1a8599bd8SSage Weil #include "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" 12a8599bd8SSage Weil #include "decode.h" 13a8599bd8SSage Weil #include "messenger.h" 14a8599bd8SSage Weil 15a8599bd8SSage Weil /* 16a8599bd8SSage Weil * Capability management 17a8599bd8SSage Weil * 18a8599bd8SSage Weil * The Ceph metadata servers control client access to inode metadata 19a8599bd8SSage Weil * and file data by issuing capabilities, granting clients permission 20a8599bd8SSage Weil * to read and/or write both inode field and file data to OSDs 21a8599bd8SSage Weil * (storage nodes). Each capability consists of a set of bits 22a8599bd8SSage Weil * indicating which operations are allowed. 23a8599bd8SSage Weil * 24a8599bd8SSage Weil * If the client holds a *_SHARED cap, the client has a coherent value 25a8599bd8SSage Weil * that can be safely read from the cached inode. 26a8599bd8SSage Weil * 27a8599bd8SSage Weil * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the 28a8599bd8SSage Weil * client is allowed to change inode attributes (e.g., file size, 29a8599bd8SSage Weil * mtime), note its dirty state in the ceph_cap, and asynchronously 30a8599bd8SSage Weil * flush that metadata change to the MDS. 31a8599bd8SSage Weil * 32a8599bd8SSage Weil * In the event of a conflicting operation (perhaps by another 33a8599bd8SSage Weil * client), the MDS will revoke the conflicting client capabilities. 34a8599bd8SSage Weil * 35a8599bd8SSage Weil * In order for a client to cache an inode, it must hold a capability 36a8599bd8SSage Weil * with at least one MDS server. When inodes are released, release 37a8599bd8SSage Weil * notifications are batched and periodically sent en masse to the MDS 38a8599bd8SSage Weil * cluster to release server state. 39a8599bd8SSage Weil */ 40a8599bd8SSage Weil 41a8599bd8SSage Weil 42a8599bd8SSage Weil /* 43a8599bd8SSage Weil * Generate readable cap strings for debugging output. 44a8599bd8SSage Weil */ 45a8599bd8SSage Weil #define MAX_CAP_STR 20 46a8599bd8SSage Weil static char cap_str[MAX_CAP_STR][40]; 47a8599bd8SSage Weil static DEFINE_SPINLOCK(cap_str_lock); 48a8599bd8SSage Weil static int last_cap_str; 49a8599bd8SSage Weil 50a8599bd8SSage Weil static char *gcap_string(char *s, int c) 51a8599bd8SSage Weil { 52a8599bd8SSage Weil if (c & CEPH_CAP_GSHARED) 53a8599bd8SSage Weil *s++ = 's'; 54a8599bd8SSage Weil if (c & CEPH_CAP_GEXCL) 55a8599bd8SSage Weil *s++ = 'x'; 56a8599bd8SSage Weil if (c & CEPH_CAP_GCACHE) 57a8599bd8SSage Weil *s++ = 'c'; 58a8599bd8SSage Weil if (c & CEPH_CAP_GRD) 59a8599bd8SSage Weil *s++ = 'r'; 60a8599bd8SSage Weil if (c & CEPH_CAP_GWR) 61a8599bd8SSage Weil *s++ = 'w'; 62a8599bd8SSage Weil if (c & CEPH_CAP_GBUFFER) 63a8599bd8SSage Weil *s++ = 'b'; 64a8599bd8SSage Weil if (c & CEPH_CAP_GLAZYIO) 65a8599bd8SSage Weil *s++ = 'l'; 66a8599bd8SSage Weil return s; 67a8599bd8SSage Weil } 68a8599bd8SSage Weil 69a8599bd8SSage Weil const char *ceph_cap_string(int caps) 70a8599bd8SSage Weil { 71a8599bd8SSage Weil int i; 72a8599bd8SSage Weil char *s; 73a8599bd8SSage Weil int c; 74a8599bd8SSage Weil 75a8599bd8SSage Weil spin_lock(&cap_str_lock); 76a8599bd8SSage Weil i = last_cap_str++; 77a8599bd8SSage Weil if (last_cap_str == MAX_CAP_STR) 78a8599bd8SSage Weil last_cap_str = 0; 79a8599bd8SSage Weil spin_unlock(&cap_str_lock); 80a8599bd8SSage Weil 81a8599bd8SSage Weil s = cap_str[i]; 82a8599bd8SSage Weil 83a8599bd8SSage Weil if (caps & CEPH_CAP_PIN) 84a8599bd8SSage Weil *s++ = 'p'; 85a8599bd8SSage Weil 86a8599bd8SSage Weil c = (caps >> CEPH_CAP_SAUTH) & 3; 87a8599bd8SSage Weil if (c) { 88a8599bd8SSage Weil *s++ = 'A'; 89a8599bd8SSage Weil s = gcap_string(s, c); 90a8599bd8SSage Weil } 91a8599bd8SSage Weil 92a8599bd8SSage Weil c = (caps >> CEPH_CAP_SLINK) & 3; 93a8599bd8SSage Weil if (c) { 94a8599bd8SSage Weil *s++ = 'L'; 95a8599bd8SSage Weil s = gcap_string(s, c); 96a8599bd8SSage Weil } 97a8599bd8SSage Weil 98a8599bd8SSage Weil c = (caps >> CEPH_CAP_SXATTR) & 3; 99a8599bd8SSage Weil if (c) { 100a8599bd8SSage Weil *s++ = 'X'; 101a8599bd8SSage Weil s = gcap_string(s, c); 102a8599bd8SSage Weil } 103a8599bd8SSage Weil 104a8599bd8SSage Weil c = caps >> CEPH_CAP_SFILE; 105a8599bd8SSage Weil if (c) { 106a8599bd8SSage Weil *s++ = 'F'; 107a8599bd8SSage Weil s = gcap_string(s, c); 108a8599bd8SSage Weil } 109a8599bd8SSage Weil 110a8599bd8SSage Weil if (s == cap_str[i]) 111a8599bd8SSage Weil *s++ = '-'; 112a8599bd8SSage Weil *s = 0; 113a8599bd8SSage Weil return cap_str[i]; 114a8599bd8SSage Weil } 115a8599bd8SSage Weil 11637151668SYehuda Sadeh void ceph_caps_init(struct ceph_mds_client *mdsc) 117a8599bd8SSage Weil { 11837151668SYehuda Sadeh INIT_LIST_HEAD(&mdsc->caps_list); 11937151668SYehuda Sadeh spin_lock_init(&mdsc->caps_list_lock); 120a8599bd8SSage Weil } 121a8599bd8SSage Weil 12237151668SYehuda Sadeh void ceph_caps_finalize(struct ceph_mds_client *mdsc) 123a8599bd8SSage Weil { 124a8599bd8SSage Weil struct ceph_cap *cap; 125a8599bd8SSage Weil 12637151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 12737151668SYehuda Sadeh while (!list_empty(&mdsc->caps_list)) { 12837151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, 12937151668SYehuda Sadeh struct ceph_cap, caps_item); 130a8599bd8SSage Weil list_del(&cap->caps_item); 131a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 132a8599bd8SSage Weil } 13337151668SYehuda Sadeh mdsc->caps_total_count = 0; 13437151668SYehuda Sadeh mdsc->caps_avail_count = 0; 13537151668SYehuda Sadeh mdsc->caps_use_count = 0; 13637151668SYehuda Sadeh mdsc->caps_reserve_count = 0; 13737151668SYehuda Sadeh mdsc->caps_min_count = 0; 13837151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 13985ccce43SSage Weil } 14085ccce43SSage Weil 14137151668SYehuda Sadeh void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta) 14285ccce43SSage Weil { 14337151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 14437151668SYehuda Sadeh mdsc->caps_min_count += delta; 14537151668SYehuda Sadeh BUG_ON(mdsc->caps_min_count < 0); 14637151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 147a8599bd8SSage Weil } 148a8599bd8SSage Weil 14937151668SYehuda Sadeh int ceph_reserve_caps(struct ceph_mds_client *mdsc, 15037151668SYehuda Sadeh struct ceph_cap_reservation *ctx, int need) 151a8599bd8SSage Weil { 152a8599bd8SSage Weil int i; 153a8599bd8SSage Weil struct ceph_cap *cap; 154a8599bd8SSage Weil int have; 155a8599bd8SSage Weil int alloc = 0; 156a8599bd8SSage Weil LIST_HEAD(newcaps); 157a8599bd8SSage Weil int ret = 0; 158a8599bd8SSage Weil 159a8599bd8SSage Weil dout("reserve caps ctx=%p need=%d\n", ctx, need); 160a8599bd8SSage Weil 161a8599bd8SSage Weil /* first reserve any caps that are already allocated */ 16237151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 16337151668SYehuda Sadeh if (mdsc->caps_avail_count >= need) 164a8599bd8SSage Weil have = need; 165a8599bd8SSage Weil else 16637151668SYehuda Sadeh have = mdsc->caps_avail_count; 16737151668SYehuda Sadeh mdsc->caps_avail_count -= have; 16837151668SYehuda Sadeh mdsc->caps_reserve_count += have; 16937151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 17037151668SYehuda Sadeh mdsc->caps_reserve_count + 17137151668SYehuda Sadeh mdsc->caps_avail_count); 17237151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 173a8599bd8SSage Weil 174a8599bd8SSage Weil for (i = have; i < need; i++) { 175a8599bd8SSage Weil cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); 176a8599bd8SSage Weil if (!cap) { 177a8599bd8SSage Weil ret = -ENOMEM; 178a8599bd8SSage Weil goto out_alloc_count; 179a8599bd8SSage Weil } 180a8599bd8SSage Weil list_add(&cap->caps_item, &newcaps); 181a8599bd8SSage Weil alloc++; 182a8599bd8SSage Weil } 183a8599bd8SSage Weil BUG_ON(have + alloc != need); 184a8599bd8SSage Weil 18537151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 18637151668SYehuda Sadeh mdsc->caps_total_count += alloc; 18737151668SYehuda Sadeh mdsc->caps_reserve_count += alloc; 18837151668SYehuda Sadeh list_splice(&newcaps, &mdsc->caps_list); 189a8599bd8SSage Weil 19037151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 19137151668SYehuda Sadeh mdsc->caps_reserve_count + 19237151668SYehuda Sadeh mdsc->caps_avail_count); 19337151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 194a8599bd8SSage Weil 195a8599bd8SSage Weil ctx->count = need; 196a8599bd8SSage Weil dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n", 19737151668SYehuda Sadeh ctx, mdsc->caps_total_count, mdsc->caps_use_count, 19837151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 199a8599bd8SSage Weil return 0; 200a8599bd8SSage Weil 201a8599bd8SSage Weil out_alloc_count: 202a8599bd8SSage Weil /* we didn't manage to reserve as much as we needed */ 203a8599bd8SSage Weil pr_warning("reserve caps ctx=%p ENOMEM need=%d got=%d\n", 204a8599bd8SSage Weil ctx, need, have); 205a8599bd8SSage Weil return ret; 206a8599bd8SSage Weil } 207a8599bd8SSage Weil 20837151668SYehuda Sadeh int ceph_unreserve_caps(struct ceph_mds_client *mdsc, 20937151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 210a8599bd8SSage Weil { 211a8599bd8SSage Weil dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count); 212a8599bd8SSage Weil if (ctx->count) { 21337151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 21437151668SYehuda Sadeh BUG_ON(mdsc->caps_reserve_count < ctx->count); 21537151668SYehuda Sadeh mdsc->caps_reserve_count -= ctx->count; 21637151668SYehuda Sadeh mdsc->caps_avail_count += ctx->count; 217a8599bd8SSage Weil ctx->count = 0; 218a8599bd8SSage Weil dout("unreserve caps %d = %d used + %d resv + %d avail\n", 21937151668SYehuda Sadeh mdsc->caps_total_count, mdsc->caps_use_count, 22037151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 22137151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 22237151668SYehuda Sadeh mdsc->caps_reserve_count + 22337151668SYehuda Sadeh mdsc->caps_avail_count); 22437151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 225a8599bd8SSage Weil } 226a8599bd8SSage Weil return 0; 227a8599bd8SSage Weil } 228a8599bd8SSage Weil 22937151668SYehuda Sadeh static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc, 23037151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 231a8599bd8SSage Weil { 232a8599bd8SSage Weil struct ceph_cap *cap = NULL; 233a8599bd8SSage Weil 234a8599bd8SSage Weil /* temporary, until we do something about cap import/export */ 235443b3760SSage Weil if (!ctx) { 236443b3760SSage Weil cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); 237443b3760SSage Weil if (cap) { 23837151668SYehuda Sadeh mdsc->caps_use_count++; 23937151668SYehuda Sadeh mdsc->caps_total_count++; 240443b3760SSage Weil } 241443b3760SSage Weil return cap; 242443b3760SSage Weil } 243a8599bd8SSage Weil 24437151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 245a8599bd8SSage Weil dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n", 24637151668SYehuda Sadeh ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count, 24737151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 248a8599bd8SSage Weil BUG_ON(!ctx->count); 24937151668SYehuda Sadeh BUG_ON(ctx->count > mdsc->caps_reserve_count); 25037151668SYehuda Sadeh BUG_ON(list_empty(&mdsc->caps_list)); 251a8599bd8SSage Weil 252a8599bd8SSage Weil ctx->count--; 25337151668SYehuda Sadeh mdsc->caps_reserve_count--; 25437151668SYehuda Sadeh mdsc->caps_use_count++; 255a8599bd8SSage Weil 25637151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item); 257a8599bd8SSage Weil list_del(&cap->caps_item); 258a8599bd8SSage Weil 25937151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 26037151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 26137151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 262a8599bd8SSage Weil return cap; 263a8599bd8SSage Weil } 264a8599bd8SSage Weil 26537151668SYehuda Sadeh void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap) 266a8599bd8SSage Weil { 26737151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 2687c1332b8SSage Weil dout("put_cap %p %d = %d used + %d resv + %d avail\n", 26937151668SYehuda Sadeh cap, mdsc->caps_total_count, mdsc->caps_use_count, 27037151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 27137151668SYehuda Sadeh mdsc->caps_use_count--; 272a8599bd8SSage Weil /* 27385ccce43SSage Weil * Keep some preallocated caps around (ceph_min_count), to 27485ccce43SSage Weil * avoid lots of free/alloc churn. 275a8599bd8SSage Weil */ 27637151668SYehuda Sadeh if (mdsc->caps_avail_count >= mdsc->caps_reserve_count + 27737151668SYehuda Sadeh mdsc->caps_min_count) { 27837151668SYehuda Sadeh mdsc->caps_total_count--; 279a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 280a8599bd8SSage Weil } else { 28137151668SYehuda Sadeh mdsc->caps_avail_count++; 28237151668SYehuda Sadeh list_add(&cap->caps_item, &mdsc->caps_list); 283a8599bd8SSage Weil } 284a8599bd8SSage Weil 28537151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 28637151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 28737151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 288a8599bd8SSage Weil } 289a8599bd8SSage Weil 290a8599bd8SSage Weil void ceph_reservation_status(struct ceph_client *client, 29185ccce43SSage Weil int *total, int *avail, int *used, int *reserved, 29285ccce43SSage Weil int *min) 293a8599bd8SSage Weil { 29437151668SYehuda Sadeh struct ceph_mds_client *mdsc = &client->mdsc; 29537151668SYehuda Sadeh 296a8599bd8SSage Weil if (total) 29737151668SYehuda Sadeh *total = mdsc->caps_total_count; 298a8599bd8SSage Weil if (avail) 29937151668SYehuda Sadeh *avail = mdsc->caps_avail_count; 300a8599bd8SSage Weil if (used) 30137151668SYehuda Sadeh *used = mdsc->caps_use_count; 302a8599bd8SSage Weil if (reserved) 30337151668SYehuda Sadeh *reserved = mdsc->caps_reserve_count; 30485ccce43SSage Weil if (min) 30537151668SYehuda Sadeh *min = mdsc->caps_min_count; 306a8599bd8SSage Weil } 307a8599bd8SSage Weil 308a8599bd8SSage Weil /* 309a8599bd8SSage Weil * Find ceph_cap for given mds, if any. 310a8599bd8SSage Weil * 311a8599bd8SSage Weil * Called with i_lock held. 312a8599bd8SSage Weil */ 313a8599bd8SSage Weil static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds) 314a8599bd8SSage Weil { 315a8599bd8SSage Weil struct ceph_cap *cap; 316a8599bd8SSage Weil struct rb_node *n = ci->i_caps.rb_node; 317a8599bd8SSage Weil 318a8599bd8SSage Weil while (n) { 319a8599bd8SSage Weil cap = rb_entry(n, struct ceph_cap, ci_node); 320a8599bd8SSage Weil if (mds < cap->mds) 321a8599bd8SSage Weil n = n->rb_left; 322a8599bd8SSage Weil else if (mds > cap->mds) 323a8599bd8SSage Weil n = n->rb_right; 324a8599bd8SSage Weil else 325a8599bd8SSage Weil return cap; 326a8599bd8SSage Weil } 327a8599bd8SSage Weil return NULL; 328a8599bd8SSage Weil } 329a8599bd8SSage Weil 3302bc50259SGreg Farnum struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds) 3312bc50259SGreg Farnum { 3322bc50259SGreg Farnum struct ceph_cap *cap; 3332bc50259SGreg Farnum 3342bc50259SGreg Farnum spin_lock(&ci->vfs_inode.i_lock); 3352bc50259SGreg Farnum cap = __get_cap_for_mds(ci, mds); 3362bc50259SGreg Farnum spin_unlock(&ci->vfs_inode.i_lock); 3372bc50259SGreg Farnum return cap; 3382bc50259SGreg Farnum } 3392bc50259SGreg Farnum 340a8599bd8SSage Weil /* 34133caad32SSage Weil * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1. 342a8599bd8SSage Weil */ 343ca81f3f6SSage Weil static int __ceph_get_cap_mds(struct ceph_inode_info *ci) 344a8599bd8SSage Weil { 345a8599bd8SSage Weil struct ceph_cap *cap; 346a8599bd8SSage Weil int mds = -1; 347a8599bd8SSage Weil struct rb_node *p; 348a8599bd8SSage Weil 34933caad32SSage Weil /* prefer mds with WR|BUFFER|EXCL caps */ 350a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 351a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 352a8599bd8SSage Weil mds = cap->mds; 353a8599bd8SSage Weil if (cap->issued & (CEPH_CAP_FILE_WR | 354a8599bd8SSage Weil CEPH_CAP_FILE_BUFFER | 355a8599bd8SSage Weil CEPH_CAP_FILE_EXCL)) 356a8599bd8SSage Weil break; 357a8599bd8SSage Weil } 358a8599bd8SSage Weil return mds; 359a8599bd8SSage Weil } 360a8599bd8SSage Weil 361a8599bd8SSage Weil int ceph_get_cap_mds(struct inode *inode) 362a8599bd8SSage Weil { 363a8599bd8SSage Weil int mds; 364a8599bd8SSage Weil spin_lock(&inode->i_lock); 365ca81f3f6SSage Weil mds = __ceph_get_cap_mds(ceph_inode(inode)); 366a8599bd8SSage Weil spin_unlock(&inode->i_lock); 367a8599bd8SSage Weil return mds; 368a8599bd8SSage Weil } 369a8599bd8SSage Weil 370a8599bd8SSage Weil /* 371a8599bd8SSage Weil * Called under i_lock. 372a8599bd8SSage Weil */ 373a8599bd8SSage Weil static void __insert_cap_node(struct ceph_inode_info *ci, 374a8599bd8SSage Weil struct ceph_cap *new) 375a8599bd8SSage Weil { 376a8599bd8SSage Weil struct rb_node **p = &ci->i_caps.rb_node; 377a8599bd8SSage Weil struct rb_node *parent = NULL; 378a8599bd8SSage Weil struct ceph_cap *cap = NULL; 379a8599bd8SSage Weil 380a8599bd8SSage Weil while (*p) { 381a8599bd8SSage Weil parent = *p; 382a8599bd8SSage Weil cap = rb_entry(parent, struct ceph_cap, ci_node); 383a8599bd8SSage Weil if (new->mds < cap->mds) 384a8599bd8SSage Weil p = &(*p)->rb_left; 385a8599bd8SSage Weil else if (new->mds > cap->mds) 386a8599bd8SSage Weil p = &(*p)->rb_right; 387a8599bd8SSage Weil else 388a8599bd8SSage Weil BUG(); 389a8599bd8SSage Weil } 390a8599bd8SSage Weil 391a8599bd8SSage Weil rb_link_node(&new->ci_node, parent, p); 392a8599bd8SSage Weil rb_insert_color(&new->ci_node, &ci->i_caps); 393a8599bd8SSage Weil } 394a8599bd8SSage Weil 395a8599bd8SSage Weil /* 396a8599bd8SSage Weil * (re)set cap hold timeouts, which control the delayed release 397a8599bd8SSage Weil * of unused caps back to the MDS. Should be called on cap use. 398a8599bd8SSage Weil */ 399a8599bd8SSage Weil static void __cap_set_timeouts(struct ceph_mds_client *mdsc, 400a8599bd8SSage Weil struct ceph_inode_info *ci) 401a8599bd8SSage Weil { 4026b805185SSage Weil struct ceph_mount_args *ma = mdsc->client->mount_args; 403a8599bd8SSage Weil 404a8599bd8SSage Weil ci->i_hold_caps_min = round_jiffies(jiffies + 405a8599bd8SSage Weil ma->caps_wanted_delay_min * HZ); 406a8599bd8SSage Weil ci->i_hold_caps_max = round_jiffies(jiffies + 407a8599bd8SSage Weil ma->caps_wanted_delay_max * HZ); 408a8599bd8SSage Weil dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode, 409a8599bd8SSage Weil ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies); 