13d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 2a8599bd8SSage Weil 3a8599bd8SSage Weil #include <linux/fs.h> 4a8599bd8SSage Weil #include <linux/kernel.h> 5a8599bd8SSage Weil #include <linux/sched.h> 65a0e3ad6STejun Heo #include <linux/slab.h> 7a8599bd8SSage Weil #include <linux/vmalloc.h> 8a8599bd8SSage Weil #include <linux/wait.h> 9f1a3d572SStephen Rothwell #include <linux/writeback.h> 10a8599bd8SSage Weil 11a8599bd8SSage Weil #include "super.h" 123d14c5d2SYehuda Sadeh #include "mds_client.h" 133d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 143d14c5d2SYehuda Sadeh #include <linux/ceph/messenger.h> 15a8599bd8SSage Weil 16a8599bd8SSage Weil /* 17a8599bd8SSage Weil * Capability management 18a8599bd8SSage Weil * 19a8599bd8SSage Weil * The Ceph metadata servers control client access to inode metadata 20a8599bd8SSage Weil * and file data by issuing capabilities, granting clients permission 21a8599bd8SSage Weil * to read and/or write both inode field and file data to OSDs 22a8599bd8SSage Weil * (storage nodes). Each capability consists of a set of bits 23a8599bd8SSage Weil * indicating which operations are allowed. 24a8599bd8SSage Weil * 25a8599bd8SSage Weil * If the client holds a *_SHARED cap, the client has a coherent value 26a8599bd8SSage Weil * that can be safely read from the cached inode. 27a8599bd8SSage Weil * 28a8599bd8SSage Weil * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the 29a8599bd8SSage Weil * client is allowed to change inode attributes (e.g., file size, 30a8599bd8SSage Weil * mtime), note its dirty state in the ceph_cap, and asynchronously 31a8599bd8SSage Weil * flush that metadata change to the MDS. 32a8599bd8SSage Weil * 33a8599bd8SSage Weil * In the event of a conflicting operation (perhaps by another 34a8599bd8SSage Weil * client), the MDS will revoke the conflicting client capabilities. 35a8599bd8SSage Weil * 36a8599bd8SSage Weil * In order for a client to cache an inode, it must hold a capability 37a8599bd8SSage Weil * with at least one MDS server. When inodes are released, release 38a8599bd8SSage Weil * notifications are batched and periodically sent en masse to the MDS 39a8599bd8SSage Weil * cluster to release server state. 40a8599bd8SSage Weil */ 41a8599bd8SSage Weil 42a8599bd8SSage Weil 43a8599bd8SSage Weil /* 44a8599bd8SSage Weil * Generate readable cap strings for debugging output. 45a8599bd8SSage Weil */ 46a8599bd8SSage Weil #define MAX_CAP_STR 20 47a8599bd8SSage Weil static char cap_str[MAX_CAP_STR][40]; 48a8599bd8SSage Weil static DEFINE_SPINLOCK(cap_str_lock); 49a8599bd8SSage Weil static int last_cap_str; 50a8599bd8SSage Weil 51a8599bd8SSage Weil static char *gcap_string(char *s, int c) 52a8599bd8SSage Weil { 53a8599bd8SSage Weil if (c & CEPH_CAP_GSHARED) 54a8599bd8SSage Weil *s++ = 's'; 55a8599bd8SSage Weil if (c & CEPH_CAP_GEXCL) 56a8599bd8SSage Weil *s++ = 'x'; 57a8599bd8SSage Weil if (c & CEPH_CAP_GCACHE) 58a8599bd8SSage Weil *s++ = 'c'; 59a8599bd8SSage Weil if (c & CEPH_CAP_GRD) 60a8599bd8SSage Weil *s++ = 'r'; 61a8599bd8SSage Weil if (c & CEPH_CAP_GWR) 62a8599bd8SSage Weil *s++ = 'w'; 63a8599bd8SSage Weil if (c & CEPH_CAP_GBUFFER) 64a8599bd8SSage Weil *s++ = 'b'; 65a8599bd8SSage Weil if (c & CEPH_CAP_GLAZYIO) 66a8599bd8SSage Weil *s++ = 'l'; 67a8599bd8SSage Weil return s; 68a8599bd8SSage Weil } 69a8599bd8SSage Weil 70a8599bd8SSage Weil const char *ceph_cap_string(int caps) 71a8599bd8SSage Weil { 72a8599bd8SSage Weil int i; 73a8599bd8SSage Weil char *s; 74a8599bd8SSage Weil int c; 75a8599bd8SSage Weil 76a8599bd8SSage Weil spin_lock(&cap_str_lock); 77a8599bd8SSage Weil i = last_cap_str++; 78a8599bd8SSage Weil if (last_cap_str == MAX_CAP_STR) 79a8599bd8SSage Weil last_cap_str = 0; 80a8599bd8SSage Weil spin_unlock(&cap_str_lock); 81a8599bd8SSage Weil 82a8599bd8SSage Weil s = cap_str[i]; 83a8599bd8SSage Weil 84a8599bd8SSage Weil if (caps & CEPH_CAP_PIN) 85a8599bd8SSage Weil *s++ = 'p'; 86a8599bd8SSage Weil 87a8599bd8SSage Weil c = (caps >> CEPH_CAP_SAUTH) & 3; 88a8599bd8SSage Weil if (c) { 89a8599bd8SSage Weil *s++ = 'A'; 90a8599bd8SSage Weil s = gcap_string(s, c); 91a8599bd8SSage Weil } 92a8599bd8SSage Weil 93a8599bd8SSage Weil c = (caps >> CEPH_CAP_SLINK) & 3; 94a8599bd8SSage Weil if (c) { 95a8599bd8SSage Weil *s++ = 'L'; 96a8599bd8SSage Weil s = gcap_string(s, c); 97a8599bd8SSage Weil } 98a8599bd8SSage Weil 99a8599bd8SSage Weil c = (caps >> CEPH_CAP_SXATTR) & 3; 100a8599bd8SSage Weil if (c) { 101a8599bd8SSage Weil *s++ = 'X'; 102a8599bd8SSage Weil s = gcap_string(s, c); 103a8599bd8SSage Weil } 104a8599bd8SSage Weil 105a8599bd8SSage Weil c = caps >> CEPH_CAP_SFILE; 106a8599bd8SSage Weil if (c) { 107a8599bd8SSage Weil *s++ = 'F'; 108a8599bd8SSage Weil s = gcap_string(s, c); 109a8599bd8SSage Weil } 110a8599bd8SSage Weil 111a8599bd8SSage Weil if (s == cap_str[i]) 112a8599bd8SSage Weil *s++ = '-'; 113a8599bd8SSage Weil *s = 0; 114a8599bd8SSage Weil return cap_str[i]; 115a8599bd8SSage Weil } 116a8599bd8SSage Weil 11737151668SYehuda Sadeh void ceph_caps_init(struct ceph_mds_client *mdsc) 118a8599bd8SSage Weil { 11937151668SYehuda Sadeh INIT_LIST_HEAD(&mdsc->caps_list); 12037151668SYehuda Sadeh spin_lock_init(&mdsc->caps_list_lock); 121a8599bd8SSage Weil } 122a8599bd8SSage Weil 12337151668SYehuda Sadeh void ceph_caps_finalize(struct ceph_mds_client *mdsc) 124a8599bd8SSage Weil { 125a8599bd8SSage Weil struct ceph_cap *cap; 126a8599bd8SSage Weil 12737151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 12837151668SYehuda Sadeh while (!list_empty(&mdsc->caps_list)) { 12937151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, 13037151668SYehuda Sadeh struct ceph_cap, caps_item); 131a8599bd8SSage Weil list_del(&cap->caps_item); 132a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 133a8599bd8SSage Weil } 13437151668SYehuda Sadeh mdsc->caps_total_count = 0; 13537151668SYehuda Sadeh mdsc->caps_avail_count = 0; 13637151668SYehuda Sadeh mdsc->caps_use_count = 0; 13737151668SYehuda Sadeh mdsc->caps_reserve_count = 0; 13837151668SYehuda Sadeh mdsc->caps_min_count = 0; 13937151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 14085ccce43SSage Weil } 14185ccce43SSage Weil 14237151668SYehuda Sadeh void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta) 14385ccce43SSage Weil { 14437151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 14537151668SYehuda Sadeh mdsc->caps_min_count += delta; 14637151668SYehuda Sadeh BUG_ON(mdsc->caps_min_count < 0); 14737151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 148a8599bd8SSage Weil } 149a8599bd8SSage Weil 15093faca6eSmajianpeng void ceph_reserve_caps(struct ceph_mds_client *mdsc, 15137151668SYehuda Sadeh struct ceph_cap_reservation *ctx, int need) 152a8599bd8SSage Weil { 153a8599bd8SSage Weil int i; 154a8599bd8SSage Weil struct ceph_cap *cap; 155a8599bd8SSage Weil int have; 156a8599bd8SSage Weil int alloc = 0; 157a8599bd8SSage Weil LIST_HEAD(newcaps); 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); 17693faca6eSmajianpeng if (!cap) 17793faca6eSmajianpeng break; 178a8599bd8SSage Weil list_add(&cap->caps_item, &newcaps); 179a8599bd8SSage Weil alloc++; 180a8599bd8SSage Weil } 18193faca6eSmajianpeng /* we didn't manage to reserve as much as we needed */ 18293faca6eSmajianpeng if (have + alloc != need) 18393faca6eSmajianpeng pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n", 18493faca6eSmajianpeng ctx, need, have + alloc); 185a8599bd8SSage Weil 18637151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 18737151668SYehuda Sadeh mdsc->caps_total_count += alloc; 18837151668SYehuda Sadeh mdsc->caps_reserve_count += alloc; 18937151668SYehuda Sadeh list_splice(&newcaps, &mdsc->caps_list); 190a8599bd8SSage Weil 19137151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 19237151668SYehuda Sadeh mdsc->caps_reserve_count + 19337151668SYehuda Sadeh mdsc->caps_avail_count); 19437151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 195a8599bd8SSage Weil 196a8599bd8SSage Weil ctx->count = need; 197a8599bd8SSage Weil dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n", 19837151668SYehuda Sadeh ctx, mdsc->caps_total_count, mdsc->caps_use_count, 19937151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 200a8599bd8SSage Weil } 201a8599bd8SSage Weil 20237151668SYehuda Sadeh int ceph_unreserve_caps(struct ceph_mds_client *mdsc, 20337151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 204a8599bd8SSage Weil { 205a8599bd8SSage Weil dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count); 206a8599bd8SSage Weil if (ctx->count) { 20737151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 20837151668SYehuda Sadeh BUG_ON(mdsc->caps_reserve_count < ctx->count); 20937151668SYehuda Sadeh mdsc->caps_reserve_count -= ctx->count; 21037151668SYehuda Sadeh mdsc->caps_avail_count += ctx->count; 211a8599bd8SSage Weil ctx->count = 0; 212a8599bd8SSage Weil dout("unreserve caps %d = %d used + %d resv + %d avail\n", 21337151668SYehuda Sadeh mdsc->caps_total_count, mdsc->caps_use_count, 21437151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 21537151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 21637151668SYehuda Sadeh mdsc->caps_reserve_count + 21737151668SYehuda Sadeh mdsc->caps_avail_count); 21837151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 219a8599bd8SSage Weil } 220a8599bd8SSage Weil return 0; 221a8599bd8SSage Weil } 222a8599bd8SSage Weil 22337151668SYehuda Sadeh static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc, 22437151668SYehuda Sadeh struct ceph_cap_reservation *ctx) 225a8599bd8SSage Weil { 226a8599bd8SSage Weil struct ceph_cap *cap = NULL; 227a8599bd8SSage Weil 228a8599bd8SSage Weil /* temporary, until we do something about cap import/export */ 229443b3760SSage Weil if (!ctx) { 230443b3760SSage Weil cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); 231443b3760SSage Weil if (cap) { 2324d1d0534SYan, Zheng spin_lock(&mdsc->caps_list_lock); 23337151668SYehuda Sadeh mdsc->caps_use_count++; 23437151668SYehuda Sadeh mdsc->caps_total_count++; 2354d1d0534SYan, Zheng spin_unlock(&mdsc->caps_list_lock); 236443b3760SSage Weil } 237443b3760SSage Weil return cap; 238443b3760SSage Weil } 239a8599bd8SSage Weil 24037151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 241a8599bd8SSage Weil dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n", 24237151668SYehuda Sadeh ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count, 24337151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 244a8599bd8SSage Weil BUG_ON(!ctx->count); 24537151668SYehuda Sadeh BUG_ON(ctx->count > mdsc->caps_reserve_count); 24637151668SYehuda Sadeh BUG_ON(list_empty(&mdsc->caps_list)); 247a8599bd8SSage Weil 248a8599bd8SSage Weil ctx->count--; 24937151668SYehuda Sadeh mdsc->caps_reserve_count--; 25037151668SYehuda Sadeh mdsc->caps_use_count++; 251a8599bd8SSage Weil 25237151668SYehuda Sadeh cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item); 253a8599bd8SSage Weil list_del(&cap->caps_item); 254a8599bd8SSage Weil 25537151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 25637151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 25737151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 258a8599bd8SSage Weil return cap; 259a8599bd8SSage Weil } 260a8599bd8SSage Weil 26137151668SYehuda Sadeh void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap) 262a8599bd8SSage Weil { 26337151668SYehuda Sadeh spin_lock(&mdsc->caps_list_lock); 2647c1332b8SSage Weil dout("put_cap %p %d = %d used + %d resv + %d avail\n", 26537151668SYehuda Sadeh cap, mdsc->caps_total_count, mdsc->caps_use_count, 26637151668SYehuda Sadeh mdsc->caps_reserve_count, mdsc->caps_avail_count); 26737151668SYehuda Sadeh mdsc->caps_use_count--; 268a8599bd8SSage Weil /* 26985ccce43SSage Weil * Keep some preallocated caps around (ceph_min_count), to 27085ccce43SSage Weil * avoid lots of free/alloc churn. 