13d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 2a8599bd8SSage Weil 3a8599bd8SSage Weil #include <linux/fs.h> 4a8599bd8SSage Weil #include <linux/kernel.h> 5a8599bd8SSage Weil #include <linux/sched.h> 65a0e3ad6STejun Heo #include <linux/slab.h> 7a8599bd8SSage Weil #include <linux/vmalloc.h> 8a8599bd8SSage Weil #include <linux/wait.h> 9f1a3d572SStephen Rothwell #include <linux/writeback.h> 10a8599bd8SSage Weil 11a8599bd8SSage Weil #include "super.h" 123d14c5d2SYehuda Sadeh #include "mds_client.h" 1399ccbd22SMilosz Tanski #include "cache.h" 143d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 153d14c5d2SYehuda Sadeh #include <linux/ceph/messenger.h> 16a8599bd8SSage Weil 17a8599bd8SSage Weil /* 18a8599bd8SSage Weil * Capability management 19a8599bd8SSage Weil * 20a8599bd8SSage Weil * The Ceph metadata servers control client access to inode metadata 21a8599bd8SSage Weil * and file data by issuing capabilities, granting clients permission 22a8599bd8SSage Weil * to read and/or write both inode field and file data to OSDs 23a8599bd8SSage Weil * (storage nodes). Each capability consists of a set of bits 24a8599bd8SSage Weil * indicating which operations are allowed. 25a8599bd8SSage Weil * 26a8599bd8SSage Weil * If the client holds a *_SHARED cap, the client has a coherent value 27a8599bd8SSage Weil * that can be safely read from the cached inode. 28a8599bd8SSage Weil * 29a8599bd8SSage Weil * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the 30a8599bd8SSage Weil * client is allowed to change inode attributes (e.g., file size, 31a8599bd8SSage Weil * mtime), note its dirty state in the ceph_cap, and asynchronously 32a8599bd8SSage Weil * flush that metadata change to the MDS. 33a8599bd8SSage Weil * 34a8599bd8SSage Weil * In the event of a conflicting operation (perhaps by another 35a8599bd8SSage Weil * client), the MDS will revoke the conflicting client capabilities. 36a8599bd8SSage Weil * 37a8599bd8SSage Weil * In order for a client to cache an inode, it must hold a capability 38a8599bd8SSage Weil * with at least one MDS server. When inodes are released, release 39a8599bd8SSage Weil * notifications are batched and periodically sent en masse to the MDS 40a8599bd8SSage Weil * cluster to release server state. 41a8599bd8SSage Weil */ 42a8599bd8SSage Weil 43a8599bd8SSage Weil 44a8599bd8SSage Weil /* 45a8599bd8SSage Weil * Generate readable cap strings for debugging output. 46a8599bd8SSage Weil */ 47a8599bd8SSage Weil #define MAX_CAP_STR 20 48a8599bd8SSage Weil static char cap_str[MAX_CAP_STR][40]; 49a8599bd8SSage Weil static DEFINE_SPINLOCK(cap_str_lock); 50a8599bd8SSage Weil static int last_cap_str; 51a8599bd8SSage Weil 52a8599bd8SSage Weil static char *gcap_string(char *s, int c) 53a8599bd8SSage Weil { 54a8599bd8SSage Weil if (c & CEPH_CAP_GSHARED) 55a8599bd8SSage Weil *s++ = 's'; 56a8599bd8SSage Weil if (c & CEPH_CAP_GEXCL) 57a8599bd8SSage Weil *s++ = 'x'; 58a8599bd8SSage Weil if (c & CEPH_CAP_GCACHE) 59a8599bd8SSage Weil *s++ = 'c'; 60a8599bd8SSage Weil if (c & CEPH_CAP_GRD) 61a8599bd8SSage Weil *s++ = 'r'; 62a8599bd8SSage Weil if (c & CEPH_CAP_GWR) 63a8599bd8SSage Weil *s++ = 'w'; 64a8599bd8SSage Weil if (c & CEPH_CAP_GBUFFER) 65a8599bd8SSage Weil *s++ = 'b'; 66a8599bd8SSage Weil if (c & CEPH_CAP_GLAZYIO) 67a8599bd8SSage Weil *s++ = 'l'; 68a8599bd8SSage Weil return s; 69a8599bd8SSage Weil } 70a8599bd8SSage Weil 71a8599bd8SSage Weil const char *ceph_cap_string(int caps) 72a8599bd8SSage Weil { 73a8599bd8SSage Weil int i; 74a8599bd8SSage Weil char *s; 75a8599bd8SSage Weil int c; 76a8599bd8SSage Weil 77a8599bd8SSage Weil spin_lock(&cap_str_lock); 78a8599bd8SSage Weil i = last_cap_str++; 79a8599bd8SSage Weil if (last_cap_str == MAX_CAP_STR) 80a8599bd8SSage Weil last_cap_str = 0; 81a8599bd8SSage Weil spin_unlock(&cap_str_lock); 82a8599bd8SSage Weil 83a8599bd8SSage Weil s = cap_str[i]; 84a8599bd8SSage Weil 85a8599bd8SSage Weil if (caps & CEPH_CAP_PIN) 86a8599bd8SSage Weil *s++ = 'p'; 87a8599bd8SSage Weil 88a8599bd8SSage Weil c = (caps >> CEPH_CAP_SAUTH) & 3; 89a8599bd8SSage Weil if (c) { 90a8599bd8SSage Weil *s++ = 'A'; 91a8599bd8SSage Weil s = gcap_string(s, c); 92a8599bd8SSage Weil } 93a8599bd8SSage Weil 94a8599bd8SSage Weil c = (caps >> CEPH_CAP_SLINK) & 3; 95a8599bd8SSage Weil if (c) { 96a8599bd8SSage Weil *s++ = 'L'; 97a8599bd8SSage Weil s = gcap_string(s, c); 98a8599bd8SSage Weil } 99a8599bd8SSage Weil 100a8599bd8SSage Weil c = (caps >> CEPH_CAP_SXATTR) & 3; 101a8599bd8SSage Weil if (c) { 102a8599bd8SSage Weil *s++ = 'X'; 103a8599bd8SSage Weil s = gcap_string(s, c); 104a8599bd8SSage Weil } 105a8599bd8SSage Weil 106a8599bd8SSage Weil c = caps >> CEPH_CAP_SFILE; 107a8599bd8SSage Weil if (c) { 108a8599bd8SSage Weil *s++ = 'F'; 109a8599bd8SSage Weil s = gcap_string(s, c); 110a8599bd8SSage Weil } 111a8599bd8SSage Weil 112a8599bd8SSage Weil if (s == cap_str[i]) 113a8599bd8SSage Weil *s++ = '-'; 114a8599bd8SSage Weil *s = 0; 115a8599bd8SSage Weil return cap_str[i]; 116a8599bd8SSage Weil } 117a8599bd8SSage Weil 11837151668SYehuda Sadeh void ceph_caps_init(struct ceph_mds_client *mdsc) 119a8599bd8SSage Weil { 12037151668SYehuda Sadeh INIT_LIST_HEAD(&mdsc->caps_list); 12137151668SYehuda Sadeh spin_lock_init(&mdsc->caps_list_lock); 122a8599bd8SSage Weil } 123a8599bd8SSage Weil 12437151668SYehuda Sadeh void ceph_caps_finalize(struct ceph_mds_client *mdsc) 125a8599bd8SSage Weil { 126a8599bd8SSage Weil struct ceph_cap *cap; 127a8599bd8SSage Weil 12837151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 12937151668SYehuda Sadeh while (!list_empty(&mdsc->caps_list)) { 13037151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, 13137151668SYehuda Sadeh struct ceph_cap, caps_item); 132a8599bd8SSage Weil list_del(&cap->caps_item); 133a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 134a8599bd8SSage Weil } 13537151668SYehuda Sadeh mdsc->caps_total_count = 0; 13637151668SYehuda Sadeh mdsc->caps_avail_count = 0; 13737151668SYehuda Sadeh mdsc->caps_use_count = 0; 13837151668SYehuda Sadeh mdsc->caps_reserve_count = 0; 13937151668SYehuda Sadeh mdsc->caps_min_count = 0; 14037151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 14185ccce43SSage Weil } 14285ccce43SSage Weil 14337151668SYehuda Sadeh void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta) 14485ccce43SSage Weil { 14537151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 14637151668SYehuda Sadeh mdsc->caps_min_count += delta; 14737151668SYehuda Sadeh BUG_ON(mdsc->caps_min_count < 0); 14837151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 149a8599bd8SSage Weil } 150a8599bd8SSage Weil 15193faca6eSmajianpeng void ceph_reserve_caps(struct ceph_mds_client *mdsc, 15237151668SYehuda Sadeh struct ceph_cap_reservation *ctx, int need) 153a8599bd8SSage Weil { 154a8599bd8SSage Weil int i; 155a8599bd8SSage Weil struct ceph_cap *cap; 156a8599bd8SSage Weil int have; 157a8599bd8SSage Weil int alloc = 0; 158a8599bd8SSage Weil LIST_HEAD(newcaps); 159a8599bd8SSage Weil 160a8599bd8SSage Weil dout("reserve caps ctx=%p need=%d\n", ctx, need); 161a8599bd8SSage Weil 162a8599bd8SSage Weil /* first reserve any caps that are already allocated */ 16337151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 16437151668SYehuda Sadeh if (mdsc->caps_avail_count >= need) 165a8599bd8SSage Weil have = need; 166a8599bd8SSage Weil else 16737151668SYehuda Sadeh have = mdsc->caps_avail_count; 16837151668SYehuda Sadeh mdsc->caps_avail_count -= have; 16937151668SYehuda Sadeh mdsc->caps_reserve_count += have; 17037151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 17137151668SYehuda Sadeh mdsc->caps_reserve_count + 17237151668SYehuda Sadeh mdsc->caps_avail_count); 17337151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 174a8599bd8SSage Weil 175a8599bd8SSage Weil for (i = have; i < need; i++) { 176a8599bd8SSage Weil cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); 17793faca6eSmajianpeng if (!cap) 17893faca6eSmajianpeng break; 179a8599bd8SSage Weil list_add(&cap->caps_item, &newcaps); 180a8599bd8SSage Weil alloc++; 181a8599bd8SSage Weil } 18293faca6eSmajianpeng /* we didn't manage to reserve as much as we needed */ 18393faca6eSmajianpeng if (have + alloc != need) 18493faca6eSmajianpeng pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n", 18593faca6eSmajianpeng ctx, need, have + alloc); 186a8599bd8SSage Weil 18737151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 18837151668SYehuda Sadeh mdsc->caps_total_count += alloc; 18937151668SYehuda Sadeh mdsc->caps_reserve_count += alloc; 19037151668SYehuda Sadeh list_splice(&newcaps, &mdsc->caps_list); 191a8599bd8SSage Weil 19237151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 19337151668SYehuda Sadeh mdsc->caps_reserve_count + 19437151668SYehuda Sadeh mdsc->caps_avail_count); 19537151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 196a8599bd8SSage Weil 197a8599bd8SSage Weil ctx->count = need; 198a8599bd8SSage Weil dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n", 19937151668SYehuda Sadeh ctx, mdsc->caps_total_count, mdsc->caps_use_count, 20037151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 201a8599bd8SSage Weil } 202a8599bd8SSage Weil 20337151668SYehuda Sadeh int ceph_unreserve_caps(struct ceph_mds_client *mdsc, 20437151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 205a8599bd8SSage Weil { 206a8599bd8SSage Weil dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count); 207a8599bd8SSage Weil if (ctx->count) { 20837151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 20937151668SYehuda Sadeh BUG_ON(mdsc->caps_reserve_count < ctx->count); 21037151668SYehuda Sadeh mdsc->caps_reserve_count -= ctx->count; 21137151668SYehuda Sadeh mdsc->caps_avail_count += ctx->count; 212a8599bd8SSage Weil ctx->count = 0; 213a8599bd8SSage Weil dout("unreserve caps %d = %d used + %d resv + %d avail\n", 21437151668SYehuda Sadeh mdsc->caps_total_count, mdsc->caps_use_count, 21537151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 21637151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 21737151668SYehuda Sadeh mdsc->caps_reserve_count + 21837151668SYehuda Sadeh mdsc->caps_avail_count); 21937151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 220a8599bd8SSage Weil } 221a8599bd8SSage Weil return 0; 222a8599bd8SSage Weil } 223a8599bd8SSage Weil 224d9df2783SYan, Zheng struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc, 22537151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 226a8599bd8SSage Weil { 227a8599bd8SSage Weil struct ceph_cap *cap = NULL; 228a8599bd8SSage Weil 229a8599bd8SSage Weil /* temporary, until we do something about cap import/export */ 230443b3760SSage Weil if (!ctx) { 231443b3760SSage Weil cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); 232443b3760SSage Weil if (cap) { 2334d1d0534SYan, Zheng spin_lock(&mdsc->caps_list_lock); 23437151668SYehuda Sadeh mdsc->caps_use_count++; 23537151668SYehuda Sadeh mdsc->caps_total_count++; 2364d1d0534SYan, Zheng spin_unlock(&mdsc->caps_list_lock); 237443b3760SSage Weil } 238443b3760SSage Weil return cap; 239443b3760SSage Weil } 240a8599bd8SSage Weil 24137151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 242a8599bd8SSage Weil dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n", 24337151668SYehuda Sadeh ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count, 24437151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 245a8599bd8SSage Weil BUG_ON(!ctx->count); 24637151668SYehuda Sadeh BUG_ON(ctx->count > mdsc->caps_reserve_count); 24737151668SYehuda Sadeh BUG_ON(list_empty(&mdsc->caps_list)); 248a8599bd8SSage Weil 249a8599bd8SSage Weil ctx->count--; 25037151668SYehuda Sadeh mdsc->caps_reserve_count--; 25137151668SYehuda Sadeh mdsc->caps_use_count++; 252a8599bd8SSage Weil 25337151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item); 254a8599bd8SSage Weil list_del(&cap->caps_item); 255a8599bd8SSage Weil 25637151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 25737151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 25837151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 259a8599bd8SSage Weil return cap; 260a8599bd8SSage Weil } 261a8599bd8SSage Weil 26237151668SYehuda Sadeh void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap) 263a8599bd8SSage Weil { 26437151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 2657c1332b8SSage Weil dout("put_cap %p %d = %d used + %d resv + %d avail\n", 26637151668SYehuda Sadeh cap, mdsc->caps_total_count, mdsc->caps_use_count, 26737151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 26837151668SYehuda Sadeh mdsc->caps_use_count--; 269a8599bd8SSage Weil /* 27085ccce43SSage Weil * Keep some preallocated caps around (ceph_min_count), to 27185ccce43SSage Weil * avoid lots of free/alloc churn. 272a8599bd8SSage Weil */ 27337151668SYehuda Sadeh if (mdsc->caps_avail_count >= mdsc->caps_reserve_count + 27437151668SYehuda Sadeh mdsc->caps_min_count) { 27537151668SYehuda Sadeh mdsc->caps_total_count--; 276a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 277a8599bd8SSage Weil } else { 27837151668SYehuda Sadeh mdsc->caps_avail_count++; 27937151668SYehuda Sadeh list_add(&cap->caps_item, &mdsc->caps_list); 280a8599bd8SSage Weil } 281a8599bd8SSage Weil 28237151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 28337151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 28437151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 285a8599bd8SSage Weil } 286a8599bd8SSage Weil 2873d14c5d2SYehuda Sadeh void ceph_reservation_status(struct ceph_fs_client *fsc, 28885ccce43SSage Weil int *total, int *avail, int *used, int *reserved, 28985ccce43SSage Weil int *min) 290a8599bd8SSage Weil { 2913d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 29237151668SYehuda Sadeh 293a8599bd8SSage Weil if (total) 29437151668SYehuda Sadeh *total = mdsc->caps_total_count; 295a8599bd8SSage Weil if (avail) 29637151668SYehuda Sadeh *avail = mdsc->caps_avail_count; 297a8599bd8SSage Weil if (used) 29837151668SYehuda Sadeh *used = mdsc->caps_use_count; 299a8599bd8SSage Weil if (reserved) 30037151668SYehuda Sadeh *reserved = mdsc->caps_reserve_count; 30185ccce43SSage Weil if (min) 30237151668SYehuda Sadeh *min = mdsc->caps_min_count; 303a8599bd8SSage Weil } 304a8599bd8SSage Weil 305a8599bd8SSage Weil /* 306a8599bd8SSage Weil * Find ceph_cap for given mds, if any. 307a8599bd8SSage Weil * 308be655596SSage Weil * Called with i_ceph_lock held. 309a8599bd8SSage Weil */ 310a8599bd8SSage Weil static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds) 311a8599bd8SSage Weil { 312a8599bd8SSage Weil struct ceph_cap *cap; 313a8599bd8SSage Weil struct rb_node *n = ci->i_caps.rb_node; 314a8599bd8SSage Weil 315a8599bd8SSage Weil while (n) { 316a8599bd8SSage Weil cap = rb_entry(n, struct ceph_cap, ci_node); 317a8599bd8SSage Weil if (mds < cap->mds) 318a8599bd8SSage Weil n = n->rb_left; 319a8599bd8SSage Weil else if (mds > cap->mds) 320a8599bd8SSage Weil n = n->rb_right; 321a8599bd8SSage Weil else 322a8599bd8SSage Weil return cap; 323a8599bd8SSage Weil } 324a8599bd8SSage Weil return NULL; 325a8599bd8SSage Weil } 326a8599bd8SSage Weil 3272bc50259SGreg Farnum struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds) 3282bc50259SGreg Farnum { 3292bc50259SGreg Farnum struct ceph_cap *cap; 3302bc50259SGreg Farnum 331be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3322bc50259SGreg Farnum cap = __get_cap_for_mds(ci, mds); 333be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3342bc50259SGreg Farnum return cap; 3352bc50259SGreg Farnum } 3362bc50259SGreg Farnum 337a8599bd8SSage Weil /* 33833caad32SSage Weil * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1. 339a8599bd8SSage Weil */ 340ca81f3f6SSage Weil static int __ceph_get_cap_mds(struct ceph_inode_info *ci) 341a8599bd8SSage Weil { 342a8599bd8SSage Weil struct ceph_cap *cap; 343a8599bd8SSage Weil int mds = -1; 344a8599bd8SSage Weil struct rb_node *p; 345a8599bd8SSage Weil 34633caad32SSage Weil /* prefer mds with WR|BUFFER|EXCL caps */ 347a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 348a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 349a8599bd8SSage Weil mds = cap->mds; 350a8599bd8SSage Weil if (cap->issued & (CEPH_CAP_FILE_WR | 351a8599bd8SSage Weil CEPH_CAP_FILE_BUFFER | 352a8599bd8SSage Weil CEPH_CAP_FILE_EXCL)) 353a8599bd8SSage Weil break; 354a8599bd8SSage Weil } 355a8599bd8SSage Weil return mds; 356a8599bd8SSage Weil } 357a8599bd8SSage Weil 358a8599bd8SSage Weil int ceph_get_cap_mds(struct inode *inode) 359a8599bd8SSage Weil { 360be655596SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 361a8599bd8SSage Weil int mds; 362be655596SSage Weil spin_lock(&ci->i_ceph_lock); 363ca81f3f6SSage Weil mds = __ceph_get_cap_mds(ceph_inode(inode)); 364be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 365a8599bd8SSage Weil return mds; 366a8599bd8SSage Weil } 367a8599bd8SSage Weil 368a8599bd8SSage Weil /* 369be655596SSage Weil * Called under i_ceph_lock. 370a8599bd8SSage Weil */ 371a8599bd8SSage Weil static void __insert_cap_node(struct ceph_inode_info *ci, 372a8599bd8SSage Weil struct ceph_cap *new) 373a8599bd8SSage Weil { 374a8599bd8SSage Weil struct rb_node **p = &ci->i_caps.rb_node; 375a8599bd8SSage Weil struct rb_node *parent = NULL; 376a8599bd8SSage Weil struct ceph_cap *cap = NULL; 377a8599bd8SSage Weil 378a8599bd8SSage Weil while (*p) { 379a8599bd8SSage Weil parent = *p; 380a8599bd8SSage Weil cap = rb_entry(parent, struct ceph_cap, ci_node); 381a8599bd8SSage Weil if (new->mds < cap->mds) 382a8599bd8SSage Weil p = &(*p)->rb_left; 383a8599bd8SSage Weil else if (new->mds > cap->mds) 384a8599bd8SSage Weil p = &(*p)->rb_right; 385a8599bd8SSage Weil else 386a8599bd8SSage Weil BUG(); 387a8599bd8SSage Weil } 388a8599bd8SSage Weil 389a8599bd8SSage Weil rb_link_node(&new->ci_node, parent, p); 390a8599bd8SSage Weil rb_insert_color(&new->ci_node, &ci->i_caps); 391a8599bd8SSage Weil } 392a8599bd8SSage Weil 393a8599bd8SSage Weil /* 394a8599bd8SSage Weil * (re)set cap hold timeouts, which control the delayed release 395a8599bd8SSage Weil * of unused caps back to the MDS. Should be called on cap use. 396a8599bd8SSage Weil */ 397a8599bd8SSage Weil static void __cap_set_timeouts(struct ceph_mds_client *mdsc, 398a8599bd8SSage Weil struct ceph_inode_info *ci) 399a8599bd8SSage Weil { 4003d14c5d2SYehuda Sadeh struct ceph_mount_options *ma = mdsc->fsc->mount_options; 401a8599bd8SSage Weil 402a8599bd8SSage Weil ci->i_hold_caps_min = round_jiffies(jiffies + 403a8599bd8SSage Weil ma->caps_wanted_delay_min * HZ); 404a8599bd8SSage Weil ci->i_hold_caps_max = round_jiffies(jiffies + 405a8599bd8SSage Weil ma->caps_wanted_delay_max * HZ); 406a8599bd8SSage Weil dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode, 407a8599bd8SSage Weil ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies); 408a8599bd8SSage Weil } 409a8599bd8SSage Weil 410a8599bd8SSage Weil /* 411a8599bd8SSage Weil * (Re)queue cap at the end of the delayed cap release list. 412a8599bd8SSage Weil * 413a8599bd8SSage Weil * If I_FLUSH is set, leave the inode at the front of the list. 414a8599bd8SSage Weil * 415be655596SSage Weil * Caller holds i_ceph_lock 416a8599bd8SSage Weil * -> we take mdsc->cap_delay_lock 417a8599bd8SSage Weil */ 418a8599bd8SSage Weil static void __cap_delay_requeue(struct ceph_mds_client *mdsc, 419a8599bd8SSage Weil struct ceph_inode_info *ci) 420a8599bd8SSage Weil { 421a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 422a8599bd8SSage Weil dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode, 423a8599bd8SSage Weil ci->i_ceph_flags, ci->i_hold_caps_max); 424a8599bd8SSage Weil if (!mdsc->stopping) { 425a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 426a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) { 427a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 428a8599bd8SSage Weil goto no_change; 429a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 430a8599bd8SSage Weil } 431a8599bd8SSage Weil list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 432a8599bd8SSage Weil no_change: 433a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 434a8599bd8SSage Weil } 435a8599bd8SSage Weil } 436a8599bd8SSage Weil 437a8599bd8SSage Weil /* 438a8599bd8SSage Weil * Queue an inode for immediate writeback. Mark inode with I_FLUSH, 439a8599bd8SSage Weil * indicating we should send a cap message to flush dirty metadata 440a8599bd8SSage Weil * asap, and move to the front of the delayed cap list. 441a8599bd8SSage Weil */ 442a8599bd8SSage Weil static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc, 443a8599bd8SSage Weil struct ceph_inode_info *ci) 444a8599bd8SSage Weil { 445a8599bd8SSage Weil dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode); 446a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 447a8599bd8SSage Weil ci->i_ceph_flags |= CEPH_I_FLUSH; 448a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) 449a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 450a8599bd8SSage Weil list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 451a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 452a8599bd8SSage Weil } 453a8599bd8SSage Weil 454a8599bd8SSage Weil /* 455a8599bd8SSage Weil * Cancel delayed work on cap. 456a8599bd8SSage Weil * 457be655596SSage Weil * Caller must hold i_ceph_lock. 458a8599bd8SSage Weil */ 459a8599bd8SSage Weil static void __cap_delay_cancel(struct ceph_mds_client *mdsc, 460a8599bd8SSage Weil struct ceph_inode_info *ci) 461a8599bd8SSage Weil { 462a8599bd8SSage Weil dout("__cap_delay_cancel %p\n", &ci->vfs_inode); 463a8599bd8SSage Weil if (list_empty(&ci->i_cap_delay_list)) 464a8599bd8SSage Weil return; 465a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 466a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 467a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 468a8599bd8SSage Weil } 469a8599bd8SSage Weil 470a8599bd8SSage Weil /* 471a8599bd8SSage Weil * Common issue checks for add_cap, handle_cap_grant. 