410a8599bd8SSage Weil } 411a8599bd8SSage Weil 412a8599bd8SSage Weil /* 413a8599bd8SSage Weil * (Re)queue cap at the end of the delayed cap release list. 414a8599bd8SSage Weil * 415a8599bd8SSage Weil * If I_FLUSH is set, leave the inode at the front of the list. 416a8599bd8SSage Weil * 417a8599bd8SSage Weil * Caller holds i_lock 418a8599bd8SSage Weil * -> we take mdsc->cap_delay_lock 419a8599bd8SSage Weil */ 420a8599bd8SSage Weil static void __cap_delay_requeue(struct ceph_mds_client *mdsc, 421a8599bd8SSage Weil struct ceph_inode_info *ci) 422a8599bd8SSage Weil { 423a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 424a8599bd8SSage Weil dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode, 425a8599bd8SSage Weil ci->i_ceph_flags, ci->i_hold_caps_max); 426a8599bd8SSage Weil if (!mdsc->stopping) { 427a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 428a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) { 429a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 430a8599bd8SSage Weil goto no_change; 431a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 432a8599bd8SSage Weil } 433a8599bd8SSage Weil list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 434a8599bd8SSage Weil no_change: 435a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 436a8599bd8SSage Weil } 437a8599bd8SSage Weil } 438a8599bd8SSage Weil 439a8599bd8SSage Weil /* 440a8599bd8SSage Weil * Queue an inode for immediate writeback. Mark inode with I_FLUSH, 441a8599bd8SSage Weil * indicating we should send a cap message to flush dirty metadata 442a8599bd8SSage Weil * asap, and move to the front of the delayed cap list. 443a8599bd8SSage Weil */ 444a8599bd8SSage Weil static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc, 445a8599bd8SSage Weil struct ceph_inode_info *ci) 446a8599bd8SSage Weil { 447a8599bd8SSage Weil dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode); 448a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 449a8599bd8SSage Weil ci->i_ceph_flags |= CEPH_I_FLUSH; 450a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) 451a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 452a8599bd8SSage Weil list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 453a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 454a8599bd8SSage Weil } 455a8599bd8SSage Weil 456a8599bd8SSage Weil /* 457a8599bd8SSage Weil * Cancel delayed work on cap. 458a8599bd8SSage Weil * 459a8599bd8SSage Weil * Caller must hold i_lock. 460a8599bd8SSage Weil */ 461a8599bd8SSage Weil static void __cap_delay_cancel(struct ceph_mds_client *mdsc, 462a8599bd8SSage Weil struct ceph_inode_info *ci) 463a8599bd8SSage Weil { 464a8599bd8SSage Weil dout("__cap_delay_cancel %p\n", &ci->vfs_inode); 465a8599bd8SSage Weil if (list_empty(&ci->i_cap_delay_list)) 466a8599bd8SSage Weil return; 467a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 468a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 469a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 470a8599bd8SSage Weil } 471a8599bd8SSage Weil 472a8599bd8SSage Weil /* 473a8599bd8SSage Weil * Common issue checks for add_cap, handle_cap_grant. 474a8599bd8SSage Weil */ 475a8599bd8SSage Weil static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap, 476a8599bd8SSage Weil unsigned issued) 477a8599bd8SSage Weil { 478a8599bd8SSage Weil unsigned had = __ceph_caps_issued(ci, NULL); 479a8599bd8SSage Weil 480a8599bd8SSage Weil /* 481a8599bd8SSage Weil * Each time we receive FILE_CACHE anew, we increment 482a8599bd8SSage Weil * i_rdcache_gen. 483a8599bd8SSage Weil */ 4842962507cSSage Weil if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && 4852962507cSSage Weil (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) 486a8599bd8SSage Weil ci->i_rdcache_gen++; 487a8599bd8SSage Weil 488a8599bd8SSage Weil /* 489a8599bd8SSage Weil * if we are newly issued FILE_SHARED, clear I_COMPLETE; we 490a8599bd8SSage Weil * don't know what happened to this directory while we didn't 491a8599bd8SSage Weil * have the cap. 492a8599bd8SSage Weil */ 493a8599bd8SSage Weil if ((issued & CEPH_CAP_FILE_SHARED) && 494a8599bd8SSage Weil (had & CEPH_CAP_FILE_SHARED) == 0) { 495a8599bd8SSage Weil ci->i_shared_gen++; 496a8599bd8SSage Weil if (S_ISDIR(ci->vfs_inode.i_mode)) { 497a8599bd8SSage Weil dout(" marking %p NOT complete\n", &ci->vfs_inode); 498a8599bd8SSage Weil ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 499a8599bd8SSage Weil } 500a8599bd8SSage Weil } 501a8599bd8SSage Weil } 502a8599bd8SSage Weil 503a8599bd8SSage Weil /* 504a8599bd8SSage Weil * Add a capability under the given MDS session. 505a8599bd8SSage Weil * 506a8599bd8SSage Weil * Caller should hold session snap_rwsem (read) and s_mutex. 507a8599bd8SSage Weil * 508a8599bd8SSage Weil * @fmode is the open file mode, if we are opening a file, otherwise 509a8599bd8SSage Weil * it is < 0. (This is so we can atomically add the cap and add an 510a8599bd8SSage Weil * open file reference to it.) 511a8599bd8SSage Weil */ 512a8599bd8SSage Weil int ceph_add_cap(struct inode *inode, 513a8599bd8SSage Weil struct ceph_mds_session *session, u64 cap_id, 514a8599bd8SSage Weil int fmode, unsigned issued, unsigned wanted, 515a8599bd8SSage Weil unsigned seq, unsigned mseq, u64 realmino, int flags, 516a8599bd8SSage Weil struct ceph_cap_reservation *caps_reservation) 517a8599bd8SSage Weil { 518a8599bd8SSage Weil struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc; 519a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 520a8599bd8SSage Weil struct ceph_cap *new_cap = NULL; 521a8599bd8SSage Weil struct ceph_cap *cap; 522a8599bd8SSage Weil int mds = session->s_mds; 523a8599bd8SSage Weil int actual_wanted; 524a8599bd8SSage Weil 525a8599bd8SSage Weil dout("add_cap %p mds%d cap %llx %s seq %d\n", inode, 526a8599bd8SSage Weil session->s_mds, cap_id, ceph_cap_string(issued), seq); 527a8599bd8SSage Weil 528a8599bd8SSage Weil /* 529a8599bd8SSage Weil * If we are opening the file, include file mode wanted bits 530a8599bd8SSage Weil * in wanted. 531a8599bd8SSage Weil */ 532a8599bd8SSage Weil if (fmode >= 0) 533a8599bd8SSage Weil wanted |= ceph_caps_for_mode(fmode); 534a8599bd8SSage Weil 535a8599bd8SSage Weil retry: 536a8599bd8SSage Weil spin_lock(&inode->i_lock); 537a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 538a8599bd8SSage Weil if (!cap) { 539a8599bd8SSage Weil if (new_cap) { 540a8599bd8SSage Weil cap = new_cap; 541a8599bd8SSage Weil new_cap = NULL; 542a8599bd8SSage Weil } else { 543a8599bd8SSage Weil spin_unlock(&inode->i_lock); 54437151668SYehuda Sadeh new_cap = get_cap(mdsc, caps_reservation); 545a8599bd8SSage Weil if (new_cap == NULL) 546a8599bd8SSage Weil return -ENOMEM; 547a8599bd8SSage Weil goto retry; 548a8599bd8SSage Weil } 549a8599bd8SSage Weil 550a8599bd8SSage Weil cap->issued = 0; 551a8599bd8SSage Weil cap->implemented = 0; 552a8599bd8SSage Weil cap->mds = mds; 553a8599bd8SSage Weil cap->mds_wanted = 0; 554a8599bd8SSage Weil 555a8599bd8SSage Weil cap->ci = ci; 556a8599bd8SSage Weil __insert_cap_node(ci, cap); 557a8599bd8SSage Weil 558a8599bd8SSage Weil /* clear out old exporting info? (i.e. on cap import) */ 559a8599bd8SSage Weil if (ci->i_cap_exporting_mds == mds) { 560a8599bd8SSage Weil ci->i_cap_exporting_issued = 0; 561a8599bd8SSage Weil ci->i_cap_exporting_mseq = 0; 562a8599bd8SSage Weil ci->i_cap_exporting_mds = -1; 563a8599bd8SSage Weil } 564a8599bd8SSage Weil 565a8599bd8SSage Weil /* add to session cap list */ 566a8599bd8SSage Weil cap->session = session; 567a8599bd8SSage Weil spin_lock(&session->s_cap_lock); 568a8599bd8SSage Weil list_add_tail(&cap->session_caps, &session->s_caps); 569a8599bd8SSage Weil session->s_nr_caps++; 570a8599bd8SSage Weil spin_unlock(&session->s_cap_lock); 571a8599bd8SSage Weil } 572a8599bd8SSage Weil 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 ceph_get_snap_realm(mdsc, realm); 581a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 582a8599bd8SSage Weil ci->i_snap_realm = realm; 583a8599bd8SSage Weil list_add(&ci->i_snap_realm_item, 584a8599bd8SSage Weil &realm->inodes_with_caps); 585a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 586a8599bd8SSage Weil } else { 587a8599bd8SSage Weil pr_err("ceph_add_cap: couldn't find snap realm %llx\n", 588a8599bd8SSage Weil realmino); 589b8cd07e7SSage Weil WARN_ON(!realm); 590a8599bd8SSage Weil } 591a8599bd8SSage Weil } 592a8599bd8SSage Weil 593a8599bd8SSage Weil __check_cap_issue(ci, cap, issued); 594a8599bd8SSage Weil 595a8599bd8SSage Weil /* 596a8599bd8SSage Weil * If we are issued caps we don't want, or the mds' wanted 597a8599bd8SSage Weil * value appears to be off, queue a check so we'll release 598a8599bd8SSage Weil * later and/or update the mds wanted value. 599a8599bd8SSage Weil */ 600a8599bd8SSage Weil actual_wanted = __ceph_caps_wanted(ci); 601a8599bd8SSage Weil if ((wanted & ~actual_wanted) || 602a8599bd8SSage Weil (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) { 603a8599bd8SSage Weil dout(" issued %s, mds wanted %s, actual %s, queueing\n", 604a8599bd8SSage Weil ceph_cap_string(issued), ceph_cap_string(wanted), 605a8599bd8SSage Weil ceph_cap_string(actual_wanted)); 606a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 607a8599bd8SSage Weil } 608a8599bd8SSage Weil 609a8599bd8SSage Weil if (flags & CEPH_CAP_FLAG_AUTH) 610a8599bd8SSage Weil ci->i_auth_cap = cap; 611a8599bd8SSage Weil else if (ci->i_auth_cap == cap) 612a8599bd8SSage Weil ci->i_auth_cap = NULL; 613a8599bd8SSage Weil 614a8599bd8SSage Weil dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n", 615a8599bd8SSage Weil inode, ceph_vinop(inode), cap, ceph_cap_string(issued), 616a8599bd8SSage Weil ceph_cap_string(issued|cap->issued), seq, mds); 617a8599bd8SSage Weil cap->cap_id = cap_id; 618a8599bd8SSage Weil cap->issued = issued; 619a8599bd8SSage Weil cap->implemented |= issued; 620a8599bd8SSage Weil cap->mds_wanted |= wanted; 621a8599bd8SSage Weil cap->seq = seq; 622a8599bd8SSage Weil cap->issue_seq = seq; 623a8599bd8SSage Weil cap->mseq = mseq; 624685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 625a8599bd8SSage Weil 626a8599bd8SSage Weil if (fmode >= 0) 627a8599bd8SSage Weil __ceph_get_fmode(ci, fmode); 628a8599bd8SSage Weil spin_unlock(&inode->i_lock); 62903066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 630a8599bd8SSage Weil return 0; 631a8599bd8SSage Weil } 632a8599bd8SSage Weil 633a8599bd8SSage Weil /* 634a8599bd8SSage Weil * Return true if cap has not timed out and belongs to the current 635a8599bd8SSage Weil * generation of the MDS session (i.e. has not gone 'stale' due to 636a8599bd8SSage Weil * us losing touch with the mds). 637a8599bd8SSage Weil */ 638a8599bd8SSage Weil static int __cap_is_valid(struct ceph_cap *cap) 639a8599bd8SSage Weil { 640a8599bd8SSage Weil unsigned long ttl; 641cdac8303SSage Weil u32 gen; 642a8599bd8SSage Weil 643a8599bd8SSage Weil spin_lock(&cap->session->s_cap_lock); 644a8599bd8SSage Weil gen = cap->session->s_cap_gen; 645a8599bd8SSage Weil ttl = cap->session->s_cap_ttl; 646a8599bd8SSage Weil spin_unlock(&cap->session->s_cap_lock); 647a8599bd8SSage Weil 648685f9a5dSSage Weil if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) { 649a8599bd8SSage Weil dout("__cap_is_valid %p cap %p issued %s " 650a8599bd8SSage Weil "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode, 651685f9a5dSSage Weil cap, ceph_cap_string(cap->issued), cap->cap_gen, gen); 652a8599bd8SSage Weil return 0; 653a8599bd8SSage Weil } 654a8599bd8SSage Weil 655a8599bd8SSage Weil return 1; 656a8599bd8SSage Weil } 657a8599bd8SSage Weil 658a8599bd8SSage Weil /* 659a8599bd8SSage Weil * Return set of valid cap bits issued to us. Note that caps time 660a8599bd8SSage Weil * out, and may be invalidated in bulk if the client session times out 661a8599bd8SSage Weil * and session->s_cap_gen is bumped. 662a8599bd8SSage Weil */ 663a8599bd8SSage Weil int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented) 664a8599bd8SSage Weil { 6657af8f1e4SSage Weil int have = ci->i_snap_caps | ci->i_cap_exporting_issued; 666a8599bd8SSage Weil struct ceph_cap *cap; 667a8599bd8SSage Weil struct rb_node *p; 668a8599bd8SSage Weil 669a8599bd8SSage Weil if (implemented) 670a8599bd8SSage Weil *implemented = 0; 671a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 672a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 673a8599bd8SSage Weil if (!__cap_is_valid(cap)) 674a8599bd8SSage Weil continue; 675a8599bd8SSage Weil dout("__ceph_caps_issued %p cap %p issued %s\n", 676a8599bd8SSage Weil &ci->vfs_inode, cap, ceph_cap_string(cap->issued)); 677a8599bd8SSage Weil have |= cap->issued; 678a8599bd8SSage Weil if (implemented) 679a8599bd8SSage Weil *implemented |= cap->implemented; 680a8599bd8SSage Weil } 681a8599bd8SSage Weil return have; 682a8599bd8SSage Weil } 683a8599bd8SSage Weil 684a8599bd8SSage Weil /* 685a8599bd8SSage Weil * Get cap bits issued by caps other than @ocap 686a8599bd8SSage Weil */ 687a8599bd8SSage Weil int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap) 688a8599bd8SSage Weil { 689a8599bd8SSage Weil int have = ci->i_snap_caps; 690a8599bd8SSage Weil struct ceph_cap *cap; 691a8599bd8SSage Weil struct rb_node *p; 692a8599bd8SSage Weil 693a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 694a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 695a8599bd8SSage Weil if (cap == ocap) 696a8599bd8SSage Weil continue; 697a8599bd8SSage Weil if (!__cap_is_valid(cap)) 698a8599bd8SSage Weil continue; 699a8599bd8SSage Weil have |= cap->issued; 700a8599bd8SSage Weil } 701a8599bd8SSage Weil return have; 702a8599bd8SSage Weil } 703a8599bd8SSage Weil 704a8599bd8SSage Weil /* 705a8599bd8SSage Weil * Move a cap to the end of the LRU (oldest caps at list head, newest 706a8599bd8SSage Weil * at list tail). 707a8599bd8SSage Weil */ 708a8599bd8SSage Weil static void __touch_cap(struct ceph_cap *cap) 709a8599bd8SSage Weil { 710a8599bd8SSage Weil struct ceph_mds_session *s = cap->session; 711a8599bd8SSage Weil 7125dacf091SSage Weil spin_lock(&s->s_cap_lock); 7137c1332b8SSage Weil if (s->s_cap_iterator == NULL) { 714a8599bd8SSage Weil dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap, 715a8599bd8SSage Weil s->s_mds); 716a8599bd8SSage Weil list_move_tail(&cap->session_caps, &s->s_caps); 7175dacf091SSage Weil } else { 7185dacf091SSage Weil dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n", 7195dacf091SSage Weil &cap->ci->vfs_inode, cap, s->s_mds); 7205dacf091SSage Weil } 721a8599bd8SSage Weil spin_unlock(&s->s_cap_lock); 722a8599bd8SSage Weil } 723a8599bd8SSage Weil 724a8599bd8SSage Weil /* 725a8599bd8SSage Weil * Check if we hold the given mask. If so, move the cap(s) to the 726a8599bd8SSage Weil * front of their respective LRUs. (This is the preferred way for 727a8599bd8SSage Weil * callers to check for caps they want.) 728a8599bd8SSage Weil */ 729a8599bd8SSage Weil int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) 730a8599bd8SSage Weil { 731a8599bd8SSage Weil struct ceph_cap *cap; 732a8599bd8SSage Weil struct rb_node *p; 733a8599bd8SSage Weil int have = ci->i_snap_caps; 734a8599bd8SSage Weil 735a8599bd8SSage Weil if ((have & mask) == mask) { 736a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p snap issued %s" 737a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 738a8599bd8SSage Weil ceph_cap_string(have), 739a8599bd8SSage Weil ceph_cap_string(mask)); 740a8599bd8SSage Weil return 1; 741a8599bd8SSage Weil } 742a8599bd8SSage Weil 743a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 744a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 745a8599bd8SSage Weil if (!__cap_is_valid(cap)) 746a8599bd8SSage Weil continue; 747a8599bd8SSage Weil if ((cap->issued & mask) == mask) { 748a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p cap %p issued %s" 749a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, cap, 750a8599bd8SSage Weil ceph_cap_string(cap->issued), 751a8599bd8SSage Weil ceph_cap_string(mask)); 752a8599bd8SSage Weil if (touch) 753a8599bd8SSage Weil __touch_cap(cap); 754a8599bd8SSage Weil return 1; 755a8599bd8SSage Weil } 756a8599bd8SSage Weil 757a8599bd8SSage Weil /* does a combination of caps satisfy mask? */ 758a8599bd8SSage Weil have |= cap->issued; 759a8599bd8SSage Weil if ((have & mask) == mask) { 760a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p combo issued %s" 761a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 762a8599bd8SSage Weil ceph_cap_string(cap->issued), 763a8599bd8SSage Weil ceph_cap_string(mask)); 764a8599bd8SSage Weil if (touch) { 765a8599bd8SSage Weil struct rb_node *q; 766a8599bd8SSage Weil 767a8599bd8SSage Weil /* touch this + preceeding caps */ 768a8599bd8SSage Weil __touch_cap(cap); 769a8599bd8SSage Weil for (q = rb_first(&ci->i_caps); q != p; 770a8599bd8SSage Weil q = rb_next(q)) { 771a8599bd8SSage Weil cap = rb_entry(q, struct ceph_cap, 772a8599bd8SSage Weil ci_node); 773a8599bd8SSage Weil if (!__cap_is_valid(cap)) 774a8599bd8SSage Weil continue; 775a8599bd8SSage Weil __touch_cap(cap); 776a8599bd8SSage Weil } 777a8599bd8SSage Weil } 778a8599bd8SSage Weil return 1; 779a8599bd8SSage Weil } 780a8599bd8SSage Weil } 781a8599bd8SSage Weil 782a8599bd8SSage Weil return 0; 783a8599bd8SSage Weil } 784a8599bd8SSage Weil 785a8599bd8SSage Weil /* 786a8599bd8SSage Weil * Return true if mask caps are currently being revoked by an MDS. 787a8599bd8SSage Weil */ 788a8599bd8SSage Weil int ceph_caps_revoking(struct ceph_inode_info *ci, int mask) 789a8599bd8SSage Weil { 790a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 791a8599bd8SSage Weil struct ceph_cap *cap; 792a8599bd8SSage Weil struct rb_node *p; 793a8599bd8SSage Weil int ret = 0; 794a8599bd8SSage Weil 795a8599bd8SSage Weil spin_lock(&inode->i_lock); 796a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 797a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 798a8599bd8SSage Weil if (__cap_is_valid(cap) && 799a8599bd8SSage Weil (cap->implemented & ~cap->issued & mask)) { 800a8599bd8SSage Weil ret = 1; 801a8599bd8SSage Weil break; 802a8599bd8SSage Weil } 803a8599bd8SSage Weil } 804a8599bd8SSage Weil spin_unlock(&inode->i_lock); 805a8599bd8SSage Weil dout("ceph_caps_revoking %p %s = %d\n", inode, 806a8599bd8SSage Weil ceph_cap_string(mask), ret); 807a8599bd8SSage Weil return ret; 808a8599bd8SSage Weil } 809a8599bd8SSage Weil 810a8599bd8SSage Weil int __ceph_caps_used(struct ceph_inode_info *ci) 811a8599bd8SSage Weil { 812a8599bd8SSage Weil int used = 0; 813a8599bd8SSage Weil if (ci->i_pin_ref) 814a8599bd8SSage Weil used |= CEPH_CAP_PIN; 815a8599bd8SSage Weil if (ci->i_rd_ref) 816a8599bd8SSage Weil used |= CEPH_CAP_FILE_RD; 817a43fb731SSage Weil if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages) 818a8599bd8SSage Weil used |= CEPH_CAP_FILE_CACHE; 819a8599bd8SSage Weil if (ci->i_wr_ref) 820a8599bd8SSage Weil used |= CEPH_CAP_FILE_WR; 821a8599bd8SSage Weil if (ci->i_wrbuffer_ref) 822a8599bd8SSage Weil used |= CEPH_CAP_FILE_BUFFER; 823a8599bd8SSage Weil return used; 824a8599bd8SSage Weil } 825a8599bd8SSage Weil 826a8599bd8SSage Weil /* 827a8599bd8SSage Weil * wanted, by virtue of open file modes 828a8599bd8SSage Weil */ 829a8599bd8SSage Weil int __ceph_caps_file_wanted(struct ceph_inode_info *ci) 830a8599bd8SSage Weil { 831a8599bd8SSage Weil int want = 0; 832a8599bd8SSage Weil int mode; 83333caad32SSage Weil for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++) 834a8599bd8SSage Weil if (ci->i_nr_by_mode[mode]) 835a8599bd8SSage Weil want |= ceph_caps_for_mode(mode); 836a8599bd8SSage Weil return want; 837a8599bd8SSage Weil } 838a8599bd8SSage Weil 839a8599bd8SSage Weil /* 840a8599bd8SSage Weil * Return caps we have registered with the MDS(s) as 'wanted'. 