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 899db40cc17SYan, Zheng static void drop_inode_snap_realm(struct ceph_inode_info *ci) 900db40cc17SYan, Zheng { 901db40cc17SYan, Zheng struct ceph_snap_realm *realm = ci->i_snap_realm; 902db40cc17SYan, Zheng spin_lock(&realm->inodes_with_caps_lock); 903db40cc17SYan, Zheng list_del_init(&ci->i_snap_realm_item); 904db40cc17SYan, Zheng ci->i_snap_realm_counter++; 905db40cc17SYan, Zheng ci->i_snap_realm = NULL; 906db40cc17SYan, Zheng spin_unlock(&realm->inodes_with_caps_lock); 907db40cc17SYan, Zheng ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc, 908db40cc17SYan, Zheng realm); 909db40cc17SYan, Zheng } 910db40cc17SYan, Zheng 911a8599bd8SSage Weil /* 912f818a736SSage Weil * Remove a cap. Take steps to deal with a racing iterate_session_caps. 913f818a736SSage Weil * 914be655596SSage Weil * caller should hold i_ceph_lock. 915a6369741SSage Weil * caller will not hold session s_mutex if called from destroy_inode. 916a8599bd8SSage Weil */ 917a096b09aSYan, Zheng void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) 918a8599bd8SSage Weil { 919a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 920a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 921640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 9223d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 923f818a736SSage Weil int removed = 0; 924a8599bd8SSage Weil 925a8599bd8SSage Weil dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); 926a8599bd8SSage Weil 9277c1332b8SSage Weil /* remove from session list */ 9287c1332b8SSage Weil spin_lock(&session->s_cap_lock); 92999a9c273SYan, Zheng /* 93099a9c273SYan, Zheng * s_cap_reconnect is protected by s_cap_lock. no one changes 93199a9c273SYan, Zheng * s_cap_gen while session is in the reconnect state. 93299a9c273SYan, Zheng */ 93399a9c273SYan, Zheng if (queue_release && 93499a9c273SYan, Zheng (!session->s_cap_reconnect || 93599a9c273SYan, Zheng cap->cap_gen == session->s_cap_gen)) 936a096b09aSYan, Zheng __queue_cap_release(session, ci->i_vino.ino, cap->cap_id, 937a096b09aSYan, Zheng cap->mseq, cap->issue_seq); 938a096b09aSYan, Zheng 9397c1332b8SSage Weil if (session->s_cap_iterator == cap) { 9407c1332b8SSage Weil /* not yet, we are iterating over this very cap */ 9417c1332b8SSage Weil dout("__ceph_remove_cap delaying %p removal from session %p\n", 9427c1332b8SSage Weil cap, cap->session); 9437c1332b8SSage Weil } else { 9447c1332b8SSage Weil list_del_init(&cap->session_caps); 9457c1332b8SSage Weil session->s_nr_caps--; 9467c1332b8SSage Weil cap->session = NULL; 947f818a736SSage Weil removed = 1; 9487c1332b8SSage Weil } 949f818a736SSage Weil /* protect backpointer with s_cap_lock: see iterate_session_caps */ 950f818a736SSage Weil cap->ci = NULL; 9517c1332b8SSage Weil spin_unlock(&session->s_cap_lock); 9527c1332b8SSage Weil 953f818a736SSage Weil /* remove from inode list */ 954f818a736SSage Weil rb_erase(&cap->ci_node, &ci->i_caps); 955f818a736SSage Weil if (ci->i_auth_cap == cap) 956f818a736SSage Weil ci->i_auth_cap = NULL; 957f818a736SSage Weil 958f818a736SSage Weil if (removed) 95937151668SYehuda Sadeh ceph_put_cap(mdsc, cap); 960a8599bd8SSage Weil 961db40cc17SYan, Zheng /* when reconnect denied, we remove session caps forcibly, 962db40cc17SYan, Zheng * i_wr_ref can be non-zero. If there are ongoing write, 963db40cc17SYan, Zheng * keep i_snap_realm. 964db40cc17SYan, Zheng */ 965db40cc17SYan, Zheng if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm) 966db40cc17SYan, Zheng drop_inode_snap_realm(ci); 967db40cc17SYan, Zheng 968a8599bd8SSage Weil if (!__ceph_is_any_real_caps(ci)) 969a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 970a8599bd8SSage Weil } 971a8599bd8SSage Weil 972a8599bd8SSage Weil /* 973a8599bd8SSage Weil * Build and send a cap message to the given MDS. 974a8599bd8SSage Weil * 975a8599bd8SSage Weil * Caller should be holding s_mutex. 976a8599bd8SSage Weil */ 977a8599bd8SSage Weil static int send_cap_msg(struct ceph_mds_session *session, 978a8599bd8SSage Weil u64 ino, u64 cid, int op, 979a8599bd8SSage Weil int caps, int wanted, int dirty, 980a8599bd8SSage Weil u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq, 981a8599bd8SSage Weil u64 size, u64 max_size, 982a8599bd8SSage Weil struct timespec *mtime, struct timespec *atime, 983a8599bd8SSage Weil u64 time_warp_seq, 98405cb11c1SEric W. Biederman kuid_t uid, kgid_t gid, umode_t mode, 985a8599bd8SSage Weil u64 xattr_version, 986a8599bd8SSage Weil struct ceph_buffer *xattrs_buf, 987e20d258dSYan, Zheng u64 follows, bool inline_data) 988a8599bd8SSage Weil { 989a8599bd8SSage Weil struct ceph_mds_caps *fc; 990a8599bd8SSage Weil struct ceph_msg *msg; 991e20d258dSYan, Zheng void *p; 992e20d258dSYan, Zheng size_t extra_len; 993a8599bd8SSage Weil 994a8599bd8SSage Weil dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" 995a8599bd8SSage Weil " seq %u/%u mseq %u follows %lld size %llu/%llu" 996a8599bd8SSage Weil " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op), 997a8599bd8SSage Weil cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted), 998a8599bd8SSage Weil ceph_cap_string(dirty), 999a8599bd8SSage Weil seq, issue_seq, mseq, follows, size, max_size, 1000a8599bd8SSage Weil xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0); 1001a8599bd8SSage Weil 1002e20d258dSYan, Zheng /* flock buffer size + inline version + inline data size */ 1003e20d258dSYan, Zheng extra_len = 4 + 8 + 4; 1004e20d258dSYan, Zheng msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len, 1005e20d258dSYan, Zheng GFP_NOFS, false); 1006a79832f2SSage Weil if (!msg) 1007a79832f2SSage Weil return -ENOMEM; 1008a8599bd8SSage Weil 10096df058c0SSage Weil msg->hdr.tid = cpu_to_le64(flush_tid); 1010a8599bd8SSage Weil 10116df058c0SSage Weil fc = msg->front.iov_base; 1012a8599bd8SSage Weil memset(fc, 0, sizeof(*fc)); 1013a8599bd8SSage Weil 1014a8599bd8SSage Weil fc->cap_id = cpu_to_le64(cid); 1015a8599bd8SSage Weil fc->op = cpu_to_le32(op); 1016a8599bd8SSage Weil fc->seq = cpu_to_le32(seq); 1017a8599bd8SSage Weil fc->issue_seq = cpu_to_le32(issue_seq); 1018a8599bd8SSage Weil fc->migrate_seq = cpu_to_le32(mseq); 1019a8599bd8SSage Weil fc->caps = cpu_to_le32(caps); 1020a8599bd8SSage Weil fc->wanted = cpu_to_le32(wanted); 1021a8599bd8SSage Weil fc->dirty = cpu_to_le32(dirty); 1022a8599bd8SSage Weil fc->ino = cpu_to_le64(ino); 1023a8599bd8SSage Weil fc->snap_follows = cpu_to_le64(follows); 1024a8599bd8SSage Weil 1025a8599bd8SSage Weil fc->size = cpu_to_le64(size); 1026a8599bd8SSage Weil fc->max_size = cpu_to_le64(max_size); 1027a8599bd8SSage Weil if (mtime) 1028a8599bd8SSage Weil ceph_encode_timespec(&fc->mtime, mtime); 1029a8599bd8SSage Weil if (atime) 1030a8599bd8SSage Weil ceph_encode_timespec(&fc->atime, atime); 1031a8599bd8SSage Weil fc->time_warp_seq = cpu_to_le32(time_warp_seq); 1032a8599bd8SSage Weil 103305cb11c1SEric W. Biederman fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid)); 103405cb11c1SEric W. Biederman fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid)); 1035a8599bd8SSage Weil fc->mode = cpu_to_le32(mode); 1036a8599bd8SSage Weil 1037e20d258dSYan, Zheng p = fc + 1; 1038e20d258dSYan, Zheng /* flock buffer size */ 1039e20d258dSYan, Zheng ceph_encode_32(&p, 0); 1040e20d258dSYan, Zheng /* inline version */ 1041e20d258dSYan, Zheng ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE); 1042e20d258dSYan, Zheng /* inline data size */ 1043e20d258dSYan, Zheng ceph_encode_32(&p, 0); 1044e20d258dSYan, Zheng 1045a8599bd8SSage Weil fc->xattr_version = cpu_to_le64(xattr_version); 1046a8599bd8SSage Weil if (xattrs_buf) { 1047a8599bd8SSage Weil msg->middle = ceph_buffer_get(xattrs_buf); 1048a8599bd8SSage Weil fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len); 1049a8599bd8SSage Weil msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len); 1050a8599bd8SSage Weil } 1051a8599bd8SSage Weil 1052a8599bd8SSage Weil ceph_con_send(&session->s_con, msg); 1053a8599bd8SSage Weil return 0; 1054a8599bd8SSage Weil } 1055a8599bd8SSage Weil 1056d40ee0dcSYan, Zheng void __queue_cap_release(struct ceph_mds_session *session, 10573d7ded4dSSage Weil u64 ino, u64 cap_id, u32 migrate_seq, 10583d7ded4dSSage Weil u32 issue_seq) 10593d7ded4dSSage Weil { 10603d7ded4dSSage Weil struct ceph_msg *msg; 10613d7ded4dSSage Weil struct ceph_mds_cap_release *head; 10623d7ded4dSSage Weil struct ceph_mds_cap_item *item; 10633d7ded4dSSage Weil 10643d7ded4dSSage Weil BUG_ON(!session->s_num_cap_releases); 10653d7ded4dSSage Weil msg = list_first_entry(&session->s_cap_releases, 10663d7ded4dSSage Weil struct ceph_msg, list_head); 10673d7ded4dSSage Weil 10683d7ded4dSSage Weil dout(" adding %llx release to mds%d msg %p (%d left)\n", 10693d7ded4dSSage Weil ino, session->s_mds, msg, session->s_num_cap_releases); 10703d7ded4dSSage Weil 10713d7ded4dSSage Weil BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE); 10723d7ded4dSSage Weil head = msg->front.iov_base; 1073b905a7f8SWei Yongjun le32_add_cpu(&head->num, 1); 10743d7ded4dSSage Weil item = msg->front.iov_base + msg->front.iov_len; 10753d7ded4dSSage Weil item->ino = cpu_to_le64(ino); 10763d7ded4dSSage Weil item->cap_id = cpu_to_le64(cap_id); 10773d7ded4dSSage Weil item->migrate_seq = cpu_to_le32(migrate_seq); 10783d7ded4dSSage Weil item->seq = cpu_to_le32(issue_seq); 10793d7ded4dSSage Weil 10803d7ded4dSSage Weil session->s_num_cap_releases--; 10813d7ded4dSSage Weil 10823d7ded4dSSage Weil msg->front.iov_len += sizeof(*item); 10833d7ded4dSSage Weil if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) { 10843d7ded4dSSage Weil dout(" release msg %p full\n", msg); 10853d7ded4dSSage Weil list_move_tail(&msg->list_head, &session->s_cap_releases_done); 10863d7ded4dSSage Weil } else { 10873d7ded4dSSage Weil dout(" release msg %p at %d/%d (%d)\n", msg, 10883d7ded4dSSage Weil (int)le32_to_cpu(head->num), 10893d7ded4dSSage Weil (int)CEPH_CAPS_PER_RELEASE, 10903d7ded4dSSage Weil (int)msg->front.iov_len); 10913d7ded4dSSage Weil } 10923d7ded4dSSage Weil } 10933d7ded4dSSage Weil 1094a8599bd8SSage Weil /* 1095a6369741SSage Weil * Queue cap releases when an inode is dropped from our cache. Since 1096be655596SSage Weil * inode is about to be destroyed, there is no need for i_ceph_lock. 1097a8599bd8SSage Weil */ 1098a8599bd8SSage Weil void ceph_queue_caps_release(struct inode *inode) 1099a8599bd8SSage Weil { 1100a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1101a8599bd8SSage Weil struct rb_node *p; 1102a8599bd8SSage Weil 1103a8599bd8SSage Weil p = rb_first(&ci->i_caps); 1104a8599bd8SSage Weil while (p) { 1105a8599bd8SSage Weil struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); 1106a8599bd8SSage Weil p = rb_next(p); 1107a096b09aSYan, Zheng __ceph_remove_cap(cap, true); 1108a8599bd8SSage Weil } 1109a8599bd8SSage Weil } 1110a8599bd8SSage Weil 1111a8599bd8SSage Weil /* 1112a8599bd8SSage Weil * Send a cap msg on the given inode. Update our caps state, then 1113be655596SSage Weil * drop i_ceph_lock and send the message. 1114a8599bd8SSage Weil * 1115a8599bd8SSage Weil * Make note of max_size reported/requested from mds, revoked caps 1116a8599bd8SSage Weil * that have now been implemented. 1117a8599bd8SSage Weil * 1118a8599bd8SSage Weil * Make half-hearted attempt ot to invalidate page cache if we are 1119a8599bd8SSage Weil * dropping RDCACHE. Note that this will leave behind locked pages 1120a8599bd8SSage Weil * that we'll then need to deal with elsewhere. 1121a8599bd8SSage Weil * 1122a8599bd8SSage Weil * Return non-zero if delayed release, or we experienced an error 1123a8599bd8SSage Weil * such that the caller should requeue + retry later. 1124a8599bd8SSage Weil * 1125be655596SSage Weil * called with i_ceph_lock, then drops it. 1126a8599bd8SSage Weil * caller should hold snap_rwsem (read), s_mutex. 1127a8599bd8SSage Weil */ 1128a8599bd8SSage Weil static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, 1129a8599bd8SSage Weil int op, int used, int want, int retain, int flushing, 1130a8599bd8SSage Weil unsigned *pflush_tid) 1131be655596SSage Weil __releases(cap->ci->i_ceph_lock) 1132a8599bd8SSage Weil { 1133a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 1134a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1135a8599bd8SSage Weil u64 cap_id = cap->cap_id; 113668c28323SSage Weil int held, revoking, dropping, keep; 1137a8599bd8SSage Weil u64 seq, issue_seq, mseq, time_warp_seq, follows; 1138a8599bd8SSage Weil u64 size, max_size; 1139a8599bd8SSage Weil struct timespec mtime, atime; 1140a8599bd8SSage Weil int wake = 0; 11415706b27dSAl Viro umode_t mode; 114205cb11c1SEric W. Biederman kuid_t uid; 114305cb11c1SEric W. Biederman kgid_t gid; 1144a8599bd8SSage Weil struct ceph_mds_session *session; 1145a8599bd8SSage Weil u64 xattr_version = 0; 1146082afec9SSage Weil struct ceph_buffer *xattr_blob = NULL; 1147a8599bd8SSage Weil int delayed = 0; 1148a8599bd8SSage Weil u64 flush_tid = 0; 1149a8599bd8SSage Weil int i; 1150a8599bd8SSage Weil int ret; 1151e20d258dSYan, Zheng bool inline_data; 1152a8599bd8SSage Weil 115368c28323SSage Weil held = cap->issued | cap->implemented; 115468c28323SSage Weil revoking = cap->implemented & ~cap->issued; 115568c28323SSage Weil retain &= ~revoking; 115668c28323SSage Weil dropping = cap->issued & ~retain; 115768c28323SSage Weil 1158a8599bd8SSage Weil dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", 1159a8599bd8SSage Weil inode, cap, cap->session, 1160a8599bd8SSage Weil ceph_cap_string(held), ceph_cap_string(held & retain), 1161a8599bd8SSage Weil ceph_cap_string(revoking)); 1162a8599bd8SSage Weil BUG_ON((retain & CEPH_CAP_PIN) == 0); 1163a8599bd8SSage Weil 1164a8599bd8SSage Weil session = cap->session; 1165a8599bd8SSage Weil 1166a8599bd8SSage Weil /* don't release wanted unless we've waited a bit. */ 1167a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1168a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_min)) { 1169a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", 1170a8599bd8SSage Weil ceph_cap_string(cap->issued), 1171a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1172a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1173a8599bd8SSage Weil ceph_cap_string(want)); 1174a8599bd8SSage Weil want |= cap->mds_wanted; 1175a8599bd8SSage Weil retain |= cap->issued; 1176a8599bd8SSage Weil delayed = 1; 1177a8599bd8SSage Weil } 1178a8599bd8SSage Weil ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); 1179a8599bd8SSage Weil 1180a8599bd8SSage Weil cap->issued &= retain; /* drop bits we don't want */ 1181a8599bd8SSage Weil if (cap->implemented & ~cap->issued) { 1182a8599bd8SSage Weil /* 1183a8599bd8SSage Weil * Wake up any waiters on wanted -> needed transition. 1184a8599bd8SSage Weil * This is due to the weird transition from buffered 1185a8599bd8SSage Weil * to sync IO... we need to flush dirty pages _before_ 1186a8599bd8SSage Weil * allowing sync writes to avoid reordering. 