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> 13c9af9fb6SYehuda Sadeh #include <linux/pagevec.h> 14355da1ebSSage Weil 15355da1ebSSage Weil #include "super.h" 16355da1ebSSage Weil #include "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)) 72355da1ebSSage Weil return ERR_PTR(PTR_ERR(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 300355da1ebSSage Weil ci->i_fragtree = RB_ROOT; 301355da1ebSSage Weil mutex_init(&ci->i_fragtree_mutex); 302355da1ebSSage Weil 303355da1ebSSage Weil ci->i_xattrs.blob = NULL; 304355da1ebSSage Weil ci->i_xattrs.prealloc_blob = NULL; 305355da1ebSSage Weil ci->i_xattrs.dirty = false; 306355da1ebSSage Weil ci->i_xattrs.index = RB_ROOT; 307355da1ebSSage Weil ci->i_xattrs.count = 0; 308355da1ebSSage Weil ci->i_xattrs.names_size = 0; 309355da1ebSSage Weil ci->i_xattrs.vals_size = 0; 310355da1ebSSage Weil ci->i_xattrs.version = 0; 311355da1ebSSage Weil ci->i_xattrs.index_version = 0; 312355da1ebSSage Weil 313355da1ebSSage Weil ci->i_caps = RB_ROOT; 314355da1ebSSage Weil ci->i_auth_cap = NULL; 315355da1ebSSage Weil ci->i_dirty_caps = 0; 316355da1ebSSage Weil ci->i_flushing_caps = 0; 317355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_dirty_item); 318355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_flushing_item); 319355da1ebSSage Weil ci->i_cap_flush_seq = 0; 320355da1ebSSage Weil ci->i_cap_flush_last_tid = 0; 321355da1ebSSage Weil memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 322355da1ebSSage Weil init_waitqueue_head(&ci->i_cap_wq); 323355da1ebSSage Weil ci->i_hold_caps_min = 0; 324355da1ebSSage Weil ci->i_hold_caps_max = 0; 325355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_delay_list); 326355da1ebSSage Weil ci->i_cap_exporting_mds = 0; 327355da1ebSSage Weil ci->i_cap_exporting_mseq = 0; 328355da1ebSSage Weil ci->i_cap_exporting_issued = 0; 329355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_cap_snaps); 330355da1ebSSage Weil ci->i_head_snapc = NULL; 331355da1ebSSage Weil ci->i_snap_caps = 0; 332355da1ebSSage Weil 333355da1ebSSage Weil for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 334355da1ebSSage Weil ci->i_nr_by_mode[i] = 0; 335355da1ebSSage Weil 336355da1ebSSage Weil ci->i_truncate_seq = 0; 337355da1ebSSage Weil ci->i_truncate_size = 0; 338355da1ebSSage Weil ci->i_truncate_pending = 0; 339355da1ebSSage Weil 340355da1ebSSage Weil ci->i_max_size = 0; 341355da1ebSSage Weil ci->i_reported_size = 0; 342355da1ebSSage Weil ci->i_wanted_max_size = 0; 343355da1ebSSage Weil ci->i_requested_max_size = 0; 344355da1ebSSage Weil 345355da1ebSSage Weil ci->i_pin_ref = 0; 346355da1ebSSage Weil ci->i_rd_ref = 0; 347355da1ebSSage Weil ci->i_rdcache_ref = 0; 348355da1ebSSage Weil ci->i_wr_ref = 0; 349355da1ebSSage Weil ci->i_wrbuffer_ref = 0; 350355da1ebSSage Weil ci->i_wrbuffer_ref_head = 0; 351355da1ebSSage Weil ci->i_shared_gen = 0; 352355da1ebSSage Weil ci->i_rdcache_gen = 0; 353355da1ebSSage Weil ci->i_rdcache_revoking = 0; 354355da1ebSSage Weil 355355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_writes); 356355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_unsafe_dirops); 357355da1ebSSage Weil spin_lock_init(&ci->i_unsafe_lock); 358355da1ebSSage Weil 359355da1ebSSage Weil ci->i_snap_realm = NULL; 360355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_realm_item); 361355da1ebSSage Weil INIT_LIST_HEAD(&ci->i_snap_flush_item); 362355da1ebSSage Weil 3633c6f6b79SSage Weil INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 3643c6f6b79SSage Weil INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 365355da1ebSSage Weil 366355da1ebSSage Weil INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 367355da1ebSSage Weil 368355da1ebSSage Weil return &ci->vfs_inode; 369355da1ebSSage Weil } 370355da1ebSSage Weil 371355da1ebSSage Weil void ceph_destroy_inode(struct inode *inode) 372355da1ebSSage Weil { 373355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 374355da1ebSSage Weil struct ceph_inode_frag *frag; 375355da1ebSSage Weil struct rb_node *n; 376355da1ebSSage Weil 377355da1ebSSage Weil dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 378355da1ebSSage Weil 379355da1ebSSage Weil ceph_queue_caps_release(inode); 380355da1ebSSage Weil 3818b218b8aSSage Weil /* 3828b218b8aSSage Weil * we may still have a snap_realm reference if there are stray 3838b218b8aSSage Weil * caps in i_cap_exporting_issued or i_snap_caps. 3848b218b8aSSage Weil */ 3858b218b8aSSage Weil if (ci->i_snap_realm) { 3868b218b8aSSage Weil struct ceph_mds_client *mdsc = 3878b218b8aSSage Weil &ceph_client(ci->vfs_inode.i_sb)->mdsc; 3888b218b8aSSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 3898b218b8aSSage Weil 3908b218b8aSSage Weil dout(" dropping residual ref to snap realm %p\n", realm); 3918b218b8aSSage Weil spin_lock(&realm->inodes_with_caps_lock); 3928b218b8aSSage Weil list_del_init(&ci->i_snap_realm_item); 3938b218b8aSSage Weil spin_unlock(&realm->inodes_with_caps_lock); 3948b218b8aSSage Weil ceph_put_snap_realm(mdsc, realm); 3958b218b8aSSage Weil } 3968b218b8aSSage Weil 397355da1ebSSage Weil kfree(ci->i_symlink); 398355da1ebSSage Weil while ((n = rb_first(&ci->i_fragtree)) != NULL) { 399355da1ebSSage Weil frag = rb_entry(n, struct ceph_inode_frag, node); 400355da1ebSSage Weil rb_erase(n, &ci->i_fragtree); 401355da1ebSSage Weil kfree(frag); 402355da1ebSSage Weil } 403355da1ebSSage Weil 404355da1ebSSage Weil __ceph_destroy_xattrs(ci); 405b6c1d5b8SSage Weil if (ci->i_xattrs.blob) 406355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 407b6c1d5b8SSage Weil if (ci->i_xattrs.prealloc_blob) 408355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.prealloc_blob); 409355da1ebSSage Weil 410355da1ebSSage Weil kmem_cache_free(ceph_inode_cachep, ci); 411355da1ebSSage Weil } 412355da1ebSSage Weil 413355da1ebSSage Weil 414355da1ebSSage Weil /* 415355da1ebSSage Weil * Helpers to fill in size, ctime, mtime, and atime. We have to be 416355da1ebSSage Weil * careful because either the client or MDS may have more up to date 417355da1ebSSage Weil * info, depending on which capabilities are held, and whether 418355da1ebSSage Weil * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 419355da1ebSSage Weil * and size are monotonically increasing, except when utimes() or 420355da1ebSSage Weil * truncate() increments the corresponding _seq values.) 421355da1ebSSage Weil */ 422355da1ebSSage Weil int ceph_fill_file_size(struct inode *inode, int issued, 423355da1ebSSage Weil u32 truncate_seq, u64 truncate_size, u64 size) 424355da1ebSSage Weil { 425355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 426355da1ebSSage Weil int queue_trunc = 0; 427355da1ebSSage Weil 428355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 429355da1ebSSage Weil (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 430355da1ebSSage Weil dout("size %lld -> %llu\n", inode->i_size, size); 431355da1ebSSage Weil inode->i_size = size; 432355da1ebSSage Weil inode->i_blocks = (size + (1<<9) - 1) >> 9; 433355da1ebSSage Weil ci->i_reported_size = size; 434355da1ebSSage Weil if (truncate_seq != ci->i_truncate_seq) { 435355da1ebSSage Weil dout("truncate_seq %u -> %u\n", 436355da1ebSSage Weil ci->i_truncate_seq, truncate_seq); 437355da1ebSSage Weil ci->i_truncate_seq = truncate_seq; 4383d497d85SYehuda Sadeh /* 4393d497d85SYehuda Sadeh * If we hold relevant caps, or in the case where we're 4403d497d85SYehuda Sadeh * not the only client referencing this file and we 4413d497d85SYehuda Sadeh * don't hold those caps, then we need to check whether 4423d497d85SYehuda Sadeh * the file is either opened or mmaped 4433d497d85SYehuda Sadeh */ 4443d497d85SYehuda Sadeh if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 445355da1ebSSage Weil CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 4463d497d85SYehuda Sadeh CEPH_CAP_FILE_EXCL)) || 4473d497d85SYehuda Sadeh mapping_mapped(inode->i_mapping) || 4483d497d85SYehuda Sadeh __ceph_caps_file_wanted(ci)) { 449355da1ebSSage Weil ci->i_truncate_pending++; 450355da1ebSSage Weil queue_trunc = 1; 451355da1ebSSage Weil } 452355da1ebSSage Weil } 453355da1ebSSage Weil } 454355da1ebSSage Weil if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 455355da1ebSSage Weil ci->i_truncate_size != truncate_size) { 456355da1ebSSage Weil dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 457355da1ebSSage Weil truncate_size); 458355da1ebSSage Weil ci->i_truncate_size = truncate_size; 459355da1ebSSage Weil } 460355da1ebSSage Weil return queue_trunc; 461355da1ebSSage Weil } 462355da1ebSSage Weil 463355da1ebSSage Weil void ceph_fill_file_time(struct inode *inode, int issued, 464355da1ebSSage Weil u64 time_warp_seq, struct timespec *ctime, 465355da1ebSSage Weil struct timespec *mtime, struct timespec *atime) 466355da1ebSSage Weil { 467355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 468355da1ebSSage Weil int warn = 0; 469355da1ebSSage Weil 470355da1ebSSage Weil if (issued & (CEPH_CAP_FILE_EXCL| 471355da1ebSSage Weil CEPH_CAP_FILE_WR| 