841a8599bd8SSage Weil */ 842a8599bd8SSage Weil int __ceph_caps_mds_wanted(struct ceph_inode_info *ci) 843a8599bd8SSage Weil { 844a8599bd8SSage Weil struct ceph_cap *cap; 845a8599bd8SSage Weil struct rb_node *p; 846a8599bd8SSage Weil int mds_wanted = 0; 847a8599bd8SSage Weil 848a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 849a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 850a8599bd8SSage Weil if (!__cap_is_valid(cap)) 851a8599bd8SSage Weil continue; 852a8599bd8SSage Weil mds_wanted |= cap->mds_wanted; 853a8599bd8SSage Weil } 854a8599bd8SSage Weil return mds_wanted; 855a8599bd8SSage Weil } 856a8599bd8SSage Weil 857a8599bd8SSage Weil /* 858a8599bd8SSage Weil * called under i_lock 859a8599bd8SSage Weil */ 860a8599bd8SSage Weil static int __ceph_is_any_caps(struct ceph_inode_info *ci) 861a8599bd8SSage Weil { 862a8599bd8SSage Weil return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0; 863a8599bd8SSage Weil } 864a8599bd8SSage Weil 865a8599bd8SSage Weil /* 866f818a736SSage Weil * Remove a cap. Take steps to deal with a racing iterate_session_caps. 867f818a736SSage Weil * 868a6369741SSage Weil * caller should hold i_lock. 869a6369741SSage Weil * caller will not hold session s_mutex if called from destroy_inode. 870a8599bd8SSage Weil */ 8717c1332b8SSage Weil void __ceph_remove_cap(struct ceph_cap *cap) 872a8599bd8SSage Weil { 873a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 874a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 875640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 876640ef79dSCheng Renquan &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 877f818a736SSage Weil int removed = 0; 878a8599bd8SSage Weil 879a8599bd8SSage Weil dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); 880a8599bd8SSage Weil 8817c1332b8SSage Weil /* remove from session list */ 8827c1332b8SSage Weil spin_lock(&session->s_cap_lock); 8837c1332b8SSage Weil if (session->s_cap_iterator == cap) { 8847c1332b8SSage Weil /* not yet, we are iterating over this very cap */ 8857c1332b8SSage Weil dout("__ceph_remove_cap delaying %p removal from session %p\n", 8867c1332b8SSage Weil cap, cap->session); 8877c1332b8SSage Weil } else { 8887c1332b8SSage Weil list_del_init(&cap->session_caps); 8897c1332b8SSage Weil session->s_nr_caps--; 8907c1332b8SSage Weil cap->session = NULL; 891f818a736SSage Weil removed = 1; 8927c1332b8SSage Weil } 893f818a736SSage Weil /* protect backpointer with s_cap_lock: see iterate_session_caps */ 894f818a736SSage Weil cap->ci = NULL; 8957c1332b8SSage Weil spin_unlock(&session->s_cap_lock); 8967c1332b8SSage Weil 897f818a736SSage Weil /* remove from inode list */ 898f818a736SSage Weil rb_erase(&cap->ci_node, &ci->i_caps); 899f818a736SSage Weil if (ci->i_auth_cap == cap) 900f818a736SSage Weil ci->i_auth_cap = NULL; 901f818a736SSage Weil 902f818a736SSage Weil if (removed) 90337151668SYehuda Sadeh ceph_put_cap(mdsc, cap); 904a8599bd8SSage Weil 905a8599bd8SSage Weil if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) { 906a8599bd8SSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 907a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 908a8599bd8SSage Weil list_del_init(&ci->i_snap_realm_item); 909a8599bd8SSage Weil ci->i_snap_realm_counter++; 910a8599bd8SSage Weil ci->i_snap_realm = NULL; 911a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 912a8599bd8SSage Weil ceph_put_snap_realm(mdsc, realm); 913a8599bd8SSage Weil } 914a8599bd8SSage Weil if (!__ceph_is_any_real_caps(ci)) 915a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 916a8599bd8SSage Weil } 917a8599bd8SSage Weil 918a8599bd8SSage Weil /* 919a8599bd8SSage Weil * Build and send a cap message to the given MDS. 920a8599bd8SSage Weil * 921a8599bd8SSage Weil * Caller should be holding s_mutex. 922a8599bd8SSage Weil */ 923a8599bd8SSage Weil static int send_cap_msg(struct ceph_mds_session *session, 924a8599bd8SSage Weil u64 ino, u64 cid, int op, 925a8599bd8SSage Weil int caps, int wanted, int dirty, 926a8599bd8SSage Weil u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq, 927a8599bd8SSage Weil u64 size, u64 max_size, 928a8599bd8SSage Weil struct timespec *mtime, struct timespec *atime, 929a8599bd8SSage Weil u64 time_warp_seq, 930a8599bd8SSage Weil uid_t uid, gid_t gid, mode_t mode, 931a8599bd8SSage Weil u64 xattr_version, 932a8599bd8SSage Weil struct ceph_buffer *xattrs_buf, 933a8599bd8SSage Weil u64 follows) 934a8599bd8SSage Weil { 935a8599bd8SSage Weil struct ceph_mds_caps *fc; 936a8599bd8SSage Weil struct ceph_msg *msg; 937a8599bd8SSage Weil 938a8599bd8SSage Weil dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" 939a8599bd8SSage Weil " seq %u/%u mseq %u follows %lld size %llu/%llu" 940a8599bd8SSage Weil " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op), 941a8599bd8SSage Weil cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted), 942a8599bd8SSage Weil ceph_cap_string(dirty), 943a8599bd8SSage Weil seq, issue_seq, mseq, follows, size, max_size, 944a8599bd8SSage Weil xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0); 945a8599bd8SSage Weil 94634d23762SYehuda Sadeh msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS); 947a79832f2SSage Weil if (!msg) 948a79832f2SSage Weil return -ENOMEM; 949a8599bd8SSage Weil 9506df058c0SSage Weil msg->hdr.tid = cpu_to_le64(flush_tid); 951a8599bd8SSage Weil 9526df058c0SSage Weil fc = msg->front.iov_base; 953a8599bd8SSage Weil memset(fc, 0, sizeof(*fc)); 954a8599bd8SSage Weil 955a8599bd8SSage Weil fc->cap_id = cpu_to_le64(cid); 956a8599bd8SSage Weil fc->op = cpu_to_le32(op); 957a8599bd8SSage Weil fc->seq = cpu_to_le32(seq); 958a8599bd8SSage Weil fc->issue_seq = cpu_to_le32(issue_seq); 959a8599bd8SSage Weil fc->migrate_seq = cpu_to_le32(mseq); 960a8599bd8SSage Weil fc->caps = cpu_to_le32(caps); 961a8599bd8SSage Weil fc->wanted = cpu_to_le32(wanted); 962a8599bd8SSage Weil fc->dirty = cpu_to_le32(dirty); 963a8599bd8SSage Weil fc->ino = cpu_to_le64(ino); 964a8599bd8SSage Weil fc->snap_follows = cpu_to_le64(follows); 965a8599bd8SSage Weil 966a8599bd8SSage Weil fc->size = cpu_to_le64(size); 967a8599bd8SSage Weil fc->max_size = cpu_to_le64(max_size); 968a8599bd8SSage Weil if (mtime) 969a8599bd8SSage Weil ceph_encode_timespec(&fc->mtime, mtime); 970a8599bd8SSage Weil if (atime) 971a8599bd8SSage Weil ceph_encode_timespec(&fc->atime, atime); 972a8599bd8SSage Weil fc->time_warp_seq = cpu_to_le32(time_warp_seq); 973a8599bd8SSage Weil 974a8599bd8SSage Weil fc->uid = cpu_to_le32(uid); 975a8599bd8SSage Weil fc->gid = cpu_to_le32(gid); 976a8599bd8SSage Weil fc->mode = cpu_to_le32(mode); 977a8599bd8SSage Weil 978a8599bd8SSage Weil fc->xattr_version = cpu_to_le64(xattr_version); 979a8599bd8SSage Weil if (xattrs_buf) { 980a8599bd8SSage Weil msg->middle = ceph_buffer_get(xattrs_buf); 981a8599bd8SSage Weil fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len); 982a8599bd8SSage Weil msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len); 983a8599bd8SSage Weil } 984a8599bd8SSage Weil 985a8599bd8SSage Weil ceph_con_send(&session->s_con, msg); 986a8599bd8SSage Weil return 0; 987a8599bd8SSage Weil } 988a8599bd8SSage Weil 9893d7ded4dSSage Weil static void __queue_cap_release(struct ceph_mds_session *session, 9903d7ded4dSSage Weil u64 ino, u64 cap_id, u32 migrate_seq, 9913d7ded4dSSage Weil u32 issue_seq) 9923d7ded4dSSage Weil { 9933d7ded4dSSage Weil struct ceph_msg *msg; 9943d7ded4dSSage Weil struct ceph_mds_cap_release *head; 9953d7ded4dSSage Weil struct ceph_mds_cap_item *item; 9963d7ded4dSSage Weil 9973d7ded4dSSage Weil spin_lock(&session->s_cap_lock); 9983d7ded4dSSage Weil BUG_ON(!session->s_num_cap_releases); 9993d7ded4dSSage Weil msg = list_first_entry(&session->s_cap_releases, 10003d7ded4dSSage Weil struct ceph_msg, list_head); 10013d7ded4dSSage Weil 10023d7ded4dSSage Weil dout(" adding %llx release to mds%d msg %p (%d left)\n", 10033d7ded4dSSage Weil ino, session->s_mds, msg, session->s_num_cap_releases); 10043d7ded4dSSage Weil 10053d7ded4dSSage Weil BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE); 10063d7ded4dSSage Weil head = msg->front.iov_base; 10073d7ded4dSSage Weil head->num = cpu_to_le32(le32_to_cpu(head->num) + 1); 10083d7ded4dSSage Weil item = msg->front.iov_base + msg->front.iov_len; 10093d7ded4dSSage Weil item->ino = cpu_to_le64(ino); 10103d7ded4dSSage Weil item->cap_id = cpu_to_le64(cap_id); 10113d7ded4dSSage Weil item->migrate_seq = cpu_to_le32(migrate_seq); 10123d7ded4dSSage Weil item->seq = cpu_to_le32(issue_seq); 10133d7ded4dSSage Weil 10143d7ded4dSSage Weil session->s_num_cap_releases--; 10153d7ded4dSSage Weil 10163d7ded4dSSage Weil msg->front.iov_len += sizeof(*item); 10173d7ded4dSSage Weil if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) { 10183d7ded4dSSage Weil dout(" release msg %p full\n", msg); 10193d7ded4dSSage Weil list_move_tail(&msg->list_head, &session->s_cap_releases_done); 10203d7ded4dSSage Weil } else { 10213d7ded4dSSage Weil dout(" release msg %p at %d/%d (%d)\n", msg, 10223d7ded4dSSage Weil (int)le32_to_cpu(head->num), 10233d7ded4dSSage Weil (int)CEPH_CAPS_PER_RELEASE, 10243d7ded4dSSage Weil (int)msg->front.iov_len); 10253d7ded4dSSage Weil } 10263d7ded4dSSage Weil spin_unlock(&session->s_cap_lock); 10273d7ded4dSSage Weil } 10283d7ded4dSSage Weil 1029a8599bd8SSage Weil /* 1030a6369741SSage Weil * Queue cap releases when an inode is dropped from our cache. Since 1031a6369741SSage Weil * inode is about to be destroyed, there is no need for i_lock. 1032a8599bd8SSage Weil */ 1033a8599bd8SSage Weil void ceph_queue_caps_release(struct inode *inode) 1034a8599bd8SSage Weil { 1035a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1036a8599bd8SSage Weil struct rb_node *p; 1037a8599bd8SSage Weil 1038a8599bd8SSage Weil p = rb_first(&ci->i_caps); 1039a8599bd8SSage Weil while (p) { 1040a8599bd8SSage Weil struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); 1041a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 1042a8599bd8SSage Weil 10433d7ded4dSSage Weil __queue_cap_release(session, ceph_ino(inode), cap->cap_id, 10443d7ded4dSSage Weil cap->mseq, cap->issue_seq); 1045a8599bd8SSage Weil p = rb_next(p); 10467c1332b8SSage Weil __ceph_remove_cap(cap); 1047a8599bd8SSage Weil } 1048a8599bd8SSage Weil } 1049a8599bd8SSage Weil 1050a8599bd8SSage Weil /* 1051a8599bd8SSage Weil * Send a cap msg on the given inode. Update our caps state, then 1052a8599bd8SSage Weil * drop i_lock and send the message. 1053a8599bd8SSage Weil * 1054a8599bd8SSage Weil * Make note of max_size reported/requested from mds, revoked caps 1055a8599bd8SSage Weil * that have now been implemented. 1056a8599bd8SSage Weil * 1057a8599bd8SSage Weil * Make half-hearted attempt ot to invalidate page cache if we are 1058a8599bd8SSage Weil * dropping RDCACHE. Note that this will leave behind locked pages 1059a8599bd8SSage Weil * that we'll then need to deal with elsewhere. 1060a8599bd8SSage Weil * 1061a8599bd8SSage Weil * Return non-zero if delayed release, or we experienced an error 1062a8599bd8SSage Weil * such that the caller should requeue + retry later. 1063a8599bd8SSage Weil * 1064a8599bd8SSage Weil * called with i_lock, then drops it. 1065a8599bd8SSage Weil * caller should hold snap_rwsem (read), s_mutex. 1066a8599bd8SSage Weil */ 1067a8599bd8SSage Weil static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, 1068a8599bd8SSage Weil int op, int used, int want, int retain, int flushing, 1069a8599bd8SSage Weil unsigned *pflush_tid) 1070a8599bd8SSage Weil __releases(cap->ci->vfs_inode->i_lock) 1071a8599bd8SSage Weil { 1072a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 1073a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1074a8599bd8SSage Weil u64 cap_id = cap->cap_id; 107568c28323SSage Weil int held, revoking, dropping, keep; 1076a8599bd8SSage Weil u64 seq, issue_seq, mseq, time_warp_seq, follows; 1077a8599bd8SSage Weil u64 size, max_size; 1078a8599bd8SSage Weil struct timespec mtime, atime; 1079a8599bd8SSage Weil int wake = 0; 1080a8599bd8SSage Weil mode_t mode; 1081a8599bd8SSage Weil uid_t uid; 1082a8599bd8SSage Weil gid_t gid; 1083a8599bd8SSage Weil struct ceph_mds_session *session; 1084a8599bd8SSage Weil u64 xattr_version = 0; 1085082afec9SSage Weil struct ceph_buffer *xattr_blob = NULL; 1086a8599bd8SSage Weil int delayed = 0; 1087a8599bd8SSage Weil u64 flush_tid = 0; 1088a8599bd8SSage Weil int i; 1089a8599bd8SSage Weil int ret; 1090a8599bd8SSage Weil 109168c28323SSage Weil held = cap->issued | cap->implemented; 109268c28323SSage Weil revoking = cap->implemented & ~cap->issued; 109368c28323SSage Weil retain &= ~revoking; 109468c28323SSage Weil dropping = cap->issued & ~retain; 109568c28323SSage Weil 1096a8599bd8SSage Weil dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", 1097a8599bd8SSage Weil inode, cap, cap->session, 1098a8599bd8SSage Weil ceph_cap_string(held), ceph_cap_string(held & retain), 1099a8599bd8SSage Weil ceph_cap_string(revoking)); 1100a8599bd8SSage Weil BUG_ON((retain & CEPH_CAP_PIN) == 0); 1101a8599bd8SSage Weil 1102a8599bd8SSage Weil session = cap->session; 1103a8599bd8SSage Weil 1104a8599bd8SSage Weil /* don't release wanted unless we've waited a bit. */ 1105a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1106a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_min)) { 1107a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", 1108a8599bd8SSage Weil ceph_cap_string(cap->issued), 1109a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1110a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1111a8599bd8SSage Weil ceph_cap_string(want)); 1112a8599bd8SSage Weil want |= cap->mds_wanted; 1113a8599bd8SSage Weil retain |= cap->issued; 1114a8599bd8SSage Weil delayed = 1; 1115a8599bd8SSage Weil } 1116a8599bd8SSage Weil ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); 1117a8599bd8SSage Weil 1118a8599bd8SSage Weil cap->issued &= retain; /* drop bits we don't want */ 1119a8599bd8SSage Weil if (cap->implemented & ~cap->issued) { 1120a8599bd8SSage Weil /* 1121a8599bd8SSage Weil * Wake up any waiters on wanted -> needed transition. 1122a8599bd8SSage Weil * This is due to the weird transition from buffered 1123a8599bd8SSage Weil * to sync IO... we need to flush dirty pages _before_ 1124a8599bd8SSage Weil * allowing sync writes to avoid reordering. 1125a8599bd8SSage Weil */ 1126a8599bd8SSage Weil wake = 1; 1127a8599bd8SSage Weil } 1128a8599bd8SSage Weil cap->implemented &= cap->issued | used; 1129a8599bd8SSage Weil cap->mds_wanted = want; 1130a8599bd8SSage Weil 1131a8599bd8SSage Weil if (flushing) { 1132a8599bd8SSage Weil /* 1133a8599bd8SSage Weil * assign a tid for flush operations so we can avoid 1134a8599bd8SSage Weil * flush1 -> dirty1 -> flush2 -> flushack1 -> mark 1135a8599bd8SSage Weil * clean type races. track latest tid for every bit 1136a8599bd8SSage Weil * so we can handle flush AxFw, flush Fw, and have the 1137a8599bd8SSage Weil * first ack clean Ax. 1138a8599bd8SSage Weil */ 1139a8599bd8SSage Weil flush_tid = ++ci->i_cap_flush_last_tid; 1140a8599bd8SSage Weil if (pflush_tid) 1141a8599bd8SSage Weil *pflush_tid = flush_tid; 1142a8599bd8SSage Weil dout(" cap_flush_tid %d\n", (int)flush_tid); 1143a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1144a8599bd8SSage Weil if (flushing & (1 << i)) 1145a8599bd8SSage Weil ci->i_cap_flush_tid[i] = flush_tid; 11467d8cb26dSSage Weil 11477d8cb26dSSage Weil follows = ci->i_head_snapc->seq; 11487d8cb26dSSage Weil } else { 11497d8cb26dSSage Weil follows = 0; 1150a8599bd8SSage Weil } 1151a8599bd8SSage Weil 1152a8599bd8SSage Weil keep = cap->implemented; 1153a8599bd8SSage Weil seq = cap->seq; 1154a8599bd8SSage Weil issue_seq = cap->issue_seq; 1155a8599bd8SSage Weil mseq = cap->mseq; 1156a8599bd8SSage Weil size = inode->i_size; 1157a8599bd8SSage Weil ci->i_reported_size = size; 1158a8599bd8SSage Weil max_size = ci->i_wanted_max_size; 1159a8599bd8SSage Weil ci->i_requested_max_size = max_size; 1160a8599bd8SSage Weil mtime = inode->i_mtime; 1161a8599bd8SSage Weil atime = inode->i_atime; 1162a8599bd8SSage Weil time_warp_seq = ci->i_time_warp_seq; 1163a8599bd8SSage Weil uid = inode->i_uid; 1164a8599bd8SSage Weil gid = inode->i_gid; 1165a8599bd8SSage Weil mode = inode->i_mode; 1166a8599bd8SSage Weil 1167082afec9SSage Weil if (flushing & CEPH_CAP_XATTR_EXCL) { 1168a8599bd8SSage Weil __ceph_build_xattrs_blob(ci); 1169082afec9SSage Weil xattr_blob = ci->i_xattrs.blob; 1170082afec9SSage Weil xattr_version = ci->i_xattrs.version; 1171a8599bd8SSage Weil } 1172a8599bd8SSage Weil 1173a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1174a8599bd8SSage Weil 1175a8599bd8SSage Weil ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id, 1176a8599bd8SSage Weil op, keep, want, flushing, seq, flush_tid, issue_seq, mseq, 1177a8599bd8SSage Weil size, max_size, &mtime, &atime, time_warp_seq, 1178082afec9SSage Weil uid, gid, mode, xattr_version, xattr_blob, 1179a8599bd8SSage Weil follows); 1180a8599bd8SSage Weil if (ret < 0) { 1181a8599bd8SSage Weil dout("error sending cap msg, must requeue %p\n", inode); 1182a8599bd8SSage Weil delayed = 1; 1183a8599bd8SSage Weil } 1184a8599bd8SSage Weil 1185a8599bd8SSage Weil if (wake) 118603066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1187a8599bd8SSage Weil 1188a8599bd8SSage Weil return delayed; 1189a8599bd8SSage Weil } 1190a8599bd8SSage Weil 1191a8599bd8SSage Weil /* 1192a8599bd8SSage Weil * When a snapshot is taken, clients accumulate dirty metadata on 1193a8599bd8SSage Weil * inodes with capabilities in ceph_cap_snaps to describe the file 1194a8599bd8SSage Weil * state at the time the snapshot was taken. This must be flushed 1195a8599bd8SSage Weil * asynchronously back to the MDS once sync writes complete and dirty 1196a8599bd8SSage Weil * data is written out. 1197a8599bd8SSage Weil * 1198e835124cSSage Weil * Unless @again is true, skip cap_snaps that were already sent to 1199e835124cSSage Weil * the MDS (i.e., during this session). 