1187a8599bd8SSage Weil */ 1188a8599bd8SSage Weil wake = 1; 1189a8599bd8SSage Weil } 1190a8599bd8SSage Weil cap->implemented &= cap->issued | used; 1191a8599bd8SSage Weil cap->mds_wanted = want; 1192a8599bd8SSage Weil 1193a8599bd8SSage Weil if (flushing) { 1194a8599bd8SSage Weil /* 1195a8599bd8SSage Weil * assign a tid for flush operations so we can avoid 1196a8599bd8SSage Weil * flush1 -> dirty1 -> flush2 -> flushack1 -> mark 1197a8599bd8SSage Weil * clean type races. track latest tid for every bit 1198a8599bd8SSage Weil * so we can handle flush AxFw, flush Fw, and have the 1199a8599bd8SSage Weil * first ack clean Ax. 1200a8599bd8SSage Weil */ 1201a8599bd8SSage Weil flush_tid = ++ci->i_cap_flush_last_tid; 1202a8599bd8SSage Weil if (pflush_tid) 1203a8599bd8SSage Weil *pflush_tid = flush_tid; 1204a8599bd8SSage Weil dout(" cap_flush_tid %d\n", (int)flush_tid); 1205a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1206a8599bd8SSage Weil if (flushing & (1 << i)) 1207a8599bd8SSage Weil ci->i_cap_flush_tid[i] = flush_tid; 12087d8cb26dSSage Weil 12097d8cb26dSSage Weil follows = ci->i_head_snapc->seq; 12107d8cb26dSSage Weil } else { 12117d8cb26dSSage Weil follows = 0; 1212a8599bd8SSage Weil } 1213a8599bd8SSage Weil 1214a8599bd8SSage Weil keep = cap->implemented; 1215a8599bd8SSage Weil seq = cap->seq; 1216a8599bd8SSage Weil issue_seq = cap->issue_seq; 1217a8599bd8SSage Weil mseq = cap->mseq; 1218a8599bd8SSage Weil size = inode->i_size; 1219a8599bd8SSage Weil ci->i_reported_size = size; 1220a8599bd8SSage Weil max_size = ci->i_wanted_max_size; 1221a8599bd8SSage Weil ci->i_requested_max_size = max_size; 1222a8599bd8SSage Weil mtime = inode->i_mtime; 1223a8599bd8SSage Weil atime = inode->i_atime; 1224a8599bd8SSage Weil time_warp_seq = ci->i_time_warp_seq; 1225a8599bd8SSage Weil uid = inode->i_uid; 1226a8599bd8SSage Weil gid = inode->i_gid; 1227a8599bd8SSage Weil mode = inode->i_mode; 1228a8599bd8SSage Weil 1229082afec9SSage Weil if (flushing & CEPH_CAP_XATTR_EXCL) { 1230a8599bd8SSage Weil __ceph_build_xattrs_blob(ci); 1231082afec9SSage Weil xattr_blob = ci->i_xattrs.blob; 1232082afec9SSage Weil xattr_version = ci->i_xattrs.version; 1233a8599bd8SSage Weil } 1234a8599bd8SSage Weil 1235e20d258dSYan, Zheng inline_data = ci->i_inline_version != CEPH_INLINE_NONE; 1236e20d258dSYan, Zheng 1237be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1238a8599bd8SSage Weil 1239a8599bd8SSage Weil ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id, 1240a8599bd8SSage Weil op, keep, want, flushing, seq, flush_tid, issue_seq, mseq, 1241a8599bd8SSage Weil size, max_size, &mtime, &atime, time_warp_seq, 1242082afec9SSage Weil uid, gid, mode, xattr_version, xattr_blob, 1243e20d258dSYan, Zheng follows, inline_data); 1244a8599bd8SSage Weil if (ret < 0) { 1245a8599bd8SSage Weil dout("error sending cap msg, must requeue %p\n", inode); 1246a8599bd8SSage Weil delayed = 1; 1247a8599bd8SSage Weil } 1248a8599bd8SSage Weil 1249a8599bd8SSage Weil if (wake) 125003066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1251a8599bd8SSage Weil 1252a8599bd8SSage Weil return delayed; 1253a8599bd8SSage Weil } 1254a8599bd8SSage Weil 1255a8599bd8SSage Weil /* 1256a8599bd8SSage Weil * When a snapshot is taken, clients accumulate dirty metadata on 1257a8599bd8SSage Weil * inodes with capabilities in ceph_cap_snaps to describe the file 1258a8599bd8SSage Weil * state at the time the snapshot was taken. This must be flushed 1259a8599bd8SSage Weil * asynchronously back to the MDS once sync writes complete and dirty 1260a8599bd8SSage Weil * data is written out. 1261a8599bd8SSage Weil * 1262e835124cSSage Weil * Unless @again is true, skip cap_snaps that were already sent to 1263e835124cSSage Weil * the MDS (i.e., during this session). 1264e835124cSSage Weil * 1265be655596SSage Weil * Called under i_ceph_lock. Takes s_mutex as needed. 1266a8599bd8SSage Weil */ 1267a8599bd8SSage Weil void __ceph_flush_snaps(struct ceph_inode_info *ci, 1268e835124cSSage Weil struct ceph_mds_session **psession, 1269e835124cSSage Weil int again) 1270be655596SSage Weil __releases(ci->i_ceph_lock) 1271be655596SSage Weil __acquires(ci->i_ceph_lock) 1272a8599bd8SSage Weil { 1273a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1274a8599bd8SSage Weil int mds; 1275a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1276a8599bd8SSage Weil u32 mseq; 12773d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 1278a8599bd8SSage Weil struct ceph_mds_session *session = NULL; /* if session != NULL, we hold 1279a8599bd8SSage Weil session->s_mutex */ 1280a8599bd8SSage Weil u64 next_follows = 0; /* keep track of how far we've gotten through the 1281a8599bd8SSage Weil i_cap_snaps list, and skip these entries next time 1282a8599bd8SSage Weil around to avoid an infinite loop */ 1283a8599bd8SSage Weil 1284a8599bd8SSage Weil if (psession) 1285a8599bd8SSage Weil session = *psession; 1286a8599bd8SSage Weil 1287a8599bd8SSage Weil dout("__flush_snaps %p\n", inode); 1288a8599bd8SSage Weil retry: 1289a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 1290a8599bd8SSage Weil /* avoid an infiniute loop after retry */ 1291a8599bd8SSage Weil if (capsnap->follows < next_follows) 1292a8599bd8SSage Weil continue; 1293a8599bd8SSage Weil /* 1294a8599bd8SSage Weil * we need to wait for sync writes to complete and for dirty 1295a8599bd8SSage Weil * pages to be written out. 1296a8599bd8SSage Weil */ 1297a8599bd8SSage Weil if (capsnap->dirty_pages || capsnap->writing) 1298cfc0bf66SSage Weil break; 1299a8599bd8SSage Weil 1300819ccbfaSSage Weil /* 1301819ccbfaSSage Weil * if cap writeback already occurred, we should have dropped 1302819ccbfaSSage Weil * the capsnap in ceph_put_wrbuffer_cap_refs. 1303819ccbfaSSage Weil */ 1304819ccbfaSSage Weil BUG_ON(capsnap->dirty == 0); 1305819ccbfaSSage Weil 1306a8599bd8SSage Weil /* pick mds, take s_mutex */ 1307ca81f3f6SSage Weil if (ci->i_auth_cap == NULL) { 1308ca81f3f6SSage Weil dout("no auth cap (migrating?), doing nothing\n"); 1309ca81f3f6SSage Weil goto out; 1310ca81f3f6SSage Weil } 1311e835124cSSage Weil 1312e835124cSSage Weil /* only flush each capsnap once */ 1313e835124cSSage Weil if (!again && !list_empty(&capsnap->flushing_item)) { 1314e835124cSSage Weil dout("already flushed %p, skipping\n", capsnap); 1315e835124cSSage Weil continue; 1316e835124cSSage Weil } 1317e835124cSSage Weil 1318ca81f3f6SSage Weil mds = ci->i_auth_cap->session->s_mds; 1319ca81f3f6SSage Weil mseq = ci->i_auth_cap->mseq; 1320ca81f3f6SSage Weil 1321a8599bd8SSage Weil if (session && session->s_mds != mds) { 1322a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1323a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1324a8599bd8SSage Weil ceph_put_mds_session(session); 1325a8599bd8SSage Weil session = NULL; 1326a8599bd8SSage Weil } 1327a8599bd8SSage Weil if (!session) { 1328be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1329a8599bd8SSage Weil mutex_lock(&mdsc->mutex); 1330a8599bd8SSage Weil session = __ceph_lookup_mds_session(mdsc, mds); 1331a8599bd8SSage Weil mutex_unlock(&mdsc->mutex); 1332a8599bd8SSage Weil if (session) { 1333a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1334a8599bd8SSage Weil session); 1335a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1336a8599bd8SSage Weil } 1337a8599bd8SSage Weil /* 1338a8599bd8SSage Weil * if session == NULL, we raced against a cap 1339ca81f3f6SSage Weil * deletion or migration. retry, and we'll 1340ca81f3f6SSage Weil * get a better @mds value next time. 1341a8599bd8SSage Weil */ 1342be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1343a8599bd8SSage Weil goto retry; 1344a8599bd8SSage Weil } 1345a8599bd8SSage Weil 1346a8599bd8SSage Weil capsnap->flush_tid = ++ci->i_cap_flush_last_tid; 1347a8599bd8SSage Weil atomic_inc(&capsnap->nref); 1348a8599bd8SSage Weil if (!list_empty(&capsnap->flushing_item)) 1349a8599bd8SSage Weil list_del_init(&capsnap->flushing_item); 1350a8599bd8SSage Weil list_add_tail(&capsnap->flushing_item, 1351a8599bd8SSage Weil &session->s_cap_snaps_flushing); 1352be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1353a8599bd8SSage Weil 1354cfc0bf66SSage Weil dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n", 1355cfc0bf66SSage Weil inode, capsnap, capsnap->follows, capsnap->flush_tid); 1356a8599bd8SSage Weil send_cap_msg(session, ceph_vino(inode).ino, 0, 1357a8599bd8SSage Weil CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0, 1358a8599bd8SSage Weil capsnap->dirty, 0, capsnap->flush_tid, 0, mseq, 1359a8599bd8SSage Weil capsnap->size, 0, 1360a8599bd8SSage Weil &capsnap->mtime, &capsnap->atime, 1361a8599bd8SSage Weil capsnap->time_warp_seq, 1362a8599bd8SSage Weil capsnap->uid, capsnap->gid, capsnap->mode, 13634a625be4SSage Weil capsnap->xattr_version, capsnap->xattr_blob, 1364e20d258dSYan, Zheng capsnap->follows, capsnap->inline_data); 1365a8599bd8SSage Weil 1366a8599bd8SSage Weil next_follows = capsnap->follows + 1; 1367a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 1368a8599bd8SSage Weil 1369be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1370a8599bd8SSage Weil goto retry; 1371a8599bd8SSage Weil } 1372a8599bd8SSage Weil 1373a8599bd8SSage Weil /* we flushed them all; remove this inode from the queue */ 1374a8599bd8SSage Weil spin_lock(&mdsc->snap_flush_lock); 1375a8599bd8SSage Weil list_del_init(&ci->i_snap_flush_item); 1376a8599bd8SSage Weil spin_unlock(&mdsc->snap_flush_lock); 1377a8599bd8SSage Weil 1378ca81f3f6SSage Weil out: 1379a8599bd8SSage Weil if (psession) 1380a8599bd8SSage Weil *psession = session; 1381a8599bd8SSage Weil else if (session) { 1382a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1383a8599bd8SSage Weil ceph_put_mds_session(session); 1384a8599bd8SSage Weil } 1385a8599bd8SSage Weil } 1386a8599bd8SSage Weil 1387a8599bd8SSage Weil static void ceph_flush_snaps(struct ceph_inode_info *ci) 1388a8599bd8SSage Weil { 1389be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1390e835124cSSage Weil __ceph_flush_snaps(ci, NULL, 0); 1391be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1392a8599bd8SSage Weil } 1393a8599bd8SSage Weil 1394a8599bd8SSage Weil /* 1395fca65b4aSSage Weil * Mark caps dirty. If inode is newly dirty, return the dirty flags. 1396fca65b4aSSage Weil * Caller is then responsible for calling __mark_inode_dirty with the 1397fca65b4aSSage Weil * returned flags value. 139876e3b390SSage Weil */ 1399fca65b4aSSage Weil int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask) 140076e3b390SSage Weil { 1401640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 14023d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 140376e3b390SSage Weil struct inode *inode = &ci->vfs_inode; 140476e3b390SSage Weil int was = ci->i_dirty_caps; 140576e3b390SSage Weil int dirty = 0; 140676e3b390SSage Weil 1407571ade33SYan, Zheng if (!ci->i_auth_cap) { 1408571ade33SYan, Zheng pr_warn("__mark_dirty_caps %p %llx mask %s, " 1409571ade33SYan, Zheng "but no auth cap (session was closed?)\n", 1410571ade33SYan, Zheng inode, ceph_ino(inode), ceph_cap_string(mask)); 1411571ade33SYan, Zheng return 0; 1412571ade33SYan, Zheng } 1413571ade33SYan, Zheng 141476e3b390SSage Weil dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, 141576e3b390SSage Weil ceph_cap_string(mask), ceph_cap_string(was), 141676e3b390SSage Weil ceph_cap_string(was | mask)); 141776e3b390SSage Weil ci->i_dirty_caps |= mask; 141876e3b390SSage Weil if (was == 0) { 14197d8cb26dSSage Weil if (!ci->i_head_snapc) 14207d8cb26dSSage Weil ci->i_head_snapc = ceph_get_snap_context( 14217d8cb26dSSage Weil ci->i_snap_realm->cached_context); 14220685235fSYan, Zheng dout(" inode %p now dirty snapc %p auth cap %p\n", 14230685235fSYan, Zheng &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap); 142476e3b390SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 142576e3b390SSage Weil spin_lock(&mdsc->cap_dirty_lock); 142676e3b390SSage Weil list_add(&ci->i_dirty_item, &mdsc->cap_dirty); 142776e3b390SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 142876e3b390SSage Weil if (ci->i_flushing_caps == 0) { 14293772d26dSSage Weil ihold(inode); 143076e3b390SSage Weil dirty |= I_DIRTY_SYNC; 143176e3b390SSage Weil } 143276e3b390SSage Weil } 143376e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 143476e3b390SSage Weil if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && 143576e3b390SSage Weil (mask & CEPH_CAP_FILE_BUFFER)) 143676e3b390SSage Weil dirty |= I_DIRTY_DATASYNC; 143776e3b390SSage Weil __cap_delay_requeue(mdsc, ci); 1438fca65b4aSSage Weil return dirty; 143976e3b390SSage Weil } 144076e3b390SSage Weil 144176e3b390SSage Weil /* 1442a8599bd8SSage Weil * Add dirty inode to the flushing list. Assigned a seq number so we 1443a8599bd8SSage Weil * can wait for caps to flush without starving. 1444cdc35f96SSage Weil * 1445be655596SSage Weil * Called under i_ceph_lock. 1446a8599bd8SSage Weil */ 1447cdc35f96SSage Weil static int __mark_caps_flushing(struct inode *inode, 1448a8599bd8SSage Weil struct ceph_mds_session *session) 1449a8599bd8SSage Weil { 14503d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1451a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1452cdc35f96SSage Weil int flushing; 1453a8599bd8SSage Weil 1454cdc35f96SSage Weil BUG_ON(ci->i_dirty_caps == 0); 1455a8599bd8SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 1456cdc35f96SSage Weil 1457cdc35f96SSage Weil flushing = ci->i_dirty_caps; 1458cdc35f96SSage Weil dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", 1459cdc35f96SSage Weil ceph_cap_string(flushing), 1460cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps), 1461cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps | flushing)); 1462cdc35f96SSage Weil ci->i_flushing_caps |= flushing; 1463cdc35f96SSage Weil ci->i_dirty_caps = 0; 1464afcdaea3SSage Weil dout(" inode %p now !dirty\n", inode); 1465cdc35f96SSage Weil 1466a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 1467cdc35f96SSage Weil list_del_init(&ci->i_dirty_item); 1468afcdaea3SSage Weil 1469afcdaea3SSage Weil if (list_empty(&ci->i_flushing_item)) { 1470d3383a8eSYan, Zheng ci->i_cap_flush_seq = ++mdsc->cap_flush_seq; 1471a8599bd8SSage Weil list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1472a8599bd8SSage Weil mdsc->num_cap_flushing++; 1473afcdaea3SSage Weil dout(" inode %p now flushing seq %lld\n", inode, 1474afcdaea3SSage Weil ci->i_cap_flush_seq); 1475afcdaea3SSage Weil } else { 1476afcdaea3SSage Weil list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1477afcdaea3SSage Weil dout(" inode %p now flushing (more) seq %lld\n", inode, 1478a8599bd8SSage Weil ci->i_cap_flush_seq); 1479a8599bd8SSage Weil } 1480a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 1481cdc35f96SSage Weil 1482cdc35f96SSage Weil return flushing; 1483a8599bd8SSage Weil } 1484a8599bd8SSage Weil 1485a8599bd8SSage Weil /* 14865ecad6fdSSage Weil * try to invalidate mapping pages without blocking. 14875ecad6fdSSage Weil */ 14885ecad6fdSSage Weil static int try_nonblocking_invalidate(struct inode *inode) 14895ecad6fdSSage Weil { 14905ecad6fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14915ecad6fdSSage Weil u32 invalidating_gen = ci->i_rdcache_gen; 14925ecad6fdSSage Weil 1493be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 14945ecad6fdSSage Weil invalidate_mapping_pages(&inode->i_data, 0, -1); 1495be655596SSage Weil spin_lock(&ci->i_ceph_lock); 14965ecad6fdSSage Weil 149718a38193SSage Weil if (inode->i_data.nrpages == 0 && 14985ecad6fdSSage Weil invalidating_gen == ci->i_rdcache_gen) { 14995ecad6fdSSage Weil /* success. */ 15005ecad6fdSSage Weil dout("try_nonblocking_invalidate %p success\n", inode); 1501cd045cb4SSage Weil /* save any racing async invalidate some trouble */ 1502cd045cb4SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen - 1; 15035ecad6fdSSage Weil return 0; 15045ecad6fdSSage Weil } 15055ecad6fdSSage Weil dout("try_nonblocking_invalidate %p failed\n", inode); 15065ecad6fdSSage Weil return -1; 15075ecad6fdSSage Weil } 15085ecad6fdSSage Weil 15095ecad6fdSSage Weil /* 1510a8599bd8SSage Weil * Swiss army knife function to examine currently used and wanted 1511a8599bd8SSage Weil * versus held caps. Release, flush, ack revoked caps to mds as 1512a8599bd8SSage Weil * appropriate. 1513a8599bd8SSage Weil * 1514a8599bd8SSage Weil * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay 1515a8599bd8SSage Weil * cap release further. 1516a8599bd8SSage Weil * CHECK_CAPS_AUTHONLY - we should only check the auth cap 1517a8599bd8SSage Weil * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without 1518a8599bd8SSage Weil * further delay. 