540355da1ebSSage Weil * provided version will be odd if inode value is projected, 541355da1ebSSage Weil * even if stable. skip the update if we have a newer info 542355da1ebSSage Weil * (e.g., due to inode info racing form multiple MDSs), or if 543355da1ebSSage Weil * we are getting projected (unstable) inode info. 544355da1ebSSage Weil */ 545355da1ebSSage Weil if (le64_to_cpu(info->version) > 0 && 546355da1ebSSage Weil (ci->i_version & ~1) > le64_to_cpu(info->version)) 547355da1ebSSage Weil goto no_change; 548355da1ebSSage Weil 549355da1ebSSage Weil issued = __ceph_caps_issued(ci, &implemented); 550355da1ebSSage Weil issued |= implemented | __ceph_caps_dirty(ci); 551355da1ebSSage Weil 552355da1ebSSage Weil /* update inode */ 553355da1ebSSage Weil ci->i_version = le64_to_cpu(info->version); 554355da1ebSSage Weil inode->i_version++; 555355da1ebSSage Weil inode->i_rdev = le32_to_cpu(info->rdev); 556355da1ebSSage Weil 557355da1ebSSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 558355da1ebSSage Weil inode->i_mode = le32_to_cpu(info->mode); 559355da1ebSSage Weil inode->i_uid = le32_to_cpu(info->uid); 560355da1ebSSage Weil inode->i_gid = le32_to_cpu(info->gid); 561355da1ebSSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 562355da1ebSSage Weil inode->i_uid, inode->i_gid); 563355da1ebSSage Weil } 564355da1ebSSage Weil 565355da1ebSSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 566355da1ebSSage Weil inode->i_nlink = le32_to_cpu(info->nlink); 567355da1ebSSage Weil 568355da1ebSSage Weil /* be careful with mtime, atime, size */ 569355da1ebSSage Weil ceph_decode_timespec(&atime, &info->atime); 570355da1ebSSage Weil ceph_decode_timespec(&mtime, &info->mtime); 571355da1ebSSage Weil ceph_decode_timespec(&ctime, &info->ctime); 572355da1ebSSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 573355da1ebSSage Weil le32_to_cpu(info->truncate_seq), 574355da1ebSSage Weil le64_to_cpu(info->truncate_size), 575355da1ebSSage Weil le64_to_cpu(info->size)); 576355da1ebSSage Weil ceph_fill_file_time(inode, issued, 577355da1ebSSage Weil le32_to_cpu(info->time_warp_seq), 578355da1ebSSage Weil &ctime, &mtime, &atime); 579355da1ebSSage Weil 580355da1ebSSage Weil ci->i_max_size = le64_to_cpu(info->max_size); 581355da1ebSSage Weil ci->i_layout = info->layout; 582355da1ebSSage Weil inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 583355da1ebSSage Weil 584355da1ebSSage Weil /* xattrs */ 585355da1ebSSage Weil /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 586355da1ebSSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 587355da1ebSSage Weil le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 588355da1ebSSage Weil if (ci->i_xattrs.blob) 589355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 590355da1ebSSage Weil ci->i_xattrs.blob = xattr_blob; 591355da1ebSSage Weil if (xattr_blob) 592355da1ebSSage Weil memcpy(ci->i_xattrs.blob->vec.iov_base, 593355da1ebSSage Weil iinfo->xattr_data, iinfo->xattr_len); 594355da1ebSSage Weil ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 595355da1ebSSage Weil } 596355da1ebSSage Weil 597355da1ebSSage Weil inode->i_mapping->a_ops = &ceph_aops; 598355da1ebSSage Weil inode->i_mapping->backing_dev_info = 599355da1ebSSage Weil &ceph_client(inode->i_sb)->backing_dev_info; 600355da1ebSSage Weil 601355da1ebSSage Weil switch (inode->i_mode & S_IFMT) { 602355da1ebSSage Weil case S_IFIFO: 603355da1ebSSage Weil case S_IFBLK: 604355da1ebSSage Weil case S_IFCHR: 605355da1ebSSage Weil case S_IFSOCK: 606355da1ebSSage Weil init_special_inode(inode, inode->i_mode, inode->i_rdev); 607355da1ebSSage Weil inode->i_op = &ceph_file_iops; 608355da1ebSSage Weil break; 609355da1ebSSage Weil case S_IFREG: 610355da1ebSSage Weil inode->i_op = &ceph_file_iops; 611355da1ebSSage Weil inode->i_fop = &ceph_file_fops; 612355da1ebSSage Weil break; 613355da1ebSSage Weil case S_IFLNK: 614355da1ebSSage Weil inode->i_op = &ceph_symlink_iops; 615355da1ebSSage Weil if (!ci->i_symlink) { 616355da1ebSSage Weil int symlen = iinfo->symlink_len; 617355da1ebSSage Weil char *sym; 618355da1ebSSage Weil 619355da1ebSSage Weil BUG_ON(symlen != inode->i_size); 620355da1ebSSage Weil spin_unlock(&inode->i_lock); 621355da1ebSSage Weil 622355da1ebSSage Weil err = -ENOMEM; 623355da1ebSSage Weil sym = kmalloc(symlen+1, GFP_NOFS); 624355da1ebSSage Weil if (!sym) 625355da1ebSSage Weil goto out; 626355da1ebSSage Weil memcpy(sym, iinfo->symlink, symlen); 627355da1ebSSage Weil sym[symlen] = 0; 628355da1ebSSage Weil 629355da1ebSSage Weil spin_lock(&inode->i_lock); 630355da1ebSSage Weil if (!ci->i_symlink) 631355da1ebSSage Weil ci->i_symlink = sym; 632355da1ebSSage Weil else 633355da1ebSSage Weil kfree(sym); /* lost a race */ 634355da1ebSSage Weil } 635355da1ebSSage Weil break; 636355da1ebSSage Weil case S_IFDIR: 637355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 638355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 639355da1ebSSage Weil 640355da1ebSSage Weil ci->i_files = le64_to_cpu(info->files); 641355da1ebSSage Weil ci->i_subdirs = le64_to_cpu(info->subdirs); 642355da1ebSSage Weil ci->i_rbytes = le64_to_cpu(info->rbytes); 643355da1ebSSage Weil ci->i_rfiles = le64_to_cpu(info->rfiles); 644355da1ebSSage Weil ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 645355da1ebSSage Weil ceph_decode_timespec(&ci->i_rctime, &info->rctime); 646355da1ebSSage Weil 647355da1ebSSage Weil /* set dir completion flag? */ 648355da1ebSSage Weil if (ci->i_files == 0 && ci->i_subdirs == 0 && 649355da1ebSSage Weil ceph_snap(inode) == CEPH_NOSNAP && 650355da1ebSSage Weil (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) { 651355da1ebSSage Weil dout(" marking %p complete (empty)\n", inode); 652355da1ebSSage Weil ci->i_ceph_flags |= CEPH_I_COMPLETE; 653355da1ebSSage Weil ci->i_max_offset = 2; 654355da1ebSSage Weil } 655355da1ebSSage Weil 656355da1ebSSage Weil /* it may be better to set st_size in getattr instead? */ 657355da1ebSSage Weil if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES)) 658355da1ebSSage Weil inode->i_size = ci->i_rbytes; 659355da1ebSSage Weil break; 660355da1ebSSage Weil default: 661355da1ebSSage Weil pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 662355da1ebSSage Weil ceph_vinop(inode), inode->i_mode); 663355da1ebSSage Weil } 664355da1ebSSage Weil 665355da1ebSSage Weil no_change: 666355da1ebSSage Weil spin_unlock(&inode->i_lock); 667355da1ebSSage Weil 668355da1ebSSage Weil /* queue truncate if we saw i_size decrease */ 669355da1ebSSage Weil if (queue_trunc) 670355da1ebSSage Weil if (queue_work(ceph_client(inode->i_sb)->trunc_wq, 671355da1ebSSage Weil &ci->i_vmtruncate_work)) 672355da1ebSSage Weil igrab(inode); 673355da1ebSSage Weil 674355da1ebSSage Weil /* populate frag tree */ 675355da1ebSSage Weil /* FIXME: move me up, if/when version reflects fragtree changes */ 676355da1ebSSage Weil nsplits = le32_to_cpu(info->fragtree.nsplits); 677355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 678355da1ebSSage Weil for (i = 0; i < nsplits; i++) { 679355da1ebSSage Weil u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 680355da1ebSSage Weil struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 681355da1ebSSage Weil 682355da1ebSSage Weil if (IS_ERR(frag)) 683355da1ebSSage Weil continue; 684355da1ebSSage Weil frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 685355da1ebSSage Weil dout(" frag %x split by %d\n", frag->frag, frag->split_by); 686355da1ebSSage Weil } 687355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 688355da1ebSSage Weil 689355da1ebSSage Weil /* were we issued a capability? */ 690355da1ebSSage Weil if (info->cap.caps) { 691355da1ebSSage Weil if (ceph_snap(inode) == CEPH_NOSNAP) { 692355da1ebSSage Weil ceph_add_cap(inode, session, 693355da1ebSSage Weil le64_to_cpu(info->cap.cap_id), 694355da1ebSSage Weil cap_fmode, 695355da1ebSSage Weil le32_to_cpu(info->cap.caps), 696355da1ebSSage Weil le32_to_cpu(info->cap.wanted), 697355da1ebSSage Weil le32_to_cpu(info->cap.seq), 698355da1ebSSage Weil le32_to_cpu(info->cap.mseq), 699355da1ebSSage Weil le64_to_cpu(info->cap.realm), 700355da1ebSSage Weil info->cap.flags, 701355da1ebSSage Weil caps_reservation); 702355da1ebSSage Weil } else { 703355da1ebSSage Weil spin_lock(&inode->i_lock); 704355da1ebSSage Weil dout(" %p got snap_caps %s\n", inode, 705355da1ebSSage Weil ceph_cap_string(le32_to_cpu(info->cap.caps))); 706355da1ebSSage Weil ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 707355da1ebSSage Weil if (cap_fmode >= 0) 708355da1ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 709355da1ebSSage Weil spin_unlock(&inode->i_lock); 710355da1ebSSage Weil } 711355da1ebSSage Weil } 712355da1ebSSage Weil 713355da1ebSSage Weil /* update delegation info? */ 714355da1ebSSage Weil if (dirinfo) 715355da1ebSSage Weil ceph_fill_dirfrag(inode, dirinfo); 716355da1ebSSage Weil 717355da1ebSSage Weil err = 0; 718355da1ebSSage Weil 719355da1ebSSage Weil out: 720b6c1d5b8SSage Weil if (xattr_blob) 721355da1ebSSage Weil ceph_buffer_put(xattr_blob); 722355da1ebSSage Weil return err; 723355da1ebSSage Weil } 724355da1ebSSage Weil 725355da1ebSSage Weil /* 726355da1ebSSage Weil * caller should hold session s_mutex. 727355da1ebSSage Weil */ 728355da1ebSSage Weil static void update_dentry_lease(struct dentry *dentry, 729355da1ebSSage Weil struct ceph_mds_reply_lease *lease, 730355da1ebSSage Weil struct ceph_mds_session *session, 731355da1ebSSage Weil unsigned long from_time) 732355da1ebSSage Weil { 733355da1ebSSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 734355da1ebSSage Weil long unsigned duration = le32_to_cpu(lease->duration_ms); 735355da1ebSSage Weil long unsigned ttl = from_time + (duration * HZ) / 1000; 736355da1ebSSage Weil long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 737355da1ebSSage Weil struct inode *dir; 738355da1ebSSage Weil 739355da1ebSSage Weil /* only track leases on regular dentries */ 740355da1ebSSage Weil if (dentry->d_op != &ceph_dentry_ops) 741355da1ebSSage Weil return; 742355da1ebSSage Weil 743355da1ebSSage Weil spin_lock(&dentry->d_lock); 744355da1ebSSage Weil dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 745355da1ebSSage Weil dentry, le16_to_cpu(lease->mask), duration, ttl); 746355da1ebSSage Weil 747355da1ebSSage Weil /* make lease_rdcache_gen match directory */ 748355da1ebSSage Weil dir = dentry->d_parent->d_inode; 749355da1ebSSage Weil di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 750355da1ebSSage Weil 751355da1ebSSage Weil if (lease->mask == 0) 752355da1ebSSage Weil goto out_unlock; 753355da1ebSSage Weil 754355da1ebSSage Weil if (di->lease_gen == session->s_cap_gen && 755355da1ebSSage Weil time_before(ttl, dentry->d_time)) 756355da1ebSSage Weil goto out_unlock; /* we already have a newer lease. */ 757355da1ebSSage Weil 758355da1ebSSage Weil if (di->lease_session && di->lease_session != session) 759355da1ebSSage Weil goto out_unlock; 760355da1ebSSage Weil 761355da1ebSSage Weil ceph_dentry_lru_touch(dentry); 762355da1ebSSage Weil 763355da1ebSSage Weil if (!di->lease_session) 764355da1ebSSage Weil di->lease_session = ceph_get_mds_session(session); 765355da1ebSSage Weil di->lease_gen = session->s_cap_gen; 766355da1ebSSage Weil di->lease_seq = le32_to_cpu(lease->seq); 767355da1ebSSage Weil di->lease_renew_after = half_ttl; 768355da1ebSSage Weil di->lease_renew_from = 0; 769355da1ebSSage Weil dentry->d_time = ttl; 770355da1ebSSage Weil out_unlock: 771355da1ebSSage Weil spin_unlock(&dentry->d_lock); 772355da1ebSSage Weil return; 773355da1ebSSage Weil } 774355da1ebSSage Weil 775355da1ebSSage Weil /* 776355da1ebSSage Weil * splice a dentry to an inode. 777355da1ebSSage Weil * caller must hold directory i_mutex for this to be safe. 778355da1ebSSage Weil * 779355da1ebSSage Weil * we will only rehash the resulting dentry if @prehash is 780355da1ebSSage Weil * true; @prehash will be set to false (for the benefit of 781355da1ebSSage Weil * the caller) if we fail. 782355da1ebSSage Weil */ 783355da1ebSSage Weil static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 784355da1ebSSage Weil bool *prehash) 785355da1ebSSage Weil { 786355da1ebSSage Weil struct dentry *realdn; 787355da1ebSSage Weil 788355da1ebSSage Weil /* dn must be unhashed */ 789355da1ebSSage Weil if (!d_unhashed(dn)) 790355da1ebSSage Weil d_drop(dn); 791355da1ebSSage Weil realdn = d_materialise_unique(dn, in); 792355da1ebSSage Weil if (IS_ERR(realdn)) { 793355da1ebSSage Weil pr_err("splice_dentry error %p inode %p ino %llx.%llx\n", 794355da1ebSSage Weil dn, in, ceph_vinop(in)); 795355da1ebSSage Weil if (prehash) 796355da1ebSSage Weil *prehash = false; /* don't rehash on error */ 797355da1ebSSage Weil dn = realdn; /* note realdn contains the error */ 798355da1ebSSage Weil goto out; 799355da1ebSSage Weil } else if (realdn) { 800355da1ebSSage Weil dout("dn %p (%d) spliced with %p (%d) " 801355da1ebSSage Weil "inode %p ino %llx.