472355da1ebSSage Weil CEPH_CAP_FILE_BUFFER)) { 473355da1ebSSage Weil if (timespec_compare(ctime, &inode->i_ctime) > 0) { 474355da1ebSSage Weil dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 475355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 476355da1ebSSage Weil ctime->tv_sec, ctime->tv_nsec); 477355da1ebSSage Weil inode->i_ctime = *ctime; 478355da1ebSSage Weil } 479355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 480355da1ebSSage Weil /* the MDS did a utimes() */ 481355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld " 482355da1ebSSage Weil "tw %d -> %d\n", 483355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 484355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec, 485355da1ebSSage Weil ci->i_time_warp_seq, (int)time_warp_seq); 486355da1ebSSage Weil 487355da1ebSSage Weil inode->i_mtime = *mtime; 488355da1ebSSage Weil inode->i_atime = *atime; 489355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 490355da1ebSSage Weil } else if (time_warp_seq == ci->i_time_warp_seq) { 491355da1ebSSage Weil /* nobody did utimes(); take the max */ 492355da1ebSSage Weil if (timespec_compare(mtime, &inode->i_mtime) > 0) { 493355da1ebSSage Weil dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 494355da1ebSSage Weil inode->i_mtime.tv_sec, 495355da1ebSSage Weil inode->i_mtime.tv_nsec, 496355da1ebSSage Weil mtime->tv_sec, mtime->tv_nsec); 497355da1ebSSage Weil inode->i_mtime = *mtime; 498355da1ebSSage Weil } 499355da1ebSSage Weil if (timespec_compare(atime, &inode->i_atime) > 0) { 500355da1ebSSage Weil dout("atime %ld.%09ld -> %ld.%09ld inc\n", 501355da1ebSSage Weil inode->i_atime.tv_sec, 502355da1ebSSage Weil inode->i_atime.tv_nsec, 503355da1ebSSage Weil atime->tv_sec, atime->tv_nsec); 504355da1ebSSage Weil inode->i_atime = *atime; 505355da1ebSSage Weil } 506355da1ebSSage Weil } else if (issued & CEPH_CAP_FILE_EXCL) { 507355da1ebSSage Weil /* we did a utimes(); ignore mds values */ 508355da1ebSSage Weil } else { 509355da1ebSSage Weil warn = 1; 510355da1ebSSage Weil } 511355da1ebSSage Weil } else { 512355da1ebSSage Weil /* we have no write caps; whatever the MDS says is true */ 513355da1ebSSage Weil if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 514355da1ebSSage Weil inode->i_ctime = *ctime; 515355da1ebSSage Weil inode->i_mtime = *mtime; 516355da1ebSSage Weil inode->i_atime = *atime; 517355da1ebSSage Weil ci->i_time_warp_seq = time_warp_seq; 518355da1ebSSage Weil } else { 519355da1ebSSage Weil warn = 1; 520355da1ebSSage Weil } 521355da1ebSSage Weil } 522355da1ebSSage Weil if (warn) /* time_warp_seq shouldn't go backwards */ 523355da1ebSSage Weil dout("%p mds time_warp_seq %llu < %u\n", 524355da1ebSSage Weil inode, time_warp_seq, ci->i_time_warp_seq); 525355da1ebSSage Weil } 526355da1ebSSage Weil 527355da1ebSSage Weil /* 528355da1ebSSage Weil * Populate an inode based on info from mds. May be called on new or 529355da1ebSSage Weil * existing inodes. 530355da1ebSSage Weil */ 531355da1ebSSage Weil static int fill_inode(struct inode *inode, 532355da1ebSSage Weil struct ceph_mds_reply_info_in *iinfo, 533355da1ebSSage Weil struct ceph_mds_reply_dirfrag *dirinfo, 534355da1ebSSage Weil struct ceph_mds_session *session, 535355da1ebSSage Weil unsigned long ttl_from, int cap_fmode, 536355da1ebSSage Weil struct ceph_cap_reservation *caps_reservation) 537355da1ebSSage Weil { 538355da1ebSSage Weil struct ceph_mds_reply_inode *info = iinfo->in; 539355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 540355da1ebSSage Weil int i; 541355da1ebSSage Weil int issued, implemented; 542355da1ebSSage Weil struct timespec mtime, atime, ctime; 543355da1ebSSage Weil u32 nsplits; 544355da1ebSSage Weil struct ceph_buffer *xattr_blob = NULL; 545355da1ebSSage Weil int err = 0; 546355da1ebSSage Weil int queue_trunc = 0; 547355da1ebSSage Weil 548355da1ebSSage Weil dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 549355da1ebSSage Weil inode, ceph_vinop(inode), le64_to_cpu(info->version), 550355da1ebSSage Weil ci->i_version); 551355da1ebSSage Weil 552355da1ebSSage Weil /* 553355da1ebSSage Weil * prealloc xattr data, if it looks like we'll need it. only 554355da1ebSSage Weil * if len > 4 (meaning there are actually xattrs; the first 4 555355da1ebSSage Weil * bytes are the xattr count). 556355da1ebSSage Weil */ 557355da1ebSSage Weil if (iinfo->xattr_len > 4) { 558b6c1d5b8SSage Weil xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 559355da1ebSSage Weil if (!xattr_blob) 560355da1ebSSage Weil pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 561355da1ebSSage Weil iinfo->xattr_len); 562355da1ebSSage Weil } 563355da1ebSSage Weil 564355da1ebSSage Weil spin_lock(&inode->i_lock); 565355da1ebSSage Weil 566355da1ebSSage Weil /* 567355da1ebSSage Weil * provided version will be odd if inode value is projected, 568355da1ebSSage Weil * even if stable. skip the update if we have a newer info 569355da1ebSSage Weil * (e.g., due to inode info racing form multiple MDSs), or if 570355da1ebSSage Weil * we are getting projected (unstable) inode info. 571355da1ebSSage Weil */ 572355da1ebSSage Weil if (le64_to_cpu(info->version) > 0 && 573355da1ebSSage Weil (ci->i_version & ~1) > le64_to_cpu(info->version)) 574355da1ebSSage Weil goto no_change; 575355da1ebSSage Weil 576355da1ebSSage Weil issued = __ceph_caps_issued(ci, &implemented); 577355da1ebSSage Weil issued |= implemented | __ceph_caps_dirty(ci); 578355da1ebSSage Weil 579355da1ebSSage Weil /* update inode */ 580355da1ebSSage Weil ci->i_version = le64_to_cpu(info->version); 581355da1ebSSage Weil inode->i_version++; 582355da1ebSSage Weil inode->i_rdev = le32_to_cpu(info->rdev); 583355da1ebSSage Weil 584355da1ebSSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 585355da1ebSSage Weil inode->i_mode = le32_to_cpu(info->mode); 586355da1ebSSage Weil inode->i_uid = le32_to_cpu(info->uid); 587355da1ebSSage Weil inode->i_gid = le32_to_cpu(info->gid); 588355da1ebSSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 589355da1ebSSage Weil inode->i_uid, inode->i_gid); 590355da1ebSSage Weil } 591355da1ebSSage Weil 592355da1ebSSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 593355da1ebSSage Weil inode->i_nlink = le32_to_cpu(info->nlink); 594355da1ebSSage Weil 595355da1ebSSage Weil /* be careful with mtime, atime, size */ 596355da1ebSSage Weil ceph_decode_timespec(&atime, &info->atime); 597355da1ebSSage Weil ceph_decode_timespec(&mtime, &info->mtime); 598355da1ebSSage Weil ceph_decode_timespec(&ctime, &info->ctime); 599355da1ebSSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 600355da1ebSSage Weil le32_to_cpu(info->truncate_seq), 601355da1ebSSage Weil le64_to_cpu(info->truncate_size), 602355da1ebSSage Weil le64_to_cpu(info->size)); 603355da1ebSSage Weil ceph_fill_file_time(inode, issued, 604355da1ebSSage Weil le32_to_cpu(info->time_warp_seq), 605355da1ebSSage Weil &ctime, &mtime, &atime); 606355da1ebSSage Weil 607355da1ebSSage Weil ci->i_max_size = le64_to_cpu(info->max_size); 608355da1ebSSage Weil ci->i_layout = info->layout; 609355da1ebSSage Weil inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 610355da1ebSSage Weil 611355da1ebSSage Weil /* xattrs */ 612355da1ebSSage Weil /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 613355da1ebSSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 614355da1ebSSage Weil le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 615355da1ebSSage Weil if (ci->i_xattrs.blob) 616355da1ebSSage Weil ceph_buffer_put(ci->i_xattrs.blob); 617355da1ebSSage Weil ci->i_xattrs.blob = xattr_blob; 618355da1ebSSage Weil if (xattr_blob) 619355da1ebSSage Weil memcpy(ci->i_xattrs.blob->vec.iov_base, 620355da1ebSSage Weil iinfo->xattr_data, iinfo->xattr_len); 621355da1ebSSage Weil ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 622355da1ebSSage Weil } 623355da1ebSSage Weil 624355da1ebSSage Weil inode->i_mapping->a_ops = &ceph_aops; 625355da1ebSSage Weil inode->i_mapping->backing_dev_info = 626355da1ebSSage Weil &ceph_client(inode->i_sb)->backing_dev_info; 627355da1ebSSage Weil 628355da1ebSSage Weil switch (inode->i_mode & S_IFMT) { 629355da1ebSSage Weil case S_IFIFO: 630355da1ebSSage Weil case S_IFBLK: 631355da1ebSSage Weil case S_IFCHR: 632355da1ebSSage Weil case S_IFSOCK: 633355da1ebSSage Weil init_special_inode(inode, inode->i_mode, inode->i_rdev); 634355da1ebSSage Weil inode->i_op = &ceph_file_iops; 635355da1ebSSage Weil break; 636355da1ebSSage Weil case S_IFREG: 637355da1ebSSage Weil inode->i_op = &ceph_file_iops; 638355da1ebSSage Weil inode->i_fop = &ceph_file_fops; 639355da1ebSSage Weil break; 640355da1ebSSage Weil case S_IFLNK: 641355da1ebSSage Weil inode->i_op = &ceph_symlink_iops; 642355da1ebSSage Weil if (!ci->i_symlink) { 643355da1ebSSage Weil int symlen = iinfo->symlink_len; 644355da1ebSSage Weil char *sym; 645355da1ebSSage Weil 646355da1ebSSage Weil BUG_ON(symlen != inode->i_size); 647355da1ebSSage Weil spin_unlock(&inode->i_lock); 648355da1ebSSage Weil 649355da1ebSSage Weil err = -ENOMEM; 