1519a8599bd8SSage Weil */ 1520a8599bd8SSage Weil void ceph_check_caps(struct ceph_inode_info *ci, int flags, 1521a8599bd8SSage Weil struct ceph_mds_session *session) 1522a8599bd8SSage Weil { 15233d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 15243d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1525a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1526a8599bd8SSage Weil struct ceph_cap *cap; 1527395c312bSYan, Zheng int file_wanted, used, cap_used; 1528a8599bd8SSage Weil int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ 1529cbd03635SSage Weil int issued, implemented, want, retain, revoking, flushing = 0; 1530a8599bd8SSage Weil int mds = -1; /* keep track of how far we've gone through i_caps list 1531a8599bd8SSage Weil to avoid an infinite loop on retry */ 1532a8599bd8SSage Weil struct rb_node *p; 1533a8599bd8SSage Weil int tried_invalidate = 0; 1534a8599bd8SSage Weil int delayed = 0, sent = 0, force_requeue = 0, num; 1535cbd03635SSage Weil int queue_invalidate = 0; 1536a8599bd8SSage Weil int is_delayed = flags & CHECK_CAPS_NODELAY; 1537a8599bd8SSage Weil 1538a8599bd8SSage Weil /* if we are unmounting, flush any unused caps immediately. */ 1539a8599bd8SSage Weil if (mdsc->stopping) 1540a8599bd8SSage Weil is_delayed = 1; 1541a8599bd8SSage Weil 1542be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1543a8599bd8SSage Weil 1544a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 1545a8599bd8SSage Weil flags |= CHECK_CAPS_FLUSH; 1546a8599bd8SSage Weil 1547a8599bd8SSage Weil /* flush snaps first time around only */ 1548a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) 1549e835124cSSage Weil __ceph_flush_snaps(ci, &session, 0); 1550a8599bd8SSage Weil goto retry_locked; 1551a8599bd8SSage Weil retry: 1552be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1553a8599bd8SSage Weil retry_locked: 1554a8599bd8SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 1555a8599bd8SSage Weil used = __ceph_caps_used(ci); 1556a8599bd8SSage Weil want = file_wanted | used; 1557cbd03635SSage Weil issued = __ceph_caps_issued(ci, &implemented); 1558cbd03635SSage Weil revoking = implemented & ~issued; 1559a8599bd8SSage Weil 1560a8599bd8SSage Weil retain = want | CEPH_CAP_PIN; 1561a8599bd8SSage Weil if (!mdsc->stopping && inode->i_nlink > 0) { 1562a8599bd8SSage Weil if (want) { 1563a8599bd8SSage Weil retain |= CEPH_CAP_ANY; /* be greedy */ 156432ec4397SYan, Zheng } else if (S_ISDIR(inode->i_mode) && 156532ec4397SYan, Zheng (issued & CEPH_CAP_FILE_SHARED) && 156632ec4397SYan, Zheng __ceph_dir_is_complete(ci)) { 156732ec4397SYan, Zheng /* 156832ec4397SYan, Zheng * If a directory is complete, we want to keep 156932ec4397SYan, Zheng * the exclusive cap. So that MDS does not end up 157032ec4397SYan, Zheng * revoking the shared cap on every create/unlink 157132ec4397SYan, Zheng * operation. 157232ec4397SYan, Zheng */ 157332ec4397SYan, Zheng want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; 157432ec4397SYan, Zheng retain |= want; 1575a8599bd8SSage Weil } else { 157632ec4397SYan, Zheng 1577a8599bd8SSage Weil retain |= CEPH_CAP_ANY_SHARED; 1578a8599bd8SSage Weil /* 1579a8599bd8SSage Weil * keep RD only if we didn't have the file open RW, 1580a8599bd8SSage Weil * because then the mds would revoke it anyway to 1581a8599bd8SSage Weil * journal max_size=0. 1582a8599bd8SSage Weil */ 1583a8599bd8SSage Weil if (ci->i_max_size == 0) 1584a8599bd8SSage Weil retain |= CEPH_CAP_ANY_RD; 1585a8599bd8SSage Weil } 1586a8599bd8SSage Weil } 1587a8599bd8SSage Weil 1588a8599bd8SSage Weil dout("check_caps %p file_want %s used %s dirty %s flushing %s" 1589cbd03635SSage Weil " issued %s revoking %s retain %s %s%s%s\n", inode, 1590a8599bd8SSage Weil ceph_cap_string(file_wanted), 1591a8599bd8SSage Weil ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), 1592a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps), 1593cbd03635SSage Weil ceph_cap_string(issued), ceph_cap_string(revoking), 1594a8599bd8SSage Weil ceph_cap_string(retain), 1595a8599bd8SSage Weil (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", 1596a8599bd8SSage Weil (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", 1597a8599bd8SSage Weil (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); 1598a8599bd8SSage Weil 1599a8599bd8SSage Weil /* 1600a8599bd8SSage Weil * If we no longer need to hold onto old our caps, and we may 1601a8599bd8SSage Weil * have cached pages, but don't want them, then try to invalidate. 1602a8599bd8SSage Weil * If we fail, it's because pages are locked.... try again later. 1603a8599bd8SSage Weil */ 1604a8599bd8SSage Weil if ((!is_delayed || mdsc->stopping) && 1605a8599bd8SSage Weil ci->i_wrbuffer_ref == 0 && /* no dirty pages... */ 160693afd449SSage Weil inode->i_data.nrpages && /* have cached pages */ 1607cbd03635SSage Weil (file_wanted == 0 || /* no open files */ 16082962507cSSage Weil (revoking & (CEPH_CAP_FILE_CACHE| 16092962507cSSage Weil CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */ 1610a8599bd8SSage Weil !tried_invalidate) { 1611a8599bd8SSage Weil dout("check_caps trying to invalidate on %p\n", inode); 16125ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) < 0) { 16132962507cSSage Weil if (revoking & (CEPH_CAP_FILE_CACHE| 16142962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) { 1615cbd03635SSage Weil dout("check_caps queuing invalidate\n"); 1616cbd03635SSage Weil queue_invalidate = 1; 1617cbd03635SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 1618a8599bd8SSage Weil } else { 1619a8599bd8SSage Weil dout("check_caps failed to invalidate pages\n"); 1620a8599bd8SSage Weil /* we failed to invalidate pages. check these 1621a8599bd8SSage Weil caps again later. */ 1622a8599bd8SSage Weil force_requeue = 1; 1623a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 1624a8599bd8SSage Weil } 16255ecad6fdSSage Weil } 1626a8599bd8SSage Weil tried_invalidate = 1; 1627a8599bd8SSage Weil goto retry_locked; 1628a8599bd8SSage Weil } 1629a8599bd8SSage Weil 1630a8599bd8SSage Weil num = 0; 1631a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 1632a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 1633a8599bd8SSage Weil num++; 1634a8599bd8SSage Weil 1635a8599bd8SSage Weil /* avoid looping forever */ 1636a8599bd8SSage Weil if (mds >= cap->mds || 1637a8599bd8SSage Weil ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) 1638a8599bd8SSage Weil continue; 1639a8599bd8SSage Weil 1640a8599bd8SSage Weil /* NOTE: no side-effects allowed, until we take s_mutex */ 1641a8599bd8SSage Weil 1642395c312bSYan, Zheng cap_used = used; 1643395c312bSYan, Zheng if (ci->i_auth_cap && cap != ci->i_auth_cap) 1644395c312bSYan, Zheng cap_used &= ~ci->i_auth_cap->issued; 1645395c312bSYan, Zheng 1646a8599bd8SSage Weil revoking = cap->implemented & ~cap->issued; 1647395c312bSYan, Zheng dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n", 1648088b3f5eSSage Weil cap->mds, cap, ceph_cap_string(cap->issued), 1649395c312bSYan, Zheng ceph_cap_string(cap_used), 1650088b3f5eSSage Weil ceph_cap_string(cap->implemented), 1651a8599bd8SSage Weil ceph_cap_string(revoking)); 1652a8599bd8SSage Weil 1653a8599bd8SSage Weil if (cap == ci->i_auth_cap && 1654a8599bd8SSage Weil (cap->issued & CEPH_CAP_FILE_WR)) { 1655a8599bd8SSage Weil /* request larger max_size from MDS? */ 1656a8599bd8SSage Weil if (ci->i_wanted_max_size > ci->i_max_size && 1657a8599bd8SSage Weil ci->i_wanted_max_size > ci->i_requested_max_size) { 1658a8599bd8SSage Weil dout("requesting new max_size\n"); 1659a8599bd8SSage Weil goto ack; 1660a8599bd8SSage Weil } 1661a8599bd8SSage Weil 1662a8599bd8SSage Weil /* approaching file_max? */ 1663a8599bd8SSage Weil if ((inode->i_size << 1) >= ci->i_max_size && 1664a8599bd8SSage Weil (ci->i_reported_size << 1) < ci->i_max_size) { 1665a8599bd8SSage Weil dout("i_size approaching max_size\n"); 1666a8599bd8SSage Weil goto ack; 1667a8599bd8SSage Weil } 1668a8599bd8SSage Weil } 1669a8599bd8SSage Weil /* flush anything dirty? */ 1670a8599bd8SSage Weil if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) && 1671a8599bd8SSage Weil ci->i_dirty_caps) { 1672a8599bd8SSage Weil dout("flushing dirty caps\n"); 1673a8599bd8SSage Weil goto ack; 1674a8599bd8SSage Weil } 1675a8599bd8SSage Weil 1676a8599bd8SSage Weil /* completed revocation? going down and there are no caps? */ 1677395c312bSYan, Zheng if (revoking && (revoking & cap_used) == 0) { 1678a8599bd8SSage Weil dout("completed revocation of %s\n", 1679a8599bd8SSage Weil ceph_cap_string(cap->implemented & ~cap->issued)); 1680a8599bd8SSage Weil goto ack; 1681a8599bd8SSage Weil } 1682a8599bd8SSage Weil 1683a8599bd8SSage Weil /* want more caps from mds? */ 1684a8599bd8SSage Weil if (want & ~(cap->mds_wanted | cap->issued)) 1685a8599bd8SSage Weil goto ack; 1686a8599bd8SSage Weil 1687a8599bd8SSage Weil /* things we might delay */ 1688a8599bd8SSage Weil if ((cap->issued & ~retain) == 0 && 1689a8599bd8SSage Weil cap->mds_wanted == want) 1690a8599bd8SSage Weil continue; /* nope, all good */ 1691a8599bd8SSage Weil 1692a8599bd8SSage Weil if (is_delayed) 1693a8599bd8SSage Weil goto ack; 1694a8599bd8SSage Weil 1695a8599bd8SSage Weil /* delay? */ 1696a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1697a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) { 1698a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s\n", 1699a8599bd8SSage Weil ceph_cap_string(cap->issued), 1700a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1701a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1702a8599bd8SSage Weil ceph_cap_string(want)); 1703a8599bd8SSage Weil delayed++; 1704a8599bd8SSage Weil continue; 1705a8599bd8SSage Weil } 1706a8599bd8SSage Weil 1707a8599bd8SSage Weil ack: 1708e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1709e9964c10SSage Weil dout(" skipping %p I_NOFLUSH set\n", inode); 1710e9964c10SSage Weil continue; 1711e9964c10SSage Weil } 1712e9964c10SSage Weil 1713a8599bd8SSage Weil if (session && session != cap->session) { 1714a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1715a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1716a8599bd8SSage Weil session = NULL; 1717a8599bd8SSage Weil } 1718a8599bd8SSage Weil if (!session) { 1719a8599bd8SSage Weil session = cap->session; 1720a8599bd8SSage Weil if (mutex_trylock(&session->s_mutex) == 0) { 1721a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1722a8599bd8SSage Weil session); 1723be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1724a8599bd8SSage Weil if (took_snap_rwsem) { 1725a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1726a8599bd8SSage Weil took_snap_rwsem = 0; 1727a8599bd8SSage Weil } 1728a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1729a8599bd8SSage Weil goto retry; 1730a8599bd8SSage Weil } 1731a8599bd8SSage Weil } 1732a8599bd8SSage Weil /* take snap_rwsem after session mutex */ 1733a8599bd8SSage Weil if (!took_snap_rwsem) { 1734a8599bd8SSage Weil if (down_read_trylock(&mdsc->snap_rwsem) == 0) { 1735a8599bd8SSage Weil dout("inverting snap/in locks on %p\n", 1736a8599bd8SSage Weil inode); 1737be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1738a8599bd8SSage Weil down_read(&mdsc->snap_rwsem); 1739a8599bd8SSage Weil took_snap_rwsem = 1; 1740a8599bd8SSage Weil goto retry; 1741a8599bd8SSage Weil } 1742a8599bd8SSage Weil took_snap_rwsem = 1; 1743a8599bd8SSage Weil } 1744a8599bd8SSage Weil 1745cdc35f96SSage Weil if (cap == ci->i_auth_cap && ci->i_dirty_caps) 1746cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 174724be0c48SSage Weil else 174824be0c48SSage Weil flushing = 0; 1749a8599bd8SSage Weil 1750a8599bd8SSage Weil mds = cap->mds; /* remember mds, so we don't repeat */ 1751a8599bd8SSage Weil sent++; 1752a8599bd8SSage Weil 1753be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1754395c312bSYan, Zheng delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used, 1755395c312bSYan, Zheng want, retain, flushing, NULL); 1756be655596SSage Weil goto retry; /* retake i_ceph_lock and restart our cap scan. */ 1757a8599bd8SSage Weil } 1758a8599bd8SSage Weil 1759a8599bd8SSage Weil /* 1760a8599bd8SSage Weil * Reschedule delayed caps release if we delayed anything, 1761a8599bd8SSage Weil * otherwise cancel. 1762a8599bd8SSage Weil */ 1763a8599bd8SSage Weil if (delayed && is_delayed) 1764a8599bd8SSage Weil force_requeue = 1; /* __send_cap delayed release; requeue */ 1765a8599bd8SSage Weil if (!delayed && !is_delayed) 1766a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 1767a8599bd8SSage Weil else if (!is_delayed || force_requeue) 1768a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1769a8599bd8SSage Weil 1770be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1771a8599bd8SSage Weil 1772cbd03635SSage Weil if (queue_invalidate) 17733c6f6b79SSage Weil ceph_queue_invalidate(inode); 1774cbd03635SSage Weil 1775cdc2ce05SSage Weil if (session) 1776a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1777a8599bd8SSage Weil if (took_snap_rwsem) 1778a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1779a8599bd8SSage Weil } 1780a8599bd8SSage Weil 1781a8599bd8SSage Weil /* 1782a8599bd8SSage Weil * Try to flush dirty caps back to the auth mds. 1783a8599bd8SSage Weil */ 17844fe59789SYan, Zheng static int try_flush_caps(struct inode *inode, unsigned *flush_tid) 1785a8599bd8SSage Weil { 17863d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1787a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1788a8599bd8SSage Weil int flushing = 0; 17894fe59789SYan, Zheng struct ceph_mds_session *session = NULL; 1790a8599bd8SSage Weil 1791a8599bd8SSage Weil retry: 1792be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1793e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1794e9964c10SSage Weil dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode); 1795e9964c10SSage Weil goto out; 1796e9964c10SSage Weil } 1797a8599bd8SSage Weil if (ci->i_dirty_caps && ci->i_auth_cap) { 1798a8599bd8SSage Weil struct ceph_cap *cap = ci->i_auth_cap; 1799a8599bd8SSage Weil int used = __ceph_caps_used(ci); 1800a8599bd8SSage Weil int want = __ceph_caps_wanted(ci); 1801a8599bd8SSage Weil int delayed; 1802a8599bd8SSage Weil 18034fe59789SYan, Zheng if (!session || session != cap->session) { 1804be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 18054fe59789SYan, Zheng if (session) 18064fe59789SYan, Zheng mutex_unlock(&session->s_mutex); 1807a8599bd8SSage Weil session = cap->session; 1808a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1809a8599bd8SSage Weil goto retry; 1810a8599bd8SSage Weil } 1811a8599bd8SSage Weil if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) 1812a8599bd8SSage Weil goto out; 1813a8599bd8SSage Weil 1814cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 1815a8599bd8SSage Weil 1816be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1817a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want, 1818a8599bd8SSage Weil cap->issued | cap->implemented, flushing, 1819a8599bd8SSage Weil flush_tid); 1820a8599bd8SSage Weil if (!delayed) 1821a8599bd8SSage Weil goto out_unlocked; 1822a8599bd8SSage Weil 1823be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1824a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1825a8599bd8SSage Weil } 1826a8599bd8SSage Weil out: 1827be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1828a8599bd8SSage Weil out_unlocked: 18294fe59789SYan, Zheng if (session) 1830a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1831a8599bd8SSage Weil return flushing; 1832a8599bd8SSage Weil } 1833a8599bd8SSage Weil 1834a8599bd8SSage Weil /* 1835a8599bd8SSage Weil * Return true if we've flushed caps through the given flush_tid. 1836a8599bd8SSage Weil */ 1837a8599bd8SSage Weil static int caps_are_flushed(struct inode *inode, unsigned tid) 1838a8599bd8SSage Weil { 1839a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1840a5ee751cSDan Carpenter int i, ret = 1; 1841a8599bd8SSage Weil 1842be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1843a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1844a8599bd8SSage Weil if ((ci->i_flushing_caps & (1 << i)) && 1845a8599bd8SSage Weil ci->i_cap_flush_tid[i] <= tid) { 1846a8599bd8SSage Weil /* still flushing this bit */ 1847a8599bd8SSage Weil ret = 0; 1848a8599bd8SSage Weil break; 1849a8599bd8SSage Weil } 1850be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1851a8599bd8SSage Weil return ret; 1852a8599bd8SSage Weil } 1853a8599bd8SSage Weil 1854a8599bd8SSage Weil /* 1855a8599bd8SSage Weil * Wait on any unsafe replies for the given inode. First wait on the 1856a8599bd8SSage Weil * newest request, and make that the upper bound. Then, if there are 1857a8599bd8SSage Weil * more requests, keep waiting on the oldest as long as it is still older 1858a8599bd8SSage Weil * than the original request. 