271a8599bd8SSage Weil */ 27237151668SYehuda Sadeh if (mdsc->caps_avail_count >= mdsc->caps_reserve_count + 27337151668SYehuda Sadeh mdsc->caps_min_count) { 27437151668SYehuda Sadeh mdsc->caps_total_count--; 275a8599bd8SSage Weil kmem_cache_free(ceph_cap_cachep, cap); 276a8599bd8SSage Weil } else { 27737151668SYehuda Sadeh mdsc->caps_avail_count++; 27837151668SYehuda Sadeh list_add(&cap->caps_item, &mdsc->caps_list); 279a8599bd8SSage Weil } 280a8599bd8SSage Weil 28137151668SYehuda Sadeh BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + 28237151668SYehuda Sadeh mdsc->caps_reserve_count + mdsc->caps_avail_count); 28337151668SYehuda Sadeh spin_unlock(&mdsc->caps_list_lock); 284a8599bd8SSage Weil } 285a8599bd8SSage Weil 2863d14c5d2SYehuda Sadeh void ceph_reservation_status(struct ceph_fs_client *fsc, 28785ccce43SSage Weil int *total, int *avail, int *used, int *reserved, 28885ccce43SSage Weil int *min) 289a8599bd8SSage Weil { 2903d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 29137151668SYehuda Sadeh 292a8599bd8SSage Weil if (total) 29337151668SYehuda Sadeh *total = mdsc->caps_total_count; 294a8599bd8SSage Weil if (avail) 29537151668SYehuda Sadeh *avail = mdsc->caps_avail_count; 296a8599bd8SSage Weil if (used) 29737151668SYehuda Sadeh *used = mdsc->caps_use_count; 298a8599bd8SSage Weil if (reserved) 29937151668SYehuda Sadeh *reserved = mdsc->caps_reserve_count; 30085ccce43SSage Weil if (min) 30137151668SYehuda Sadeh *min = mdsc->caps_min_count; 302a8599bd8SSage Weil } 303a8599bd8SSage Weil 304a8599bd8SSage Weil /* 305a8599bd8SSage Weil * Find ceph_cap for given mds, if any. 306a8599bd8SSage Weil * 307be655596SSage Weil * Called with i_ceph_lock held. 308a8599bd8SSage Weil */ 309a8599bd8SSage Weil static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds) 310a8599bd8SSage Weil { 311a8599bd8SSage Weil struct ceph_cap *cap; 312a8599bd8SSage Weil struct rb_node *n = ci->i_caps.rb_node; 313a8599bd8SSage Weil 314a8599bd8SSage Weil while (n) { 315a8599bd8SSage Weil cap = rb_entry(n, struct ceph_cap, ci_node); 316a8599bd8SSage Weil if (mds < cap->mds) 317a8599bd8SSage Weil n = n->rb_left; 318a8599bd8SSage Weil else if (mds > cap->mds) 319a8599bd8SSage Weil n = n->rb_right; 320a8599bd8SSage Weil else 321a8599bd8SSage Weil return cap; 322a8599bd8SSage Weil } 323a8599bd8SSage Weil return NULL; 324a8599bd8SSage Weil } 325a8599bd8SSage Weil 3262bc50259SGreg Farnum struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds) 3272bc50259SGreg Farnum { 3282bc50259SGreg Farnum struct ceph_cap *cap; 3292bc50259SGreg Farnum 330be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3312bc50259SGreg Farnum cap = __get_cap_for_mds(ci, mds); 332be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3332bc50259SGreg Farnum return cap; 3342bc50259SGreg Farnum } 3352bc50259SGreg Farnum 336a8599bd8SSage Weil /* 33733caad32SSage Weil * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1. 338a8599bd8SSage Weil */ 339ca81f3f6SSage Weil static int __ceph_get_cap_mds(struct ceph_inode_info *ci) 340a8599bd8SSage Weil { 341a8599bd8SSage Weil struct ceph_cap *cap; 342a8599bd8SSage Weil int mds = -1; 343a8599bd8SSage Weil struct rb_node *p; 344a8599bd8SSage Weil 34533caad32SSage Weil /* prefer mds with WR|BUFFER|EXCL caps */ 346a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 347a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 348a8599bd8SSage Weil mds = cap->mds; 349a8599bd8SSage Weil if (cap->issued & (CEPH_CAP_FILE_WR | 350a8599bd8SSage Weil CEPH_CAP_FILE_BUFFER | 351a8599bd8SSage Weil CEPH_CAP_FILE_EXCL)) 352a8599bd8SSage Weil break; 353a8599bd8SSage Weil } 354a8599bd8SSage Weil return mds; 355a8599bd8SSage Weil } 356a8599bd8SSage Weil 357a8599bd8SSage Weil int ceph_get_cap_mds(struct inode *inode) 358a8599bd8SSage Weil { 359be655596SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 360a8599bd8SSage Weil int mds; 361be655596SSage Weil spin_lock(&ci->i_ceph_lock); 362ca81f3f6SSage Weil mds = __ceph_get_cap_mds(ceph_inode(inode)); 363be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 364a8599bd8SSage Weil return mds; 365a8599bd8SSage Weil } 366a8599bd8SSage Weil 367a8599bd8SSage Weil /* 368be655596SSage Weil * Called under i_ceph_lock. 369a8599bd8SSage Weil */ 370a8599bd8SSage Weil static void __insert_cap_node(struct ceph_inode_info *ci, 371a8599bd8SSage Weil struct ceph_cap *new) 372a8599bd8SSage Weil { 373a8599bd8SSage Weil struct rb_node **p = &ci->i_caps.rb_node; 374a8599bd8SSage Weil struct rb_node *parent = NULL; 375a8599bd8SSage Weil struct ceph_cap *cap = NULL; 376a8599bd8SSage Weil 377a8599bd8SSage Weil while (*p) { 378a8599bd8SSage Weil parent = *p; 379a8599bd8SSage Weil cap = rb_entry(parent, struct ceph_cap, ci_node); 380a8599bd8SSage Weil if (new->mds < cap->mds) 381a8599bd8SSage Weil p = &(*p)->rb_left; 382a8599bd8SSage Weil else if (new->mds > cap->mds) 383a8599bd8SSage Weil p = &(*p)->rb_right; 384a8599bd8SSage Weil else 385a8599bd8SSage Weil BUG(); 386a8599bd8SSage Weil } 387a8599bd8SSage Weil 388a8599bd8SSage Weil rb_link_node(&new->ci_node, parent, p); 389a8599bd8SSage Weil rb_insert_color(&new->ci_node, &ci->i_caps); 390a8599bd8SSage Weil } 391a8599bd8SSage Weil 392a8599bd8SSage Weil /* 393a8599bd8SSage Weil * (re)set cap hold timeouts, which control the delayed release 394a8599bd8SSage Weil * of unused caps back to the MDS. Should be called on cap use. 395a8599bd8SSage Weil */ 396a8599bd8SSage Weil static void __cap_set_timeouts(struct ceph_mds_client *mdsc, 397a8599bd8SSage Weil struct ceph_inode_info *ci) 398a8599bd8SSage Weil { 3993d14c5d2SYehuda Sadeh struct ceph_mount_options *ma = mdsc->fsc->mount_options; 400a8599bd8SSage Weil 401a8599bd8SSage Weil ci->i_hold_caps_min = round_jiffies(jiffies + 402a8599bd8SSage Weil ma->caps_wanted_delay_min * HZ); 403a8599bd8SSage Weil ci->i_hold_caps_max = round_jiffies(jiffies + 404a8599bd8SSage Weil ma->caps_wanted_delay_max * HZ); 405a8599bd8SSage Weil dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode, 406a8599bd8SSage Weil ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies); 407a8599bd8SSage Weil } 408a8599bd8SSage Weil 409a8599bd8SSage Weil /* 410a8599bd8SSage Weil * (Re)queue cap at the end of the delayed cap release list. 411a8599bd8SSage Weil * 412a8599bd8SSage Weil * If I_FLUSH is set, leave the inode at the front of the list. 413a8599bd8SSage Weil * 414be655596SSage Weil * Caller holds i_ceph_lock 415a8599bd8SSage Weil * -> we take mdsc->cap_delay_lock 416a8599bd8SSage Weil */ 417a8599bd8SSage Weil static void __cap_delay_requeue(struct ceph_mds_client *mdsc, 418a8599bd8SSage Weil struct ceph_inode_info *ci) 419a8599bd8SSage Weil { 420a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 421a8599bd8SSage Weil dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode, 422a8599bd8SSage Weil ci->i_ceph_flags, ci->i_hold_caps_max); 423a8599bd8SSage Weil if (!mdsc->stopping) { 424a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 425a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) { 426a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 427a8599bd8SSage Weil goto no_change; 428a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 429a8599bd8SSage Weil } 430a8599bd8SSage Weil list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 431a8599bd8SSage Weil no_change: 432a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 433a8599bd8SSage Weil } 434a8599bd8SSage Weil } 435a8599bd8SSage Weil 436a8599bd8SSage Weil /* 437a8599bd8SSage Weil * Queue an inode for immediate writeback. Mark inode with I_FLUSH, 438a8599bd8SSage Weil * indicating we should send a cap message to flush dirty metadata 439a8599bd8SSage Weil * asap, and move to the front of the delayed cap list. 440a8599bd8SSage Weil */ 441a8599bd8SSage Weil static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc, 442a8599bd8SSage Weil struct ceph_inode_info *ci) 443a8599bd8SSage Weil { 444a8599bd8SSage Weil dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode); 445a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 446a8599bd8SSage Weil ci->i_ceph_flags |= CEPH_I_FLUSH; 447a8599bd8SSage Weil if (!list_empty(&ci->i_cap_delay_list)) 448a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 449a8599bd8SSage Weil list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list); 450a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 451a8599bd8SSage Weil } 452a8599bd8SSage Weil 453a8599bd8SSage Weil /* 454a8599bd8SSage Weil * Cancel delayed work on cap. 455a8599bd8SSage Weil * 456be655596SSage Weil * Caller must hold i_ceph_lock. 457a8599bd8SSage Weil */ 458a8599bd8SSage Weil static void __cap_delay_cancel(struct ceph_mds_client *mdsc, 459a8599bd8SSage Weil struct ceph_inode_info *ci) 460a8599bd8SSage Weil { 461a8599bd8SSage Weil dout("__cap_delay_cancel %p\n", &ci->vfs_inode); 462a8599bd8SSage Weil if (list_empty(&ci->i_cap_delay_list)) 463a8599bd8SSage Weil return; 464a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 465a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 466a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 467a8599bd8SSage Weil } 468a8599bd8SSage Weil 469a8599bd8SSage Weil /* 470a8599bd8SSage Weil * Common issue checks for add_cap, handle_cap_grant. 471a8599bd8SSage Weil */ 472a8599bd8SSage Weil static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap, 473a8599bd8SSage Weil unsigned issued) 474a8599bd8SSage Weil { 475a8599bd8SSage Weil unsigned had = __ceph_caps_issued(ci, NULL); 476a8599bd8SSage Weil 477a8599bd8SSage Weil /* 478a8599bd8SSage Weil * Each time we receive FILE_CACHE anew, we increment 479a8599bd8SSage Weil * i_rdcache_gen. 480a8599bd8SSage Weil */ 4812962507cSSage Weil if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && 4822962507cSSage Weil (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) 483a8599bd8SSage Weil ci->i_rdcache_gen++; 484a8599bd8SSage Weil 485a8599bd8SSage Weil /* 4862f276c51SYan, Zheng * if we are newly issued FILE_SHARED, mark dir not complete; we 487a8599bd8SSage Weil * don't know what happened to this directory while we didn't 488a8599bd8SSage Weil * have the cap. 489a8599bd8SSage Weil */ 490a8599bd8SSage Weil if ((issued & CEPH_CAP_FILE_SHARED) && 491a8599bd8SSage Weil (had & CEPH_CAP_FILE_SHARED) == 0) { 492a8599bd8SSage Weil ci->i_shared_gen++; 493a8673d61SYan, Zheng if (S_ISDIR(ci->vfs_inode.i_mode)) { 494a8673d61SYan, Zheng dout(" marking %p NOT complete\n", &ci->vfs_inode); 4952f276c51SYan, Zheng __ceph_dir_clear_complete(ci); 496a8673d61SYan, Zheng } 497a8599bd8SSage Weil } 498a8599bd8SSage Weil } 499a8599bd8SSage Weil 500a8599bd8SSage Weil /* 501a8599bd8SSage Weil * Add a capability under the given MDS session. 502a8599bd8SSage Weil * 503a8599bd8SSage Weil * Caller should hold session snap_rwsem (read) and s_mutex. 504a8599bd8SSage Weil * 505a8599bd8SSage Weil * @fmode is the open file mode, if we are opening a file, otherwise 506a8599bd8SSage Weil * it is < 0. (This is so we can atomically add the cap and add an 507a8599bd8SSage Weil * open file reference to it.) 508a8599bd8SSage Weil */ 509a8599bd8SSage Weil int ceph_add_cap(struct inode *inode, 510a8599bd8SSage Weil struct ceph_mds_session *session, u64 cap_id, 511a8599bd8SSage Weil int fmode, unsigned issued, unsigned wanted, 512a8599bd8SSage Weil unsigned seq, unsigned mseq, u64 realmino, int flags, 513a8599bd8SSage Weil struct ceph_cap_reservation *caps_reservation) 514a8599bd8SSage Weil { 5153d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 516a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 517a8599bd8SSage Weil struct ceph_cap *new_cap = NULL; 518a8599bd8SSage Weil struct ceph_cap *cap; 519a8599bd8SSage Weil int mds = session->s_mds; 520a8599bd8SSage Weil int actual_wanted; 521a8599bd8SSage Weil 522a8599bd8SSage Weil dout("add_cap %p mds%d cap %llx %s seq %d\n", inode, 523a8599bd8SSage Weil session->s_mds, cap_id, ceph_cap_string(issued), seq); 524a8599bd8SSage Weil 525a8599bd8SSage Weil /* 526a8599bd8SSage Weil * If we are opening the file, include file mode wanted bits 527a8599bd8SSage Weil * in wanted. 528a8599bd8SSage Weil */ 529a8599bd8SSage Weil if (fmode >= 0) 530a8599bd8SSage Weil wanted |= ceph_caps_for_mode(fmode); 531a8599bd8SSage Weil 532a8599bd8SSage Weil retry: 533be655596SSage Weil spin_lock(&ci->i_ceph_lock); 534a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 535a8599bd8SSage Weil if (!cap) { 536a8599bd8SSage Weil if (new_cap) { 537a8599bd8SSage Weil cap = new_cap; 538a8599bd8SSage Weil new_cap = NULL; 539a8599bd8SSage Weil } else { 540be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 54137151668SYehuda Sadeh new_cap = get_cap(mdsc, caps_reservation); 542a8599bd8SSage Weil if (new_cap == NULL) 543a8599bd8SSage Weil return -ENOMEM; 544a8599bd8SSage Weil goto retry; 545a8599bd8SSage Weil } 546a8599bd8SSage Weil 547a8599bd8SSage Weil cap->issued = 0; 548a8599bd8SSage Weil cap->implemented = 0; 549a8599bd8SSage Weil cap->mds = mds; 550a8599bd8SSage Weil cap->mds_wanted = 0; 551964266ccSYan, Zheng cap->mseq = 0; 552a8599bd8SSage Weil 553a8599bd8SSage Weil cap->ci = ci; 554a8599bd8SSage Weil __insert_cap_node(ci, cap); 555a8599bd8SSage Weil 556a8599bd8SSage Weil /* clear out old exporting info? (i.e. on cap import) */ 557a8599bd8SSage Weil if (ci->i_cap_exporting_mds == mds) { 558a8599bd8SSage Weil ci->i_cap_exporting_issued = 0; 559a8599bd8SSage Weil ci->i_cap_exporting_mseq = 0; 560a8599bd8SSage Weil ci->i_cap_exporting_mds = -1; 561a8599bd8SSage Weil } 562a8599bd8SSage Weil 563a8599bd8SSage Weil /* add to session cap list */ 564a8599bd8SSage Weil cap->session = session; 565a8599bd8SSage Weil spin_lock(&session->s_cap_lock); 566a8599bd8SSage Weil list_add_tail(&cap->session_caps, &session->s_caps); 567a8599bd8SSage Weil session->s_nr_caps++; 568a8599bd8SSage Weil spin_unlock(&session->s_cap_lock); 5693540303fSSage Weil } else if (new_cap) 5703540303fSSage Weil ceph_put_cap(mdsc, new_cap); 571a8599bd8SSage Weil 572a8599bd8SSage Weil if (!ci->i_snap_realm) { 573a8599bd8SSage Weil /* 574a8599bd8SSage Weil * add this inode to the appropriate snap realm 575a8599bd8SSage Weil */ 576a8599bd8SSage Weil struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc, 577a8599bd8SSage Weil realmino); 578a8599bd8SSage Weil if (realm) { 579a8599bd8SSage Weil ceph_get_snap_realm(mdsc, realm); 580a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 581a8599bd8SSage Weil ci->i_snap_realm = realm; 582a8599bd8SSage Weil list_add(&ci->i_snap_realm_item, 583a8599bd8SSage Weil &realm->inodes_with_caps); 584a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 585a8599bd8SSage Weil } else { 586a8599bd8SSage Weil pr_err("ceph_add_cap: couldn't find snap realm %llx\n", 587a8599bd8SSage Weil realmino); 588b8cd07e7SSage Weil WARN_ON(!realm); 589a8599bd8SSage Weil } 590a8599bd8SSage Weil } 591a8599bd8SSage Weil 592a8599bd8SSage Weil __check_cap_issue(ci, cap, issued); 593a8599bd8SSage Weil 594a8599bd8SSage Weil /* 595a8599bd8SSage Weil * If we are issued caps we don't want, or the mds' wanted 596a8599bd8SSage Weil * value appears to be off, queue a check so we'll release 597a8599bd8SSage Weil * later and/or update the mds wanted value. 