650355da1ebSSage Weil sym = kmalloc(symlen+1, GFP_NOFS); 651355da1ebSSage Weil if (!sym) 652355da1ebSSage Weil goto out; 653355da1ebSSage Weil memcpy(sym, iinfo->symlink, symlen); 654355da1ebSSage Weil sym[symlen] = 0; 655355da1ebSSage Weil 656355da1ebSSage Weil spin_lock(&inode->i_lock); 657355da1ebSSage Weil if (!ci->i_symlink) 658355da1ebSSage Weil ci->i_symlink = sym; 659355da1ebSSage Weil else 660355da1ebSSage Weil kfree(sym); /* lost a race */ 661355da1ebSSage Weil } 662355da1ebSSage Weil break; 663355da1ebSSage Weil case S_IFDIR: 664355da1ebSSage Weil inode->i_op = &ceph_dir_iops; 665355da1ebSSage Weil inode->i_fop = &ceph_dir_fops; 666355da1ebSSage Weil 667355da1ebSSage Weil ci->i_files = le64_to_cpu(info->files); 668355da1ebSSage Weil ci->i_subdirs = le64_to_cpu(info->subdirs); 669355da1ebSSage Weil ci->i_rbytes = le64_to_cpu(info->rbytes); 670355da1ebSSage Weil ci->i_rfiles = le64_to_cpu(info->rfiles); 671355da1ebSSage Weil ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 672355da1ebSSage Weil ceph_decode_timespec(&ci->i_rctime, &info->rctime); 673355da1ebSSage Weil 674355da1ebSSage Weil /* set dir completion flag? */ 675355da1ebSSage Weil if (ci->i_files == 0 && ci->i_subdirs == 0 && 676355da1ebSSage Weil ceph_snap(inode) == CEPH_NOSNAP && 677355da1ebSSage Weil (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) { 678355da1ebSSage Weil dout(" marking %p complete (empty)\n", inode); 679355da1ebSSage Weil ci->i_ceph_flags |= CEPH_I_COMPLETE; 680355da1ebSSage Weil ci->i_max_offset = 2; 681355da1ebSSage Weil } 682355da1ebSSage Weil 683355da1ebSSage Weil /* it may be better to set st_size in getattr instead? */ 684355da1ebSSage Weil if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES)) 685355da1ebSSage Weil inode->i_size = ci->i_rbytes; 686355da1ebSSage Weil break; 687355da1ebSSage Weil default: 688355da1ebSSage Weil pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 689355da1ebSSage Weil ceph_vinop(inode), inode->i_mode); 690355da1ebSSage Weil } 691355da1ebSSage Weil 692355da1ebSSage Weil no_change: 693355da1ebSSage Weil spin_unlock(&inode->i_lock); 694355da1ebSSage Weil 695355da1ebSSage Weil /* queue truncate if we saw i_size decrease */ 696355da1ebSSage Weil if (queue_trunc) 6973c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 698355da1ebSSage Weil 699355da1ebSSage Weil /* populate frag tree */ 700355da1ebSSage Weil /* FIXME: move me up, if/when version reflects fragtree changes */ 701355da1ebSSage Weil nsplits = le32_to_cpu(info->fragtree.nsplits); 702355da1ebSSage Weil mutex_lock(&ci->i_fragtree_mutex); 703355da1ebSSage Weil for (i = 0; i < nsplits; i++) { 704355da1ebSSage Weil u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 705355da1ebSSage Weil struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 706355da1ebSSage Weil 707355da1ebSSage Weil if (IS_ERR(frag)) 708355da1ebSSage Weil continue; 709355da1ebSSage Weil frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 710355da1ebSSage Weil dout(" frag %x split by %d\n", frag->frag, frag->split_by); 711355da1ebSSage Weil } 712355da1ebSSage Weil mutex_unlock(&ci->i_fragtree_mutex); 713355da1ebSSage Weil 714355da1ebSSage Weil /* were we issued a capability? */ 715355da1ebSSage Weil if (info->cap.caps) { 716355da1ebSSage Weil if (ceph_snap(inode) == CEPH_NOSNAP) { 717355da1ebSSage Weil ceph_add_cap(inode, session, 718355da1ebSSage Weil le64_to_cpu(info->cap.cap_id), 719355da1ebSSage Weil cap_fmode, 720355da1ebSSage Weil le32_to_cpu(info->cap.caps), 721355da1ebSSage Weil le32_to_cpu(info->cap.wanted), 722355da1ebSSage Weil le32_to_cpu(info->cap.seq), 723355da1ebSSage Weil le32_to_cpu(info->cap.mseq), 724355da1ebSSage Weil le64_to_cpu(info->cap.realm), 725355da1ebSSage Weil info->cap.flags, 726355da1ebSSage Weil caps_reservation); 727355da1ebSSage Weil } else { 728355da1ebSSage Weil spin_lock(&inode->i_lock); 729355da1ebSSage Weil dout(" %p got snap_caps %s\n", inode, 730355da1ebSSage Weil ceph_cap_string(le32_to_cpu(info->cap.caps))); 731355da1ebSSage Weil ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 732355da1ebSSage Weil if (cap_fmode >= 0) 733355da1ebSSage Weil __ceph_get_fmode(ci, cap_fmode); 734355da1ebSSage Weil spin_unlock(&inode->i_lock); 735355da1ebSSage Weil } 736355da1ebSSage Weil } 737355da1ebSSage Weil 738355da1ebSSage Weil /* update delegation info? */ 739355da1ebSSage Weil if (dirinfo) 740355da1ebSSage Weil ceph_fill_dirfrag(inode, dirinfo); 741355da1ebSSage Weil 742355da1ebSSage Weil err = 0; 743355da1ebSSage Weil 744355da1ebSSage Weil out: 745b6c1d5b8SSage Weil if (xattr_blob) 746355da1ebSSage Weil ceph_buffer_put(xattr_blob); 747355da1ebSSage Weil return err; 748355da1ebSSage Weil } 749355da1ebSSage Weil 750355da1ebSSage Weil /* 751355da1ebSSage Weil * caller should hold session s_mutex. 752355da1ebSSage Weil */ 753355da1ebSSage Weil static void update_dentry_lease(struct dentry *dentry, 754355da1ebSSage Weil struct ceph_mds_reply_lease *lease, 755355da1ebSSage Weil struct ceph_mds_session *session, 756355da1ebSSage Weil unsigned long from_time) 757355da1ebSSage Weil { 758355da1ebSSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 759355da1ebSSage Weil long unsigned duration = le32_to_cpu(lease->duration_ms); 760355da1ebSSage Weil long unsigned ttl = from_time + (duration * HZ) / 1000; 761355da1ebSSage Weil long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 762355da1ebSSage Weil struct inode *dir; 763355da1ebSSage Weil 764355da1ebSSage Weil /* only track leases on regular dentries */ 765355da1ebSSage Weil if (dentry->d_op != &ceph_dentry_ops) 766355da1ebSSage Weil return; 767355da1ebSSage Weil 768355da1ebSSage Weil spin_lock(&dentry->d_lock); 769355da1ebSSage Weil dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 770355da1ebSSage Weil dentry, le16_to_cpu(lease->mask), duration, ttl); 771355da1ebSSage Weil 772355da1ebSSage Weil /* make lease_rdcache_gen match directory */ 773355da1ebSSage Weil dir = dentry->d_parent->d_inode; 774355da1ebSSage Weil di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 775355da1ebSSage Weil 776355da1ebSSage Weil if (lease->mask == 0) 777355da1ebSSage Weil goto out_unlock; 778355da1ebSSage Weil 779355da1ebSSage Weil if (di->lease_gen == session->s_cap_gen && 780355da1ebSSage Weil time_before(ttl, dentry->d_time)) 781355da1ebSSage Weil goto out_unlock; /* we already have a newer lease. */ 782355da1ebSSage Weil 783355da1ebSSage Weil if (di->lease_session && di->lease_session != session) 784355da1ebSSage Weil goto out_unlock; 785355da1ebSSage Weil 786355da1ebSSage Weil ceph_dentry_lru_touch(dentry); 787355da1ebSSage Weil 788355da1ebSSage Weil if (!di->lease_session) 789355da1ebSSage Weil di->lease_session = ceph_get_mds_session(session); 790355da1ebSSage Weil di->lease_gen = session->s_cap_gen; 791355da1ebSSage Weil di->lease_seq = le32_to_cpu(lease->seq); 792355da1ebSSage Weil di->lease_renew_after = half_ttl; 793355da1ebSSage Weil di->lease_renew_from = 0; 794355da1ebSSage Weil dentry->d_time = ttl; 795355da1ebSSage Weil out_unlock: 796355da1ebSSage Weil spin_unlock(&dentry->d_lock); 797355da1ebSSage Weil return; 798355da1ebSSage Weil } 799355da1ebSSage Weil 800355da1ebSSage Weil /* 801355da1ebSSage Weil * splice a dentry to an inode. 802355da1ebSSage Weil * caller must hold directory i_mutex for this to be safe. 803355da1ebSSage Weil * 804355da1ebSSage Weil * we will only rehash the resulting dentry if @prehash is 805355da1ebSSage Weil * true; @prehash will be set to false (for the benefit of 806355da1ebSSage Weil * the caller) if we fail. 807355da1ebSSage Weil */ 808355da1ebSSage Weil static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 809355da1ebSSage Weil bool *prehash) 810355da1ebSSage Weil { 811355da1ebSSage Weil struct dentry *realdn; 812355da1ebSSage Weil 813355da1ebSSage Weil /* dn must be unhashed */ 814355da1ebSSage Weil if (!d_unhashed(dn)) 815355da1ebSSage Weil d_drop(dn); 816355da1ebSSage Weil realdn = d_materialise_unique(dn, in); 817355da1ebSSage Weil if (IS_ERR(realdn)) { 818355da1ebSSage Weil pr_err("splice_dentry error %p inode %p ino %llx.%llx\n", 819355da1ebSSage Weil dn, in, ceph_vinop(in)); 820355da1ebSSage Weil if (prehash) 821355da1ebSSage Weil *prehash = false; /* don't rehash on error */ 822355da1ebSSage Weil dn = realdn; /* note realdn contains the error */ 823355da1ebSSage Weil goto out; 824355da1ebSSage Weil } else if (realdn) { 825355da1ebSSage Weil dout("dn %p (%d) spliced with %p (%d) " 826355da1ebSSage Weil "inode %p ino %llx.%llx\n", 827355da1ebSSage Weil dn, atomic_read(&dn->d_count), 828355da1ebSSage Weil realdn, atomic_read(&realdn->d_count), 829355da1ebSSage Weil realdn->d_inode, ceph_vinop(realdn->d_inode)); 830355da1ebSSage Weil dput(dn); 831355da1ebSSage Weil dn = realdn; 832355da1ebSSage Weil } else { 833355da1ebSSage Weil BUG_ON(!ceph_dentry(dn)); 834355da1ebSSage Weil 835355da1ebSSage Weil dout("dn %p attached to %p ino %llx.