1859a8599bd8SSage Weil */ 1860a8599bd8SSage Weil static void sync_write_wait(struct inode *inode) 1861a8599bd8SSage Weil { 1862a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1863a8599bd8SSage Weil struct list_head *head = &ci->i_unsafe_writes; 1864a8599bd8SSage Weil struct ceph_osd_request *req; 1865a8599bd8SSage Weil u64 last_tid; 1866a8599bd8SSage Weil 1867a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1868a8599bd8SSage Weil if (list_empty(head)) 1869a8599bd8SSage Weil goto out; 1870a8599bd8SSage Weil 1871a8599bd8SSage Weil /* set upper bound as _last_ entry in chain */ 1872a8599bd8SSage Weil req = list_entry(head->prev, struct ceph_osd_request, 1873a8599bd8SSage Weil r_unsafe_item); 1874a8599bd8SSage Weil last_tid = req->r_tid; 1875a8599bd8SSage Weil 1876a8599bd8SSage Weil do { 1877a8599bd8SSage Weil ceph_osdc_get_request(req); 1878a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1879a8599bd8SSage Weil dout("sync_write_wait on tid %llu (until %llu)\n", 1880a8599bd8SSage Weil req->r_tid, last_tid); 1881a8599bd8SSage Weil wait_for_completion(&req->r_safe_completion); 1882a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1883a8599bd8SSage Weil ceph_osdc_put_request(req); 1884a8599bd8SSage Weil 1885a8599bd8SSage Weil /* 1886a8599bd8SSage Weil * from here on look at first entry in chain, since we 1887a8599bd8SSage Weil * only want to wait for anything older than last_tid 1888a8599bd8SSage Weil */ 1889a8599bd8SSage Weil if (list_empty(head)) 1890a8599bd8SSage Weil break; 1891a8599bd8SSage Weil req = list_entry(head->next, struct ceph_osd_request, 1892a8599bd8SSage Weil r_unsafe_item); 1893a8599bd8SSage Weil } while (req->r_tid < last_tid); 1894a8599bd8SSage Weil out: 1895a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1896a8599bd8SSage Weil } 1897a8599bd8SSage Weil 189802c24a82SJosef Bacik int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync) 1899a8599bd8SSage Weil { 19007ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 1901a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1902a8599bd8SSage Weil unsigned flush_tid; 1903a8599bd8SSage Weil int ret; 1904a8599bd8SSage Weil int dirty; 1905a8599bd8SSage Weil 1906a8599bd8SSage Weil dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); 1907a8599bd8SSage Weil sync_write_wait(inode); 1908a8599bd8SSage Weil 190902c24a82SJosef Bacik ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 1910a8599bd8SSage Weil if (ret < 0) 1911a8599bd8SSage Weil return ret; 191202c24a82SJosef Bacik mutex_lock(&inode->i_mutex); 1913a8599bd8SSage Weil 19144fe59789SYan, Zheng dirty = try_flush_caps(inode, &flush_tid); 1915a8599bd8SSage Weil dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); 1916a8599bd8SSage Weil 1917a8599bd8SSage Weil /* 1918a8599bd8SSage Weil * only wait on non-file metadata writeback (the mds 1919a8599bd8SSage Weil * can recover size and mtime, so we don't need to 1920a8599bd8SSage Weil * wait for that) 1921a8599bd8SSage Weil */ 1922a8599bd8SSage Weil if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { 1923a8599bd8SSage Weil dout("fsync waiting for flush_tid %u\n", flush_tid); 1924a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 1925a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1926a8599bd8SSage Weil } 1927a8599bd8SSage Weil 1928a8599bd8SSage Weil dout("fsync %p%s done\n", inode, datasync ? " datasync" : ""); 192902c24a82SJosef Bacik mutex_unlock(&inode->i_mutex); 1930a8599bd8SSage Weil return ret; 1931a8599bd8SSage Weil } 1932a8599bd8SSage Weil 1933a8599bd8SSage Weil /* 1934a8599bd8SSage Weil * Flush any dirty caps back to the mds. If we aren't asked to wait, 1935a8599bd8SSage Weil * queue inode for flush but don't do so immediately, because we can 1936a8599bd8SSage Weil * get by with fewer MDS messages if we wait for data writeback to 1937a8599bd8SSage Weil * complete first. 1938a8599bd8SSage Weil */ 1939f1a3d572SStephen Rothwell int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) 1940a8599bd8SSage Weil { 1941a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1942a8599bd8SSage Weil unsigned flush_tid; 1943a8599bd8SSage Weil int err = 0; 1944a8599bd8SSage Weil int dirty; 1945f1a3d572SStephen Rothwell int wait = wbc->sync_mode == WB_SYNC_ALL; 1946a8599bd8SSage Weil 1947a8599bd8SSage Weil dout("write_inode %p wait=%d\n", inode, wait); 1948a8599bd8SSage Weil if (wait) { 19494fe59789SYan, Zheng dirty = try_flush_caps(inode, &flush_tid); 1950a8599bd8SSage Weil if (dirty) 1951a8599bd8SSage Weil err = wait_event_interruptible(ci->i_cap_wq, 1952a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1953a8599bd8SSage Weil } else { 1954640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 19553d14c5d2SYehuda Sadeh ceph_sb_to_client(inode->i_sb)->mdsc; 1956a8599bd8SSage Weil 1957be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1958a8599bd8SSage Weil if (__ceph_caps_dirty(ci)) 1959a8599bd8SSage Weil __cap_delay_requeue_front(mdsc, ci); 1960be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1961a8599bd8SSage Weil } 1962a8599bd8SSage Weil return err; 1963a8599bd8SSage Weil } 1964a8599bd8SSage Weil 1965a8599bd8SSage Weil /* 1966a8599bd8SSage Weil * After a recovering MDS goes active, we need to resend any caps 1967a8599bd8SSage Weil * we were flushing. 1968a8599bd8SSage Weil * 1969a8599bd8SSage Weil * Caller holds session->s_mutex. 1970a8599bd8SSage Weil */ 1971a8599bd8SSage Weil static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc, 1972a8599bd8SSage Weil struct ceph_mds_session *session) 1973a8599bd8SSage Weil { 1974a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1975a8599bd8SSage Weil 1976a8599bd8SSage Weil dout("kick_flushing_capsnaps mds%d\n", session->s_mds); 1977a8599bd8SSage Weil list_for_each_entry(capsnap, &session->s_cap_snaps_flushing, 1978a8599bd8SSage Weil flushing_item) { 1979a8599bd8SSage Weil struct ceph_inode_info *ci = capsnap->ci; 1980a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1981a8599bd8SSage Weil struct ceph_cap *cap; 1982a8599bd8SSage Weil 1983be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1984a8599bd8SSage Weil cap = ci->i_auth_cap; 1985a8599bd8SSage Weil if (cap && cap->session == session) { 1986a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p capsnap %p\n", inode, 1987a8599bd8SSage Weil cap, capsnap); 1988e835124cSSage Weil __ceph_flush_snaps(ci, &session, 1); 1989a8599bd8SSage Weil } else { 1990a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1991a8599bd8SSage Weil cap, session->s_mds); 1992a8599bd8SSage Weil } 1993be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1994a8599bd8SSage Weil } 1995a8599bd8SSage Weil } 1996a8599bd8SSage Weil 1997a8599bd8SSage Weil void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, 1998a8599bd8SSage Weil struct ceph_mds_session *session) 1999a8599bd8SSage Weil { 2000a8599bd8SSage Weil struct ceph_inode_info *ci; 2001a8599bd8SSage Weil 2002a8599bd8SSage Weil kick_flushing_capsnaps(mdsc, session); 2003a8599bd8SSage Weil 2004a8599bd8SSage Weil dout("kick_flushing_caps mds%d\n", session->s_mds); 2005a8599bd8SSage Weil list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { 2006a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2007a8599bd8SSage Weil struct ceph_cap *cap; 2008a8599bd8SSage Weil int delayed = 0; 2009a8599bd8SSage Weil 2010be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2011a8599bd8SSage Weil cap = ci->i_auth_cap; 2012a8599bd8SSage Weil if (cap && cap->session == session) { 2013a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p %s\n", inode, 2014a8599bd8SSage Weil cap, ceph_cap_string(ci->i_flushing_caps)); 2015a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 2016a8599bd8SSage Weil __ceph_caps_used(ci), 2017a8599bd8SSage Weil __ceph_caps_wanted(ci), 2018a8599bd8SSage Weil cap->issued | cap->implemented, 2019a8599bd8SSage Weil ci->i_flushing_caps, NULL); 2020a8599bd8SSage Weil if (delayed) { 2021be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2022a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 2023be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2024a8599bd8SSage Weil } 2025a8599bd8SSage Weil } else { 2026a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 2027a8599bd8SSage Weil cap, session->s_mds); 2028be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2029a8599bd8SSage Weil } 2030a8599bd8SSage Weil } 2031a8599bd8SSage Weil } 2032a8599bd8SSage Weil 2033088b3f5eSSage Weil static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc, 2034088b3f5eSSage Weil struct ceph_mds_session *session, 2035088b3f5eSSage Weil struct inode *inode) 2036088b3f5eSSage Weil { 2037088b3f5eSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2038088b3f5eSSage Weil struct ceph_cap *cap; 2039088b3f5eSSage Weil int delayed = 0; 2040088b3f5eSSage Weil 2041be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2042088b3f5eSSage Weil cap = ci->i_auth_cap; 2043088b3f5eSSage Weil dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode, 2044088b3f5eSSage Weil ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq); 2045005c4697SYan, Zheng 2046088b3f5eSSage Weil __ceph_flush_snaps(ci, &session, 1); 2047005c4697SYan, Zheng 2048088b3f5eSSage Weil if (ci->i_flushing_caps) { 2049005c4697SYan, Zheng spin_lock(&mdsc->cap_dirty_lock); 2050005c4697SYan, Zheng list_move_tail(&ci->i_flushing_item, 2051005c4697SYan, Zheng &cap->session->s_cap_flushing); 2052005c4697SYan, Zheng spin_unlock(&mdsc->cap_dirty_lock); 2053005c4697SYan, Zheng 2054088b3f5eSSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 2055088b3f5eSSage Weil __ceph_caps_used(ci), 2056088b3f5eSSage Weil __ceph_caps_wanted(ci), 2057088b3f5eSSage Weil cap->issued | cap->implemented, 2058088b3f5eSSage Weil ci->i_flushing_caps, NULL); 2059088b3f5eSSage Weil if (delayed) { 2060be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2061088b3f5eSSage Weil __cap_delay_requeue(mdsc, ci); 2062be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2063088b3f5eSSage Weil } 2064088b3f5eSSage Weil } else { 2065be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2066088b3f5eSSage Weil } 2067088b3f5eSSage Weil } 2068088b3f5eSSage Weil 2069a8599bd8SSage Weil 2070a8599bd8SSage Weil /* 2071a8599bd8SSage Weil * Take references to capabilities we hold, so that we don't release 2072a8599bd8SSage Weil * them to the MDS prematurely. 2073a8599bd8SSage Weil * 2074be655596SSage Weil * Protected by i_ceph_lock. 2075a8599bd8SSage Weil */ 20765dda377cSYan, Zheng static void __take_cap_refs(struct ceph_inode_info *ci, int got, 20775dda377cSYan, Zheng bool snap_rwsem_locked) 2078a8599bd8SSage Weil { 2079a8599bd8SSage Weil if (got & CEPH_CAP_PIN) 2080a8599bd8SSage Weil ci->i_pin_ref++; 2081a8599bd8SSage Weil if (got & CEPH_CAP_FILE_RD) 2082a8599bd8SSage Weil ci->i_rd_ref++; 2083a8599bd8SSage Weil if (got & CEPH_CAP_FILE_CACHE) 2084a8599bd8SSage Weil ci->i_rdcache_ref++; 20855dda377cSYan, Zheng if (got & CEPH_CAP_FILE_WR) { 20865dda377cSYan, Zheng if (ci->i_wr_ref == 0 && !ci->i_head_snapc) { 20875dda377cSYan, Zheng BUG_ON(!snap_rwsem_locked); 20885dda377cSYan, Zheng ci->i_head_snapc = ceph_get_snap_context( 20895dda377cSYan, Zheng ci->i_snap_realm->cached_context); 20905dda377cSYan, Zheng } 2091a8599bd8SSage Weil ci->i_wr_ref++; 20925dda377cSYan, Zheng } 2093a8599bd8SSage Weil if (got & CEPH_CAP_FILE_BUFFER) { 2094d3d0720dSHenry C Chang if (ci->i_wb_ref == 0) 20953772d26dSSage Weil ihold(&ci->vfs_inode); 2096d3d0720dSHenry C Chang ci->i_wb_ref++; 2097d3d0720dSHenry C Chang dout("__take_cap_refs %p wb %d -> %d (?)\n", 2098d3d0720dSHenry C Chang &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref); 2099a8599bd8SSage Weil } 2100a8599bd8SSage Weil } 2101a8599bd8SSage Weil 2102a8599bd8SSage Weil /* 2103a8599bd8SSage Weil * Try to grab cap references. Specify those refs we @want, and the 2104a8599bd8SSage Weil * minimal set we @need. Also include the larger offset we are writing 2105a8599bd8SSage Weil * to (when applicable), and check against max_size here as well. 2106a8599bd8SSage Weil * Note that caller is responsible for ensuring max_size increases are 2107a8599bd8SSage Weil * requested from the MDS. 2108a8599bd8SSage Weil */ 2109a8599bd8SSage Weil static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want, 21105dda377cSYan, Zheng loff_t endoff, bool nonblock, int *got, int *err) 2111a8599bd8SSage Weil { 2112a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 21135dda377cSYan, Zheng struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 2114a8599bd8SSage Weil int ret = 0; 2115c4d4a582SYan, Zheng int have, implemented; 2116195d3ce2SSage Weil int file_wanted; 21175dda377cSYan, Zheng bool snap_rwsem_locked = false; 2118a8599bd8SSage Weil 2119a8599bd8SSage Weil dout("get_cap_refs %p need %s want %s\n", inode, 2120a8599bd8SSage Weil ceph_cap_string(need), ceph_cap_string(want)); 2121c4d4a582SYan, Zheng 21225dda377cSYan, Zheng again: 2123be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2124a8599bd8SSage Weil 2125195d3ce2SSage Weil /* make sure file is actually open */ 2126195d3ce2SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 2127195d3ce2SSage Weil if ((file_wanted & need) == 0) { 2128195d3ce2SSage Weil dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", 2129195d3ce2SSage Weil ceph_cap_string(need), ceph_cap_string(file_wanted)); 2130a8599bd8SSage Weil *err = -EBADF; 2131a8599bd8SSage Weil ret = 1; 21323738daa6SYan, Zheng goto out_unlock; 2133a8599bd8SSage Weil } 2134a8599bd8SSage Weil 213537505d57SYan, Zheng /* finish pending truncate */ 213637505d57SYan, Zheng while (ci->i_truncate_pending) { 213737505d57SYan, Zheng spin_unlock(&ci->i_ceph_lock); 21385dda377cSYan, Zheng if (snap_rwsem_locked) { 21395dda377cSYan, Zheng up_read(&mdsc->snap_rwsem); 21405dda377cSYan, Zheng snap_rwsem_locked = false; 21415dda377cSYan, Zheng } 2142b415bf4fSYan, Zheng __ceph_do_pending_vmtruncate(inode); 214337505d57SYan, Zheng spin_lock(&ci->i_ceph_lock); 214437505d57SYan, Zheng } 214537505d57SYan, Zheng 21463871cbb9SYan, Zheng have = __ceph_caps_issued(ci, &implemented); 21473871cbb9SYan, Zheng 21483871cbb9SYan, Zheng if (have & need & CEPH_CAP_FILE_WR) { 2149a8599bd8SSage Weil if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { 2150a8599bd8SSage Weil dout("get_cap_refs %p endoff %llu > maxsize %llu\n", 2151a8599bd8SSage Weil inode, endoff, ci->i_max_size); 21523871cbb9SYan, Zheng if (endoff > ci->i_requested_max_size) { 21535dda377cSYan, Zheng *err = -EAGAIN; 2154a8599bd8SSage Weil ret = 1; 2155a8599bd8SSage Weil } 21563738daa6SYan, Zheng goto out_unlock; 2157a8599bd8SSage Weil } 2158a8599bd8SSage Weil /* 2159a8599bd8SSage Weil * If a sync write is in progress, we must wait, so that we 2160a8599bd8SSage Weil * can get a final snapshot value for size+mtime. 2161a8599bd8SSage Weil */ 2162a8599bd8SSage Weil if (__ceph_have_pending_cap_snap(ci)) { 2163a8599bd8SSage Weil dout("get_cap_refs %p cap_snap_pending\n", inode); 21643738daa6SYan, Zheng goto out_unlock; 2165a8599bd8SSage Weil } 2166a8599bd8SSage Weil } 2167a8599bd8SSage Weil 2168a8599bd8SSage Weil if ((have & need) == need) { 2169a8599bd8SSage Weil /* 2170a8599bd8SSage Weil * Look at (implemented & ~have & not) so that we keep waiting 2171a8599bd8SSage Weil * on transition from wanted -> needed caps. This is needed 2172a8599bd8SSage Weil * for WRBUFFER|WR -> WR to avoid a new WR sync write from 2173a8599bd8SSage Weil * going before a prior buffered writeback happens. 