598a8599bd8SSage Weil */ 599a8599bd8SSage Weil actual_wanted = __ceph_caps_wanted(ci); 600a8599bd8SSage Weil if ((wanted & ~actual_wanted) || 601a8599bd8SSage Weil (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) { 602a8599bd8SSage Weil dout(" issued %s, mds wanted %s, actual %s, queueing\n", 603a8599bd8SSage Weil ceph_cap_string(issued), ceph_cap_string(wanted), 604a8599bd8SSage Weil ceph_cap_string(actual_wanted)); 605a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 606a8599bd8SSage Weil } 607a8599bd8SSage Weil 608b8c2f3aeSYan, Zheng if (flags & CEPH_CAP_FLAG_AUTH) { 609b8c2f3aeSYan, Zheng if (ci->i_auth_cap == NULL || 610b8c2f3aeSYan, Zheng ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) 611a8599bd8SSage Weil ci->i_auth_cap = cap; 612b8c2f3aeSYan, Zheng } else if (ci->i_auth_cap == cap) { 613a8599bd8SSage Weil ci->i_auth_cap = NULL; 6148a92a119SYan, Zheng spin_lock(&mdsc->cap_dirty_lock); 6158a92a119SYan, Zheng if (!list_empty(&ci->i_dirty_item)) { 6168a92a119SYan, Zheng dout(" moving %p to cap_dirty_migrating\n", inode); 6178a92a119SYan, Zheng list_move(&ci->i_dirty_item, 6188a92a119SYan, Zheng &mdsc->cap_dirty_migrating); 6198a92a119SYan, Zheng } 6208a92a119SYan, Zheng spin_unlock(&mdsc->cap_dirty_lock); 6218a92a119SYan, Zheng } 622a8599bd8SSage Weil 623a8599bd8SSage Weil dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n", 624a8599bd8SSage Weil inode, ceph_vinop(inode), cap, ceph_cap_string(issued), 625a8599bd8SSage Weil ceph_cap_string(issued|cap->issued), seq, mds); 626a8599bd8SSage Weil cap->cap_id = cap_id; 627a8599bd8SSage Weil cap->issued = issued; 628a8599bd8SSage Weil cap->implemented |= issued; 629964266ccSYan, Zheng if (mseq > cap->mseq) 630964266ccSYan, Zheng cap->mds_wanted = wanted; 631964266ccSYan, Zheng else 632a8599bd8SSage Weil cap->mds_wanted |= wanted; 633a8599bd8SSage Weil cap->seq = seq; 634a8599bd8SSage Weil cap->issue_seq = seq; 635a8599bd8SSage Weil cap->mseq = mseq; 636685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 637a8599bd8SSage Weil 638a8599bd8SSage Weil if (fmode >= 0) 639a8599bd8SSage Weil __ceph_get_fmode(ci, fmode); 640be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 64103066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 642a8599bd8SSage Weil return 0; 643a8599bd8SSage Weil } 644a8599bd8SSage Weil 645a8599bd8SSage Weil /* 646a8599bd8SSage Weil * Return true if cap has not timed out and belongs to the current 647a8599bd8SSage Weil * generation of the MDS session (i.e. has not gone 'stale' due to 648a8599bd8SSage Weil * us losing touch with the mds). 649a8599bd8SSage Weil */ 650a8599bd8SSage Weil static int __cap_is_valid(struct ceph_cap *cap) 651a8599bd8SSage Weil { 652a8599bd8SSage Weil unsigned long ttl; 653cdac8303SSage Weil u32 gen; 654a8599bd8SSage Weil 655d8fb02abSAlex Elder spin_lock(&cap->session->s_gen_ttl_lock); 656a8599bd8SSage Weil gen = cap->session->s_cap_gen; 657a8599bd8SSage Weil ttl = cap->session->s_cap_ttl; 658d8fb02abSAlex Elder spin_unlock(&cap->session->s_gen_ttl_lock); 659a8599bd8SSage Weil 660685f9a5dSSage Weil if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) { 661a8599bd8SSage Weil dout("__cap_is_valid %p cap %p issued %s " 662a8599bd8SSage Weil "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode, 663685f9a5dSSage Weil cap, ceph_cap_string(cap->issued), cap->cap_gen, gen); 664a8599bd8SSage Weil return 0; 665a8599bd8SSage Weil } 666a8599bd8SSage Weil 667a8599bd8SSage Weil return 1; 668a8599bd8SSage Weil } 669a8599bd8SSage Weil 670a8599bd8SSage Weil /* 671a8599bd8SSage Weil * Return set of valid cap bits issued to us. Note that caps time 672a8599bd8SSage Weil * out, and may be invalidated in bulk if the client session times out 673a8599bd8SSage Weil * and session->s_cap_gen is bumped. 674a8599bd8SSage Weil */ 675a8599bd8SSage Weil int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented) 676a8599bd8SSage Weil { 6777af8f1e4SSage Weil int have = ci->i_snap_caps | ci->i_cap_exporting_issued; 678a8599bd8SSage Weil struct ceph_cap *cap; 679a8599bd8SSage Weil struct rb_node *p; 680a8599bd8SSage Weil 681a8599bd8SSage Weil if (implemented) 682a8599bd8SSage Weil *implemented = 0; 683a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 684a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 685a8599bd8SSage Weil if (!__cap_is_valid(cap)) 686a8599bd8SSage Weil continue; 687a8599bd8SSage Weil dout("__ceph_caps_issued %p cap %p issued %s\n", 688a8599bd8SSage Weil &ci->vfs_inode, cap, ceph_cap_string(cap->issued)); 689a8599bd8SSage Weil have |= cap->issued; 690a8599bd8SSage Weil if (implemented) 691a8599bd8SSage Weil *implemented |= cap->implemented; 692a8599bd8SSage Weil } 693a8599bd8SSage Weil return have; 694a8599bd8SSage Weil } 695a8599bd8SSage Weil 696a8599bd8SSage Weil /* 697a8599bd8SSage Weil * Get cap bits issued by caps other than @ocap 698a8599bd8SSage Weil */ 699a8599bd8SSage Weil int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap) 700a8599bd8SSage Weil { 701a8599bd8SSage Weil int have = ci->i_snap_caps; 702a8599bd8SSage Weil struct ceph_cap *cap; 703a8599bd8SSage Weil struct rb_node *p; 704a8599bd8SSage Weil 705a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 706a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 707a8599bd8SSage Weil if (cap == ocap) 708a8599bd8SSage Weil continue; 709a8599bd8SSage Weil if (!__cap_is_valid(cap)) 710a8599bd8SSage Weil continue; 711a8599bd8SSage Weil have |= cap->issued; 712a8599bd8SSage Weil } 713a8599bd8SSage Weil return have; 714a8599bd8SSage Weil } 715a8599bd8SSage Weil 716a8599bd8SSage Weil /* 717a8599bd8SSage Weil * Move a cap to the end of the LRU (oldest caps at list head, newest 718a8599bd8SSage Weil * at list tail). 719a8599bd8SSage Weil */ 720a8599bd8SSage Weil static void __touch_cap(struct ceph_cap *cap) 721a8599bd8SSage Weil { 722a8599bd8SSage Weil struct ceph_mds_session *s = cap->session; 723a8599bd8SSage Weil 7245dacf091SSage Weil spin_lock(&s->s_cap_lock); 7257c1332b8SSage Weil if (s->s_cap_iterator == NULL) { 726a8599bd8SSage Weil dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap, 727a8599bd8SSage Weil s->s_mds); 728a8599bd8SSage Weil list_move_tail(&cap->session_caps, &s->s_caps); 7295dacf091SSage Weil } else { 7305dacf091SSage Weil dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n", 7315dacf091SSage Weil &cap->ci->vfs_inode, cap, s->s_mds); 7325dacf091SSage Weil } 733a8599bd8SSage Weil spin_unlock(&s->s_cap_lock); 734a8599bd8SSage Weil } 735a8599bd8SSage Weil 736a8599bd8SSage Weil /* 737a8599bd8SSage Weil * Check if we hold the given mask. If so, move the cap(s) to the 738a8599bd8SSage Weil * front of their respective LRUs. (This is the preferred way for 739a8599bd8SSage Weil * callers to check for caps they want.) 740a8599bd8SSage Weil */ 741a8599bd8SSage Weil int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) 742a8599bd8SSage Weil { 743a8599bd8SSage Weil struct ceph_cap *cap; 744a8599bd8SSage Weil struct rb_node *p; 745a8599bd8SSage Weil int have = ci->i_snap_caps; 746a8599bd8SSage Weil 747a8599bd8SSage Weil if ((have & mask) == mask) { 748a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p snap issued %s" 749a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 750a8599bd8SSage Weil ceph_cap_string(have), 751a8599bd8SSage Weil ceph_cap_string(mask)); 752a8599bd8SSage Weil return 1; 753a8599bd8SSage Weil } 754a8599bd8SSage Weil 755a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 756a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 757a8599bd8SSage Weil if (!__cap_is_valid(cap)) 758a8599bd8SSage Weil continue; 759a8599bd8SSage Weil if ((cap->issued & mask) == mask) { 760a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p cap %p issued %s" 761a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, cap, 762a8599bd8SSage Weil ceph_cap_string(cap->issued), 763a8599bd8SSage Weil ceph_cap_string(mask)); 764a8599bd8SSage Weil if (touch) 765a8599bd8SSage Weil __touch_cap(cap); 766a8599bd8SSage Weil return 1; 767a8599bd8SSage Weil } 768a8599bd8SSage Weil 769a8599bd8SSage Weil /* does a combination of caps satisfy mask? */ 770a8599bd8SSage Weil have |= cap->issued; 771a8599bd8SSage Weil if ((have & mask) == mask) { 772a8599bd8SSage Weil dout("__ceph_caps_issued_mask %p combo issued %s" 773a8599bd8SSage Weil " (mask %s)\n", &ci->vfs_inode, 774a8599bd8SSage Weil ceph_cap_string(cap->issued), 775a8599bd8SSage Weil ceph_cap_string(mask)); 776a8599bd8SSage Weil if (touch) { 777a8599bd8SSage Weil struct rb_node *q; 778a8599bd8SSage Weil 77925985edcSLucas De Marchi /* touch this + preceding caps */ 780a8599bd8SSage Weil __touch_cap(cap); 781a8599bd8SSage Weil for (q = rb_first(&ci->i_caps); q != p; 782a8599bd8SSage Weil q = rb_next(q)) { 783a8599bd8SSage Weil cap = rb_entry(q, struct ceph_cap, 784a8599bd8SSage Weil ci_node); 785a8599bd8SSage Weil if (!__cap_is_valid(cap)) 786a8599bd8SSage Weil continue; 787a8599bd8SSage Weil __touch_cap(cap); 788a8599bd8SSage Weil } 789a8599bd8SSage Weil } 790a8599bd8SSage Weil return 1; 791a8599bd8SSage Weil } 792a8599bd8SSage Weil } 793a8599bd8SSage Weil 794a8599bd8SSage Weil return 0; 795a8599bd8SSage Weil } 796a8599bd8SSage Weil 797a8599bd8SSage Weil /* 798a8599bd8SSage Weil * Return true if mask caps are currently being revoked by an MDS. 799a8599bd8SSage Weil */ 800a8599bd8SSage Weil int ceph_caps_revoking(struct ceph_inode_info *ci, int mask) 801a8599bd8SSage Weil { 802a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 803a8599bd8SSage Weil struct ceph_cap *cap; 804a8599bd8SSage Weil struct rb_node *p; 805a8599bd8SSage Weil int ret = 0; 806a8599bd8SSage Weil 807be655596SSage Weil spin_lock(&ci->i_ceph_lock); 808a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 809a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 810a8599bd8SSage Weil if (__cap_is_valid(cap) && 811a8599bd8SSage Weil (cap->implemented & ~cap->issued & mask)) { 812a8599bd8SSage Weil ret = 1; 813a8599bd8SSage Weil break; 814a8599bd8SSage Weil } 815a8599bd8SSage Weil } 816be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 817a8599bd8SSage Weil dout("ceph_caps_revoking %p %s = %d\n", inode, 818a8599bd8SSage Weil ceph_cap_string(mask), ret); 819a8599bd8SSage Weil return ret; 820a8599bd8SSage Weil } 821a8599bd8SSage Weil 822a8599bd8SSage Weil int __ceph_caps_used(struct ceph_inode_info *ci) 823a8599bd8SSage Weil { 824a8599bd8SSage Weil int used = 0; 825a8599bd8SSage Weil if (ci->i_pin_ref) 826a8599bd8SSage Weil used |= CEPH_CAP_PIN; 827a8599bd8SSage Weil if (ci->i_rd_ref) 828a8599bd8SSage Weil used |= CEPH_CAP_FILE_RD; 829a43fb731SSage Weil if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages) 830a8599bd8SSage Weil used |= CEPH_CAP_FILE_CACHE; 831a8599bd8SSage Weil if (ci->i_wr_ref) 832a8599bd8SSage Weil used |= CEPH_CAP_FILE_WR; 833d3d0720dSHenry C Chang if (ci->i_wb_ref || ci->i_wrbuffer_ref) 834a8599bd8SSage Weil used |= CEPH_CAP_FILE_BUFFER; 835a8599bd8SSage Weil return used; 836a8599bd8SSage Weil } 837a8599bd8SSage Weil 838a8599bd8SSage Weil /* 839a8599bd8SSage Weil * wanted, by virtue of open file modes 840a8599bd8SSage Weil */ 841a8599bd8SSage Weil int __ceph_caps_file_wanted(struct ceph_inode_info *ci) 842a8599bd8SSage Weil { 843a8599bd8SSage Weil int want = 0; 844a8599bd8SSage Weil int mode; 84533caad32SSage Weil for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++) 846a8599bd8SSage Weil if (ci->i_nr_by_mode[mode]) 847a8599bd8SSage Weil want |= ceph_caps_for_mode(mode); 848a8599bd8SSage Weil return want; 849a8599bd8SSage Weil } 850a8599bd8SSage Weil 851a8599bd8SSage Weil /* 852a8599bd8SSage Weil * Return caps we have registered with the MDS(s) as 'wanted'. 853a8599bd8SSage Weil */ 854a8599bd8SSage Weil int __ceph_caps_mds_wanted(struct ceph_inode_info *ci) 855a8599bd8SSage Weil { 856a8599bd8SSage Weil struct ceph_cap *cap; 857a8599bd8SSage Weil struct rb_node *p; 858a8599bd8SSage Weil int mds_wanted = 0; 859a8599bd8SSage Weil 860a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 861a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 862a8599bd8SSage Weil if (!