%llx\n", 836355da1ebSSage Weil dn, dn->d_inode, ceph_vinop(dn->d_inode)); 837355da1ebSSage Weil } 838355da1ebSSage Weil if ((!prehash || *prehash) && d_unhashed(dn)) 839355da1ebSSage Weil d_rehash(dn); 840355da1ebSSage Weil out: 841355da1ebSSage Weil return dn; 842355da1ebSSage Weil } 843355da1ebSSage Weil 844355da1ebSSage Weil /* 8454baa75efSYehuda Sadeh * Set dentry's directory position based on the current dir's max, and 8464baa75efSYehuda Sadeh * order it in d_subdirs, so that dcache_readdir behaves. 8474baa75efSYehuda Sadeh */ 8484baa75efSYehuda Sadeh static void ceph_set_dentry_offset(struct dentry *dn) 8494baa75efSYehuda Sadeh { 8504baa75efSYehuda Sadeh struct dentry *dir = dn->d_parent; 8514baa75efSYehuda Sadeh struct inode *inode = dn->d_parent->d_inode; 8524baa75efSYehuda Sadeh struct ceph_dentry_info *di; 8534baa75efSYehuda Sadeh 8544baa75efSYehuda Sadeh BUG_ON(!inode); 8554baa75efSYehuda Sadeh 8564baa75efSYehuda Sadeh di = ceph_dentry(dn); 8574baa75efSYehuda Sadeh 8584baa75efSYehuda Sadeh spin_lock(&inode->i_lock); 8594baa75efSYehuda Sadeh di->offset = ceph_inode(inode)->i_max_offset++; 8604baa75efSYehuda Sadeh spin_unlock(&inode->i_lock); 8614baa75efSYehuda Sadeh 8624baa75efSYehuda Sadeh spin_lock(&dcache_lock); 8634baa75efSYehuda Sadeh spin_lock(&dn->d_lock); 8644baa75efSYehuda Sadeh list_move_tail(&dir->d_subdirs, &dn->d_u.d_child); 8654baa75efSYehuda Sadeh dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 8664baa75efSYehuda Sadeh dn->d_u.d_child.prev, dn->d_u.d_child.next); 8674baa75efSYehuda Sadeh spin_unlock(&dn->d_lock); 8684baa75efSYehuda Sadeh spin_unlock(&dcache_lock); 8694baa75efSYehuda Sadeh } 8704baa75efSYehuda Sadeh 8714baa75efSYehuda Sadeh /* 872355da1ebSSage Weil * Incorporate results into the local cache. This is either just 873355da1ebSSage Weil * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 874355da1ebSSage Weil * after a lookup). 875355da1ebSSage Weil * 876355da1ebSSage Weil * A reply may contain 877355da1ebSSage Weil * a directory inode along with a dentry. 878355da1ebSSage Weil * and/or a target inode 879355da1ebSSage Weil * 880355da1ebSSage Weil * Called with snap_rwsem (read). 881355da1ebSSage Weil */ 882355da1ebSSage Weil int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 883355da1ebSSage Weil struct ceph_mds_session *session) 884355da1ebSSage Weil { 885355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 886355da1ebSSage Weil struct inode *in = NULL; 887355da1ebSSage Weil struct ceph_mds_reply_inode *ininfo; 888355da1ebSSage Weil struct ceph_vino vino; 8899358c6d4SSage Weil struct ceph_client *client = ceph_sb_to_client(sb); 890355da1ebSSage Weil int i = 0; 891355da1ebSSage Weil int err = 0; 892355da1ebSSage Weil 893355da1ebSSage Weil dout("fill_trace %p is_dentry %d is_target %d\n", req, 894355da1ebSSage Weil rinfo->head->is_dentry, rinfo->head->is_target); 895355da1ebSSage Weil 896355da1ebSSage Weil #if 0 897355da1ebSSage Weil /* 898355da1ebSSage Weil * Debugging hook: 899355da1ebSSage Weil * 900355da1ebSSage Weil * If we resend completed ops to a recovering mds, we get no 901355da1ebSSage Weil * trace. Since that is very rare, pretend this is the case 902355da1ebSSage Weil * to ensure the 'no trace' handlers in the callers behave. 903355da1ebSSage Weil * 904355da1ebSSage Weil * Fill in inodes unconditionally to avoid breaking cap 905355da1ebSSage Weil * invariants. 906355da1ebSSage Weil */ 907355da1ebSSage Weil if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 908355da1ebSSage Weil pr_info("fill_trace faking empty trace on %lld %s\n", 909355da1ebSSage Weil req->r_tid, ceph_mds_op_name(rinfo->head->op)); 910355da1ebSSage Weil if (rinfo->head->is_dentry) { 911355da1ebSSage Weil rinfo->head->is_dentry = 0; 912355da1ebSSage Weil err = fill_inode(req->r_locked_dir, 913355da1ebSSage Weil &rinfo->diri, rinfo->dirfrag, 914355da1ebSSage Weil session, req->r_request_started, -1); 915355da1ebSSage Weil } 916355da1ebSSage Weil if (rinfo->head->is_target) { 917355da1ebSSage Weil rinfo->head->is_target = 0; 918355da1ebSSage Weil ininfo = rinfo->targeti.in; 919355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 920355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 921355da1ebSSage Weil in = ceph_get_inode(sb, vino); 922355da1ebSSage Weil err = fill_inode(in, &rinfo->targeti, NULL, 923355da1ebSSage Weil session, req->r_request_started, 924355da1ebSSage Weil req->r_fmode); 925355da1ebSSage Weil iput(in); 926355da1ebSSage Weil } 927355da1ebSSage Weil } 928355da1ebSSage Weil #endif 929355da1ebSSage Weil 930355da1ebSSage Weil if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 931355da1ebSSage Weil dout("fill_trace reply is empty!\n"); 932355da1ebSSage Weil if (rinfo->head->result == 0 && req->r_locked_dir) { 933355da1ebSSage Weil struct ceph_inode_info *ci = 934355da1ebSSage Weil ceph_inode(req->r_locked_dir); 935355da1ebSSage Weil dout(" clearing %p complete (empty trace)\n", 936355da1ebSSage Weil req->r_locked_dir); 937355da1ebSSage Weil ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 938355da1ebSSage Weil ci->i_release_count++; 939355da1ebSSage Weil } 940355da1ebSSage Weil return 0; 941355da1ebSSage Weil } 942355da1ebSSage Weil 943355da1ebSSage Weil if (rinfo->head->is_dentry) { 9445b1daecdSSage Weil struct inode *dir = req->r_locked_dir; 9455b1daecdSSage Weil 9465b1daecdSSage Weil err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 9475b1daecdSSage Weil session, req->r_request_started, -1, 9485b1daecdSSage Weil &req->r_caps_reservation); 9495b1daecdSSage Weil if (err < 0) 9505b1daecdSSage Weil return err; 9515b1daecdSSage Weil } 9525b1daecdSSage Weil 9539358c6d4SSage Weil /* 9549358c6d4SSage Weil * ignore null lease/binding on snapdir ENOENT, or else we 9559358c6d4SSage Weil * will have trouble splicing in the virtual snapdir later 9569358c6d4SSage Weil */ 9579358c6d4SSage Weil if (rinfo->head->is_dentry && !req->r_aborted && 9589358c6d4SSage Weil (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 9599358c6d4SSage Weil client->mount_args->snapdir_name, 9609358c6d4SSage Weil req->r_dentry->d_name.len))) { 961355da1ebSSage Weil /* 962355da1ebSSage Weil * lookup link rename : null -> possibly existing inode 963355da1ebSSage Weil * mknod symlink mkdir : null -> new inode 964355da1ebSSage Weil * unlink : linked -> null 965355da1ebSSage Weil */ 966355da1ebSSage Weil struct inode *dir = req->r_locked_dir; 967355da1ebSSage Weil struct dentry *dn = req->r_dentry; 968355da1ebSSage Weil bool have_dir_cap, have_lease; 969355da1ebSSage Weil 970355da1ebSSage Weil BUG_ON(!dn); 971355da1ebSSage Weil BUG_ON(!dir); 972355da1ebSSage Weil BUG_ON(dn->d_parent->d_inode != dir); 973355da1ebSSage Weil BUG_ON(ceph_ino(dir) != 974355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->ino)); 975355da1ebSSage Weil BUG_ON(ceph_snap(dir) != 976355da1ebSSage Weil le64_to_cpu(rinfo->diri.in->snapid)); 977355da1ebSSage Weil 978355da1ebSSage Weil /* do we have a lease on the whole dir? */ 979355da1ebSSage Weil have_dir_cap = 980355da1ebSSage Weil (le32_to_cpu(rinfo->diri.in->cap.caps) & 981355da1ebSSage Weil CEPH_CAP_FILE_SHARED); 982355da1ebSSage Weil 983355da1ebSSage Weil /* do we have a dn lease? */ 984355da1ebSSage Weil have_lease = have_dir_cap || 985355da1ebSSage Weil (le16_to_cpu(rinfo->dlease->mask) & 986355da1ebSSage Weil CEPH_LOCK_DN); 987355da1ebSSage Weil 988355da1ebSSage Weil if (!have_lease) 989355da1ebSSage Weil dout("fill_trace no dentry lease or dir cap\n"); 990355da1ebSSage Weil 991355da1ebSSage Weil /* rename? */ 992355da1ebSSage Weil if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 993355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 994355da1ebSSage Weil req->r_old_dentry, 995355da1ebSSage Weil req->r_old_dentry->d_name.len, 996355da1ebSSage Weil req->r_old_dentry->d_name.name, 997355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 998355da1ebSSage Weil dout("fill_trace doing d_move %p -> %p\n", 999355da1ebSSage Weil req->r_old_dentry, dn); 1000355da1ebSSage Weil d_move(req->r_old_dentry, dn); 1001355da1ebSSage Weil dout(" src %p '%.*s' dst %p '%.