2174a8599bd8SSage Weil */ 2175a8599bd8SSage Weil int not = want & ~(have & need); 2176a8599bd8SSage Weil int revoking = implemented & ~have; 2177a8599bd8SSage Weil dout("get_cap_refs %p have %s but not %s (revoking %s)\n", 2178a8599bd8SSage Weil inode, ceph_cap_string(have), ceph_cap_string(not), 2179a8599bd8SSage Weil ceph_cap_string(revoking)); 2180a8599bd8SSage Weil if ((revoking & not) == 0) { 21815dda377cSYan, Zheng if (!snap_rwsem_locked && 21825dda377cSYan, Zheng !ci->i_head_snapc && 21835dda377cSYan, Zheng (need & CEPH_CAP_FILE_WR)) { 21845dda377cSYan, Zheng if (!down_read_trylock(&mdsc->snap_rwsem)) { 21855dda377cSYan, Zheng /* 21865dda377cSYan, Zheng * we can not call down_read() when 21875dda377cSYan, Zheng * task isn't in TASK_RUNNING state 21885dda377cSYan, Zheng */ 21895dda377cSYan, Zheng if (nonblock) { 21905dda377cSYan, Zheng *err = -EAGAIN; 21915dda377cSYan, Zheng ret = 1; 21925dda377cSYan, Zheng goto out_unlock; 21935dda377cSYan, Zheng } 21945dda377cSYan, Zheng 21955dda377cSYan, Zheng spin_unlock(&ci->i_ceph_lock); 21965dda377cSYan, Zheng down_read(&mdsc->snap_rwsem); 21975dda377cSYan, Zheng snap_rwsem_locked = true; 21985dda377cSYan, Zheng goto again; 21995dda377cSYan, Zheng } 22005dda377cSYan, Zheng snap_rwsem_locked = true; 22015dda377cSYan, Zheng } 2202c4d4a582SYan, Zheng *got = need | (have & want); 22035dda377cSYan, Zheng __take_cap_refs(ci, *got, true); 2204a8599bd8SSage Weil ret = 1; 2205a8599bd8SSage Weil } 2206a8599bd8SSage Weil } else { 220703f4fcb0SYan, Zheng int session_readonly = false; 220803f4fcb0SYan, Zheng if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) { 220903f4fcb0SYan, Zheng struct ceph_mds_session *s = ci->i_auth_cap->session; 221003f4fcb0SYan, Zheng spin_lock(&s->s_cap_lock); 221103f4fcb0SYan, Zheng session_readonly = s->s_readonly; 221203f4fcb0SYan, Zheng spin_unlock(&s->s_cap_lock); 221303f4fcb0SYan, Zheng } 221403f4fcb0SYan, Zheng if (session_readonly) { 221503f4fcb0SYan, Zheng dout("get_cap_refs %p needed %s but mds%d readonly\n", 221603f4fcb0SYan, Zheng inode, ceph_cap_string(need), ci->i_auth_cap->mds); 221703f4fcb0SYan, Zheng *err = -EROFS; 221803f4fcb0SYan, Zheng ret = 1; 221903f4fcb0SYan, Zheng goto out_unlock; 222003f4fcb0SYan, Zheng } 222103f4fcb0SYan, Zheng 2222a8599bd8SSage Weil dout("get_cap_refs %p have %s needed %s\n", inode, 2223a8599bd8SSage Weil ceph_cap_string(have), ceph_cap_string(need)); 2224a8599bd8SSage Weil } 22253738daa6SYan, Zheng out_unlock: 2226be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 22275dda377cSYan, Zheng if (snap_rwsem_locked) 22285dda377cSYan, Zheng up_read(&mdsc->snap_rwsem); 22293738daa6SYan, Zheng 2230a8599bd8SSage Weil dout("get_cap_refs %p ret %d got %s\n", inode, 2231c4d4a582SYan, Zheng ret, ceph_cap_string(*got)); 2232a8599bd8SSage Weil return ret; 2233a8599bd8SSage Weil } 2234a8599bd8SSage Weil 2235a8599bd8SSage Weil /* 2236a8599bd8SSage Weil * Check the offset we are writing up to against our current 2237a8599bd8SSage Weil * max_size. If necessary, tell the MDS we want to write to 2238a8599bd8SSage Weil * a larger offset. 2239a8599bd8SSage Weil */ 2240a8599bd8SSage Weil static void check_max_size(struct inode *inode, loff_t endoff) 2241a8599bd8SSage Weil { 2242a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2243a8599bd8SSage Weil int check = 0; 2244a8599bd8SSage Weil 2245a8599bd8SSage Weil /* do we need to explicitly request a larger max_size? */ 2246be655596SSage Weil spin_lock(&ci->i_ceph_lock); 22473871cbb9SYan, Zheng if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) { 2248a8599bd8SSage Weil dout("write %p at large endoff %llu, req max_size\n", 2249a8599bd8SSage Weil inode, endoff); 2250a8599bd8SSage Weil ci->i_wanted_max_size = endoff; 2251a8599bd8SSage Weil } 22523871cbb9SYan, Zheng /* duplicate ceph_check_caps()'s logic */ 22533871cbb9SYan, Zheng if (ci->i_auth_cap && 22543871cbb9SYan, Zheng (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) && 22553871cbb9SYan, Zheng ci->i_wanted_max_size > ci->i_max_size && 22563871cbb9SYan, Zheng ci->i_wanted_max_size > ci->i_requested_max_size) 22573871cbb9SYan, Zheng check = 1; 2258be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2259a8599bd8SSage Weil if (check) 2260a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2261a8599bd8SSage Weil } 2262a8599bd8SSage Weil 2263a8599bd8SSage Weil /* 2264a8599bd8SSage Weil * Wait for caps, and take cap references. If we can't get a WR cap 2265a8599bd8SSage Weil * due to a small max_size, make sure we check_max_size (and possibly 2266a8599bd8SSage Weil * ask the mds) so we don't get hung up indefinitely. 2267a8599bd8SSage Weil */ 22683738daa6SYan, Zheng int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, 22693738daa6SYan, Zheng loff_t endoff, int *got, struct page **pinned_page) 2270a8599bd8SSage Weil { 22715dda377cSYan, Zheng int _got, ret, err = 0; 2272a8599bd8SSage Weil 227310183a69SYan, Zheng ret = ceph_pool_perm_check(ci, need); 227410183a69SYan, Zheng if (ret < 0) 227510183a69SYan, Zheng return ret; 227610183a69SYan, Zheng 22775dda377cSYan, Zheng while (true) { 2278a8599bd8SSage Weil if (endoff > 0) 2279a8599bd8SSage Weil check_max_size(&ci->vfs_inode, endoff); 22805dda377cSYan, Zheng 22815dda377cSYan, Zheng err = 0; 2282c4d4a582SYan, Zheng _got = 0; 22835dda377cSYan, Zheng ret = try_get_cap_refs(ci, need, want, endoff, 22845dda377cSYan, Zheng false, &_got, &err); 22855dda377cSYan, Zheng if (ret) { 22865dda377cSYan, Zheng if (err == -EAGAIN) 22875dda377cSYan, Zheng continue; 22885dda377cSYan, Zheng if (err < 0) 22895dda377cSYan, Zheng return err; 22905dda377cSYan, Zheng } else { 2291a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 22923738daa6SYan, Zheng try_get_cap_refs(ci, need, want, endoff, 22935dda377cSYan, Zheng true, &_got, &err)); 22945dda377cSYan, Zheng if (err == -EAGAIN) 22955dda377cSYan, Zheng continue; 22965dda377cSYan, Zheng if (err < 0) 2297a8599bd8SSage Weil ret = err; 2298c4d4a582SYan, Zheng if (ret < 0) 2299c4d4a582SYan, Zheng return ret; 23005dda377cSYan, Zheng } 2301c4d4a582SYan, Zheng 2302c4d4a582SYan, Zheng if (ci->i_inline_version != CEPH_INLINE_NONE && 2303c4d4a582SYan, Zheng (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && 2304c4d4a582SYan, Zheng i_size_read(&ci->vfs_inode) > 0) { 23055dda377cSYan, Zheng struct page *page = 23065dda377cSYan, Zheng find_get_page(ci->vfs_inode.i_mapping, 0); 2307c4d4a582SYan, Zheng if (page) { 2308c4d4a582SYan, Zheng if (PageUptodate(page)) { 2309c4d4a582SYan, Zheng *pinned_page = page; 23105dda377cSYan, Zheng break; 2311c4d4a582SYan, Zheng } 2312c4d4a582SYan, Zheng page_cache_release(page); 2313c4d4a582SYan, Zheng } 2314c4d4a582SYan, Zheng /* 23155dda377cSYan, Zheng * drop cap refs first because getattr while 23165dda377cSYan, Zheng * holding * caps refs can cause deadlock. 2317c4d4a582SYan, Zheng */ 2318c4d4a582SYan, Zheng ceph_put_cap_refs(ci, _got); 2319c4d4a582SYan, Zheng _got = 0; 2320c4d4a582SYan, Zheng 23215dda377cSYan, Zheng /* 23225dda377cSYan, Zheng * getattr request will bring inline data into 23235dda377cSYan, Zheng * page cache 23245dda377cSYan, Zheng */ 2325c4d4a582SYan, Zheng ret = __ceph_do_getattr(&ci->vfs_inode, NULL, 23265dda377cSYan, Zheng CEPH_STAT_CAP_INLINE_DATA, 23275dda377cSYan, Zheng true); 2328c4d4a582SYan, Zheng if (ret < 0) 2329a8599bd8SSage Weil return ret; 23305dda377cSYan, Zheng continue; 2331c4d4a582SYan, Zheng } 23325dda377cSYan, Zheng break; 23335dda377cSYan, Zheng } 23345dda377cSYan, Zheng 2335c4d4a582SYan, Zheng *got = _got; 2336c4d4a582SYan, Zheng return 0; 2337a8599bd8SSage Weil } 2338a8599bd8SSage Weil 2339a8599bd8SSage Weil /* 2340a8599bd8SSage Weil * Take cap refs. Caller must already know we hold at least one ref 2341a8599bd8SSage Weil * on the caps in question or we don't know this is safe. 2342a8599bd8SSage Weil */ 2343a8599bd8SSage Weil void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) 2344a8599bd8SSage Weil { 2345be655596SSage Weil spin_lock(&ci->i_ceph_lock); 23465dda377cSYan, Zheng __take_cap_refs(ci, caps, false); 2347be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2348a8599bd8SSage Weil } 2349a8599bd8SSage Weil 2350a8599bd8SSage Weil /* 2351a8599bd8SSage Weil * Release cap refs. 2352a8599bd8SSage Weil * 2353a8599bd8SSage Weil * If we released the last ref on any given cap, call ceph_check_caps 2354a8599bd8SSage Weil * to release (or schedule a release). 2355a8599bd8SSage Weil * 2356a8599bd8SSage Weil * If we are releasing a WR cap (from a sync write), finalize any affected 2357a8599bd8SSage Weil * cap_snap, and wake up any waiters. 2358a8599bd8SSage Weil */ 2359a8599bd8SSage Weil void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) 2360a8599bd8SSage Weil { 2361a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2362a8599bd8SSage Weil int last = 0, put = 0, flushsnaps = 0, wake = 0; 2363a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2364a8599bd8SSage Weil 2365be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2366a8599bd8SSage Weil if (had & CEPH_CAP_PIN) 2367a8599bd8SSage Weil --ci->i_pin_ref; 2368a8599bd8SSage Weil if (had & CEPH_CAP_FILE_RD) 2369a8599bd8SSage Weil if (--ci->i_rd_ref == 0) 2370a8599bd8SSage Weil last++; 2371a8599bd8SSage Weil if (had & CEPH_CAP_FILE_CACHE) 2372a8599bd8SSage Weil if (--ci->i_rdcache_ref == 0) 2373a8599bd8SSage Weil last++; 2374a8599bd8SSage Weil if (had & CEPH_CAP_FILE_BUFFER) { 2375d3d0720dSHenry C Chang if (--ci->i_wb_ref == 0) { 2376a8599bd8SSage Weil last++; 2377a8599bd8SSage Weil put++; 2378a8599bd8SSage Weil } 2379d3d0720dSHenry C Chang dout("put_cap_refs %p wb %d -> %d (?)\n", 2380d3d0720dSHenry C Chang inode, ci->i_wb_ref+1, ci->i_wb_ref); 2381a8599bd8SSage Weil } 2382a8599bd8SSage Weil if (had & CEPH_CAP_FILE_WR) 2383a8599bd8SSage Weil if (--ci->i_wr_ref == 0) { 2384a8599bd8SSage Weil last++; 2385a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) { 2386a8599bd8SSage Weil capsnap = list_first_entry(&ci->i_cap_snaps, 2387a8599bd8SSage Weil struct ceph_cap_snap, 2388a8599bd8SSage Weil ci_item); 2389a8599bd8SSage Weil if (capsnap->writing) { 2390a8599bd8SSage Weil capsnap->writing = 0; 2391a8599bd8SSage Weil flushsnaps = 2392a8599bd8SSage Weil __ceph_finish_cap_snap(ci, 2393a8599bd8SSage Weil capsnap); 2394a8599bd8SSage Weil wake = 1; 2395a8599bd8SSage Weil } 2396a8599bd8SSage Weil } 23975dda377cSYan, Zheng if (ci->i_wrbuffer_ref_head == 0 && 23985dda377cSYan, Zheng ci->i_dirty_caps == 0 && 23995dda377cSYan, Zheng ci->i_flushing_caps == 0) { 24005dda377cSYan, Zheng BUG_ON(!ci->i_head_snapc); 24015dda377cSYan, Zheng ceph_put_snap_context(ci->i_head_snapc); 24025dda377cSYan, Zheng ci->i_head_snapc = NULL; 24035dda377cSYan, Zheng } 2404db40cc17SYan, Zheng /* see comment in __ceph_remove_cap() */ 2405db40cc17SYan, Zheng if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) 2406db40cc17SYan, Zheng drop_inode_snap_realm(ci); 2407a8599bd8SSage Weil } 2408be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2409a8599bd8SSage Weil 2410819ccbfaSSage Weil dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), 2411819ccbfaSSage Weil last ? " last" : "", put ? " put" : ""); 2412a8599bd8SSage Weil 2413a8599bd8SSage Weil if (last && !flushsnaps) 2414a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 2415a8599bd8SSage Weil else if (flushsnaps) 2416a8599bd8SSage Weil ceph_flush_snaps(ci); 2417a8599bd8SSage Weil if (wake) 241803066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2419a8599bd8SSage Weil if (put) 2420a8599bd8SSage Weil iput(inode); 2421a8599bd8SSage Weil } 2422a8599bd8SSage Weil 2423a8599bd8SSage Weil /* 2424a8599bd8SSage Weil * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap 2425a8599bd8SSage Weil * context. Adjust per-snap dirty page accounting as appropriate. 2426a8599bd8SSage Weil * Once all dirty data for a cap_snap is flushed, flush snapped file 2427a8599bd8SSage Weil * metadata back to the MDS. If we dropped the last ref, call 2428a8599bd8SSage Weil * ceph_check_caps. 2429a8599bd8SSage Weil */ 2430a8599bd8SSage Weil void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, 2431a8599bd8SSage Weil struct ceph_snap_context *snapc) 2432a8599bd8SSage Weil { 2433a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2434a8599bd8SSage Weil int last = 0; 2435819ccbfaSSage Weil int complete_capsnap = 0; 2436819ccbfaSSage Weil int drop_capsnap = 0; 2437a8599bd8SSage Weil int found = 0; 2438a8599bd8SSage Weil struct ceph_cap_snap *capsnap = NULL; 2439a8599bd8SSage Weil 2440be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2441a8599bd8SSage Weil ci->i_wrbuffer_ref -= nr; 2442a8599bd8SSage Weil last = !ci->i_wrbuffer_ref; 2443a8599bd8SSage Weil 2444a8599bd8SSage Weil if (ci->i_head_snapc == snapc) { 2445a8599bd8SSage Weil ci->i_wrbuffer_ref_head -= nr; 24467d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0 && 24475dda377cSYan, Zheng ci->i_wr_ref == 0 && 24485dda377cSYan, Zheng ci->i_dirty_caps == 0 && 24495dda377cSYan, Zheng ci->i_flushing_caps == 0) { 24507d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 2451a8599bd8SSage Weil ceph_put_snap_context(ci->i_head_snapc); 2452a8599bd8SSage Weil ci->i_head_snapc = NULL; 2453a8599bd8SSage Weil } 2454a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", 2455a8599bd8SSage Weil inode, 2456a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, 2457a8599bd8SSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 2458a8599bd8SSage Weil last ? " LAST" : ""); 2459a8599bd8SSage Weil } else { 2460a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2461a8599bd8SSage Weil if (capsnap->context == snapc) { 2462a8599bd8SSage Weil found = 1; 2463a8599bd8SSage Weil break; 2464a8599bd8SSage Weil } 2465a8599bd8SSage Weil } 2466a8599bd8SSage Weil BUG_ON(!found); 2467819ccbfaSSage Weil capsnap->dirty_pages -= nr; 2468819ccbfaSSage Weil if (capsnap->dirty_pages == 0) { 2469819ccbfaSSage Weil complete_capsnap = 1; 2470819ccbfaSSage Weil if (capsnap->dirty == 0) 2471819ccbfaSSage Weil /* cap writeback completed before we created 2472819ccbfaSSage Weil * the cap_snap; no FLUSHSNAP is needed */ 2473819ccbfaSSage Weil drop_capsnap = 1; 2474819ccbfaSSage Weil } 2475a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p cap_snap %p " 2476819ccbfaSSage Weil " snap %lld %d/%d -> %d/%d %s%s%s\n", 2477a8599bd8SSage Weil inode, capsnap, capsnap->context->seq, 2478a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, 2479a8599bd8SSage Weil ci->i_wrbuffer_ref, capsnap->dirty_pages, 2480a8599bd8SSage Weil last ? " (wrbuffer last)" : "", 2481819ccbfaSSage Weil complete_capsnap ? " (complete capsnap)" : "", 2482819ccbfaSSage Weil drop_capsnap ? " (drop capsnap)" : ""); 2483819ccbfaSSage Weil if (drop_capsnap) { 2484819ccbfaSSage Weil ceph_put_snap_context(capsnap->context); 2485819ccbfaSSage Weil list_del(&capsnap->ci_item); 2486819ccbfaSSage Weil list_del(&capsnap->flushing_item); 2487819ccbfaSSage Weil ceph_put_cap_snap(capsnap); 2488819ccbfaSSage Weil } 2489a8599bd8SSage Weil } 2490a8599bd8SSage Weil 2491be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2492a8599bd8SSage Weil 2493a8599bd8SSage Weil if (last) { 2494a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2495a8599bd8SSage Weil iput(inode); 2496819ccbfaSSage Weil } else if (complete_capsnap) { 2497a8599bd8SSage Weil ceph_flush_snaps(ci); 249803066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2499a8599bd8SSage Weil } 2500819ccbfaSSage Weil if (drop_capsnap) 2501819ccbfaSSage Weil iput(inode); 2502a8599bd8SSage Weil } 2503a8599bd8SSage Weil 2504a8599bd8SSage Weil /* 2505ca20c991SYan, Zheng * Invalidate unlinked inode's aliases, so we can drop the inode ASAP. 2506ca20c991SYan, Zheng */ 2507ca20c991SYan, Zheng static void invalidate_aliases(struct inode *inode) 2508ca20c991SYan, Zheng { 2509ca20c991SYan, Zheng struct dentry *dn, *prev = NULL; 2510ca20c991SYan, Zheng 2511ca20c991SYan, Zheng dout("invalidate_aliases inode %p\n", inode); 2512ca20c991SYan, Zheng d_prune_aliases(inode); 2513ca20c991SYan, Zheng /* 2514ca20c991SYan, Zheng * For non-directory inode, d_find_alias() only returns 2515fc12c80aSJ. Bruce Fields * hashed dentry. After calling d_invalidate(), the 2516fc12c80aSJ. Bruce Fields * dentry becomes unhashed. 