__cap_is_valid(cap)) 863a8599bd8SSage Weil continue; 864a8599bd8SSage Weil mds_wanted |= cap->mds_wanted; 865a8599bd8SSage Weil } 866a8599bd8SSage Weil return mds_wanted; 867a8599bd8SSage Weil } 868a8599bd8SSage Weil 869a8599bd8SSage Weil /* 870be655596SSage Weil * called under i_ceph_lock 871a8599bd8SSage Weil */ 872a8599bd8SSage Weil static int __ceph_is_any_caps(struct ceph_inode_info *ci) 873a8599bd8SSage Weil { 874a8599bd8SSage Weil return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0; 875a8599bd8SSage Weil } 876a8599bd8SSage Weil 877a8599bd8SSage Weil /* 878f818a736SSage Weil * Remove a cap. Take steps to deal with a racing iterate_session_caps. 879f818a736SSage Weil * 880be655596SSage Weil * caller should hold i_ceph_lock. 881a6369741SSage Weil * caller will not hold session s_mutex if called from destroy_inode. 882a8599bd8SSage Weil */ 8837c1332b8SSage Weil void __ceph_remove_cap(struct ceph_cap *cap) 884a8599bd8SSage Weil { 885a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 886a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 887640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 8883d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 889f818a736SSage Weil int removed = 0; 890a8599bd8SSage Weil 891a8599bd8SSage Weil dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); 892a8599bd8SSage Weil 8937c1332b8SSage Weil /* remove from session list */ 8947c1332b8SSage Weil spin_lock(&session->s_cap_lock); 8957c1332b8SSage Weil if (session->s_cap_iterator == cap) { 8967c1332b8SSage Weil /* not yet, we are iterating over this very cap */ 8977c1332b8SSage Weil dout("__ceph_remove_cap delaying %p removal from session %p\n", 8987c1332b8SSage Weil cap, cap->session); 8997c1332b8SSage Weil } else { 9007c1332b8SSage Weil list_del_init(&cap->session_caps); 9017c1332b8SSage Weil session->s_nr_caps--; 9027c1332b8SSage Weil cap->session = NULL; 903f818a736SSage Weil removed = 1; 9047c1332b8SSage Weil } 905f818a736SSage Weil /* protect backpointer with s_cap_lock: see iterate_session_caps */ 906f818a736SSage Weil cap->ci = NULL; 9077c1332b8SSage Weil spin_unlock(&session->s_cap_lock); 9087c1332b8SSage Weil 909f818a736SSage Weil /* remove from inode list */ 910f818a736SSage Weil rb_erase(&cap->ci_node, &ci->i_caps); 911f818a736SSage Weil if (ci->i_auth_cap == cap) 912f818a736SSage Weil ci->i_auth_cap = NULL; 913f818a736SSage Weil 914f818a736SSage Weil if (removed) 91537151668SYehuda Sadeh ceph_put_cap(mdsc, cap); 916a8599bd8SSage Weil 917a8599bd8SSage Weil if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) { 918a8599bd8SSage Weil struct ceph_snap_realm *realm = ci->i_snap_realm; 919a8599bd8SSage Weil spin_lock(&realm->inodes_with_caps_lock); 920a8599bd8SSage Weil list_del_init(&ci->i_snap_realm_item); 921a8599bd8SSage Weil ci->i_snap_realm_counter++; 922a8599bd8SSage Weil ci->i_snap_realm = NULL; 923a8599bd8SSage Weil spin_unlock(&realm->inodes_with_caps_lock); 924a8599bd8SSage Weil ceph_put_snap_realm(mdsc, realm); 925a8599bd8SSage Weil } 926a8599bd8SSage Weil if (!__ceph_is_any_real_caps(ci)) 927a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 928a8599bd8SSage Weil } 929a8599bd8SSage Weil 930a8599bd8SSage Weil /* 931a8599bd8SSage Weil * Build and send a cap message to the given MDS. 932a8599bd8SSage Weil * 933a8599bd8SSage Weil * Caller should be holding s_mutex. 934a8599bd8SSage Weil */ 935a8599bd8SSage Weil static int send_cap_msg(struct ceph_mds_session *session, 936a8599bd8SSage Weil u64 ino, u64 cid, int op, 937a8599bd8SSage Weil int caps, int wanted, int dirty, 938a8599bd8SSage Weil u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq, 939a8599bd8SSage Weil u64 size, u64 max_size, 940a8599bd8SSage Weil struct timespec *mtime, struct timespec *atime, 941a8599bd8SSage Weil u64 time_warp_seq, 94205cb11c1SEric W. Biederman kuid_t uid, kgid_t gid, umode_t mode, 943a8599bd8SSage Weil u64 xattr_version, 944a8599bd8SSage Weil struct ceph_buffer *xattrs_buf, 945a8599bd8SSage Weil u64 follows) 946a8599bd8SSage Weil { 947a8599bd8SSage Weil struct ceph_mds_caps *fc; 948a8599bd8SSage Weil struct ceph_msg *msg; 949a8599bd8SSage Weil 950a8599bd8SSage Weil dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" 951a8599bd8SSage Weil " seq %u/%u mseq %u follows %lld size %llu/%llu" 952a8599bd8SSage Weil " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op), 953a8599bd8SSage Weil cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted), 954a8599bd8SSage Weil ceph_cap_string(dirty), 955a8599bd8SSage Weil seq, issue_seq, mseq, follows, size, max_size, 956a8599bd8SSage Weil xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0); 957a8599bd8SSage Weil 958b61c2763SSage Weil msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false); 959a79832f2SSage Weil if (!msg) 960a79832f2SSage Weil return -ENOMEM; 961a8599bd8SSage Weil 9626df058c0SSage Weil msg->hdr.tid = cpu_to_le64(flush_tid); 963a8599bd8SSage Weil 9646df058c0SSage Weil fc = msg->front.iov_base; 965a8599bd8SSage Weil memset(fc, 0, sizeof(*fc)); 966a8599bd8SSage Weil 967a8599bd8SSage Weil fc->cap_id = cpu_to_le64(cid); 968a8599bd8SSage Weil fc->op = cpu_to_le32(op); 969a8599bd8SSage Weil fc->seq = cpu_to_le32(seq); 970a8599bd8SSage Weil fc->issue_seq = cpu_to_le32(issue_seq); 971a8599bd8SSage Weil fc->migrate_seq = cpu_to_le32(mseq); 972a8599bd8SSage Weil fc->caps = cpu_to_le32(caps); 973a8599bd8SSage Weil fc->wanted = cpu_to_le32(wanted); 974a8599bd8SSage Weil fc->dirty = cpu_to_le32(dirty); 975a8599bd8SSage Weil fc->ino = cpu_to_le64(ino); 976a8599bd8SSage Weil fc->snap_follows = cpu_to_le64(follows); 977a8599bd8SSage Weil 978a8599bd8SSage Weil fc->size = cpu_to_le64(size); 979a8599bd8SSage Weil fc->max_size = cpu_to_le64(max_size); 980a8599bd8SSage Weil if (mtime) 981a8599bd8SSage Weil ceph_encode_timespec(&fc->mtime, mtime); 982a8599bd8SSage Weil if (atime) 983a8599bd8SSage Weil ceph_encode_timespec(&fc->atime, atime); 984a8599bd8SSage Weil fc->time_warp_seq = cpu_to_le32(time_warp_seq); 985a8599bd8SSage Weil 98605cb11c1SEric W. Biederman fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid)); 98705cb11c1SEric W. Biederman fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid)); 988a8599bd8SSage Weil fc->mode = cpu_to_le32(mode); 989a8599bd8SSage Weil 990a8599bd8SSage Weil fc->xattr_version = cpu_to_le64(xattr_version); 991a8599bd8SSage Weil if (xattrs_buf) { 992a8599bd8SSage Weil msg->middle = ceph_buffer_get(xattrs_buf); 993a8599bd8SSage Weil fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len); 994a8599bd8SSage Weil msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len); 995a8599bd8SSage Weil } 996a8599bd8SSage Weil 997a8599bd8SSage Weil ceph_con_send(&session->s_con, msg); 998a8599bd8SSage Weil return 0; 999a8599bd8SSage Weil } 1000a8599bd8SSage Weil 1001d40ee0dcSYan, Zheng void __queue_cap_release(struct ceph_mds_session *session, 10023d7ded4dSSage Weil u64 ino, u64 cap_id, u32 migrate_seq, 10033d7ded4dSSage Weil u32 issue_seq) 10043d7ded4dSSage Weil { 10053d7ded4dSSage Weil struct ceph_msg *msg; 10063d7ded4dSSage Weil struct ceph_mds_cap_release *head; 10073d7ded4dSSage Weil struct ceph_mds_cap_item *item; 10083d7ded4dSSage Weil 10093d7ded4dSSage Weil spin_lock(&session->s_cap_lock); 10103d7ded4dSSage Weil BUG_ON(!session->s_num_cap_releases); 10113d7ded4dSSage Weil msg = list_first_entry(&session->s_cap_releases, 10123d7ded4dSSage Weil struct ceph_msg, list_head); 10133d7ded4dSSage Weil 10143d7ded4dSSage Weil dout(" adding %llx release to mds%d msg %p (%d left)\n", 10153d7ded4dSSage Weil ino, session->s_mds, msg, session->s_num_cap_releases); 10163d7ded4dSSage Weil 10173d7ded4dSSage Weil BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE); 10183d7ded4dSSage Weil head = msg->front.iov_base; 1019b905a7f8SWei Yongjun le32_add_cpu(&head->num, 1); 10203d7ded4dSSage Weil item = msg->front.iov_base + msg->front.iov_len; 10213d7ded4dSSage Weil item->ino = cpu_to_le64(ino); 10223d7ded4dSSage Weil item->cap_id = cpu_to_le64(cap_id); 10233d7ded4dSSage Weil item->migrate_seq = cpu_to_le32(migrate_seq); 10243d7ded4dSSage Weil item->seq = cpu_to_le32(issue_seq); 10253d7ded4dSSage Weil 10263d7ded4dSSage Weil session->s_num_cap_releases--; 10273d7ded4dSSage Weil 10283d7ded4dSSage Weil msg->front.iov_len += sizeof(*item); 10293d7ded4dSSage Weil if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) { 10303d7ded4dSSage Weil dout(" release msg %p full\n", msg); 10313d7ded4dSSage Weil list_move_tail(&msg->list_head, &session->s_cap_releases_done); 10323d7ded4dSSage Weil } else { 10333d7ded4dSSage Weil dout(" release msg %p at %d/%d (%d)\n", msg, 10343d7ded4dSSage Weil (int)le32_to_cpu(head->num), 10353d7ded4dSSage Weil (int)CEPH_CAPS_PER_RELEASE, 10363d7ded4dSSage Weil (int)msg->front.iov_len); 10373d7ded4dSSage Weil } 10383d7ded4dSSage Weil spin_unlock(&session->s_cap_lock); 10393d7ded4dSSage Weil } 10403d7ded4dSSage Weil 1041a8599bd8SSage Weil /* 1042a6369741SSage Weil * Queue cap releases when an inode is dropped from our cache. Since 1043be655596SSage Weil * inode is about to be destroyed, there is no need for i_ceph_lock. 1044a8599bd8SSage Weil */ 1045a8599bd8SSage Weil void ceph_queue_caps_release(struct inode *inode) 1046a8599bd8SSage Weil { 1047a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1048a8599bd8SSage Weil struct rb_node *p; 1049a8599bd8SSage Weil 1050a8599bd8SSage Weil p = rb_first(&ci->i_caps); 1051a8599bd8SSage Weil while (p) { 1052a8599bd8SSage Weil struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); 1053a8599bd8SSage Weil struct ceph_mds_session *session = cap->session; 1054a8599bd8SSage Weil 10553d7ded4dSSage Weil __queue_cap_release(session, ceph_ino(inode), cap->cap_id, 10563d7ded4dSSage Weil cap->mseq, cap->issue_seq); 1057a8599bd8SSage Weil p = rb_next(p); 10587c1332b8SSage Weil __ceph_remove_cap(cap); 1059a8599bd8SSage Weil } 1060a8599bd8SSage Weil } 1061a8599bd8SSage Weil 1062a8599bd8SSage Weil /* 1063a8599bd8SSage Weil * Send a cap msg on the given inode. Update our caps state, then 1064be655596SSage Weil * drop i_ceph_lock and send the message. 1065a8599bd8SSage Weil * 1066a8599bd8SSage Weil * Make note of max_size reported/requested from mds, revoked caps 1067a8599bd8SSage Weil * that have now been implemented. 1068a8599bd8SSage Weil * 1069a8599bd8SSage Weil * Make half-hearted attempt ot to invalidate page cache if we are 1070a8599bd8SSage Weil * dropping RDCACHE. Note that this will leave behind locked pages 1071a8599bd8SSage Weil * that we'll then need to deal with elsewhere. 1072a8599bd8SSage Weil * 1073a8599bd8SSage Weil * Return non-zero if delayed release, or we experienced an error 1074a8599bd8SSage Weil * such that the caller should requeue + retry later. 1075a8599bd8SSage Weil * 1076be655596SSage Weil * called with i_ceph_lock, then drops it. 1077a8599bd8SSage Weil * caller should hold snap_rwsem (read), s_mutex. 1078a8599bd8SSage Weil */ 1079a8599bd8SSage Weil static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, 1080a8599bd8SSage Weil int op, int used, int want, int retain, int flushing, 1081a8599bd8SSage Weil unsigned *pflush_tid) 1082be655596SSage Weil __releases(cap->ci->i_ceph_lock) 1083a8599bd8SSage Weil { 1084a8599bd8SSage Weil struct ceph_inode_info *ci = cap->ci; 1085a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1086a8599bd8SSage Weil u64 cap_id = cap->cap_id; 108768c28323SSage Weil int held, revoking, dropping, keep; 1088a8599bd8SSage Weil u64 seq, issue_seq, mseq, time_warp_seq, follows; 1089a8599bd8SSage Weil u64 size, max_size; 1090a8599bd8SSage Weil struct timespec mtime, atime; 1091a8599bd8SSage Weil int wake = 0; 10925706b27dSAl Viro umode_t mode; 109305cb11c1SEric W. Biederman kuid_t uid; 109405cb11c1SEric W. Biederman kgid_t gid; 1095a8599bd8SSage Weil struct ceph_mds_session *session; 1096a8599bd8SSage Weil u64 xattr_version = 0; 1097082afec9SSage Weil struct ceph_buffer *xattr_blob = NULL; 1098a8599bd8SSage Weil int delayed = 0; 1099a8599bd8SSage Weil u64 flush_tid = 0; 1100a8599bd8SSage Weil int i; 1101a8599bd8SSage Weil int ret; 1102a8599bd8SSage Weil 110368c28323SSage Weil held = cap->issued | cap->implemented; 110468c28323SSage Weil revoking = cap->implemented & ~cap->issued; 110568c28323SSage Weil retain &= ~revoking; 110668c28323SSage Weil dropping = cap->issued & ~retain; 110768c28323SSage Weil 1108a8599bd8SSage Weil dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", 1109a8599bd8SSage Weil inode, cap, cap->session, 1110a8599bd8SSage Weil ceph_cap_string(held), ceph_cap_string(held & retain), 1111a8599bd8SSage Weil ceph_cap_string(revoking)); 1112a8599bd8SSage Weil BUG_ON((retain & CEPH_CAP_PIN) == 0); 1113a8599bd8SSage Weil 1114a8599bd8SSage Weil session = cap->session; 1115a8599bd8SSage Weil 1116a8599bd8SSage Weil /* don't release wanted unless we've waited a bit. */ 1117a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1118a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_min)) { 1119a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", 1120a8599bd8SSage Weil ceph_cap_string(cap->issued), 1121a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1122a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1123a8599bd8SSage Weil ceph_cap_string(want)); 1124a8599bd8SSage Weil want |= cap->mds_wanted; 1125a8599bd8SSage Weil retain |= cap->issued; 1126a8599bd8SSage Weil delayed = 1; 1127a8599bd8SSage Weil } 1128a8599bd8SSage Weil ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); 1129a8599bd8SSage Weil 1130a8599bd8SSage Weil cap->issued &= retain; /* drop bits we don't want */ 1131a8599bd8SSage Weil if (cap->implemented & ~cap->issued) { 1132a8599bd8SSage Weil /* 1133a8599bd8SSage Weil * Wake up any waiters on wanted -> needed transition. 