*s'\n", 1002355da1ebSSage Weil req->r_old_dentry, 1003355da1ebSSage Weil req->r_old_dentry->d_name.len, 1004355da1ebSSage Weil req->r_old_dentry->d_name.name, 1005355da1ebSSage Weil dn, dn->d_name.len, dn->d_name.name); 1006c4a29f26SSage Weil /* ensure target dentry is invalidated, despite 1007c4a29f26SSage Weil rehashing bug in vfs_rename_dir */ 1008c4a29f26SSage Weil dn->d_time = jiffies; 1009c4a29f26SSage Weil ceph_dentry(dn)->lease_shared_gen = 0; 1010355da1ebSSage Weil /* take overwritten dentry's readdir offset */ 1011355da1ebSSage Weil ceph_dentry(req->r_old_dentry)->offset = 1012355da1ebSSage Weil ceph_dentry(dn)->offset; 1013355da1ebSSage Weil dn = req->r_old_dentry; /* use old_dentry */ 1014355da1ebSSage Weil in = dn->d_inode; 1015355da1ebSSage Weil } 1016355da1ebSSage Weil 1017355da1ebSSage Weil /* null dentry? */ 1018355da1ebSSage Weil if (!rinfo->head->is_target) { 1019355da1ebSSage Weil dout("fill_trace null dentry\n"); 1020355da1ebSSage Weil if (dn->d_inode) { 1021355da1ebSSage Weil dout("d_delete %p\n", dn); 1022355da1ebSSage Weil d_delete(dn); 1023355da1ebSSage Weil } else { 1024355da1ebSSage Weil dout("d_instantiate %p NULL\n", dn); 1025355da1ebSSage Weil d_instantiate(dn, NULL); 1026355da1ebSSage Weil if (have_lease && d_unhashed(dn)) 1027355da1ebSSage Weil d_rehash(dn); 1028355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, 1029355da1ebSSage Weil session, 1030355da1ebSSage Weil req->r_request_started); 1031355da1ebSSage Weil } 1032355da1ebSSage Weil goto done; 1033355da1ebSSage Weil } 1034355da1ebSSage Weil 1035355da1ebSSage Weil /* attach proper inode */ 1036355da1ebSSage Weil ininfo = rinfo->targeti.in; 1037355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1038355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1039355da1ebSSage Weil if (!dn->d_inode) { 1040355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1041355da1ebSSage Weil if (IS_ERR(in)) { 1042355da1ebSSage Weil pr_err("fill_trace bad get_inode " 1043355da1ebSSage Weil "%llx.%llx\n", vino.ino, vino.snap); 1044355da1ebSSage Weil err = PTR_ERR(in); 1045355da1ebSSage Weil d_delete(dn); 1046355da1ebSSage Weil goto done; 1047355da1ebSSage Weil } 1048355da1ebSSage Weil dn = splice_dentry(dn, in, &have_lease); 1049355da1ebSSage Weil if (IS_ERR(dn)) { 1050355da1ebSSage Weil err = PTR_ERR(dn); 1051355da1ebSSage Weil goto done; 1052355da1ebSSage Weil } 1053355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 10544baa75efSYehuda Sadeh ceph_set_dentry_offset(dn); 1055355da1ebSSage Weil igrab(in); 1056355da1ebSSage Weil } else if (ceph_ino(in) == vino.ino && 1057355da1ebSSage Weil ceph_snap(in) == vino.snap) { 1058355da1ebSSage Weil igrab(in); 1059355da1ebSSage Weil } else { 1060355da1ebSSage Weil dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1061355da1ebSSage Weil dn, in, ceph_ino(in), ceph_snap(in), 1062355da1ebSSage Weil vino.ino, vino.snap); 1063355da1ebSSage Weil have_lease = false; 1064355da1ebSSage Weil in = NULL; 1065355da1ebSSage Weil } 1066355da1ebSSage Weil 1067355da1ebSSage Weil if (have_lease) 1068355da1ebSSage Weil update_dentry_lease(dn, rinfo->dlease, session, 1069355da1ebSSage Weil req->r_request_started); 1070355da1ebSSage Weil dout(" final dn %p\n", dn); 1071355da1ebSSage Weil i++; 1072355da1ebSSage Weil } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1073355da1ebSSage Weil req->r_op == CEPH_MDS_OP_MKSNAP) { 1074355da1ebSSage Weil struct dentry *dn = req->r_dentry; 1075355da1ebSSage Weil 1076355da1ebSSage Weil /* fill out a snapdir LOOKUPSNAP dentry */ 1077355da1ebSSage Weil BUG_ON(!dn); 1078355da1ebSSage Weil BUG_ON(!req->r_locked_dir); 1079355da1ebSSage Weil BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1080355da1ebSSage Weil ininfo = rinfo->targeti.in; 1081355da1ebSSage Weil vino.ino = le64_to_cpu(ininfo->ino); 1082355da1ebSSage Weil vino.snap = le64_to_cpu(ininfo->snapid); 1083355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1084355da1ebSSage Weil if (IS_ERR(in)) { 1085355da1ebSSage Weil pr_err("fill_inode get_inode badness %llx.%llx\n", 1086355da1ebSSage Weil vino.ino, vino.snap); 1087355da1ebSSage Weil err = PTR_ERR(in); 1088355da1ebSSage Weil d_delete(dn); 1089355da1ebSSage Weil goto done; 1090355da1ebSSage Weil } 1091355da1ebSSage Weil dout(" linking snapped dir %p to dn %p\n", in, dn); 1092355da1ebSSage Weil dn = splice_dentry(dn, in, NULL); 1093355da1ebSSage Weil if (IS_ERR(dn)) { 1094355da1ebSSage Weil err = PTR_ERR(dn); 1095355da1ebSSage Weil goto done; 1096355da1ebSSage Weil } 10974baa75efSYehuda Sadeh ceph_set_dentry_offset(dn); 1098355da1ebSSage Weil req->r_dentry = dn; /* may have spliced */ 1099355da1ebSSage Weil igrab(in); 1100355da1ebSSage Weil rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1101355da1ebSSage Weil } 1102355da1ebSSage Weil 1103355da1ebSSage Weil if (rinfo->head->is_target) { 1104355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1105355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1106355da1ebSSage Weil 1107355da1ebSSage Weil if (in == NULL || ceph_ino(in) != vino.ino || 1108355da1ebSSage Weil ceph_snap(in) != vino.snap) { 1109355da1ebSSage Weil in = ceph_get_inode(sb, vino); 1110355da1ebSSage Weil if (IS_ERR(in)) { 1111355da1ebSSage Weil err = PTR_ERR(in); 1112355da1ebSSage Weil goto done; 1113355da1ebSSage Weil } 1114355da1ebSSage Weil } 1115355da1ebSSage Weil req->r_target_inode = in; 1116355da1ebSSage Weil 1117355da1ebSSage Weil err = fill_inode(in, 1118355da1ebSSage Weil &rinfo->targeti, NULL, 1119355da1ebSSage Weil session, req->r_request_started, 1120355da1ebSSage Weil (le32_to_cpu(rinfo->head->result) == 0) ? 1121355da1ebSSage Weil req->r_fmode : -1, 1122355da1ebSSage Weil &req->r_caps_reservation); 1123355da1ebSSage Weil if (err < 0) { 1124355da1ebSSage Weil pr_err("fill_inode badness %p %llx.%llx\n", 1125355da1ebSSage Weil in, ceph_vinop(in)); 1126355da1ebSSage Weil goto done; 1127355da1ebSSage Weil } 1128355da1ebSSage Weil } 1129355da1ebSSage Weil 1130355da1ebSSage Weil done: 1131355da1ebSSage Weil dout("fill_trace done err=%d\n", err); 1132355da1ebSSage Weil return err; 1133355da1ebSSage Weil } 1134355da1ebSSage Weil 1135355da1ebSSage Weil /* 1136355da1ebSSage Weil * Prepopulate our cache with readdir results, leases, etc. 1137355da1ebSSage Weil */ 1138355da1ebSSage Weil int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1139355da1ebSSage Weil struct ceph_mds_session *session) 1140355da1ebSSage Weil { 1141355da1ebSSage Weil struct dentry *parent = req->r_dentry; 1142355da1ebSSage Weil struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1143355da1ebSSage Weil struct qstr dname; 1144355da1ebSSage Weil struct dentry *dn; 1145355da1ebSSage Weil struct inode *in; 1146355da1ebSSage Weil int err = 0, i; 1147355da1ebSSage Weil struct inode *snapdir = NULL; 1148355da1ebSSage Weil struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1149355da1ebSSage Weil u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1150355da1ebSSage Weil struct ceph_dentry_info *di; 1151355da1ebSSage Weil 1152355da1ebSSage Weil if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1153355da1ebSSage Weil snapdir = ceph_get_snapdir(parent->d_inode); 1154355da1ebSSage Weil parent = d_find_alias(snapdir); 1155355da1ebSSage Weil dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1156355da1ebSSage Weil rinfo->dir_nr, parent); 1157355da1ebSSage Weil } else { 1158355da1ebSSage Weil dout("readdir_prepopulate %d items under dn %p\n", 1159355da1ebSSage Weil rinfo->dir_nr, parent); 1160355da1ebSSage Weil if (rinfo->dir_dir) 1161355da1ebSSage Weil ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1162355da1ebSSage Weil } 1163355da1ebSSage Weil 1164355da1ebSSage Weil for (i = 0; i < rinfo->dir_nr; i++) { 1165355da1ebSSage Weil struct ceph_vino vino; 1166355da1ebSSage Weil 1167355da1ebSSage Weil dname.name = rinfo->dir_dname[i]; 1168355da1ebSSage Weil dname.len = rinfo->dir_dname_len[i]; 1169355da1ebSSage Weil dname.hash = full_name_hash(dname.name, dname.len); 1170355da1ebSSage Weil 1171355da1ebSSage Weil vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1172355da1ebSSage Weil vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1173355da1ebSSage Weil 1174355da1ebSSage Weil retry_lookup: 1175355da1ebSSage Weil dn = d_lookup(parent, &dname); 1176355da1ebSSage Weil dout("d_lookup on parent=%p name=%.*s got %p\n", 1177355da1ebSSage Weil parent, dname.len, dname.name, dn); 1178355da1ebSSage Weil 1179355da1ebSSage Weil if (!dn) { 1180355da1ebSSage Weil dn = d_alloc(parent, &dname); 1181355da1ebSSage Weil dout("d_alloc %p '%.