%llx\n", 802355da1ebSSage Weil dn, atomic_read(&dn->d_count), 803355da1ebSSage Weil realdn, atomic_read(&realdn->d_count), 804355da1ebSSage Weil realdn->d_inode, ceph_vinop(realdn->d_inode)); 805355da1ebSSage Weil dput(dn); 806355da1ebSSage Weil dn = realdn; 807355da1ebSSage Weil } else { 808355da1ebSSage Weil BUG_ON(!ceph_dentry(dn)); 809355da1ebSSage Weil 810355da1ebSSage Weil dout("dn %p attached to %p ino %llx.%llx\n", 811355da1ebSSage Weil dn, dn->d_inode, ceph_vinop(dn->d_inode)); 812355da1ebSSage Weil } 813355da1ebSSage Weil if ((!prehash || *prehash) && d_unhashed(dn)) 814355da1ebSSage Weil d_rehash(dn); 815355da1ebSSage Weil out: 816355da1ebSSage Weil return dn; 817355da1ebSSage Weil } 818355da1ebSSage Weil 819355da1ebSSage Weil /* 820355da1ebSSage Weil * Incorporate results into the local cache. This is either just 821355da1ebSSage Weil * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 822355da1ebSSage Weil * after a lookup). 823355da1ebSSage Weil * 824355da1ebSSage Weil * A reply may contain 825355da1ebSSage Weil * a directory inode along with a dentry. 826355da1ebSSage Weil * and/or a target inode 827355da1ebSSage Weil * 828355da1ebSSage Weil * Called with snap_rwsem (read). 829355da1ebSSage Weil */ 830355da1ebSSage Weil int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 831355da1ebSSage Weil struct ceph_mds_session *session) 832355da1ebSSage Weil { 833355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 834355da1ebSSage Weil struct inode *in = NULL; 835355da1ebSSage Weil struct ceph_mds_reply_inode *ininfo; 836355da1ebSSage Weil struct ceph_vino vino; 837355da1ebSSage Weil int i = 0; 838355da1ebSSage Weil int err = 0; 839355da1ebSSage Weil 840355da1ebSSage Weil dout("fill_trace %p is_dentry %d is_target %d\n", req, 841355da1ebSSage Weil rinfo->head->is_dentry, rinfo->head->is_target); 842355da1ebSSage Weil 843355da1ebSSage Weil #if 0 844355da1ebSSage Weil /* 845355da1ebSSage Weil * Debugging hook: 846355da1ebSSage Weil * 847355da1ebSSage Weil * If we resend completed ops to a recovering mds, we get no 848355da1ebSSage Weil * trace. Since that is very rare, pretend this is the case 849355da1ebSSage Weil * to ensure the 'no trace' handlers in the callers behave. 850355da1ebSSage Weil * 851355da1ebSSage Weil * Fill in inodes unconditionally to avoid breaking cap 852355da1ebSSage Weil * invariants. 853355da1ebSSage Weil */ 854355da1ebSSage Weil if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 855355da1ebSSage Weil pr_info("fill_trace faking empty trace on %lld %s\n", 856355da1ebSSage Weil req->r_tid, ceph_mds_op_name(rinfo->head->op)); 857355da1ebSSage Weil if (rinfo->head->is_dentry) { 858355da1ebSSage Weil rinfo->head->is_dentry = 0; 859355da1ebSSage Weil err = fill_inode(req->r_locked_dir, 860355da1ebSSage Weil &rinfo->diri, rinfo->dirfrag, 861355da1ebSSage Weil session, req->r_request_started, -1); 862355da1ebSSage Weil } 863355da1ebSSage Weil if (rinfo->head->is_target) { 864355da1ebSSage Weil rinfo->head->is_target = 0; 865355da1ebSSage Weil ininfo = rinfo->targeti.in; 866355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 867355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 868355da1ebSSage Weil in = ceph_get_inode(sb, vino); 869355da1ebSSage Weil err = fill_inode(in, &rinfo->targeti, NULL, 870355da1ebSSage Weil session, req->r_request_started, 871355da1ebSSage Weil req->r_fmode); 872355da1ebSSage Weil iput(in); 873355da1ebSSage Weil } 874355da1ebSSage Weil } 875355da1ebSSage Weil #endif 876355da1ebSSage Weil 877355da1ebSSage Weil if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 878355da1ebSSage Weil dout("fill_trace reply is empty!\n"); 879355da1ebSSage Weil if (rinfo->head->result == 0 && req->r_locked_dir) { 880355da1ebSSage Weil struct ceph_inode_info *ci = 881355da1ebSSage Weil ceph_inode(req->r_locked_dir); 882355da1ebSSage Weil dout(" clearing %p complete (empty trace)\n", 883355da1ebSSage Weil req->r_locked_dir); 884355da1ebSSage Weil ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 885355da1ebSSage Weil ci->i_release_count++; 886355da1ebSSage Weil } 887355da1ebSSage Weil return 0; 888355da1ebSSage Weil } 889355da1ebSSage Weil 890355da1ebSSage Weil if (rinfo->head->is_dentry) { 891355da1ebSSage Weil /* 892355da1ebSSage Weil * lookup link rename : null -> possibly existing inode 893355da1ebSSage Weil * mknod symlink mkdir : null -> new inode 894355da1ebSSage Weil * unlink : linked -> null 895355da1ebSSage Weil */ 896355da1ebSSage Weil struct inode *dir = req->r_locked_dir; 897355da1ebSSage Weil struct dentry *dn = req->r_dentry; 898355da1ebSSage Weil bool have_dir_cap, have_lease; 899355da1ebSSage Weil 900355da1ebSSage Weil BUG_ON(!dn); 901355da1ebSSage Weil BUG_ON(!dir); 902355da1ebSSage Weil BUG_ON(dn->d_parent->d_inode != dir); 903355da1ebSSage Weil BUG_ON(ceph_ino(dir) != 904355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->ino)); 905355da1ebSSage Weil BUG_ON(ceph_snap(dir) != 906355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->snapid)); 907355da1ebSSage Weil 908355da1ebSSage Weil err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 909355da1ebSSage Weil session, req->r_request_started, -1, 910355da1ebSSage Weil &req->r_caps_reservation); 911355da1ebSSage Weil if (err < 0) 912355da1ebSSage Weil return err; 913355da1ebSSage Weil 914355da1ebSSage Weil /* do we have a lease on the whole dir? */ 915355da1ebSSage Weil have_dir_cap = 916355da1ebSSage Weil (le32_to_cpu(rinfo->diri.in->cap.caps) & 917355da1ebSSage Weil CEPH_CAP_FILE_SHARED); 918355da1ebSSage Weil 919355da1ebSSage Weil /* do we have a dn lease? */ 920355da1ebSSage Weil have_lease = have_dir_cap || 921355da1ebSSage Weil (le16_to_cpu(rinfo->dlease->mask) & 922355da1ebSSage Weil CEPH_LOCK_DN); 923355da1ebSSage Weil 924355da1ebSSage Weil if (!have_lease) 925355da1ebSSage Weil dout("fill_trace no dentry lease or dir cap\n"); 926355da1ebSSage Weil 927355da1ebSSage Weil /* rename? */ 928355da1ebSSage Weil if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 929355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 930355da1ebSSage Weil req->r_old_dentry, 931355da1ebSSage Weil req->r_old_dentry->d_name.len, 932355da1ebSSage Weil req->r_old_dentry->d_name.name, 933355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 934355da1ebSSage Weil dout("fill_trace doing d_move %p -> %p\n", 935355da1ebSSage Weil req->r_old_dentry, dn); 936355da1ebSSage Weil d_move(req->r_old_dentry, dn); 937355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 938355da1ebSSage Weil req->r_old_dentry, 939355da1ebSSage Weil req->r_old_dentry->d_name.len, 940355da1ebSSage Weil req->r_old_dentry->d_name.name, 941355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 