2517ca20c991SYan, Zheng * 2518a8d436f0SYan, Zheng * For directory inode, d_find_alias() can return 2519fc12c80aSJ. Bruce Fields * unhashed dentry. But directory inode should have 2520ca20c991SYan, Zheng * one alias at most. 2521ca20c991SYan, Zheng */ 2522ca20c991SYan, Zheng while ((dn = d_find_alias(inode))) { 2523ca20c991SYan, Zheng if (dn == prev) { 2524ca20c991SYan, Zheng dput(dn); 2525ca20c991SYan, Zheng break; 2526ca20c991SYan, Zheng } 2527a8d436f0SYan, Zheng d_invalidate(dn); 2528ca20c991SYan, Zheng if (prev) 2529ca20c991SYan, Zheng dput(prev); 2530ca20c991SYan, Zheng prev = dn; 2531ca20c991SYan, Zheng } 2532ca20c991SYan, Zheng if (prev) 2533ca20c991SYan, Zheng dput(prev); 2534ca20c991SYan, Zheng } 2535ca20c991SYan, Zheng 2536ca20c991SYan, Zheng /* 2537a8599bd8SSage Weil * Handle a cap GRANT message from the MDS. (Note that a GRANT may 2538a8599bd8SSage Weil * actually be a revocation if it specifies a smaller cap set.) 2539a8599bd8SSage Weil * 2540be655596SSage Weil * caller holds s_mutex and i_ceph_lock, we drop both. 2541a8599bd8SSage Weil */ 25422cd698beSYan, Zheng static void handle_cap_grant(struct ceph_mds_client *mdsc, 25432cd698beSYan, Zheng struct inode *inode, struct ceph_mds_caps *grant, 2544fb01d1f8SYan, Zheng u64 inline_version, 2545fb01d1f8SYan, Zheng void *inline_data, int inline_len, 25462cd698beSYan, Zheng struct ceph_buffer *xattr_buf, 2547a8599bd8SSage Weil struct ceph_mds_session *session, 25482cd698beSYan, Zheng struct ceph_cap *cap, int issued) 2549be655596SSage Weil __releases(ci->i_ceph_lock) 2550982d6011SYan, Zheng __releases(mdsc->snap_rwsem) 2551a8599bd8SSage Weil { 2552a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2553a8599bd8SSage Weil int mds = session->s_mds; 25542f56f56aSSage Weil int seq = le32_to_cpu(grant->seq); 2555a8599bd8SSage Weil int newcaps = le32_to_cpu(grant->caps); 25562cd698beSYan, Zheng int used, wanted, dirty; 2557a8599bd8SSage Weil u64 size = le64_to_cpu(grant->size); 2558a8599bd8SSage Weil u64 max_size = le64_to_cpu(grant->max_size); 2559a8599bd8SSage Weil struct timespec mtime, atime, ctime; 256015637c8bSSage Weil int check_caps = 0; 2561ab6c2c3eSFabian Frederick bool wake = false; 2562ab6c2c3eSFabian Frederick bool writeback = false; 2563ab6c2c3eSFabian Frederick bool queue_trunc = false; 2564ab6c2c3eSFabian Frederick bool queue_invalidate = false; 2565ab6c2c3eSFabian Frederick bool queue_revalidate = false; 2566ab6c2c3eSFabian Frederick bool deleted_inode = false; 256731c542a1SYan, Zheng bool fill_inline = false; 2568a8599bd8SSage Weil 25692f56f56aSSage Weil dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", 25702f56f56aSSage Weil inode, cap, mds, seq, ceph_cap_string(newcaps)); 2571a8599bd8SSage Weil dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, 2572a8599bd8SSage Weil inode->i_size); 2573a8599bd8SSage Weil 257411df2dfbSYan, Zheng 257511df2dfbSYan, Zheng /* 257611df2dfbSYan, Zheng * auth mds of the inode changed. we received the cap export message, 257711df2dfbSYan, Zheng * but still haven't received the cap import message. handle_cap_export 257811df2dfbSYan, Zheng * updated the new auth MDS' cap. 257911df2dfbSYan, Zheng * 258011df2dfbSYan, Zheng * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message 258111df2dfbSYan, Zheng * that was sent before the cap import message. So don't remove caps. 258211df2dfbSYan, Zheng */ 258311df2dfbSYan, Zheng if (ceph_seq_cmp(seq, cap->seq) <= 0) { 258411df2dfbSYan, Zheng WARN_ON(cap != ci->i_auth_cap); 258511df2dfbSYan, Zheng WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id)); 258611df2dfbSYan, Zheng seq = cap->seq; 258711df2dfbSYan, Zheng newcaps |= cap->issued; 258811df2dfbSYan, Zheng } 258911df2dfbSYan, Zheng 2590a8599bd8SSage Weil /* 2591a8599bd8SSage Weil * If CACHE is being revoked, and we have no dirty buffers, 2592a8599bd8SSage Weil * try to invalidate (once). (If there are dirty buffers, we 2593a8599bd8SSage Weil * will invalidate _after_ writeback.) 2594a8599bd8SSage Weil */ 25953b454c49SSage Weil if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && 25963b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 2597bcd2cbd1SYehuda Sadeh !ci->i_wrbuffer_ref) { 2598e9075743SLi Wang if (try_nonblocking_invalidate(inode)) { 2599a8599bd8SSage Weil /* there were locked pages.. invalidate later 2600a8599bd8SSage Weil in a separate thread. */ 2601a8599bd8SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 2602ab6c2c3eSFabian Frederick queue_invalidate = true; 2603a8599bd8SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 2604a8599bd8SSage Weil } 2605a8599bd8SSage Weil } 260699ccbd22SMilosz Tanski 260799ccbd22SMilosz Tanski ceph_fscache_invalidate(inode); 2608a8599bd8SSage Weil } 2609a8599bd8SSage Weil 2610a8599bd8SSage Weil /* side effects now are allowed */ 2611685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 261211df2dfbSYan, Zheng cap->seq = seq; 2613a8599bd8SSage Weil 2614a8599bd8SSage Weil __check_cap_issue(ci, cap, newcaps); 2615a8599bd8SSage Weil 2616f98a128aSYan, Zheng if ((newcaps & CEPH_CAP_AUTH_SHARED) && 2617f98a128aSYan, Zheng (issued & CEPH_CAP_AUTH_EXCL) == 0) { 2618a8599bd8SSage Weil inode->i_mode = le32_to_cpu(grant->mode); 261905cb11c1SEric W. Biederman inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid)); 262005cb11c1SEric W. Biederman inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid)); 2621a8599bd8SSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 2622bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 2623bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid)); 2624a8599bd8SSage Weil } 2625a8599bd8SSage Weil 2626f98a128aSYan, Zheng if ((newcaps & CEPH_CAP_AUTH_SHARED) && 2627f98a128aSYan, Zheng (issued & CEPH_CAP_LINK_EXCL) == 0) { 2628bfe86848SMiklos Szeredi set_nlink(inode, le32_to_cpu(grant->nlink)); 2629ca20c991SYan, Zheng if (inode->i_nlink == 0 && 2630ca20c991SYan, Zheng (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) 2631ab6c2c3eSFabian Frederick deleted_inode = true; 2632ca20c991SYan, Zheng } 2633a8599bd8SSage Weil 2634a8599bd8SSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) { 2635a8599bd8SSage Weil int len = le32_to_cpu(grant->xattr_len); 2636a8599bd8SSage Weil u64 version = le64_to_cpu(grant->xattr_version); 2637a8599bd8SSage Weil 2638a8599bd8SSage Weil if (version > ci->i_xattrs.version) { 2639a8599bd8SSage Weil dout(" got new xattrs v%llu on %p len %d\n", 2640a8599bd8SSage Weil version, inode, len); 2641a8599bd8SSage Weil if (ci->i_xattrs.blob) 2642a8599bd8SSage Weil ceph_buffer_put(ci->i_xattrs.blob); 2643a8599bd8SSage Weil ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); 2644a8599bd8SSage Weil ci->i_xattrs.version = version; 26457221fe4cSGuangliang Zhao ceph_forget_all_cached_acls(inode); 2646a8599bd8SSage Weil } 2647a8599bd8SSage Weil } 2648a8599bd8SSage Weil 264999ccbd22SMilosz Tanski /* Do we need to revalidate our fscache cookie. Don't bother on the 265099ccbd22SMilosz Tanski * first cache cap as we already validate at cookie creation time. */ 265199ccbd22SMilosz Tanski if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1) 2652ab6c2c3eSFabian Frederick queue_revalidate = true; 265399ccbd22SMilosz Tanski 2654f98a128aSYan, Zheng if (newcaps & CEPH_CAP_ANY_RD) { 2655f98a128aSYan, Zheng /* ctime/mtime/atime? */ 2656a8599bd8SSage Weil ceph_decode_timespec(&mtime, &grant->mtime); 2657a8599bd8SSage Weil ceph_decode_timespec(&atime, &grant->atime); 2658a8599bd8SSage Weil ceph_decode_timespec(&ctime, &grant->ctime); 2659a8599bd8SSage Weil ceph_fill_file_time(inode, issued, 2660f98a128aSYan, Zheng le32_to_cpu(grant->time_warp_seq), 2661f98a128aSYan, Zheng &ctime, &mtime, &atime); 2662f98a128aSYan, Zheng } 2663a8599bd8SSage Weil 2664f98a128aSYan, Zheng if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { 266511df2dfbSYan, Zheng /* file layout may have changed */ 266611df2dfbSYan, Zheng ci->i_layout = grant->layout; 2667f98a128aSYan, Zheng /* size/truncate_seq? */ 2668f98a128aSYan, Zheng queue_trunc = ceph_fill_file_size(inode, issued, 2669f98a128aSYan, Zheng le32_to_cpu(grant->truncate_seq), 2670f98a128aSYan, Zheng le64_to_cpu(grant->truncate_size), 2671f98a128aSYan, Zheng size); 2672a8599bd8SSage Weil /* max size increase? */ 26735e62ad30SYan, Zheng if (ci->i_auth_cap == cap && max_size != ci->i_max_size) { 2674f98a128aSYan, Zheng dout("max_size %lld -> %llu\n", 2675f98a128aSYan, Zheng ci->i_max_size, max_size); 2676a8599bd8SSage Weil ci->i_max_size = max_size; 2677a8599bd8SSage Weil if (max_size >= ci->i_wanted_max_size) { 2678a8599bd8SSage Weil ci->i_wanted_max_size = 0; /* reset */ 2679a8599bd8SSage Weil ci->i_requested_max_size = 0; 2680a8599bd8SSage Weil } 2681ab6c2c3eSFabian Frederick wake = true; 2682a8599bd8SSage Weil } 2683f98a128aSYan, Zheng } 2684a8599bd8SSage Weil 2685a8599bd8SSage Weil /* check cap bits */ 2686a8599bd8SSage Weil wanted = __ceph_caps_wanted(ci); 2687a8599bd8SSage Weil used = __ceph_caps_used(ci); 2688a8599bd8SSage Weil dirty = __ceph_caps_dirty(ci); 2689a8599bd8SSage Weil dout(" my wanted = %s, used = %s, dirty %s\n", 2690a8599bd8SSage Weil ceph_cap_string(wanted), 2691a8599bd8SSage Weil ceph_cap_string(used), 2692a8599bd8SSage Weil ceph_cap_string(dirty)); 2693a8599bd8SSage Weil if (wanted != le32_to_cpu(grant->wanted)) { 2694a8599bd8SSage Weil dout("mds wanted %s -> %s\n", 2695a8599bd8SSage Weil ceph_cap_string(le32_to_cpu(grant->wanted)), 2696a8599bd8SSage Weil ceph_cap_string(wanted)); 2697390306c3SYan, Zheng /* imported cap may not have correct mds_wanted */ 2698390306c3SYan, Zheng if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) 2699390306c3SYan, Zheng check_caps = 1; 2700a8599bd8SSage Weil } 2701a8599bd8SSage Weil 2702a8599bd8SSage Weil /* revocation, grant, or no-op? */ 2703a8599bd8SSage Weil if (cap->issued & ~newcaps) { 27043b454c49SSage Weil int revoking = cap->issued & ~newcaps; 27053b454c49SSage Weil 27063b454c49SSage Weil dout("revocation: %s -> %s (revoking %s)\n", 27073b454c49SSage Weil ceph_cap_string(cap->issued), 27083b454c49SSage Weil ceph_cap_string(newcaps), 27093b454c49SSage Weil ceph_cap_string(revoking)); 27100eb6cd49SSage Weil if (revoking & used & CEPH_CAP_FILE_BUFFER) 2711ab6c2c3eSFabian Frederick writeback = true; /* initiate writeback; will delay ack */ 27123b454c49SSage Weil else if (revoking == CEPH_CAP_FILE_CACHE && 27133b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 27143b454c49SSage Weil queue_invalidate) 27153b454c49SSage Weil ; /* do nothing yet, invalidation will be queued */ 27163b454c49SSage Weil else if (cap == ci->i_auth_cap) 27173b454c49SSage Weil check_caps = 1; /* check auth cap only */ 27183b454c49SSage Weil else 27193b454c49SSage Weil check_caps = 2; /* check all caps */ 2720a8599bd8SSage Weil cap->issued = newcaps; 2721978097c9SSage Weil cap->implemented |= newcaps; 2722a8599bd8SSage Weil } else if (cap->issued == newcaps) { 2723a8599bd8SSage Weil dout("caps unchanged: %s -> %s\n", 2724a8599bd8SSage Weil ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); 2725a8599bd8SSage Weil } else { 2726a8599bd8SSage Weil dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), 2727a8599bd8SSage Weil ceph_cap_string(newcaps)); 27286ee6b953SYan, Zheng /* non-auth MDS is revoking the newly grant caps ? */ 27296ee6b953SYan, Zheng if (cap == ci->i_auth_cap && 27306ee6b953SYan, Zheng __ceph_caps_revoking_other(ci, cap, newcaps)) 27316ee6b953SYan, Zheng check_caps = 2; 27326ee6b953SYan, Zheng 2733a8599bd8SSage Weil cap->issued = newcaps; 2734a8599bd8SSage Weil cap->implemented |= newcaps; /* add bits only, to 2735a8599bd8SSage Weil * avoid stepping on a 2736a8599bd8SSage Weil * pending revocation */ 2737ab6c2c3eSFabian Frederick wake = true; 2738a8599bd8SSage Weil } 2739978097c9SSage Weil BUG_ON(cap->issued & ~cap->implemented); 2740a8599bd8SSage Weil 274131c542a1SYan, Zheng if (inline_version > 0 && inline_version >= ci->i_inline_version) { 274231c542a1SYan, Zheng ci->i_inline_version = inline_version; 274331c542a1SYan, Zheng if (ci->i_inline_version != CEPH_INLINE_NONE && 274431c542a1SYan, Zheng (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO))) 274531c542a1SYan, Zheng fill_inline = true; 274631c542a1SYan, Zheng } 274731c542a1SYan, Zheng 2748be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 274999ccbd22SMilosz Tanski 27502cd698beSYan, Zheng if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) { 27512cd698beSYan, Zheng kick_flushing_inode_caps(mdsc, session, inode); 27522cd698beSYan, Zheng up_read(&mdsc->snap_rwsem); 27532cd698beSYan, Zheng if (newcaps & ~issued) 2754ab6c2c3eSFabian Frederick wake = true; 27552cd698beSYan, Zheng } 27562cd698beSYan, Zheng 275731c542a1SYan, Zheng if (fill_inline) 275831c542a1SYan, Zheng ceph_fill_inline_data(inode, NULL, inline_data, inline_len); 275931c542a1SYan, Zheng 2760c6bcda6fSYan, Zheng if (queue_trunc) { 2761c6bcda6fSYan, Zheng ceph_queue_vmtruncate(inode); 2762c6bcda6fSYan, Zheng ceph_queue_revalidate(inode); 2763c6bcda6fSYan, Zheng } else if (queue_revalidate) 2764c6bcda6fSYan, Zheng ceph_queue_revalidate(inode); 2765c6bcda6fSYan, Zheng 27663c6f6b79SSage Weil if (writeback) 2767a8599bd8SSage Weil /* 2768a8599bd8SSage Weil * queue inode for writeback: we can't actually call 2769a8599bd8SSage Weil * filemap_write_and_wait, etc. from message handler 2770a8599bd8SSage Weil * context. 2771a8599bd8SSage Weil */ 27723c6f6b79SSage Weil ceph_queue_writeback(inode); 27733c6f6b79SSage Weil if (queue_invalidate) 27743c6f6b79SSage Weil ceph_queue_invalidate(inode); 2775ca20c991SYan, Zheng if (deleted_inode) 2776ca20c991SYan, Zheng invalidate_aliases(inode); 2777a8599bd8SSage Weil if (wake) 277803066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 277915637c8bSSage Weil 278015637c8bSSage Weil if (check_caps == 1) 278115637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, 278215637c8bSSage Weil session); 278315637c8bSSage Weil else if (check_caps == 2) 278415637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); 278515637c8bSSage Weil else 278615637c8bSSage Weil mutex_unlock(&session->s_mutex); 2787a8599bd8SSage Weil } 2788a8599bd8SSage Weil 2789a8599bd8SSage Weil /* 2790a8599bd8SSage Weil * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the 2791a8599bd8SSage Weil * MDS has been safely committed. 2792a8599bd8SSage Weil */ 27936df058c0SSage Weil static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, 2794a8599bd8SSage Weil struct ceph_mds_caps *m, 2795a8599bd8SSage Weil struct ceph_mds_session *session, 2796a8599bd8SSage Weil struct ceph_cap *cap) 2797be655596SSage Weil __releases(ci->i_ceph_lock) 2798a8599bd8SSage Weil { 2799a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 28003d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 2801a8599bd8SSage Weil unsigned seq = le32_to_cpu(m->seq); 2802a8599bd8SSage Weil int dirty = le32_to_cpu(m->dirty); 2803a8599bd8SSage Weil int cleaned = 0; 2804afcdaea3SSage Weil int drop = 0; 2805a8599bd8SSage Weil int i; 2806a8599bd8SSage Weil 2807a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 2808a8599bd8SSage Weil if ((dirty & (1 << i)) && 28093231300bSYan, Zheng (u16)flush_tid == ci->i_cap_flush_tid[i]) 2810a8599bd8SSage Weil cleaned |= 1 << i; 2811a8599bd8SSage Weil 2812a8599bd8SSage Weil dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," 2813a8599bd8SSage Weil " flushing %s -> %s\n", 2814a8599bd8SSage Weil inode, session->s_mds, seq, ceph_cap_string(dirty), 2815a8599bd8SSage Weil ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), 2816a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps & ~cleaned)); 2817a8599bd8SSage Weil 2818a8599bd8SSage Weil if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned)) 2819a8599bd8SSage Weil goto out; 2820a8599bd8SSage Weil 2821a8599bd8SSage Weil ci->i_flushing_caps &= ~cleaned; 2822a8599bd8SSage Weil 2823a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2824a8599bd8SSage Weil if (ci->i_flushing_caps == 0) { 2825a8599bd8SSage Weil list_del_init(&ci->i_flushing_item); 2826a8599bd8SSage Weil if (!list_empty(&session->s_cap_flushing)) 2827a8599bd8SSage Weil dout(" mds%d still flushing cap on %p\n", 2828a8599bd8SSage Weil session->s_mds, 2829a8599bd8SSage Weil &list_entry(session->s_cap_flushing.next, 2830a8599bd8SSage Weil struct ceph_inode_info, 2831a8599bd8SSage Weil i_flushing_item)->vfs_inode); 2832a8599bd8SSage Weil mdsc->num_cap_flushing--; 283303066f23SYehuda Sadeh wake_up_all(&mdsc->cap_flushing_wq); 2834a8599bd8SSage Weil dout(" inode %p now !flushing\n", inode); 2835afcdaea3SSage Weil 2836afcdaea3SSage Weil if (ci->i_dirty_caps == 0) { 2837a8599bd8SSage Weil dout(" inode %p now clean\n", inode); 2838afcdaea3SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 2839afcdaea3SSage Weil drop = 1; 28405dda377cSYan, Zheng if (ci->i_wr_ref == 0 && 28415dda377cSYan, Zheng ci->i_wrbuffer_ref_head == 0) { 28427d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 28437d8cb26dSSage Weil ceph_put_snap_context(ci->i_head_snapc); 28447d8cb26dSSage Weil ci->i_head_snapc = NULL; 28457d8cb26dSSage Weil } 284676e3b390SSage Weil } else { 284776e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 2848afcdaea3SSage Weil } 2849a8599bd8SSage Weil } 2850a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 285103066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2852a8599bd8SSage Weil 2853a8599bd8SSage Weil out: 2854be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2855afcdaea3SSage Weil if (drop) 2856a8599bd8SSage Weil iput(inode); 2857a8599bd8SSage Weil } 2858a8599bd8SSage Weil 2859a8599bd8SSage Weil /* 2860a8599bd8SSage Weil * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can 2861a8599bd8SSage Weil * throw away our cap_snap. 2862a8599bd8SSage Weil * 2863a8599bd8SSage Weil * Caller hold s_mutex. 