472a8599bd8SSage Weil */ 473a8599bd8SSage Weil static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap, 474a8599bd8SSage Weil unsigned issued) 475a8599bd8SSage Weil { 476a8599bd8SSage Weil unsigned had = __ceph_caps_issued(ci, NULL); 477a8599bd8SSage Weil 478a8599bd8SSage Weil /* 479a8599bd8SSage Weil * Each time we receive FILE_CACHE anew, we increment 480a8599bd8SSage Weil * i_rdcache_gen. 481a8599bd8SSage Weil */ 4822962507cSSage Weil if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && 48399ccbd22SMilosz Tanski (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) { 484a8599bd8SSage Weil ci->i_rdcache_gen++; 48599ccbd22SMilosz Tanski } 486a8599bd8SSage Weil 487a8599bd8SSage Weil /* 4882f276c51SYan, Zheng * if we are newly issued FILE_SHARED, mark dir not complete; we 489a8599bd8SSage Weil * don't know what happened to this directory while we didn't 490a8599bd8SSage Weil * have the cap. 491a8599bd8SSage Weil */ 492a8599bd8SSage Weil if ((issued & CEPH_CAP_FILE_SHARED) && 493a8599bd8SSage Weil (had & CEPH_CAP_FILE_SHARED) == 0) { 494a8599bd8SSage Weil ci->i_shared_gen++; 495a8673d61SYan, Zheng if (S_ISDIR(ci->vfs_inode.i_mode)) { 496a8673d61SYan, Zheng dout(" marking %p NOT complete\n", &ci->vfs_inode); 4972f276c51SYan, Zheng __ceph_dir_clear_complete(ci); 498a8673d61SYan, Zheng } 499a8599bd8SSage Weil } 500a8599bd8SSage Weil } 501a8599bd8SSage Weil 502a8599bd8SSage Weil /* 503a8599bd8SSage Weil * Add a capability under the given MDS session. 504a8599bd8SSage Weil * 505a8599bd8SSage Weil * Caller should hold session snap_rwsem (read) and s_mutex. 506a8599bd8SSage Weil * 507a8599bd8SSage Weil * @fmode is the open file mode, if we are opening a file, otherwise 508a8599bd8SSage Weil * it is < 0. (This is so we can atomically add the cap and add an 509a8599bd8SSage Weil * open file reference to it.) 510a8599bd8SSage Weil */ 511d9df2783SYan, Zheng void ceph_add_cap(struct inode *inode, 512a8599bd8SSage Weil struct ceph_mds_session *session, u64 cap_id, 513a8599bd8SSage Weil int fmode, unsigned issued, unsigned wanted, 514a8599bd8SSage Weil unsigned seq, unsigned mseq, u64 realmino, int flags, 515d9df2783SYan, Zheng struct ceph_cap **new_cap) 516a8599bd8SSage Weil { 5173d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 518a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 519a8599bd8SSage Weil struct ceph_cap *cap; 520a8599bd8SSage Weil int mds = session->s_mds; 521a8599bd8SSage Weil int actual_wanted; 522a8599bd8SSage Weil 523a8599bd8SSage Weil dout("add_cap %p mds%d cap %llx %s seq %d\n", inode, 524a8599bd8SSage Weil session->s_mds, cap_id, ceph_cap_string(issued), seq); 525a8599bd8SSage Weil 526a8599bd8SSage Weil /* 527a8599bd8SSage Weil * If we are opening the file, include file mode wanted bits 528a8599bd8SSage Weil * in wanted. 529a8599bd8SSage Weil */ 530a8599bd8SSage Weil if (fmode >= 0) 531a8599bd8SSage Weil wanted |= ceph_caps_for_mode(fmode); 532a8599bd8SSage Weil 533a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 534a8599bd8SSage Weil if (!cap) { 535d9df2783SYan, Zheng cap = *new_cap; 536d9df2783SYan, Zheng *new_cap = NULL; 537a8599bd8SSage Weil 538a8599bd8SSage Weil cap->issued = 0; 539a8599bd8SSage Weil cap->implemented = 0; 540a8599bd8SSage Weil cap->mds = mds; 541a8599bd8SSage Weil cap->mds_wanted = 0; 542964266ccSYan, Zheng cap->mseq = 0; 543a8599bd8SSage Weil 544a8599bd8SSage Weil cap->ci = ci; 545a8599bd8SSage Weil __insert_cap_node(ci, cap); 546a8599bd8SSage Weil 547a8599bd8SSage Weil /* add to session cap list */ 548a8599bd8SSage Weil cap->session = session; 549a8599bd8SSage Weil spin_lock(&session->s_cap_lock); 550a8599bd8SSage Weil list_add_tail(&cap->session_caps, &session->s_caps); 551a8599bd8SSage Weil session->s_nr_caps++; 552a8599bd8SSage Weil spin_unlock(&session->s_cap_lock); 55311df2dfbSYan, Zheng } else { 55411df2dfbSYan, Zheng /* 55511df2dfbSYan, Zheng * auth mds of the inode changed. we received the cap export 55611df2dfbSYan, Zheng * message, but still haven't received the cap import message. 55711df2dfbSYan, Zheng * handle_cap_export() updated the new auth MDS' cap. 55811df2dfbSYan, Zheng * 55911df2dfbSYan, Zheng * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing 56011df2dfbSYan, Zheng * a message that was send before the cap import message. So 56111df2dfbSYan, Zheng * don't remove caps. 56211df2dfbSYan, Zheng */ 56311df2dfbSYan, Zheng if (ceph_seq_cmp(seq, cap->seq) <= 0) { 56411df2dfbSYan, Zheng WARN_ON(cap != ci->i_auth_cap); 56511df2dfbSYan, Zheng WARN_ON(cap->cap_id != cap_id); 56611df2dfbSYan, Zheng seq = cap->seq; 56711df2dfbSYan, Zheng mseq = cap->mseq; 56811df2dfbSYan, Zheng issued |= cap->issued; 56911df2dfbSYan, Zheng flags |= CEPH_CAP_FLAG_AUTH; 57011df2dfbSYan, Zheng } 57111df2dfbSYan, Zheng } 57211df2dfbSYan, Zheng 573a8599bd8SSage Weil if (!ci->i_snap_realm) { 574a8599bd8SSage Weil /* 575a8599bd8SSage Weil * add this inode to the appropriate snap realm 576a8599bd8SSage Weil */ 577a8599bd8SSage Weil struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc, 578a8599bd8SSage Weil realmino); 579a8599bd8SSage Weil if (realm) { 580a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 581a8599bd8SSage Weil ci->i_snap_realm = realm; 582a8599bd8SSage Weil list_add(&ci->i_snap_realm_item, 583a8599bd8SSage Weil &realm->inodes_with_caps); 584a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 585a8599bd8SSage Weil } else { 586a8599bd8SSage Weil pr_err("ceph_add_cap: couldn't find snap realm %llx\n", 587a8599bd8SSage Weil realmino); 588b8cd07e7SSage Weil WARN_ON(!realm); 589a8599bd8SSage Weil } 590a8599bd8SSage Weil } 591a8599bd8SSage Weil 592a8599bd8SSage Weil __check_cap_issue(ci, cap, issued); 593a8599bd8SSage Weil 594a8599bd8SSage Weil /* 595a8599bd8SSage Weil * If we are issued caps we don't want, or the mds' wanted 596a8599bd8SSage Weil * value appears to be off, queue a check so we'll release 597a8599bd8SSage Weil * later and/or update the mds wanted value. 598a8599bd8SSage Weil */ 599a8599bd8SSage Weil actual_wanted = __ceph_caps_wanted(ci); 600a8599bd8SSage Weil if ((wanted & ~actual_wanted) || 601a8599bd8SSage Weil (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) { 602a8599bd8SSage Weil dout(" issued %s, mds wanted %s, actual %s, queueing\n", 603a8599bd8SSage Weil ceph_cap_string(issued), ceph_cap_string(wanted), 604a8599bd8SSage Weil ceph_cap_string(actual_wanted)); 605a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 606a8599bd8SSage Weil } 607a8599bd8SSage Weil 608b8c2f3aeSYan, Zheng if (flags & CEPH_CAP_FLAG_AUTH) { 609b8c2f3aeSYan, Zheng if (ci->i_auth_cap == NULL || 610d9ffc4f7SYan, Zheng ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) { 611a8599bd8SSage Weil ci->i_auth_cap = cap; 612d9ffc4f7SYan, Zheng cap->mds_wanted = wanted; 613d9ffc4f7SYan, Zheng } 61411df2dfbSYan, Zheng } else { 61511df2dfbSYan, Zheng WARN_ON(ci->i_auth_cap == cap); 6168a92a119SYan, Zheng } 617a8599bd8SSage Weil 618a8599bd8SSage Weil dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n", 619a8599bd8SSage Weil inode, ceph_vinop(inode), cap, ceph_cap_string(issued), 620a8599bd8SSage Weil ceph_cap_string(issued|cap->issued), seq, mds); 621a8599bd8SSage Weil cap->cap_id = cap_id; 622a8599bd8SSage Weil cap->issued = issued; 623a8599bd8SSage Weil cap->implemented |= issued; 624d1b87809SYan, Zheng if (ceph_seq_cmp(mseq, cap->mseq) > 0) 625964266ccSYan, Zheng cap->mds_wanted = wanted; 626964266ccSYan, Zheng else 627a8599bd8SSage Weil cap->mds_wanted |= wanted; 628a8599bd8SSage Weil cap->seq = seq; 629a8599bd8SSage Weil cap->issue_seq = seq; 630a8599bd8SSage Weil cap->mseq = mseq; 631685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 632a8599bd8SSage Weil 633a8599bd8SSage Weil if (fmode >= 0) 634a8599bd8SSage Weil __ceph_get_fmode(ci, fmode); 635a8599bd8SSage Weil } 636a8599bd8SSage Weil 637a8599bd8SSage Weil /* 638a8599bd8SSage Weil * Return true if cap has not timed out and belongs to the current 639a8599bd8SSage Weil * generation of the MDS session (i.e. has not gone 'stale' due to 640a8599bd8SSage Weil * us losing touch with the mds). 641a8599bd8SSage Weil */ 642a8599bd8SSage Weil static int __cap_is_valid(struct ceph_cap *cap) 643a8599bd8SSage Weil { 644a8599bd8SSage Weil unsigned long ttl; 645cdac8303SSage Weil u32 gen; 646a8599bd8SSage Weil 647d8fb02abSAlex Elder spin_lock(&cap->session->s_gen_ttl_lock); 648a8599bd8SSage Weil gen = cap->session->s_cap_gen; 649a8599bd8SSage Weil ttl = cap->session->s_cap_ttl; 650d8fb02abSAlex Elder spin_unlock(&cap->session->s_gen_ttl_lock); 651a8599bd8SSage Weil 652685f9a5dSSage Weil if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) { 653a8599bd8SSage Weil dout("__cap_is_valid %p cap %p issued %s " 654a8599bd8SSage Weil "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode, 655685f9a5dSSage Weil cap, ceph_cap_string(cap->issued), cap->cap_gen, gen); 656a8599bd8SSage Weil return 0; 657a8599bd8SSage Weil } 658a8599bd8SSage Weil 659a8599bd8SSage Weil return 1; 660a8599bd8SSage Weil } 661a8599bd8SSage Weil 662a8599bd8SSage Weil /* 663a8599bd8SSage Weil * Return set of valid cap bits issued to us. Note that caps time 664a8599bd8SSage Weil * out, and may be invalidated in bulk if the client session times out 665a8599bd8SSage Weil * and session->s_cap_gen is bumped. 666a8599bd8SSage Weil */ 667a8599bd8SSage Weil int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented) 668a8599bd8SSage Weil { 669d9df2783SYan, Zheng int have = ci->i_snap_caps; 670a8599bd8SSage Weil struct ceph_cap *cap; 671a8599bd8SSage Weil struct rb_node *p; 672a8599bd8SSage Weil 673a8599bd8SSage Weil if (implemented) 674a8599bd8SSage Weil *implemented = 0; 675a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 676a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 677a8599bd8SSage Weil if (!__cap_is_valid(cap)) 678a8599bd8SSage Weil continue; 679a8599bd8SSage Weil dout("__ceph_caps_issued %p cap %p issued %s\n", 680a8599bd8SSage Weil &ci->vfs_inode, cap, ceph_cap_string(cap->issued)); 681a8599bd8SSage Weil have |= cap->issued; 682a8599bd8SSage Weil if (implemented) 683a8599bd8SSage Weil *implemented |= cap->implemented; 684a8599bd8SSage Weil } 685b1530f57SYan, Zheng /* 686b1530f57SYan, Zheng * exclude caps issued by non-auth MDS, but are been revoking 687b1530f57SYan, Zheng * by the auth MDS. The non-auth MDS should be revoking/exporting 688b1530f57SYan, Zheng * these caps, but the message is delayed. 689b1530f57SYan, Zheng */ 690b1530f57SYan, Zheng if (ci->i_auth_cap) { 691b1530f57SYan, Zheng cap = ci->i_auth_cap; 692b1530f57SYan, Zheng have &= ~cap->implemented | cap->issued; 693b1530f57SYan, Zheng } 694a8599bd8SSage Weil return have; 695a8599bd8SSage Weil } 696a8599bd8SSage Weil 697a8599bd8SSage Weil /* 698a8599bd8SSage Weil * Get cap bits issued by caps other than @ocap 699a8599bd8SSage Weil */ 700a8599bd8SSage Weil int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap) 701a8599bd8SSage Weil { 702a8599bd8SSage Weil int have = ci->i_snap_caps; 703a8599bd8SSage Weil struct ceph_cap *cap; 704a8599bd8SSage Weil struct rb_node *p; 705a8599bd8SSage Weil 706a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 707a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 708a8599bd8SSage Weil if (cap == ocap) 709a8599bd8SSage Weil continue; 710a8599bd8SSage Weil if (!__cap_is_valid(cap)) 711a8599bd8SSage Weil continue; 712a8599bd8SSage Weil have |= cap->issued; 713a8599bd8SSage Weil } 714a8599bd8SSage Weil return have; 715a8599bd8SSage Weil } 716a8599bd8SSage Weil 717a8599bd8SSage Weil /* 718a8599bd8SSage Weil * Move a cap to the end of the LRU (oldest caps at list head, newest 719a8599bd8SSage Weil * at list tail). 720a8599bd8SSage Weil */ 721a8599bd8SSage Weil static void __touch_cap(struct ceph_cap *cap) 722a8599bd8SSage Weil { 723a8599bd8SSage Weil struct ceph_mds_session *s = cap->session; 724a8599bd8SSage Weil 7255dacf091SSage Weil spin_lock(&s->s_cap_lock); 7267c1332b8SSage Weil if (s->s_cap_iterator == NULL) { 727a8599bd8SSage Weil dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap, 728a8599bd8SSage Weil s->s_mds); 729a8599bd8SSage Weil list_move_tail(&cap->session_caps, &s->s_caps); 7305dacf091SSage Weil } else { 7315dacf091SSage Weil dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n", 7325dacf091SSage Weil &cap->ci->vfs_inode, cap, s->s_mds); 7335dacf091SSage Weil } 734a8599bd8SSage Weil spin_unlock(&s->s_cap_lock); 735a8599bd8SSage Weil } 736a8599bd8SSage Weil 737a8599bd8SSage Weil /* 738a8599bd8SSage Weil * Check if we hold the given mask. If so, move the cap(s) to the 739a8599bd8SSage Weil * front of their respective LRUs. (This is the preferred way for 740a8599bd8SSage Weil * callers to check for caps they want.) 741a8599bd8SSage Weil */ 742a8599bd8SSage Weil int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) 743a8599bd8SSage Weil { 744a8599bd8SSage Weil struct ceph_cap *cap; 745a8599bd8SSage Weil struct rb_node *p; 746a8599bd8SSage Weil int have = ci->i_snap_caps; 747a8599bd8SSage Weil 748a8599bd8SSage Weil if ((have & mask) == mask) { 749a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p snap issued %s" 750a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 751a8599bd8SSage Weil ceph_cap_string(have), 752a8599bd8SSage Weil ceph_cap_string(mask)); 753a8599bd8SSage Weil return 1; 754a8599bd8SSage Weil } 755a8599bd8SSage Weil 756a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 757a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 758a8599bd8SSage Weil if (!__cap_is_valid(cap)) 759a8599bd8SSage Weil continue; 760a8599bd8SSage Weil if ((cap->issued & mask) == mask) { 761a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p cap %p issued %s" 762a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, cap, 763a8599bd8SSage Weil ceph_cap_string(cap->issued), 764a8599bd8SSage Weil ceph_cap_string(mask)); 765a8599bd8SSage Weil if (touch) 766a8599bd8SSage Weil __touch_cap(cap); 767a8599bd8SSage Weil return 1; 768a8599bd8SSage Weil } 769a8599bd8SSage Weil 770a8599bd8SSage Weil /* does a combination of caps satisfy mask? */ 771a8599bd8SSage Weil have |= cap->issued; 772a8599bd8SSage Weil if ((have & mask) == mask) { 773a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p combo issued %s" 774a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 775a8599bd8SSage Weil ceph_cap_string(cap->issued), 776a8599bd8SSage Weil ceph_cap_string(mask)); 777a8599bd8SSage Weil if (touch) { 778a8599bd8SSage Weil struct rb_node *q; 779a8599bd8SSage Weil 78025985edcSLucas De Marchi /* touch this + preceding caps */ 781a8599bd8SSage Weil __touch_cap(cap); 782a8599bd8SSage Weil for (q = rb_first(&ci->i_caps); q != p; 783a8599bd8SSage Weil q = rb_next(q)) { 784a8599bd8SSage Weil cap = rb_entry(q, struct ceph_cap, 785a8599bd8SSage Weil ci_node); 786a8599bd8SSage Weil if (!__cap_is_valid(cap)) 787a8599bd8SSage Weil continue; 788a8599bd8SSage Weil __touch_cap(cap); 789a8599bd8SSage Weil } 790a8599bd8SSage Weil } 791a8599bd8SSage Weil return 1; 792a8599bd8SSage Weil } 793a8599bd8SSage Weil } 794a8599bd8SSage Weil 795a8599bd8SSage Weil return 0; 796a8599bd8SSage Weil } 797a8599bd8SSage Weil 798a8599bd8SSage Weil /* 799a8599bd8SSage Weil * Return true if mask caps are currently being revoked by an MDS. 800a8599bd8SSage Weil */ 8016ee6b953SYan, Zheng int __ceph_caps_revoking_other(struct ceph_inode_info *ci, 8026ee6b953SYan, Zheng struct ceph_cap *ocap, int mask) 8036ee6b953SYan, Zheng { 8046ee6b953SYan, Zheng struct ceph_cap *cap; 8056ee6b953SYan, Zheng struct rb_node *p; 8066ee6b953SYan, Zheng 8076ee6b953SYan, Zheng for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 8086ee6b953SYan, Zheng cap = rb_entry(p, struct ceph_cap, ci_node); 8099563f88cSYan, Zheng if (cap != ocap && 8106ee6b953SYan, Zheng (cap->implemented & ~cap->issued & mask)) 8116ee6b953SYan, Zheng return 1; 8126ee6b953SYan, Zheng } 8136ee6b953SYan, Zheng return 0; 8146ee6b953SYan, Zheng } 8156ee6b953SYan, Zheng 816a8599bd8SSage Weil int ceph_caps_revoking(struct ceph_inode_info *ci, int mask) 817a8599bd8SSage Weil { 818a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 8196ee6b953SYan, Zheng int ret; 820a8599bd8SSage Weil 821be655596SSage Weil spin_lock(&ci->i_ceph_lock); 8226ee6b953SYan, Zheng ret = __ceph_caps_revoking_other(ci, NULL, mask); 823be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 824a8599bd8SSage Weil dout("ceph_caps_revoking %p %s = %d\n", inode, 825a8599bd8SSage Weil ceph_cap_string(mask), ret); 826a8599bd8SSage Weil return ret; 827a8599bd8SSage Weil } 828a8599bd8SSage Weil 829a8599bd8SSage Weil int __ceph_caps_used(struct ceph_inode_info *ci) 830a8599bd8SSage Weil { 831a8599bd8SSage Weil int used = 0; 832a8599bd8SSage Weil if (ci->i_pin_ref) 833a8599bd8SSage Weil used |= CEPH_CAP_PIN; 834a8599bd8SSage Weil if (ci->i_rd_ref) 835a8599bd8SSage Weil used |= CEPH_CAP_FILE_RD; 836a43fb731SSage Weil if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages) 837a8599bd8SSage Weil used |= CEPH_CAP_FILE_CACHE; 838a8599bd8SSage Weil if (ci->i_wr_ref) 839a8599bd8SSage Weil used |= CEPH_CAP_FILE_WR; 840d3d0720dSHenry C Chang if (ci->i_wb_ref || ci->i_wrbuffer_ref) 841a8599bd8SSage Weil used |= CEPH_CAP_FILE_BUFFER; 842a8599bd8SSage Weil return used; 843a8599bd8SSage Weil } 844a8599bd8SSage Weil 845a8599bd8SSage Weil /* 846a8599bd8SSage Weil * wanted, by virtue of open file modes 847a8599bd8SSage Weil */ 848a8599bd8SSage Weil int __ceph_caps_file_wanted(struct ceph_inode_info *ci) 849a8599bd8SSage Weil { 850a8599bd8SSage Weil int want = 0; 851a8599bd8SSage Weil int mode; 85233caad32SSage Weil for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++) 853a8599bd8SSage Weil if (ci->i_nr_by_mode[mode]) 854a8599bd8SSage Weil want |= ceph_caps_for_mode(mode); 855a8599bd8SSage Weil return want; 856a8599bd8SSage Weil } 857a8599bd8SSage Weil 858a8599bd8SSage Weil /* 859a8599bd8SSage Weil * Return caps we have registered with the MDS(s) as 'wanted'. 860a8599bd8SSage Weil */ 861a8599bd8SSage Weil int __ceph_caps_mds_wanted(struct ceph_inode_info *ci) 862a8599bd8SSage Weil { 863a8599bd8SSage Weil struct ceph_cap *cap; 864a8599bd8SSage Weil struct rb_node *p; 865a8599bd8SSage Weil int mds_wanted = 0; 866a8599bd8SSage Weil 867a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 868a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 869a8599bd8SSage Weil if (!__cap_is_valid(cap)) 870a8599bd8SSage Weil continue; 871a2550604SYan, Zheng if (cap == ci->i_auth_cap) 872a8599bd8SSage Weil mds_wanted |= cap->mds_wanted; 873a2550604SYan, Zheng else 874a2550604SYan, Zheng mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR); 875a8599bd8SSage Weil } 876a8599bd8SSage Weil return mds_wanted; 877a8599bd8SSage Weil } 878a8599bd8SSage Weil 879a8599bd8SSage Weil /* 880be655596SSage Weil * called under i_ceph_lock 881a8599bd8SSage Weil */ 882a8599bd8SSage Weil static int __ceph_is_any_caps(struct ceph_inode_info *ci) 883a8599bd8SSage Weil { 884d9df2783SYan, Zheng return !RB_EMPTY_ROOT(&ci->i_caps); 885a8599bd8SSage Weil } 886a8599bd8SSage Weil 8879215aeeaSYan, Zheng int ceph_is_any_caps(struct inode *inode) 8889215aeeaSYan, Zheng { 8899215aeeaSYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 8909215aeeaSYan, Zheng int ret; 8919215aeeaSYan, Zheng 8929215aeeaSYan, Zheng spin_lock(&ci->i_ceph_lock); 8939215aeeaSYan, Zheng ret = __ceph_is_any_caps(ci); 8949215aeeaSYan, Zheng spin_unlock(&ci->i_ceph_lock); 8959215aeeaSYan, Zheng 8969215aeeaSYan, Zheng return ret; 8979215aeeaSYan, Zheng } 8989215aeeaSYan, Zheng 899a8599bd8SSage Weil /* 900f818a736SSage Weil * Remove a cap. Take steps to deal with a racing iterate_session_caps. 901f818a736SSage Weil * 902be655596SSage Weil * caller should hold i_ceph_lock. 903a6369741SSage Weil * caller will not hold session s_mutex if called from destroy_inode. 