942c4a29f26SSage Weil /* ensure target dentry is invalidated, despite 943c4a29f26SSage Weil rehashing bug in vfs_rename_dir */ 944c4a29f26SSage Weil dn->d_time = jiffies; 945c4a29f26SSage Weil ceph_dentry(dn)->lease_shared_gen = 0; 946355da1ebSSage Weil /* take overwritten dentry's readdir offset */ 947355da1ebSSage Weil ceph_dentry(req->r_old_dentry)->offset = 948355da1ebSSage Weil ceph_dentry(dn)->offset; 949355da1ebSSage Weil dn = req->r_old_dentry; /* use old_dentry */ 950355da1ebSSage Weil in = dn->d_inode; 951355da1ebSSage Weil } 952355da1ebSSage Weil 953355da1ebSSage Weil /* null dentry? */ 954355da1ebSSage Weil if (!rinfo->head->is_target) { 955355da1ebSSage Weil dout("fill_trace null dentry\n"); 956355da1ebSSage Weil if (dn->d_inode) { 957355da1ebSSage Weil dout("d_delete %p\n", dn); 958355da1ebSSage Weil d_delete(dn); 959355da1ebSSage Weil } else { 960355da1ebSSage Weil dout("d_instantiate %p NULL\n", dn); 961355da1ebSSage Weil d_instantiate(dn, NULL); 962355da1ebSSage Weil if (have_lease && d_unhashed(dn)) 963355da1ebSSage Weil d_rehash(dn); 964355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, 965355da1ebSSage Weil session, 966355da1ebSSage Weil req->r_request_started); 967355da1ebSSage Weil } 968355da1ebSSage Weil goto done; 969355da1ebSSage Weil } 970355da1ebSSage Weil 971355da1ebSSage Weil /* attach proper inode */ 972355da1ebSSage Weil ininfo = rinfo->targeti.in; 973355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 974355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 975355da1ebSSage Weil if (!dn->d_inode) { 976355da1ebSSage Weil in = ceph_get_inode(sb, vino); 977355da1ebSSage Weil if (IS_ERR(in)) { 978355da1ebSSage Weil pr_err("fill_trace bad get_inode " 979355da1ebSSage Weil "%llx.%llx\n", vino.ino, vino.snap); 980355da1ebSSage Weil err = PTR_ERR(in); 981355da1ebSSage Weil d_delete(dn); 982355da1ebSSage Weil goto done; 983355da1ebSSage Weil } 984355da1ebSSage Weil dn = splice_dentry(dn, in, &have_lease); 985355da1ebSSage Weil if (IS_ERR(dn)) { 986355da1ebSSage Weil err = PTR_ERR(dn); 987355da1ebSSage Weil goto done; 988355da1ebSSage Weil } 989355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 990355da1ebSSage Weil igrab(in); 991355da1ebSSage Weil } else if (ceph_ino(in) == vino.ino && 992355da1ebSSage Weil ceph_snap(in) == vino.snap) { 993355da1ebSSage Weil igrab(in); 994355da1ebSSage Weil } else { 995355da1ebSSage Weil dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 996355da1ebSSage Weil dn, in, ceph_ino(in), ceph_snap(in), 997355da1ebSSage Weil vino.ino, vino.snap); 998355da1ebSSage Weil have_lease = false; 999355da1ebSSage Weil in = NULL; 1000355da1ebSSage Weil } 1001355da1ebSSage Weil 1002355da1ebSSage Weil if (have_lease) 1003355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, session, 1004355da1ebSSage Weil req->r_request_started); 1005355da1ebSSage Weil dout(" final dn %p\n", dn); 1006355da1ebSSage Weil i++; 1007355da1ebSSage Weil } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1008355da1ebSSage Weil req->r_op == CEPH_MDS_OP_MKSNAP) { 1009355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1010355da1ebSSage Weil 1011355da1ebSSage Weil /* fill out a snapdir LOOKUPSNAP dentry */ 1012355da1ebSSage Weil BUG_ON(!dn); 1013355da1ebSSage Weil BUG_ON(!req->r_locked_dir); 1014355da1ebSSage Weil BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1015355da1ebSSage Weil ininfo = rinfo->targeti.in; 1016355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1017355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1018355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1019355da1ebSSage Weil if (IS_ERR(in)) { 1020355da1ebSSage Weil pr_err("fill_inode get_inode badness %llx.%llx\n", 1021355da1ebSSage Weil vino.ino, vino.snap); 1022355da1ebSSage Weil err = PTR_ERR(in); 1023355da1ebSSage Weil d_delete(dn); 1024355da1ebSSage Weil goto done; 1025355da1ebSSage Weil } 1026355da1ebSSage Weil dout(" linking snapped dir %p to dn %p\n", in, dn); 1027355da1ebSSage Weil dn = splice_dentry(dn, in, NULL); 1028355da1ebSSage Weil if (IS_ERR(dn)) { 1029355da1ebSSage Weil err = PTR_ERR(dn); 1030355da1ebSSage Weil goto done; 1031355da1ebSSage Weil } 1032355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 1033355da1ebSSage Weil igrab(in); 1034355da1ebSSage Weil rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1035355da1ebSSage Weil } 1036355da1ebSSage Weil 1037355da1ebSSage Weil if (rinfo->head->is_target) { 1038355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1039355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1040355da1ebSSage Weil 1041355da1ebSSage Weil if (in == NULL || ceph_ino(in) != vino.ino || 1042355da1ebSSage Weil ceph_snap(in) != vino.snap) { 1043355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1044355da1ebSSage Weil if (IS_ERR(in)) { 1045355da1ebSSage Weil err = PTR_ERR(in); 1046355da1ebSSage Weil goto done; 1047355da1ebSSage Weil } 1048355da1ebSSage Weil } 1049355da1ebSSage Weil req->r_target_inode = in; 1050355da1ebSSage Weil 1051355da1ebSSage Weil err = fill_inode(in, 1052355da1ebSSage Weil &rinfo->targeti, NULL, 1053355da1ebSSage Weil session, req->r_request_started, 1054355da1ebSSage Weil (le32_to_cpu(rinfo->head->result) == 0) ? 1055355da1ebSSage Weil req->r_fmode : -1, 1056355da1ebSSage Weil &req->r_caps_reservation); 1057355da1ebSSage Weil if (err < 0) { 1058355da1ebSSage Weil pr_err("fill_inode badness %p %llx.%llx\n", 1059355da1ebSSage Weil in, ceph_vinop(in)); 1060355da1ebSSage Weil goto done; 1061355da1ebSSage Weil } 1062355da1ebSSage Weil } 1063355da1ebSSage Weil 1064355da1ebSSage Weil done: 1065355da1ebSSage Weil dout("fill_trace done err=%d\n", err); 1066355da1ebSSage Weil return err; 1067355da1ebSSage Weil } 1068355da1ebSSage Weil 1069355da1ebSSage Weil /* 1070355da1ebSSage Weil * Prepopulate our cache with readdir results, leases, etc. 1071355da1ebSSage Weil */ 1072355da1ebSSage Weil int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1073355da1ebSSage Weil struct ceph_mds_session *session) 1074355da1ebSSage Weil { 1075355da1ebSSage Weil struct dentry *parent = req->r_dentry; 1076355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1077355da1ebSSage Weil struct qstr dname; 1078355da1ebSSage Weil struct dentry *dn; 1079355da1ebSSage Weil struct inode *in; 1080355da1ebSSage Weil int err = 0, i; 1081355da1ebSSage Weil struct inode *snapdir = NULL; 1082355da1ebSSage Weil struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1083355da1ebSSage Weil