*s' = %p\n", parent, 1182355da1ebSSage Weil dname.len, dname.name, dn); 1183355da1ebSSage Weil if (dn == NULL) { 1184355da1ebSSage Weil dout("d_alloc badness\n"); 1185355da1ebSSage Weil err = -ENOMEM; 1186355da1ebSSage Weil goto out; 1187355da1ebSSage Weil } 1188355da1ebSSage Weil err = ceph_init_dentry(dn); 1189355da1ebSSage Weil if (err < 0) 1190355da1ebSSage Weil goto out; 1191355da1ebSSage Weil } else if (dn->d_inode && 1192355da1ebSSage Weil (ceph_ino(dn->d_inode) != vino.ino || 1193355da1ebSSage Weil ceph_snap(dn->d_inode) != vino.snap)) { 1194355da1ebSSage Weil dout(" dn %p points to wrong inode %p\n", 1195355da1ebSSage Weil dn, dn->d_inode); 1196355da1ebSSage Weil d_delete(dn); 1197355da1ebSSage Weil dput(dn); 1198355da1ebSSage Weil goto retry_lookup; 1199355da1ebSSage Weil } else { 1200355da1ebSSage Weil /* reorder parent's d_subdirs */ 1201355da1ebSSage Weil spin_lock(&dcache_lock); 1202355da1ebSSage Weil spin_lock(&dn->d_lock); 1203355da1ebSSage Weil list_move(&dn->d_u.d_child, &parent->d_subdirs); 1204355da1ebSSage Weil spin_unlock(&dn->d_lock); 1205355da1ebSSage Weil spin_unlock(&dcache_lock); 1206355da1ebSSage Weil } 1207355da1ebSSage Weil 1208355da1ebSSage Weil di = dn->d_fsdata; 1209355da1ebSSage Weil di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1210355da1ebSSage Weil 1211355da1ebSSage Weil /* inode */ 1212355da1ebSSage Weil if (dn->d_inode) { 1213355da1ebSSage Weil in = dn->d_inode; 1214355da1ebSSage Weil } else { 1215355da1ebSSage Weil in = ceph_get_inode(parent->d_sb, vino); 1216355da1ebSSage Weil if (in == NULL) { 1217355da1ebSSage Weil dout("new_inode badness\n"); 1218355da1ebSSage Weil d_delete(dn); 1219355da1ebSSage Weil dput(dn); 1220355da1ebSSage Weil err = -ENOMEM; 1221355da1ebSSage Weil goto out; 1222355da1ebSSage Weil } 1223355da1ebSSage Weil dn = splice_dentry(dn, in, NULL); 1224355da1ebSSage Weil } 1225355da1ebSSage Weil 1226355da1ebSSage Weil if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1227355da1ebSSage Weil req->r_request_started, -1, 1228355da1ebSSage Weil &req->r_caps_reservation) < 0) { 1229355da1ebSSage Weil pr_err("fill_inode badness on %p\n", in); 1230355da1ebSSage Weil dput(dn); 1231355da1ebSSage Weil continue; 1232355da1ebSSage Weil } 1233355da1ebSSage Weil update_dentry_lease(dn, rinfo->dir_dlease[i], 1234355da1ebSSage Weil req->r_session, req->r_request_started); 1235355da1ebSSage Weil dput(dn); 1236355da1ebSSage Weil } 1237355da1ebSSage Weil req->r_did_prepopulate = true; 1238355da1ebSSage Weil 1239355da1ebSSage Weil out: 1240355da1ebSSage Weil if (snapdir) { 1241355da1ebSSage Weil iput(snapdir); 1242355da1ebSSage Weil dput(parent); 1243355da1ebSSage Weil } 1244355da1ebSSage Weil dout("readdir_prepopulate done\n"); 1245355da1ebSSage Weil return err; 1246355da1ebSSage Weil } 1247355da1ebSSage Weil 1248355da1ebSSage Weil int ceph_inode_set_size(struct inode *inode, loff_t size) 1249355da1ebSSage Weil { 1250355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1251355da1ebSSage Weil int ret = 0; 1252355da1ebSSage Weil 1253355da1ebSSage Weil spin_lock(&inode->i_lock); 1254355da1ebSSage Weil dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1255355da1ebSSage Weil inode->i_size = size; 1256355da1ebSSage Weil inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1257355da1ebSSage Weil 1258355da1ebSSage Weil /* tell the MDS if we are approaching max_size */ 1259355da1ebSSage Weil if ((size << 1) >= ci->i_max_size && 1260355da1ebSSage Weil (ci->i_reported_size << 1) < ci->i_max_size) 1261355da1ebSSage Weil ret = 1; 1262355da1ebSSage Weil 1263355da1ebSSage Weil spin_unlock(&inode->i_lock); 1264355da1ebSSage Weil return ret; 1265355da1ebSSage Weil } 1266355da1ebSSage Weil 1267355da1ebSSage Weil /* 1268355da1ebSSage Weil * Write back inode data in a worker thread. (This can't be done 1269355da1ebSSage Weil * in the message handler context.) 1270355da1ebSSage Weil */ 12713c6f6b79SSage Weil void ceph_queue_writeback(struct inode *inode) 12723c6f6b79SSage Weil { 12733c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->wb_wq, 12743c6f6b79SSage Weil &ceph_inode(inode)->i_wb_work)) { 12752c27c9a5SSage Weil dout("ceph_queue_writeback %p\n", inode); 12763c6f6b79SSage Weil igrab(inode); 12773c6f6b79SSage Weil } else { 12782c27c9a5SSage Weil dout("ceph_queue_writeback %p failed\n", inode); 12793c6f6b79SSage Weil } 12803c6f6b79SSage Weil } 12813c6f6b79SSage Weil 12823c6f6b79SSage Weil static void ceph_writeback_work(struct work_struct *work) 1283355da1ebSSage Weil { 1284355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1285355da1ebSSage Weil i_wb_work); 1286355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1287355da1ebSSage Weil 1288355da1ebSSage Weil dout("writeback %p\n", inode); 1289355da1ebSSage Weil filemap_fdatawrite(&inode->i_data); 1290355da1ebSSage Weil iput(inode); 1291355da1ebSSage Weil } 1292355da1ebSSage Weil 1293355da1ebSSage Weil /* 12943c6f6b79SSage Weil * queue an async invalidation 12953c6f6b79SSage Weil */ 12963c6f6b79SSage Weil void ceph_queue_invalidate(struct inode *inode) 12973c6f6b79SSage Weil { 12983c6f6b79SSage Weil if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 12993c6f6b79SSage Weil &ceph_inode(inode)->i_pg_inv_work)) { 13003c6f6b79SSage Weil dout("ceph_queue_invalidate %p\n", inode); 13013c6f6b79SSage Weil igrab(inode); 13023c6f6b79SSage Weil } else { 13033c6f6b79SSage Weil dout("ceph_queue_invalidate %p failed\n", inode); 13043c6f6b79SSage Weil } 13053c6f6b79SSage Weil } 13063c6f6b79SSage Weil 13073c6f6b79SSage Weil /* 1308c9af9fb6SYehuda Sadeh * invalidate any pages that are not dirty or under writeback. this 1309c9af9fb6SYehuda Sadeh * includes pages that are clean and mapped. 1310c9af9fb6SYehuda Sadeh */ 1311c9af9fb6SYehuda Sadeh static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1312c9af9fb6SYehuda Sadeh { 1313c9af9fb6SYehuda Sadeh struct pagevec pvec; 1314c9af9fb6SYehuda Sadeh pgoff_t next = 0; 1315c9af9fb6SYehuda Sadeh int i; 1316c9af9fb6SYehuda Sadeh 1317c9af9fb6SYehuda Sadeh pagevec_init(&pvec, 0); 1318c9af9fb6SYehuda Sadeh while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1319c9af9fb6SYehuda Sadeh for (i = 0; i < pagevec_count(&pvec); i++) { 1320c9af9fb6SYehuda Sadeh struct page *page = pvec.pages[i]; 1321c9af9fb6SYehuda Sadeh pgoff_t index; 1322c9af9fb6SYehuda Sadeh int skip_page = 1323c9af9fb6SYehuda Sadeh (PageDirty(page) || PageWriteback(page)); 1324c9af9fb6SYehuda Sadeh 1325c9af9fb6SYehuda Sadeh if (!skip_page) 1326c9af9fb6SYehuda Sadeh skip_page = !trylock_page(page); 1327c9af9fb6SYehuda Sadeh 1328c9af9fb6SYehuda Sadeh /* 1329c9af9fb6SYehuda Sadeh * We really shouldn't be looking at the ->index of an 1330c9af9fb6SYehuda Sadeh * unlocked page. But we're not allowed to lock these 1331c9af9fb6SYehuda Sadeh * pages. So we rely upon nobody altering the ->index 1332c9af9fb6SYehuda Sadeh * of this (pinned-by-us) page. 1333c9af9fb6SYehuda Sadeh */ 1334c9af9fb6SYehuda Sadeh index = page->index; 1335c9af9fb6SYehuda Sadeh if (index > next) 1336c9af9fb6SYehuda Sadeh next = index; 1337c9af9fb6SYehuda Sadeh next++; 1338c9af9fb6SYehuda Sadeh 1339c9af9fb6SYehuda Sadeh if (skip_page) 1340c9af9fb6SYehuda Sadeh continue; 1341c9af9fb6SYehuda Sadeh 1342c9af9fb6SYehuda Sadeh generic_error_remove_page(mapping, page); 1343c9af9fb6SYehuda Sadeh unlock_page(page); 1344c9af9fb6SYehuda Sadeh } 1345c9af9fb6SYehuda Sadeh pagevec_release(&pvec); 1346c9af9fb6SYehuda Sadeh cond_resched(); 1347c9af9fb6SYehuda Sadeh } 1348c9af9fb6SYehuda Sadeh } 1349c9af9fb6SYehuda Sadeh 1350c9af9fb6SYehuda Sadeh /* 1351355da1ebSSage Weil * Invalidate inode pages in a worker thread. (This can't be done 1352355da1ebSSage Weil * in the message handler context.) 1353355da1ebSSage Weil */ 13543c6f6b79SSage Weil static void ceph_invalidate_work(struct work_struct *work) 1355355da1ebSSage Weil { 1356355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1357355da1ebSSage Weil i_pg_inv_work); 1358355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1359355da1ebSSage Weil u32 orig_gen; 1360355da1ebSSage Weil int check = 0; 1361355da1ebSSage Weil 1362355da1ebSSage Weil spin_lock(&inode->i_lock); 1363355da1ebSSage Weil dout("invalidate_pages %p gen %d revoking %d\n", inode, 1364355da1ebSSage Weil ci->i_rdcache_gen, ci->i_rdcache_revoking); 1365355da1ebSSage Weil if (ci->i_rdcache_gen == 0 || 1366355da1ebSSage Weil ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1367355da1ebSSage Weil BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen); 1368355da1ebSSage Weil /* nevermind! */ 1369355da1ebSSage Weil ci->i_rdcache_revoking = 0; 1370355da1ebSSage Weil spin_unlock(&inode->i_lock); 1371355da1ebSSage Weil goto out; 1372355da1ebSSage Weil } 1373355da1ebSSage Weil orig_gen = ci->i_rdcache_gen; 1374355da1ebSSage Weil spin_unlock(&inode->i_lock); 1375355da1ebSSage Weil 1376c9af9fb6SYehuda Sadeh ceph_invalidate_nondirty_pages(inode->i_mapping); 1377355da1ebSSage Weil 1378355da1ebSSage Weil spin_lock(&inode->i_lock); 1379355da1ebSSage Weil if (orig_gen == ci->i_rdcache_gen) { 1380355da1ebSSage Weil dout("invalidate_pages %p gen %d successful\n", inode, 1381355da1ebSSage Weil ci->i_rdcache_gen); 1382355da1ebSSage Weil ci->i_rdcache_gen = 0; 1383355da1ebSSage Weil ci->i_rdcache_revoking = 0; 1384355da1ebSSage Weil check = 1; 1385355da1ebSSage Weil } else { 1386355da1ebSSage Weil dout("invalidate_pages %p gen %d raced, gen now %d\n", 1387355da1ebSSage Weil inode, orig_gen, ci->i_rdcache_gen); 1388355da1ebSSage Weil } 1389355da1ebSSage Weil spin_unlock(&inode->i_lock); 1390355da1ebSSage Weil 1391355da1ebSSage Weil if (check) 1392355da1ebSSage Weil ceph_check_caps(ci, 0, NULL); 1393355da1ebSSage Weil out: 1394355da1ebSSage Weil iput(inode); 1395355da1ebSSage Weil } 1396355da1ebSSage Weil 1397355da1ebSSage Weil 1398355da1ebSSage Weil /* 1399355da1ebSSage Weil * called by trunc_wq; take i_mutex ourselves 1400355da1ebSSage Weil * 1401355da1ebSSage Weil * We also truncate in a separate thread as well. 1402355da1ebSSage Weil */ 14033c6f6b79SSage Weil static void ceph_vmtruncate_work(struct work_struct *work) 1404355da1ebSSage Weil { 1405355da1ebSSage Weil struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1406355da1ebSSage Weil i_vmtruncate_work); 1407355da1ebSSage Weil struct inode *inode = &ci->vfs_inode; 1408355da1ebSSage Weil 1409355da1ebSSage Weil dout("vmtruncate_work %p\n", inode); 1410355da1ebSSage Weil mutex_lock(&inode->i_mutex); 1411355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1412355da1ebSSage Weil mutex_unlock(&inode->i_mutex); 1413355da1ebSSage Weil iput(inode); 1414355da1ebSSage Weil } 1415355da1ebSSage Weil 1416355da1ebSSage Weil /* 14173c6f6b79SSage Weil * Queue an async vmtruncate. If we fail to queue work, we will handle 14183c6f6b79SSage Weil * the truncation the next time we call __ceph_do_pending_vmtruncate. 