904a8599bd8SSage Weil */ 905a096b09aSYan, Zheng void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) 906a8599bd8SSage Weil { 907a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 908a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 909640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 9103d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 911f818a736SSage Weil int removed = 0; 912a8599bd8SSage Weil 913a8599bd8SSage Weil dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); 914a8599bd8SSage Weil 9157c1332b8SSage Weil /* remove from session list */ 9167c1332b8SSage Weil spin_lock(&session->s_cap_lock); 91799a9c273SYan, Zheng /* 91899a9c273SYan, Zheng * s_cap_reconnect is protected by s_cap_lock. no one changes 91999a9c273SYan, Zheng * s_cap_gen while session is in the reconnect state. 92099a9c273SYan, Zheng */ 92199a9c273SYan, Zheng if (queue_release && 92299a9c273SYan, Zheng (!session->s_cap_reconnect || 92399a9c273SYan, Zheng cap->cap_gen == session->s_cap_gen)) 924a096b09aSYan, Zheng __queue_cap_release(session, ci->i_vino.ino, cap->cap_id, 925a096b09aSYan, Zheng cap->mseq, cap->issue_seq); 926a096b09aSYan, Zheng 9277c1332b8SSage Weil if (session->s_cap_iterator == cap) { 9287c1332b8SSage Weil /* not yet, we are iterating over this very cap */ 9297c1332b8SSage Weil dout("__ceph_remove_cap delaying %p removal from session %p\n", 9307c1332b8SSage Weil cap, cap->session); 9317c1332b8SSage Weil } else { 9327c1332b8SSage Weil list_del_init(&cap->session_caps); 9337c1332b8SSage Weil session->s_nr_caps--; 9347c1332b8SSage Weil cap->session = NULL; 935f818a736SSage Weil removed = 1; 9367c1332b8SSage Weil } 937f818a736SSage Weil /* protect backpointer with s_cap_lock: see iterate_session_caps */ 938f818a736SSage Weil cap->ci = NULL; 9397c1332b8SSage Weil spin_unlock(&session->s_cap_lock); 9407c1332b8SSage Weil 941f818a736SSage Weil /* remove from inode list */ 942f818a736SSage Weil rb_erase(&cap->ci_node, &ci->i_caps); 943f818a736SSage Weil if (ci->i_auth_cap == cap) 944f818a736SSage Weil ci->i_auth_cap = NULL; 945f818a736SSage Weil 946f818a736SSage Weil if (removed) 94737151668SYehuda Sadeh ceph_put_cap(mdsc, cap); 948a8599bd8SSage Weil 949a8599bd8SSage Weil if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) { 950a8599bd8SSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 951a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 952a8599bd8SSage Weil list_del_init(&ci->i_snap_realm_item); 953a8599bd8SSage Weil ci->i_snap_realm_counter++; 954a8599bd8SSage Weil ci->i_snap_realm = NULL; 955a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 956a8599bd8SSage Weil ceph_put_snap_realm(mdsc, realm); 957a8599bd8SSage Weil } 958a8599bd8SSage Weil if (!__ceph_is_any_real_caps(ci)) 959a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 960a8599bd8SSage Weil } 961a8599bd8SSage Weil 962a8599bd8SSage Weil /* 963a8599bd8SSage Weil * Build and send a cap message to the given MDS. 964a8599bd8SSage Weil * 965a8599bd8SSage Weil * Caller should be holding s_mutex. 966a8599bd8SSage Weil */ 967a8599bd8SSage Weil static int send_cap_msg(struct ceph_mds_session *session, 968a8599bd8SSage Weil u64 ino, u64 cid, int op, 969a8599bd8SSage Weil int caps, int wanted, int dirty, 970a8599bd8SSage Weil u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq, 971a8599bd8SSage Weil u64 size, u64 max_size, 972a8599bd8SSage Weil struct timespec *mtime, struct timespec *atime, 973a8599bd8SSage Weil u64 time_warp_seq, 97405cb11c1SEric W. Biederman kuid_t uid, kgid_t gid, umode_t mode, 975a8599bd8SSage Weil u64 xattr_version, 976a8599bd8SSage Weil struct ceph_buffer *xattrs_buf, 977e20d258dSYan, Zheng u64 follows, bool inline_data) 978a8599bd8SSage Weil { 979a8599bd8SSage Weil struct ceph_mds_caps *fc; 980a8599bd8SSage Weil struct ceph_msg *msg; 981e20d258dSYan, Zheng void *p; 982e20d258dSYan, Zheng size_t extra_len; 983a8599bd8SSage Weil 984a8599bd8SSage Weil dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" 985a8599bd8SSage Weil " seq %u/%u mseq %u follows %lld size %llu/%llu" 986a8599bd8SSage Weil " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op), 987a8599bd8SSage Weil cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted), 988a8599bd8SSage Weil ceph_cap_string(dirty), 989a8599bd8SSage Weil seq, issue_seq, mseq, follows, size, max_size, 990a8599bd8SSage Weil xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0); 991a8599bd8SSage Weil 992e20d258dSYan, Zheng /* flock buffer size + inline version + inline data size */ 993e20d258dSYan, Zheng extra_len = 4 + 8 + 4; 994e20d258dSYan, Zheng msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len, 995e20d258dSYan, Zheng GFP_NOFS, false); 996a79832f2SSage Weil if (!msg) 997a79832f2SSage Weil return -ENOMEM; 998a8599bd8SSage Weil 9996df058c0SSage Weil msg->hdr.tid = cpu_to_le64(flush_tid); 1000a8599bd8SSage Weil 10016df058c0SSage Weil fc = msg->front.iov_base; 1002a8599bd8SSage Weil memset(fc, 0, sizeof(*fc)); 1003a8599bd8SSage Weil 1004a8599bd8SSage Weil fc->cap_id = cpu_to_le64(cid); 1005a8599bd8SSage Weil fc->op = cpu_to_le32(op); 1006a8599bd8SSage Weil fc->seq = cpu_to_le32(seq); 1007a8599bd8SSage Weil fc->issue_seq = cpu_to_le32(issue_seq); 1008a8599bd8SSage Weil fc->migrate_seq = cpu_to_le32(mseq); 1009a8599bd8SSage Weil fc->caps = cpu_to_le32(caps); 1010a8599bd8SSage Weil fc->wanted = cpu_to_le32(wanted); 1011a8599bd8SSage Weil fc->dirty = cpu_to_le32(dirty); 1012a8599bd8SSage Weil fc->ino = cpu_to_le64(ino); 1013a8599bd8SSage Weil fc->snap_follows = cpu_to_le64(follows); 1014a8599bd8SSage Weil 1015a8599bd8SSage Weil fc->size = cpu_to_le64(size); 1016a8599bd8SSage Weil fc->max_size = cpu_to_le64(max_size); 1017a8599bd8SSage Weil if (mtime) 1018a8599bd8SSage Weil ceph_encode_timespec(&fc->mtime, mtime); 1019a8599bd8SSage Weil if (atime) 1020a8599bd8SSage Weil ceph_encode_timespec(&fc->atime, atime); 1021a8599bd8SSage Weil fc->time_warp_seq = cpu_to_le32(time_warp_seq); 1022a8599bd8SSage Weil 102305cb11c1SEric W. Biederman fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid)); 102405cb11c1SEric W. Biederman fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid)); 1025a8599bd8SSage Weil fc->mode = cpu_to_le32(mode); 1026a8599bd8SSage Weil 1027e20d258dSYan, Zheng p = fc + 1; 1028e20d258dSYan, Zheng /* flock buffer size */ 1029e20d258dSYan, Zheng ceph_encode_32(&p, 0); 1030e20d258dSYan, Zheng /* inline version */ 1031e20d258dSYan, Zheng ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE); 1032e20d258dSYan, Zheng /* inline data size */ 1033e20d258dSYan, Zheng ceph_encode_32(&p, 0); 1034e20d258dSYan, Zheng 1035a8599bd8SSage Weil fc->xattr_version = cpu_to_le64(xattr_version); 1036a8599bd8SSage Weil if (xattrs_buf) { 1037a8599bd8SSage Weil msg->middle = ceph_buffer_get(xattrs_buf); 1038a8599bd8SSage Weil fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len); 1039a8599bd8SSage Weil msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len); 1040a8599bd8SSage Weil } 1041a8599bd8SSage Weil 1042a8599bd8SSage Weil ceph_con_send(&session->s_con, msg); 1043a8599bd8SSage Weil return 0; 1044a8599bd8SSage Weil } 1045a8599bd8SSage Weil 1046d40ee0dcSYan, Zheng void __queue_cap_release(struct ceph_mds_session *session, 10473d7ded4dSSage Weil u64 ino, u64 cap_id, u32 migrate_seq, 10483d7ded4dSSage Weil u32 issue_seq) 10493d7ded4dSSage Weil { 10503d7ded4dSSage Weil struct ceph_msg *msg; 10513d7ded4dSSage Weil struct ceph_mds_cap_release *head; 10523d7ded4dSSage Weil struct ceph_mds_cap_item *item; 10533d7ded4dSSage Weil 10543d7ded4dSSage Weil BUG_ON(!session->s_num_cap_releases); 10553d7ded4dSSage Weil msg = list_first_entry(&session->s_cap_releases, 10563d7ded4dSSage Weil struct ceph_msg, list_head); 10573d7ded4dSSage Weil 10583d7ded4dSSage Weil dout(" adding %llx release to mds%d msg %p (%d left)\n", 10593d7ded4dSSage Weil ino, session->s_mds, msg, session->s_num_cap_releases); 10603d7ded4dSSage Weil 10613d7ded4dSSage Weil BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE); 10623d7ded4dSSage Weil head = msg->front.iov_base; 1063b905a7f8SWei Yongjun le32_add_cpu(&head->num, 1); 10643d7ded4dSSage Weil item = msg->front.iov_base + msg->front.iov_len; 10653d7ded4dSSage Weil item->ino = cpu_to_le64(ino); 10663d7ded4dSSage Weil item->cap_id = cpu_to_le64(cap_id); 10673d7ded4dSSage Weil item->migrate_seq = cpu_to_le32(migrate_seq); 10683d7ded4dSSage Weil item->seq = cpu_to_le32(issue_seq); 10693d7ded4dSSage Weil 10703d7ded4dSSage Weil session->s_num_cap_releases--; 10713d7ded4dSSage Weil 10723d7ded4dSSage Weil msg->front.iov_len += sizeof(*item); 10733d7ded4dSSage Weil if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) { 10743d7ded4dSSage Weil dout(" release msg %p full\n", msg); 10753d7ded4dSSage Weil list_move_tail(&msg->list_head, &session->s_cap_releases_done); 10763d7ded4dSSage Weil } else { 10773d7ded4dSSage Weil dout(" release msg %p at %d/%d (%d)\n", msg, 10783d7ded4dSSage Weil (int)le32_to_cpu(head->num), 10793d7ded4dSSage Weil (int)CEPH_CAPS_PER_RELEASE, 10803d7ded4dSSage Weil (int)msg->front.iov_len); 10813d7ded4dSSage Weil } 10823d7ded4dSSage Weil } 10833d7ded4dSSage Weil 1084a8599bd8SSage Weil /* 1085a6369741SSage Weil * Queue cap releases when an inode is dropped from our cache. Since 1086be655596SSage Weil * inode is about to be destroyed, there is no need for i_ceph_lock. 1087a8599bd8SSage Weil */ 1088a8599bd8SSage Weil void ceph_queue_caps_release(struct inode *inode) 1089a8599bd8SSage Weil { 1090a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1091a8599bd8SSage Weil struct rb_node *p; 1092a8599bd8SSage Weil 1093a8599bd8SSage Weil p = rb_first(&ci->i_caps); 1094a8599bd8SSage Weil while (p) { 1095a8599bd8SSage Weil struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); 1096a8599bd8SSage Weil p = rb_next(p); 1097a096b09aSYan, Zheng __ceph_remove_cap(cap, true); 1098a8599bd8SSage Weil } 1099a8599bd8SSage Weil } 1100a8599bd8SSage Weil 1101a8599bd8SSage Weil /* 1102a8599bd8SSage Weil * Send a cap msg on the given inode. Update our caps state, then 1103be655596SSage Weil * drop i_ceph_lock and send the message. 1104a8599bd8SSage Weil * 1105a8599bd8SSage Weil * Make note of max_size reported/requested from mds, revoked caps 1106a8599bd8SSage Weil * that have now been implemented. 1107a8599bd8SSage Weil * 1108a8599bd8SSage Weil * Make half-hearted attempt ot to invalidate page cache if we are 1109a8599bd8SSage Weil * dropping RDCACHE. Note that this will leave behind locked pages 1110a8599bd8SSage Weil * that we'll then need to deal with elsewhere. 1111a8599bd8SSage Weil * 1112a8599bd8SSage Weil * Return non-zero if delayed release, or we experienced an error 1113a8599bd8SSage Weil * such that the caller should requeue + retry later. 1114a8599bd8SSage Weil * 1115be655596SSage Weil * called with i_ceph_lock, then drops it. 1116a8599bd8SSage Weil * caller should hold snap_rwsem (read), s_mutex. 1117a8599bd8SSage Weil */ 1118a8599bd8SSage Weil static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, 1119a8599bd8SSage Weil int op, int used, int want, int retain, int flushing, 1120a8599bd8SSage Weil unsigned *pflush_tid) 1121be655596SSage Weil __releases(cap->ci->i_ceph_lock) 1122a8599bd8SSage Weil { 1123a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 1124a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1125a8599bd8SSage Weil u64 cap_id = cap->cap_id; 112668c28323SSage Weil int held, revoking, dropping, keep; 1127a8599bd8SSage Weil u64 seq, issue_seq, mseq, time_warp_seq, follows; 1128a8599bd8SSage Weil u64 size, max_size; 1129a8599bd8SSage Weil struct timespec mtime, atime; 1130a8599bd8SSage Weil int wake = 0; 11315706b27dSAl Viro umode_t mode; 113205cb11c1SEric W. Biederman kuid_t uid; 113305cb11c1SEric W. Biederman kgid_t gid; 1134a8599bd8SSage Weil struct ceph_mds_session *session; 1135a8599bd8SSage Weil u64 xattr_version = 0; 1136082afec9SSage Weil struct ceph_buffer *xattr_blob = NULL; 1137a8599bd8SSage Weil int delayed = 0; 1138a8599bd8SSage Weil u64 flush_tid = 0; 1139a8599bd8SSage Weil int i; 1140a8599bd8SSage Weil int ret; 1141e20d258dSYan, Zheng bool inline_data; 1142a8599bd8SSage Weil 114368c28323SSage Weil held = cap->issued | cap->implemented; 114468c28323SSage Weil revoking = cap->implemented & ~cap->issued; 114568c28323SSage Weil retain &= ~revoking; 114668c28323SSage Weil dropping = cap->issued & ~retain; 114768c28323SSage Weil 1148a8599bd8SSage Weil dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", 1149a8599bd8SSage Weil inode, cap, cap->session, 1150a8599bd8SSage Weil ceph_cap_string(held), ceph_cap_string(held & retain), 1151a8599bd8SSage Weil ceph_cap_string(revoking)); 1152a8599bd8SSage Weil BUG_ON((retain & CEPH_CAP_PIN) == 0); 1153a8599bd8SSage Weil 1154a8599bd8SSage Weil session = cap->session; 1155a8599bd8SSage Weil 1156a8599bd8SSage Weil /* don't release wanted unless we've waited a bit. */ 1157a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1158a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_min)) { 1159a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", 1160a8599bd8SSage Weil ceph_cap_string(cap->issued), 1161a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1162a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1163a8599bd8SSage Weil ceph_cap_string(want)); 1164a8599bd8SSage Weil want |= cap->mds_wanted; 1165a8599bd8SSage Weil retain |= cap->issued; 1166a8599bd8SSage Weil delayed = 1; 1167a8599bd8SSage Weil } 1168a8599bd8SSage Weil ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); 1169a8599bd8SSage Weil 1170a8599bd8SSage Weil cap->issued &= retain; /* drop bits we don't want */ 1171a8599bd8SSage Weil if (cap->implemented & ~cap->issued) { 1172a8599bd8SSage Weil /* 1173a8599bd8SSage Weil * Wake up any waiters on wanted -> needed transition. 1174a8599bd8SSage Weil * This is due to the weird transition from buffered 1175a8599bd8SSage Weil * to sync IO... we need to flush dirty pages _before_ 1176a8599bd8SSage Weil * allowing sync writes to avoid reordering. 1177a8599bd8SSage Weil */ 1178a8599bd8SSage Weil wake = 1; 1179a8599bd8SSage Weil } 1180a8599bd8SSage Weil cap->implemented &= cap->issued | used; 1181a8599bd8SSage Weil cap->mds_wanted = want; 1182a8599bd8SSage Weil 1183a8599bd8SSage Weil if (flushing) { 1184a8599bd8SSage Weil /* 1185a8599bd8SSage Weil * assign a tid for flush operations so we can avoid 1186a8599bd8SSage Weil * flush1 -> dirty1 -> flush2 -> flushack1 -> mark 1187a8599bd8SSage Weil * clean type races. track latest tid for every bit 1188a8599bd8SSage Weil * so we can handle flush AxFw, flush Fw, and have the 1189a8599bd8SSage Weil * first ack clean Ax. 1190a8599bd8SSage Weil */ 1191a8599bd8SSage Weil flush_tid = ++ci->i_cap_flush_last_tid; 1192a8599bd8SSage Weil if (pflush_tid) 1193a8599bd8SSage Weil *pflush_tid = flush_tid; 1194a8599bd8SSage Weil dout(" cap_flush_tid %d\n", (int)flush_tid); 1195a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1196a8599bd8SSage Weil if (flushing & (1 << i)) 1197a8599bd8SSage Weil ci->i_cap_flush_tid[i] = flush_tid; 11987d8cb26dSSage Weil 11997d8cb26dSSage Weil follows = ci->i_head_snapc->seq; 12007d8cb26dSSage Weil } else { 12017d8cb26dSSage Weil follows = 0; 1202a8599bd8SSage Weil } 1203a8599bd8SSage Weil 1204a8599bd8SSage Weil keep = cap->implemented; 1205a8599bd8SSage Weil seq = cap->seq; 1206a8599bd8SSage Weil issue_seq = cap->issue_seq; 1207a8599bd8SSage Weil mseq = cap->mseq; 1208a8599bd8SSage Weil size = inode->i_size; 1209a8599bd8SSage Weil ci->i_reported_size = size; 1210a8599bd8SSage Weil max_size = ci->i_wanted_max_size; 1211a8599bd8SSage Weil ci->i_requested_max_size = max_size; 1212a8599bd8SSage Weil mtime = inode->i_mtime; 1213a8599bd8SSage Weil atime = inode->i_atime; 1214a8599bd8SSage Weil time_warp_seq = ci->i_time_warp_seq; 1215a8599bd8SSage Weil uid = inode->i_uid; 1216a8599bd8SSage Weil gid = inode->i_gid; 1217a8599bd8SSage Weil mode = inode->i_mode; 1218a8599bd8SSage Weil 1219082afec9SSage Weil if (flushing & CEPH_CAP_XATTR_EXCL) { 1220a8599bd8SSage Weil __ceph_build_xattrs_blob(ci); 1221082afec9SSage Weil xattr_blob = ci->i_xattrs.blob; 1222082afec9SSage Weil xattr_version = ci->i_xattrs.version; 1223a8599bd8SSage Weil } 1224a8599bd8SSage Weil 1225e20d258dSYan, Zheng inline_data = ci->i_inline_version != CEPH_INLINE_NONE; 1226e20d258dSYan, Zheng 1227be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1228a8599bd8SSage Weil 1229a8599bd8SSage Weil ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id, 1230a8599bd8SSage Weil op, keep, want, flushing, seq, flush_tid, issue_seq, mseq, 1231a8599bd8SSage Weil size, max_size, &mtime, &atime, time_warp_seq, 1232082afec9SSage Weil uid, gid, mode, xattr_version, xattr_blob, 1233e20d258dSYan, Zheng follows, inline_data); 1234a8599bd8SSage Weil if (ret < 0) { 1235a8599bd8SSage Weil dout("error sending cap msg, must requeue %p\n", inode); 1236a8599bd8SSage Weil delayed = 1; 1237a8599bd8SSage Weil } 1238a8599bd8SSage Weil 1239a8599bd8SSage Weil if (wake) 124003066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1241a8599bd8SSage Weil 1242a8599bd8SSage Weil return delayed; 1243a8599bd8SSage Weil } 1244a8599bd8SSage Weil 1245a8599bd8SSage Weil /* 1246a8599bd8SSage Weil * When a snapshot is taken, clients accumulate dirty metadata on 1247a8599bd8SSage Weil * inodes with capabilities in ceph_cap_snaps to describe the file 1248a8599bd8SSage Weil * state at the time the snapshot was taken. This must be flushed 1249a8599bd8SSage Weil * asynchronously back to the MDS once sync writes complete and dirty 1250a8599bd8SSage Weil * data is written out. 1251a8599bd8SSage Weil * 1252e835124cSSage Weil * Unless @again is true, skip cap_snaps that were already sent to 1253e835124cSSage Weil * the MDS (i.e., during this session). 1254e835124cSSage Weil * 1255be655596SSage Weil * Called under i_ceph_lock. Takes s_mutex as needed. 