u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1084355da1ebSSage Weil struct ceph_dentry_info *di; 1085355da1ebSSage Weil 1086355da1ebSSage Weil if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1087355da1ebSSage Weil snapdir = ceph_get_snapdir(parent->d_inode); 1088355da1ebSSage Weil parent = d_find_alias(snapdir); 1089355da1ebSSage Weil dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1090355da1ebSSage Weil rinfo->dir_nr, parent); 1091355da1ebSSage Weil } else { 1092355da1ebSSage Weil dout("readdir_prepopulate %d items under dn %p\n", 1093355da1ebSSage Weil rinfo->dir_nr, parent); 1094355da1ebSSage Weil if (rinfo->dir_dir) 1095355da1ebSSage Weil ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1096355da1ebSSage Weil } 1097355da1ebSSage Weil 1098355da1ebSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 1099355da1ebSSage Weil struct ceph_vino vino; 1100355da1ebSSage Weil 1101355da1ebSSage Weil dname.name = rinfo->dir_dname[i]; 1102355da1ebSSage Weil dname.len = rinfo->dir_dname_len[i]; 1103355da1ebSSage Weil dname.hash = full_name_hash(dname.name, dname.len); 1104355da1ebSSage Weil 1105355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1106355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1107355da1ebSSage Weil 1108355da1ebSSage Weil retry_lookup: 1109355da1ebSSage Weil dn = d_lookup(parent, &dname); 1110355da1ebSSage Weil dout("d_lookup on parent=%p name=%.*s got %p\n", 1111355da1ebSSage Weil parent, dname.len, dname.name, dn); 1112355da1ebSSage Weil 1113355da1ebSSage Weil if (!dn) { 1114355da1ebSSage Weil dn = d_alloc(parent, &dname); 1115355da1ebSSage Weil dout("d_alloc %p '%.*s' = %p\n", parent, 1116355da1ebSSage Weil dname.len, dname.name, dn); 1117355da1ebSSage Weil if (dn == NULL) { 1118355da1ebSSage Weil dout("d_alloc badness\n"); 1119355da1ebSSage Weil err = -ENOMEM; 1120355da1ebSSage Weil goto out; 1121355da1ebSSage Weil } 1122355da1ebSSage Weil err = ceph_init_dentry(dn); 1123355da1ebSSage Weil if (err < 0) 1124355da1ebSSage Weil goto out; 1125355da1ebSSage Weil } else if (dn->d_inode && 1126355da1ebSSage Weil (ceph_ino(dn->d_inode) != vino.ino || 1127355da1ebSSage Weil ceph_snap(dn->d_inode) != vino.snap)) { 1128355da1ebSSage Weil dout(" dn %p points to wrong inode %p\n", 1129355da1ebSSage Weil dn, dn->d_inode); 1130355da1ebSSage Weil d_delete(dn); 1131355da1ebSSage Weil dput(dn); 1132355da1ebSSage Weil goto retry_lookup; 1133355da1ebSSage Weil } else { 1134355da1ebSSage Weil /* reorder parent's d_subdirs */ 1135355da1ebSSage Weil spin_lock(&dcache_lock); 1136355da1ebSSage Weil spin_lock(&dn->d_lock); 1137355da1ebSSage Weil list_move(&dn->d_u.d_child, &parent->d_subdirs); 1138355da1ebSSage Weil spin_unlock(&dn->d_lock); 1139355da1ebSSage Weil spin_unlock(&dcache_lock); 1140355da1ebSSage Weil } 1141355da1ebSSage Weil 1142355da1ebSSage Weil di = dn->d_fsdata; 1143355da1ebSSage Weil di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1144355da1ebSSage Weil 1145355da1ebSSage Weil /* inode */ 1146355da1ebSSage Weil if (dn->d_inode) { 1147355da1ebSSage Weil in = dn->d_inode; 1148355da1ebSSage Weil } else { 1149355da1ebSSage Weil in = ceph_get_inode(parent->d_sb, vino); 1150355da1ebSSage Weil if (in == NULL) { 1151355da1ebSSage Weil dout("new_inode badness\n"); 1152355da1ebSSage Weil d_delete(dn); 1153355da1ebSSage Weil dput(dn); 1154355da1ebSSage Weil err = -ENOMEM; 1155355da1ebSSage Weil goto out; 1156355da1ebSSage Weil } 1157355da1ebSSage Weil dn = splice_dentry(dn, in, NULL); 1158355da1ebSSage Weil } 1159355da1ebSSage Weil 1160355da1ebSSage Weil if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1161355da1ebSSage Weil req->r_request_started, -1, 1162355da1ebSSage Weil &req->r_caps_reservation) < 0) { 1163355da1ebSSage Weil pr_err("fill_inode badness on %p\n", in); 1164355da1ebSSage Weil dput(dn); 1165355da1ebSSage Weil continue; 1166355da1ebSSage Weil } 1167355da1ebSSage Weil update_dentry_lease(dn, rinfo->dir_dlease[i], 1168355da1ebSSage Weil req->r_session, req->r_request_started); 1169355da1ebSSage Weil dput(dn); 1170355da1ebSSage Weil } 1171355da1ebSSage Weil req->r_did_prepopulate = true; 1172355da1ebSSage Weil 1173355da1ebSSage Weil out: 1174355da1ebSSage Weil if (snapdir) { 1175355da1ebSSage Weil iput(snapdir); 1176355da1ebSSage Weil dput(parent); 1177355da1ebSSage Weil } 1178355da1ebSSage Weil dout("readdir_prepopulate done\n"); 1179355da1ebSSage Weil return err; 1180355da1ebSSage Weil } 1181355da1ebSSage Weil 1182355da1ebSSage Weil int ceph_inode_set_size(struct inode *inode, loff_t size) 1183355da1ebSSage Weil { 1184355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1185355da1ebSSage Weil int ret = 0; 1186355da1ebSSage Weil 1187355da1ebSSage Weil spin_lock(&inode->i_lock); 1188355da1ebSSage Weil dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1189355da1ebSSage Weil inode->i_size = size; 1190355da1ebSSage Weil inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1191355da1ebSSage Weil 1192355da1ebSSage Weil /* tell the MDS if we are approaching max_size */ 1193355da1ebSSage Weil if ((size << 1) >= ci->i_max_size && 1194355da1ebSSage Weil (ci->i_reported_size << 1) < ci->i_max_size) 1195355da1ebSSage Weil ret = 1; 1196355da1ebSSage Weil 1197355da1ebSSage Weil spin_unlock(&inode->i_lock); 1198355da1ebSSage Weil return ret; 1199355da1ebSSage Weil } 1200355da1ebSSage Weil 1201355da1ebSSage Weil /* 1202355da1ebSSage Weil * Write back inode data in a worker thread. (This can't be done 1203355da1ebSSage Weil * in the message handler context.) 1204355da1ebSSage Weil */ 1205355da1ebSSage Weil void ceph_inode_writeback(struct work_struct *work) 1206355da1ebSSage Weil { 1207355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1208355da1ebSSage Weil i_wb_work); 1209355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1210355da1ebSSage Weil 1211355da1ebSSage Weil dout("writeback %p\n", inode); 1212355da1ebSSage Weil filemap_fdatawrite(&inode->i_data); 1213355da1ebSSage Weil iput(inode); 1214355da1ebSSage Weil } 1215355da1ebSSage Weil 1216355da1ebSSage Weil /* 1217355da1ebSSage Weil * Invalidate inode pages in a worker thread. (This can't be done 1218355da1ebSSage Weil * in the message handler context.) 