14193c6f6b79SSage Weil */ 14203c6f6b79SSage Weil void ceph_queue_vmtruncate(struct inode *inode) 14213c6f6b79SSage Weil { 14223c6f6b79SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14233c6f6b79SSage Weil 14243c6f6b79SSage Weil if (queue_work(ceph_client(inode->i_sb)->trunc_wq, 14253c6f6b79SSage Weil &ci->i_vmtruncate_work)) { 14263c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p\n", inode); 14273c6f6b79SSage Weil igrab(inode); 14283c6f6b79SSage Weil } else { 14293c6f6b79SSage Weil dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 14303c6f6b79SSage Weil inode, ci->i_truncate_pending); 14313c6f6b79SSage Weil } 14323c6f6b79SSage Weil } 14333c6f6b79SSage Weil 14343c6f6b79SSage Weil /* 1435355da1ebSSage Weil * called with i_mutex held. 1436355da1ebSSage Weil * 1437355da1ebSSage Weil * Make sure any pending truncation is applied before doing anything 1438355da1ebSSage Weil * that may depend on it. 1439355da1ebSSage Weil */ 1440355da1ebSSage Weil void __ceph_do_pending_vmtruncate(struct inode *inode) 1441355da1ebSSage Weil { 1442355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1443355da1ebSSage Weil u64 to; 1444355da1ebSSage Weil int wrbuffer_refs, wake = 0; 1445355da1ebSSage Weil 1446355da1ebSSage Weil retry: 1447355da1ebSSage Weil spin_lock(&inode->i_lock); 1448355da1ebSSage Weil if (ci->i_truncate_pending == 0) { 1449355da1ebSSage Weil dout("__do_pending_vmtruncate %p none pending\n", inode); 1450355da1ebSSage Weil spin_unlock(&inode->i_lock); 1451355da1ebSSage Weil return; 1452355da1ebSSage Weil } 1453355da1ebSSage Weil 1454355da1ebSSage Weil /* 1455355da1ebSSage Weil * make sure any dirty snapped pages are flushed before we 1456355da1ebSSage Weil * possibly truncate them.. so write AND block! 1457355da1ebSSage Weil */ 1458355da1ebSSage Weil if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1459355da1ebSSage Weil dout("__do_pending_vmtruncate %p flushing snaps first\n", 1460355da1ebSSage Weil inode); 1461355da1ebSSage Weil spin_unlock(&inode->i_lock); 1462355da1ebSSage Weil filemap_write_and_wait_range(&inode->i_data, 0, 1463355da1ebSSage Weil inode->i_sb->s_maxbytes); 1464355da1ebSSage Weil goto retry; 1465355da1ebSSage Weil } 1466355da1ebSSage Weil 1467355da1ebSSage Weil to = ci->i_truncate_size; 1468355da1ebSSage Weil wrbuffer_refs = ci->i_wrbuffer_ref; 1469355da1ebSSage Weil dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1470355da1ebSSage Weil ci->i_truncate_pending, to); 1471355da1ebSSage Weil spin_unlock(&inode->i_lock); 1472355da1ebSSage Weil 1473355da1ebSSage Weil truncate_inode_pages(inode->i_mapping, to); 1474355da1ebSSage Weil 1475355da1ebSSage Weil spin_lock(&inode->i_lock); 1476355da1ebSSage Weil ci->i_truncate_pending--; 1477355da1ebSSage Weil if (ci->i_truncate_pending == 0) 1478355da1ebSSage Weil wake = 1; 1479355da1ebSSage Weil spin_unlock(&inode->i_lock); 1480355da1ebSSage Weil 1481355da1ebSSage Weil if (wrbuffer_refs == 0) 1482355da1ebSSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1483355da1ebSSage Weil if (wake) 1484355da1ebSSage Weil wake_up(&ci->i_cap_wq); 1485355da1ebSSage Weil } 1486355da1ebSSage Weil 1487355da1ebSSage Weil 1488355da1ebSSage Weil /* 1489355da1ebSSage Weil * symlinks 1490355da1ebSSage Weil */ 1491355da1ebSSage Weil static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1492355da1ebSSage Weil { 1493355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1494355da1ebSSage Weil nd_set_link(nd, ci->i_symlink); 1495355da1ebSSage Weil return NULL; 1496355da1ebSSage Weil } 1497355da1ebSSage Weil 1498355da1ebSSage Weil static const struct inode_operations ceph_symlink_iops = { 1499355da1ebSSage Weil .readlink = generic_readlink, 1500355da1ebSSage Weil .follow_link = ceph_sym_follow_link, 1501355da1ebSSage Weil }; 1502355da1ebSSage Weil 1503355da1ebSSage Weil /* 1504355da1ebSSage Weil * setattr 1505355da1ebSSage Weil */ 1506355da1ebSSage Weil int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1507355da1ebSSage Weil { 1508355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1509355da1ebSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1510355da1ebSSage Weil struct inode *parent_inode = dentry->d_parent->d_inode; 1511355da1ebSSage Weil const unsigned int ia_valid = attr->ia_valid; 1512355da1ebSSage Weil struct ceph_mds_request *req; 1513355da1ebSSage Weil struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc; 1514355da1ebSSage Weil int issued; 1515355da1ebSSage Weil int release = 0, dirtied = 0; 1516355da1ebSSage Weil int mask = 0; 1517355da1ebSSage Weil int err = 0; 1518355da1ebSSage Weil 1519355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1520355da1ebSSage Weil return -EROFS; 1521355da1ebSSage Weil 1522355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1523355da1ebSSage Weil 1524355da1ebSSage Weil err = inode_change_ok(inode, attr); 1525355da1ebSSage Weil if (err != 0) 1526355da1ebSSage Weil return err; 1527355da1ebSSage Weil 1528355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1529355da1ebSSage Weil USE_AUTH_MDS); 1530355da1ebSSage Weil if (IS_ERR(req)) 1531355da1ebSSage Weil return PTR_ERR(req); 1532355da1ebSSage Weil 1533355da1ebSSage Weil spin_lock(&inode->i_lock); 1534355da1ebSSage Weil issued = __ceph_caps_issued(ci, NULL); 1535355da1ebSSage Weil dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1536355da1ebSSage Weil 1537355da1ebSSage Weil if (ia_valid & ATTR_UID) { 1538355da1ebSSage Weil dout("setattr %p uid %d -> %d\n", inode, 1539355da1ebSSage Weil inode->i_uid, attr->ia_uid); 1540355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1541355da1ebSSage Weil inode->i_uid = attr->ia_uid; 1542355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1543355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1544355da1ebSSage Weil attr->ia_uid != inode->i_uid) { 1545355da1ebSSage Weil req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1546355da1ebSSage Weil mask |= CEPH_SETATTR_UID; 1547355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1548355da1ebSSage Weil } 1549355da1ebSSage Weil } 1550355da1ebSSage Weil if (ia_valid & ATTR_GID) { 1551355da1ebSSage Weil dout("setattr %p gid %d -> %d\n", inode, 1552355da1ebSSage Weil inode->i_gid, attr->ia_gid); 1553355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1554355da1ebSSage Weil inode->i_gid = attr->ia_gid; 1555355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1556355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1557355da1ebSSage Weil attr->ia_gid != inode->i_gid) { 1558355da1ebSSage Weil req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1559355da1ebSSage Weil mask |= CEPH_SETATTR_GID; 1560355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1561355da1ebSSage Weil } 1562355da1ebSSage Weil } 1563355da1ebSSage Weil if (ia_valid & ATTR_MODE) { 1564355da1ebSSage Weil dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1565355da1ebSSage Weil attr->ia_mode); 1566355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) { 1567355da1ebSSage Weil inode->i_mode = attr->ia_mode; 1568355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1569355da1ebSSage Weil } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1570355da1ebSSage Weil attr->ia_mode != inode->i_mode) { 1571355da1ebSSage Weil req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1572355da1ebSSage Weil mask |= CEPH_SETATTR_MODE; 1573355da1ebSSage Weil release |= CEPH_CAP_AUTH_SHARED; 1574355da1ebSSage Weil } 1575355da1ebSSage Weil } 1576355da1ebSSage Weil 1577355da1ebSSage Weil if (ia_valid & ATTR_ATIME) { 1578355da1ebSSage Weil dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1579355da1ebSSage Weil inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1580355da1ebSSage Weil attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1581355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1582355da1ebSSage Weil ci->i_time_warp_seq++; 1583355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1584355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1585355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1586355da1ebSSage Weil timespec_compare(&inode->i_atime, 1587355da1ebSSage Weil &attr->ia_atime) < 0) { 1588355da1ebSSage Weil inode->i_atime = attr->ia_atime; 1589355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1590355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1591355da1ebSSage Weil !