2864a8599bd8SSage Weil */ 28656df058c0SSage Weil static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, 2866a8599bd8SSage Weil struct ceph_mds_caps *m, 2867a8599bd8SSage Weil struct ceph_mds_session *session) 2868a8599bd8SSage Weil { 2869a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2870a8599bd8SSage Weil u64 follows = le64_to_cpu(m->snap_follows); 2871a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2872a8599bd8SSage Weil int drop = 0; 2873a8599bd8SSage Weil 2874a8599bd8SSage Weil dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", 2875a8599bd8SSage Weil inode, ci, session->s_mds, follows); 2876a8599bd8SSage Weil 2877be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2878a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2879a8599bd8SSage Weil if (capsnap->follows == follows) { 2880a8599bd8SSage Weil if (capsnap->flush_tid != flush_tid) { 2881a8599bd8SSage Weil dout(" cap_snap %p follows %lld tid %lld !=" 2882a8599bd8SSage Weil " %lld\n", capsnap, follows, 2883a8599bd8SSage Weil flush_tid, capsnap->flush_tid); 2884a8599bd8SSage Weil break; 2885a8599bd8SSage Weil } 2886a8599bd8SSage Weil WARN_ON(capsnap->dirty_pages || capsnap->writing); 2887819ccbfaSSage Weil dout(" removing %p cap_snap %p follows %lld\n", 2888819ccbfaSSage Weil inode, capsnap, follows); 2889a8599bd8SSage Weil ceph_put_snap_context(capsnap->context); 2890a8599bd8SSage Weil list_del(&capsnap->ci_item); 2891a8599bd8SSage Weil list_del(&capsnap->flushing_item); 2892a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 2893a8599bd8SSage Weil drop = 1; 2894a8599bd8SSage Weil break; 2895a8599bd8SSage Weil } else { 2896a8599bd8SSage Weil dout(" skipping cap_snap %p follows %lld\n", 2897a8599bd8SSage Weil capsnap, capsnap->follows); 2898a8599bd8SSage Weil } 2899a8599bd8SSage Weil } 2900be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2901a8599bd8SSage Weil if (drop) 2902a8599bd8SSage Weil iput(inode); 2903a8599bd8SSage Weil } 2904a8599bd8SSage Weil 2905a8599bd8SSage Weil /* 2906a8599bd8SSage Weil * Handle TRUNC from MDS, indicating file truncation. 2907a8599bd8SSage Weil * 2908a8599bd8SSage Weil * caller hold s_mutex. 2909a8599bd8SSage Weil */ 2910a8599bd8SSage Weil static void handle_cap_trunc(struct inode *inode, 2911a8599bd8SSage Weil struct ceph_mds_caps *trunc, 2912a8599bd8SSage Weil struct ceph_mds_session *session) 2913be655596SSage Weil __releases(ci->i_ceph_lock) 2914a8599bd8SSage Weil { 2915a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2916a8599bd8SSage Weil int mds = session->s_mds; 2917a8599bd8SSage Weil int seq = le32_to_cpu(trunc->seq); 2918a8599bd8SSage Weil u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); 2919a8599bd8SSage Weil u64 truncate_size = le64_to_cpu(trunc->truncate_size); 2920a8599bd8SSage Weil u64 size = le64_to_cpu(trunc->size); 2921a8599bd8SSage Weil int implemented = 0; 2922a8599bd8SSage Weil int dirty = __ceph_caps_dirty(ci); 2923a8599bd8SSage Weil int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); 2924a8599bd8SSage Weil int queue_trunc = 0; 2925a8599bd8SSage Weil 2926a8599bd8SSage Weil issued |= implemented | dirty; 2927a8599bd8SSage Weil 2928a8599bd8SSage Weil dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", 2929a8599bd8SSage Weil inode, mds, seq, truncate_size, truncate_seq); 2930a8599bd8SSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 2931a8599bd8SSage Weil truncate_seq, truncate_size, size); 2932be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2933a8599bd8SSage Weil 293499ccbd22SMilosz Tanski if (queue_trunc) { 29353c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 293699ccbd22SMilosz Tanski ceph_fscache_invalidate(inode); 293799ccbd22SMilosz Tanski } 2938a8599bd8SSage Weil } 2939a8599bd8SSage Weil 2940a8599bd8SSage Weil /* 2941a8599bd8SSage Weil * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a 2942a8599bd8SSage Weil * different one. If we are the most recent migration we've seen (as 2943a8599bd8SSage Weil * indicated by mseq), make note of the migrating cap bits for the 2944a8599bd8SSage Weil * duration (until we see the corresponding IMPORT). 2945a8599bd8SSage Weil * 2946a8599bd8SSage Weil * caller holds s_mutex 2947a8599bd8SSage Weil */ 2948a8599bd8SSage Weil static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, 294911df2dfbSYan, Zheng struct ceph_mds_cap_peer *ph, 295011df2dfbSYan, Zheng struct ceph_mds_session *session) 2951a8599bd8SSage Weil { 2952db354052SSage Weil struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 295311df2dfbSYan, Zheng struct ceph_mds_session *tsession = NULL; 2954d9df2783SYan, Zheng struct ceph_cap *cap, *tcap, *new_cap = NULL; 2955a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 295611df2dfbSYan, Zheng u64 t_cap_id; 2957a8599bd8SSage Weil unsigned mseq = le32_to_cpu(ex->migrate_seq); 295811df2dfbSYan, Zheng unsigned t_seq, t_mseq; 295911df2dfbSYan, Zheng int target, issued; 296011df2dfbSYan, Zheng int mds = session->s_mds; 2961a8599bd8SSage Weil 296211df2dfbSYan, Zheng if (ph) { 296311df2dfbSYan, Zheng t_cap_id = le64_to_cpu(ph->cap_id); 296411df2dfbSYan, Zheng t_seq = le32_to_cpu(ph->seq); 296511df2dfbSYan, Zheng t_mseq = le32_to_cpu(ph->mseq); 296611df2dfbSYan, Zheng target = le32_to_cpu(ph->mds); 296711df2dfbSYan, Zheng } else { 296811df2dfbSYan, Zheng t_cap_id = t_seq = t_mseq = 0; 296911df2dfbSYan, Zheng target = -1; 297011df2dfbSYan, Zheng } 2971a8599bd8SSage Weil 297211df2dfbSYan, Zheng dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n", 297311df2dfbSYan, Zheng inode, ci, mds, mseq, target); 297411df2dfbSYan, Zheng retry: 2975be655596SSage Weil spin_lock(&ci->i_ceph_lock); 297611df2dfbSYan, Zheng cap = __get_cap_for_mds(ci, mds); 2977ca665e02SYan, Zheng if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id)) 297811df2dfbSYan, Zheng goto out_unlock; 2979a8599bd8SSage Weil 298011df2dfbSYan, Zheng if (target < 0) { 298111df2dfbSYan, Zheng __ceph_remove_cap(cap, false); 298211df2dfbSYan, Zheng goto out_unlock; 2983a8599bd8SSage Weil } 2984a8599bd8SSage Weil 2985154f42c2SSage Weil /* 298611df2dfbSYan, Zheng * now we know we haven't received the cap import message yet 298711df2dfbSYan, Zheng * because the exported cap still exist. 2988154f42c2SSage Weil */ 2989db354052SSage Weil 299011df2dfbSYan, Zheng issued = cap->issued; 299111df2dfbSYan, Zheng WARN_ON(issued != cap->implemented); 299211df2dfbSYan, Zheng 299311df2dfbSYan, Zheng tcap = __get_cap_for_mds(ci, target); 299411df2dfbSYan, Zheng if (tcap) { 299511df2dfbSYan, Zheng /* already have caps from the target */ 299611df2dfbSYan, Zheng if (tcap->cap_id != t_cap_id || 299711df2dfbSYan, Zheng ceph_seq_cmp(tcap->seq, t_seq) < 0) { 299811df2dfbSYan, Zheng dout(" updating import cap %p mds%d\n", tcap, target); 299911df2dfbSYan, Zheng tcap->cap_id = t_cap_id; 300011df2dfbSYan, Zheng tcap->seq = t_seq - 1; 300111df2dfbSYan, Zheng tcap->issue_seq = t_seq - 1; 300211df2dfbSYan, Zheng tcap->mseq = t_mseq; 300311df2dfbSYan, Zheng tcap->issued |= issued; 300411df2dfbSYan, Zheng tcap->implemented |= issued; 300511df2dfbSYan, Zheng if (cap == ci->i_auth_cap) 300611df2dfbSYan, Zheng ci->i_auth_cap = tcap; 300711df2dfbSYan, Zheng if (ci->i_flushing_caps && ci->i_auth_cap == tcap) { 3008db354052SSage Weil spin_lock(&mdsc->cap_dirty_lock); 300911df2dfbSYan, Zheng list_move_tail(&ci->i_flushing_item, 301011df2dfbSYan, Zheng &tcap->session->s_cap_flushing); 3011db354052SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 3012a8599bd8SSage Weil } 3013a8599bd8SSage Weil } 301411df2dfbSYan, Zheng __ceph_remove_cap(cap, false); 301511df2dfbSYan, Zheng goto out_unlock; 3016d9df2783SYan, Zheng } else if (tsession) { 301711df2dfbSYan, Zheng /* add placeholder for the export tagert */ 3018d9df2783SYan, Zheng int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0; 301911df2dfbSYan, Zheng ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0, 3020d9df2783SYan, Zheng t_seq - 1, t_mseq, (u64)-1, flag, &new_cap); 3021d9df2783SYan, Zheng 3022d9df2783SYan, Zheng __ceph_remove_cap(cap, false); 3023d9df2783SYan, Zheng goto out_unlock; 302411df2dfbSYan, Zheng } 3025a8599bd8SSage Weil 3026be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 302711df2dfbSYan, Zheng mutex_unlock(&session->s_mutex); 302811df2dfbSYan, Zheng 302911df2dfbSYan, Zheng /* open target session */ 303011df2dfbSYan, Zheng tsession = ceph_mdsc_open_export_target_session(mdsc, target); 303111df2dfbSYan, Zheng if (!IS_ERR(tsession)) { 303211df2dfbSYan, Zheng if (mds > target) { 303311df2dfbSYan, Zheng mutex_lock(&session->s_mutex); 303411df2dfbSYan, Zheng mutex_lock_nested(&tsession->s_mutex, 303511df2dfbSYan, Zheng SINGLE_DEPTH_NESTING); 303611df2dfbSYan, Zheng } else { 303711df2dfbSYan, Zheng mutex_lock(&tsession->s_mutex); 303811df2dfbSYan, Zheng mutex_lock_nested(&session->s_mutex, 303911df2dfbSYan, Zheng SINGLE_DEPTH_NESTING); 304011df2dfbSYan, Zheng } 304111df2dfbSYan, Zheng ceph_add_cap_releases(mdsc, tsession); 3042d9df2783SYan, Zheng new_cap = ceph_get_cap(mdsc, NULL); 304311df2dfbSYan, Zheng } else { 304411df2dfbSYan, Zheng WARN_ON(1); 304511df2dfbSYan, Zheng tsession = NULL; 304611df2dfbSYan, Zheng target = -1; 304711df2dfbSYan, Zheng } 304811df2dfbSYan, Zheng goto retry; 304911df2dfbSYan, Zheng 305011df2dfbSYan, Zheng out_unlock: 305111df2dfbSYan, Zheng spin_unlock(&ci->i_ceph_lock); 305211df2dfbSYan, Zheng mutex_unlock(&session->s_mutex); 305311df2dfbSYan, Zheng if (tsession) { 305411df2dfbSYan, Zheng mutex_unlock(&tsession->s_mutex); 305511df2dfbSYan, Zheng ceph_put_mds_session(tsession); 305611df2dfbSYan, Zheng } 3057d9df2783SYan, Zheng if (new_cap) 3058d9df2783SYan, Zheng ceph_put_cap(mdsc, new_cap); 3059a8599bd8SSage Weil } 3060a8599bd8SSage Weil 3061a8599bd8SSage Weil /* 30622cd698beSYan, Zheng * Handle cap IMPORT. 3063a8599bd8SSage Weil * 30642cd698beSYan, Zheng * caller holds s_mutex. acquires i_ceph_lock 3065a8599bd8SSage Weil */ 3066a8599bd8SSage Weil static void handle_cap_import(struct ceph_mds_client *mdsc, 3067a8599bd8SSage Weil struct inode *inode, struct ceph_mds_caps *im, 30684ee6a914SYan, Zheng struct ceph_mds_cap_peer *ph, 3069a8599bd8SSage Weil struct ceph_mds_session *session, 30702cd698beSYan, Zheng struct ceph_cap **target_cap, int *old_issued) 30712cd698beSYan, Zheng __acquires(ci->i_ceph_lock) 3072a8599bd8SSage Weil { 3073a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 30742cd698beSYan, Zheng struct ceph_cap *cap, *ocap, *new_cap = NULL; 3075a8599bd8SSage Weil int mds = session->s_mds; 30762cd698beSYan, Zheng int issued; 30772cd698beSYan, Zheng unsigned caps = le32_to_cpu(im->caps); 3078a8599bd8SSage Weil unsigned wanted = le32_to_cpu(im->wanted); 3079a8599bd8SSage Weil unsigned seq = le32_to_cpu(im->seq); 3080a8599bd8SSage Weil unsigned mseq = le32_to_cpu(im->migrate_seq); 3081a8599bd8SSage Weil u64 realmino = le64_to_cpu(im->realm); 3082a8599bd8SSage Weil u64 cap_id = le64_to_cpu(im->cap_id); 30834ee6a914SYan, Zheng u64 p_cap_id; 30844ee6a914SYan, Zheng int peer; 3085a8599bd8SSage Weil 30864ee6a914SYan, Zheng if (ph) { 30874ee6a914SYan, Zheng p_cap_id = le64_to_cpu(ph->cap_id); 30884ee6a914SYan, Zheng peer = le32_to_cpu(ph->mds); 3089a8599bd8SSage Weil } else { 30904ee6a914SYan, Zheng p_cap_id = 0; 30914ee6a914SYan, Zheng peer = -1; 3092a8599bd8SSage Weil } 3093a8599bd8SSage Weil 30944ee6a914SYan, Zheng dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n", 30954ee6a914SYan, Zheng inode, ci, mds, mseq, peer); 30964ee6a914SYan, Zheng 3097d9df2783SYan, Zheng retry: 30984ee6a914SYan, Zheng spin_lock(&ci->i_ceph_lock); 3099d9df2783SYan, Zheng cap = __get_cap_for_mds(ci, mds); 3100d9df2783SYan, Zheng if (!cap) { 3101d9df2783SYan, Zheng if (!new_cap) { 3102d9df2783SYan, Zheng spin_unlock(&ci->i_ceph_lock); 3103d9df2783SYan, Zheng new_cap = ceph_get_cap(mdsc, NULL); 3104d9df2783SYan, Zheng goto retry; 3105d9df2783SYan, Zheng } 31062cd698beSYan, Zheng cap = new_cap; 31072cd698beSYan, Zheng } else { 31082cd698beSYan, Zheng if (new_cap) { 31092cd698beSYan, Zheng ceph_put_cap(mdsc, new_cap); 31102cd698beSYan, Zheng new_cap = NULL; 31112cd698beSYan, Zheng } 3112d9df2783SYan, Zheng } 3113d9df2783SYan, Zheng 31142cd698beSYan, Zheng __ceph_caps_issued(ci, &issued); 31152cd698beSYan, Zheng issued |= __ceph_caps_dirty(ci); 31162cd698beSYan, Zheng 31172cd698beSYan, Zheng ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq, 3118d9df2783SYan, Zheng realmino, CEPH_CAP_FLAG_AUTH, &new_cap); 3119d9df2783SYan, Zheng 31202cd698beSYan, Zheng ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL; 31212cd698beSYan, Zheng if (ocap && ocap->cap_id == p_cap_id) { 31224ee6a914SYan, Zheng dout(" remove export cap %p mds%d flags %d\n", 31232cd698beSYan, Zheng ocap, peer, ph->flags); 31244ee6a914SYan, Zheng if ((ph->flags & CEPH_CAP_FLAG_AUTH) && 31252cd698beSYan, Zheng (ocap->seq != le32_to_cpu(ph->seq) || 31262cd698beSYan, Zheng ocap->mseq != le32_to_cpu(ph->mseq))) { 31274ee6a914SYan, Zheng pr_err("handle_cap_import: mismatched seq/mseq: " 31284ee6a914SYan, Zheng "ino (%llx.%llx) mds%d seq %d mseq %d " 31294ee6a914SYan, Zheng "importer mds%d has peer seq %d mseq %d\n", 31302cd698beSYan, Zheng ceph_vinop(inode), peer, ocap->seq, 31312cd698beSYan, Zheng ocap->mseq, mds, le32_to_cpu(ph->seq), 31324ee6a914SYan, Zheng le32_to_cpu(ph->mseq)); 31334ee6a914SYan, Zheng } 31342cd698beSYan, Zheng __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE)); 31354ee6a914SYan, Zheng } 31364ee6a914SYan, Zheng 31374ee6a914SYan, Zheng /* make sure we re-request max_size, if necessary */ 31384ee6a914SYan, Zheng ci->i_wanted_max_size = 0; 31394ee6a914SYan, Zheng ci->i_requested_max_size = 0; 31404ee6a914SYan, Zheng 31412cd698beSYan, Zheng *old_issued = issued; 31422cd698beSYan, Zheng *target_cap = cap; 3143a8599bd8SSage Weil } 3144a8599bd8SSage Weil 3145a8599bd8SSage Weil /* 3146a8599bd8SSage Weil * Handle a caps message from the MDS. 3147a8599bd8SSage Weil * 3148a8599bd8SSage Weil * Identify the appropriate session, inode, and call the right handler 3149a8599bd8SSage Weil * based on the cap op. 3150a8599bd8SSage Weil */ 3151a8599bd8SSage Weil void ceph_handle_caps(struct ceph_mds_session *session, 3152a8599bd8SSage Weil struct ceph_msg *msg) 3153a8599bd8SSage Weil { 3154a8599bd8SSage Weil struct ceph_mds_client *mdsc = session->s_mdsc; 31553d14c5d2SYehuda Sadeh struct super_block *sb = mdsc->fsc->sb; 3156a8599bd8SSage Weil struct inode *inode; 3157be655596SSage Weil struct ceph_inode_info *ci; 3158a8599bd8SSage Weil struct ceph_cap *cap; 3159a8599bd8SSage Weil struct ceph_mds_caps *h; 31604ee6a914SYan, Zheng struct ceph_mds_cap_peer *peer = NULL; 3161982d6011SYan, Zheng struct ceph_snap_realm *realm; 31622600d2ddSSage Weil int mds = session->s_mds; 31632cd698beSYan, Zheng int op, issued; 31643d7ded4dSSage Weil u32 seq, mseq; 3165a8599bd8SSage Weil struct ceph_vino vino; 3166a8599bd8SSage Weil u64 cap_id; 3167a8599bd8SSage Weil u64 size, max_size; 31686df058c0SSage Weil u64 tid; 3169fb01d1f8SYan, Zheng u64 inline_version = 0; 3170fb01d1f8SYan, Zheng void *inline_data = NULL; 3171fb01d1f8SYan, Zheng u32 inline_len = 0; 317270edb55bSSage Weil void *snaptrace; 3173ce1fbc8dSSage Weil size_t snaptrace_len; 3174fb01d1f8SYan, Zheng void *p, *end; 3175a8599bd8SSage Weil 3176a8599bd8SSage Weil dout("handle_caps from mds%d\n", mds); 3177a8599bd8SSage Weil 3178a8599bd8SSage Weil /* decode */ 31794ee6a914SYan, Zheng end = msg->front.iov_base + msg->front.iov_len; 31806df058c0SSage Weil tid = le64_to_cpu(msg->hdr.tid); 3181a8599bd8SSage Weil if (msg->front.iov_len < sizeof(*h)) 3182a8599bd8SSage Weil goto bad; 3183a8599bd8SSage Weil h = msg->front.iov_base; 3184a8599bd8SSage Weil op = le32_to_cpu(h->op); 3185a8599bd8SSage Weil vino.ino = le64_to_cpu(h->ino); 3186a8599bd8SSage Weil vino.snap = CEPH_NOSNAP; 3187a8599bd8SSage Weil cap_id = le64_to_cpu(h->cap_id); 3188a8599bd8SSage Weil seq = le32_to_cpu(h->seq); 31893d7ded4dSSage Weil mseq = le32_to_cpu(h->migrate_seq); 3190a8599bd8SSage Weil size = le64_to_cpu(h->size); 3191a8599bd8SSage Weil max_size = le64_to_cpu(h->max_size); 3192a8599bd8SSage Weil 3193ce1fbc8dSSage Weil snaptrace = h + 1; 3194ce1fbc8dSSage Weil snaptrace_len = le32_to_cpu(h->snap_trace_len); 3195fb01d1f8SYan, Zheng p = snaptrace + snaptrace_len; 3196ce1fbc8dSSage Weil 3197ce1fbc8dSSage Weil if (le16_to_cpu(msg->hdr.version) >= 2) { 3198fb01d1f8SYan, Zheng u32 flock_len; 3199ce1fbc8dSSage Weil ceph_decode_32_safe(&p, end, flock_len, bad); 32004ee6a914SYan, Zheng if (p + flock_len > end) 32014ee6a914SYan, Zheng goto bad; 3202fb01d1f8SYan, Zheng p += flock_len; 3203ce1fbc8dSSage Weil } 3204ce1fbc8dSSage Weil 32054ee6a914SYan, Zheng if (le16_to_cpu(msg->hdr.version) >= 3) { 32064ee6a914SYan, Zheng if (op == CEPH_CAP_OP_IMPORT) { 32074ee6a914SYan, Zheng if (p + sizeof(*peer) > end) 32084ee6a914SYan, Zheng goto bad; 32094ee6a914SYan, Zheng peer = p; 3210fb01d1f8SYan, Zheng p += sizeof(*peer); 321111df2dfbSYan, Zheng } else if (op == CEPH_CAP_OP_EXPORT) { 321211df2dfbSYan, Zheng /* recorded in unused fields */ 321311df2dfbSYan, Zheng peer = (void *)&h->size; 32144ee6a914SYan, Zheng } 32154ee6a914SYan, Zheng } 32164ee6a914SYan, Zheng 3217fb01d1f8SYan, Zheng if (le16_to_cpu(msg->hdr.version) >= 4) { 3218fb01d1f8SYan, Zheng ceph_decode_64_safe(&p, end, inline_version, bad); 3219fb01d1f8SYan, Zheng ceph_decode_32_safe(&p, end, inline_len, bad); 3220fb01d1f8SYan, Zheng if (p + inline_len > end) 3221fb01d1f8SYan, Zheng goto bad; 3222fb01d1f8SYan, Zheng inline_data = p; 3223fb01d1f8SYan, Zheng p += inline_len; 3224fb01d1f8SYan, Zheng } 3225fb01d1f8SYan, Zheng 32266cd3bcadSYan, Zheng /* lookup ino */ 32276cd3bcadSYan, Zheng inode = ceph_find_inode(sb, vino); 32286cd3bcadSYan, Zheng ci = ceph_inode(inode); 32296cd3bcadSYan, Zheng dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, 32306cd3bcadSYan, Zheng vino.snap, inode); 32316cd3bcadSYan, Zheng 3232a8599bd8SSage Weil mutex_lock(&session->s_mutex); 3233a8599bd8SSage Weil session->s_seq++; 3234a8599bd8SSage Weil dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, 3235a8599bd8SSage Weil (unsigned)seq); 3236a8599bd8SSage Weil 323766f58691SYan, Zheng if (op == CEPH_CAP_OP_IMPORT) 323866f58691SYan, Zheng ceph_add_cap_releases(mdsc, session); 323966f58691SYan, Zheng 3240a8599bd8SSage Weil if (!inode) { 3241a8599bd8SSage Weil dout(" i don't have ino %llx\n", vino.ino); 32423d7ded4dSSage Weil 3243a096b09aSYan, Zheng if (op == CEPH_CAP_OP_IMPORT) { 3244a096b09aSYan, Zheng spin_lock(&session->s_cap_lock); 32453d7ded4dSSage Weil __queue_cap_release(session, vino.ino, cap_id, 32463d7ded4dSSage Weil mseq, seq); 3247a096b09aSYan, Zheng spin_unlock(&session->s_cap_lock); 3248a096b09aSYan, Zheng } 324921b559deSGreg Farnum goto flush_cap_releases; 3250a8599bd8SSage Weil } 3251a8599bd8SSage Weil 3252a8599bd8SSage Weil /* these will work even if we don't have a cap yet */ 3253a8599bd8SSage Weil switch (op) { 3254a8599bd8SSage Weil case CEPH_CAP_OP_FLUSHSNAP_ACK: 32556df058c0SSage Weil handle_cap_flushsnap_ack(inode, tid, h, session); 3256a8599bd8SSage Weil goto done; 3257a8599bd8SSage Weil 3258a8599bd8SSage Weil case CEPH_CAP_OP_EXPORT: 325911df2dfbSYan, Zheng handle_cap_export(inode, h, peer, session); 326011df2dfbSYan, Zheng goto done_unlocked; 3261a8599bd8SSage Weil 3262a8599bd8SSage Weil case CEPH_CAP_OP_IMPORT: 3263982d6011SYan, Zheng realm = NULL; 3264982d6011SYan, Zheng if (snaptrace_len) { 3265982d6011SYan, Zheng down_write(&mdsc->snap_rwsem); 3266982d6011SYan, Zheng ceph_update_snap_trace(mdsc, snaptrace, 3267982d6011SYan, Zheng snaptrace + snaptrace_len, 3268982d6011SYan, Zheng false, &realm); 3269982d6011SYan, Zheng downgrade_write(&mdsc->snap_rwsem); 3270982d6011SYan, Zheng } else { 3271982d6011SYan, Zheng down_read(&mdsc->snap_rwsem); 3272982d6011SYan, Zheng } 32734ee6a914SYan, Zheng handle_cap_import(mdsc, inode, h, peer, session, 32742cd698beSYan, Zheng &cap, &issued); 3275982d6011SYan, Zheng handle_cap_grant(mdsc, inode, h, 3276fb01d1f8SYan, Zheng inline_version, inline_data, inline_len, 32772cd698beSYan, Zheng msg->middle, session, cap, issued); 3278982d6011SYan, Zheng if (realm) 3279982d6011SYan, Zheng ceph_put_snap_realm(mdsc, realm); 32802cd698beSYan, Zheng goto done_unlocked; 3281a8599bd8SSage Weil } 3282a8599bd8SSage Weil 3283a8599bd8SSage Weil /* the rest require a cap */ 3284be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3285a8599bd8SSage Weil cap = __get_cap_for_mds(ceph_inode(inode), mds); 3286a8599bd8SSage Weil if (!cap) { 32879dbd412fSSage Weil dout(" no cap on %p ino %llx.%llx from mds%d\n", 3288a8599bd8SSage Weil inode, ceph_ino(inode), ceph_snap(inode), mds); 3289be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 329021b559deSGreg Farnum goto flush_cap_releases; 3291a8599bd8SSage Weil } 3292a8599bd8SSage Weil 3293be655596SSage Weil /* note that each of these drops i_ceph_lock for us */ 3294a8599bd8SSage Weil switch (op) { 3295a8599bd8SSage Weil case CEPH_CAP_OP_REVOKE: 3296a8599bd8SSage Weil case CEPH_CAP_OP_GRANT: 32972cd698beSYan, Zheng __ceph_caps_issued(ci, &issued); 32982cd698beSYan, Zheng issued |= __ceph_caps_dirty(ci); 3299982d6011SYan, Zheng handle_cap_grant(mdsc, inode, h, 3300fb01d1f8SYan, Zheng inline_version, inline_data, inline_len, 3301fb01d1f8SYan, Zheng msg->middle, session, cap, issued); 330215637c8bSSage Weil goto done_unlocked; 3303a8599bd8SSage Weil 3304a8599bd8SSage Weil case CEPH_CAP_OP_FLUSH_ACK: 33056df058c0SSage Weil handle_cap_flush_ack(inode, tid, h, session, cap); 3306a8599bd8SSage Weil break; 3307a8599bd8SSage Weil 3308a8599bd8SSage Weil case CEPH_CAP_OP_TRUNC: 3309a8599bd8SSage Weil handle_cap_trunc(inode, h, session); 3310a8599bd8SSage Weil break; 3311a8599bd8SSage Weil 3312a8599bd8SSage Weil default: 3313be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3314a8599bd8SSage Weil pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, 3315a8599bd8SSage Weil ceph_cap_op_name(op)); 3316a8599bd8SSage Weil } 3317a8599bd8SSage Weil 331821b559deSGreg Farnum goto done; 331921b559deSGreg Farnum 332021b559deSGreg Farnum flush_cap_releases: 332121b559deSGreg Farnum /* 332221b559deSGreg Farnum * send any full release message to try to move things 332321b559deSGreg Farnum * along for the mds (who clearly thinks we still have this 332421b559deSGreg Farnum * cap). 332521b559deSGreg Farnum */ 332621b559deSGreg Farnum ceph_add_cap_releases(mdsc, session); 332721b559deSGreg Farnum ceph_send_cap_releases(mdsc, session); 332821b559deSGreg Farnum 3329a8599bd8SSage Weil done: 3330a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 333115637c8bSSage Weil done_unlocked: 3332a8599bd8SSage Weil iput(inode); 3333a8599bd8SSage Weil return; 3334a8599bd8SSage Weil 3335a8599bd8SSage Weil bad: 3336a8599bd8SSage Weil pr_err("ceph_handle_caps: corrupt message\n"); 33379ec7cab1SSage Weil ceph_msg_dump(msg); 3338a8599bd8SSage Weil return; 3339a8599bd8SSage Weil } 3340a8599bd8SSage Weil 3341a8599bd8SSage Weil /* 3342a8599bd8SSage Weil * Delayed work handler to process end of delayed cap release LRU list. 3343a8599bd8SSage Weil */ 3344afcdaea3SSage Weil void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) 3345a8599bd8SSage Weil { 3346a8599bd8SSage Weil struct ceph_inode_info *ci; 3347a8599bd8SSage Weil int flags = CHECK_CAPS_NODELAY; 3348a8599bd8SSage Weil 3349a8599bd8SSage Weil dout("check_delayed_caps\n"); 3350a8599bd8SSage Weil while (1) { 3351a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 3352a8599bd8SSage Weil if (list_empty(&mdsc->cap_delay_list)) 3353a8599bd8SSage Weil break; 3354a8599bd8SSage Weil ci = list_first_entry(&mdsc->cap_delay_list, 3355a8599bd8SSage Weil struct ceph_inode_info, 3356a8599bd8SSage Weil i_cap_delay_list); 3357a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && 3358a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) 3359a8599bd8SSage Weil break; 3360a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 3361a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 3362a8599bd8SSage Weil dout("check_delayed_caps on %p\n", &ci->vfs_inode); 3363a8599bd8SSage Weil ceph_check_caps(ci, flags, NULL); 3364a8599bd8SSage Weil } 3365a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 3366a8599bd8SSage Weil } 3367a8599bd8SSage Weil 3368a8599bd8SSage Weil /* 3369afcdaea3SSage Weil * Flush all dirty caps to the mds 3370afcdaea3SSage Weil */ 3371afcdaea3SSage Weil void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) 3372afcdaea3SSage Weil { 3373db354052SSage Weil struct ceph_inode_info *ci; 3374db354052SSage Weil struct inode *inode; 3375afcdaea3SSage Weil 3376afcdaea3SSage Weil dout("flush_dirty_caps\n"); 3377afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 3378db354052SSage Weil while (!list_empty(&mdsc->cap_dirty)) { 3379db354052SSage Weil ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info, 3380afcdaea3SSage Weil i_dirty_item); 338170b666c3SSage Weil inode = &ci->vfs_inode; 338270b666c3SSage Weil ihold(inode); 3383db354052SSage Weil dout("flush_dirty_caps %p\n", inode); 3384afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 338570b666c3SSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL); 3386afcdaea3SSage Weil iput(inode); 3387afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 3388afcdaea3SSage Weil } 3389afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 3390db354052SSage Weil dout("flush_dirty_caps done\n"); 3391afcdaea3SSage Weil } 3392afcdaea3SSage Weil 3393afcdaea3SSage Weil /* 3394a8599bd8SSage Weil * Drop open file reference. If we were the last open file, 3395a8599bd8SSage Weil * we may need to release capabilities to the MDS (or schedule 3396a8599bd8SSage Weil * their delayed release). 3397a8599bd8SSage Weil */ 3398a8599bd8SSage Weil void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) 3399a8599bd8SSage Weil { 3400a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 3401a8599bd8SSage Weil int last = 0; 3402a8599bd8SSage Weil 3403be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3404a8599bd8SSage Weil dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode, 3405a8599bd8SSage Weil ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1); 3406a8599bd8SSage Weil BUG_ON(ci->i_nr_by_mode[fmode] == 0); 3407a8599bd8SSage Weil if (--ci->i_nr_by_mode[fmode] == 0) 3408a8599bd8SSage Weil last++; 3409be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3410a8599bd8SSage Weil 3411a8599bd8SSage Weil if (last && ci->i_vino.snap == CEPH_NOSNAP) 3412a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 3413a8599bd8SSage Weil } 3414a8599bd8SSage Weil 3415a8599bd8SSage Weil /* 3416a8599bd8SSage Weil * Helpers for embedding cap and dentry lease releases into mds 3417a8599bd8SSage Weil * requests. 3418a8599bd8SSage Weil * 3419a8599bd8SSage Weil * @force is used by dentry_release (below) to force inclusion of a 3420a8599bd8SSage Weil * record for the directory inode, even when there aren't any caps to 3421a8599bd8SSage Weil * drop. 3422a8599bd8SSage Weil */ 3423a8599bd8SSage Weil int ceph_encode_inode_release(void **p, struct inode *inode, 3424a8599bd8SSage Weil int mds, int drop, int unless, int force) 3425a8599bd8SSage Weil { 3426a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 3427a8599bd8SSage Weil struct ceph_cap *cap; 3428a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3429ec97f88bSSage Weil int used, dirty; 3430a8599bd8SSage Weil int ret = 0; 3431a8599bd8SSage Weil 3432be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3433916623daSSage Weil used = __ceph_caps_used(ci); 3434ec97f88bSSage Weil dirty = __ceph_caps_dirty(ci); 3435916623daSSage Weil 3436ec97f88bSSage Weil dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", 3437ec97f88bSSage Weil inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), 3438916623daSSage Weil ceph_cap_string(unless)); 3439916623daSSage Weil 3440ec97f88bSSage Weil /* only drop unused, clean caps */ 3441ec97f88bSSage Weil drop &= ~(used | dirty); 3442916623daSSage Weil 3443a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 3444a8599bd8SSage Weil if (cap && __cap_is_valid(cap)) { 3445a8599bd8SSage Weil if (force || 3446a8599bd8SSage Weil ((cap->issued & drop) && 3447a8599bd8SSage Weil (cap->issued & unless) == 0)) { 3448a8599bd8SSage Weil if ((cap->issued & drop) && 3449a8599bd8SSage Weil (cap->issued & unless) == 0) { 3450bb137f84SYan, Zheng int wanted = __ceph_caps_wanted(ci); 3451bb137f84SYan, Zheng if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0) 3452bb137f84SYan, Zheng wanted |= cap->mds_wanted; 3453bb137f84SYan, Zheng dout("encode_inode_release %p cap %p " 3454bb137f84SYan, Zheng "%s -> %s, wanted %s -> %s\n", inode, cap, 3455a8599bd8SSage Weil ceph_cap_string(cap->issued), 3456bb137f84SYan, Zheng ceph_cap_string(cap->issued & ~drop), 3457bb137f84SYan, Zheng ceph_cap_string(cap->mds_wanted), 3458bb137f84SYan, Zheng ceph_cap_string(wanted)); 3459bb137f84SYan, Zheng 3460a8599bd8SSage Weil cap->issued &= ~drop; 3461a8599bd8SSage Weil cap->implemented &= ~drop; 3462bb137f84SYan, Zheng cap->mds_wanted = wanted; 3463a8599bd8SSage Weil } else { 3464a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s" 3465a8599bd8SSage Weil " (force)\n", inode, cap, 3466a8599bd8SSage Weil ceph_cap_string(cap->issued)); 3467a8599bd8SSage Weil } 3468a8599bd8SSage Weil 3469a8599bd8SSage Weil rel->ino = cpu_to_le64(ceph_ino(inode)); 3470a8599bd8SSage Weil rel->cap_id = cpu_to_le64(cap->cap_id); 3471a8599bd8SSage Weil rel->seq = cpu_to_le32(cap->seq); 347208a0f24eSHimangi Saraogi rel->issue_seq = cpu_to_le32(cap->issue_seq); 3473a8599bd8SSage Weil rel->mseq = cpu_to_le32(cap->mseq); 3474fd7b95cdSYan, Zheng rel->caps = cpu_to_le32(cap->implemented); 3475a8599bd8SSage Weil rel->wanted = cpu_to_le32(cap->mds_wanted); 3476a8599bd8SSage Weil rel->dname_len = 0; 3477a8599bd8SSage Weil rel->dname_seq = 0; 3478a8599bd8SSage Weil *p += sizeof(*rel); 3479a8599bd8SSage Weil ret = 1; 3480a8599bd8SSage Weil } else { 3481a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s\n", 3482a8599bd8SSage Weil inode, cap, ceph_cap_string(cap->issued)); 3483a8599bd8SSage Weil } 3484a8599bd8SSage Weil } 3485be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3486a8599bd8SSage Weil return ret; 3487a8599bd8SSage Weil } 3488a8599bd8SSage Weil 3489a8599bd8SSage Weil int ceph_encode_dentry_release(void **p, struct dentry *dentry, 3490a8599bd8SSage Weil int mds, int drop, int unless) 3491a8599bd8SSage Weil { 34922b0143b5SDavid Howells struct inode *dir = d_inode(dentry->d_parent); 3493a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3494a8599bd8SSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 3495a8599bd8SSage Weil int force = 0; 3496a8599bd8SSage Weil int ret; 3497a8599bd8SSage Weil 3498a8599bd8SSage Weil /* 3499a8599bd8SSage Weil * force an record for the directory caps if we have a dentry lease. 3500be655596SSage Weil * this is racy (can't take i_ceph_lock and d_lock together), but it 3501a8599bd8SSage Weil * doesn't have to be perfect; the mds will revoke anything we don't 3502a8599bd8SSage Weil * release. 3503a8599bd8SSage Weil */ 3504a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3505a8599bd8SSage Weil if (di->lease_session && di->lease_session->s_mds == mds) 3506a8599bd8SSage Weil force = 1; 3507a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3508a8599bd8SSage Weil 3509a8599bd8SSage Weil ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); 3510a8599bd8SSage Weil 3511a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3512a8599bd8SSage Weil if (ret && di->lease_session && di->lease_session->s_mds == mds) { 3513a8599bd8SSage Weil dout("encode_dentry_release %p mds%d seq %d\n", 3514a8599bd8SSage Weil dentry, mds, (int)di->lease_seq); 3515a8599bd8SSage Weil rel->dname_len = cpu_to_le32(dentry->d_name.len); 3516a8599bd8SSage Weil memcpy(*p, dentry->d_name.name, dentry->d_name.len); 3517a8599bd8SSage Weil *p += dentry->d_name.len; 3518a8599bd8SSage Weil rel->dname_seq = cpu_to_le32(di->lease_seq); 35191dadcce3SSage Weil __ceph_mdsc_drop_dentry_lease(dentry); 3520a8599bd8SSage Weil } 3521a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3522a8599bd8SSage Weil return ret; 3523a8599bd8SSage Weil } 3524