1219355da1ebSSage Weil */ 1220355da1ebSSage Weil static void ceph_inode_invalidate_pages(struct work_struct *work) 1221355da1ebSSage Weil { 1222355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1223355da1ebSSage Weil i_pg_inv_work); 1224355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1225355da1ebSSage Weil u32 orig_gen; 1226355da1ebSSage Weil int check = 0; 1227355da1ebSSage Weil 1228355da1ebSSage Weil spin_lock(&inode->i_lock); 1229355da1ebSSage Weil dout("invalidate_pages %p gen %d revoking %d\n", inode, 1230355da1ebSSage Weil ci->i_rdcache_gen, ci->i_rdcache_revoking); 1231355da1ebSSage Weil if (ci->i_rdcache_gen == 0 || 1232355da1ebSSage Weil ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1233355da1ebSSage Weil BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen); 1234355da1ebSSage Weil /* nevermind! */ 1235355da1ebSSage Weil ci->i_rdcache_revoking = 0; 1236355da1ebSSage Weil spin_unlock(&inode->i_lock); 1237355da1ebSSage Weil goto out; 1238355da1ebSSage Weil } 1239355da1ebSSage Weil orig_gen = ci->i_rdcache_gen; 1240355da1ebSSage Weil spin_unlock(&inode->i_lock); 1241355da1ebSSage Weil 1242355da1ebSSage Weil truncate_inode_pages(&inode->i_data, 0); 1243355da1ebSSage Weil 1244355da1ebSSage Weil spin_lock(&inode->i_lock); 1245355da1ebSSage Weil if (orig_gen == ci->i_rdcache_gen) { 1246355da1ebSSage Weil dout("invalidate_pages %p gen %d successful\n", inode, 1247355da1ebSSage Weil ci->i_rdcache_gen); 1248355da1ebSSage Weil ci->i_rdcache_gen = 0; 1249355da1ebSSage Weil ci->i_rdcache_revoking = 0; 1250355da1ebSSage Weil check = 1; 1251355da1ebSSage Weil } else { 1252355da1ebSSage Weil dout("invalidate_pages %p gen %d raced, gen now %d\n", 1253355da1ebSSage Weil inode, orig_gen, ci->i_rdcache_gen); 1254355da1ebSSage Weil } 1255355da1ebSSage Weil spin_unlock(&inode->i_lock); 1256355da1ebSSage Weil 1257355da1ebSSage Weil if (check) 1258355da1ebSSage Weil ceph_check_caps(ci, 0, NULL); 1259355da1ebSSage Weil out: 1260355da1ebSSage Weil iput(inode); 1261355da1ebSSage Weil } 1262355da1ebSSage Weil 1263355da1ebSSage Weil 1264355da1ebSSage Weil /* 1265355da1ebSSage Weil * called by trunc_wq; take i_mutex ourselves 1266355da1ebSSage Weil * 1267355da1ebSSage Weil * We also truncate in a separate thread as well. 1268355da1ebSSage Weil */ 1269355da1ebSSage Weil void ceph_vmtruncate_work(struct work_struct *work) 1270355da1ebSSage Weil { 1271355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1272355da1ebSSage Weil i_vmtruncate_work); 1273355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1274355da1ebSSage Weil 1275355da1ebSSage Weil dout("vmtruncate_work %p\n", inode); 1276355da1ebSSage Weil mutex_lock(&inode->i_mutex); 1277355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1278355da1ebSSage Weil mutex_unlock(&inode->i_mutex); 1279355da1ebSSage Weil iput(inode); 1280355da1ebSSage Weil } 1281355da1ebSSage Weil 1282355da1ebSSage Weil /* 1283355da1ebSSage Weil * called with i_mutex held. 1284355da1ebSSage Weil * 1285355da1ebSSage Weil * Make sure any pending truncation is applied before doing anything 1286355da1ebSSage Weil * that may depend on it. 1287355da1ebSSage Weil */ 1288355da1ebSSage Weil void __ceph_do_pending_vmtruncate(struct inode *inode) 1289355da1ebSSage Weil { 1290355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1291355da1ebSSage Weil u64 to; 1292355da1ebSSage Weil int wrbuffer_refs, wake = 0; 1293355da1ebSSage Weil 1294355da1ebSSage Weil retry: 1295355da1ebSSage Weil spin_lock(&inode->i_lock); 1296355da1ebSSage Weil if (ci->i_truncate_pending == 0) { 1297355da1ebSSage Weil dout("__do_pending_vmtruncate %p none pending\n", inode); 1298355da1ebSSage Weil spin_unlock(&inode->i_lock); 1299355da1ebSSage Weil return; 1300355da1ebSSage Weil } 1301355da1ebSSage Weil 1302355da1ebSSage Weil /* 1303355da1ebSSage Weil * make sure any dirty snapped pages are flushed before we 1304355da1ebSSage Weil * possibly truncate them.. so write AND block! 1305355da1ebSSage Weil */ 1306355da1ebSSage Weil if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1307355da1ebSSage Weil dout("__do_pending_vmtruncate %p flushing snaps first\n", 1308355da1ebSSage Weil inode); 1309355da1ebSSage Weil spin_unlock(&inode->i_lock); 1310355da1ebSSage Weil filemap_write_and_wait_range(&inode->i_data, 0, 1311355da1ebSSage Weil inode->i_sb->s_maxbytes); 1312355da1ebSSage Weil goto retry; 1313355da1ebSSage Weil } 1314355da1ebSSage Weil 1315355da1ebSSage Weil to = ci->i_truncate_size; 1316355da1ebSSage Weil wrbuffer_refs = ci->i_wrbuffer_ref; 1317355da1ebSSage Weil dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1318355da1ebSSage Weil ci->i_truncate_pending, to); 1319355da1ebSSage Weil spin_unlock(&inode->i_lock); 1320355da1ebSSage Weil 1321355da1ebSSage Weil truncate_inode_pages(inode->i_mapping, to); 1322355da1ebSSage Weil 1323355da1ebSSage Weil spin_lock(&inode->i_lock); 1324355da1ebSSage Weil ci->i_truncate_pending--; 1325355da1ebSSage Weil if (ci->i_truncate_pending == 0) 1326355da1ebSSage Weil wake = 1; 1327355da1ebSSage Weil spin_unlock(&inode->i_lock); 1328355da1ebSSage Weil 1329355da1ebSSage Weil if (wrbuffer_refs == 0) 1330355da1ebSSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1331355da1ebSSage Weil if (wake) 1332355da1ebSSage Weil wake_up(&ci->i_cap_wq); 1333355da1ebSSage Weil } 1334355da1ebSSage Weil 1335355da1ebSSage Weil 1336355da1ebSSage Weil /* 1337355da1ebSSage Weil * symlinks 1338355da1ebSSage Weil */ 1339355da1ebSSage Weil static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1340355da1ebSSage Weil { 1341355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1342355da1ebSSage Weil nd_set_link(nd, ci->i_symlink); 1343355da1ebSSage Weil return NULL; 1344355da1ebSSage Weil } 1345355da1ebSSage Weil 1346355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops = { 1347355da1ebSSage Weil .readlink = generic_readlink, 1348355da1ebSSage Weil .follow_link = ceph_sym_follow_link, 1349355da1ebSSage Weil }; 1350355da1ebSSage Weil 1351355da1ebSSage Weil /* 1352355da1ebSSage Weil * setattr 1353355da1ebSSage Weil */ 1354355da1ebSSage Weil int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1355355da1ebSSage Weil { 1356355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1357355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1358355da1ebSSage Weil struct inode *parent_inode = dentry->d_parent->d_inode; 1359355da1ebSSage Weil const unsigned int ia_valid = attr->ia_valid; 1360355da1ebSSage Weil struct ceph_mds_request *req; 1361355da1ebSSage Weil struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc; 1362355da1ebSSage Weil int issued; 1363355da1ebSSage Weil int release = 0, dirtied = 0; 1364355da1ebSSage Weil int mask = 0; 1365355da1ebSSage Weil int err = 0; 1366355da1ebSSage Weil int queue_trunc = 0; 1367355da1ebSSage Weil 1368355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1369355da1ebSSage