1256a8599bd8SSage Weil */ 1257a8599bd8SSage Weil void __ceph_flush_snaps(struct ceph_inode_info *ci, 1258e835124cSSage Weil struct ceph_mds_session **psession, 1259e835124cSSage Weil int again) 1260be655596SSage Weil __releases(ci->i_ceph_lock) 1261be655596SSage Weil __acquires(ci->i_ceph_lock) 1262a8599bd8SSage Weil { 1263a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1264a8599bd8SSage Weil int mds; 1265a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1266a8599bd8SSage Weil u32 mseq; 12673d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 1268a8599bd8SSage Weil struct ceph_mds_session *session = NULL; /* if session != NULL, we hold 1269a8599bd8SSage Weil session->s_mutex */ 1270a8599bd8SSage Weil u64 next_follows = 0; /* keep track of how far we've gotten through the 1271a8599bd8SSage Weil i_cap_snaps list, and skip these entries next time 1272a8599bd8SSage Weil around to avoid an infinite loop */ 1273a8599bd8SSage Weil 1274a8599bd8SSage Weil if (psession) 1275a8599bd8SSage Weil session = *psession; 1276a8599bd8SSage Weil 1277a8599bd8SSage Weil dout("__flush_snaps %p\n", inode); 1278a8599bd8SSage Weil retry: 1279a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 1280a8599bd8SSage Weil /* avoid an infiniute loop after retry */ 1281a8599bd8SSage Weil if (capsnap->follows < next_follows) 1282a8599bd8SSage Weil continue; 1283a8599bd8SSage Weil /* 1284a8599bd8SSage Weil * we need to wait for sync writes to complete and for dirty 1285a8599bd8SSage Weil * pages to be written out. 1286a8599bd8SSage Weil */ 1287a8599bd8SSage Weil if (capsnap->dirty_pages || capsnap->writing) 1288cfc0bf66SSage Weil break; 1289a8599bd8SSage Weil 1290819ccbfaSSage Weil /* 1291819ccbfaSSage Weil * if cap writeback already occurred, we should have dropped 1292819ccbfaSSage Weil * the capsnap in ceph_put_wrbuffer_cap_refs. 1293819ccbfaSSage Weil */ 1294819ccbfaSSage Weil BUG_ON(capsnap->dirty == 0); 1295819ccbfaSSage Weil 1296a8599bd8SSage Weil /* pick mds, take s_mutex */ 1297ca81f3f6SSage Weil if (ci->i_auth_cap == NULL) { 1298ca81f3f6SSage Weil dout("no auth cap (migrating?), doing nothing\n"); 1299ca81f3f6SSage Weil goto out; 1300ca81f3f6SSage Weil } 1301e835124cSSage Weil 1302e835124cSSage Weil /* only flush each capsnap once */ 1303e835124cSSage Weil if (!again && !list_empty(&capsnap->flushing_item)) { 1304e835124cSSage Weil dout("already flushed %p, skipping\n", capsnap); 1305e835124cSSage Weil continue; 1306e835124cSSage Weil } 1307e835124cSSage Weil 1308ca81f3f6SSage Weil mds = ci->i_auth_cap->session->s_mds; 1309ca81f3f6SSage Weil mseq = ci->i_auth_cap->mseq; 1310ca81f3f6SSage Weil 1311a8599bd8SSage Weil if (session && session->s_mds != mds) { 1312a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1313a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1314a8599bd8SSage Weil ceph_put_mds_session(session); 1315a8599bd8SSage Weil session = NULL; 1316a8599bd8SSage Weil } 1317a8599bd8SSage Weil if (!session) { 1318be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1319a8599bd8SSage Weil mutex_lock(&mdsc->mutex); 1320a8599bd8SSage Weil session = __ceph_lookup_mds_session(mdsc, mds); 1321a8599bd8SSage Weil mutex_unlock(&mdsc->mutex); 1322a8599bd8SSage Weil if (session) { 1323a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1324a8599bd8SSage Weil session); 1325a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1326a8599bd8SSage Weil } 1327a8599bd8SSage Weil /* 1328a8599bd8SSage Weil * if session == NULL, we raced against a cap 1329ca81f3f6SSage Weil * deletion or migration. retry, and we'll 1330ca81f3f6SSage Weil * get a better @mds value next time. 1331a8599bd8SSage Weil */ 1332be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1333a8599bd8SSage Weil goto retry; 1334a8599bd8SSage Weil } 1335a8599bd8SSage Weil 1336a8599bd8SSage Weil capsnap->flush_tid = ++ci->i_cap_flush_last_tid; 1337a8599bd8SSage Weil atomic_inc(&capsnap->nref); 1338a8599bd8SSage Weil if (!list_empty(&capsnap->flushing_item)) 1339a8599bd8SSage Weil list_del_init(&capsnap->flushing_item); 1340a8599bd8SSage Weil list_add_tail(&capsnap->flushing_item, 1341a8599bd8SSage Weil &session->s_cap_snaps_flushing); 1342be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1343a8599bd8SSage Weil 1344cfc0bf66SSage Weil dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n", 1345cfc0bf66SSage Weil inode, capsnap, capsnap->follows, capsnap->flush_tid); 1346a8599bd8SSage Weil send_cap_msg(session, ceph_vino(inode).ino, 0, 1347a8599bd8SSage Weil CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0, 1348a8599bd8SSage Weil capsnap->dirty, 0, capsnap->flush_tid, 0, mseq, 1349a8599bd8SSage Weil capsnap->size, 0, 1350a8599bd8SSage Weil &capsnap->mtime, &capsnap->atime, 1351a8599bd8SSage Weil capsnap->time_warp_seq, 1352a8599bd8SSage Weil capsnap->uid, capsnap->gid, capsnap->mode, 13534a625be4SSage Weil capsnap->xattr_version, capsnap->xattr_blob, 1354e20d258dSYan, Zheng capsnap->follows, capsnap->inline_data); 1355a8599bd8SSage Weil 1356a8599bd8SSage Weil next_follows = capsnap->follows + 1; 1357a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 1358a8599bd8SSage Weil 1359be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1360a8599bd8SSage Weil goto retry; 1361a8599bd8SSage Weil } 1362a8599bd8SSage Weil 1363a8599bd8SSage Weil /* we flushed them all; remove this inode from the queue */ 1364a8599bd8SSage Weil spin_lock(&mdsc->snap_flush_lock); 1365a8599bd8SSage Weil list_del_init(&ci->i_snap_flush_item); 1366a8599bd8SSage Weil spin_unlock(&mdsc->snap_flush_lock); 1367a8599bd8SSage Weil 1368ca81f3f6SSage Weil out: 1369a8599bd8SSage Weil if (psession) 1370a8599bd8SSage Weil *psession = session; 1371a8599bd8SSage Weil else if (session) { 1372a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1373a8599bd8SSage Weil ceph_put_mds_session(session); 1374a8599bd8SSage Weil } 1375a8599bd8SSage Weil } 1376a8599bd8SSage Weil 1377a8599bd8SSage Weil static void ceph_flush_snaps(struct ceph_inode_info *ci) 1378a8599bd8SSage Weil { 1379be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1380e835124cSSage Weil __ceph_flush_snaps(ci, NULL, 0); 1381be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1382a8599bd8SSage Weil } 1383a8599bd8SSage Weil 1384a8599bd8SSage Weil /* 1385fca65b4aSSage Weil * Mark caps dirty. If inode is newly dirty, return the dirty flags. 1386fca65b4aSSage Weil * Caller is then responsible for calling __mark_inode_dirty with the 1387fca65b4aSSage Weil * returned flags value. 138876e3b390SSage Weil */ 1389fca65b4aSSage Weil int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask) 139076e3b390SSage Weil { 1391640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 13923d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 139376e3b390SSage Weil struct inode *inode = &ci->vfs_inode; 139476e3b390SSage Weil int was = ci->i_dirty_caps; 139576e3b390SSage Weil int dirty = 0; 139676e3b390SSage Weil 139776e3b390SSage Weil dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, 139876e3b390SSage Weil ceph_cap_string(mask), ceph_cap_string(was), 139976e3b390SSage Weil ceph_cap_string(was | mask)); 140076e3b390SSage Weil ci->i_dirty_caps |= mask; 140176e3b390SSage Weil if (was == 0) { 14027d8cb26dSSage Weil if (!ci->i_head_snapc) 14037d8cb26dSSage Weil ci->i_head_snapc = ceph_get_snap_context( 14047d8cb26dSSage Weil ci->i_snap_realm->cached_context); 14050685235fSYan, Zheng dout(" inode %p now dirty snapc %p auth cap %p\n", 14060685235fSYan, Zheng &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap); 140711df2dfbSYan, Zheng WARN_ON(!ci->i_auth_cap); 140876e3b390SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 140976e3b390SSage Weil spin_lock(&mdsc->cap_dirty_lock); 141076e3b390SSage Weil list_add(&ci->i_dirty_item, &mdsc->cap_dirty); 141176e3b390SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 141276e3b390SSage Weil if (ci->i_flushing_caps == 0) { 14133772d26dSSage Weil ihold(inode); 141476e3b390SSage Weil dirty |= I_DIRTY_SYNC; 141576e3b390SSage Weil } 141676e3b390SSage Weil } 141776e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 141876e3b390SSage Weil if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && 141976e3b390SSage Weil (mask & CEPH_CAP_FILE_BUFFER)) 142076e3b390SSage Weil dirty |= I_DIRTY_DATASYNC; 142176e3b390SSage Weil __cap_delay_requeue(mdsc, ci); 1422fca65b4aSSage Weil return dirty; 142376e3b390SSage Weil } 142476e3b390SSage Weil 142576e3b390SSage Weil /* 1426a8599bd8SSage Weil * Add dirty inode to the flushing list. Assigned a seq number so we 1427a8599bd8SSage Weil * can wait for caps to flush without starving. 1428cdc35f96SSage Weil * 1429be655596SSage Weil * Called under i_ceph_lock. 1430a8599bd8SSage Weil */ 1431cdc35f96SSage Weil static int __mark_caps_flushing(struct inode *inode, 1432a8599bd8SSage Weil struct ceph_mds_session *session) 1433a8599bd8SSage Weil { 14343d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1435a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1436cdc35f96SSage Weil int flushing; 1437a8599bd8SSage Weil 1438cdc35f96SSage Weil BUG_ON(ci->i_dirty_caps == 0); 1439a8599bd8SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 1440cdc35f96SSage Weil 1441cdc35f96SSage Weil flushing = ci->i_dirty_caps; 1442cdc35f96SSage Weil dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", 1443cdc35f96SSage Weil ceph_cap_string(flushing), 1444cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps), 1445cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps | flushing)); 1446cdc35f96SSage Weil ci->i_flushing_caps |= flushing; 1447cdc35f96SSage Weil ci->i_dirty_caps = 0; 1448afcdaea3SSage Weil dout(" inode %p now !dirty\n", inode); 1449cdc35f96SSage Weil 1450a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 1451cdc35f96SSage Weil list_del_init(&ci->i_dirty_item); 1452afcdaea3SSage Weil 1453afcdaea3SSage Weil if (list_empty(&ci->i_flushing_item)) { 1454d3383a8eSYan, Zheng ci->i_cap_flush_seq = ++mdsc->cap_flush_seq; 1455a8599bd8SSage Weil list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1456a8599bd8SSage Weil mdsc->num_cap_flushing++; 1457afcdaea3SSage Weil dout(" inode %p now flushing seq %lld\n", inode, 1458afcdaea3SSage Weil ci->i_cap_flush_seq); 1459afcdaea3SSage Weil } else { 1460afcdaea3SSage Weil list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1461afcdaea3SSage Weil dout(" inode %p now flushing (more) seq %lld\n", inode, 1462a8599bd8SSage Weil ci->i_cap_flush_seq); 1463a8599bd8SSage Weil } 1464a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 1465cdc35f96SSage Weil 1466cdc35f96SSage Weil return flushing; 1467a8599bd8SSage Weil } 1468a8599bd8SSage Weil 1469a8599bd8SSage Weil /* 14705ecad6fdSSage Weil * try to invalidate mapping pages without blocking. 14715ecad6fdSSage Weil */ 14725ecad6fdSSage Weil static int try_nonblocking_invalidate(struct inode *inode) 14735ecad6fdSSage Weil { 14745ecad6fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14755ecad6fdSSage Weil u32 invalidating_gen = ci->i_rdcache_gen; 14765ecad6fdSSage Weil 1477be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 14785ecad6fdSSage Weil invalidate_mapping_pages(&inode->i_data, 0, -1); 1479be655596SSage Weil spin_lock(&ci->i_ceph_lock); 14805ecad6fdSSage Weil 148118a38193SSage Weil if (inode->i_data.nrpages == 0 && 14825ecad6fdSSage Weil invalidating_gen == ci->i_rdcache_gen) { 14835ecad6fdSSage Weil /* success. */ 14845ecad6fdSSage Weil dout("try_nonblocking_invalidate %p success\n", inode); 1485cd045cb4SSage Weil /* save any racing async invalidate some trouble */ 1486cd045cb4SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen - 1; 14875ecad6fdSSage Weil return 0; 14885ecad6fdSSage Weil } 14895ecad6fdSSage Weil dout("try_nonblocking_invalidate %p failed\n", inode); 14905ecad6fdSSage Weil return -1; 14915ecad6fdSSage Weil } 14925ecad6fdSSage Weil 14935ecad6fdSSage Weil /* 1494a8599bd8SSage Weil * Swiss army knife function to examine currently used and wanted 1495a8599bd8SSage Weil * versus held caps. Release, flush, ack revoked caps to mds as 1496a8599bd8SSage Weil * appropriate. 1497a8599bd8SSage Weil * 1498a8599bd8SSage Weil * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay 1499a8599bd8SSage Weil * cap release further. 1500a8599bd8SSage Weil * CHECK_CAPS_AUTHONLY - we should only check the auth cap 1501a8599bd8SSage Weil * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without 1502a8599bd8SSage Weil * further delay. 1503a8599bd8SSage Weil */ 1504a8599bd8SSage Weil void ceph_check_caps(struct ceph_inode_info *ci, int flags, 1505a8599bd8SSage Weil struct ceph_mds_session *session) 1506a8599bd8SSage Weil { 15073d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 15083d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1509a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1510a8599bd8SSage Weil struct ceph_cap *cap; 1511395c312bSYan, Zheng int file_wanted, used, cap_used; 1512a8599bd8SSage Weil int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ 1513cbd03635SSage Weil int issued, implemented, want, retain, revoking, flushing = 0; 1514a8599bd8SSage Weil int mds = -1; /* keep track of how far we've gone through i_caps list 1515a8599bd8SSage Weil to avoid an infinite loop on retry */ 1516a8599bd8SSage Weil struct rb_node *p; 1517a8599bd8SSage Weil int tried_invalidate = 0; 1518a8599bd8SSage Weil int delayed = 0, sent = 0, force_requeue = 0, num; 1519cbd03635SSage Weil int queue_invalidate = 0; 1520a8599bd8SSage Weil int is_delayed = flags & CHECK_CAPS_NODELAY; 1521a8599bd8SSage Weil 1522a8599bd8SSage Weil /* if we are unmounting, flush any unused caps immediately. */ 1523a8599bd8SSage Weil if (mdsc->stopping) 1524a8599bd8SSage Weil is_delayed = 1; 1525a8599bd8SSage Weil 1526be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1527a8599bd8SSage Weil 1528a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 1529a8599bd8SSage Weil flags |= CHECK_CAPS_FLUSH; 1530a8599bd8SSage Weil 1531a8599bd8SSage Weil /* flush snaps first time around only */ 1532a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) 1533e835124cSSage Weil __ceph_flush_snaps(ci, &session, 0); 1534a8599bd8SSage Weil goto retry_locked; 1535a8599bd8SSage Weil retry: 1536be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1537a8599bd8SSage Weil retry_locked: 1538a8599bd8SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 1539a8599bd8SSage Weil used = __ceph_caps_used(ci); 1540a8599bd8SSage Weil want = file_wanted | used; 1541cbd03635SSage Weil issued = __ceph_caps_issued(ci, &implemented); 1542cbd03635SSage Weil revoking = implemented & ~issued; 1543a8599bd8SSage Weil 1544a8599bd8SSage Weil retain = want | CEPH_CAP_PIN; 1545a8599bd8SSage Weil if (!mdsc->stopping && inode->i_nlink > 0) { 1546a8599bd8SSage Weil if (want) { 1547a8599bd8SSage Weil retain |= CEPH_CAP_ANY; /* be greedy */ 1548a8599bd8SSage Weil } else { 1549a8599bd8SSage Weil retain |= CEPH_CAP_ANY_SHARED; 1550a8599bd8SSage Weil /* 1551a8599bd8SSage Weil * keep RD only if we didn't have the file open RW, 1552a8599bd8SSage Weil * because then the mds would revoke it anyway to 1553a8599bd8SSage Weil * journal max_size=0. 1554a8599bd8SSage Weil */ 1555a8599bd8SSage Weil if (ci->i_max_size == 0) 1556a8599bd8SSage Weil retain |= CEPH_CAP_ANY_RD; 1557a8599bd8SSage Weil } 1558a8599bd8SSage Weil } 1559a8599bd8SSage Weil 1560a8599bd8SSage Weil dout("check_caps %p file_want %s used %s dirty %s flushing %s" 1561cbd03635SSage Weil " issued %s revoking %s retain %s %s%s%s\n", inode, 1562a8599bd8SSage Weil ceph_cap_string(file_wanted), 1563a8599bd8SSage Weil ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), 1564a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps), 1565cbd03635SSage Weil ceph_cap_string(issued), ceph_cap_string(revoking), 1566a8599bd8SSage Weil ceph_cap_string(retain), 1567a8599bd8SSage Weil (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", 1568a8599bd8SSage Weil (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", 1569a8599bd8SSage Weil (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); 1570a8599bd8SSage Weil 1571a8599bd8SSage Weil /* 1572a8599bd8SSage Weil * If we no longer need to hold onto old our caps, and we may 1573a8599bd8SSage Weil * have cached pages, but don't want them, then try to invalidate. 1574a8599bd8SSage Weil * If we fail, it's because pages are locked.... try again later. 1575a8599bd8SSage Weil */ 1576a8599bd8SSage Weil if ((!is_delayed || mdsc->stopping) && 1577a8599bd8SSage Weil ci->i_wrbuffer_ref == 0 && /* no dirty pages... */ 157893afd449SSage Weil inode->i_data.nrpages && /* have cached pages */ 1579cbd03635SSage Weil (file_wanted == 0 || /* no open files */ 15802962507cSSage Weil (revoking & (CEPH_CAP_FILE_CACHE| 15812962507cSSage Weil CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */ 1582a8599bd8SSage Weil !tried_invalidate) { 1583a8599bd8SSage Weil dout("check_caps trying to invalidate on %p\n", inode); 15845ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) < 0) { 15852962507cSSage Weil if (revoking & (CEPH_CAP_FILE_CACHE| 15862962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) { 1587cbd03635SSage Weil dout("check_caps queuing invalidate\n"); 1588cbd03635SSage Weil queue_invalidate = 1; 1589cbd03635SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 1590a8599bd8SSage Weil } else { 1591a8599bd8SSage Weil dout("check_caps failed to invalidate pages\n"); 1592a8599bd8SSage Weil /* we failed to invalidate pages. check these 1593a8599bd8SSage Weil caps again later. */ 1594a8599bd8SSage Weil force_requeue = 1; 1595a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 1596a8599bd8SSage Weil } 15975ecad6fdSSage Weil } 1598a8599bd8SSage Weil tried_invalidate = 1; 1599a8599bd8SSage Weil goto retry_locked; 1600a8599bd8SSage Weil } 1601a8599bd8SSage Weil 1602a8599bd8SSage Weil num = 0; 1603a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 1604a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 1605a8599bd8SSage Weil num++; 1606a8599bd8SSage Weil 1607a8599bd8SSage Weil /* avoid looping forever */ 1608a8599bd8SSage Weil if (mds >= cap->mds || 1609a8599bd8SSage Weil ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) 1610a8599bd8SSage Weil continue; 1611a8599bd8SSage Weil 1612a8599bd8SSage Weil /* NOTE: no side-effects allowed, until we take s_mutex */ 1613a8599bd8SSage Weil 1614395c312bSYan, Zheng cap_used = used; 1615395c312bSYan, Zheng if (ci->i_auth_cap && cap != ci->i_auth_cap) 1616395c312bSYan, Zheng cap_used &= ~ci->i_auth_cap->issued; 1617395c312bSYan, Zheng 1618a8599bd8SSage Weil revoking = cap->implemented & ~cap->issued; 1619395c312bSYan, Zheng dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n", 1620088b3f5eSSage Weil cap->mds, cap, ceph_cap_string(cap->issued), 1621395c312bSYan, Zheng ceph_cap_string(cap_used), 1622088b3f5eSSage Weil ceph_cap_string(cap->implemented), 1623a8599bd8SSage Weil ceph_cap_string(revoking)); 1624a8599bd8SSage Weil 1625a8599bd8SSage Weil if (cap == ci->i_auth_cap && 1626a8599bd8SSage Weil (cap->issued & CEPH_CAP_FILE_WR)) { 1627a8599bd8SSage Weil /* request larger max_size from MDS? */ 1628a8599bd8SSage Weil if (ci->i_wanted_max_size > ci->i_max_size && 1629a8599bd8SSage Weil ci->i_wanted_max_size > ci->i_requested_max_size) { 1630a8599bd8SSage Weil dout("requesting new max_size\n"); 1631a8599bd8SSage Weil goto ack; 1632a8599bd8SSage Weil } 1633a8599bd8SSage Weil 1634a8599bd8SSage Weil /* approaching file_max? */ 1635a8599bd8SSage Weil if ((inode->i_size << 1) >= ci->i_max_size && 1636a8599bd8SSage Weil (ci->i_reported_size << 1) < ci->i_max_size) { 1637a8599bd8SSage Weil dout("i_size approaching max_size\n"); 1638a8599bd8SSage Weil goto ack; 1639a8599bd8SSage Weil } 1640a8599bd8SSage Weil } 1641a8599bd8SSage Weil /* flush anything dirty? */ 1642a8599bd8SSage Weil if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) && 1643a8599bd8SSage Weil ci->i_dirty_caps) { 1644a8599bd8SSage Weil dout("flushing dirty caps\n"); 1645a8599bd8SSage Weil goto ack; 1646a8599bd8SSage Weil } 1647a8599bd8SSage Weil 1648a8599bd8SSage Weil /* completed revocation? going down and there are no caps? */ 1649395c312bSYan, Zheng if (revoking && (revoking & cap_used) == 0) { 1650a8599bd8SSage Weil dout("completed revocation of %s\n", 1651a8599bd8SSage Weil ceph_cap_string(cap->implemented & ~cap->issued)); 1652a8599bd8SSage Weil goto ack; 1653a8599bd8SSage Weil } 1654a8599bd8SSage Weil 1655a8599bd8SSage Weil /* want more caps from mds? */ 1656a8599bd8SSage Weil if (want & ~(cap->mds_wanted | cap->issued)) 1657a8599bd8SSage Weil goto ack; 1658a8599bd8SSage Weil 1659a8599bd8SSage Weil /* things we might delay */ 1660a8599bd8SSage Weil if ((cap->issued & ~retain) == 0 && 1661a8599bd8SSage Weil cap->mds_wanted == want) 1662a8599bd8SSage Weil continue; /* nope, all good */ 1663a8599bd8SSage Weil 1664a8599bd8SSage Weil if (is_delayed) 1665a8599bd8SSage Weil goto ack; 1666a8599bd8SSage Weil 1667a8599bd8SSage Weil /* delay? */ 1668a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1669a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) { 1670a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s\n", 1671a8599bd8SSage Weil ceph_cap_string(cap->issued), 1672a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1673a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1674a8599bd8SSage Weil ceph_cap_string(want)); 1675a8599bd8SSage Weil delayed++; 1676a8599bd8SSage Weil continue; 1677a8599bd8SSage Weil } 1678a8599bd8SSage Weil 1679a8599bd8SSage Weil ack: 1680e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1681e9964c10SSage Weil dout(" skipping %p I_NOFLUSH set\n", inode); 1682e9964c10SSage Weil continue; 1683e9964c10SSage Weil } 1684e9964c10SSage Weil 1685a8599bd8SSage Weil if (session && session != cap->session) { 1686a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1687a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1688a8599bd8SSage Weil session = NULL; 1689a8599bd8SSage Weil } 1690a8599bd8SSage Weil if (!session) { 1691a8599bd8SSage Weil session = cap->session; 1692a8599bd8SSage Weil if (mutex_trylock(&session->s_mutex) == 0) { 1693a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1694a8599bd8SSage Weil session); 1695be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1696a8599bd8SSage Weil if (took_snap_rwsem) { 1697a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1698a8599bd8SSage Weil took_snap_rwsem = 0; 1699a8599bd8SSage Weil } 1700a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1701a8599bd8SSage Weil goto retry; 1702a8599bd8SSage Weil } 1703a8599bd8SSage Weil } 1704a8599bd8SSage Weil /* take snap_rwsem after session mutex */ 1705a8599bd8SSage Weil if (!took_snap_rwsem) { 1706a8599bd8SSage Weil if (down_read_trylock(&mdsc->snap_rwsem) == 0) { 1707a8599bd8SSage Weil dout("inverting snap/in locks on %p\n", 1708a8599bd8SSage Weil inode); 1709be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1710a8599bd8SSage Weil down_read(&mdsc->snap_rwsem); 1711a8599bd8SSage Weil took_snap_rwsem = 1; 1712a8599bd8SSage Weil goto retry; 1713a8599bd8SSage Weil } 1714a8599bd8SSage Weil took_snap_rwsem = 1; 1715a8599bd8SSage Weil } 1716a8599bd8SSage Weil 1717cdc35f96SSage Weil if (cap == ci->i_auth_cap && ci->i_dirty_caps) 1718cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 171924be0c48SSage Weil else 172024be0c48SSage Weil flushing = 0; 1721a8599bd8SSage Weil 1722a8599bd8SSage Weil mds = cap->mds; /* remember mds, so we don't repeat */ 1723a8599bd8SSage Weil sent++; 1724a8599bd8SSage Weil 1725be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1726395c312bSYan, Zheng delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used, 1727395c312bSYan, Zheng want, retain, flushing, NULL); 1728be655596SSage Weil goto retry; /* retake i_ceph_lock and restart our cap scan. */ 1729a8599bd8SSage Weil } 1730a8599bd8SSage Weil 1731a8599bd8SSage Weil /* 1732a8599bd8SSage Weil * Reschedule delayed caps release if we delayed anything, 1733a8599bd8SSage Weil * otherwise cancel. 