1134a8599bd8SSage Weil * This is due to the weird transition from buffered 1135a8599bd8SSage Weil * to sync IO... we need to flush dirty pages _before_ 1136a8599bd8SSage Weil * allowing sync writes to avoid reordering. 1137a8599bd8SSage Weil */ 1138a8599bd8SSage Weil wake = 1; 1139a8599bd8SSage Weil } 1140a8599bd8SSage Weil cap->implemented &= cap->issued | used; 1141a8599bd8SSage Weil cap->mds_wanted = want; 1142a8599bd8SSage Weil 1143a8599bd8SSage Weil if (flushing) { 1144a8599bd8SSage Weil /* 1145a8599bd8SSage Weil * assign a tid for flush operations so we can avoid 1146a8599bd8SSage Weil * flush1 -> dirty1 -> flush2 -> flushack1 -> mark 1147a8599bd8SSage Weil * clean type races. track latest tid for every bit 1148a8599bd8SSage Weil * so we can handle flush AxFw, flush Fw, and have the 1149a8599bd8SSage Weil * first ack clean Ax. 1150a8599bd8SSage Weil */ 1151a8599bd8SSage Weil flush_tid = ++ci->i_cap_flush_last_tid; 1152a8599bd8SSage Weil if (pflush_tid) 1153a8599bd8SSage Weil *pflush_tid = flush_tid; 1154a8599bd8SSage Weil dout(" cap_flush_tid %d\n", (int)flush_tid); 1155a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1156a8599bd8SSage Weil if (flushing & (1 << i)) 1157a8599bd8SSage Weil ci->i_cap_flush_tid[i] = flush_tid; 11587d8cb26dSSage Weil 11597d8cb26dSSage Weil follows = ci->i_head_snapc->seq; 11607d8cb26dSSage Weil } else { 11617d8cb26dSSage Weil follows = 0; 1162a8599bd8SSage Weil } 1163a8599bd8SSage Weil 1164a8599bd8SSage Weil keep = cap->implemented; 1165a8599bd8SSage Weil seq = cap->seq; 1166a8599bd8SSage Weil issue_seq = cap->issue_seq; 1167a8599bd8SSage Weil mseq = cap->mseq; 1168a8599bd8SSage Weil size = inode->i_size; 1169a8599bd8SSage Weil ci->i_reported_size = size; 1170a8599bd8SSage Weil max_size = ci->i_wanted_max_size; 1171a8599bd8SSage Weil ci->i_requested_max_size = max_size; 1172a8599bd8SSage Weil mtime = inode->i_mtime; 1173a8599bd8SSage Weil atime = inode->i_atime; 1174a8599bd8SSage Weil time_warp_seq = ci->i_time_warp_seq; 1175a8599bd8SSage Weil uid = inode->i_uid; 1176a8599bd8SSage Weil gid = inode->i_gid; 1177a8599bd8SSage Weil mode = inode->i_mode; 1178a8599bd8SSage Weil 1179082afec9SSage Weil if (flushing & CEPH_CAP_XATTR_EXCL) { 1180a8599bd8SSage Weil __ceph_build_xattrs_blob(ci); 1181082afec9SSage Weil xattr_blob = ci->i_xattrs.blob; 1182082afec9SSage Weil xattr_version = ci->i_xattrs.version; 1183a8599bd8SSage Weil } 1184a8599bd8SSage Weil 1185be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1186a8599bd8SSage Weil 1187a8599bd8SSage Weil ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id, 1188a8599bd8SSage Weil op, keep, want, flushing, seq, flush_tid, issue_seq, mseq, 1189a8599bd8SSage Weil size, max_size, &mtime, &atime, time_warp_seq, 1190082afec9SSage Weil uid, gid, mode, xattr_version, xattr_blob, 1191a8599bd8SSage Weil follows); 1192a8599bd8SSage Weil if (ret < 0) { 1193a8599bd8SSage Weil dout("error sending cap msg, must requeue %p\n", inode); 1194a8599bd8SSage Weil delayed = 1; 1195a8599bd8SSage Weil } 1196a8599bd8SSage Weil 1197a8599bd8SSage Weil if (wake) 119803066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 1199a8599bd8SSage Weil 1200a8599bd8SSage Weil return delayed; 1201a8599bd8SSage Weil } 1202a8599bd8SSage Weil 1203a8599bd8SSage Weil /* 1204a8599bd8SSage Weil * When a snapshot is taken, clients accumulate dirty metadata on 1205a8599bd8SSage Weil * inodes with capabilities in ceph_cap_snaps to describe the file 1206a8599bd8SSage Weil * state at the time the snapshot was taken. This must be flushed 1207a8599bd8SSage Weil * asynchronously back to the MDS once sync writes complete and dirty 1208a8599bd8SSage Weil * data is written out. 1209a8599bd8SSage Weil * 1210e835124cSSage Weil * Unless @again is true, skip cap_snaps that were already sent to 1211e835124cSSage Weil * the MDS (i.e., during this session). 1212e835124cSSage Weil * 1213be655596SSage Weil * Called under i_ceph_lock. Takes s_mutex as needed. 1214a8599bd8SSage Weil */ 1215a8599bd8SSage Weil void __ceph_flush_snaps(struct ceph_inode_info *ci, 1216e835124cSSage Weil struct ceph_mds_session **psession, 1217e835124cSSage Weil int again) 1218be655596SSage Weil __releases(ci->i_ceph_lock) 1219be655596SSage Weil __acquires(ci->i_ceph_lock) 1220a8599bd8SSage Weil { 1221a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1222a8599bd8SSage Weil int mds; 1223a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1224a8599bd8SSage Weil u32 mseq; 12253d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 1226a8599bd8SSage Weil struct ceph_mds_session *session = NULL; /* if session != NULL, we hold 1227a8599bd8SSage Weil session->s_mutex */ 1228a8599bd8SSage Weil u64 next_follows = 0; /* keep track of how far we've gotten through the 1229a8599bd8SSage Weil i_cap_snaps list, and skip these entries next time 1230a8599bd8SSage Weil around to avoid an infinite loop */ 1231a8599bd8SSage Weil 1232a8599bd8SSage Weil if (psession) 1233a8599bd8SSage Weil session = *psession; 1234a8599bd8SSage Weil 1235a8599bd8SSage Weil dout("__flush_snaps %p\n", inode); 1236a8599bd8SSage Weil retry: 1237a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 1238a8599bd8SSage Weil /* avoid an infiniute loop after retry */ 1239a8599bd8SSage Weil if (capsnap->follows < next_follows) 1240a8599bd8SSage Weil continue; 1241a8599bd8SSage Weil /* 1242a8599bd8SSage Weil * we need to wait for sync writes to complete and for dirty 1243a8599bd8SSage Weil * pages to be written out. 1244a8599bd8SSage Weil */ 1245a8599bd8SSage Weil if (capsnap->dirty_pages || capsnap->writing) 1246cfc0bf66SSage Weil break; 1247a8599bd8SSage Weil 1248819ccbfaSSage Weil /* 1249819ccbfaSSage Weil * if cap writeback already occurred, we should have dropped 1250819ccbfaSSage Weil * the capsnap in ceph_put_wrbuffer_cap_refs. 1251819ccbfaSSage Weil */ 1252819ccbfaSSage Weil BUG_ON(capsnap->dirty == 0); 1253819ccbfaSSage Weil 1254a8599bd8SSage Weil /* pick mds, take s_mutex */ 1255ca81f3f6SSage Weil if (ci->i_auth_cap == NULL) { 1256ca81f3f6SSage Weil dout("no auth cap (migrating?), doing nothing\n"); 1257ca81f3f6SSage Weil goto out; 1258ca81f3f6SSage Weil } 1259e835124cSSage Weil 1260e835124cSSage Weil /* only flush each capsnap once */ 1261e835124cSSage Weil if (!again && !list_empty(&capsnap->flushing_item)) { 1262e835124cSSage Weil dout("already flushed %p, skipping\n", capsnap); 1263e835124cSSage Weil continue; 1264e835124cSSage Weil } 1265e835124cSSage Weil 1266ca81f3f6SSage Weil mds = ci->i_auth_cap->session->s_mds; 1267ca81f3f6SSage Weil mseq = ci->i_auth_cap->mseq; 1268ca81f3f6SSage Weil 1269a8599bd8SSage Weil if (session && session->s_mds != mds) { 1270a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1271a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1272a8599bd8SSage Weil ceph_put_mds_session(session); 1273a8599bd8SSage Weil session = NULL; 1274a8599bd8SSage Weil } 1275a8599bd8SSage Weil if (!session) { 1276be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1277a8599bd8SSage Weil mutex_lock(&mdsc->mutex); 1278a8599bd8SSage Weil session = __ceph_lookup_mds_session(mdsc, mds); 1279a8599bd8SSage Weil mutex_unlock(&mdsc->mutex); 1280a8599bd8SSage Weil if (session) { 1281a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1282a8599bd8SSage Weil session); 1283a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1284a8599bd8SSage Weil } 1285a8599bd8SSage Weil /* 1286a8599bd8SSage Weil * if session == NULL, we raced against a cap 1287ca81f3f6SSage Weil * deletion or migration. retry, and we'll 1288ca81f3f6SSage Weil * get a better @mds value next time. 1289a8599bd8SSage Weil */ 1290be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1291a8599bd8SSage Weil goto retry; 1292a8599bd8SSage Weil } 1293a8599bd8SSage Weil 1294a8599bd8SSage Weil capsnap->flush_tid = ++ci->i_cap_flush_last_tid; 1295a8599bd8SSage Weil atomic_inc(&capsnap->nref); 1296a8599bd8SSage Weil if (!list_empty(&capsnap->flushing_item)) 1297a8599bd8SSage Weil list_del_init(&capsnap->flushing_item); 1298a8599bd8SSage Weil list_add_tail(&capsnap->flushing_item, 1299a8599bd8SSage Weil &session->s_cap_snaps_flushing); 1300be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1301a8599bd8SSage Weil 1302cfc0bf66SSage Weil dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n", 1303cfc0bf66SSage Weil inode, capsnap, capsnap->follows, capsnap->flush_tid); 1304a8599bd8SSage Weil send_cap_msg(session, ceph_vino(inode).ino, 0, 1305a8599bd8SSage Weil CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0, 1306a8599bd8SSage Weil capsnap->dirty, 0, capsnap->flush_tid, 0, mseq, 1307a8599bd8SSage Weil capsnap->size, 0, 1308a8599bd8SSage Weil &capsnap->mtime, &capsnap->atime, 1309a8599bd8SSage Weil capsnap->time_warp_seq, 1310a8599bd8SSage Weil capsnap->uid, capsnap->gid, capsnap->mode, 13114a625be4SSage Weil capsnap->xattr_version, capsnap->xattr_blob, 1312a8599bd8SSage Weil capsnap->follows); 1313a8599bd8SSage Weil 1314a8599bd8SSage Weil next_follows = capsnap->follows + 1; 1315a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 1316a8599bd8SSage Weil 1317be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1318a8599bd8SSage Weil goto retry; 1319a8599bd8SSage Weil } 1320a8599bd8SSage Weil 1321a8599bd8SSage Weil /* we flushed them all; remove this inode from the queue */ 1322a8599bd8SSage Weil spin_lock(&mdsc->snap_flush_lock); 1323a8599bd8SSage Weil list_del_init(&ci->i_snap_flush_item); 1324a8599bd8SSage Weil spin_unlock(&mdsc->snap_flush_lock); 1325a8599bd8SSage Weil 1326ca81f3f6SSage Weil out: 1327a8599bd8SSage Weil if (psession) 1328a8599bd8SSage Weil *psession = session; 1329a8599bd8SSage Weil else if (session) { 1330a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1331a8599bd8SSage Weil ceph_put_mds_session(session); 1332a8599bd8SSage Weil } 1333a8599bd8SSage Weil } 1334a8599bd8SSage Weil 1335a8599bd8SSage Weil static void ceph_flush_snaps(struct ceph_inode_info *ci) 1336a8599bd8SSage Weil { 1337be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1338e835124cSSage Weil __ceph_flush_snaps(ci, NULL, 0); 1339be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1340a8599bd8SSage Weil } 1341a8599bd8SSage Weil 1342a8599bd8SSage Weil /* 1343fca65b4aSSage Weil * Mark caps dirty. If inode is newly dirty, return the dirty flags. 1344fca65b4aSSage Weil * Caller is then responsible for calling __mark_inode_dirty with the 1345fca65b4aSSage Weil * returned flags value. 134676e3b390SSage Weil */ 1347fca65b4aSSage Weil int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask) 134876e3b390SSage Weil { 1349640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 13503d14c5d2SYehuda Sadeh ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 135176e3b390SSage Weil struct inode *inode = &ci->vfs_inode; 135276e3b390SSage Weil int was = ci->i_dirty_caps; 135376e3b390SSage Weil int dirty = 0; 135476e3b390SSage Weil 135576e3b390SSage Weil dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, 135676e3b390SSage Weil ceph_cap_string(mask), ceph_cap_string(was), 135776e3b390SSage Weil ceph_cap_string(was | mask)); 135876e3b390SSage Weil ci->i_dirty_caps |= mask; 135976e3b390SSage Weil if (was == 0) { 13607d8cb26dSSage Weil if (!ci->i_head_snapc) 13617d8cb26dSSage Weil ci->i_head_snapc = ceph_get_snap_context( 13627d8cb26dSSage Weil ci->i_snap_realm->cached_context); 13630685235fSYan, Zheng dout(" inode %p now dirty snapc %p auth cap %p\n", 13640685235fSYan, Zheng &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap); 136576e3b390SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 136676e3b390SSage Weil spin_lock(&mdsc->cap_dirty_lock); 13670685235fSYan, Zheng if (ci->i_auth_cap) 136876e3b390SSage Weil list_add(&ci->i_dirty_item, &mdsc->cap_dirty); 13690685235fSYan, Zheng else 13700685235fSYan, Zheng list_add(&ci->i_dirty_item, 13710685235fSYan, Zheng &mdsc->cap_dirty_migrating); 137276e3b390SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 137376e3b390SSage Weil if (ci->i_flushing_caps == 0) { 13743772d26dSSage Weil ihold(inode); 137576e3b390SSage Weil dirty |= I_DIRTY_SYNC; 137676e3b390SSage Weil } 137776e3b390SSage Weil } 137876e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 137976e3b390SSage Weil if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && 138076e3b390SSage Weil (mask & CEPH_CAP_FILE_BUFFER)) 138176e3b390SSage Weil dirty |= I_DIRTY_DATASYNC; 138276e3b390SSage Weil __cap_delay_requeue(mdsc, ci); 1383fca65b4aSSage Weil return dirty; 138476e3b390SSage Weil } 138576e3b390SSage Weil 138676e3b390SSage Weil /* 1387a8599bd8SSage Weil * Add dirty inode to the flushing list. Assigned a seq number so we 1388a8599bd8SSage Weil * can wait for caps to flush without starving. 