1200e835124cSSage Weil * 1201a8599bd8SSage Weil * Called under i_lock. Takes s_mutex as needed. 1202a8599bd8SSage Weil */ 1203a8599bd8SSage Weil void __ceph_flush_snaps(struct ceph_inode_info *ci, 1204e835124cSSage Weil struct ceph_mds_session **psession, 1205e835124cSSage Weil int again) 1206cd84db6eSYehuda Sadeh __releases(ci->vfs_inode->i_lock) 1207cd84db6eSYehuda Sadeh __acquires(ci->vfs_inode->i_lock) 1208a8599bd8SSage Weil { 1209a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1210a8599bd8SSage Weil int mds; 1211a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1212a8599bd8SSage Weil u32 mseq; 1213a8599bd8SSage Weil struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc; 1214a8599bd8SSage Weil struct ceph_mds_session *session = NULL; /* if session != NULL, we hold 1215a8599bd8SSage Weil session->s_mutex */ 1216a8599bd8SSage Weil u64 next_follows = 0; /* keep track of how far we've gotten through the 1217a8599bd8SSage Weil i_cap_snaps list, and skip these entries next time 1218a8599bd8SSage Weil around to avoid an infinite loop */ 1219a8599bd8SSage Weil 1220a8599bd8SSage Weil if (psession) 1221a8599bd8SSage Weil session = *psession; 1222a8599bd8SSage Weil 1223a8599bd8SSage Weil dout("__flush_snaps %p\n", inode); 1224a8599bd8SSage Weil retry: 1225a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 1226a8599bd8SSage Weil /* avoid an infiniute loop after retry */ 1227a8599bd8SSage Weil if (capsnap->follows < next_follows) 1228a8599bd8SSage Weil continue; 1229a8599bd8SSage Weil /* 1230a8599bd8SSage Weil * we need to wait for sync writes to complete and for dirty 1231a8599bd8SSage Weil * pages to be written out. 1232a8599bd8SSage Weil */ 1233a8599bd8SSage Weil if (capsnap->dirty_pages || capsnap->writing) 1234cfc0bf66SSage Weil break; 1235a8599bd8SSage Weil 1236819ccbfaSSage Weil /* 1237819ccbfaSSage Weil * if cap writeback already occurred, we should have dropped 1238819ccbfaSSage Weil * the capsnap in ceph_put_wrbuffer_cap_refs. 1239819ccbfaSSage Weil */ 1240819ccbfaSSage Weil BUG_ON(capsnap->dirty == 0); 1241819ccbfaSSage Weil 1242a8599bd8SSage Weil /* pick mds, take s_mutex */ 1243ca81f3f6SSage Weil if (ci->i_auth_cap == NULL) { 1244ca81f3f6SSage Weil dout("no auth cap (migrating?), doing nothing\n"); 1245ca81f3f6SSage Weil goto out; 1246ca81f3f6SSage Weil } 1247e835124cSSage Weil 1248e835124cSSage Weil /* only flush each capsnap once */ 1249e835124cSSage Weil if (!again && !list_empty(&capsnap->flushing_item)) { 1250e835124cSSage Weil dout("already flushed %p, skipping\n", capsnap); 1251e835124cSSage Weil continue; 1252e835124cSSage Weil } 1253e835124cSSage Weil 1254ca81f3f6SSage Weil mds = ci->i_auth_cap->session->s_mds; 1255ca81f3f6SSage Weil mseq = ci->i_auth_cap->mseq; 1256ca81f3f6SSage Weil 1257a8599bd8SSage Weil if (session && session->s_mds != mds) { 1258a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1259a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1260a8599bd8SSage Weil ceph_put_mds_session(session); 1261a8599bd8SSage Weil session = NULL; 1262a8599bd8SSage Weil } 1263a8599bd8SSage Weil if (!session) { 1264a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1265a8599bd8SSage Weil mutex_lock(&mdsc->mutex); 1266a8599bd8SSage Weil session = __ceph_lookup_mds_session(mdsc, mds); 1267a8599bd8SSage Weil mutex_unlock(&mdsc->mutex); 1268a8599bd8SSage Weil if (session) { 1269a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1270a8599bd8SSage Weil session); 1271a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1272a8599bd8SSage Weil } 1273a8599bd8SSage Weil /* 1274a8599bd8SSage Weil * if session == NULL, we raced against a cap 1275ca81f3f6SSage Weil * deletion or migration. retry, and we'll 1276ca81f3f6SSage Weil * get a better @mds value next time. 1277a8599bd8SSage Weil */ 1278a8599bd8SSage Weil spin_lock(&inode->i_lock); 1279a8599bd8SSage Weil goto retry; 1280a8599bd8SSage Weil } 1281a8599bd8SSage Weil 1282a8599bd8SSage Weil capsnap->flush_tid = ++ci->i_cap_flush_last_tid; 1283a8599bd8SSage Weil atomic_inc(&capsnap->nref); 1284a8599bd8SSage Weil if (!list_empty(&capsnap->flushing_item)) 1285a8599bd8SSage Weil list_del_init(&capsnap->flushing_item); 1286a8599bd8SSage Weil list_add_tail(&capsnap->flushing_item, 1287a8599bd8SSage Weil &session->s_cap_snaps_flushing); 1288a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1289a8599bd8SSage Weil 1290cfc0bf66SSage Weil dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n", 1291cfc0bf66SSage Weil inode, capsnap, capsnap->follows, capsnap->flush_tid); 1292a8599bd8SSage Weil send_cap_msg(session, ceph_vino(inode).ino, 0, 1293a8599bd8SSage Weil CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0, 1294a8599bd8SSage Weil capsnap->dirty, 0, capsnap->flush_tid, 0, mseq, 1295a8599bd8SSage Weil capsnap->size, 0, 1296a8599bd8SSage Weil &capsnap->mtime, &capsnap->atime, 1297a8599bd8SSage Weil capsnap->time_warp_seq, 1298a8599bd8SSage Weil capsnap->uid, capsnap->gid, capsnap->mode, 12994a625be4SSage Weil capsnap->xattr_version, capsnap->xattr_blob, 1300a8599bd8SSage Weil capsnap->follows); 1301a8599bd8SSage Weil 1302a8599bd8SSage Weil next_follows = capsnap->follows + 1; 1303a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 1304a8599bd8SSage Weil 1305a8599bd8SSage Weil spin_lock(&inode->i_lock); 1306a8599bd8SSage Weil goto retry; 1307a8599bd8SSage Weil } 1308a8599bd8SSage Weil 1309a8599bd8SSage Weil /* we flushed them all; remove this inode from the queue */ 1310a8599bd8SSage Weil spin_lock(&mdsc->snap_flush_lock); 1311a8599bd8SSage Weil list_del_init(&ci->i_snap_flush_item); 1312a8599bd8SSage Weil spin_unlock(&mdsc->snap_flush_lock); 1313a8599bd8SSage Weil 1314ca81f3f6SSage Weil out: 1315a8599bd8SSage Weil if (psession) 1316a8599bd8SSage Weil *psession = session; 1317a8599bd8SSage Weil else if (session) { 1318a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1319a8599bd8SSage Weil ceph_put_mds_session(session); 1320a8599bd8SSage Weil } 1321a8599bd8SSage Weil } 1322a8599bd8SSage Weil 1323a8599bd8SSage Weil static void ceph_flush_snaps(struct ceph_inode_info *ci) 1324a8599bd8SSage Weil { 1325a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1326a8599bd8SSage Weil 1327a8599bd8SSage Weil spin_lock(&inode->i_lock); 1328e835124cSSage Weil __ceph_flush_snaps(ci, NULL, 0); 1329a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1330a8599bd8SSage Weil } 1331a8599bd8SSage Weil 1332a8599bd8SSage Weil /* 133376e3b390SSage Weil * Mark caps dirty. If inode is newly dirty, add to the global dirty 133476e3b390SSage Weil * list. 133576e3b390SSage Weil */ 133676e3b390SSage Weil void __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask) 133776e3b390SSage Weil { 1338640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 1339640ef79dSCheng Renquan &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 134076e3b390SSage Weil struct inode *inode = &ci->vfs_inode; 134176e3b390SSage Weil int was = ci->i_dirty_caps; 134276e3b390SSage Weil int dirty = 0; 134376e3b390SSage Weil 134476e3b390SSage Weil dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, 134576e3b390SSage Weil ceph_cap_string(mask), ceph_cap_string(was), 134676e3b390SSage Weil ceph_cap_string(was | mask)); 134776e3b390SSage Weil ci->i_dirty_caps |= mask; 134876e3b390SSage Weil if (was == 0) { 13497d8cb26dSSage Weil if (!ci->i_head_snapc) 13507d8cb26dSSage Weil ci->i_head_snapc = ceph_get_snap_context( 13517d8cb26dSSage Weil ci->i_snap_realm->cached_context); 13527d8cb26dSSage Weil dout(" inode %p now dirty snapc %p\n", &ci->vfs_inode, 13537d8cb26dSSage Weil ci->i_head_snapc); 135476e3b390SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 135576e3b390SSage Weil spin_lock(&mdsc->cap_dirty_lock); 135676e3b390SSage Weil list_add(&ci->i_dirty_item, &mdsc->cap_dirty); 135776e3b390SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 135876e3b390SSage Weil if (ci->i_flushing_caps == 0) { 135976e3b390SSage Weil igrab(inode); 136076e3b390SSage Weil dirty |= I_DIRTY_SYNC; 136176e3b390SSage Weil } 136276e3b390SSage Weil } 136376e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 136476e3b390SSage Weil if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && 136576e3b390SSage Weil (mask & CEPH_CAP_FILE_BUFFER)) 136676e3b390SSage Weil dirty |= I_DIRTY_DATASYNC; 136776e3b390SSage Weil if (dirty) 136876e3b390SSage Weil __mark_inode_dirty(inode, dirty); 136976e3b390SSage Weil __cap_delay_requeue(mdsc, ci); 137076e3b390SSage Weil } 137176e3b390SSage Weil 137276e3b390SSage Weil /* 1373a8599bd8SSage Weil * Add dirty inode to the flushing list. Assigned a seq number so we 1374a8599bd8SSage Weil * can wait for caps to flush without starving. 1375cdc35f96SSage Weil * 1376cdc35f96SSage Weil * Called under i_lock. 1377a8599bd8SSage Weil */ 1378cdc35f96SSage Weil static int __mark_caps_flushing(struct inode *inode, 1379a8599bd8SSage Weil struct ceph_mds_session *session) 1380a8599bd8SSage Weil { 1381640ef79dSCheng Renquan struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc; 1382a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1383cdc35f96SSage Weil int flushing; 1384a8599bd8SSage Weil 1385cdc35f96SSage Weil BUG_ON(ci->i_dirty_caps == 0); 1386a8599bd8SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 1387cdc35f96SSage Weil 1388cdc35f96SSage Weil flushing = ci->i_dirty_caps; 1389cdc35f96SSage Weil dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", 1390cdc35f96SSage Weil ceph_cap_string(flushing), 1391cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps), 1392cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps | flushing)); 1393cdc35f96SSage Weil ci->i_flushing_caps |= flushing; 1394cdc35f96SSage Weil ci->i_dirty_caps = 0; 1395afcdaea3SSage Weil dout(" inode %p now !dirty\n", inode); 1396cdc35f96SSage Weil 1397a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 1398cdc35f96SSage Weil list_del_init(&ci->i_dirty_item); 1399afcdaea3SSage Weil 1400afcdaea3SSage Weil ci->i_cap_flush_seq = ++mdsc->cap_flush_seq; 1401afcdaea3SSage Weil if (list_empty(&ci->i_flushing_item)) { 1402a8599bd8SSage Weil list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1403a8599bd8SSage Weil mdsc->num_cap_flushing++; 1404afcdaea3SSage Weil dout(" inode %p now flushing seq %lld\n", inode, 1405afcdaea3SSage Weil ci->i_cap_flush_seq); 1406afcdaea3SSage Weil } else { 1407afcdaea3SSage Weil list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1408afcdaea3SSage Weil dout(" inode %p now flushing (more) seq %lld\n", inode, 1409a8599bd8SSage Weil ci->i_cap_flush_seq); 1410a8599bd8SSage Weil } 1411a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 1412cdc35f96SSage Weil 1413cdc35f96SSage Weil return flushing; 1414a8599bd8SSage Weil } 1415a8599bd8SSage Weil 1416a8599bd8SSage Weil /* 14175ecad6fdSSage Weil * try to invalidate mapping pages without blocking. 14185ecad6fdSSage Weil */ 14195ecad6fdSSage Weil static int mapping_is_empty(struct address_space *mapping) 14205ecad6fdSSage Weil { 14215ecad6fdSSage Weil struct page *page = find_get_page(mapping, 0); 14225ecad6fdSSage Weil 14235ecad6fdSSage Weil if (!page) 14245ecad6fdSSage Weil return 1; 14255ecad6fdSSage Weil 14265ecad6fdSSage Weil put_page(page); 14275ecad6fdSSage Weil return 0; 14285ecad6fdSSage Weil } 14295ecad6fdSSage Weil 14305ecad6fdSSage Weil static int try_nonblocking_invalidate(struct inode *inode) 14315ecad6fdSSage Weil { 14325ecad6fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14335ecad6fdSSage Weil u32 invalidating_gen = ci->i_rdcache_gen; 14345ecad6fdSSage Weil 14355ecad6fdSSage Weil spin_unlock(&inode->i_lock); 14365ecad6fdSSage Weil invalidate_mapping_pages(&inode->i_data, 0, -1); 14375ecad6fdSSage Weil spin_lock(&inode->i_lock); 14385ecad6fdSSage Weil 14395ecad6fdSSage Weil if (mapping_is_empty(&inode->i_data) && 14405ecad6fdSSage Weil invalidating_gen == ci->i_rdcache_gen) { 14415ecad6fdSSage Weil /* success. */ 14425ecad6fdSSage Weil dout("try_nonblocking_invalidate %p success\n", inode); 14435ecad6fdSSage Weil ci->i_rdcache_gen = 0; 14445ecad6fdSSage Weil ci->i_rdcache_revoking = 0; 14455ecad6fdSSage Weil return 0; 14465ecad6fdSSage Weil } 14475ecad6fdSSage Weil dout("try_nonblocking_invalidate %p failed\n", inode); 14485ecad6fdSSage Weil return -1; 14495ecad6fdSSage Weil } 14505ecad6fdSSage Weil 14515ecad6fdSSage Weil /* 1452a8599bd8SSage Weil * Swiss army knife function to examine currently used and wanted 1453a8599bd8SSage Weil * versus held caps. Release, flush, ack revoked caps to mds as 1454a8599bd8SSage Weil * appropriate. 1455a8599bd8SSage Weil * 1456a8599bd8SSage Weil * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay 1457a8599bd8SSage Weil * cap release further. 1458a8599bd8SSage Weil * CHECK_CAPS_AUTHONLY - we should only check the auth cap 1459a8599bd8SSage Weil * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without 1460a8599bd8SSage Weil * further delay. 1461a8599bd8SSage Weil */ 1462a8599bd8SSage Weil void ceph_check_caps(struct ceph_inode_info *ci, int flags, 1463a8599bd8SSage Weil struct ceph_mds_session *session) 1464a8599bd8SSage Weil { 1465a8599bd8SSage Weil struct ceph_client *client = ceph_inode_to_client(&ci->vfs_inode); 1466a8599bd8SSage Weil struct ceph_mds_client *mdsc = &client->mdsc; 1467a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1468a8599bd8SSage Weil struct ceph_cap *cap; 1469a8599bd8SSage Weil int file_wanted, used; 1470a8599bd8SSage Weil int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ 1471cbd03635SSage Weil int issued, implemented, want, retain, revoking, flushing = 0; 1472a8599bd8SSage Weil int mds = -1; /* keep track of how far we've gone through i_caps list 1473a8599bd8SSage Weil to avoid an infinite loop on retry */ 1474a8599bd8SSage Weil struct rb_node *p; 1475a8599bd8SSage Weil int tried_invalidate = 0; 1476a8599bd8SSage Weil int delayed = 0, sent = 0, force_requeue = 0, num; 1477cbd03635SSage Weil int queue_invalidate = 0; 1478a8599bd8SSage Weil int is_delayed = flags & CHECK_CAPS_NODELAY; 1479a8599bd8SSage Weil 1480a8599bd8SSage Weil /* if we are unmounting, flush any unused caps immediately. */ 1481a8599bd8SSage Weil if (mdsc->stopping) 1482a8599bd8SSage Weil is_delayed = 1; 1483a8599bd8SSage Weil 1484a8599bd8SSage Weil spin_lock(&inode->i_lock); 1485a8599bd8SSage Weil 1486a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 1487a8599bd8SSage Weil flags |= CHECK_CAPS_FLUSH; 1488a8599bd8SSage Weil 1489a8599bd8SSage Weil /* flush snaps first time around only */ 1490a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) 1491e835124cSSage Weil __ceph_flush_snaps(ci, &session, 0); 1492a8599bd8SSage Weil goto retry_locked; 1493a8599bd8SSage Weil retry: 1494a8599bd8SSage Weil spin_lock(&inode->i_lock); 1495a8599bd8SSage Weil retry_locked: 1496a8599bd8SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 1497a8599bd8SSage Weil used = __ceph_caps_used(ci); 1498a8599bd8SSage Weil want = file_wanted | used; 1499cbd03635SSage Weil issued = __ceph_caps_issued(ci, &implemented); 1500cbd03635SSage Weil revoking = implemented & ~issued; 1501a8599bd8SSage Weil 1502a8599bd8SSage Weil retain = want | CEPH_CAP_PIN; 1503a8599bd8SSage Weil if (!mdsc->stopping && inode->i_nlink > 0) { 1504a8599bd8SSage Weil if (want) { 1505a8599bd8SSage Weil retain |= CEPH_CAP_ANY; /* be greedy */ 1506a8599bd8SSage Weil } else { 1507a8599bd8SSage Weil retain |= CEPH_CAP_ANY_SHARED; 1508a8599bd8SSage Weil /* 1509a8599bd8SSage Weil * keep RD only if we didn't have the file open RW, 1510a8599bd8SSage Weil * because then the mds would revoke it anyway to 1511a8599bd8SSage Weil * journal max_size=0. 1512a8599bd8SSage Weil */ 1513a8599bd8SSage Weil if (ci->i_max_size == 0) 1514a8599bd8SSage Weil retain |= CEPH_CAP_ANY_RD; 1515a8599bd8SSage Weil } 1516a8599bd8SSage Weil } 1517a8599bd8SSage Weil 1518a8599bd8SSage Weil dout("check_caps %p file_want %s used %s dirty %s flushing %s" 1519cbd03635SSage Weil " issued %s revoking %s retain %s %s%s%s\n", inode, 1520a8599bd8SSage Weil ceph_cap_string(file_wanted), 1521a8599bd8SSage Weil ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), 1522a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps), 1523cbd03635SSage Weil ceph_cap_string(issued), ceph_cap_string(revoking), 1524a8599bd8SSage Weil ceph_cap_string(retain), 1525a8599bd8SSage Weil (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", 1526a8599bd8SSage Weil (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", 1527a8599bd8SSage Weil (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); 1528a8599bd8SSage Weil 1529a8599bd8SSage Weil /* 1530a8599bd8SSage Weil * If we no longer need to hold onto old our caps, and we may 1531a8599bd8SSage Weil * have cached pages, but don't want them, then try to invalidate. 1532a8599bd8SSage Weil * If we fail, it's because pages are locked.... try again later. 