1734a8599bd8SSage Weil */ 1735a8599bd8SSage Weil if (delayed && is_delayed) 1736a8599bd8SSage Weil force_requeue = 1; /* __send_cap delayed release; requeue */ 1737a8599bd8SSage Weil if (!delayed && !is_delayed) 1738a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 1739a8599bd8SSage Weil else if (!is_delayed || force_requeue) 1740a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1741a8599bd8SSage Weil 1742be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1743a8599bd8SSage Weil 1744cbd03635SSage Weil if (queue_invalidate) 17453c6f6b79SSage Weil ceph_queue_invalidate(inode); 1746cbd03635SSage Weil 1747cdc2ce05SSage Weil if (session) 1748a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1749a8599bd8SSage Weil if (took_snap_rwsem) 1750a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1751a8599bd8SSage Weil } 1752a8599bd8SSage Weil 1753a8599bd8SSage Weil /* 1754a8599bd8SSage Weil * Try to flush dirty caps back to the auth mds. 1755a8599bd8SSage Weil */ 17564fe59789SYan, Zheng static int try_flush_caps(struct inode *inode, unsigned *flush_tid) 1757a8599bd8SSage Weil { 17583d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1759a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1760a8599bd8SSage Weil int flushing = 0; 17614fe59789SYan, Zheng struct ceph_mds_session *session = NULL; 1762a8599bd8SSage Weil 1763a8599bd8SSage Weil retry: 1764be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1765e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1766e9964c10SSage Weil dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode); 1767e9964c10SSage Weil goto out; 1768e9964c10SSage Weil } 1769a8599bd8SSage Weil if (ci->i_dirty_caps && ci->i_auth_cap) { 1770a8599bd8SSage Weil struct ceph_cap *cap = ci->i_auth_cap; 1771a8599bd8SSage Weil int used = __ceph_caps_used(ci); 1772a8599bd8SSage Weil int want = __ceph_caps_wanted(ci); 1773a8599bd8SSage Weil int delayed; 1774a8599bd8SSage Weil 17754fe59789SYan, Zheng if (!session || session != cap->session) { 1776be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 17774fe59789SYan, Zheng if (session) 17784fe59789SYan, Zheng mutex_unlock(&session->s_mutex); 1779a8599bd8SSage Weil session = cap->session; 1780a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1781a8599bd8SSage Weil goto retry; 1782a8599bd8SSage Weil } 1783a8599bd8SSage Weil if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) 1784a8599bd8SSage Weil goto out; 1785a8599bd8SSage Weil 1786cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 1787a8599bd8SSage Weil 1788be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1789a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want, 1790a8599bd8SSage Weil cap->issued | cap->implemented, flushing, 1791a8599bd8SSage Weil flush_tid); 1792a8599bd8SSage Weil if (!delayed) 1793a8599bd8SSage Weil goto out_unlocked; 1794a8599bd8SSage Weil 1795be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1796a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1797a8599bd8SSage Weil } 1798a8599bd8SSage Weil out: 1799be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1800a8599bd8SSage Weil out_unlocked: 18014fe59789SYan, Zheng if (session) 1802a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1803a8599bd8SSage Weil return flushing; 1804a8599bd8SSage Weil } 1805a8599bd8SSage Weil 1806a8599bd8SSage Weil /* 1807a8599bd8SSage Weil * Return true if we've flushed caps through the given flush_tid. 1808a8599bd8SSage Weil */ 1809a8599bd8SSage Weil static int caps_are_flushed(struct inode *inode, unsigned tid) 1810a8599bd8SSage Weil { 1811a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1812a5ee751cSDan Carpenter int i, ret = 1; 1813a8599bd8SSage Weil 1814be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1815a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1816a8599bd8SSage Weil if ((ci->i_flushing_caps & (1 << i)) && 1817a8599bd8SSage Weil ci->i_cap_flush_tid[i] <= tid) { 1818a8599bd8SSage Weil /* still flushing this bit */ 1819a8599bd8SSage Weil ret = 0; 1820a8599bd8SSage Weil break; 1821a8599bd8SSage Weil } 1822be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1823a8599bd8SSage Weil return ret; 1824a8599bd8SSage Weil } 1825a8599bd8SSage Weil 1826a8599bd8SSage Weil /* 1827a8599bd8SSage Weil * Wait on any unsafe replies for the given inode. First wait on the 1828a8599bd8SSage Weil * newest request, and make that the upper bound. Then, if there are 1829a8599bd8SSage Weil * more requests, keep waiting on the oldest as long as it is still older 1830a8599bd8SSage Weil * than the original request. 1831a8599bd8SSage Weil */ 1832a8599bd8SSage Weil static void sync_write_wait(struct inode *inode) 1833a8599bd8SSage Weil { 1834a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1835a8599bd8SSage Weil struct list_head *head = &ci->i_unsafe_writes; 1836a8599bd8SSage Weil struct ceph_osd_request *req; 1837a8599bd8SSage Weil u64 last_tid; 1838a8599bd8SSage Weil 1839a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1840a8599bd8SSage Weil if (list_empty(head)) 1841a8599bd8SSage Weil goto out; 1842a8599bd8SSage Weil 1843a8599bd8SSage Weil /* set upper bound as _last_ entry in chain */ 1844a8599bd8SSage Weil req = list_entry(head->prev, struct ceph_osd_request, 1845a8599bd8SSage Weil r_unsafe_item); 1846a8599bd8SSage Weil last_tid = req->r_tid; 1847a8599bd8SSage Weil 1848a8599bd8SSage Weil do { 1849a8599bd8SSage Weil ceph_osdc_get_request(req); 1850a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1851a8599bd8SSage Weil dout("sync_write_wait on tid %llu (until %llu)\n", 1852a8599bd8SSage Weil req->r_tid, last_tid); 1853a8599bd8SSage Weil wait_for_completion(&req->r_safe_completion); 1854a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1855a8599bd8SSage Weil ceph_osdc_put_request(req); 1856a8599bd8SSage Weil 1857a8599bd8SSage Weil /* 1858a8599bd8SSage Weil * from here on look at first entry in chain, since we 1859a8599bd8SSage Weil * only want to wait for anything older than last_tid 1860a8599bd8SSage Weil */ 1861a8599bd8SSage Weil if (list_empty(head)) 1862a8599bd8SSage Weil break; 1863a8599bd8SSage Weil req = list_entry(head->next, struct ceph_osd_request, 1864a8599bd8SSage Weil r_unsafe_item); 1865a8599bd8SSage Weil } while (req->r_tid < last_tid); 1866a8599bd8SSage Weil out: 1867a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1868a8599bd8SSage Weil } 1869a8599bd8SSage Weil 187002c24a82SJosef Bacik int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync) 1871a8599bd8SSage Weil { 18727ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 1873a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1874a8599bd8SSage Weil unsigned flush_tid; 1875a8599bd8SSage Weil int ret; 1876a8599bd8SSage Weil int dirty; 1877a8599bd8SSage Weil 1878a8599bd8SSage Weil dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); 1879a8599bd8SSage Weil sync_write_wait(inode); 1880a8599bd8SSage Weil 188102c24a82SJosef Bacik ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 1882a8599bd8SSage Weil if (ret < 0) 1883a8599bd8SSage Weil return ret; 188402c24a82SJosef Bacik mutex_lock(&inode->i_mutex); 1885a8599bd8SSage Weil 18864fe59789SYan, Zheng dirty = try_flush_caps(inode, &flush_tid); 1887a8599bd8SSage Weil dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); 1888a8599bd8SSage Weil 1889a8599bd8SSage Weil /* 1890a8599bd8SSage Weil * only wait on non-file metadata writeback (the mds 1891a8599bd8SSage Weil * can recover size and mtime, so we don't need to 1892a8599bd8SSage Weil * wait for that) 1893a8599bd8SSage Weil */ 1894a8599bd8SSage Weil if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { 1895a8599bd8SSage Weil dout("fsync waiting for flush_tid %u\n", flush_tid); 1896a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 1897a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1898a8599bd8SSage Weil } 1899a8599bd8SSage Weil 1900a8599bd8SSage Weil dout("fsync %p%s done\n", inode, datasync ? " datasync" : ""); 190102c24a82SJosef Bacik mutex_unlock(&inode->i_mutex); 1902a8599bd8SSage Weil return ret; 1903a8599bd8SSage Weil } 1904a8599bd8SSage Weil 1905a8599bd8SSage Weil /* 1906a8599bd8SSage Weil * Flush any dirty caps back to the mds. If we aren't asked to wait, 1907a8599bd8SSage Weil * queue inode for flush but don't do so immediately, because we can 1908a8599bd8SSage Weil * get by with fewer MDS messages if we wait for data writeback to 1909a8599bd8SSage Weil * complete first. 1910a8599bd8SSage Weil */ 1911f1a3d572SStephen Rothwell int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) 1912a8599bd8SSage Weil { 1913a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1914a8599bd8SSage Weil unsigned flush_tid; 1915a8599bd8SSage Weil int err = 0; 1916a8599bd8SSage Weil int dirty; 1917f1a3d572SStephen Rothwell int wait = wbc->sync_mode == WB_SYNC_ALL; 1918a8599bd8SSage Weil 1919a8599bd8SSage Weil dout("write_inode %p wait=%d\n", inode, wait); 1920a8599bd8SSage Weil if (wait) { 19214fe59789SYan, Zheng dirty = try_flush_caps(inode, &flush_tid); 1922a8599bd8SSage Weil if (dirty) 1923a8599bd8SSage Weil err = wait_event_interruptible(ci->i_cap_wq, 1924a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1925a8599bd8SSage Weil } else { 1926640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 19273d14c5d2SYehuda Sadeh ceph_sb_to_client(inode->i_sb)->mdsc; 1928a8599bd8SSage Weil 1929be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1930a8599bd8SSage Weil if (__ceph_caps_dirty(ci)) 1931a8599bd8SSage Weil __cap_delay_requeue_front(mdsc, ci); 1932be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1933a8599bd8SSage Weil } 1934a8599bd8SSage Weil return err; 1935a8599bd8SSage Weil } 1936a8599bd8SSage Weil 1937a8599bd8SSage Weil /* 1938a8599bd8SSage Weil * After a recovering MDS goes active, we need to resend any caps 1939a8599bd8SSage Weil * we were flushing. 1940a8599bd8SSage Weil * 1941a8599bd8SSage Weil * Caller holds session->s_mutex. 1942a8599bd8SSage Weil */ 1943a8599bd8SSage Weil static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc, 1944a8599bd8SSage Weil struct ceph_mds_session *session) 1945a8599bd8SSage Weil { 1946a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1947a8599bd8SSage Weil 1948a8599bd8SSage Weil dout("kick_flushing_capsnaps mds%d\n", session->s_mds); 1949a8599bd8SSage Weil list_for_each_entry(capsnap, &session->s_cap_snaps_flushing, 1950a8599bd8SSage Weil flushing_item) { 1951a8599bd8SSage Weil struct ceph_inode_info *ci = capsnap->ci; 1952a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1953a8599bd8SSage Weil struct ceph_cap *cap; 1954a8599bd8SSage Weil 1955be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1956a8599bd8SSage Weil cap = ci->i_auth_cap; 1957a8599bd8SSage Weil if (cap && cap->session == session) { 1958a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p capsnap %p\n", inode, 1959a8599bd8SSage Weil cap, capsnap); 1960e835124cSSage Weil __ceph_flush_snaps(ci, &session, 1); 1961a8599bd8SSage Weil } else { 1962a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1963a8599bd8SSage Weil cap, session->s_mds); 1964a8599bd8SSage Weil } 1965be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1966a8599bd8SSage Weil } 1967a8599bd8SSage Weil } 1968a8599bd8SSage Weil 1969a8599bd8SSage Weil void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, 1970a8599bd8SSage Weil struct ceph_mds_session *session) 1971a8599bd8SSage Weil { 1972a8599bd8SSage Weil struct ceph_inode_info *ci; 1973a8599bd8SSage Weil 1974a8599bd8SSage Weil kick_flushing_capsnaps(mdsc, session); 1975a8599bd8SSage Weil 1976a8599bd8SSage Weil dout("kick_flushing_caps mds%d\n", session->s_mds); 1977a8599bd8SSage Weil list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { 1978a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1979a8599bd8SSage Weil struct ceph_cap *cap; 1980a8599bd8SSage Weil int delayed = 0; 1981a8599bd8SSage Weil 1982be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1983a8599bd8SSage Weil cap = ci->i_auth_cap; 1984a8599bd8SSage Weil if (cap && cap->session == session) { 1985a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p %s\n", inode, 1986a8599bd8SSage Weil cap, ceph_cap_string(ci->i_flushing_caps)); 1987a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 1988a8599bd8SSage Weil __ceph_caps_used(ci), 1989a8599bd8SSage Weil __ceph_caps_wanted(ci), 1990a8599bd8SSage Weil cap->issued | cap->implemented, 1991a8599bd8SSage Weil ci->i_flushing_caps, NULL); 1992a8599bd8SSage Weil if (delayed) { 1993be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1994a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1995be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1996a8599bd8SSage Weil } 1997a8599bd8SSage Weil } else { 1998a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1999a8599bd8SSage Weil cap, session->s_mds); 2000be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2001a8599bd8SSage Weil } 2002a8599bd8SSage Weil } 2003a8599bd8SSage Weil } 2004a8599bd8SSage Weil 2005088b3f5eSSage Weil static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc, 2006088b3f5eSSage Weil struct ceph_mds_session *session, 2007088b3f5eSSage Weil struct inode *inode) 2008088b3f5eSSage Weil { 2009088b3f5eSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2010088b3f5eSSage Weil struct ceph_cap *cap; 2011088b3f5eSSage Weil int delayed = 0; 2012088b3f5eSSage Weil 2013be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2014088b3f5eSSage Weil cap = ci->i_auth_cap; 2015088b3f5eSSage Weil dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode, 2016088b3f5eSSage Weil ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq); 2017005c4697SYan, Zheng 2018088b3f5eSSage Weil __ceph_flush_snaps(ci, &session, 1); 2019005c4697SYan, Zheng 2020088b3f5eSSage Weil if (ci->i_flushing_caps) { 2021005c4697SYan, Zheng spin_lock(&mdsc->cap_dirty_lock); 2022005c4697SYan, Zheng list_move_tail(&ci->i_flushing_item, 2023005c4697SYan, Zheng &cap->session->s_cap_flushing); 2024005c4697SYan, Zheng spin_unlock(&mdsc->cap_dirty_lock); 2025005c4697SYan, Zheng 2026088b3f5eSSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 2027088b3f5eSSage Weil __ceph_caps_used(ci), 2028088b3f5eSSage Weil __ceph_caps_wanted(ci), 2029088b3f5eSSage Weil cap->issued | cap->implemented, 2030088b3f5eSSage Weil ci->i_flushing_caps, NULL); 2031088b3f5eSSage Weil if (delayed) { 2032be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2033088b3f5eSSage Weil __cap_delay_requeue(mdsc, ci); 2034be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2035088b3f5eSSage Weil } 2036088b3f5eSSage Weil } else { 2037be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2038088b3f5eSSage Weil } 2039088b3f5eSSage Weil } 2040088b3f5eSSage Weil 2041a8599bd8SSage Weil 2042a8599bd8SSage Weil /* 2043a8599bd8SSage Weil * Take references to capabilities we hold, so that we don't release 2044a8599bd8SSage Weil * them to the MDS prematurely. 2045a8599bd8SSage Weil * 2046be655596SSage Weil * Protected by i_ceph_lock. 2047a8599bd8SSage Weil */ 2048a8599bd8SSage Weil static void __take_cap_refs(struct ceph_inode_info *ci, int got) 2049a8599bd8SSage Weil { 2050a8599bd8SSage Weil if (got & CEPH_CAP_PIN) 2051a8599bd8SSage Weil ci->i_pin_ref++; 2052a8599bd8SSage Weil if (got & CEPH_CAP_FILE_RD) 2053a8599bd8SSage Weil ci->i_rd_ref++; 2054a8599bd8SSage Weil if (got & CEPH_CAP_FILE_CACHE) 2055a8599bd8SSage Weil ci->i_rdcache_ref++; 2056a8599bd8SSage Weil if (got & CEPH_CAP_FILE_WR) 2057a8599bd8SSage Weil ci->i_wr_ref++; 2058a8599bd8SSage Weil if (got & CEPH_CAP_FILE_BUFFER) { 2059d3d0720dSHenry C Chang if (ci->i_wb_ref == 0) 20603772d26dSSage Weil ihold(&ci->vfs_inode); 2061d3d0720dSHenry C Chang ci->i_wb_ref++; 2062d3d0720dSHenry C Chang dout("__take_cap_refs %p wb %d -> %d (?)\n", 2063d3d0720dSHenry C Chang &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref); 2064a8599bd8SSage Weil } 2065a8599bd8SSage Weil } 2066a8599bd8SSage Weil 2067a8599bd8SSage Weil /* 2068a8599bd8SSage Weil * Try to grab cap references. Specify those refs we @want, and the 2069a8599bd8SSage Weil * minimal set we @need. Also include the larger offset we are writing 2070a8599bd8SSage Weil * to (when applicable), and check against max_size here as well. 2071a8599bd8SSage Weil * Note that caller is responsible for ensuring max_size increases are 2072a8599bd8SSage Weil * requested from the MDS. 2073a8599bd8SSage Weil */ 2074a8599bd8SSage Weil static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want, 20753738daa6SYan, Zheng loff_t endoff, int *got, struct page **pinned_page, 20763738daa6SYan, Zheng int *check_max, int *err) 2077a8599bd8SSage Weil { 2078a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2079a8599bd8SSage Weil int ret = 0; 20803738daa6SYan, Zheng int have, implemented, _got = 0; 2081195d3ce2SSage Weil int file_wanted; 2082a8599bd8SSage Weil 2083a8599bd8SSage Weil dout("get_cap_refs %p need %s want %s\n", inode, 2084a8599bd8SSage Weil ceph_cap_string(need), ceph_cap_string(want)); 20853738daa6SYan, Zheng again: 2086be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2087a8599bd8SSage Weil 2088195d3ce2SSage Weil /* make sure file is actually open */ 2089195d3ce2SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 2090195d3ce2SSage Weil if ((file_wanted & need) == 0) { 2091195d3ce2SSage Weil dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", 2092195d3ce2SSage Weil ceph_cap_string(need), ceph_cap_string(file_wanted)); 2093a8599bd8SSage Weil *err = -EBADF; 2094a8599bd8SSage Weil ret = 1; 20953738daa6SYan, Zheng goto out_unlock; 2096a8599bd8SSage Weil } 2097a8599bd8SSage Weil 209837505d57SYan, Zheng /* finish pending truncate */ 209937505d57SYan, Zheng while (ci->i_truncate_pending) { 210037505d57SYan, Zheng spin_unlock(&ci->i_ceph_lock); 2101b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 210237505d57SYan, Zheng spin_lock(&ci->i_ceph_lock); 210337505d57SYan, Zheng } 210437505d57SYan, Zheng 21053871cbb9SYan, Zheng have = __ceph_caps_issued(ci, &implemented); 21063871cbb9SYan, Zheng 21073871cbb9SYan, Zheng if (have & need & CEPH_CAP_FILE_WR) { 2108a8599bd8SSage Weil if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { 2109a8599bd8SSage Weil dout("get_cap_refs %p endoff %llu > maxsize %llu\n", 2110a8599bd8SSage Weil inode, endoff, ci->i_max_size); 21113871cbb9SYan, Zheng if (endoff > ci->i_requested_max_size) { 2112a8599bd8SSage Weil *check_max = 1; 2113a8599bd8SSage Weil ret = 1; 2114a8599bd8SSage Weil } 21153738daa6SYan, Zheng goto out_unlock; 2116a8599bd8SSage Weil } 2117a8599bd8SSage Weil /* 2118a8599bd8SSage Weil * If a sync write is in progress, we must wait, so that we 2119a8599bd8SSage Weil * can get a final snapshot value for size+mtime. 2120a8599bd8SSage Weil */ 2121a8599bd8SSage Weil if (__ceph_have_pending_cap_snap(ci)) { 2122a8599bd8SSage Weil dout("get_cap_refs %p cap_snap_pending\n", inode); 21233738daa6SYan, Zheng goto out_unlock; 2124a8599bd8SSage Weil } 2125a8599bd8SSage Weil } 2126a8599bd8SSage Weil 2127a8599bd8SSage Weil if ((have & need) == need) { 2128a8599bd8SSage Weil /* 2129a8599bd8SSage Weil * Look at (implemented & ~have & not) so that we keep waiting 2130a8599bd8SSage Weil * on transition from wanted -> needed caps. This is needed 2131a8599bd8SSage Weil * for WRBUFFER|WR -> WR to avoid a new WR sync write from 2132a8599bd8SSage Weil * going before a prior buffered writeback happens. 2133a8599bd8SSage Weil */ 2134a8599bd8SSage Weil int not = want & ~(have & need); 2135a8599bd8SSage Weil int revoking = implemented & ~have; 2136a8599bd8SSage Weil dout("get_cap_refs %p have %s but not %s (revoking %s)\n", 2137a8599bd8SSage Weil inode, ceph_cap_string(have), ceph_cap_string(not), 2138a8599bd8SSage Weil ceph_cap_string(revoking)); 2139a8599bd8SSage Weil if ((revoking & not) == 0) { 21403738daa6SYan, Zheng _got = need | (have & want); 21413738daa6SYan, Zheng __take_cap_refs(ci, _got); 2142a8599bd8SSage Weil ret = 1; 2143a8599bd8SSage Weil } 2144a8599bd8SSage Weil } else { 214503f4fcb0SYan, Zheng int session_readonly = false; 214603f4fcb0SYan, Zheng if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) { 214703f4fcb0SYan, Zheng struct ceph_mds_session *s = ci->i_auth_cap->session; 214803f4fcb0SYan, Zheng spin_lock(&s->s_cap_lock); 214903f4fcb0SYan, Zheng session_readonly = s->s_readonly; 215003f4fcb0SYan, Zheng spin_unlock(&s->s_cap_lock); 215103f4fcb0SYan, Zheng } 215203f4fcb0SYan, Zheng if (session_readonly) { 215303f4fcb0SYan, Zheng dout("get_cap_refs %p needed %s but mds%d readonly\n", 215403f4fcb0SYan, Zheng inode, ceph_cap_string(need), ci->i_auth_cap->mds); 215503f4fcb0SYan, Zheng *err = -EROFS; 215603f4fcb0SYan, Zheng ret = 1; 215703f4fcb0SYan, Zheng goto out_unlock; 215803f4fcb0SYan, Zheng } 215903f4fcb0SYan, Zheng 2160a8599bd8SSage Weil dout("get_cap_refs %p have %s needed %s\n", inode, 2161a8599bd8SSage Weil ceph_cap_string(have), ceph_cap_string(need)); 2162a8599bd8SSage Weil } 21633738daa6SYan, Zheng out_unlock: 2164be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 21653738daa6SYan, Zheng 21663738daa6SYan, Zheng if (ci->i_inline_version != CEPH_INLINE_NONE && 21673738daa6SYan, Zheng (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && 21683738daa6SYan, Zheng i_size_read(inode) > 0) { 21693738daa6SYan, Zheng int ret1; 21703738daa6SYan, Zheng struct page *page = find_get_page(inode->i_mapping, 0); 21713738daa6SYan, Zheng if (page) { 21723738daa6SYan, Zheng if (PageUptodate(page)) { 21733738daa6SYan, Zheng *pinned_page = page; 21743738daa6SYan, Zheng goto out; 21753738daa6SYan, Zheng } 21763738daa6SYan, Zheng page_cache_release(page); 21773738daa6SYan, Zheng } 21783738daa6SYan, Zheng /* 21793738daa6SYan, Zheng * drop cap refs first because getattr while holding 21803738daa6SYan, Zheng * caps refs can cause deadlock. 