1389cdc35f96SSage Weil * 1390be655596SSage Weil * Called under i_ceph_lock. 1391a8599bd8SSage Weil */ 1392cdc35f96SSage Weil static int __mark_caps_flushing(struct inode *inode, 1393a8599bd8SSage Weil struct ceph_mds_session *session) 1394a8599bd8SSage Weil { 13953d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1396a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1397cdc35f96SSage Weil int flushing; 1398a8599bd8SSage Weil 1399cdc35f96SSage Weil BUG_ON(ci->i_dirty_caps == 0); 1400a8599bd8SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 1401cdc35f96SSage Weil 1402cdc35f96SSage Weil flushing = ci->i_dirty_caps; 1403cdc35f96SSage Weil dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", 1404cdc35f96SSage Weil ceph_cap_string(flushing), 1405cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps), 1406cdc35f96SSage Weil ceph_cap_string(ci->i_flushing_caps | flushing)); 1407cdc35f96SSage Weil ci->i_flushing_caps |= flushing; 1408cdc35f96SSage Weil ci->i_dirty_caps = 0; 1409afcdaea3SSage Weil dout(" inode %p now !dirty\n", inode); 1410cdc35f96SSage Weil 1411a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 1412cdc35f96SSage Weil list_del_init(&ci->i_dirty_item); 1413afcdaea3SSage Weil 1414afcdaea3SSage Weil ci->i_cap_flush_seq = ++mdsc->cap_flush_seq; 1415afcdaea3SSage Weil if (list_empty(&ci->i_flushing_item)) { 1416a8599bd8SSage Weil list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1417a8599bd8SSage Weil mdsc->num_cap_flushing++; 1418afcdaea3SSage Weil dout(" inode %p now flushing seq %lld\n", inode, 1419afcdaea3SSage Weil ci->i_cap_flush_seq); 1420afcdaea3SSage Weil } else { 1421afcdaea3SSage Weil list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing); 1422afcdaea3SSage Weil dout(" inode %p now flushing (more) seq %lld\n", inode, 1423a8599bd8SSage Weil ci->i_cap_flush_seq); 1424a8599bd8SSage Weil } 1425a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 1426cdc35f96SSage Weil 1427cdc35f96SSage Weil return flushing; 1428a8599bd8SSage Weil } 1429a8599bd8SSage Weil 1430a8599bd8SSage Weil /* 14315ecad6fdSSage Weil * try to invalidate mapping pages without blocking. 14325ecad6fdSSage Weil */ 14335ecad6fdSSage Weil static int try_nonblocking_invalidate(struct inode *inode) 14345ecad6fdSSage Weil { 14355ecad6fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 14365ecad6fdSSage Weil u32 invalidating_gen = ci->i_rdcache_gen; 14375ecad6fdSSage Weil 1438be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 14395ecad6fdSSage Weil invalidate_mapping_pages(&inode->i_data, 0, -1); 1440be655596SSage Weil spin_lock(&ci->i_ceph_lock); 14415ecad6fdSSage Weil 144218a38193SSage Weil if (inode->i_data.nrpages == 0 && 14435ecad6fdSSage Weil invalidating_gen == ci->i_rdcache_gen) { 14445ecad6fdSSage Weil /* success. */ 14455ecad6fdSSage Weil dout("try_nonblocking_invalidate %p success\n", inode); 1446cd045cb4SSage Weil /* save any racing async invalidate some trouble */ 1447cd045cb4SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen - 1; 14485ecad6fdSSage Weil return 0; 14495ecad6fdSSage Weil } 14505ecad6fdSSage Weil dout("try_nonblocking_invalidate %p failed\n", inode); 14515ecad6fdSSage Weil return -1; 14525ecad6fdSSage Weil } 14535ecad6fdSSage Weil 14545ecad6fdSSage Weil /* 1455a8599bd8SSage Weil * Swiss army knife function to examine currently used and wanted 1456a8599bd8SSage Weil * versus held caps. Release, flush, ack revoked caps to mds as 1457a8599bd8SSage Weil * appropriate. 1458a8599bd8SSage Weil * 1459a8599bd8SSage Weil * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay 1460a8599bd8SSage Weil * cap release further. 1461a8599bd8SSage Weil * CHECK_CAPS_AUTHONLY - we should only check the auth cap 1462a8599bd8SSage Weil * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without 1463a8599bd8SSage Weil * further delay. 1464a8599bd8SSage Weil */ 1465a8599bd8SSage Weil void ceph_check_caps(struct ceph_inode_info *ci, int flags, 1466a8599bd8SSage Weil struct ceph_mds_session *session) 1467a8599bd8SSage Weil { 14683d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 14693d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 1470a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1471a8599bd8SSage Weil struct ceph_cap *cap; 1472395c312bSYan, Zheng int file_wanted, used, cap_used; 1473a8599bd8SSage Weil int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ 1474cbd03635SSage Weil int issued, implemented, want, retain, revoking, flushing = 0; 1475a8599bd8SSage Weil int mds = -1; /* keep track of how far we've gone through i_caps list 1476a8599bd8SSage Weil to avoid an infinite loop on retry */ 1477a8599bd8SSage Weil struct rb_node *p; 1478a8599bd8SSage Weil int tried_invalidate = 0; 1479a8599bd8SSage Weil int delayed = 0, sent = 0, force_requeue = 0, num; 1480cbd03635SSage Weil int queue_invalidate = 0; 1481a8599bd8SSage Weil int is_delayed = flags & CHECK_CAPS_NODELAY; 1482a8599bd8SSage Weil 1483a8599bd8SSage Weil /* if we are unmounting, flush any unused caps immediately. */ 1484a8599bd8SSage Weil if (mdsc->stopping) 1485a8599bd8SSage Weil is_delayed = 1; 1486a8599bd8SSage Weil 1487be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1488a8599bd8SSage Weil 1489a8599bd8SSage Weil if (ci->i_ceph_flags & CEPH_I_FLUSH) 1490a8599bd8SSage Weil flags |= CHECK_CAPS_FLUSH; 1491a8599bd8SSage Weil 1492a8599bd8SSage Weil /* flush snaps first time around only */ 1493a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) 1494e835124cSSage Weil __ceph_flush_snaps(ci, &session, 0); 1495a8599bd8SSage Weil goto retry_locked; 1496a8599bd8SSage Weil retry: 1497be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1498a8599bd8SSage Weil retry_locked: 1499a8599bd8SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 1500a8599bd8SSage Weil used = __ceph_caps_used(ci); 1501a8599bd8SSage Weil want = file_wanted | used; 1502cbd03635SSage Weil issued = __ceph_caps_issued(ci, &implemented); 1503cbd03635SSage Weil revoking = implemented & ~issued; 1504a8599bd8SSage Weil 1505a8599bd8SSage Weil retain = want | CEPH_CAP_PIN; 1506a8599bd8SSage Weil if (!mdsc->stopping && inode->i_nlink > 0) { 1507a8599bd8SSage Weil if (want) { 1508a8599bd8SSage Weil retain |= CEPH_CAP_ANY; /* be greedy */ 1509a8599bd8SSage Weil } else { 1510a8599bd8SSage Weil retain |= CEPH_CAP_ANY_SHARED; 1511a8599bd8SSage Weil /* 1512a8599bd8SSage Weil * keep RD only if we didn't have the file open RW, 1513a8599bd8SSage Weil * because then the mds would revoke it anyway to 1514a8599bd8SSage Weil * journal max_size=0. 1515a8599bd8SSage Weil */ 1516a8599bd8SSage Weil if (ci->i_max_size == 0) 1517a8599bd8SSage Weil retain |= CEPH_CAP_ANY_RD; 1518a8599bd8SSage Weil } 1519a8599bd8SSage Weil } 1520a8599bd8SSage Weil 1521a8599bd8SSage Weil dout("check_caps %p file_want %s used %s dirty %s flushing %s" 1522cbd03635SSage Weil " issued %s revoking %s retain %s %s%s%s\n", inode, 1523a8599bd8SSage Weil ceph_cap_string(file_wanted), 1524a8599bd8SSage Weil ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), 1525a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps), 1526cbd03635SSage Weil ceph_cap_string(issued), ceph_cap_string(revoking), 1527a8599bd8SSage Weil ceph_cap_string(retain), 1528a8599bd8SSage Weil (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", 1529a8599bd8SSage Weil (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", 1530a8599bd8SSage Weil (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); 1531a8599bd8SSage Weil 1532a8599bd8SSage Weil /* 1533a8599bd8SSage Weil * If we no longer need to hold onto old our caps, and we may 1534a8599bd8SSage Weil * have cached pages, but don't want them, then try to invalidate. 1535a8599bd8SSage Weil * If we fail, it's because pages are locked.... try again later. 1536a8599bd8SSage Weil */ 1537a8599bd8SSage Weil if ((!is_delayed || mdsc->stopping) && 1538a8599bd8SSage Weil ci->i_wrbuffer_ref == 0 && /* no dirty pages... */ 153993afd449SSage Weil inode->i_data.nrpages && /* have cached pages */ 1540cbd03635SSage Weil (file_wanted == 0 || /* no open files */ 15412962507cSSage Weil (revoking & (CEPH_CAP_FILE_CACHE| 15422962507cSSage Weil CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */ 1543a8599bd8SSage Weil !tried_invalidate) { 1544a8599bd8SSage Weil dout("check_caps trying to invalidate on %p\n", inode); 15455ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) < 0) { 15462962507cSSage Weil if (revoking & (CEPH_CAP_FILE_CACHE| 15472962507cSSage Weil CEPH_CAP_FILE_LAZYIO)) { 1548cbd03635SSage Weil dout("check_caps queuing invalidate\n"); 1549cbd03635SSage Weil queue_invalidate = 1; 1550cbd03635SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 1551a8599bd8SSage Weil } else { 1552a8599bd8SSage Weil dout("check_caps failed to invalidate pages\n"); 1553a8599bd8SSage Weil /* we failed to invalidate pages. check these 1554a8599bd8SSage Weil caps again later. */ 1555a8599bd8SSage Weil force_requeue = 1; 1556a8599bd8SSage Weil __cap_set_timeouts(mdsc, ci); 1557a8599bd8SSage Weil } 15585ecad6fdSSage Weil } 1559a8599bd8SSage Weil tried_invalidate = 1; 1560a8599bd8SSage Weil goto retry_locked; 1561a8599bd8SSage Weil } 1562a8599bd8SSage Weil 1563a8599bd8SSage Weil num = 0; 1564a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 1565a8599bd8SSage Weil cap = rb_entry(p, struct ceph_cap, ci_node); 1566a8599bd8SSage Weil num++; 1567a8599bd8SSage Weil 1568a8599bd8SSage Weil /* avoid looping forever */ 1569a8599bd8SSage Weil if (mds >= cap->mds || 1570a8599bd8SSage Weil ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) 1571a8599bd8SSage Weil continue; 1572a8599bd8SSage Weil 1573a8599bd8SSage Weil /* NOTE: no side-effects allowed, until we take s_mutex */ 1574a8599bd8SSage Weil 1575395c312bSYan, Zheng cap_used = used; 1576395c312bSYan, Zheng if (ci->i_auth_cap && cap != ci->i_auth_cap) 1577395c312bSYan, Zheng cap_used &= ~ci->i_auth_cap->issued; 1578395c312bSYan, Zheng 1579a8599bd8SSage Weil revoking = cap->implemented & ~cap->issued; 1580395c312bSYan, Zheng dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n", 1581088b3f5eSSage Weil cap->mds, cap, ceph_cap_string(cap->issued), 1582395c312bSYan, Zheng ceph_cap_string(cap_used), 1583088b3f5eSSage Weil ceph_cap_string(cap->implemented), 1584a8599bd8SSage Weil ceph_cap_string(revoking)); 1585a8599bd8SSage Weil 1586a8599bd8SSage Weil if (cap == ci->i_auth_cap && 1587a8599bd8SSage Weil (cap->issued & CEPH_CAP_FILE_WR)) { 1588a8599bd8SSage Weil /* request larger max_size from MDS? */ 1589a8599bd8SSage Weil if (ci->i_wanted_max_size > ci->i_max_size && 1590a8599bd8SSage Weil ci->i_wanted_max_size > ci->i_requested_max_size) { 1591a8599bd8SSage Weil dout("requesting new max_size\n"); 1592a8599bd8SSage Weil goto ack; 1593a8599bd8SSage Weil } 1594a8599bd8SSage Weil 1595a8599bd8SSage Weil /* approaching file_max? */ 1596a8599bd8SSage Weil if ((inode->i_size << 1) >= ci->i_max_size && 1597a8599bd8SSage Weil (ci->i_reported_size << 1) < ci->i_max_size) { 1598a8599bd8SSage Weil dout("i_size approaching max_size\n"); 1599a8599bd8SSage Weil goto ack; 1600a8599bd8SSage Weil } 1601a8599bd8SSage Weil } 1602a8599bd8SSage Weil /* flush anything dirty? */ 1603a8599bd8SSage Weil if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) && 1604a8599bd8SSage Weil ci->i_dirty_caps) { 1605a8599bd8SSage Weil dout("flushing dirty caps\n"); 1606a8599bd8SSage Weil goto ack; 1607a8599bd8SSage Weil } 1608a8599bd8SSage Weil 1609a8599bd8SSage Weil /* completed revocation? going down and there are no caps? */ 1610395c312bSYan, Zheng if (revoking && (revoking & cap_used) == 0) { 1611a8599bd8SSage Weil dout("completed revocation of %s\n", 1612a8599bd8SSage Weil ceph_cap_string(cap->implemented & ~cap->issued)); 1613a8599bd8SSage Weil goto ack; 1614a8599bd8SSage Weil } 1615a8599bd8SSage Weil 1616a8599bd8SSage Weil /* want more caps from mds? */ 1617a8599bd8SSage Weil if (want & ~(cap->mds_wanted | cap->issued)) 1618a8599bd8SSage Weil goto ack; 1619a8599bd8SSage Weil 1620a8599bd8SSage Weil /* things we might delay */ 1621a8599bd8SSage Weil if ((cap->issued & ~retain) == 0 && 1622a8599bd8SSage Weil cap->mds_wanted == want) 1623a8599bd8SSage Weil continue; /* nope, all good */ 1624a8599bd8SSage Weil 1625a8599bd8SSage Weil if (is_delayed) 1626a8599bd8SSage Weil goto ack; 1627a8599bd8SSage Weil 1628a8599bd8SSage Weil /* delay? */ 1629a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && 1630a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) { 1631a8599bd8SSage Weil dout(" delaying issued %s -> %s, wanted %s -> %s\n", 1632a8599bd8SSage Weil ceph_cap_string(cap->issued), 1633a8599bd8SSage Weil ceph_cap_string(cap->issued & retain), 1634a8599bd8SSage Weil ceph_cap_string(cap->mds_wanted), 1635a8599bd8SSage Weil ceph_cap_string(want)); 1636a8599bd8SSage Weil delayed++; 1637a8599bd8SSage Weil continue; 1638a8599bd8SSage Weil } 1639a8599bd8SSage Weil 1640a8599bd8SSage Weil ack: 1641e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1642e9964c10SSage Weil dout(" skipping %p I_NOFLUSH set\n", inode); 1643e9964c10SSage Weil continue; 1644e9964c10SSage Weil } 1645e9964c10SSage Weil 1646a8599bd8SSage Weil if (session && session != cap->session) { 1647a8599bd8SSage Weil dout("oops, wrong session %p mutex\n", session); 1648a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1649a8599bd8SSage Weil session = NULL; 1650a8599bd8SSage Weil } 1651a8599bd8SSage Weil if (!session) { 1652a8599bd8SSage