1533a8599bd8SSage Weil */ 1534a8599bd8SSage Weil if ((!is_delayed || mdsc->stopping) && 1535a8599bd8SSage Weil ci->i_wrbuffer_ref == 0 && /* no dirty pages... */ 1536a8599bd8SSage Weil ci->i_rdcache_gen && /* may have cached pages */ 1537cbd03635SSage Weil (file_wanted == 0 || /* no open files */ 15382962507cSSage Weil (revoking & (CEPH_CAP_FILE_CACHE| 15392962507cSSage Weil CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */ 1540a8599bd8SSage Weil !tried_invalidate) { 1541a8599bd8SSage Weil dout("check_caps trying to invalidate on %p\n", inode); 15425ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) < 0) { 15432962507cSSage Weil if (revoking & (CEPH_CAP_FILE_CACHE| 15442962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) { 1545cbd03635SSage Weil dout("check_caps queuing invalidate\n"); 1546cbd03635SSage Weil queue_invalidate = 1; 1547cbd03635SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 1548a8599bd8SSage Weil } else { 1549a8599bd8SSage Weil dout("check_caps failed to invalidate pages\n"); 1550a8599bd8SSage Weil /* we failed to invalidate pages. check these 1551a8599bd8SSage Weil caps again later. */ 1552a8599bd8SSage Weil force_requeue = 1; 1553a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 1554a8599bd8SSage Weil } 15555ecad6fdSSage Weil } 1556a8599bd8SSage Weil tried_invalidate = 1; 1557a8599bd8SSage Weil goto retry_locked; 1558a8599bd8SSage Weil } 1559a8599bd8SSage Weil 1560a8599bd8SSage Weil num = 0; 1561a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 1562a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 1563a8599bd8SSage Weil num++; 1564a8599bd8SSage Weil 1565a8599bd8SSage Weil /* avoid looping forever */ 1566a8599bd8SSage Weil if (mds >= cap->mds || 1567a8599bd8SSage Weil ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) 1568a8599bd8SSage Weil continue; 1569a8599bd8SSage Weil 1570a8599bd8SSage Weil /* NOTE: no side-effects allowed, until we take s_mutex */ 1571a8599bd8SSage Weil 1572a8599bd8SSage Weil revoking = cap->implemented & ~cap->issued; 1573a8599bd8SSage Weil if (revoking) 1574a8599bd8SSage Weil dout(" mds%d revoking %s\n", cap->mds, 1575a8599bd8SSage Weil ceph_cap_string(revoking)); 1576a8599bd8SSage Weil 1577a8599bd8SSage Weil if (cap == ci->i_auth_cap && 1578a8599bd8SSage Weil (cap->issued & CEPH_CAP_FILE_WR)) { 1579a8599bd8SSage Weil /* request larger max_size from MDS? */ 1580a8599bd8SSage Weil if (ci->i_wanted_max_size > ci->i_max_size && 1581a8599bd8SSage Weil ci->i_wanted_max_size > ci->i_requested_max_size) { 1582a8599bd8SSage Weil dout("requesting new max_size\n"); 1583a8599bd8SSage Weil goto ack; 1584a8599bd8SSage Weil } 1585a8599bd8SSage Weil 1586a8599bd8SSage Weil /* approaching file_max? */ 1587a8599bd8SSage Weil if ((inode->i_size << 1) >= ci->i_max_size && 1588a8599bd8SSage Weil (ci->i_reported_size << 1) < ci->i_max_size) { 1589a8599bd8SSage Weil dout("i_size approaching max_size\n"); 1590a8599bd8SSage Weil goto ack; 1591a8599bd8SSage Weil } 1592a8599bd8SSage Weil } 1593a8599bd8SSage Weil /* flush anything dirty? */ 1594a8599bd8SSage Weil if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) && 1595a8599bd8SSage Weil ci->i_dirty_caps) { 1596a8599bd8SSage Weil dout("flushing dirty caps\n"); 1597a8599bd8SSage Weil goto ack; 1598a8599bd8SSage Weil } 1599a8599bd8SSage Weil 1600a8599bd8SSage Weil /* completed revocation? going down and there are no caps? */ 1601a8599bd8SSage Weil if (revoking && (revoking & used) == 0) { 1602a8599bd8SSage Weil dout("completed revocation of %s\n", 1603a8599bd8SSage Weil ceph_cap_string(cap->implemented & ~cap->issued)); 1604a8599bd8SSage Weil goto ack; 1605a8599bd8SSage Weil } 1606a8599bd8SSage Weil 1607a8599bd8SSage Weil /* want more caps from mds? */ 1608a8599bd8SSage Weil if (want & ~(cap->mds_wanted | cap->issued)) 1609a8599bd8SSage Weil goto ack; 1610a8599bd8SSage Weil 1611a8599bd8SSage Weil /* things we might delay */ 1612a8599bd8SSage Weil if ((cap->issued & ~retain) == 0 && 1613a8599bd8SSage Weil cap->mds_wanted == want) 1614a8599bd8SSage Weil continue; /* nope, all good */ 1615a8599bd8SSage Weil 1616a8599bd8SSage Weil if (is_delayed) 1617a8599bd8SSage Weil goto ack; 1618a8599bd8SSage Weil 1619a8599bd8SSage Weil /* delay? */ 1620a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1621a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) { 1622a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s\n", 1623a8599bd8SSage Weil ceph_cap_string(cap->issued), 1624a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1625a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1626a8599bd8SSage Weil ceph_cap_string(want)); 1627a8599bd8SSage Weil delayed++; 1628a8599bd8SSage Weil continue; 1629a8599bd8SSage Weil } 1630a8599bd8SSage Weil 1631a8599bd8SSage Weil ack: 1632e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1633e9964c10SSage Weil dout(" skipping %p I_NOFLUSH set\n", inode); 1634e9964c10SSage Weil continue; 1635e9964c10SSage Weil } 1636e9964c10SSage Weil 1637a8599bd8SSage Weil if (session && session != cap->session) { 1638a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1639a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1640a8599bd8SSage Weil session = NULL; 1641a8599bd8SSage Weil } 1642a8599bd8SSage Weil if (!session) { 1643a8599bd8SSage Weil session = cap->session; 1644a8599bd8SSage Weil if (mutex_trylock(&session->s_mutex) == 0) { 1645a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1646a8599bd8SSage Weil session); 1647a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1648a8599bd8SSage Weil if (took_snap_rwsem) { 1649a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1650a8599bd8SSage Weil took_snap_rwsem = 0; 1651a8599bd8SSage Weil } 1652a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1653a8599bd8SSage Weil goto retry; 1654a8599bd8SSage Weil } 1655a8599bd8SSage Weil } 1656a8599bd8SSage Weil /* take snap_rwsem after session mutex */ 1657a8599bd8SSage Weil if (!took_snap_rwsem) { 1658a8599bd8SSage Weil if (down_read_trylock(&mdsc->snap_rwsem) == 0) { 1659a8599bd8SSage Weil dout("inverting snap/in locks on %p\n", 1660a8599bd8SSage Weil inode); 1661a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1662a8599bd8SSage Weil down_read(&mdsc->snap_rwsem); 1663a8599bd8SSage Weil took_snap_rwsem = 1; 1664a8599bd8SSage Weil goto retry; 1665a8599bd8SSage Weil } 1666a8599bd8SSage Weil took_snap_rwsem = 1; 1667a8599bd8SSage Weil } 1668a8599bd8SSage Weil 1669cdc35f96SSage Weil if (cap == ci->i_auth_cap && ci->i_dirty_caps) 1670cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 1671a8599bd8SSage Weil 1672a8599bd8SSage Weil mds = cap->mds; /* remember mds, so we don't repeat */ 1673a8599bd8SSage Weil sent++; 1674a8599bd8SSage Weil 1675a8599bd8SSage Weil /* __send_cap drops i_lock */ 1676a8599bd8SSage Weil delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, used, want, 1677a8599bd8SSage Weil retain, flushing, NULL); 1678a8599bd8SSage Weil goto retry; /* retake i_lock and restart our cap scan. */ 1679a8599bd8SSage Weil } 1680a8599bd8SSage Weil 1681a8599bd8SSage Weil /* 1682a8599bd8SSage Weil * Reschedule delayed caps release if we delayed anything, 1683a8599bd8SSage Weil * otherwise cancel. 1684a8599bd8SSage Weil */ 1685a8599bd8SSage Weil if (delayed && is_delayed) 1686a8599bd8SSage Weil force_requeue = 1; /* __send_cap delayed release; requeue */ 1687a8599bd8SSage Weil if (!delayed && !is_delayed) 1688a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 1689a8599bd8SSage Weil else if (!is_delayed || force_requeue) 1690a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1691a8599bd8SSage Weil 1692a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1693a8599bd8SSage Weil 1694cbd03635SSage Weil if (queue_invalidate) 16953c6f6b79SSage Weil ceph_queue_invalidate(inode); 1696cbd03635SSage Weil 1697cdc2ce05SSage Weil if (session) 1698a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1699a8599bd8SSage Weil if (took_snap_rwsem) 1700a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1701a8599bd8SSage Weil } 1702a8599bd8SSage Weil 1703a8599bd8SSage Weil /* 1704a8599bd8SSage Weil * Try to flush dirty caps back to the auth mds. 1705a8599bd8SSage Weil */ 1706a8599bd8SSage Weil static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session, 1707a8599bd8SSage Weil unsigned *flush_tid) 1708a8599bd8SSage Weil { 1709640ef79dSCheng Renquan struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc; 1710a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1711a8599bd8SSage Weil int unlock_session = session ? 0 : 1; 1712a8599bd8SSage Weil int flushing = 0; 1713a8599bd8SSage Weil 1714a8599bd8SSage Weil retry: 1715a8599bd8SSage Weil spin_lock(&inode->i_lock); 1716e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1717e9964c10SSage Weil dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode); 1718e9964c10SSage Weil goto out; 1719e9964c10SSage Weil } 1720a8599bd8SSage Weil if (ci->i_dirty_caps && ci->i_auth_cap) { 1721a8599bd8SSage Weil struct ceph_cap *cap = ci->i_auth_cap; 1722a8599bd8SSage Weil int used = __ceph_caps_used(ci); 1723a8599bd8SSage Weil int want = __ceph_caps_wanted(ci); 1724a8599bd8SSage Weil int delayed; 1725a8599bd8SSage Weil 1726a8599bd8SSage Weil if (!session) { 1727a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1728a8599bd8SSage Weil session = cap->session; 1729a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1730a8599bd8SSage Weil goto retry; 1731a8599bd8SSage Weil } 1732a8599bd8SSage Weil BUG_ON(session != cap->session); 1733a8599bd8SSage Weil if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) 1734a8599bd8SSage Weil goto out; 1735a8599bd8SSage Weil 1736cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 1737a8599bd8SSage Weil 1738a8599bd8SSage Weil /* __send_cap drops i_lock */ 1739a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want, 1740a8599bd8SSage Weil cap->issued | cap->implemented, flushing, 1741a8599bd8SSage Weil flush_tid); 1742a8599bd8SSage Weil if (!delayed) 1743a8599bd8SSage Weil goto out_unlocked; 1744a8599bd8SSage Weil 1745a8599bd8SSage Weil spin_lock(&inode->i_lock); 1746a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1747a8599bd8SSage Weil } 1748a8599bd8SSage Weil out: 1749a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1750a8599bd8SSage Weil out_unlocked: 1751a8599bd8SSage Weil if (session && unlock_session) 1752a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1753a8599bd8SSage Weil return flushing; 1754a8599bd8SSage Weil } 1755a8599bd8SSage Weil 1756a8599bd8SSage Weil /* 1757a8599bd8SSage Weil * Return true if we've flushed caps through the given flush_tid. 1758a8599bd8SSage Weil */ 1759a8599bd8SSage Weil static int caps_are_flushed(struct inode *inode, unsigned tid) 1760a8599bd8SSage Weil { 1761a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1762a5ee751cSDan Carpenter int i, ret = 1; 1763a8599bd8SSage Weil 1764a8599bd8SSage Weil spin_lock(&inode->i_lock); 1765a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1766a8599bd8SSage Weil if ((ci->i_flushing_caps & (1 << i)) && 1767a8599bd8SSage Weil ci->i_cap_flush_tid[i] <= tid) { 1768a8599bd8SSage Weil /* still flushing this bit */ 1769a8599bd8SSage Weil ret = 0; 1770a8599bd8SSage Weil break; 1771a8599bd8SSage Weil } 1772a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1773a8599bd8SSage Weil return ret; 1774a8599bd8SSage Weil } 1775a8599bd8SSage Weil 1776a8599bd8SSage Weil /* 1777a8599bd8SSage Weil * Wait on any unsafe replies for the given inode. First wait on the 1778a8599bd8SSage Weil * newest request, and make that the upper bound. Then, if there are 1779a8599bd8SSage Weil * more requests, keep waiting on the oldest as long as it is still older 1780a8599bd8SSage Weil * than the original request. 1781a8599bd8SSage Weil */ 1782a8599bd8SSage Weil static void sync_write_wait(struct inode *inode) 1783a8599bd8SSage Weil { 1784a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1785a8599bd8SSage Weil struct list_head *head = &ci->i_unsafe_writes; 1786a8599bd8SSage Weil struct ceph_osd_request *req; 1787a8599bd8SSage Weil u64 last_tid; 1788a8599bd8SSage Weil 1789a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1790a8599bd8SSage Weil if (list_empty(head)) 1791a8599bd8SSage Weil goto out; 1792a8599bd8SSage Weil 1793a8599bd8SSage Weil /* set upper bound as _last_ entry in chain */ 1794a8599bd8SSage Weil req = list_entry(head->prev, struct ceph_osd_request, 1795a8599bd8SSage Weil r_unsafe_item); 1796a8599bd8SSage Weil last_tid = req->r_tid; 1797a8599bd8SSage Weil 1798a8599bd8SSage Weil do { 1799a8599bd8SSage Weil ceph_osdc_get_request(req); 1800a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1801a8599bd8SSage Weil dout("sync_write_wait on tid %llu (until %llu)\n", 1802a8599bd8SSage Weil req->r_tid, last_tid); 1803a8599bd8SSage Weil wait_for_completion(&req->r_safe_completion); 1804a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1805a8599bd8SSage Weil ceph_osdc_put_request(req); 1806a8599bd8SSage Weil 1807a8599bd8SSage Weil /* 1808a8599bd8SSage Weil * from here on look at first entry in chain, since we 1809a8599bd8SSage Weil * only want to wait for anything older than last_tid 1810a8599bd8SSage Weil */ 1811a8599bd8SSage Weil if (list_empty(head)) 1812a8599bd8SSage Weil break; 1813a8599bd8SSage Weil req = list_entry(head->next, struct ceph_osd_request, 1814a8599bd8SSage Weil r_unsafe_item); 1815a8599bd8SSage Weil } while (req->r_tid < last_tid); 1816a8599bd8SSage Weil out: 1817a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1818a8599bd8SSage Weil } 1819a8599bd8SSage Weil 18207ea80859SChristoph Hellwig int ceph_fsync(struct file *file, int datasync) 1821a8599bd8SSage Weil { 18227ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 1823a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1824a8599bd8SSage Weil unsigned flush_tid; 1825a8599bd8SSage Weil int ret; 1826a8599bd8SSage Weil int dirty; 1827a8599bd8SSage Weil 1828a8599bd8SSage Weil dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); 1829a8599bd8SSage Weil sync_write_wait(inode); 1830a8599bd8SSage Weil 1831a8599bd8SSage Weil ret = filemap_write_and_wait(inode->i_mapping); 1832a8599bd8SSage Weil if (ret < 0) 1833a8599bd8SSage Weil return ret; 1834a8599bd8SSage Weil 1835a8599bd8SSage Weil dirty = try_flush_caps(inode, NULL, &flush_tid); 1836a8599bd8SSage Weil dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); 1837a8599bd8SSage Weil 1838a8599bd8SSage Weil /* 1839a8599bd8SSage Weil * only wait on non-file metadata writeback (the mds 1840a8599bd8SSage Weil * can recover size and mtime, so we don't need to 1841a8599bd8SSage Weil * wait for that) 1842a8599bd8SSage Weil */ 1843a8599bd8SSage Weil if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { 1844a8599bd8SSage Weil dout("fsync waiting for flush_tid %u\n", flush_tid); 1845a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 1846a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1847a8599bd8SSage Weil } 1848a8599bd8SSage Weil 1849a8599bd8SSage Weil dout("fsync %p%s done\n", inode, datasync ? " datasync" : ""); 1850a8599bd8SSage Weil return ret; 1851a8599bd8SSage Weil } 1852a8599bd8SSage Weil 1853a8599bd8SSage Weil /* 1854a8599bd8SSage Weil * Flush any dirty caps back to the mds. If we aren't asked to wait, 1855a8599bd8SSage Weil * queue inode for flush but don't do so immediately, because we can 1856a8599bd8SSage Weil * get by with fewer MDS messages if we wait for data writeback to 1857a8599bd8SSage Weil * complete first. 1858a8599bd8SSage Weil */ 1859f1a3d572SStephen Rothwell int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) 1860a8599bd8SSage Weil { 1861a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1862a8599bd8SSage Weil unsigned flush_tid; 1863a8599bd8SSage Weil int err = 0; 1864a8599bd8SSage Weil int dirty; 1865f1a3d572SStephen Rothwell int wait = wbc->sync_mode == WB_SYNC_ALL; 1866a8599bd8SSage Weil 1867a8599bd8SSage Weil dout("write_inode %p wait=%d\n", inode, wait); 1868a8599bd8SSage Weil if (wait) { 1869a8599bd8SSage Weil dirty = try_flush_caps(inode, NULL, &flush_tid); 1870a8599bd8SSage Weil if (dirty) 1871a8599bd8SSage Weil err = wait_event_interruptible(ci->i_cap_wq, 1872a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1873a8599bd8SSage Weil } else { 1874640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 1875640ef79dSCheng Renquan &ceph_sb_to_client(inode->i_sb)->mdsc; 1876a8599bd8SSage Weil 1877a8599bd8SSage Weil spin_lock(&inode->i_lock); 1878a8599bd8SSage Weil if (__ceph_caps_dirty(ci)) 1879a8599bd8SSage Weil __cap_delay_requeue_front(mdsc, ci); 1880a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1881a8599bd8SSage Weil } 1882a8599bd8SSage Weil return err; 1883a8599bd8SSage Weil } 1884a8599bd8SSage Weil 1885a8599bd8SSage Weil /* 1886a8599bd8SSage Weil * After a recovering MDS goes active, we need to resend any caps 1887a8599bd8SSage Weil * we were flushing. 1888a8599bd8SSage Weil * 1889a8599bd8SSage Weil * Caller holds session->s_mutex. 1890a8599bd8SSage Weil */ 1891a8599bd8SSage Weil static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc, 1892a8599bd8SSage Weil struct ceph_mds_session *session) 1893a8599bd8SSage Weil { 1894a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1895a8599bd8SSage Weil 1896a8599bd8SSage Weil dout("kick_flushing_capsnaps mds%d\n", session->s_mds); 1897a8599bd8SSage Weil list_for_each_entry(capsnap, &session->s_cap_snaps_flushing, 1898a8599bd8SSage Weil flushing_item) { 1899a8599bd8SSage Weil struct ceph_inode_info *ci = capsnap->ci; 1900a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1901a8599bd8SSage Weil struct ceph_cap *cap; 1902a8599bd8SSage Weil 1903a8599bd8SSage Weil spin_lock(&inode->i_lock); 1904a8599bd8SSage Weil cap = ci->i_auth_cap; 1905a8599bd8SSage Weil if (cap && cap->session == session) { 1906a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p capsnap %p\n", inode, 1907a8599bd8SSage Weil cap, capsnap); 1908e835124cSSage Weil __ceph_flush_snaps(ci, &session, 1); 1909a8599bd8SSage Weil } else { 1910a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1911a8599bd8SSage Weil cap, session->s_mds); 1912a8599bd8SSage Weil } 19130b0c06d1SSage Weil spin_unlock(&inode->i_lock); 1914a8599bd8SSage Weil } 1915a8599bd8SSage Weil } 1916a8599bd8SSage Weil 1917a8599bd8SSage Weil void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, 1918a8599bd8SSage Weil struct ceph_mds_session *session) 1919a8599bd8SSage Weil { 1920a8599bd8SSage Weil struct ceph_inode_info *ci; 1921a8599bd8SSage Weil 1922a8599bd8SSage Weil kick_flushing_capsnaps(mdsc, session); 1923a8599bd8SSage Weil 1924a8599bd8SSage Weil dout("kick_flushing_caps mds%d\n", session->s_mds); 1925a8599bd8SSage Weil list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { 1926a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1927a8599bd8SSage Weil struct ceph_cap *cap; 1928a8599bd8SSage Weil int delayed = 0; 1929a8599bd8SSage Weil 1930a8599bd8SSage Weil spin_lock(&inode->i_lock); 1931a8599bd8SSage Weil cap = ci->i_auth_cap; 1932a8599bd8SSage Weil if (cap && cap->session == session) { 1933a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p %s\n", inode, 1934a8599bd8SSage Weil cap, ceph_cap_string(ci->i_flushing_caps)); 1935a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 1936a8599bd8SSage Weil __ceph_caps_used(ci), 1937a8599bd8SSage Weil __ceph_caps_wanted(ci), 1938a8599bd8SSage Weil cap->issued | cap->implemented, 1939a8599bd8SSage Weil ci->i_flushing_caps, NULL); 1940a8599bd8SSage Weil if (delayed) { 1941a8599bd8SSage Weil spin_lock(&inode->i_lock); 1942a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1943a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1944a8599bd8SSage Weil } 1945a8599bd8SSage Weil } else { 1946a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1947a8599bd8SSage Weil cap, session->s_mds); 1948a8599bd8SSage Weil spin_unlock(&inode->i_lock); 1949a8599bd8SSage Weil } 1950a8599bd8SSage Weil } 1951a8599bd8SSage Weil } 1952a8599bd8SSage Weil 1953a8599bd8SSage Weil 1954a8599bd8SSage Weil /* 1955a8599bd8SSage Weil * Take references to capabilities we hold, so that we don't release 1956a8599bd8SSage Weil * them to the MDS prematurely. 