21813738daa6SYan, Zheng */ 21823738daa6SYan, Zheng ceph_put_cap_refs(ci, _got); 21833738daa6SYan, Zheng _got = 0; 21843738daa6SYan, Zheng 21853738daa6SYan, Zheng /* getattr request will bring inline data into page cache */ 21863738daa6SYan, Zheng ret1 = __ceph_do_getattr(inode, NULL, 21873738daa6SYan, Zheng CEPH_STAT_CAP_INLINE_DATA, true); 21883738daa6SYan, Zheng if (ret1 >= 0) { 21893738daa6SYan, Zheng ret = 0; 21903738daa6SYan, Zheng goto again; 21913738daa6SYan, Zheng } 21923738daa6SYan, Zheng *err = ret1; 21933738daa6SYan, Zheng ret = 1; 21943738daa6SYan, Zheng } 21953738daa6SYan, Zheng out: 2196a8599bd8SSage Weil dout("get_cap_refs %p ret %d got %s\n", inode, 21973738daa6SYan, Zheng ret, ceph_cap_string(_got)); 21983738daa6SYan, Zheng *got = _got; 2199a8599bd8SSage Weil return ret; 2200a8599bd8SSage Weil } 2201a8599bd8SSage Weil 2202a8599bd8SSage Weil /* 2203a8599bd8SSage Weil * Check the offset we are writing up to against our current 2204a8599bd8SSage Weil * max_size. If necessary, tell the MDS we want to write to 2205a8599bd8SSage Weil * a larger offset. 2206a8599bd8SSage Weil */ 2207a8599bd8SSage Weil static void check_max_size(struct inode *inode, loff_t endoff) 2208a8599bd8SSage Weil { 2209a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2210a8599bd8SSage Weil int check = 0; 2211a8599bd8SSage Weil 2212a8599bd8SSage Weil /* do we need to explicitly request a larger max_size? */ 2213be655596SSage Weil spin_lock(&ci->i_ceph_lock); 22143871cbb9SYan, Zheng if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) { 2215a8599bd8SSage Weil dout("write %p at large endoff %llu, req max_size\n", 2216a8599bd8SSage Weil inode, endoff); 2217a8599bd8SSage Weil ci->i_wanted_max_size = endoff; 2218a8599bd8SSage Weil } 22193871cbb9SYan, Zheng /* duplicate ceph_check_caps()'s logic */ 22203871cbb9SYan, Zheng if (ci->i_auth_cap && 22213871cbb9SYan, Zheng (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) && 22223871cbb9SYan, Zheng ci->i_wanted_max_size > ci->i_max_size && 22233871cbb9SYan, Zheng ci->i_wanted_max_size > ci->i_requested_max_size) 22243871cbb9SYan, Zheng check = 1; 2225be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2226a8599bd8SSage Weil if (check) 2227a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2228a8599bd8SSage Weil } 2229a8599bd8SSage Weil 2230a8599bd8SSage Weil /* 2231a8599bd8SSage Weil * Wait for caps, and take cap references. If we can't get a WR cap 2232a8599bd8SSage Weil * due to a small max_size, make sure we check_max_size (and possibly 2233a8599bd8SSage Weil * ask the mds) so we don't get hung up indefinitely. 2234a8599bd8SSage Weil */ 22353738daa6SYan, Zheng int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, 22363738daa6SYan, Zheng loff_t endoff, int *got, struct page **pinned_page) 2237a8599bd8SSage Weil { 2238a8599bd8SSage Weil int check_max, ret, err; 2239a8599bd8SSage Weil 2240a8599bd8SSage Weil retry: 2241a8599bd8SSage Weil if (endoff > 0) 2242a8599bd8SSage Weil check_max_size(&ci->vfs_inode, endoff); 2243a8599bd8SSage Weil check_max = 0; 2244a8599bd8SSage Weil err = 0; 2245a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 22463738daa6SYan, Zheng try_get_cap_refs(ci, need, want, endoff, 22473738daa6SYan, Zheng got, pinned_page, 2248a8599bd8SSage Weil &check_max, &err)); 2249a8599bd8SSage Weil if (err) 2250a8599bd8SSage Weil ret = err; 2251a8599bd8SSage Weil if (check_max) 2252a8599bd8SSage Weil goto retry; 2253a8599bd8SSage Weil return ret; 2254a8599bd8SSage Weil } 2255a8599bd8SSage Weil 2256a8599bd8SSage Weil /* 2257a8599bd8SSage Weil * Take cap refs. Caller must already know we hold at least one ref 2258a8599bd8SSage Weil * on the caps in question or we don't know this is safe. 2259a8599bd8SSage Weil */ 2260a8599bd8SSage Weil void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) 2261a8599bd8SSage Weil { 2262be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2263a8599bd8SSage Weil __take_cap_refs(ci, caps); 2264be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2265a8599bd8SSage Weil } 2266a8599bd8SSage Weil 2267a8599bd8SSage Weil /* 2268a8599bd8SSage Weil * Release cap refs. 2269a8599bd8SSage Weil * 2270a8599bd8SSage Weil * If we released the last ref on any given cap, call ceph_check_caps 2271a8599bd8SSage Weil * to release (or schedule a release). 2272a8599bd8SSage Weil * 2273a8599bd8SSage Weil * If we are releasing a WR cap (from a sync write), finalize any affected 2274a8599bd8SSage Weil * cap_snap, and wake up any waiters. 2275a8599bd8SSage Weil */ 2276a8599bd8SSage Weil void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) 2277a8599bd8SSage Weil { 2278a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2279a8599bd8SSage Weil int last = 0, put = 0, flushsnaps = 0, wake = 0; 2280a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2281a8599bd8SSage Weil 2282be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2283a8599bd8SSage Weil if (had & CEPH_CAP_PIN) 2284a8599bd8SSage Weil --ci->i_pin_ref; 2285a8599bd8SSage Weil if (had & CEPH_CAP_FILE_RD) 2286a8599bd8SSage Weil if (--ci->i_rd_ref == 0) 2287a8599bd8SSage Weil last++; 2288a8599bd8SSage Weil if (had & CEPH_CAP_FILE_CACHE) 2289a8599bd8SSage Weil if (--ci->i_rdcache_ref == 0) 2290a8599bd8SSage Weil last++; 2291a8599bd8SSage Weil if (had & CEPH_CAP_FILE_BUFFER) { 2292d3d0720dSHenry C Chang if (--ci->i_wb_ref == 0) { 2293a8599bd8SSage Weil last++; 2294a8599bd8SSage Weil put++; 2295a8599bd8SSage Weil } 2296d3d0720dSHenry C Chang dout("put_cap_refs %p wb %d -> %d (?)\n", 2297d3d0720dSHenry C Chang inode, ci->i_wb_ref+1, ci->i_wb_ref); 2298a8599bd8SSage Weil } 2299a8599bd8SSage Weil if (had & CEPH_CAP_FILE_WR) 2300a8599bd8SSage Weil if (--ci->i_wr_ref == 0) { 2301a8599bd8SSage Weil last++; 2302a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) { 2303a8599bd8SSage Weil capsnap = list_first_entry(&ci->i_cap_snaps, 2304a8599bd8SSage Weil struct ceph_cap_snap, 2305a8599bd8SSage Weil ci_item); 2306a8599bd8SSage Weil if (capsnap->writing) { 2307a8599bd8SSage Weil capsnap->writing = 0; 2308a8599bd8SSage Weil flushsnaps = 2309a8599bd8SSage Weil __ceph_finish_cap_snap(ci, 2310a8599bd8SSage Weil capsnap); 2311a8599bd8SSage Weil wake = 1; 2312a8599bd8SSage Weil } 2313a8599bd8SSage Weil } 2314a8599bd8SSage Weil } 2315be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2316a8599bd8SSage Weil 2317819ccbfaSSage Weil dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), 2318819ccbfaSSage Weil last ? " last" : "", put ? " put" : ""); 2319a8599bd8SSage Weil 2320a8599bd8SSage Weil if (last && !flushsnaps) 2321a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 2322a8599bd8SSage Weil else if (flushsnaps) 2323a8599bd8SSage Weil ceph_flush_snaps(ci); 2324a8599bd8SSage Weil if (wake) 232503066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2326a8599bd8SSage Weil if (put) 2327a8599bd8SSage Weil iput(inode); 2328a8599bd8SSage Weil } 2329a8599bd8SSage Weil 2330a8599bd8SSage Weil /* 2331a8599bd8SSage Weil * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap 2332a8599bd8SSage Weil * context. Adjust per-snap dirty page accounting as appropriate. 2333a8599bd8SSage Weil * Once all dirty data for a cap_snap is flushed, flush snapped file 2334a8599bd8SSage Weil * metadata back to the MDS. If we dropped the last ref, call 2335a8599bd8SSage Weil * ceph_check_caps. 2336a8599bd8SSage Weil */ 2337a8599bd8SSage Weil void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, 2338a8599bd8SSage Weil struct ceph_snap_context *snapc) 2339a8599bd8SSage Weil { 2340a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2341a8599bd8SSage Weil int last = 0; 2342819ccbfaSSage Weil int complete_capsnap = 0; 2343819ccbfaSSage Weil int drop_capsnap = 0; 2344a8599bd8SSage Weil int found = 0; 2345a8599bd8SSage Weil struct ceph_cap_snap *capsnap = NULL; 2346a8599bd8SSage Weil 2347be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2348a8599bd8SSage Weil ci->i_wrbuffer_ref -= nr; 2349a8599bd8SSage Weil last = !ci->i_wrbuffer_ref; 2350a8599bd8SSage Weil 2351a8599bd8SSage Weil if (ci->i_head_snapc == snapc) { 2352a8599bd8SSage Weil ci->i_wrbuffer_ref_head -= nr; 23537d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0 && 23547d8cb26dSSage Weil ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) { 23557d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 2356a8599bd8SSage Weil ceph_put_snap_context(ci->i_head_snapc); 2357a8599bd8SSage Weil ci->i_head_snapc = NULL; 2358a8599bd8SSage Weil } 2359a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", 2360a8599bd8SSage Weil inode, 2361a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, 2362a8599bd8SSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 2363a8599bd8SSage Weil last ? " LAST" : ""); 2364a8599bd8SSage Weil } else { 2365a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2366a8599bd8SSage Weil if (capsnap->context == snapc) { 2367a8599bd8SSage Weil found = 1; 2368a8599bd8SSage Weil break; 2369a8599bd8SSage Weil } 2370a8599bd8SSage Weil } 2371a8599bd8SSage Weil BUG_ON(!found); 2372819ccbfaSSage Weil capsnap->dirty_pages -= nr; 2373819ccbfaSSage Weil if (capsnap->dirty_pages == 0) { 2374819ccbfaSSage Weil complete_capsnap = 1; 2375819ccbfaSSage Weil if (capsnap->dirty == 0) 2376819ccbfaSSage Weil /* cap writeback completed before we created 2377819ccbfaSSage Weil * the cap_snap; no FLUSHSNAP is needed */ 2378819ccbfaSSage Weil drop_capsnap = 1; 2379819ccbfaSSage Weil } 2380a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p cap_snap %p " 2381819ccbfaSSage Weil " snap %lld %d/%d -> %d/%d %s%s%s\n", 2382a8599bd8SSage Weil inode, capsnap, capsnap->context->seq, 2383a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, 2384a8599bd8SSage Weil ci->i_wrbuffer_ref, capsnap->dirty_pages, 2385a8599bd8SSage Weil last ? " (wrbuffer last)" : "", 2386819ccbfaSSage Weil complete_capsnap ? " (complete capsnap)" : "", 2387819ccbfaSSage Weil drop_capsnap ? " (drop capsnap)" : ""); 2388819ccbfaSSage Weil if (drop_capsnap) { 2389819ccbfaSSage Weil ceph_put_snap_context(capsnap->context); 2390819ccbfaSSage Weil list_del(&capsnap->ci_item); 2391819ccbfaSSage Weil list_del(&capsnap->flushing_item); 2392819ccbfaSSage Weil ceph_put_cap_snap(capsnap); 2393819ccbfaSSage Weil } 2394a8599bd8SSage Weil } 2395a8599bd8SSage Weil 2396be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2397a8599bd8SSage Weil 2398a8599bd8SSage Weil if (last) { 2399a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2400a8599bd8SSage Weil iput(inode); 2401819ccbfaSSage Weil } else if (complete_capsnap) { 2402a8599bd8SSage Weil ceph_flush_snaps(ci); 240303066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2404a8599bd8SSage Weil } 2405819ccbfaSSage Weil if (drop_capsnap) 2406819ccbfaSSage Weil iput(inode); 2407a8599bd8SSage Weil } 2408a8599bd8SSage Weil 2409a8599bd8SSage Weil /* 2410ca20c991SYan, Zheng * Invalidate unlinked inode's aliases, so we can drop the inode ASAP. 2411ca20c991SYan, Zheng */ 2412ca20c991SYan, Zheng static void invalidate_aliases(struct inode *inode) 2413ca20c991SYan, Zheng { 2414ca20c991SYan, Zheng struct dentry *dn, *prev = NULL; 2415ca20c991SYan, Zheng 2416ca20c991SYan, Zheng dout("invalidate_aliases inode %p\n", inode); 2417ca20c991SYan, Zheng d_prune_aliases(inode); 2418ca20c991SYan, Zheng /* 2419ca20c991SYan, Zheng * For non-directory inode, d_find_alias() only returns 2420fc12c80aSJ. Bruce Fields * hashed dentry. After calling d_invalidate(), the 2421fc12c80aSJ. Bruce Fields * dentry becomes unhashed. 2422ca20c991SYan, Zheng * 2423a8d436f0SYan, Zheng * For directory inode, d_find_alias() can return 2424fc12c80aSJ. Bruce Fields * unhashed dentry. But directory inode should have 2425ca20c991SYan, Zheng * one alias at most. 2426ca20c991SYan, Zheng */ 2427ca20c991SYan, Zheng while ((dn = d_find_alias(inode))) { 2428ca20c991SYan, Zheng if (dn == prev) { 2429ca20c991SYan, Zheng dput(dn); 2430ca20c991SYan, Zheng break; 2431ca20c991SYan, Zheng } 2432a8d436f0SYan, Zheng d_invalidate(dn); 2433ca20c991SYan, Zheng if (prev) 2434ca20c991SYan, Zheng dput(prev); 2435ca20c991SYan, Zheng prev = dn; 2436ca20c991SYan, Zheng } 2437ca20c991SYan, Zheng if (prev) 2438ca20c991SYan, Zheng dput(prev); 2439ca20c991SYan, Zheng } 2440ca20c991SYan, Zheng 2441ca20c991SYan, Zheng /* 2442a8599bd8SSage Weil * Handle a cap GRANT message from the MDS. (Note that a GRANT may 2443a8599bd8SSage Weil * actually be a revocation if it specifies a smaller cap set.) 2444a8599bd8SSage Weil * 2445be655596SSage Weil * caller holds s_mutex and i_ceph_lock, we drop both. 2446a8599bd8SSage Weil */ 24472cd698beSYan, Zheng static void handle_cap_grant(struct ceph_mds_client *mdsc, 24482cd698beSYan, Zheng struct inode *inode, struct ceph_mds_caps *grant, 2449fb01d1f8SYan, Zheng u64 inline_version, 2450fb01d1f8SYan, Zheng void *inline_data, int inline_len, 24512cd698beSYan, Zheng struct ceph_buffer *xattr_buf, 2452a8599bd8SSage Weil struct ceph_mds_session *session, 24532cd698beSYan, Zheng struct ceph_cap *cap, int issued) 2454be655596SSage Weil __releases(ci->i_ceph_lock) 2455982d6011SYan, Zheng __releases(mdsc->snap_rwsem) 2456a8599bd8SSage Weil { 2457a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2458a8599bd8SSage Weil int mds = session->s_mds; 24592f56f56aSSage Weil int seq = le32_to_cpu(grant->seq); 2460a8599bd8SSage Weil int newcaps = le32_to_cpu(grant->caps); 24612cd698beSYan, Zheng int used, wanted, dirty; 2462a8599bd8SSage Weil u64 size = le64_to_cpu(grant->size); 2463a8599bd8SSage Weil u64 max_size = le64_to_cpu(grant->max_size); 2464a8599bd8SSage Weil struct timespec mtime, atime, ctime; 246515637c8bSSage Weil int check_caps = 0; 2466ab6c2c3eSFabian Frederick bool wake = false; 2467ab6c2c3eSFabian Frederick bool writeback = false; 2468ab6c2c3eSFabian Frederick bool queue_trunc = false; 2469ab6c2c3eSFabian Frederick bool queue_invalidate = false; 2470ab6c2c3eSFabian Frederick bool queue_revalidate = false; 2471ab6c2c3eSFabian Frederick bool deleted_inode = false; 247231c542a1SYan, Zheng bool fill_inline = false; 2473a8599bd8SSage Weil 24742f56f56aSSage Weil dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", 24752f56f56aSSage Weil inode, cap, mds, seq, ceph_cap_string(newcaps)); 2476a8599bd8SSage Weil dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, 2477a8599bd8SSage Weil inode->i_size); 2478a8599bd8SSage Weil 247911df2dfbSYan, Zheng 248011df2dfbSYan, Zheng /* 248111df2dfbSYan, Zheng * auth mds of the inode changed. we received the cap export message, 248211df2dfbSYan, Zheng * but still haven't received the cap import message. handle_cap_export 248311df2dfbSYan, Zheng * updated the new auth MDS' cap. 248411df2dfbSYan, Zheng * 248511df2dfbSYan, Zheng * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message 248611df2dfbSYan, Zheng * that was sent before the cap import message. So don't remove caps. 248711df2dfbSYan, Zheng */ 248811df2dfbSYan, Zheng if (ceph_seq_cmp(seq, cap->seq) <= 0) { 248911df2dfbSYan, Zheng WARN_ON(cap != ci->i_auth_cap); 249011df2dfbSYan, Zheng WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id)); 249111df2dfbSYan, Zheng seq = cap->seq; 249211df2dfbSYan, Zheng newcaps |= cap->issued; 249311df2dfbSYan, Zheng } 249411df2dfbSYan, Zheng 2495a8599bd8SSage Weil /* 2496a8599bd8SSage Weil * If CACHE is being revoked, and we have no dirty buffers, 2497a8599bd8SSage Weil * try to invalidate (once). (If there are dirty buffers, we 2498a8599bd8SSage Weil * will invalidate _after_ writeback.) 2499a8599bd8SSage Weil */ 25003b454c49SSage Weil if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && 25013b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 2502bcd2cbd1SYehuda Sadeh !ci->i_wrbuffer_ref) { 2503e9075743SLi Wang if (try_nonblocking_invalidate(inode)) { 2504a8599bd8SSage Weil /* there were locked pages.. invalidate later 2505a8599bd8SSage Weil in a separate thread. */ 2506a8599bd8SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 2507ab6c2c3eSFabian Frederick queue_invalidate = true; 2508a8599bd8SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 2509a8599bd8SSage Weil } 2510a8599bd8SSage Weil } 251199ccbd22SMilosz Tanski 251299ccbd22SMilosz Tanski ceph_fscache_invalidate(inode); 2513a8599bd8SSage Weil } 2514a8599bd8SSage Weil 2515a8599bd8SSage Weil /* side effects now are allowed */ 2516685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 251711df2dfbSYan, Zheng cap->seq = seq; 2518a8599bd8SSage Weil 2519a8599bd8SSage Weil __check_cap_issue(ci, cap, newcaps); 2520a8599bd8SSage Weil 2521f98a128aSYan, Zheng if ((newcaps & CEPH_CAP_AUTH_SHARED) && 2522f98a128aSYan, Zheng (issued & CEPH_CAP_AUTH_EXCL) == 0) { 2523a8599bd8SSage Weil inode->i_mode = le32_to_cpu(grant->mode); 252405cb11c1SEric W. Biederman inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid)); 252505cb11c1SEric W. Biederman inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid)); 2526a8599bd8SSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 2527bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 2528bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid)); 2529a8599bd8SSage Weil } 2530a8599bd8SSage Weil 2531f98a128aSYan, Zheng if ((newcaps & CEPH_CAP_AUTH_SHARED) && 2532f98a128aSYan, Zheng (issued & CEPH_CAP_LINK_EXCL) == 0) { 2533bfe86848SMiklos Szeredi set_nlink(inode, le32_to_cpu(grant->nlink)); 2534ca20c991SYan, Zheng if (inode->i_nlink == 0 && 2535ca20c991SYan, Zheng (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) 2536ab6c2c3eSFabian Frederick deleted_inode = true; 2537ca20c991SYan, Zheng } 2538a8599bd8SSage Weil 2539a8599bd8SSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) { 2540a8599bd8SSage Weil int len = le32_to_cpu(grant->xattr_len); 2541a8599bd8SSage Weil u64 version = le64_to_cpu(grant->xattr_version); 2542a8599bd8SSage Weil 2543a8599bd8SSage Weil if (version > ci->i_xattrs.version) { 2544a8599bd8SSage Weil dout(" got new xattrs v%llu on %p len %d\n", 2545a8599bd8SSage Weil version, inode, len); 2546a8599bd8SSage Weil if (ci->i_xattrs.blob) 2547a8599bd8SSage Weil ceph_buffer_put(ci->i_xattrs.blob); 2548a8599bd8SSage Weil ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); 2549a8599bd8SSage Weil ci->i_xattrs.version = version; 25507221fe4cSGuangliang Zhao ceph_forget_all_cached_acls(inode); 2551a8599bd8SSage Weil } 2552a8599bd8SSage Weil } 2553a8599bd8SSage Weil 255499ccbd22SMilosz Tanski /* Do we need to revalidate our fscache cookie. Don't bother on the 255599ccbd22SMilosz Tanski * first cache cap as we already validate at cookie creation time. */ 255699ccbd22SMilosz Tanski if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1) 2557ab6c2c3eSFabian Frederick queue_revalidate = true; 255899ccbd22SMilosz Tanski 2559f98a128aSYan, Zheng if (newcaps & CEPH_CAP_ANY_RD) { 2560f98a128aSYan, Zheng /* ctime/mtime/atime? */ 2561a8599bd8SSage Weil ceph_decode_timespec(&mtime, &grant->mtime); 2562a8599bd8SSage Weil ceph_decode_timespec(&atime, &grant->atime); 2563a8599bd8SSage Weil ceph_decode_timespec(&ctime, &grant->ctime); 2564a8599bd8SSage Weil ceph_fill_file_time(inode, issued, 2565f98a128aSYan, Zheng le32_to_cpu(grant->time_warp_seq), 2566f98a128aSYan, Zheng &ctime, &mtime, &atime); 2567f98a128aSYan, Zheng } 2568a8599bd8SSage Weil 2569f98a128aSYan, Zheng if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { 257011df2dfbSYan, Zheng /* file layout may have changed */ 257111df2dfbSYan, Zheng ci->i_layout = grant->layout; 2572f98a128aSYan, Zheng /* size/truncate_seq? */ 2573f98a128aSYan, Zheng queue_trunc = ceph_fill_file_size(inode, issued, 2574f98a128aSYan, Zheng le32_to_cpu(grant->truncate_seq), 2575f98a128aSYan, Zheng le64_to_cpu(grant->truncate_size), 