Weil session = cap->session; 1653a8599bd8SSage Weil if (mutex_trylock(&session->s_mutex) == 0) { 1654a8599bd8SSage Weil dout("inverting session/ino locks on %p\n", 1655a8599bd8SSage Weil session); 1656be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1657a8599bd8SSage Weil if (took_snap_rwsem) { 1658a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1659a8599bd8SSage Weil took_snap_rwsem = 0; 1660a8599bd8SSage Weil } 1661a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1662a8599bd8SSage Weil goto retry; 1663a8599bd8SSage Weil } 1664a8599bd8SSage Weil } 1665a8599bd8SSage Weil /* take snap_rwsem after session mutex */ 1666a8599bd8SSage Weil if (!took_snap_rwsem) { 1667a8599bd8SSage Weil if (down_read_trylock(&mdsc->snap_rwsem) == 0) { 1668a8599bd8SSage Weil dout("inverting snap/in locks on %p\n", 1669a8599bd8SSage Weil inode); 1670be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1671a8599bd8SSage Weil down_read(&mdsc->snap_rwsem); 1672a8599bd8SSage Weil took_snap_rwsem = 1; 1673a8599bd8SSage Weil goto retry; 1674a8599bd8SSage Weil } 1675a8599bd8SSage Weil took_snap_rwsem = 1; 1676a8599bd8SSage Weil } 1677a8599bd8SSage Weil 1678cdc35f96SSage Weil if (cap == ci->i_auth_cap && ci->i_dirty_caps) 1679cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 168024be0c48SSage Weil else 168124be0c48SSage Weil flushing = 0; 1682a8599bd8SSage Weil 1683a8599bd8SSage Weil mds = cap->mds; /* remember mds, so we don't repeat */ 1684a8599bd8SSage Weil sent++; 1685a8599bd8SSage Weil 1686be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1687395c312bSYan, Zheng delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used, 1688395c312bSYan, Zheng want, retain, flushing, NULL); 1689be655596SSage Weil goto retry; /* retake i_ceph_lock and restart our cap scan. */ 1690a8599bd8SSage Weil } 1691a8599bd8SSage Weil 1692a8599bd8SSage Weil /* 1693a8599bd8SSage Weil * Reschedule delayed caps release if we delayed anything, 1694a8599bd8SSage Weil * otherwise cancel. 1695a8599bd8SSage Weil */ 1696a8599bd8SSage Weil if (delayed && is_delayed) 1697a8599bd8SSage Weil force_requeue = 1; /* __send_cap delayed release; requeue */ 1698a8599bd8SSage Weil if (!delayed && !is_delayed) 1699a8599bd8SSage Weil __cap_delay_cancel(mdsc, ci); 1700a8599bd8SSage Weil else if (!is_delayed || force_requeue) 1701a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1702a8599bd8SSage Weil 1703be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1704a8599bd8SSage Weil 1705cbd03635SSage Weil if (queue_invalidate) 17063c6f6b79SSage Weil ceph_queue_invalidate(inode); 1707cbd03635SSage Weil 1708cdc2ce05SSage Weil if (session) 1709a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1710a8599bd8SSage Weil if (took_snap_rwsem) 1711a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 1712a8599bd8SSage Weil } 1713a8599bd8SSage Weil 1714a8599bd8SSage Weil /* 1715a8599bd8SSage Weil * Try to flush dirty caps back to the auth mds. 1716a8599bd8SSage Weil */ 1717a8599bd8SSage Weil static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session, 1718a8599bd8SSage Weil unsigned *flush_tid) 1719a8599bd8SSage Weil { 17203d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1721a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1722a8599bd8SSage Weil int unlock_session = session ? 0 : 1; 1723a8599bd8SSage Weil int flushing = 0; 1724a8599bd8SSage Weil 1725a8599bd8SSage Weil retry: 1726be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1727e9964c10SSage Weil if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { 1728e9964c10SSage Weil dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode); 1729e9964c10SSage Weil goto out; 1730e9964c10SSage Weil } 1731a8599bd8SSage Weil if (ci->i_dirty_caps && ci->i_auth_cap) { 1732a8599bd8SSage Weil struct ceph_cap *cap = ci->i_auth_cap; 1733a8599bd8SSage Weil int used = __ceph_caps_used(ci); 1734a8599bd8SSage Weil int want = __ceph_caps_wanted(ci); 1735a8599bd8SSage Weil int delayed; 1736a8599bd8SSage Weil 1737a8599bd8SSage Weil if (!session) { 1738be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1739a8599bd8SSage Weil session = cap->session; 1740a8599bd8SSage Weil mutex_lock(&session->s_mutex); 1741a8599bd8SSage Weil goto retry; 1742a8599bd8SSage Weil } 1743a8599bd8SSage Weil BUG_ON(session != cap->session); 1744a8599bd8SSage Weil if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) 1745a8599bd8SSage Weil goto out; 1746a8599bd8SSage Weil 1747cdc35f96SSage Weil flushing = __mark_caps_flushing(inode, session); 1748a8599bd8SSage Weil 1749be655596SSage Weil /* __send_cap drops i_ceph_lock */ 1750a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want, 1751a8599bd8SSage Weil cap->issued | cap->implemented, flushing, 1752a8599bd8SSage Weil flush_tid); 1753a8599bd8SSage Weil if (!delayed) 1754a8599bd8SSage Weil goto out_unlocked; 1755a8599bd8SSage Weil 1756be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1757a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1758a8599bd8SSage Weil } 1759a8599bd8SSage Weil out: 1760be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1761a8599bd8SSage Weil out_unlocked: 1762a8599bd8SSage Weil if (session && unlock_session) 1763a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 1764a8599bd8SSage Weil return flushing; 1765a8599bd8SSage Weil } 1766a8599bd8SSage Weil 1767a8599bd8SSage Weil /* 1768a8599bd8SSage Weil * Return true if we've flushed caps through the given flush_tid. 1769a8599bd8SSage Weil */ 1770a8599bd8SSage Weil static int caps_are_flushed(struct inode *inode, unsigned tid) 1771a8599bd8SSage Weil { 1772a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1773a5ee751cSDan Carpenter int i, ret = 1; 1774a8599bd8SSage Weil 1775be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1776a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 1777a8599bd8SSage Weil if ((ci->i_flushing_caps & (1 << i)) && 1778a8599bd8SSage Weil ci->i_cap_flush_tid[i] <= tid) { 1779a8599bd8SSage Weil /* still flushing this bit */ 1780a8599bd8SSage Weil ret = 0; 1781a8599bd8SSage Weil break; 1782a8599bd8SSage Weil } 1783be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1784a8599bd8SSage Weil return ret; 1785a8599bd8SSage Weil } 1786a8599bd8SSage Weil 1787a8599bd8SSage Weil /* 1788a8599bd8SSage Weil * Wait on any unsafe replies for the given inode. First wait on the 1789a8599bd8SSage Weil * newest request, and make that the upper bound. Then, if there are 1790a8599bd8SSage Weil * more requests, keep waiting on the oldest as long as it is still older 1791a8599bd8SSage Weil * than the original request. 1792a8599bd8SSage Weil */ 1793a8599bd8SSage Weil static void sync_write_wait(struct inode *inode) 1794a8599bd8SSage Weil { 1795a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1796a8599bd8SSage Weil struct list_head *head = &ci->i_unsafe_writes; 1797a8599bd8SSage Weil struct ceph_osd_request *req; 1798a8599bd8SSage Weil u64 last_tid; 1799a8599bd8SSage Weil 1800a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1801a8599bd8SSage Weil if (list_empty(head)) 1802a8599bd8SSage Weil goto out; 1803a8599bd8SSage Weil 1804a8599bd8SSage Weil /* set upper bound as _last_ entry in chain */ 1805a8599bd8SSage Weil req = list_entry(head->prev, struct ceph_osd_request, 1806a8599bd8SSage Weil r_unsafe_item); 1807a8599bd8SSage Weil last_tid = req->r_tid; 1808a8599bd8SSage Weil 1809a8599bd8SSage Weil do { 1810a8599bd8SSage Weil ceph_osdc_get_request(req); 1811a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1812a8599bd8SSage Weil dout("sync_write_wait on tid %llu (until %llu)\n", 1813a8599bd8SSage Weil req->r_tid, last_tid); 1814a8599bd8SSage Weil wait_for_completion(&req->r_safe_completion); 1815a8599bd8SSage Weil spin_lock(&ci->i_unsafe_lock); 1816a8599bd8SSage Weil ceph_osdc_put_request(req); 1817a8599bd8SSage Weil 1818a8599bd8SSage Weil /* 1819a8599bd8SSage Weil * from here on look at first entry in chain, since we 1820a8599bd8SSage Weil * only want to wait for anything older than last_tid 1821a8599bd8SSage Weil */ 1822a8599bd8SSage Weil if (list_empty(head)) 1823a8599bd8SSage Weil break; 1824a8599bd8SSage Weil req = list_entry(head->next, struct ceph_osd_request, 1825a8599bd8SSage Weil r_unsafe_item); 1826a8599bd8SSage Weil } while (req->r_tid < last_tid); 1827a8599bd8SSage Weil out: 1828a8599bd8SSage Weil spin_unlock(&ci->i_unsafe_lock); 1829a8599bd8SSage Weil } 1830a8599bd8SSage Weil 183102c24a82SJosef Bacik int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync) 1832a8599bd8SSage Weil { 18337ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 1834a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1835a8599bd8SSage Weil unsigned flush_tid; 1836a8599bd8SSage Weil int ret; 1837a8599bd8SSage Weil int dirty; 1838a8599bd8SSage Weil 1839a8599bd8SSage Weil dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); 1840a8599bd8SSage Weil sync_write_wait(inode); 1841a8599bd8SSage Weil 184202c24a82SJosef Bacik ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 1843a8599bd8SSage Weil if (ret < 0) 1844a8599bd8SSage Weil return ret; 184502c24a82SJosef Bacik mutex_lock(&inode->i_mutex); 1846a8599bd8SSage Weil 1847a8599bd8SSage Weil dirty = try_flush_caps(inode, NULL, &flush_tid); 1848a8599bd8SSage Weil dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); 1849a8599bd8SSage Weil 1850a8599bd8SSage Weil /* 1851a8599bd8SSage Weil * only wait on non-file metadata writeback (the mds 1852a8599bd8SSage Weil * can recover size and mtime, so we don't need to 1853a8599bd8SSage Weil * wait for that) 1854a8599bd8SSage Weil */ 1855a8599bd8SSage Weil if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { 1856a8599bd8SSage Weil dout("fsync waiting for flush_tid %u\n", flush_tid); 1857a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 1858a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1859a8599bd8SSage Weil } 1860a8599bd8SSage Weil 1861a8599bd8SSage Weil dout("fsync %p%s done\n", inode, datasync ? " datasync" : ""); 186202c24a82SJosef Bacik mutex_unlock(&inode->i_mutex); 1863a8599bd8SSage Weil return ret; 1864a8599bd8SSage Weil } 1865a8599bd8SSage Weil 1866a8599bd8SSage Weil /* 1867a8599bd8SSage Weil * Flush any dirty caps back to the mds. If we aren't asked to wait, 1868a8599bd8SSage Weil * queue inode for flush but don't do so immediately, because we can 1869a8599bd8SSage Weil * get by with fewer MDS messages if we wait for data writeback to 1870a8599bd8SSage Weil * complete first. 1871a8599bd8SSage Weil */ 1872f1a3d572SStephen Rothwell int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) 1873a8599bd8SSage Weil { 1874a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1875a8599bd8SSage Weil unsigned flush_tid; 1876a8599bd8SSage Weil int err = 0; 1877a8599bd8SSage Weil int dirty; 1878f1a3d572SStephen Rothwell int wait = wbc->sync_mode == WB_SYNC_ALL; 1879a8599bd8SSage Weil 1880a8599bd8SSage Weil dout("write_inode %p wait=%d\n", inode, wait); 1881a8599bd8SSage Weil if (wait) { 1882a8599bd8SSage Weil dirty = try_flush_caps(inode, NULL, &flush_tid); 1883a8599bd8SSage Weil if (dirty) 1884a8599bd8SSage Weil err = wait_event_interruptible(ci->i_cap_wq, 1885a8599bd8SSage Weil caps_are_flushed(inode, flush_tid)); 1886a8599bd8SSage Weil } else { 1887640ef79dSCheng Renquan struct ceph_mds_client *mdsc = 18883d14c5d2SYehuda Sadeh ceph_sb_to_client(inode->i_sb)->mdsc; 1889a8599bd8SSage Weil 1890be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1891a8599bd8SSage Weil if (__ceph_caps_dirty(ci)) 1892a8599bd8SSage Weil __cap_delay_requeue_front(mdsc, ci); 1893be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1894a8599bd8SSage Weil } 1895a8599bd8SSage Weil return err; 1896a8599bd8SSage Weil } 1897a8599bd8SSage Weil 1898a8599bd8SSage Weil /* 1899a8599bd8SSage Weil * After a recovering MDS goes active, we need to resend any caps 1900a8599bd8SSage Weil * we were flushing. 1901a8599bd8SSage Weil * 1902a8599bd8SSage Weil * Caller holds session->s_mutex. 