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1592355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.atime, 1593355da1ebSSage Weil &attr->ia_atime); 1594355da1ebSSage Weil mask |= CEPH_SETATTR_ATIME; 1595355da1ebSSage Weil release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1596355da1ebSSage Weil CEPH_CAP_FILE_WR; 1597355da1ebSSage Weil } 1598355da1ebSSage Weil } 1599355da1ebSSage Weil if (ia_valid & ATTR_MTIME) { 1600355da1ebSSage Weil dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1601355da1ebSSage Weil inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1602355da1ebSSage Weil attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1603355da1ebSSage Weil if (issued & CEPH_CAP_FILE_EXCL) { 1604355da1ebSSage Weil ci->i_time_warp_seq++; 1605355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1606355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1607355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_WR) && 1608355da1ebSSage Weil timespec_compare(&inode->i_mtime, 1609355da1ebSSage Weil &attr->ia_mtime) < 0) { 1610355da1ebSSage Weil inode->i_mtime = attr->ia_mtime; 1611355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_WR; 1612355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1613355da1ebSSage Weil !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1614355da1ebSSage Weil ceph_encode_timespec(&req->r_args.setattr.mtime, 1615355da1ebSSage Weil &attr->ia_mtime); 1616355da1ebSSage Weil mask |= CEPH_SETATTR_MTIME; 1617355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1618355da1ebSSage Weil CEPH_CAP_FILE_WR; 1619355da1ebSSage Weil } 1620355da1ebSSage Weil } 1621355da1ebSSage Weil if (ia_valid & ATTR_SIZE) { 1622355da1ebSSage Weil dout("setattr %p size %lld -> %lld\n", inode, 1623355da1ebSSage Weil inode->i_size, attr->ia_size); 1624355da1ebSSage Weil if (attr->ia_size > inode->i_sb->s_maxbytes) { 1625355da1ebSSage Weil err = -EINVAL; 1626355da1ebSSage Weil goto out; 1627355da1ebSSage Weil } 1628355da1ebSSage Weil if ((issued & CEPH_CAP_FILE_EXCL) && 1629355da1ebSSage Weil attr->ia_size > inode->i_size) { 1630355da1ebSSage Weil inode->i_size = attr->ia_size; 1631355da1ebSSage Weil inode->i_blocks = 1632355da1ebSSage Weil (attr->ia_size + (1 << 9) - 1) >> 9; 1633355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1634355da1ebSSage Weil ci->i_reported_size = attr->ia_size; 1635355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1636355da1ebSSage Weil } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1637355da1ebSSage Weil attr->ia_size != inode->i_size) { 1638355da1ebSSage Weil req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1639355da1ebSSage Weil req->r_args.setattr.old_size = 1640355da1ebSSage Weil cpu_to_le64(inode->i_size); 1641355da1ebSSage Weil mask |= CEPH_SETATTR_SIZE; 1642355da1ebSSage Weil release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1643355da1ebSSage Weil CEPH_CAP_FILE_WR; 1644355da1ebSSage Weil } 1645355da1ebSSage Weil } 1646355da1ebSSage Weil 1647355da1ebSSage Weil /* these do nothing */ 1648355da1ebSSage Weil if (ia_valid & ATTR_CTIME) { 1649355da1ebSSage Weil bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1650355da1ebSSage Weil ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1651355da1ebSSage Weil dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1652355da1ebSSage Weil inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1653355da1ebSSage Weil attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1654355da1ebSSage Weil only ? "ctime only" : "ignored"); 1655355da1ebSSage Weil inode->i_ctime = attr->ia_ctime; 1656355da1ebSSage Weil if (only) { 1657355da1ebSSage Weil /* 1658355da1ebSSage Weil * if kernel wants to dirty ctime but nothing else, 1659355da1ebSSage Weil * we need to choose a cap to dirty under, or do 1660355da1ebSSage Weil * a almost-no-op setattr 1661355da1ebSSage Weil */ 1662355da1ebSSage Weil if (issued & CEPH_CAP_AUTH_EXCL) 1663355da1ebSSage Weil dirtied |= CEPH_CAP_AUTH_EXCL; 1664355da1ebSSage Weil else if (issued & CEPH_CAP_FILE_EXCL) 1665355da1ebSSage Weil dirtied |= CEPH_CAP_FILE_EXCL; 1666355da1ebSSage Weil else if (issued & CEPH_CAP_XATTR_EXCL) 1667355da1ebSSage Weil dirtied |= CEPH_CAP_XATTR_EXCL; 1668355da1ebSSage Weil else 1669355da1ebSSage Weil mask |= CEPH_SETATTR_CTIME; 1670355da1ebSSage Weil } 1671355da1ebSSage Weil } 1672355da1ebSSage Weil if (ia_valid & ATTR_FILE) 1673355da1ebSSage Weil dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1674355da1ebSSage Weil 1675355da1ebSSage Weil if (dirtied) { 1676355da1ebSSage Weil __ceph_mark_dirty_caps(ci, dirtied); 1677355da1ebSSage Weil inode->i_ctime = CURRENT_TIME; 1678355da1ebSSage Weil } 1679355da1ebSSage Weil 1680355da1ebSSage Weil release &= issued; 1681355da1ebSSage Weil spin_unlock(&inode->i_lock); 1682355da1ebSSage Weil 1683355da1ebSSage Weil if (mask) { 1684355da1ebSSage Weil req->r_inode = igrab(inode); 1685355da1ebSSage Weil req->r_inode_drop = release; 1686355da1ebSSage Weil req->r_args.setattr.mask = cpu_to_le32(mask); 1687355da1ebSSage Weil req->r_num_caps = 1; 1688355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1689355da1ebSSage Weil } 1690355da1ebSSage Weil dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1691355da1ebSSage Weil ceph_cap_string(dirtied), mask); 1692355da1ebSSage Weil 1693355da1ebSSage Weil ceph_mdsc_put_request(req); 1694355da1ebSSage Weil __ceph_do_pending_vmtruncate(inode); 1695355da1ebSSage Weil return err; 1696355da1ebSSage Weil out: 1697355da1ebSSage Weil spin_unlock(&inode->i_lock); 1698355da1ebSSage Weil ceph_mdsc_put_request(req); 1699355da1ebSSage Weil return err; 1700355da1ebSSage Weil } 1701355da1ebSSage Weil 1702355da1ebSSage Weil /* 1703355da1ebSSage Weil * Verify that we have a lease on the given mask. If not, 1704355da1ebSSage Weil * do a getattr against an mds. 1705355da1ebSSage Weil */ 1706355da1ebSSage Weil int ceph_do_getattr(struct inode *inode, int mask) 1707355da1ebSSage Weil { 1708355da1ebSSage Weil struct ceph_client *client = ceph_sb_to_client(inode->i_sb); 1709355da1ebSSage Weil struct ceph_mds_client *mdsc = &client->mdsc; 1710355da1ebSSage Weil struct ceph_mds_request *req; 1711355da1ebSSage Weil int err; 1712355da1ebSSage Weil 1713355da1ebSSage Weil if (ceph_snap(inode) == CEPH_SNAPDIR) { 1714355da1ebSSage Weil dout("do_getattr inode %p SNAPDIR\n", inode); 1715355da1ebSSage Weil return 0; 1716355da1ebSSage Weil } 1717355da1ebSSage Weil 1718355da1ebSSage Weil dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask)); 1719355da1ebSSage Weil if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1720355da1ebSSage Weil return 0; 1721355da1ebSSage Weil 1722355da1ebSSage Weil req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1723355da1ebSSage Weil if (IS_ERR(req)) 1724355da1ebSSage Weil return PTR_ERR(req); 1725355da1ebSSage Weil req->r_inode = igrab(inode); 1726355da1ebSSage Weil req->r_num_caps = 1; 1727355da1ebSSage Weil req->r_args.getattr.mask = cpu_to_le32(mask); 1728355da1ebSSage Weil err = ceph_mdsc_do_request(mdsc, NULL, req); 1729355da1ebSSage Weil ceph_mdsc_put_request(req); 1730355da1ebSSage Weil dout("do_getattr result=%d\n", err); 1731355da1ebSSage Weil return err; 1732355da1ebSSage Weil } 1733355da1ebSSage Weil 1734355da1ebSSage Weil 1735355da1ebSSage Weil /* 1736355da1ebSSage Weil * Check inode permissions. We verify we have a valid value for 1737355da1ebSSage Weil * the AUTH cap, then call the generic handler. 1738355da1ebSSage Weil */ 1739355da1ebSSage Weil int ceph_permission(struct inode *inode, int mask) 1740355da1ebSSage Weil { 1741355da1ebSSage Weil int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1742355da1ebSSage Weil 1743355da1ebSSage Weil if (!err) 1744355da1ebSSage Weil err = generic_permission(inode, mask, NULL); 1745355da1ebSSage Weil return err; 1746355da1ebSSage Weil } 1747355da1ebSSage Weil 1748355da1ebSSage Weil /* 1749355da1ebSSage Weil * Get all attributes. Hopefully somedata we'll have a statlite() 1750355da1ebSSage Weil * and can limit the fields we require to be accurate. 1751355da1ebSSage Weil */ 1752355da1ebSSage Weil int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1753355da1ebSSage Weil struct kstat *stat) 1754355da1ebSSage Weil { 1755355da1ebSSage Weil struct inode *inode = dentry->d_inode; 1756232d4b01SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1757355da1ebSSage Weil int err; 1758355da1ebSSage Weil 1759355da1ebSSage Weil err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1760355da1ebSSage Weil if (!err) { 1761355da1ebSSage Weil generic_fillattr(inode, stat); 1762355da1ebSSage Weil stat->ino = inode->i_ino; 1763355da1ebSSage Weil if (ceph_snap(inode) != CEPH_NOSNAP) 1764355da1ebSSage Weil stat->dev = ceph_snap(inode); 1765355da1ebSSage Weil else 1766355da1ebSSage Weil stat->dev = 0; 1767232d4b01SSage Weil if (S_ISDIR(inode->i_mode)) { 1768232d4b01SSage Weil stat->size = ci->i_rbytes; 1769232d4b01SSage Weil stat->blocks = 0; 1770355da1ebSSage Weil stat->blksize = 65536; 1771355da1ebSSage Weil } 1772232d4b01SSage Weil } 1773355da1ebSSage Weil return err; 1774355da1ebSSage Weil } 1775