Weil return -EROFS; 1370355da1ebSSage Weil 1371355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1372355da1ebSSage Weil 1373355da1ebSSage Weil err = inode_change_ok(inode, attr); 1374355da1ebSSage Weil if (err != 0) 1375355da1ebSSage Weil return err; 1376355da1ebSSage Weil 1377355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1378355da1ebSSage Weil USE_AUTH_MDS); 1379355da1ebSSage Weil if (IS_ERR(req)) 1380355da1ebSSage Weil return PTR_ERR(req); 1381355da1ebSSage Weil 1382355da1ebSSage Weil spin_lock(&inode->i_lock); 1383355da1ebSSage Weil issued = __ceph_caps_issued(ci, NULL); 1384355da1ebSSage Weil dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1385355da1ebSSage Weil 1386355da1ebSSage Weil if (ia_valid & ATTR_UID) { 1387355da1ebSSage Weil dout("setattr %p uid %d -> %d\n", inode, 1388355da1ebSSage Weil inode->i_uid, attr->ia_uid); 1389355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1390355da1ebSSage Weil inode->i_uid = attr->ia_uid; 1391355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1392355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1393355da1ebSSage Weil attr->ia_uid != inode->i_uid) { 1394355da1ebSSage Weil req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1395355da1ebSSage Weil mask |= CEPH_SETATTR_UID; 1396355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1397355da1ebSSage Weil } 1398355da1ebSSage Weil } 1399355da1ebSSage Weil if (ia_valid & ATTR_GID) { 1400355da1ebSSage Weil dout("setattr %p gid %d -> %d\n", inode, 1401355da1ebSSage Weil inode->i_gid, attr->ia_gid); 1402355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1403355da1ebSSage Weil inode->i_gid = attr->ia_gid; 1404355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1405355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1406355da1ebSSage Weil attr->ia_gid != inode->i_gid) { 1407355da1ebSSage Weil req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1408355da1ebSSage Weil mask |= CEPH_SETATTR_GID; 1409355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1410355da1ebSSage Weil } 1411355da1ebSSage Weil } 1412355da1ebSSage Weil if (ia_valid & ATTR_MODE) { 1413355da1ebSSage Weil dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1414355da1ebSSage Weil attr->ia_mode); 1415355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1416355da1ebSSage Weil inode->i_mode = attr->ia_mode; 1417355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1418355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1419355da1ebSSage Weil attr->ia_mode != inode->i_mode) { 1420355da1ebSSage Weil req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1421355da1ebSSage Weil mask |= CEPH_SETATTR_MODE; 1422355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1423355da1ebSSage Weil } 1424355da1ebSSage Weil } 1425355da1ebSSage Weil 1426355da1ebSSage Weil if (ia_valid & ATTR_ATIME) { 1427355da1ebSSage Weil dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1428355da1ebSSage Weil inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1429355da1ebSSage Weil attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1430355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1431355da1ebSSage Weil ci->i_time_warp_seq++; 1432355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1433355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1434355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1435355da1ebSSage Weil timespec_compare(&inode->i_atime, 1436355da1ebSSage Weil &attr->ia_atime) < 0) { 1437355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1438355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1439355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1440355da1ebSSage Weil !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1441355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.atime, 1442355da1ebSSage Weil &attr->ia_atime); 1443355da1ebSSage Weil mask |= CEPH_SETATTR_ATIME; 1444355da1ebSSage Weil release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1445355da1ebSSage Weil CEPH_CAP_FILE_WR; 1446355da1ebSSage Weil } 1447355da1ebSSage Weil } 1448355da1ebSSage Weil if (ia_valid & ATTR_MTIME) { 1449355da1ebSSage Weil dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1450355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1451355da1ebSSage Weil attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1452355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1453355da1ebSSage Weil ci->i_time_warp_seq++; 1454355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1455355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1456355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1457355da1ebSSage Weil timespec_compare(&inode->i_mtime, 1458355da1ebSSage Weil &attr->ia_mtime) < 0) { 1459355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1460355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1461355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1462355da1ebSSage Weil !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1463355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.mtime, 1464355da1ebSSage Weil &attr->ia_mtime); 1465355da1ebSSage Weil mask |= CEPH_SETATTR_MTIME; 1466355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1467355da1ebSSage Weil CEPH_CAP_FILE_WR; 1468355da1ebSSage Weil } 1469355da1ebSSage Weil } 1470355da1ebSSage Weil if (ia_valid & ATTR_SIZE) { 1471355da1ebSSage Weil dout("setattr %p size %lld -> %lld\n", inode, 1472355da1ebSSage Weil inode->i_size, attr->ia_size); 1473355da1ebSSage Weil if (attr->ia_size > inode->i_sb->s_maxbytes) { 1474355da1ebSSage Weil err = -EINVAL; 1475355da1ebSSage Weil goto out; 1476355da1ebSSage Weil } 1477355da1ebSSage Weil if ((issued & CEPH_CAP_FILE_EXCL) && 1478355da1ebSSage Weil attr->ia_size > inode->i_size) { 1479355da1ebSSage Weil inode->i_size = attr->ia_size; 1480355da1ebSSage Weil if (attr->ia_size < inode->i_size) { 1481355da1ebSSage Weil ci->i_truncate_size = attr->ia_size; 1482355da1ebSSage Weil ci->i_truncate_pending++; 1483355da1ebSSage Weil queue_trunc = 1; 1484355da1ebSSage Weil } 1485355da1ebSSage Weil inode->i_blocks = 1486355da1ebSSage Weil (attr->ia_size + (1 << 9) - 1) >> 9; 1487355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1488355da1ebSSage Weil ci->i_reported_size = attr->ia_size; 1489355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1490355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1491355da1ebSSage Weil attr->ia_size != inode->i_size) { 1492355da1ebSSage Weil req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1493355da1ebSSage