1957a8599bd8SSage Weil * 1958a8599bd8SSage Weil * Protected by i_lock. 1959a8599bd8SSage Weil */ 1960a8599bd8SSage Weil static void __take_cap_refs(struct ceph_inode_info *ci, int got) 1961a8599bd8SSage Weil { 1962a8599bd8SSage Weil if (got & CEPH_CAP_PIN) 1963a8599bd8SSage Weil ci->i_pin_ref++; 1964a8599bd8SSage Weil if (got & CEPH_CAP_FILE_RD) 1965a8599bd8SSage Weil ci->i_rd_ref++; 1966a8599bd8SSage Weil if (got & CEPH_CAP_FILE_CACHE) 1967a8599bd8SSage Weil ci->i_rdcache_ref++; 1968a8599bd8SSage Weil if (got & CEPH_CAP_FILE_WR) 1969a8599bd8SSage Weil ci->i_wr_ref++; 1970a8599bd8SSage Weil if (got & CEPH_CAP_FILE_BUFFER) { 1971a8599bd8SSage Weil if (ci->i_wrbuffer_ref == 0) 1972a8599bd8SSage Weil igrab(&ci->vfs_inode); 1973a8599bd8SSage Weil ci->i_wrbuffer_ref++; 1974a8599bd8SSage Weil dout("__take_cap_refs %p wrbuffer %d -> %d (?)\n", 1975a8599bd8SSage Weil &ci->vfs_inode, ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref); 1976a8599bd8SSage Weil } 1977a8599bd8SSage Weil } 1978a8599bd8SSage Weil 1979a8599bd8SSage Weil /* 1980a8599bd8SSage Weil * Try to grab cap references. Specify those refs we @want, and the 1981a8599bd8SSage Weil * minimal set we @need. Also include the larger offset we are writing 1982a8599bd8SSage Weil * to (when applicable), and check against max_size here as well. 1983a8599bd8SSage Weil * Note that caller is responsible for ensuring max_size increases are 1984a8599bd8SSage Weil * requested from the MDS. 1985a8599bd8SSage Weil */ 1986a8599bd8SSage Weil static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want, 1987a8599bd8SSage Weil int *got, loff_t endoff, int *check_max, int *err) 1988a8599bd8SSage Weil { 1989a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1990a8599bd8SSage Weil int ret = 0; 1991a8599bd8SSage Weil int have, implemented; 1992195d3ce2SSage Weil int file_wanted; 1993a8599bd8SSage Weil 1994a8599bd8SSage Weil dout("get_cap_refs %p need %s want %s\n", inode, 1995a8599bd8SSage Weil ceph_cap_string(need), ceph_cap_string(want)); 1996a8599bd8SSage Weil spin_lock(&inode->i_lock); 1997a8599bd8SSage Weil 1998195d3ce2SSage Weil /* make sure file is actually open */ 1999195d3ce2SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 2000195d3ce2SSage Weil if ((file_wanted & need) == 0) { 2001195d3ce2SSage Weil dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", 2002195d3ce2SSage Weil ceph_cap_string(need), ceph_cap_string(file_wanted)); 2003a8599bd8SSage Weil *err = -EBADF; 2004a8599bd8SSage Weil ret = 1; 2005a8599bd8SSage Weil goto out; 2006a8599bd8SSage Weil } 2007a8599bd8SSage Weil 2008a8599bd8SSage Weil if (need & CEPH_CAP_FILE_WR) { 2009a8599bd8SSage Weil if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { 2010a8599bd8SSage Weil dout("get_cap_refs %p endoff %llu > maxsize %llu\n", 2011a8599bd8SSage Weil inode, endoff, ci->i_max_size); 2012a8599bd8SSage Weil if (endoff > ci->i_wanted_max_size) { 2013a8599bd8SSage Weil *check_max = 1; 2014a8599bd8SSage Weil ret = 1; 2015a8599bd8SSage Weil } 2016a8599bd8SSage Weil goto out; 2017a8599bd8SSage Weil } 2018a8599bd8SSage Weil /* 2019a8599bd8SSage Weil * If a sync write is in progress, we must wait, so that we 2020a8599bd8SSage Weil * can get a final snapshot value for size+mtime. 2021a8599bd8SSage Weil */ 2022a8599bd8SSage Weil if (__ceph_have_pending_cap_snap(ci)) { 2023a8599bd8SSage Weil dout("get_cap_refs %p cap_snap_pending\n", inode); 2024a8599bd8SSage Weil goto out; 2025a8599bd8SSage Weil } 2026a8599bd8SSage Weil } 2027a8599bd8SSage Weil have = __ceph_caps_issued(ci, &implemented); 2028a8599bd8SSage Weil 2029a8599bd8SSage Weil /* 2030a8599bd8SSage Weil * disallow writes while a truncate is pending 2031a8599bd8SSage Weil */ 2032a8599bd8SSage Weil if (ci->i_truncate_pending) 2033a8599bd8SSage Weil have &= ~CEPH_CAP_FILE_WR; 2034a8599bd8SSage Weil 2035a8599bd8SSage Weil if ((have & need) == need) { 2036a8599bd8SSage Weil /* 2037a8599bd8SSage Weil * Look at (implemented & ~have & not) so that we keep waiting 2038a8599bd8SSage Weil * on transition from wanted -> needed caps. This is needed 2039a8599bd8SSage Weil * for WRBUFFER|WR -> WR to avoid a new WR sync write from 2040a8599bd8SSage Weil * going before a prior buffered writeback happens. 2041a8599bd8SSage Weil */ 2042a8599bd8SSage Weil int not = want & ~(have & need); 2043a8599bd8SSage Weil int revoking = implemented & ~have; 2044a8599bd8SSage Weil dout("get_cap_refs %p have %s but not %s (revoking %s)\n", 2045a8599bd8SSage Weil inode, ceph_cap_string(have), ceph_cap_string(not), 2046a8599bd8SSage Weil ceph_cap_string(revoking)); 2047a8599bd8SSage Weil if ((revoking & not) == 0) { 2048a8599bd8SSage Weil *got = need | (have & want); 2049a8599bd8SSage Weil __take_cap_refs(ci, *got); 2050a8599bd8SSage Weil ret = 1; 2051a8599bd8SSage Weil } 2052a8599bd8SSage Weil } else { 2053a8599bd8SSage Weil dout("get_cap_refs %p have %s needed %s\n", inode, 2054a8599bd8SSage Weil ceph_cap_string(have), ceph_cap_string(need)); 2055a8599bd8SSage Weil } 2056a8599bd8SSage Weil out: 2057a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2058a8599bd8SSage Weil dout("get_cap_refs %p ret %d got %s\n", inode, 2059a8599bd8SSage Weil ret, ceph_cap_string(*got)); 2060a8599bd8SSage Weil return ret; 2061a8599bd8SSage Weil } 2062a8599bd8SSage Weil 2063a8599bd8SSage Weil /* 2064a8599bd8SSage Weil * Check the offset we are writing up to against our current 2065a8599bd8SSage Weil * max_size. If necessary, tell the MDS we want to write to 2066a8599bd8SSage Weil * a larger offset. 2067a8599bd8SSage Weil */ 2068a8599bd8SSage Weil static void check_max_size(struct inode *inode, loff_t endoff) 2069a8599bd8SSage Weil { 2070a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2071a8599bd8SSage Weil int check = 0; 2072a8599bd8SSage Weil 2073a8599bd8SSage Weil /* do we need to explicitly request a larger max_size? */ 2074a8599bd8SSage Weil spin_lock(&inode->i_lock); 2075a8599bd8SSage Weil if ((endoff >= ci->i_max_size || 2076a8599bd8SSage Weil endoff > (inode->i_size << 1)) && 2077a8599bd8SSage Weil endoff > ci->i_wanted_max_size) { 2078a8599bd8SSage Weil dout("write %p at large endoff %llu, req max_size\n", 2079a8599bd8SSage Weil inode, endoff); 2080a8599bd8SSage Weil ci->i_wanted_max_size = endoff; 2081a8599bd8SSage Weil check = 1; 2082a8599bd8SSage Weil } 2083a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2084a8599bd8SSage Weil if (check) 2085a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2086a8599bd8SSage Weil } 2087a8599bd8SSage Weil 2088a8599bd8SSage Weil /* 2089a8599bd8SSage Weil * Wait for caps, and take cap references. If we can't get a WR cap 2090a8599bd8SSage Weil * due to a small max_size, make sure we check_max_size (and possibly 2091a8599bd8SSage Weil * ask the mds) so we don't get hung up indefinitely. 2092a8599bd8SSage Weil */ 2093a8599bd8SSage Weil int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got, 2094a8599bd8SSage Weil loff_t endoff) 2095a8599bd8SSage Weil { 2096a8599bd8SSage Weil int check_max, ret, err; 2097a8599bd8SSage Weil 2098a8599bd8SSage Weil retry: 2099a8599bd8SSage Weil if (endoff > 0) 2100a8599bd8SSage Weil check_max_size(&ci->vfs_inode, endoff); 2101a8599bd8SSage Weil check_max = 0; 2102a8599bd8SSage Weil err = 0; 2103a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 2104a8599bd8SSage Weil try_get_cap_refs(ci, need, want, 2105a8599bd8SSage Weil got, endoff, 2106a8599bd8SSage Weil &check_max, &err)); 2107a8599bd8SSage Weil if (err) 2108a8599bd8SSage Weil ret = err; 2109a8599bd8SSage Weil if (check_max) 2110a8599bd8SSage Weil goto retry; 2111a8599bd8SSage Weil return ret; 2112a8599bd8SSage Weil } 2113a8599bd8SSage Weil 2114a8599bd8SSage Weil /* 2115a8599bd8SSage Weil * Take cap refs. Caller must already know we hold at least one ref 2116a8599bd8SSage Weil * on the caps in question or we don't know this is safe. 2117a8599bd8SSage Weil */ 2118a8599bd8SSage Weil void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) 2119a8599bd8SSage Weil { 2120a8599bd8SSage Weil spin_lock(&ci->vfs_inode.i_lock); 2121a8599bd8SSage Weil __take_cap_refs(ci, caps); 2122a8599bd8SSage Weil spin_unlock(&ci->vfs_inode.i_lock); 2123a8599bd8SSage Weil } 2124a8599bd8SSage Weil 2125a8599bd8SSage Weil /* 2126a8599bd8SSage Weil * Release cap refs. 2127a8599bd8SSage Weil * 2128a8599bd8SSage Weil * If we released the last ref on any given cap, call ceph_check_caps 2129a8599bd8SSage Weil * to release (or schedule a release). 2130a8599bd8SSage Weil * 2131a8599bd8SSage Weil * If we are releasing a WR cap (from a sync write), finalize any affected 2132a8599bd8SSage Weil * cap_snap, and wake up any waiters. 2133a8599bd8SSage Weil */ 2134a8599bd8SSage Weil void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) 2135a8599bd8SSage Weil { 2136a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2137a8599bd8SSage Weil int last = 0, put = 0, flushsnaps = 0, wake = 0; 2138a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2139a8599bd8SSage Weil 2140a8599bd8SSage Weil spin_lock(&inode->i_lock); 2141a8599bd8SSage Weil if (had & CEPH_CAP_PIN) 2142a8599bd8SSage Weil --ci->i_pin_ref; 2143a8599bd8SSage Weil if (had & CEPH_CAP_FILE_RD) 2144a8599bd8SSage Weil if (--ci->i_rd_ref == 0) 2145a8599bd8SSage Weil last++; 2146a8599bd8SSage Weil if (had & CEPH_CAP_FILE_CACHE) 2147a8599bd8SSage Weil if (--ci->i_rdcache_ref == 0) 2148a8599bd8SSage Weil last++; 2149a8599bd8SSage Weil if (had & CEPH_CAP_FILE_BUFFER) { 2150a8599bd8SSage Weil if (--ci->i_wrbuffer_ref == 0) { 2151a8599bd8SSage Weil last++; 2152a8599bd8SSage Weil put++; 2153a8599bd8SSage Weil } 2154a8599bd8SSage Weil dout("put_cap_refs %p wrbuffer %d -> %d (?)\n", 2155a8599bd8SSage Weil inode, ci->i_wrbuffer_ref+1, ci->i_wrbuffer_ref); 2156a8599bd8SSage Weil } 2157a8599bd8SSage Weil if (had & CEPH_CAP_FILE_WR) 2158a8599bd8SSage Weil if (--ci->i_wr_ref == 0) { 2159a8599bd8SSage Weil last++; 2160a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) { 2161a8599bd8SSage Weil capsnap = list_first_entry(&ci->i_cap_snaps, 2162a8599bd8SSage Weil struct ceph_cap_snap, 2163a8599bd8SSage Weil ci_item); 2164a8599bd8SSage Weil if (capsnap->writing) { 2165a8599bd8SSage Weil capsnap->writing = 0; 2166a8599bd8SSage Weil flushsnaps = 2167a8599bd8SSage Weil __ceph_finish_cap_snap(ci, 2168a8599bd8SSage Weil capsnap); 2169a8599bd8SSage Weil wake = 1; 2170a8599bd8SSage Weil } 2171a8599bd8SSage Weil } 2172a8599bd8SSage Weil } 2173a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2174a8599bd8SSage Weil 2175819ccbfaSSage Weil dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), 2176819ccbfaSSage Weil last ? " last" : "", put ? " put" : ""); 2177a8599bd8SSage Weil 2178a8599bd8SSage Weil if (last && !flushsnaps) 2179a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 2180a8599bd8SSage Weil else if (flushsnaps) 2181a8599bd8SSage Weil ceph_flush_snaps(ci); 2182a8599bd8SSage Weil if (wake) 218303066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2184a8599bd8SSage Weil if (put) 2185a8599bd8SSage Weil iput(inode); 2186a8599bd8SSage Weil } 2187a8599bd8SSage Weil 2188a8599bd8SSage Weil /* 2189a8599bd8SSage Weil * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap 2190a8599bd8SSage Weil * context. Adjust per-snap dirty page accounting as appropriate. 2191a8599bd8SSage Weil * Once all dirty data for a cap_snap is flushed, flush snapped file 2192a8599bd8SSage Weil * metadata back to the MDS. If we dropped the last ref, call 2193a8599bd8SSage Weil * ceph_check_caps. 2194a8599bd8SSage Weil */ 2195a8599bd8SSage Weil void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, 2196a8599bd8SSage Weil struct ceph_snap_context *snapc) 2197a8599bd8SSage Weil { 2198a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2199a8599bd8SSage Weil int last = 0; 2200819ccbfaSSage Weil int complete_capsnap = 0; 2201819ccbfaSSage Weil int drop_capsnap = 0; 2202a8599bd8SSage Weil int found = 0; 2203a8599bd8SSage Weil struct ceph_cap_snap *capsnap = NULL; 2204a8599bd8SSage Weil 2205a8599bd8SSage Weil spin_lock(&inode->i_lock); 2206a8599bd8SSage Weil ci->i_wrbuffer_ref -= nr; 2207a8599bd8SSage Weil last = !ci->i_wrbuffer_ref; 2208a8599bd8SSage Weil 2209a8599bd8SSage Weil if (ci->i_head_snapc == snapc) { 2210a8599bd8SSage Weil ci->i_wrbuffer_ref_head -= nr; 22117d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0 && 22127d8cb26dSSage Weil ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) { 22137d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 2214a8599bd8SSage Weil ceph_put_snap_context(ci->i_head_snapc); 2215a8599bd8SSage Weil ci->i_head_snapc = NULL; 2216a8599bd8SSage Weil } 2217a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", 2218a8599bd8SSage Weil inode, 2219a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, 2220a8599bd8SSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 2221a8599bd8SSage Weil last ? " LAST" : ""); 2222a8599bd8SSage Weil } else { 2223a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2224a8599bd8SSage Weil if (capsnap->context == snapc) { 2225a8599bd8SSage Weil found = 1; 2226a8599bd8SSage Weil break; 2227a8599bd8SSage Weil } 2228a8599bd8SSage Weil } 2229a8599bd8SSage Weil BUG_ON(!found); 2230819ccbfaSSage Weil capsnap->dirty_pages -= nr; 2231819ccbfaSSage Weil if (capsnap->dirty_pages == 0) { 2232819ccbfaSSage Weil complete_capsnap = 1; 2233819ccbfaSSage Weil if (capsnap->dirty == 0) 2234819ccbfaSSage Weil /* cap writeback completed before we created 2235819ccbfaSSage Weil * the cap_snap; no FLUSHSNAP is needed */ 2236819ccbfaSSage Weil drop_capsnap = 1; 2237819ccbfaSSage Weil } 2238a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p cap_snap %p " 2239819ccbfaSSage Weil " snap %lld %d/%d -> %d/%d %s%s%s\n", 2240a8599bd8SSage Weil inode, capsnap, capsnap->context->seq, 2241a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, 2242a8599bd8SSage Weil ci->i_wrbuffer_ref, capsnap->dirty_pages, 2243a8599bd8SSage Weil last ? " (wrbuffer last)" : "", 2244819ccbfaSSage Weil complete_capsnap ? " (complete capsnap)" : "", 2245819ccbfaSSage Weil drop_capsnap ? " (drop capsnap)" : ""); 2246819ccbfaSSage Weil if (drop_capsnap) { 2247819ccbfaSSage Weil ceph_put_snap_context(capsnap->context); 2248819ccbfaSSage Weil list_del(&capsnap->ci_item); 2249819ccbfaSSage Weil list_del(&capsnap->flushing_item); 2250819ccbfaSSage Weil ceph_put_cap_snap(capsnap); 2251819ccbfaSSage Weil } 2252a8599bd8SSage Weil } 2253a8599bd8SSage Weil 2254a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2255a8599bd8SSage Weil 2256a8599bd8SSage Weil if (last) { 2257a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2258a8599bd8SSage Weil iput(inode); 2259819ccbfaSSage Weil } else if (complete_capsnap) { 2260a8599bd8SSage Weil ceph_flush_snaps(ci); 226103066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2262a8599bd8SSage Weil } 2263819ccbfaSSage Weil if (drop_capsnap) 2264819ccbfaSSage Weil iput(inode); 2265a8599bd8SSage Weil } 2266a8599bd8SSage Weil 2267a8599bd8SSage Weil /* 2268a8599bd8SSage Weil * Handle a cap GRANT message from the MDS. (Note that a GRANT may 2269a8599bd8SSage Weil * actually be a revocation if it specifies a smaller cap set.) 2270a8599bd8SSage Weil * 227115637c8bSSage Weil * caller holds s_mutex and i_lock, we drop both. 227215637c8bSSage Weil * 2273a8599bd8SSage Weil * return value: 2274a8599bd8SSage Weil * 0 - ok 2275a8599bd8SSage Weil * 1 - check_caps on auth cap only (writeback) 2276a8599bd8SSage Weil * 2 - check_caps (ack revoke) 2277a8599bd8SSage Weil */ 227815637c8bSSage Weil static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant, 2279a8599bd8SSage Weil struct ceph_mds_session *session, 2280a8599bd8SSage Weil struct ceph_cap *cap, 2281a8599bd8SSage Weil struct ceph_buffer *xattr_buf) 2282a8599bd8SSage Weil __releases(inode->i_lock) 2283a8599bd8SSage Weil { 2284a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2285a8599bd8SSage Weil int mds = session->s_mds; 2286a8599bd8SSage Weil int seq = le32_to_cpu(grant->seq); 2287a8599bd8SSage Weil int newcaps = le32_to_cpu(grant->caps); 2288a8599bd8SSage Weil int issued, implemented, used, wanted, dirty; 2289a8599bd8SSage Weil u64 size = le64_to_cpu(grant->size); 2290a8599bd8SSage Weil u64 max_size = le64_to_cpu(grant->max_size); 2291a8599bd8SSage Weil struct timespec mtime, atime, ctime; 229215637c8bSSage Weil int check_caps = 0; 2293a8599bd8SSage Weil int wake = 0; 2294a8599bd8SSage Weil int writeback = 0; 2295a8599bd8SSage Weil int revoked_rdcache = 0; 22963c6f6b79SSage Weil int queue_invalidate = 0; 2297a8599bd8SSage Weil 2298a8599bd8SSage Weil dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", 2299a8599bd8SSage Weil inode, cap, mds, seq, ceph_cap_string(newcaps)); 2300a8599bd8SSage Weil dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, 2301a8599bd8SSage Weil inode->i_size); 2302a8599bd8SSage Weil 2303a8599bd8SSage Weil /* 2304a8599bd8SSage Weil * If CACHE is being revoked, and we have no dirty buffers, 2305a8599bd8SSage Weil * try to invalidate (once). (If there are dirty buffers, we 2306a8599bd8SSage Weil * will invalidate _after_ writeback.) 