2576f98a128aSYan, Zheng size); 2577a8599bd8SSage Weil /* max size increase? */ 25785e62ad30SYan, Zheng if (ci->i_auth_cap == cap && max_size != ci->i_max_size) { 2579f98a128aSYan, Zheng dout("max_size %lld -> %llu\n", 2580f98a128aSYan, Zheng ci->i_max_size, max_size); 2581a8599bd8SSage Weil ci->i_max_size = max_size; 2582a8599bd8SSage Weil if (max_size >= ci->i_wanted_max_size) { 2583a8599bd8SSage Weil ci->i_wanted_max_size = 0; /* reset */ 2584a8599bd8SSage Weil ci->i_requested_max_size = 0; 2585a8599bd8SSage Weil } 2586ab6c2c3eSFabian Frederick wake = true; 2587a8599bd8SSage Weil } 2588f98a128aSYan, Zheng } 2589a8599bd8SSage Weil 2590a8599bd8SSage Weil /* check cap bits */ 2591a8599bd8SSage Weil wanted = __ceph_caps_wanted(ci); 2592a8599bd8SSage Weil used = __ceph_caps_used(ci); 2593a8599bd8SSage Weil dirty = __ceph_caps_dirty(ci); 2594a8599bd8SSage Weil dout(" my wanted = %s, used = %s, dirty %s\n", 2595a8599bd8SSage Weil ceph_cap_string(wanted), 2596a8599bd8SSage Weil ceph_cap_string(used), 2597a8599bd8SSage Weil ceph_cap_string(dirty)); 2598a8599bd8SSage Weil if (wanted != le32_to_cpu(grant->wanted)) { 2599a8599bd8SSage Weil dout("mds wanted %s -> %s\n", 2600a8599bd8SSage Weil ceph_cap_string(le32_to_cpu(grant->wanted)), 2601a8599bd8SSage Weil ceph_cap_string(wanted)); 2602390306c3SYan, Zheng /* imported cap may not have correct mds_wanted */ 2603390306c3SYan, Zheng if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) 2604390306c3SYan, Zheng check_caps = 1; 2605a8599bd8SSage Weil } 2606a8599bd8SSage Weil 2607a8599bd8SSage Weil /* revocation, grant, or no-op? */ 2608a8599bd8SSage Weil if (cap->issued & ~newcaps) { 26093b454c49SSage Weil int revoking = cap->issued & ~newcaps; 26103b454c49SSage Weil 26113b454c49SSage Weil dout("revocation: %s -> %s (revoking %s)\n", 26123b454c49SSage Weil ceph_cap_string(cap->issued), 26133b454c49SSage Weil ceph_cap_string(newcaps), 26143b454c49SSage Weil ceph_cap_string(revoking)); 26150eb6cd49SSage Weil if (revoking & used & CEPH_CAP_FILE_BUFFER) 2616ab6c2c3eSFabian Frederick writeback = true; /* initiate writeback; will delay ack */ 26173b454c49SSage Weil else if (revoking == CEPH_CAP_FILE_CACHE && 26183b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 26193b454c49SSage Weil queue_invalidate) 26203b454c49SSage Weil ; /* do nothing yet, invalidation will be queued */ 26213b454c49SSage Weil else if (cap == ci->i_auth_cap) 26223b454c49SSage Weil check_caps = 1; /* check auth cap only */ 26233b454c49SSage Weil else 26243b454c49SSage Weil check_caps = 2; /* check all caps */ 2625a8599bd8SSage Weil cap->issued = newcaps; 2626978097c9SSage Weil cap->implemented |= newcaps; 2627a8599bd8SSage Weil } else if (cap->issued == newcaps) { 2628a8599bd8SSage Weil dout("caps unchanged: %s -> %s\n", 2629a8599bd8SSage Weil ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); 2630a8599bd8SSage Weil } else { 2631a8599bd8SSage Weil dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), 2632a8599bd8SSage Weil ceph_cap_string(newcaps)); 26336ee6b953SYan, Zheng /* non-auth MDS is revoking the newly grant caps ? */ 26346ee6b953SYan, Zheng if (cap == ci->i_auth_cap && 26356ee6b953SYan, Zheng __ceph_caps_revoking_other(ci, cap, newcaps)) 26366ee6b953SYan, Zheng check_caps = 2; 26376ee6b953SYan, Zheng 2638a8599bd8SSage Weil cap->issued = newcaps; 2639a8599bd8SSage Weil cap->implemented |= newcaps; /* add bits only, to 2640a8599bd8SSage Weil * avoid stepping on a 2641a8599bd8SSage Weil * pending revocation */ 2642ab6c2c3eSFabian Frederick wake = true; 2643a8599bd8SSage Weil } 2644978097c9SSage Weil BUG_ON(cap->issued & ~cap->implemented); 2645a8599bd8SSage Weil 264631c542a1SYan, Zheng if (inline_version > 0 && inline_version >= ci->i_inline_version) { 264731c542a1SYan, Zheng ci->i_inline_version = inline_version; 264831c542a1SYan, Zheng if (ci->i_inline_version != CEPH_INLINE_NONE && 264931c542a1SYan, Zheng (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO))) 265031c542a1SYan, Zheng fill_inline = true; 265131c542a1SYan, Zheng } 265231c542a1SYan, Zheng 2653be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 265499ccbd22SMilosz Tanski 26552cd698beSYan, Zheng if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) { 26562cd698beSYan, Zheng kick_flushing_inode_caps(mdsc, session, inode); 26572cd698beSYan, Zheng up_read(&mdsc->snap_rwsem); 26582cd698beSYan, Zheng if (newcaps & ~issued) 2659ab6c2c3eSFabian Frederick wake = true; 26602cd698beSYan, Zheng } 26612cd698beSYan, Zheng 266231c542a1SYan, Zheng if (fill_inline) 266331c542a1SYan, Zheng ceph_fill_inline_data(inode, NULL, inline_data, inline_len); 266431c542a1SYan, Zheng 2665c6bcda6fSYan, Zheng if (queue_trunc) { 2666c6bcda6fSYan, Zheng ceph_queue_vmtruncate(inode); 2667c6bcda6fSYan, Zheng ceph_queue_revalidate(inode); 2668c6bcda6fSYan, Zheng } else if (queue_revalidate) 2669c6bcda6fSYan, Zheng ceph_queue_revalidate(inode); 2670c6bcda6fSYan, Zheng 26713c6f6b79SSage Weil if (writeback) 2672a8599bd8SSage Weil /* 2673a8599bd8SSage Weil * queue inode for writeback: we can't actually call 2674a8599bd8SSage Weil * filemap_write_and_wait, etc. from message handler 2675a8599bd8SSage Weil * context. 2676a8599bd8SSage Weil */ 26773c6f6b79SSage Weil ceph_queue_writeback(inode); 26783c6f6b79SSage Weil if (queue_invalidate) 26793c6f6b79SSage Weil ceph_queue_invalidate(inode); 2680ca20c991SYan, Zheng if (deleted_inode) 2681ca20c991SYan, Zheng invalidate_aliases(inode); 2682a8599bd8SSage Weil if (wake) 268303066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 268415637c8bSSage Weil 268515637c8bSSage Weil if (check_caps == 1) 268615637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, 268715637c8bSSage Weil session); 268815637c8bSSage Weil else if (check_caps == 2) 268915637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); 269015637c8bSSage Weil else 269115637c8bSSage Weil mutex_unlock(&session->s_mutex); 2692a8599bd8SSage Weil } 2693a8599bd8SSage Weil 2694a8599bd8SSage Weil /* 2695a8599bd8SSage Weil * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the 2696a8599bd8SSage Weil * MDS has been safely committed. 2697a8599bd8SSage Weil */ 26986df058c0SSage Weil static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, 2699a8599bd8SSage Weil struct ceph_mds_caps *m, 2700a8599bd8SSage Weil struct ceph_mds_session *session, 2701a8599bd8SSage Weil struct ceph_cap *cap) 2702be655596SSage Weil __releases(ci->i_ceph_lock) 2703a8599bd8SSage Weil { 2704a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 27053d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 2706a8599bd8SSage Weil unsigned seq = le32_to_cpu(m->seq); 2707a8599bd8SSage Weil int dirty = le32_to_cpu(m->dirty); 2708a8599bd8SSage Weil int cleaned = 0; 2709afcdaea3SSage Weil int drop = 0; 2710a8599bd8SSage Weil int i; 2711a8599bd8SSage Weil 2712a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 2713a8599bd8SSage Weil if ((dirty & (1 << i)) && 27143231300bSYan, Zheng (u16)flush_tid == ci->i_cap_flush_tid[i]) 2715a8599bd8SSage Weil cleaned |= 1 << i; 2716a8599bd8SSage Weil 2717a8599bd8SSage Weil dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," 2718a8599bd8SSage Weil " flushing %s -> %s\n", 2719a8599bd8SSage Weil inode, session->s_mds, seq, ceph_cap_string(dirty), 2720a8599bd8SSage Weil ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), 2721a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps & ~cleaned)); 2722a8599bd8SSage Weil 2723a8599bd8SSage Weil if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned)) 2724a8599bd8SSage Weil goto out; 2725a8599bd8SSage Weil 2726a8599bd8SSage Weil ci->i_flushing_caps &= ~cleaned; 2727a8599bd8SSage Weil 2728a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2729a8599bd8SSage Weil if (ci->i_flushing_caps == 0) { 2730a8599bd8SSage Weil list_del_init(&ci->i_flushing_item); 2731a8599bd8SSage Weil if (!list_empty(&session->s_cap_flushing)) 2732a8599bd8SSage Weil dout(" mds%d still flushing cap on %p\n", 2733a8599bd8SSage Weil session->s_mds, 2734a8599bd8SSage Weil &list_entry(session->s_cap_flushing.next, 2735a8599bd8SSage Weil struct ceph_inode_info, 2736a8599bd8SSage Weil i_flushing_item)->vfs_inode); 2737a8599bd8SSage Weil mdsc->num_cap_flushing--; 273803066f23SYehuda Sadeh wake_up_all(&mdsc->cap_flushing_wq); 2739a8599bd8SSage Weil dout(" inode %p now !flushing\n", inode); 2740afcdaea3SSage Weil 2741afcdaea3SSage Weil if (ci->i_dirty_caps == 0) { 2742a8599bd8SSage Weil dout(" inode %p now clean\n", inode); 2743afcdaea3SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 2744afcdaea3SSage Weil drop = 1; 27457d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0) { 27467d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 27477d8cb26dSSage Weil ceph_put_snap_context(ci->i_head_snapc); 27487d8cb26dSSage Weil ci->i_head_snapc = NULL; 27497d8cb26dSSage Weil } 275076e3b390SSage Weil } else { 275176e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 2752afcdaea3SSage Weil } 2753a8599bd8SSage Weil } 2754a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 275503066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2756a8599bd8SSage Weil 2757a8599bd8SSage Weil out: 2758be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2759afcdaea3SSage Weil if (drop) 2760a8599bd8SSage Weil iput(inode); 2761a8599bd8SSage Weil } 2762a8599bd8SSage Weil 2763a8599bd8SSage Weil /* 2764a8599bd8SSage Weil * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can 2765a8599bd8SSage Weil * throw away our cap_snap. 2766a8599bd8SSage Weil * 2767a8599bd8SSage Weil * Caller hold s_mutex. 2768a8599bd8SSage Weil */ 27696df058c0SSage Weil static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, 2770a8599bd8SSage Weil struct ceph_mds_caps *m, 2771a8599bd8SSage Weil struct ceph_mds_session *session) 2772a8599bd8SSage Weil { 2773a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2774a8599bd8SSage Weil u64 follows = le64_to_cpu(m->snap_follows); 2775a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2776a8599bd8SSage Weil int drop = 0; 2777a8599bd8SSage Weil 2778a8599bd8SSage Weil dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", 2779a8599bd8SSage Weil inode, ci, session->s_mds, follows); 2780a8599bd8SSage Weil 2781be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2782a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2783a8599bd8SSage Weil if (capsnap->follows == follows) { 2784a8599bd8SSage Weil if (capsnap->flush_tid != flush_tid) { 2785a8599bd8SSage Weil dout(" cap_snap %p follows %lld tid %lld !=" 2786a8599bd8SSage Weil " %lld\n", capsnap, follows, 2787a8599bd8SSage Weil flush_tid, capsnap->flush_tid); 2788a8599bd8SSage Weil break; 2789a8599bd8SSage Weil } 2790a8599bd8SSage Weil WARN_ON(capsnap->dirty_pages || capsnap->writing); 2791819ccbfaSSage Weil dout(" removing %p cap_snap %p follows %lld\n", 2792819ccbfaSSage Weil inode, capsnap, follows); 2793a8599bd8SSage Weil ceph_put_snap_context(capsnap->context); 2794a8599bd8SSage Weil list_del(&capsnap->ci_item); 2795a8599bd8SSage Weil list_del(&capsnap->flushing_item); 2796a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 2797a8599bd8SSage Weil drop = 1; 2798a8599bd8SSage Weil break; 2799a8599bd8SSage Weil } else { 2800a8599bd8SSage Weil dout(" skipping cap_snap %p follows %lld\n", 2801a8599bd8SSage Weil capsnap, capsnap->follows); 2802a8599bd8SSage Weil } 2803a8599bd8SSage Weil } 2804be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2805a8599bd8SSage Weil if (drop) 2806a8599bd8SSage Weil iput(inode); 2807a8599bd8SSage Weil } 2808a8599bd8SSage Weil 2809a8599bd8SSage Weil /* 2810a8599bd8SSage Weil * Handle TRUNC from MDS, indicating file truncation. 2811a8599bd8SSage Weil * 2812a8599bd8SSage Weil * caller hold s_mutex. 2813a8599bd8SSage Weil */ 2814a8599bd8SSage Weil static void handle_cap_trunc(struct inode *inode, 2815a8599bd8SSage Weil struct ceph_mds_caps *trunc, 2816a8599bd8SSage Weil struct ceph_mds_session *session) 2817be655596SSage Weil __releases(ci->i_ceph_lock) 2818a8599bd8SSage Weil { 2819a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2820a8599bd8SSage Weil int mds = session->s_mds; 2821a8599bd8SSage Weil int seq = le32_to_cpu(trunc->seq); 2822a8599bd8SSage Weil u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); 2823a8599bd8SSage Weil u64 truncate_size = le64_to_cpu(trunc->truncate_size); 2824a8599bd8SSage Weil u64 size = le64_to_cpu(trunc->size); 2825a8599bd8SSage Weil int implemented = 0; 2826a8599bd8SSage Weil int dirty = __ceph_caps_dirty(ci); 2827a8599bd8SSage Weil int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); 2828a8599bd8SSage Weil int queue_trunc = 0; 2829a8599bd8SSage Weil 2830a8599bd8SSage Weil issued |= implemented | dirty; 2831a8599bd8SSage Weil 2832a8599bd8SSage Weil dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", 2833a8599bd8SSage Weil inode, mds, seq, truncate_size, truncate_seq); 2834a8599bd8SSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 2835a8599bd8SSage Weil truncate_seq, truncate_size, size); 2836be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2837a8599bd8SSage Weil 283899ccbd22SMilosz Tanski if (queue_trunc) { 28393c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 284099ccbd22SMilosz Tanski ceph_fscache_invalidate(inode); 284199ccbd22SMilosz Tanski } 2842a8599bd8SSage Weil } 2843a8599bd8SSage Weil 2844a8599bd8SSage Weil /* 2845a8599bd8SSage Weil * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a 2846a8599bd8SSage Weil * different one. If we are the most recent migration we've seen (as 2847a8599bd8SSage Weil * indicated by mseq), make note of the migrating cap bits for the 2848a8599bd8SSage Weil * duration (until we see the corresponding IMPORT). 2849a8599bd8SSage Weil * 2850a8599bd8SSage Weil * caller holds s_mutex 2851a8599bd8SSage Weil */ 2852a8599bd8SSage Weil static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, 285311df2dfbSYan, Zheng struct ceph_mds_cap_peer *ph, 285411df2dfbSYan, Zheng struct ceph_mds_session *session) 2855a8599bd8SSage Weil { 2856db354052SSage Weil struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 285711df2dfbSYan, Zheng struct ceph_mds_session *tsession = NULL; 2858d9df2783SYan, Zheng struct ceph_cap *cap, *tcap, *new_cap = NULL; 2859a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 286011df2dfbSYan, Zheng u64 t_cap_id; 2861a8599bd8SSage Weil unsigned mseq = le32_to_cpu(ex->migrate_seq); 286211df2dfbSYan, Zheng unsigned t_seq, t_mseq; 286311df2dfbSYan, Zheng int target, issued; 286411df2dfbSYan, Zheng int mds = session->s_mds; 2865a8599bd8SSage Weil 286611df2dfbSYan, Zheng if (ph) { 286711df2dfbSYan, Zheng t_cap_id = le64_to_cpu(ph->cap_id); 286811df2dfbSYan, Zheng t_seq = le32_to_cpu(ph->seq); 286911df2dfbSYan, Zheng t_mseq = le32_to_cpu(ph->mseq); 287011df2dfbSYan, Zheng target = le32_to_cpu(ph->mds); 287111df2dfbSYan, Zheng } else { 287211df2dfbSYan, Zheng t_cap_id = t_seq = t_mseq = 0; 287311df2dfbSYan, Zheng target = -1; 287411df2dfbSYan, Zheng } 2875a8599bd8SSage Weil 287611df2dfbSYan, Zheng dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n", 287711df2dfbSYan, Zheng inode, ci, mds, mseq, target); 287811df2dfbSYan, Zheng retry: 2879be655596SSage Weil spin_lock(&ci->i_ceph_lock); 288011df2dfbSYan, Zheng cap = __get_cap_for_mds(ci, mds); 2881ca665e02SYan, Zheng if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id)) 288211df2dfbSYan, Zheng goto out_unlock; 2883a8599bd8SSage Weil 288411df2dfbSYan, Zheng if (target < 0) { 288511df2dfbSYan, Zheng __ceph_remove_cap(cap, false); 288611df2dfbSYan, Zheng goto out_unlock; 2887a8599bd8SSage Weil } 2888a8599bd8SSage Weil 2889154f42c2SSage Weil /* 289011df2dfbSYan, Zheng * now we know we haven't received the cap import message yet 289111df2dfbSYan, Zheng * because the exported cap still exist. 2892154f42c2SSage Weil */ 2893db354052SSage Weil 289411df2dfbSYan, Zheng issued = cap->issued; 289511df2dfbSYan, Zheng WARN_ON(issued != cap->implemented); 289611df2dfbSYan, Zheng 289711df2dfbSYan, Zheng tcap = __get_cap_for_mds(ci, target); 289811df2dfbSYan, Zheng if (tcap) { 289911df2dfbSYan, Zheng /* already have caps from the target */ 290011df2dfbSYan, Zheng if (tcap->cap_id != t_cap_id || 290111df2dfbSYan, Zheng ceph_seq_cmp(tcap->seq, t_seq) < 0) { 290211df2dfbSYan, Zheng dout(" updating import cap %p mds%d\n", tcap, target); 290311df2dfbSYan, Zheng tcap->cap_id = t_cap_id; 290411df2dfbSYan, Zheng tcap->seq = t_seq - 1; 290511df2dfbSYan, Zheng tcap->issue_seq = t_seq - 1; 290611df2dfbSYan, Zheng tcap->mseq = t_mseq; 290711df2dfbSYan, Zheng tcap->issued |= issued; 290811df2dfbSYan, Zheng tcap->implemented |= issued; 290911df2dfbSYan, Zheng if (cap == ci->i_auth_cap) 291011df2dfbSYan, Zheng ci->i_auth_cap = tcap; 291111df2dfbSYan, Zheng if (ci->i_flushing_caps && ci->i_auth_cap == tcap) { 2912db354052SSage Weil spin_lock(&mdsc->cap_dirty_lock); 291311df2dfbSYan, Zheng list_move_tail(&ci->i_flushing_item, 291411df2dfbSYan, Zheng &tcap->session->s_cap_flushing); 2915db354052SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2916a8599bd8SSage Weil } 2917a8599bd8SSage Weil } 291811df2dfbSYan, Zheng __ceph_remove_cap(cap, false); 291911df2dfbSYan, Zheng goto out_unlock; 2920d9df2783SYan, Zheng } else if (tsession) { 292111df2dfbSYan, Zheng /* add placeholder for the export tagert */ 2922d9df2783SYan, Zheng int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0; 292311df2dfbSYan, Zheng ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0, 2924d9df2783SYan, Zheng t_seq - 1, t_mseq, (u64)-1, flag, &new_cap); 2925d9df2783SYan, Zheng 2926d9df2783SYan, Zheng __ceph_remove_cap(cap, false); 2927d9df2783SYan, Zheng goto out_unlock; 292811df2dfbSYan, Zheng } 2929a8599bd8SSage Weil 2930be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 293111df2dfbSYan, Zheng mutex_unlock(&session->s_mutex); 293211df2dfbSYan, Zheng 293311df2dfbSYan, Zheng /* open target session */ 293411df2dfbSYan, Zheng tsession = ceph_mdsc_open_export_target_session(mdsc, target); 293511df2dfbSYan, Zheng if (!IS_ERR(tsession)) { 293611df2dfbSYan, Zheng if (mds > target) { 293711df2dfbSYan, Zheng mutex_lock(&session->s_mutex); 293811df2dfbSYan, Zheng mutex_lock_nested(&tsession->s_mutex, 293911df2dfbSYan, Zheng SINGLE_DEPTH_NESTING); 294011df2dfbSYan, Zheng } else { 294111df2dfbSYan, Zheng mutex_lock(&tsession->s_mutex); 294211df2dfbSYan, Zheng mutex_lock_nested(&session->s_mutex, 294311df2dfbSYan, Zheng SINGLE_DEPTH_NESTING); 294411df2dfbSYan, Zheng } 294511df2dfbSYan, Zheng ceph_add_cap_releases(mdsc, tsession); 2946d9df2783SYan, Zheng new_cap = ceph_get_cap(mdsc, NULL); 294711df2dfbSYan, Zheng } else { 294811df2dfbSYan, Zheng WARN_ON(1); 294911df2dfbSYan, Zheng tsession = NULL; 295011df2dfbSYan, Zheng target = -1; 295111df2dfbSYan, Zheng } 295211df2dfbSYan, Zheng goto retry; 295311df2dfbSYan, Zheng 295411df2dfbSYan, Zheng out_unlock: 295511df2dfbSYan, Zheng spin_unlock(&ci->i_ceph_lock); 295611df2dfbSYan, Zheng mutex_unlock(&session->s_mutex); 295711df2dfbSYan, Zheng if (tsession) { 295811df2dfbSYan, Zheng mutex_unlock(&tsession->s_mutex); 295911df2dfbSYan, Zheng ceph_put_mds_session(tsession); 296011df2dfbSYan, Zheng } 2961d9df2783SYan, Zheng if (new_cap) 2962d9df2783SYan, Zheng ceph_put_cap(mdsc, new_cap); 2963a8599bd8SSage Weil } 2964a8599bd8SSage Weil 2965a8599bd8SSage Weil /* 29662cd698beSYan, Zheng * Handle cap IMPORT. 