Weil req->r_args.setattr.old_size = 1494355da1ebSSage Weil cpu_to_le64(inode->i_size); 1495355da1ebSSage Weil mask |= CEPH_SETATTR_SIZE; 1496355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1497355da1ebSSage Weil CEPH_CAP_FILE_WR; 1498355da1ebSSage Weil } 1499355da1ebSSage Weil } 1500355da1ebSSage Weil 1501355da1ebSSage Weil /* these do nothing */ 1502355da1ebSSage Weil if (ia_valid & ATTR_CTIME) { 1503355da1ebSSage Weil bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1504355da1ebSSage Weil ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1505355da1ebSSage Weil dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1506355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1507355da1ebSSage Weil attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1508355da1ebSSage Weil only ? "ctime only" : "ignored"); 1509355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1510355da1ebSSage Weil if (only) { 1511355da1ebSSage Weil /* 1512355da1ebSSage Weil * if kernel wants to dirty ctime but nothing else, 1513355da1ebSSage Weil * we need to choose a cap to dirty under, or do 1514355da1ebSSage Weil * a almost-no-op setattr 1515355da1ebSSage Weil */ 1516355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) 1517355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1518355da1ebSSage Weil else if (issued & CEPH_CAP_FILE_EXCL) 1519355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1520355da1ebSSage Weil else if (issued & CEPH_CAP_XATTR_EXCL) 1521355da1ebSSage Weil dirtied |= CEPH_CAP_XATTR_EXCL; 1522355da1ebSSage Weil else 1523355da1ebSSage Weil mask |= CEPH_SETATTR_CTIME; 1524355da1ebSSage Weil } 1525355da1ebSSage Weil } 1526355da1ebSSage Weil if (ia_valid & ATTR_FILE) 1527355da1ebSSage Weil dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1528355da1ebSSage Weil 1529355da1ebSSage Weil if (dirtied) { 1530355da1ebSSage Weil __ceph_mark_dirty_caps(ci, dirtied); 1531355da1ebSSage Weil inode->i_ctime = CURRENT_TIME; 1532355da1ebSSage Weil } 1533355da1ebSSage Weil 1534355da1ebSSage Weil release &= issued; 1535355da1ebSSage Weil spin_unlock(&inode->i_lock); 1536355da1ebSSage Weil 1537355da1ebSSage Weil if (queue_trunc) 1538355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1539355da1ebSSage Weil 1540355da1ebSSage Weil if (mask) { 1541355da1ebSSage Weil req->r_inode = igrab(inode); 1542355da1ebSSage Weil req->r_inode_drop = release; 1543355da1ebSSage Weil req->r_args.setattr.mask = cpu_to_le32(mask); 1544355da1ebSSage Weil req->r_num_caps = 1; 1545355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1546355da1ebSSage Weil } 1547355da1ebSSage Weil dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1548355da1ebSSage Weil ceph_cap_string(dirtied), mask); 1549355da1ebSSage Weil 1550355da1ebSSage Weil ceph_mdsc_put_request(req); 1551355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1552355da1ebSSage Weil return err; 1553355da1ebSSage Weil out: 1554355da1ebSSage Weil spin_unlock(&inode->i_lock); 1555355da1ebSSage Weil ceph_mdsc_put_request(req); 1556355da1ebSSage Weil return err; 1557355da1ebSSage Weil } 1558355da1ebSSage Weil 1559355da1ebSSage Weil /* 1560355da1ebSSage Weil * Verify that we have a lease on the given mask. If not, 1561355da1ebSSage Weil * do a getattr against an mds. 1562355da1ebSSage Weil */ 1563355da1ebSSage Weil int ceph_do_getattr(struct inode *inode, int mask) 1564355da1ebSSage Weil { 1565355da1ebSSage Weil struct ceph_client *client = ceph_sb_to_client(inode->i_sb); 1566355da1ebSSage Weil struct ceph_mds_client *mdsc = &client->mdsc; 1567355da1ebSSage Weil struct ceph_mds_request *req; 1568355da1ebSSage Weil int err; 1569355da1ebSSage Weil 1570355da1ebSSage Weil if (ceph_snap(inode) == CEPH_SNAPDIR) { 1571355da1ebSSage Weil dout("do_getattr inode %p SNAPDIR\n", inode); 1572355da1ebSSage Weil return 0; 1573355da1ebSSage Weil } 1574355da1ebSSage Weil 1575355da1ebSSage Weil dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask)); 1576355da1ebSSage Weil if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1577355da1ebSSage Weil return 0; 1578355da1ebSSage Weil 1579355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1580355da1ebSSage Weil if (IS_ERR(req)) 1581355da1ebSSage Weil return PTR_ERR(req); 1582355da1ebSSage Weil req->r_inode = igrab(inode); 1583355da1ebSSage Weil req->r_num_caps = 1; 1584355da1ebSSage Weil req->r_args.getattr.mask = cpu_to_le32(mask); 1585355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1586355da1ebSSage Weil ceph_mdsc_put_request(req); 1587355da1ebSSage Weil dout("do_getattr result=%d\n", err); 1588355da1ebSSage Weil return err; 1589355da1ebSSage Weil } 1590355da1ebSSage Weil 1591355da1ebSSage Weil 1592355da1ebSSage Weil /* 1593355da1ebSSage Weil * Check inode permissions. We verify we have a valid value for 1594355da1ebSSage Weil * the AUTH cap, then call the generic handler. 1595355da1ebSSage Weil */ 1596355da1ebSSage Weil int ceph_permission(struct inode *inode, int mask) 1597355da1ebSSage Weil { 1598355da1ebSSage Weil int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1599355da1ebSSage Weil 1600355da1ebSSage Weil if (!err) 1601355da1ebSSage Weil err = generic_permission(inode, mask, NULL); 1602355da1ebSSage Weil return err; 1603355da1ebSSage Weil } 1604355da1ebSSage Weil 1605355da1ebSSage Weil /* 1606355da1ebSSage Weil * Get all attributes. Hopefully somedata we'll have a statlite() 1607355da1ebSSage Weil * and can limit the fields we require to be accurate. 1608355da1ebSSage Weil */ 1609355da1ebSSage Weil int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1610355da1ebSSage Weil struct kstat *stat) 1611355da1ebSSage Weil { 1612355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1613232d4b01SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1614355da1ebSSage Weil int err; 1615355da1ebSSage Weil 1616355da1ebSSage Weil err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1617355da1ebSSage Weil if (!err) { 1618355da1ebSSage Weil generic_fillattr(inode, stat); 1619355da1ebSSage Weil stat->ino = inode->i_ino; 1620355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1621355da1ebSSage Weil stat->dev = ceph_snap(inode); 1622355da1ebSSage Weil else 1623355da1ebSSage Weil stat->dev = 0; 1624232d4b01SSage Weil if (S_ISDIR(inode->i_mode)) { 1625232d4b01SSage Weil stat->size = ci->i_rbytes; 1626232d4b01SSage Weil stat->blocks = 0; 1627355da1ebSSage Weil stat->blksize = 65536; 1628355da1ebSSage Weil } 1629232d4b01SSage Weil } 1630355da1ebSSage Weil return err; 1631355da1ebSSage Weil } 1632