2307a8599bd8SSage Weil */ 23083b454c49SSage Weil if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && 23093b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 2310bcd2cbd1SYehuda Sadeh !ci->i_wrbuffer_ref) { 23115ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) == 0) { 23125ecad6fdSSage Weil revoked_rdcache = 1; 23135ecad6fdSSage Weil } else { 2314a8599bd8SSage Weil /* there were locked pages.. invalidate later 2315a8599bd8SSage Weil in a separate thread. */ 2316a8599bd8SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 23173c6f6b79SSage Weil queue_invalidate = 1; 2318a8599bd8SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 2319a8599bd8SSage Weil } 2320a8599bd8SSage Weil } 2321a8599bd8SSage Weil } 2322a8599bd8SSage Weil 2323a8599bd8SSage Weil /* side effects now are allowed */ 2324a8599bd8SSage Weil 2325a8599bd8SSage Weil issued = __ceph_caps_issued(ci, &implemented); 2326a8599bd8SSage Weil issued |= implemented | __ceph_caps_dirty(ci); 2327a8599bd8SSage Weil 2328685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 2329a8599bd8SSage Weil 2330a8599bd8SSage Weil __check_cap_issue(ci, cap, newcaps); 2331a8599bd8SSage Weil 2332a8599bd8SSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 2333a8599bd8SSage Weil inode->i_mode = le32_to_cpu(grant->mode); 2334a8599bd8SSage Weil inode->i_uid = le32_to_cpu(grant->uid); 2335a8599bd8SSage Weil inode->i_gid = le32_to_cpu(grant->gid); 2336a8599bd8SSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 2337a8599bd8SSage Weil inode->i_uid, inode->i_gid); 2338a8599bd8SSage Weil } 2339a8599bd8SSage Weil 2340a8599bd8SSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 2341a8599bd8SSage Weil inode->i_nlink = le32_to_cpu(grant->nlink); 2342a8599bd8SSage Weil 2343a8599bd8SSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) { 2344a8599bd8SSage Weil int len = le32_to_cpu(grant->xattr_len); 2345a8599bd8SSage Weil u64 version = le64_to_cpu(grant->xattr_version); 2346a8599bd8SSage Weil 2347a8599bd8SSage Weil if (version > ci->i_xattrs.version) { 2348a8599bd8SSage Weil dout(" got new xattrs v%llu on %p len %d\n", 2349a8599bd8SSage Weil version, inode, len); 2350a8599bd8SSage Weil if (ci->i_xattrs.blob) 2351a8599bd8SSage Weil ceph_buffer_put(ci->i_xattrs.blob); 2352a8599bd8SSage Weil ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); 2353a8599bd8SSage Weil ci->i_xattrs.version = version; 2354a8599bd8SSage Weil } 2355a8599bd8SSage Weil } 2356a8599bd8SSage Weil 2357a8599bd8SSage Weil /* size/ctime/mtime/atime? */ 2358a8599bd8SSage Weil ceph_fill_file_size(inode, issued, 2359a8599bd8SSage Weil le32_to_cpu(grant->truncate_seq), 2360a8599bd8SSage Weil le64_to_cpu(grant->truncate_size), size); 2361a8599bd8SSage Weil ceph_decode_timespec(&mtime, &grant->mtime); 2362a8599bd8SSage Weil ceph_decode_timespec(&atime, &grant->atime); 2363a8599bd8SSage Weil ceph_decode_timespec(&ctime, &grant->ctime); 2364a8599bd8SSage Weil ceph_fill_file_time(inode, issued, 2365a8599bd8SSage Weil le32_to_cpu(grant->time_warp_seq), &ctime, &mtime, 2366a8599bd8SSage Weil &atime); 2367a8599bd8SSage Weil 2368a8599bd8SSage Weil /* max size increase? */ 2369a8599bd8SSage Weil if (max_size != ci->i_max_size) { 2370a8599bd8SSage Weil dout("max_size %lld -> %llu\n", ci->i_max_size, max_size); 2371a8599bd8SSage Weil ci->i_max_size = max_size; 2372a8599bd8SSage Weil if (max_size >= ci->i_wanted_max_size) { 2373a8599bd8SSage Weil ci->i_wanted_max_size = 0; /* reset */ 2374a8599bd8SSage Weil ci->i_requested_max_size = 0; 2375a8599bd8SSage Weil } 2376a8599bd8SSage Weil wake = 1; 2377a8599bd8SSage Weil } 2378a8599bd8SSage Weil 2379a8599bd8SSage Weil /* check cap bits */ 2380a8599bd8SSage Weil wanted = __ceph_caps_wanted(ci); 2381a8599bd8SSage Weil used = __ceph_caps_used(ci); 2382a8599bd8SSage Weil dirty = __ceph_caps_dirty(ci); 2383a8599bd8SSage Weil dout(" my wanted = %s, used = %s, dirty %s\n", 2384a8599bd8SSage Weil ceph_cap_string(wanted), 2385a8599bd8SSage Weil ceph_cap_string(used), 2386a8599bd8SSage Weil ceph_cap_string(dirty)); 2387a8599bd8SSage Weil if (wanted != le32_to_cpu(grant->wanted)) { 2388a8599bd8SSage Weil dout("mds wanted %s -> %s\n", 2389a8599bd8SSage Weil ceph_cap_string(le32_to_cpu(grant->wanted)), 2390a8599bd8SSage Weil ceph_cap_string(wanted)); 2391a8599bd8SSage Weil grant->wanted = cpu_to_le32(wanted); 2392a8599bd8SSage Weil } 2393a8599bd8SSage Weil 2394a8599bd8SSage Weil cap->seq = seq; 2395a8599bd8SSage Weil 2396a8599bd8SSage Weil /* file layout may have changed */ 2397a8599bd8SSage Weil ci->i_layout = grant->layout; 2398a8599bd8SSage Weil 2399a8599bd8SSage Weil /* revocation, grant, or no-op? */ 2400a8599bd8SSage Weil if (cap->issued & ~newcaps) { 24013b454c49SSage Weil int revoking = cap->issued & ~newcaps; 24023b454c49SSage Weil 24033b454c49SSage Weil dout("revocation: %s -> %s (revoking %s)\n", 24043b454c49SSage Weil ceph_cap_string(cap->issued), 24053b454c49SSage Weil ceph_cap_string(newcaps), 24063b454c49SSage Weil ceph_cap_string(revoking)); 24070eb6cd49SSage Weil if (revoking & used & CEPH_CAP_FILE_BUFFER) 24083b454c49SSage Weil writeback = 1; /* initiate writeback; will delay ack */ 24093b454c49SSage Weil else if (revoking == CEPH_CAP_FILE_CACHE && 24103b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 24113b454c49SSage Weil queue_invalidate) 24123b454c49SSage Weil ; /* do nothing yet, invalidation will be queued */ 24133b454c49SSage Weil else if (cap == ci->i_auth_cap) 24143b454c49SSage Weil check_caps = 1; /* check auth cap only */ 24153b454c49SSage Weil else 24163b454c49SSage Weil check_caps = 2; /* check all caps */ 2417a8599bd8SSage Weil cap->issued = newcaps; 2418978097c9SSage Weil cap->implemented |= newcaps; 2419a8599bd8SSage Weil } else if (cap->issued == newcaps) { 2420a8599bd8SSage Weil dout("caps unchanged: %s -> %s\n", 2421a8599bd8SSage Weil ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); 2422a8599bd8SSage Weil } else { 2423a8599bd8SSage Weil dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), 2424a8599bd8SSage Weil ceph_cap_string(newcaps)); 2425a8599bd8SSage Weil cap->issued = newcaps; 2426a8599bd8SSage Weil cap->implemented |= newcaps; /* add bits only, to 2427a8599bd8SSage Weil * avoid stepping on a 2428a8599bd8SSage Weil * pending revocation */ 2429a8599bd8SSage Weil wake = 1; 2430a8599bd8SSage Weil } 2431978097c9SSage Weil BUG_ON(cap->issued & ~cap->implemented); 2432a8599bd8SSage Weil 2433a8599bd8SSage Weil spin_unlock(&inode->i_lock); 24343c6f6b79SSage Weil if (writeback) 2435a8599bd8SSage Weil /* 2436a8599bd8SSage Weil * queue inode for writeback: we can't actually call 2437a8599bd8SSage Weil * filemap_write_and_wait, etc. from message handler 2438a8599bd8SSage Weil * context. 2439a8599bd8SSage Weil */ 24403c6f6b79SSage Weil ceph_queue_writeback(inode); 24413c6f6b79SSage Weil if (queue_invalidate) 24423c6f6b79SSage Weil ceph_queue_invalidate(inode); 2443a8599bd8SSage Weil if (wake) 244403066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 244515637c8bSSage Weil 244615637c8bSSage Weil if (check_caps == 1) 244715637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, 244815637c8bSSage Weil session); 244915637c8bSSage Weil else if (check_caps == 2) 245015637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); 245115637c8bSSage Weil else 245215637c8bSSage Weil mutex_unlock(&session->s_mutex); 2453a8599bd8SSage Weil } 2454a8599bd8SSage Weil 2455a8599bd8SSage Weil /* 2456a8599bd8SSage Weil * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the 2457a8599bd8SSage Weil * MDS has been safely committed. 2458a8599bd8SSage Weil */ 24596df058c0SSage Weil static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, 2460a8599bd8SSage Weil struct ceph_mds_caps *m, 2461a8599bd8SSage Weil struct ceph_mds_session *session, 2462a8599bd8SSage Weil struct ceph_cap *cap) 2463a8599bd8SSage Weil __releases(inode->i_lock) 2464a8599bd8SSage Weil { 2465a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2466640ef79dSCheng Renquan struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc; 2467a8599bd8SSage Weil unsigned seq = le32_to_cpu(m->seq); 2468a8599bd8SSage Weil int dirty = le32_to_cpu(m->dirty); 2469a8599bd8SSage Weil int cleaned = 0; 2470afcdaea3SSage Weil int drop = 0; 2471a8599bd8SSage Weil int i; 2472a8599bd8SSage Weil 2473a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 2474a8599bd8SSage Weil if ((dirty & (1 << i)) && 2475a8599bd8SSage Weil flush_tid == ci->i_cap_flush_tid[i]) 2476a8599bd8SSage Weil cleaned |= 1 << i; 2477a8599bd8SSage Weil 2478a8599bd8SSage Weil dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," 2479a8599bd8SSage Weil " flushing %s -> %s\n", 2480a8599bd8SSage Weil inode, session->s_mds, seq, ceph_cap_string(dirty), 2481a8599bd8SSage Weil ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), 2482a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps & ~cleaned)); 2483a8599bd8SSage Weil 2484a8599bd8SSage Weil if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned)) 2485a8599bd8SSage Weil goto out; 2486a8599bd8SSage Weil 2487a8599bd8SSage Weil ci->i_flushing_caps &= ~cleaned; 2488a8599bd8SSage Weil 2489a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2490a8599bd8SSage Weil if (ci->i_flushing_caps == 0) { 2491a8599bd8SSage Weil list_del_init(&ci->i_flushing_item); 2492a8599bd8SSage Weil if (!list_empty(&session->s_cap_flushing)) 2493a8599bd8SSage Weil dout(" mds%d still flushing cap on %p\n", 2494a8599bd8SSage Weil session->s_mds, 2495a8599bd8SSage Weil &list_entry(session->s_cap_flushing.next, 2496a8599bd8SSage Weil struct ceph_inode_info, 2497a8599bd8SSage Weil i_flushing_item)->vfs_inode); 2498a8599bd8SSage Weil mdsc->num_cap_flushing--; 249903066f23SYehuda Sadeh wake_up_all(&mdsc->cap_flushing_wq); 2500a8599bd8SSage Weil dout(" inode %p now !flushing\n", inode); 2501afcdaea3SSage Weil 2502afcdaea3SSage Weil if (ci->i_dirty_caps == 0) { 2503a8599bd8SSage Weil dout(" inode %p now clean\n", inode); 2504afcdaea3SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 2505afcdaea3SSage Weil drop = 1; 25067d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0) { 25077d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 25087d8cb26dSSage Weil ceph_put_snap_context(ci->i_head_snapc); 25097d8cb26dSSage Weil ci->i_head_snapc = NULL; 25107d8cb26dSSage Weil } 251176e3b390SSage Weil } else { 251276e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 2513afcdaea3SSage Weil } 2514a8599bd8SSage Weil } 2515a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 251603066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2517a8599bd8SSage Weil 2518a8599bd8SSage Weil out: 2519a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2520afcdaea3SSage Weil if (drop) 2521a8599bd8SSage Weil iput(inode); 2522a8599bd8SSage Weil } 2523a8599bd8SSage Weil 2524a8599bd8SSage Weil /* 2525a8599bd8SSage Weil * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can 2526a8599bd8SSage Weil * throw away our cap_snap. 2527a8599bd8SSage Weil * 2528a8599bd8SSage Weil * Caller hold s_mutex. 2529a8599bd8SSage Weil */ 25306df058c0SSage Weil static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, 2531a8599bd8SSage Weil struct ceph_mds_caps *m, 2532a8599bd8SSage Weil struct ceph_mds_session *session) 2533a8599bd8SSage Weil { 2534a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2535a8599bd8SSage Weil u64 follows = le64_to_cpu(m->snap_follows); 2536a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2537a8599bd8SSage Weil int drop = 0; 2538a8599bd8SSage Weil 2539a8599bd8SSage Weil dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", 2540a8599bd8SSage Weil inode, ci, session->s_mds, follows); 2541a8599bd8SSage Weil 2542a8599bd8SSage Weil spin_lock(&inode->i_lock); 2543a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2544a8599bd8SSage Weil if (capsnap->follows == follows) { 2545a8599bd8SSage Weil if (capsnap->flush_tid != flush_tid) { 2546a8599bd8SSage Weil dout(" cap_snap %p follows %lld tid %lld !=" 2547a8599bd8SSage Weil " %lld\n", capsnap, follows, 2548a8599bd8SSage Weil flush_tid, capsnap->flush_tid); 2549a8599bd8SSage Weil break; 2550a8599bd8SSage Weil } 2551a8599bd8SSage Weil WARN_ON(capsnap->dirty_pages || capsnap->writing); 2552819ccbfaSSage Weil dout(" removing %p cap_snap %p follows %lld\n", 2553819ccbfaSSage Weil inode, capsnap, follows); 2554a8599bd8SSage Weil ceph_put_snap_context(capsnap->context); 2555a8599bd8SSage Weil list_del(&capsnap->ci_item); 2556a8599bd8SSage Weil list_del(&capsnap->flushing_item); 2557a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 2558a8599bd8SSage Weil drop = 1; 2559a8599bd8SSage Weil break; 2560a8599bd8SSage Weil } else { 2561a8599bd8SSage Weil dout(" skipping cap_snap %p follows %lld\n", 2562a8599bd8SSage Weil capsnap, capsnap->follows); 2563a8599bd8SSage Weil } 2564a8599bd8SSage Weil } 2565a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2566a8599bd8SSage Weil if (drop) 2567a8599bd8SSage Weil iput(inode); 2568a8599bd8SSage Weil } 2569a8599bd8SSage Weil 2570a8599bd8SSage Weil /* 2571a8599bd8SSage Weil * Handle TRUNC from MDS, indicating file truncation. 2572a8599bd8SSage Weil * 2573a8599bd8SSage Weil * caller hold s_mutex. 2574a8599bd8SSage Weil */ 2575a8599bd8SSage Weil static void handle_cap_trunc(struct inode *inode, 2576a8599bd8SSage Weil struct ceph_mds_caps *trunc, 2577a8599bd8SSage Weil struct ceph_mds_session *session) 2578a8599bd8SSage Weil __releases(inode->i_lock) 2579a8599bd8SSage Weil { 2580a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2581a8599bd8SSage Weil int mds = session->s_mds; 2582a8599bd8SSage Weil int seq = le32_to_cpu(trunc->seq); 2583a8599bd8SSage Weil u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); 2584a8599bd8SSage Weil u64 truncate_size = le64_to_cpu(trunc->truncate_size); 2585a8599bd8SSage Weil u64 size = le64_to_cpu(trunc->size); 2586a8599bd8SSage Weil int implemented = 0; 2587a8599bd8SSage Weil int dirty = __ceph_caps_dirty(ci); 2588a8599bd8SSage Weil int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); 2589a8599bd8SSage Weil int queue_trunc = 0; 2590a8599bd8SSage Weil 2591a8599bd8SSage Weil issued |= implemented | dirty; 2592a8599bd8SSage Weil 2593a8599bd8SSage Weil dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", 2594a8599bd8SSage Weil inode, mds, seq, truncate_size, truncate_seq); 2595a8599bd8SSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 2596a8599bd8SSage Weil truncate_seq, truncate_size, size); 2597a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2598a8599bd8SSage Weil 2599a8599bd8SSage Weil if (queue_trunc) 26003c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 2601a8599bd8SSage Weil } 2602a8599bd8SSage Weil 2603a8599bd8SSage Weil /* 2604a8599bd8SSage Weil * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a 2605a8599bd8SSage Weil * different one. If we are the most recent migration we've seen (as 2606a8599bd8SSage Weil * indicated by mseq), make note of the migrating cap bits for the 2607a8599bd8SSage Weil * duration (until we see the corresponding IMPORT). 2608a8599bd8SSage Weil * 2609a8599bd8SSage Weil * caller holds s_mutex 2610a8599bd8SSage Weil */ 2611a8599bd8SSage Weil static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, 2612154f42c2SSage Weil struct ceph_mds_session *session, 2613154f42c2SSage Weil int *open_target_sessions) 2614a8599bd8SSage Weil { 2615a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2616a8599bd8SSage Weil int mds = session->s_mds; 2617a8599bd8SSage Weil unsigned mseq = le32_to_cpu(ex->migrate_seq); 2618a8599bd8SSage Weil struct ceph_cap *cap = NULL, *t; 2619a8599bd8SSage Weil struct rb_node *p; 2620a8599bd8SSage Weil int remember = 1; 2621a8599bd8SSage Weil 2622a8599bd8SSage Weil dout("handle_cap_export inode %p ci %p mds%d mseq %d\n", 2623a8599bd8SSage Weil inode, ci, mds, mseq); 2624a8599bd8SSage Weil 2625a8599bd8SSage Weil spin_lock(&inode->i_lock); 2626a8599bd8SSage Weil 2627a8599bd8SSage Weil /* make sure we haven't seen a higher mseq */ 2628a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 2629a8599bd8SSage Weil t = rb_entry(p, struct ceph_cap, ci_node); 2630a8599bd8SSage Weil if (ceph_seq_cmp(t->mseq, mseq) > 0) { 2631a8599bd8SSage Weil dout(" higher mseq on cap from mds%d\n", 2632a8599bd8SSage Weil t->session->s_mds); 2633a8599bd8SSage Weil remember = 0; 2634a8599bd8SSage Weil } 2635a8599bd8SSage Weil if (t->session->s_mds == mds) 2636a8599bd8SSage Weil cap = t; 2637a8599bd8SSage Weil } 2638a8599bd8SSage Weil 2639a8599bd8SSage Weil if (cap) { 2640a8599bd8SSage Weil if (remember) { 2641a8599bd8SSage Weil /* make note */ 2642a8599bd8SSage Weil ci->i_cap_exporting_mds = mds; 2643a8599bd8SSage Weil ci->i_cap_exporting_mseq = mseq; 2644a8599bd8SSage Weil ci->i_cap_exporting_issued = cap->issued; 2645154f42c2SSage Weil 2646154f42c2SSage Weil /* 2647154f42c2SSage Weil * make sure we have open sessions with all possible 2648154f42c2SSage Weil * export targets, so that we get the matching IMPORT 2649154f42c2SSage Weil */ 2650154f42c2SSage Weil *open_target_sessions = 1; 2651a8599bd8SSage Weil } 26527c1332b8SSage Weil __ceph_remove_cap(cap); 2653a8599bd8SSage Weil } 26544ea0043aSSage Weil /* else, we already released it */ 2655a8599bd8SSage Weil 2656a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2657a8599bd8SSage Weil } 2658a8599bd8SSage Weil 2659a8599bd8SSage Weil /* 2660a8599bd8SSage Weil * Handle cap IMPORT. If there are temp bits from an older EXPORT, 2661a8599bd8SSage Weil * clean them up. 2662a8599bd8SSage Weil * 2663a8599bd8SSage Weil * caller holds s_mutex. 