2967a8599bd8SSage Weil * 29682cd698beSYan, Zheng * caller holds s_mutex. acquires i_ceph_lock 2969a8599bd8SSage Weil */ 2970a8599bd8SSage Weil static void handle_cap_import(struct ceph_mds_client *mdsc, 2971a8599bd8SSage Weil struct inode *inode, struct ceph_mds_caps *im, 29724ee6a914SYan, Zheng struct ceph_mds_cap_peer *ph, 2973a8599bd8SSage Weil struct ceph_mds_session *session, 29742cd698beSYan, Zheng struct ceph_cap **target_cap, int *old_issued) 29752cd698beSYan, Zheng __acquires(ci->i_ceph_lock) 2976a8599bd8SSage Weil { 2977a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 29782cd698beSYan, Zheng struct ceph_cap *cap, *ocap, *new_cap = NULL; 2979a8599bd8SSage Weil int mds = session->s_mds; 29802cd698beSYan, Zheng int issued; 29812cd698beSYan, Zheng unsigned caps = le32_to_cpu(im->caps); 2982a8599bd8SSage Weil unsigned wanted = le32_to_cpu(im->wanted); 2983a8599bd8SSage Weil unsigned seq = le32_to_cpu(im->seq); 2984a8599bd8SSage Weil unsigned mseq = le32_to_cpu(im->migrate_seq); 2985a8599bd8SSage Weil u64 realmino = le64_to_cpu(im->realm); 2986a8599bd8SSage Weil u64 cap_id = le64_to_cpu(im->cap_id); 29874ee6a914SYan, Zheng u64 p_cap_id; 29884ee6a914SYan, Zheng int peer; 2989a8599bd8SSage Weil 29904ee6a914SYan, Zheng if (ph) { 29914ee6a914SYan, Zheng p_cap_id = le64_to_cpu(ph->cap_id); 29924ee6a914SYan, Zheng peer = le32_to_cpu(ph->mds); 2993a8599bd8SSage Weil } else { 29944ee6a914SYan, Zheng p_cap_id = 0; 29954ee6a914SYan, Zheng peer = -1; 2996a8599bd8SSage Weil } 2997a8599bd8SSage Weil 29984ee6a914SYan, Zheng dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n", 29994ee6a914SYan, Zheng inode, ci, mds, mseq, peer); 30004ee6a914SYan, Zheng 3001d9df2783SYan, Zheng retry: 30024ee6a914SYan, Zheng spin_lock(&ci->i_ceph_lock); 3003d9df2783SYan, Zheng cap = __get_cap_for_mds(ci, mds); 3004d9df2783SYan, Zheng if (!cap) { 3005d9df2783SYan, Zheng if (!new_cap) { 3006d9df2783SYan, Zheng spin_unlock(&ci->i_ceph_lock); 3007d9df2783SYan, Zheng new_cap = ceph_get_cap(mdsc, NULL); 3008d9df2783SYan, Zheng goto retry; 3009d9df2783SYan, Zheng } 30102cd698beSYan, Zheng cap = new_cap; 30112cd698beSYan, Zheng } else { 30122cd698beSYan, Zheng if (new_cap) { 30132cd698beSYan, Zheng ceph_put_cap(mdsc, new_cap); 30142cd698beSYan, Zheng new_cap = NULL; 30152cd698beSYan, Zheng } 3016d9df2783SYan, Zheng } 3017d9df2783SYan, Zheng 30182cd698beSYan, Zheng __ceph_caps_issued(ci, &issued); 30192cd698beSYan, Zheng issued |= __ceph_caps_dirty(ci); 30202cd698beSYan, Zheng 30212cd698beSYan, Zheng ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq, 3022d9df2783SYan, Zheng realmino, CEPH_CAP_FLAG_AUTH, &new_cap); 3023d9df2783SYan, Zheng 30242cd698beSYan, Zheng ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL; 30252cd698beSYan, Zheng if (ocap && ocap->cap_id == p_cap_id) { 30264ee6a914SYan, Zheng dout(" remove export cap %p mds%d flags %d\n", 30272cd698beSYan, Zheng ocap, peer, ph->flags); 30284ee6a914SYan, Zheng if ((ph->flags & CEPH_CAP_FLAG_AUTH) && 30292cd698beSYan, Zheng (ocap->seq != le32_to_cpu(ph->seq) || 30302cd698beSYan, Zheng ocap->mseq != le32_to_cpu(ph->mseq))) { 30314ee6a914SYan, Zheng pr_err("handle_cap_import: mismatched seq/mseq: " 30324ee6a914SYan, Zheng "ino (%llx.%llx) mds%d seq %d mseq %d " 30334ee6a914SYan, Zheng "importer mds%d has peer seq %d mseq %d\n", 30342cd698beSYan, Zheng ceph_vinop(inode), peer, ocap->seq, 30352cd698beSYan, Zheng ocap->mseq, mds, le32_to_cpu(ph->seq), 30364ee6a914SYan, Zheng le32_to_cpu(ph->mseq)); 30374ee6a914SYan, Zheng } 30382cd698beSYan, Zheng __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE)); 30394ee6a914SYan, Zheng } 30404ee6a914SYan, Zheng 30414ee6a914SYan, Zheng /* make sure we re-request max_size, if necessary */ 30424ee6a914SYan, Zheng ci->i_wanted_max_size = 0; 30434ee6a914SYan, Zheng ci->i_requested_max_size = 0; 30444ee6a914SYan, Zheng 30452cd698beSYan, Zheng *old_issued = issued; 30462cd698beSYan, Zheng *target_cap = cap; 3047a8599bd8SSage Weil } 3048a8599bd8SSage Weil 3049a8599bd8SSage Weil /* 3050a8599bd8SSage Weil * Handle a caps message from the MDS. 3051a8599bd8SSage Weil * 3052a8599bd8SSage Weil * Identify the appropriate session, inode, and call the right handler 3053a8599bd8SSage Weil * based on the cap op. 3054a8599bd8SSage Weil */ 3055a8599bd8SSage Weil void ceph_handle_caps(struct ceph_mds_session *session, 3056a8599bd8SSage Weil struct ceph_msg *msg) 3057a8599bd8SSage Weil { 3058a8599bd8SSage Weil struct ceph_mds_client *mdsc = session->s_mdsc; 30593d14c5d2SYehuda Sadeh struct super_block *sb = mdsc->fsc->sb; 3060a8599bd8SSage Weil struct inode *inode; 3061be655596SSage Weil struct ceph_inode_info *ci; 3062a8599bd8SSage Weil struct ceph_cap *cap; 3063a8599bd8SSage Weil struct ceph_mds_caps *h; 30644ee6a914SYan, Zheng struct ceph_mds_cap_peer *peer = NULL; 3065982d6011SYan, Zheng struct ceph_snap_realm *realm; 30662600d2ddSSage Weil int mds = session->s_mds; 30672cd698beSYan, Zheng int op, issued; 30683d7ded4dSSage Weil u32 seq, mseq; 3069a8599bd8SSage Weil struct ceph_vino vino; 3070a8599bd8SSage Weil u64 cap_id; 3071a8599bd8SSage Weil u64 size, max_size; 30726df058c0SSage Weil u64 tid; 3073fb01d1f8SYan, Zheng u64 inline_version = 0; 3074fb01d1f8SYan, Zheng void *inline_data = NULL; 3075fb01d1f8SYan, Zheng u32 inline_len = 0; 307670edb55bSSage Weil void *snaptrace; 3077ce1fbc8dSSage Weil size_t snaptrace_len; 3078fb01d1f8SYan, Zheng void *p, *end; 3079a8599bd8SSage Weil 3080a8599bd8SSage Weil dout("handle_caps from mds%d\n", mds); 3081a8599bd8SSage Weil 3082a8599bd8SSage Weil /* decode */ 30834ee6a914SYan, Zheng end = msg->front.iov_base + msg->front.iov_len; 30846df058c0SSage Weil tid = le64_to_cpu(msg->hdr.tid); 3085a8599bd8SSage Weil if (msg->front.iov_len < sizeof(*h)) 3086a8599bd8SSage Weil goto bad; 3087a8599bd8SSage Weil h = msg->front.iov_base; 3088a8599bd8SSage Weil op = le32_to_cpu(h->op); 3089a8599bd8SSage Weil vino.ino = le64_to_cpu(h->ino); 3090a8599bd8SSage Weil vino.snap = CEPH_NOSNAP; 3091a8599bd8SSage Weil cap_id = le64_to_cpu(h->cap_id); 3092a8599bd8SSage Weil seq = le32_to_cpu(h->seq); 30933d7ded4dSSage Weil mseq = le32_to_cpu(h->migrate_seq); 3094a8599bd8SSage Weil size = le64_to_cpu(h->size); 3095a8599bd8SSage Weil max_size = le64_to_cpu(h->max_size); 3096a8599bd8SSage Weil 3097ce1fbc8dSSage Weil snaptrace = h + 1; 3098ce1fbc8dSSage Weil snaptrace_len = le32_to_cpu(h->snap_trace_len); 3099fb01d1f8SYan, Zheng p = snaptrace + snaptrace_len; 3100ce1fbc8dSSage Weil 3101ce1fbc8dSSage Weil if (le16_to_cpu(msg->hdr.version) >= 2) { 3102fb01d1f8SYan, Zheng u32 flock_len; 3103ce1fbc8dSSage Weil ceph_decode_32_safe(&p, end, flock_len, bad); 31044ee6a914SYan, Zheng if (p + flock_len > end) 31054ee6a914SYan, Zheng goto bad; 3106fb01d1f8SYan, Zheng p += flock_len; 3107ce1fbc8dSSage Weil } 3108ce1fbc8dSSage Weil 31094ee6a914SYan, Zheng if (le16_to_cpu(msg->hdr.version) >= 3) { 31104ee6a914SYan, Zheng if (op == CEPH_CAP_OP_IMPORT) { 31114ee6a914SYan, Zheng if (p + sizeof(*peer) > end) 31124ee6a914SYan, Zheng goto bad; 31134ee6a914SYan, Zheng peer = p; 3114fb01d1f8SYan, Zheng p += sizeof(*peer); 311511df2dfbSYan, Zheng } else if (op == CEPH_CAP_OP_EXPORT) { 311611df2dfbSYan, Zheng /* recorded in unused fields */ 311711df2dfbSYan, Zheng peer = (void *)&h->size; 31184ee6a914SYan, Zheng } 31194ee6a914SYan, Zheng } 31204ee6a914SYan, Zheng 3121fb01d1f8SYan, Zheng if (le16_to_cpu(msg->hdr.version) >= 4) { 3122fb01d1f8SYan, Zheng ceph_decode_64_safe(&p, end, inline_version, bad); 3123fb01d1f8SYan, Zheng ceph_decode_32_safe(&p, end, inline_len, bad); 3124fb01d1f8SYan, Zheng if (p + inline_len > end) 3125fb01d1f8SYan, Zheng goto bad; 3126fb01d1f8SYan, Zheng inline_data = p; 3127fb01d1f8SYan, Zheng p += inline_len; 3128fb01d1f8SYan, Zheng } 3129fb01d1f8SYan, Zheng 31306cd3bcadSYan, Zheng /* lookup ino */ 31316cd3bcadSYan, Zheng inode = ceph_find_inode(sb, vino); 31326cd3bcadSYan, Zheng ci = ceph_inode(inode); 31336cd3bcadSYan, Zheng dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, 31346cd3bcadSYan, Zheng vino.snap, inode); 31356cd3bcadSYan, Zheng 3136a8599bd8SSage Weil mutex_lock(&session->s_mutex); 3137a8599bd8SSage Weil session->s_seq++; 3138a8599bd8SSage Weil dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, 3139a8599bd8SSage Weil (unsigned)seq); 3140a8599bd8SSage Weil 314166f58691SYan, Zheng if (op == CEPH_CAP_OP_IMPORT) 314266f58691SYan, Zheng ceph_add_cap_releases(mdsc, session); 314366f58691SYan, Zheng 3144a8599bd8SSage Weil if (!inode) { 3145a8599bd8SSage Weil dout(" i don't have ino %llx\n", vino.ino); 31463d7ded4dSSage Weil 3147a096b09aSYan, Zheng if (op == CEPH_CAP_OP_IMPORT) { 3148a096b09aSYan, Zheng spin_lock(&session->s_cap_lock); 31493d7ded4dSSage Weil __queue_cap_release(session, vino.ino, cap_id, 31503d7ded4dSSage Weil mseq, seq); 3151a096b09aSYan, Zheng spin_unlock(&session->s_cap_lock); 3152a096b09aSYan, Zheng } 315321b559deSGreg Farnum goto flush_cap_releases; 3154a8599bd8SSage Weil } 3155a8599bd8SSage Weil 3156a8599bd8SSage Weil /* these will work even if we don't have a cap yet */ 3157a8599bd8SSage Weil switch (op) { 3158a8599bd8SSage Weil case CEPH_CAP_OP_FLUSHSNAP_ACK: 31596df058c0SSage Weil handle_cap_flushsnap_ack(inode, tid, h, session); 3160a8599bd8SSage Weil goto done; 3161a8599bd8SSage Weil 3162a8599bd8SSage Weil case CEPH_CAP_OP_EXPORT: 316311df2dfbSYan, Zheng handle_cap_export(inode, h, peer, session); 316411df2dfbSYan, Zheng goto done_unlocked; 3165a8599bd8SSage Weil 3166a8599bd8SSage Weil case CEPH_CAP_OP_IMPORT: 3167982d6011SYan, Zheng realm = NULL; 3168982d6011SYan, Zheng if (snaptrace_len) { 3169982d6011SYan, Zheng down_write(&mdsc->snap_rwsem); 3170982d6011SYan, Zheng ceph_update_snap_trace(mdsc, snaptrace, 3171982d6011SYan, Zheng snaptrace + snaptrace_len, 3172982d6011SYan, Zheng false, &realm); 3173982d6011SYan, Zheng downgrade_write(&mdsc->snap_rwsem); 3174982d6011SYan, Zheng } else { 3175982d6011SYan, Zheng down_read(&mdsc->snap_rwsem); 3176982d6011SYan, Zheng } 31774ee6a914SYan, Zheng handle_cap_import(mdsc, inode, h, peer, session, 31782cd698beSYan, Zheng &cap, &issued); 3179982d6011SYan, Zheng handle_cap_grant(mdsc, inode, h, 3180fb01d1f8SYan, Zheng inline_version, inline_data, inline_len, 31812cd698beSYan, Zheng msg->middle, session, cap, issued); 3182982d6011SYan, Zheng if (realm) 3183982d6011SYan, Zheng ceph_put_snap_realm(mdsc, realm); 31842cd698beSYan, Zheng goto done_unlocked; 3185a8599bd8SSage Weil } 3186a8599bd8SSage Weil 3187a8599bd8SSage Weil /* the rest require a cap */ 3188be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3189a8599bd8SSage Weil cap = __get_cap_for_mds(ceph_inode(inode), mds); 3190a8599bd8SSage Weil if (!cap) { 31919dbd412fSSage Weil dout(" no cap on %p ino %llx.%llx from mds%d\n", 3192a8599bd8SSage Weil inode, ceph_ino(inode), ceph_snap(inode), mds); 3193be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 319421b559deSGreg Farnum goto flush_cap_releases; 3195a8599bd8SSage Weil } 3196a8599bd8SSage Weil 3197be655596SSage Weil /* note that each of these drops i_ceph_lock for us */ 3198a8599bd8SSage Weil switch (op) { 3199a8599bd8SSage Weil case CEPH_CAP_OP_REVOKE: 3200a8599bd8SSage Weil case CEPH_CAP_OP_GRANT: 32012cd698beSYan, Zheng __ceph_caps_issued(ci, &issued); 32022cd698beSYan, Zheng issued |= __ceph_caps_dirty(ci); 3203982d6011SYan, Zheng handle_cap_grant(mdsc, inode, h, 3204fb01d1f8SYan, Zheng inline_version, inline_data, inline_len, 3205fb01d1f8SYan, Zheng msg->middle, session, cap, issued); 320615637c8bSSage Weil goto done_unlocked; 3207a8599bd8SSage Weil 3208a8599bd8SSage Weil case CEPH_CAP_OP_FLUSH_ACK: 32096df058c0SSage Weil handle_cap_flush_ack(inode, tid, h, session, cap); 3210a8599bd8SSage Weil break; 3211a8599bd8SSage Weil 3212a8599bd8SSage Weil case CEPH_CAP_OP_TRUNC: 3213a8599bd8SSage Weil handle_cap_trunc(inode, h, session); 3214a8599bd8SSage Weil break; 3215a8599bd8SSage Weil 3216a8599bd8SSage Weil default: 3217be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3218a8599bd8SSage Weil pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, 3219a8599bd8SSage Weil ceph_cap_op_name(op)); 3220a8599bd8SSage Weil } 3221a8599bd8SSage Weil 322221b559deSGreg Farnum goto done; 322321b559deSGreg Farnum 322421b559deSGreg Farnum flush_cap_releases: 322521b559deSGreg Farnum /* 322621b559deSGreg Farnum * send any full release message to try to move things 322721b559deSGreg Farnum * along for the mds (who clearly thinks we still have this 322821b559deSGreg Farnum * cap). 322921b559deSGreg Farnum */ 323021b559deSGreg Farnum ceph_add_cap_releases(mdsc, session); 323121b559deSGreg Farnum ceph_send_cap_releases(mdsc, session); 323221b559deSGreg Farnum 3233a8599bd8SSage Weil done: 3234a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 323515637c8bSSage Weil done_unlocked: 3236a8599bd8SSage Weil iput(inode); 3237a8599bd8SSage Weil return; 3238a8599bd8SSage Weil 3239a8599bd8SSage Weil bad: 3240a8599bd8SSage Weil pr_err("ceph_handle_caps: corrupt message\n"); 32419ec7cab1SSage Weil ceph_msg_dump(msg); 3242a8599bd8SSage Weil return; 3243a8599bd8SSage Weil } 3244a8599bd8SSage Weil 3245a8599bd8SSage Weil /* 3246a8599bd8SSage Weil * Delayed work handler to process end of delayed cap release LRU list. 3247a8599bd8SSage Weil */ 3248afcdaea3SSage Weil void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) 3249a8599bd8SSage Weil { 3250a8599bd8SSage Weil struct ceph_inode_info *ci; 3251a8599bd8SSage Weil int flags = CHECK_CAPS_NODELAY; 3252a8599bd8SSage Weil 3253a8599bd8SSage Weil dout("check_delayed_caps\n"); 3254a8599bd8SSage Weil while (1) { 3255a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 3256a8599bd8SSage Weil if (list_empty(&mdsc->cap_delay_list)) 3257a8599bd8SSage Weil break; 3258a8599bd8SSage Weil ci = list_first_entry(&mdsc->cap_delay_list, 3259a8599bd8SSage Weil struct ceph_inode_info, 3260a8599bd8SSage Weil i_cap_delay_list); 3261a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && 3262a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) 3263a8599bd8SSage Weil break; 3264a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 3265a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 3266a8599bd8SSage Weil dout("check_delayed_caps on %p\n", &ci->vfs_inode); 3267a8599bd8SSage Weil ceph_check_caps(ci, flags, NULL); 3268a8599bd8SSage Weil } 3269a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 3270a8599bd8SSage Weil } 3271a8599bd8SSage Weil 3272a8599bd8SSage Weil /* 3273afcdaea3SSage Weil * Flush all dirty caps to the mds 3274afcdaea3SSage Weil */ 3275afcdaea3SSage Weil void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) 3276afcdaea3SSage Weil { 3277db354052SSage Weil struct ceph_inode_info *ci; 3278db354052SSage Weil struct inode *inode; 3279afcdaea3SSage Weil 3280afcdaea3SSage Weil dout("flush_dirty_caps\n"); 3281afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 3282db354052SSage Weil while (!list_empty(&mdsc->cap_dirty)) { 3283db354052SSage Weil ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info, 3284afcdaea3SSage Weil i_dirty_item); 328570b666c3SSage Weil inode = &ci->vfs_inode; 328670b666c3SSage Weil ihold(inode); 3287db354052SSage Weil dout("flush_dirty_caps %p\n", inode); 3288afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 328970b666c3SSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL); 3290afcdaea3SSage Weil iput(inode); 3291afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 3292afcdaea3SSage Weil } 3293afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 3294db354052SSage Weil dout("flush_dirty_caps done\n"); 3295afcdaea3SSage Weil } 3296afcdaea3SSage Weil 3297afcdaea3SSage Weil /* 3298a8599bd8SSage Weil * Drop open file reference. If we were the last open file, 3299a8599bd8SSage Weil * we may need to release capabilities to the MDS (or schedule 3300a8599bd8SSage Weil * their delayed release). 3301a8599bd8SSage Weil */ 3302a8599bd8SSage Weil void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) 3303a8599bd8SSage Weil { 3304a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 3305a8599bd8SSage Weil int last = 0; 3306a8599bd8SSage Weil 3307be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3308a8599bd8SSage Weil dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode, 3309a8599bd8SSage Weil ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1); 3310a8599bd8SSage Weil BUG_ON(ci->i_nr_by_mode[fmode] == 0); 3311a8599bd8SSage Weil if (--ci->i_nr_by_mode[fmode] == 0) 3312a8599bd8SSage Weil last++; 3313be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3314a8599bd8SSage Weil 3315a8599bd8SSage Weil if (last && ci->i_vino.snap == CEPH_NOSNAP) 3316a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 3317a8599bd8SSage Weil } 3318a8599bd8SSage Weil 3319a8599bd8SSage Weil /* 3320a8599bd8SSage Weil * Helpers for embedding cap and dentry lease releases into mds 3321a8599bd8SSage Weil * requests. 3322a8599bd8SSage Weil * 3323a8599bd8SSage Weil * @force is used by dentry_release (below) to force inclusion of a 3324a8599bd8SSage Weil * record for the directory inode, even when there aren't any caps to 3325a8599bd8SSage Weil * drop. 3326a8599bd8SSage Weil */ 3327a8599bd8SSage Weil int ceph_encode_inode_release(void **p, struct inode *inode, 3328a8599bd8SSage Weil int mds, int drop, int unless, int force) 3329a8599bd8SSage Weil { 3330a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 3331a8599bd8SSage Weil struct ceph_cap *cap; 3332a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3333ec97f88bSSage Weil int used, dirty; 3334a8599bd8SSage Weil int ret = 0; 3335a8599bd8SSage Weil 3336be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3337916623daSSage Weil used = __ceph_caps_used(ci); 3338ec97f88bSSage Weil dirty = __ceph_caps_dirty(ci); 3339916623daSSage Weil 3340ec97f88bSSage Weil dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", 3341ec97f88bSSage Weil inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), 3342916623daSSage Weil ceph_cap_string(unless)); 3343916623daSSage Weil 3344ec97f88bSSage Weil /* only drop unused, clean caps */ 3345ec97f88bSSage Weil drop &= ~(used | dirty); 3346916623daSSage Weil 3347a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 3348a8599bd8SSage Weil if (cap && __cap_is_valid(cap)) { 3349a8599bd8SSage Weil if (force || 3350a8599bd8SSage Weil ((cap->issued & drop) && 3351a8599bd8SSage Weil (cap->issued & unless) == 0)) { 3352a8599bd8SSage Weil if ((cap->issued & drop) && 3353a8599bd8SSage Weil (cap->issued & unless) == 0) { 3354bb137f84SYan, Zheng int wanted = __ceph_caps_wanted(ci); 3355bb137f84SYan, Zheng if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0) 3356bb137f84SYan, Zheng wanted |= cap->mds_wanted; 3357bb137f84SYan, Zheng dout("encode_inode_release %p cap %p " 3358bb137f84SYan, Zheng "%s -> %s, wanted %s -> %s\n", inode, cap, 3359a8599bd8SSage Weil ceph_cap_string(cap->issued), 3360bb137f84SYan, Zheng ceph_cap_string(cap->issued & ~drop), 3361bb137f84SYan, Zheng ceph_cap_string(cap->mds_wanted), 3362bb137f84SYan, Zheng ceph_cap_string(wanted)); 3363bb137f84SYan, Zheng 3364a8599bd8SSage Weil cap->issued &= ~drop; 3365a8599bd8SSage Weil cap->implemented &= ~drop; 3366bb137f84SYan, Zheng cap->mds_wanted = wanted; 3367a8599bd8SSage Weil } else { 3368a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s" 3369a8599bd8SSage Weil " (force)\n", inode, cap, 3370a8599bd8SSage Weil ceph_cap_string(cap->issued)); 3371a8599bd8SSage Weil } 3372a8599bd8SSage Weil 3373a8599bd8SSage Weil rel->ino = cpu_to_le64(ceph_ino(inode)); 3374a8599bd8SSage Weil rel->cap_id = cpu_to_le64(cap->cap_id); 3375a8599bd8SSage Weil rel->seq = cpu_to_le32(cap->seq); 337608a0f24eSHimangi Saraogi rel->issue_seq = cpu_to_le32(cap->issue_seq); 3377a8599bd8SSage Weil rel->mseq = cpu_to_le32(cap->mseq); 3378fd7b95cdSYan, Zheng rel->caps = cpu_to_le32(cap->implemented); 3379a8599bd8SSage Weil rel->wanted = cpu_to_le32(cap->mds_wanted); 3380a8599bd8SSage Weil rel->dname_len = 0; 3381a8599bd8SSage Weil rel->dname_seq = 0; 3382a8599bd8SSage Weil *p += sizeof(*rel); 3383a8599bd8SSage Weil ret = 1; 3384a8599bd8SSage Weil } else { 3385a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s\n", 3386a8599bd8SSage Weil inode, cap, ceph_cap_string(cap->issued)); 3387a8599bd8SSage Weil } 3388a8599bd8SSage Weil } 3389be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3390a8599bd8SSage Weil return ret; 3391a8599bd8SSage Weil } 3392a8599bd8SSage Weil 3393a8599bd8SSage Weil int ceph_encode_dentry_release(void **p, struct dentry *dentry, 3394a8599bd8SSage Weil int mds, int drop, int unless) 3395a8599bd8SSage Weil { 3396a8599bd8SSage Weil struct inode *dir = dentry->d_parent->d_inode; 3397a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3398a8599bd8SSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 3399a8599bd8SSage Weil int force = 0; 3400a8599bd8SSage Weil int ret; 3401a8599bd8SSage Weil 3402a8599bd8SSage Weil /* 3403a8599bd8SSage Weil * force an record for the directory caps if we have a dentry lease. 3404be655596SSage Weil * this is racy (can't take i_ceph_lock and d_lock together), but it 3405a8599bd8SSage Weil * doesn't have to be perfect; the mds will revoke anything we don't 3406a8599bd8SSage Weil * release. 3407a8599bd8SSage Weil */ 3408a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3409a8599bd8SSage Weil if (di->lease_session && di->lease_session->s_mds == mds) 3410a8599bd8SSage Weil force = 1; 3411a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3412a8599bd8SSage Weil 3413a8599bd8SSage Weil ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); 3414a8599bd8SSage Weil 3415a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3416a8599bd8SSage Weil if (ret && di->lease_session && di->lease_session->s_mds == mds) { 3417a8599bd8SSage Weil dout("encode_dentry_release %p mds%d seq %d\n", 3418a8599bd8SSage Weil dentry, mds, (int)di->lease_seq); 3419a8599bd8SSage Weil rel->dname_len = cpu_to_le32(dentry->d_name.len); 3420a8599bd8SSage Weil memcpy(*p, dentry->d_name.name, dentry->d_name.len); 3421a8599bd8SSage Weil *p += dentry->d_name.len; 3422a8599bd8SSage Weil rel->dname_seq = cpu_to_le32(di->lease_seq); 34231dadcce3SSage Weil __ceph_mdsc_drop_dentry_lease(dentry); 3424a8599bd8SSage Weil } 3425a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3426a8599bd8SSage Weil return ret; 3427a8599bd8SSage Weil } 3428