2664a8599bd8SSage Weil */ 2665a8599bd8SSage Weil static void handle_cap_import(struct ceph_mds_client *mdsc, 2666a8599bd8SSage Weil struct inode *inode, struct ceph_mds_caps *im, 2667a8599bd8SSage Weil struct ceph_mds_session *session, 2668a8599bd8SSage Weil void *snaptrace, int snaptrace_len) 2669a8599bd8SSage Weil { 2670a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2671a8599bd8SSage Weil int mds = session->s_mds; 2672a8599bd8SSage Weil unsigned issued = le32_to_cpu(im->caps); 2673a8599bd8SSage Weil unsigned wanted = le32_to_cpu(im->wanted); 2674a8599bd8SSage Weil unsigned seq = le32_to_cpu(im->seq); 2675a8599bd8SSage Weil unsigned mseq = le32_to_cpu(im->migrate_seq); 2676a8599bd8SSage Weil u64 realmino = le64_to_cpu(im->realm); 2677a8599bd8SSage Weil u64 cap_id = le64_to_cpu(im->cap_id); 2678a8599bd8SSage Weil 2679a8599bd8SSage Weil if (ci->i_cap_exporting_mds >= 0 && 2680a8599bd8SSage Weil ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) { 2681a8599bd8SSage Weil dout("handle_cap_import inode %p ci %p mds%d mseq %d" 2682a8599bd8SSage Weil " - cleared exporting from mds%d\n", 2683a8599bd8SSage Weil inode, ci, mds, mseq, 2684a8599bd8SSage Weil ci->i_cap_exporting_mds); 2685a8599bd8SSage Weil ci->i_cap_exporting_issued = 0; 2686a8599bd8SSage Weil ci->i_cap_exporting_mseq = 0; 2687a8599bd8SSage Weil ci->i_cap_exporting_mds = -1; 2688a8599bd8SSage Weil } else { 2689a8599bd8SSage Weil dout("handle_cap_import inode %p ci %p mds%d mseq %d\n", 2690a8599bd8SSage Weil inode, ci, mds, mseq); 2691a8599bd8SSage Weil } 2692a8599bd8SSage Weil 2693a8599bd8SSage Weil down_write(&mdsc->snap_rwsem); 2694a8599bd8SSage Weil ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len, 2695a8599bd8SSage Weil false); 2696a8599bd8SSage Weil downgrade_write(&mdsc->snap_rwsem); 2697a8599bd8SSage Weil ceph_add_cap(inode, session, cap_id, -1, 2698a8599bd8SSage Weil issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH, 2699a8599bd8SSage Weil NULL /* no caps context */); 2700a8599bd8SSage Weil try_flush_caps(inode, session, NULL); 2701a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 2702a8599bd8SSage Weil } 2703a8599bd8SSage Weil 2704a8599bd8SSage Weil /* 2705a8599bd8SSage Weil * Handle a caps message from the MDS. 2706a8599bd8SSage Weil * 2707a8599bd8SSage Weil * Identify the appropriate session, inode, and call the right handler 2708a8599bd8SSage Weil * based on the cap op. 2709a8599bd8SSage Weil */ 2710a8599bd8SSage Weil void ceph_handle_caps(struct ceph_mds_session *session, 2711a8599bd8SSage Weil struct ceph_msg *msg) 2712a8599bd8SSage Weil { 2713a8599bd8SSage Weil struct ceph_mds_client *mdsc = session->s_mdsc; 2714a8599bd8SSage Weil struct super_block *sb = mdsc->client->sb; 2715a8599bd8SSage Weil struct inode *inode; 2716a8599bd8SSage Weil struct ceph_cap *cap; 2717a8599bd8SSage Weil struct ceph_mds_caps *h; 27182600d2ddSSage Weil int mds = session->s_mds; 2719a8599bd8SSage Weil int op; 27203d7ded4dSSage Weil u32 seq, mseq; 2721a8599bd8SSage Weil struct ceph_vino vino; 2722a8599bd8SSage Weil u64 cap_id; 2723a8599bd8SSage Weil u64 size, max_size; 27246df058c0SSage Weil u64 tid; 272570edb55bSSage Weil void *snaptrace; 2726ce1fbc8dSSage Weil size_t snaptrace_len; 2727ce1fbc8dSSage Weil void *flock; 2728ce1fbc8dSSage Weil u32 flock_len; 2729154f42c2SSage Weil int open_target_sessions = 0; 2730a8599bd8SSage Weil 2731a8599bd8SSage Weil dout("handle_caps from mds%d\n", mds); 2732a8599bd8SSage Weil 2733a8599bd8SSage Weil /* decode */ 27346df058c0SSage Weil tid = le64_to_cpu(msg->hdr.tid); 2735a8599bd8SSage Weil if (msg->front.iov_len < sizeof(*h)) 2736a8599bd8SSage Weil goto bad; 2737a8599bd8SSage Weil h = msg->front.iov_base; 2738a8599bd8SSage Weil op = le32_to_cpu(h->op); 2739a8599bd8SSage Weil vino.ino = le64_to_cpu(h->ino); 2740a8599bd8SSage Weil vino.snap = CEPH_NOSNAP; 2741a8599bd8SSage Weil cap_id = le64_to_cpu(h->cap_id); 2742a8599bd8SSage Weil seq = le32_to_cpu(h->seq); 27433d7ded4dSSage Weil mseq = le32_to_cpu(h->migrate_seq); 2744a8599bd8SSage Weil size = le64_to_cpu(h->size); 2745a8599bd8SSage Weil max_size = le64_to_cpu(h->max_size); 2746a8599bd8SSage Weil 2747ce1fbc8dSSage Weil snaptrace = h + 1; 2748ce1fbc8dSSage Weil snaptrace_len = le32_to_cpu(h->snap_trace_len); 2749ce1fbc8dSSage Weil 2750ce1fbc8dSSage Weil if (le16_to_cpu(msg->hdr.version) >= 2) { 2751ce1fbc8dSSage Weil void *p, *end; 2752ce1fbc8dSSage Weil 2753ce1fbc8dSSage Weil p = snaptrace + snaptrace_len; 2754ce1fbc8dSSage Weil end = msg->front.iov_base + msg->front.iov_len; 2755ce1fbc8dSSage Weil ceph_decode_32_safe(&p, end, flock_len, bad); 2756ce1fbc8dSSage Weil flock = p; 2757ce1fbc8dSSage Weil } else { 2758ce1fbc8dSSage Weil flock = NULL; 2759ce1fbc8dSSage Weil flock_len = 0; 2760ce1fbc8dSSage Weil } 2761ce1fbc8dSSage Weil 2762a8599bd8SSage Weil mutex_lock(&session->s_mutex); 2763a8599bd8SSage Weil session->s_seq++; 2764a8599bd8SSage Weil dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, 2765a8599bd8SSage Weil (unsigned)seq); 2766a8599bd8SSage Weil 2767a8599bd8SSage Weil /* lookup ino */ 2768a8599bd8SSage Weil inode = ceph_find_inode(sb, vino); 2769a8599bd8SSage Weil dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, 2770a8599bd8SSage Weil vino.snap, inode); 2771a8599bd8SSage Weil if (!inode) { 2772a8599bd8SSage Weil dout(" i don't have ino %llx\n", vino.ino); 27733d7ded4dSSage Weil 27743d7ded4dSSage Weil if (op == CEPH_CAP_OP_IMPORT) 27753d7ded4dSSage Weil __queue_cap_release(session, vino.ino, cap_id, 27763d7ded4dSSage Weil mseq, seq); 27773d7ded4dSSage Weil 27783d7ded4dSSage Weil /* 27793d7ded4dSSage Weil * send any full release message to try to move things 27803d7ded4dSSage Weil * along for the mds (who clearly thinks we still have this 27813d7ded4dSSage Weil * cap). 27823d7ded4dSSage Weil */ 2783ee6b272bSSage Weil ceph_add_cap_releases(mdsc, session); 27843d7ded4dSSage Weil ceph_send_cap_releases(mdsc, session); 2785a8599bd8SSage Weil goto done; 2786a8599bd8SSage Weil } 2787a8599bd8SSage Weil 2788a8599bd8SSage Weil /* these will work even if we don't have a cap yet */ 2789a8599bd8SSage Weil switch (op) { 2790a8599bd8SSage Weil case CEPH_CAP_OP_FLUSHSNAP_ACK: 27916df058c0SSage Weil handle_cap_flushsnap_ack(inode, tid, h, session); 2792a8599bd8SSage Weil goto done; 2793a8599bd8SSage Weil 2794a8599bd8SSage Weil case CEPH_CAP_OP_EXPORT: 2795154f42c2SSage Weil handle_cap_export(inode, h, session, &open_target_sessions); 2796a8599bd8SSage Weil goto done; 2797a8599bd8SSage Weil 2798a8599bd8SSage Weil case CEPH_CAP_OP_IMPORT: 2799a8599bd8SSage Weil handle_cap_import(mdsc, inode, h, session, 2800ce1fbc8dSSage Weil snaptrace, snaptrace_len); 280115637c8bSSage Weil ceph_check_caps(ceph_inode(inode), CHECK_CAPS_NODELAY, 280215637c8bSSage Weil session); 280315637c8bSSage Weil goto done_unlocked; 2804a8599bd8SSage Weil } 2805a8599bd8SSage Weil 2806a8599bd8SSage Weil /* the rest require a cap */ 2807a8599bd8SSage Weil spin_lock(&inode->i_lock); 2808a8599bd8SSage Weil cap = __get_cap_for_mds(ceph_inode(inode), mds); 2809a8599bd8SSage Weil if (!cap) { 28109dbd412fSSage Weil dout(" no cap on %p ino %llx.%llx from mds%d\n", 2811a8599bd8SSage Weil inode, ceph_ino(inode), ceph_snap(inode), mds); 2812a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2813a8599bd8SSage Weil goto done; 2814a8599bd8SSage Weil } 2815a8599bd8SSage Weil 2816a8599bd8SSage Weil /* note that each of these drops i_lock for us */ 2817a8599bd8SSage Weil switch (op) { 2818a8599bd8SSage Weil case CEPH_CAP_OP_REVOKE: 2819a8599bd8SSage Weil case CEPH_CAP_OP_GRANT: 282015637c8bSSage Weil handle_cap_grant(inode, h, session, cap, msg->middle); 282115637c8bSSage Weil goto done_unlocked; 2822a8599bd8SSage Weil 2823a8599bd8SSage Weil case CEPH_CAP_OP_FLUSH_ACK: 28246df058c0SSage Weil handle_cap_flush_ack(inode, tid, h, session, cap); 2825a8599bd8SSage Weil break; 2826a8599bd8SSage Weil 2827a8599bd8SSage Weil case CEPH_CAP_OP_TRUNC: 2828a8599bd8SSage Weil handle_cap_trunc(inode, h, session); 2829a8599bd8SSage Weil break; 2830a8599bd8SSage Weil 2831a8599bd8SSage Weil default: 2832a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2833a8599bd8SSage Weil pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, 2834a8599bd8SSage Weil ceph_cap_op_name(op)); 2835a8599bd8SSage Weil } 2836a8599bd8SSage Weil 2837a8599bd8SSage Weil done: 2838a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 283915637c8bSSage Weil done_unlocked: 2840a8599bd8SSage Weil if (inode) 2841a8599bd8SSage Weil iput(inode); 2842154f42c2SSage Weil if (open_target_sessions) 2843154f42c2SSage Weil ceph_mdsc_open_export_target_sessions(mdsc, session); 2844a8599bd8SSage Weil return; 2845a8599bd8SSage Weil 2846a8599bd8SSage Weil bad: 2847a8599bd8SSage Weil pr_err("ceph_handle_caps: corrupt message\n"); 28489ec7cab1SSage Weil ceph_msg_dump(msg); 2849a8599bd8SSage Weil return; 2850a8599bd8SSage Weil } 2851a8599bd8SSage Weil 2852a8599bd8SSage Weil /* 2853a8599bd8SSage Weil * Delayed work handler to process end of delayed cap release LRU list. 2854a8599bd8SSage Weil */ 2855afcdaea3SSage Weil void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) 2856a8599bd8SSage Weil { 2857a8599bd8SSage Weil struct ceph_inode_info *ci; 2858a8599bd8SSage Weil int flags = CHECK_CAPS_NODELAY; 2859a8599bd8SSage Weil 2860a8599bd8SSage Weil dout("check_delayed_caps\n"); 2861a8599bd8SSage Weil while (1) { 2862a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 2863a8599bd8SSage Weil if (list_empty(&mdsc->cap_delay_list)) 2864a8599bd8SSage Weil break; 2865a8599bd8SSage Weil ci = list_first_entry(&mdsc->cap_delay_list, 2866a8599bd8SSage Weil struct ceph_inode_info, 2867a8599bd8SSage Weil i_cap_delay_list); 2868a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && 2869a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) 2870a8599bd8SSage Weil break; 2871a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 2872a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 2873a8599bd8SSage Weil dout("check_delayed_caps on %p\n", &ci->vfs_inode); 2874a8599bd8SSage Weil ceph_check_caps(ci, flags, NULL); 2875a8599bd8SSage Weil } 2876a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 2877a8599bd8SSage Weil } 2878a8599bd8SSage Weil 2879a8599bd8SSage Weil /* 2880afcdaea3SSage Weil * Flush all dirty caps to the mds 2881afcdaea3SSage Weil */ 2882afcdaea3SSage Weil void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) 2883afcdaea3SSage Weil { 2884e9964c10SSage Weil struct ceph_inode_info *ci, *nci = NULL; 2885e9964c10SSage Weil struct inode *inode, *ninode = NULL; 2886e9964c10SSage Weil struct list_head *p, *n; 2887afcdaea3SSage Weil 2888afcdaea3SSage Weil dout("flush_dirty_caps\n"); 2889afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2890e9964c10SSage Weil list_for_each_safe(p, n, &mdsc->cap_dirty) { 2891e9964c10SSage Weil if (nci) { 2892e9964c10SSage Weil ci = nci; 2893e9964c10SSage Weil inode = ninode; 2894e9964c10SSage Weil ci->i_ceph_flags &= ~CEPH_I_NOFLUSH; 2895e9964c10SSage Weil dout("flush_dirty_caps inode %p (was next inode)\n", 2896e9964c10SSage Weil inode); 2897e9964c10SSage Weil } else { 2898e9964c10SSage Weil ci = list_entry(p, struct ceph_inode_info, 2899afcdaea3SSage Weil i_dirty_item); 2900afcdaea3SSage Weil inode = igrab(&ci->vfs_inode); 2901e9964c10SSage Weil BUG_ON(!inode); 2902e9964c10SSage Weil dout("flush_dirty_caps inode %p\n", inode); 2903e9964c10SSage Weil } 2904e9964c10SSage Weil if (n != &mdsc->cap_dirty) { 2905e9964c10SSage Weil nci = list_entry(n, struct ceph_inode_info, 2906e9964c10SSage Weil i_dirty_item); 2907e9964c10SSage Weil ninode = igrab(&nci->vfs_inode); 2908e9964c10SSage Weil BUG_ON(!ninode); 2909e9964c10SSage Weil nci->i_ceph_flags |= CEPH_I_NOFLUSH; 2910e9964c10SSage Weil dout("flush_dirty_caps next inode %p, noflush\n", 2911e9964c10SSage Weil ninode); 2912e9964c10SSage Weil } else { 2913e9964c10SSage Weil nci = NULL; 2914e9964c10SSage Weil ninode = NULL; 2915e9964c10SSage Weil } 2916afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2917afcdaea3SSage Weil if (inode) { 2918afcdaea3SSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, 2919afcdaea3SSage Weil NULL); 2920afcdaea3SSage Weil iput(inode); 2921afcdaea3SSage Weil } 2922afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2923afcdaea3SSage Weil } 2924afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2925afcdaea3SSage Weil } 2926afcdaea3SSage Weil 2927afcdaea3SSage Weil /* 2928a8599bd8SSage Weil * Drop open file reference. If we were the last open file, 2929a8599bd8SSage Weil * we may need to release capabilities to the MDS (or schedule 2930a8599bd8SSage Weil * their delayed release). 2931a8599bd8SSage Weil */ 2932a8599bd8SSage Weil void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) 2933a8599bd8SSage Weil { 2934a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2935a8599bd8SSage Weil int last = 0; 2936a8599bd8SSage Weil 2937a8599bd8SSage Weil spin_lock(&inode->i_lock); 2938a8599bd8SSage Weil dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode, 2939a8599bd8SSage Weil ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1); 2940a8599bd8SSage Weil BUG_ON(ci->i_nr_by_mode[fmode] == 0); 2941a8599bd8SSage Weil if (--ci->i_nr_by_mode[fmode] == 0) 2942a8599bd8SSage Weil last++; 2943a8599bd8SSage Weil spin_unlock(&inode->i_lock); 2944a8599bd8SSage Weil 2945a8599bd8SSage Weil if (last && ci->i_vino.snap == CEPH_NOSNAP) 2946a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 2947a8599bd8SSage Weil } 2948a8599bd8SSage Weil 2949a8599bd8SSage Weil /* 2950a8599bd8SSage Weil * Helpers for embedding cap and dentry lease releases into mds 2951a8599bd8SSage Weil * requests. 2952a8599bd8SSage Weil * 2953a8599bd8SSage Weil * @force is used by dentry_release (below) to force inclusion of a 2954a8599bd8SSage Weil * record for the directory inode, even when there aren't any caps to 2955a8599bd8SSage Weil * drop. 2956a8599bd8SSage Weil */ 2957a8599bd8SSage Weil int ceph_encode_inode_release(void **p, struct inode *inode, 2958a8599bd8SSage Weil int mds, int drop, int unless, int force) 2959a8599bd8SSage Weil { 2960a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2961a8599bd8SSage Weil struct ceph_cap *cap; 2962a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 2963ec97f88bSSage Weil int used, dirty; 2964a8599bd8SSage Weil int ret = 0; 2965a8599bd8SSage Weil 2966a8599bd8SSage Weil spin_lock(&inode->i_lock); 2967916623daSSage Weil used = __ceph_caps_used(ci); 2968ec97f88bSSage Weil dirty = __ceph_caps_dirty(ci); 2969916623daSSage Weil 2970ec97f88bSSage Weil dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", 2971ec97f88bSSage Weil inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), 2972916623daSSage Weil ceph_cap_string(unless)); 2973916623daSSage Weil 2974ec97f88bSSage Weil /* only drop unused, clean caps */ 2975ec97f88bSSage Weil drop &= ~(used | dirty); 2976916623daSSage Weil 2977a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 2978a8599bd8SSage Weil if (cap && __cap_is_valid(cap)) { 2979a8599bd8SSage Weil if (force || 2980a8599bd8SSage Weil ((cap->issued & drop) && 2981a8599bd8SSage Weil (cap->issued & unless) == 0)) { 2982a8599bd8SSage Weil if ((cap->issued & drop) && 2983a8599bd8SSage Weil (cap->issued & unless) == 0) { 2984a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s -> " 2985a8599bd8SSage Weil "%s\n", inode, cap, 2986a8599bd8SSage Weil ceph_cap_string(cap->issued), 2987a8599bd8SSage Weil ceph_cap_string(cap->issued & ~drop)); 2988a8599bd8SSage Weil cap->issued &= ~drop; 2989a8599bd8SSage Weil cap->implemented &= ~drop; 2990a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_NODELAY) { 2991a8599bd8SSage Weil int wanted = __ceph_caps_wanted(ci); 2992a8599bd8SSage Weil dout(" wanted %s -> %s (act %s)\n", 2993a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 2994a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted & 2995a8599bd8SSage Weil ~wanted), 2996a8599bd8SSage Weil ceph_cap_string(wanted)); 2997a8599bd8SSage Weil cap->mds_wanted &= wanted; 2998a8599bd8SSage Weil } 2999a8599bd8SSage Weil } else { 3000a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s" 3001a8599bd8SSage Weil " (force)\n", inode, cap, 3002a8599bd8SSage Weil ceph_cap_string(cap->issued)); 3003a8599bd8SSage Weil } 3004a8599bd8SSage Weil 3005a8599bd8SSage Weil rel->ino = cpu_to_le64(ceph_ino(inode)); 3006a8599bd8SSage Weil rel->cap_id = cpu_to_le64(cap->cap_id); 3007a8599bd8SSage Weil rel->seq = cpu_to_le32(cap->seq); 3008a8599bd8SSage Weil rel->issue_seq = cpu_to_le32(cap->issue_seq), 3009a8599bd8SSage Weil rel->mseq = cpu_to_le32(cap->mseq); 3010a8599bd8SSage Weil rel->caps = cpu_to_le32(cap->issued); 3011a8599bd8SSage Weil rel->wanted = cpu_to_le32(cap->mds_wanted); 3012a8599bd8SSage Weil rel->dname_len = 0; 3013a8599bd8SSage Weil rel->dname_seq = 0; 3014a8599bd8SSage Weil *p += sizeof(*rel); 3015a8599bd8SSage Weil ret = 1; 3016a8599bd8SSage Weil } else { 3017a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s\n", 3018a8599bd8SSage Weil inode, cap, ceph_cap_string(cap->issued)); 3019a8599bd8SSage Weil } 3020a8599bd8SSage Weil } 3021a8599bd8SSage Weil spin_unlock(&inode->i_lock); 3022a8599bd8SSage Weil return ret; 3023a8599bd8SSage Weil } 3024a8599bd8SSage Weil 3025a8599bd8SSage Weil int ceph_encode_dentry_release(void **p, struct dentry *dentry, 3026a8599bd8SSage Weil int mds, int drop, int unless) 3027a8599bd8SSage Weil { 3028a8599bd8SSage Weil struct inode *dir = dentry->d_parent->d_inode; 3029a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3030a8599bd8SSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 3031a8599bd8SSage Weil int force = 0; 3032a8599bd8SSage Weil int ret; 3033a8599bd8SSage Weil 3034a8599bd8SSage Weil /* 3035a8599bd8SSage Weil * force an record for the directory caps if we have a dentry lease. 3036a8599bd8SSage Weil * this is racy (can't take i_lock and d_lock together), but it 3037a8599bd8SSage Weil * doesn't have to be perfect; the mds will revoke anything we don't 3038a8599bd8SSage Weil * release. 3039a8599bd8SSage Weil */ 3040a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3041a8599bd8SSage Weil if (di->lease_session && di->lease_session->s_mds == mds) 3042a8599bd8SSage Weil force = 1; 3043a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3044a8599bd8SSage Weil 3045a8599bd8SSage Weil ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); 3046a8599bd8SSage Weil 3047a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3048a8599bd8SSage Weil if (ret && di->lease_session && di->lease_session->s_mds == mds) { 3049a8599bd8SSage Weil dout("encode_dentry_release %p mds%d seq %d\n", 3050a8599bd8SSage Weil dentry, mds, (int)di->lease_seq); 3051a8599bd8SSage Weil rel->dname_len = cpu_to_le32(dentry->d_name.len); 3052a8599bd8SSage Weil memcpy(*p, dentry->d_name.name, dentry->d_name.len); 3053a8599bd8SSage Weil *p += dentry->d_name.len; 3054a8599bd8SSage Weil rel->dname_seq = cpu_to_le32(di->lease_seq); 30551dadcce3SSage Weil __ceph_mdsc_drop_dentry_lease(dentry); 3056a8599bd8SSage Weil } 3057a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3058a8599bd8SSage Weil return ret; 3059a8599bd8SSage Weil } 3060