1903a8599bd8SSage Weil */ 1904a8599bd8SSage Weil static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc, 1905a8599bd8SSage Weil struct ceph_mds_session *session) 1906a8599bd8SSage Weil { 1907a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 1908a8599bd8SSage Weil 1909a8599bd8SSage Weil dout("kick_flushing_capsnaps mds%d\n", session->s_mds); 1910a8599bd8SSage Weil list_for_each_entry(capsnap, &session->s_cap_snaps_flushing, 1911a8599bd8SSage Weil flushing_item) { 1912a8599bd8SSage Weil struct ceph_inode_info *ci = capsnap->ci; 1913a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1914a8599bd8SSage Weil struct ceph_cap *cap; 1915a8599bd8SSage Weil 1916be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1917a8599bd8SSage Weil cap = ci->i_auth_cap; 1918a8599bd8SSage Weil if (cap && cap->session == session) { 1919a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p capsnap %p\n", inode, 1920a8599bd8SSage Weil cap, capsnap); 1921e835124cSSage Weil __ceph_flush_snaps(ci, &session, 1); 1922a8599bd8SSage Weil } else { 1923a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1924a8599bd8SSage Weil cap, session->s_mds); 1925a8599bd8SSage Weil } 1926be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1927a8599bd8SSage Weil } 1928a8599bd8SSage Weil } 1929a8599bd8SSage Weil 1930a8599bd8SSage Weil void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, 1931a8599bd8SSage Weil struct ceph_mds_session *session) 1932a8599bd8SSage Weil { 1933a8599bd8SSage Weil struct ceph_inode_info *ci; 1934a8599bd8SSage Weil 1935a8599bd8SSage Weil kick_flushing_capsnaps(mdsc, session); 1936a8599bd8SSage Weil 1937a8599bd8SSage Weil dout("kick_flushing_caps mds%d\n", session->s_mds); 1938a8599bd8SSage Weil list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { 1939a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 1940a8599bd8SSage Weil struct ceph_cap *cap; 1941a8599bd8SSage Weil int delayed = 0; 1942a8599bd8SSage Weil 1943be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1944a8599bd8SSage Weil cap = ci->i_auth_cap; 1945a8599bd8SSage Weil if (cap && cap->session == session) { 1946a8599bd8SSage Weil dout("kick_flushing_caps %p cap %p %s\n", inode, 1947a8599bd8SSage Weil cap, ceph_cap_string(ci->i_flushing_caps)); 1948a8599bd8SSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 1949a8599bd8SSage Weil __ceph_caps_used(ci), 1950a8599bd8SSage Weil __ceph_caps_wanted(ci), 1951a8599bd8SSage Weil cap->issued | cap->implemented, 1952a8599bd8SSage Weil ci->i_flushing_caps, NULL); 1953a8599bd8SSage Weil if (delayed) { 1954be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1955a8599bd8SSage Weil __cap_delay_requeue(mdsc, ci); 1956be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1957a8599bd8SSage Weil } 1958a8599bd8SSage Weil } else { 1959a8599bd8SSage Weil pr_err("%p auth cap %p not mds%d ???\n", inode, 1960a8599bd8SSage Weil cap, session->s_mds); 1961be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1962a8599bd8SSage Weil } 1963a8599bd8SSage Weil } 1964a8599bd8SSage Weil } 1965a8599bd8SSage Weil 1966088b3f5eSSage Weil static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc, 1967088b3f5eSSage Weil struct ceph_mds_session *session, 1968088b3f5eSSage Weil struct inode *inode) 1969088b3f5eSSage Weil { 1970088b3f5eSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 1971088b3f5eSSage Weil struct ceph_cap *cap; 1972088b3f5eSSage Weil int delayed = 0; 1973088b3f5eSSage Weil 1974be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1975088b3f5eSSage Weil cap = ci->i_auth_cap; 1976088b3f5eSSage Weil dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode, 1977088b3f5eSSage Weil ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq); 1978005c4697SYan, Zheng 1979088b3f5eSSage Weil __ceph_flush_snaps(ci, &session, 1); 1980005c4697SYan, Zheng 1981088b3f5eSSage Weil if (ci->i_flushing_caps) { 1982005c4697SYan, Zheng spin_lock(&mdsc->cap_dirty_lock); 1983005c4697SYan, Zheng list_move_tail(&ci->i_flushing_item, 1984005c4697SYan, Zheng &cap->session->s_cap_flushing); 1985005c4697SYan, Zheng spin_unlock(&mdsc->cap_dirty_lock); 1986005c4697SYan, Zheng 1987088b3f5eSSage Weil delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, 1988088b3f5eSSage Weil __ceph_caps_used(ci), 1989088b3f5eSSage Weil __ceph_caps_wanted(ci), 1990088b3f5eSSage Weil cap->issued | cap->implemented, 1991088b3f5eSSage Weil ci->i_flushing_caps, NULL); 1992088b3f5eSSage Weil if (delayed) { 1993be655596SSage Weil spin_lock(&ci->i_ceph_lock); 1994088b3f5eSSage Weil __cap_delay_requeue(mdsc, ci); 1995be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1996088b3f5eSSage Weil } 1997088b3f5eSSage Weil } else { 1998be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1999088b3f5eSSage Weil } 2000088b3f5eSSage Weil } 2001088b3f5eSSage Weil 2002a8599bd8SSage Weil 2003a8599bd8SSage Weil /* 2004a8599bd8SSage Weil * Take references to capabilities we hold, so that we don't release 2005a8599bd8SSage Weil * them to the MDS prematurely. 2006a8599bd8SSage Weil * 2007be655596SSage Weil * Protected by i_ceph_lock. 2008a8599bd8SSage Weil */ 2009a8599bd8SSage Weil static void __take_cap_refs(struct ceph_inode_info *ci, int got) 2010a8599bd8SSage Weil { 2011a8599bd8SSage Weil if (got & CEPH_CAP_PIN) 2012a8599bd8SSage Weil ci->i_pin_ref++; 2013a8599bd8SSage Weil if (got & CEPH_CAP_FILE_RD) 2014a8599bd8SSage Weil ci->i_rd_ref++; 2015a8599bd8SSage Weil if (got & CEPH_CAP_FILE_CACHE) 2016a8599bd8SSage Weil ci->i_rdcache_ref++; 2017a8599bd8SSage Weil if (got & CEPH_CAP_FILE_WR) 2018a8599bd8SSage Weil ci->i_wr_ref++; 2019a8599bd8SSage Weil if (got & CEPH_CAP_FILE_BUFFER) { 2020d3d0720dSHenry C Chang if (ci->i_wb_ref == 0) 20213772d26dSSage Weil ihold(&ci->vfs_inode); 2022d3d0720dSHenry C Chang ci->i_wb_ref++; 2023d3d0720dSHenry C Chang dout("__take_cap_refs %p wb %d -> %d (?)\n", 2024d3d0720dSHenry C Chang &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref); 2025a8599bd8SSage Weil } 2026a8599bd8SSage Weil } 2027a8599bd8SSage Weil 2028a8599bd8SSage Weil /* 2029a8599bd8SSage Weil * Try to grab cap references. Specify those refs we @want, and the 2030a8599bd8SSage Weil * minimal set we @need. Also include the larger offset we are writing 2031a8599bd8SSage Weil * to (when applicable), and check against max_size here as well. 2032a8599bd8SSage Weil * Note that caller is responsible for ensuring max_size increases are 2033a8599bd8SSage Weil * requested from the MDS. 2034a8599bd8SSage Weil */ 2035a8599bd8SSage Weil static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want, 2036a8599bd8SSage Weil int *got, loff_t endoff, int *check_max, int *err) 2037a8599bd8SSage Weil { 2038a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2039a8599bd8SSage Weil int ret = 0; 2040a8599bd8SSage Weil int have, implemented; 2041195d3ce2SSage Weil int file_wanted; 2042a8599bd8SSage Weil 2043a8599bd8SSage Weil dout("get_cap_refs %p need %s want %s\n", inode, 2044a8599bd8SSage Weil ceph_cap_string(need), ceph_cap_string(want)); 2045be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2046a8599bd8SSage Weil 2047195d3ce2SSage Weil /* make sure file is actually open */ 2048195d3ce2SSage Weil file_wanted = __ceph_caps_file_wanted(ci); 2049195d3ce2SSage Weil if ((file_wanted & need) == 0) { 2050195d3ce2SSage Weil dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", 2051195d3ce2SSage Weil ceph_cap_string(need), ceph_cap_string(file_wanted)); 2052a8599bd8SSage Weil *err = -EBADF; 2053a8599bd8SSage Weil ret = 1; 2054a8599bd8SSage Weil goto out; 2055a8599bd8SSage Weil } 2056a8599bd8SSage Weil 205737505d57SYan, Zheng /* finish pending truncate */ 205837505d57SYan, Zheng while (ci->i_truncate_pending) { 205937505d57SYan, Zheng spin_unlock(&ci->i_ceph_lock); 206037505d57SYan, Zheng __ceph_do_pending_vmtruncate(inode, !(need & CEPH_CAP_FILE_WR)); 206137505d57SYan, Zheng spin_lock(&ci->i_ceph_lock); 206237505d57SYan, Zheng } 206337505d57SYan, Zheng 2064a8599bd8SSage Weil if (need & CEPH_CAP_FILE_WR) { 2065a8599bd8SSage Weil if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { 2066a8599bd8SSage Weil dout("get_cap_refs %p endoff %llu > maxsize %llu\n", 2067a8599bd8SSage Weil inode, endoff, ci->i_max_size); 2068a8599bd8SSage Weil if (endoff > ci->i_wanted_max_size) { 2069a8599bd8SSage Weil *check_max = 1; 2070a8599bd8SSage Weil ret = 1; 2071a8599bd8SSage Weil } 2072a8599bd8SSage Weil goto out; 2073a8599bd8SSage Weil } 2074a8599bd8SSage Weil /* 2075a8599bd8SSage Weil * If a sync write is in progress, we must wait, so that we 2076a8599bd8SSage Weil * can get a final snapshot value for size+mtime. 2077a8599bd8SSage Weil */ 2078a8599bd8SSage Weil if (__ceph_have_pending_cap_snap(ci)) { 2079a8599bd8SSage Weil dout("get_cap_refs %p cap_snap_pending\n", inode); 2080a8599bd8SSage Weil goto out; 2081a8599bd8SSage Weil } 2082a8599bd8SSage Weil } 2083a8599bd8SSage Weil have = __ceph_caps_issued(ci, &implemented); 2084a8599bd8SSage Weil 2085a8599bd8SSage Weil if ((have & need) == need) { 2086a8599bd8SSage Weil /* 2087a8599bd8SSage Weil * Look at (implemented & ~have & not) so that we keep waiting 2088a8599bd8SSage Weil * on transition from wanted -> needed caps. This is needed 2089a8599bd8SSage Weil * for WRBUFFER|WR -> WR to avoid a new WR sync write from 2090a8599bd8SSage Weil * going before a prior buffered writeback happens. 2091a8599bd8SSage Weil */ 2092a8599bd8SSage Weil int not = want & ~(have & need); 2093a8599bd8SSage Weil int revoking = implemented & ~have; 2094a8599bd8SSage Weil dout("get_cap_refs %p have %s but not %s (revoking %s)\n", 2095a8599bd8SSage Weil inode, ceph_cap_string(have), ceph_cap_string(not), 2096a8599bd8SSage Weil ceph_cap_string(revoking)); 2097a8599bd8SSage Weil if ((revoking & not) == 0) { 2098a8599bd8SSage Weil *got = need | (have & want); 2099a8599bd8SSage Weil __take_cap_refs(ci, *got); 2100a8599bd8SSage Weil ret = 1; 2101a8599bd8SSage Weil } 2102a8599bd8SSage Weil } else { 2103a8599bd8SSage Weil dout("get_cap_refs %p have %s needed %s\n", inode, 2104a8599bd8SSage Weil ceph_cap_string(have), ceph_cap_string(need)); 2105a8599bd8SSage Weil } 2106a8599bd8SSage Weil out: 2107be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2108a8599bd8SSage Weil dout("get_cap_refs %p ret %d got %s\n", inode, 2109a8599bd8SSage Weil ret, ceph_cap_string(*got)); 2110a8599bd8SSage Weil return ret; 2111a8599bd8SSage Weil } 2112a8599bd8SSage Weil 2113a8599bd8SSage Weil /* 2114a8599bd8SSage Weil * Check the offset we are writing up to against our current 2115a8599bd8SSage Weil * max_size. If necessary, tell the MDS we want to write to 2116a8599bd8SSage Weil * a larger offset. 2117a8599bd8SSage Weil */ 2118a8599bd8SSage Weil static void check_max_size(struct inode *inode, loff_t endoff) 2119a8599bd8SSage Weil { 2120a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2121a8599bd8SSage Weil int check = 0; 2122a8599bd8SSage Weil 2123a8599bd8SSage Weil /* do we need to explicitly request a larger max_size? */ 2124be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2125a8599bd8SSage Weil if ((endoff >= ci->i_max_size || 2126a8599bd8SSage Weil endoff > (inode->i_size << 1)) && 2127a8599bd8SSage Weil endoff > ci->i_wanted_max_size) { 2128a8599bd8SSage Weil dout("write %p at large endoff %llu, req max_size\n", 2129a8599bd8SSage Weil inode, endoff); 2130a8599bd8SSage Weil ci->i_wanted_max_size = endoff; 2131a8599bd8SSage Weil check = 1; 2132a8599bd8SSage Weil } 2133be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2134a8599bd8SSage Weil if (check) 2135a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2136a8599bd8SSage Weil } 2137a8599bd8SSage Weil 2138a8599bd8SSage Weil /* 2139a8599bd8SSage Weil * Wait for caps, and take cap references. If we can't get a WR cap 2140a8599bd8SSage Weil * due to a small max_size, make sure we check_max_size (and possibly 2141a8599bd8SSage Weil * ask the mds) so we don't get hung up indefinitely. 2142a8599bd8SSage Weil */ 2143a8599bd8SSage Weil int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got, 2144a8599bd8SSage Weil loff_t endoff) 2145a8599bd8SSage Weil { 2146a8599bd8SSage Weil int check_max, ret, err; 2147a8599bd8SSage Weil 2148a8599bd8SSage Weil retry: 2149a8599bd8SSage Weil if (endoff > 0) 2150a8599bd8SSage Weil check_max_size(&ci->vfs_inode, endoff); 2151a8599bd8SSage Weil check_max = 0; 2152a8599bd8SSage Weil err = 0; 2153a8599bd8SSage Weil ret = wait_event_interruptible(ci->i_cap_wq, 2154a8599bd8SSage Weil try_get_cap_refs(ci, need, want, 2155a8599bd8SSage Weil got, endoff, 2156a8599bd8SSage Weil &check_max, &err)); 2157a8599bd8SSage Weil if (err) 2158a8599bd8SSage Weil ret = err; 2159a8599bd8SSage Weil if (check_max) 2160a8599bd8SSage Weil goto retry; 2161a8599bd8SSage Weil return ret; 2162a8599bd8SSage Weil } 2163a8599bd8SSage Weil 2164a8599bd8SSage Weil /* 2165a8599bd8SSage Weil * Take cap refs. Caller must already know we hold at least one ref 2166a8599bd8SSage Weil * on the caps in question or we don't know this is safe. 2167a8599bd8SSage Weil */ 2168a8599bd8SSage Weil void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) 2169a8599bd8SSage Weil { 2170be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2171a8599bd8SSage Weil __take_cap_refs(ci, caps); 2172be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2173a8599bd8SSage Weil } 2174a8599bd8SSage Weil 2175a8599bd8SSage Weil /* 2176a8599bd8SSage Weil * Release cap refs. 2177a8599bd8SSage Weil * 2178a8599bd8SSage Weil * If we released the last ref on any given cap, call ceph_check_caps 2179a8599bd8SSage Weil * to release (or schedule a release). 2180a8599bd8SSage Weil * 2181a8599bd8SSage Weil * If we are releasing a WR cap (from a sync write), finalize any affected 2182a8599bd8SSage Weil * cap_snap, and wake up any waiters. 2183a8599bd8SSage Weil */ 2184a8599bd8SSage Weil void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) 2185a8599bd8SSage Weil { 2186a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2187a8599bd8SSage Weil int last = 0, put = 0, flushsnaps = 0, wake = 0; 2188a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2189a8599bd8SSage Weil 2190be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2191a8599bd8SSage Weil if (had & CEPH_CAP_PIN) 2192a8599bd8SSage Weil --ci->i_pin_ref; 2193a8599bd8SSage Weil if (had & CEPH_CAP_FILE_RD) 2194a8599bd8SSage Weil if (--ci->i_rd_ref == 0) 2195a8599bd8SSage Weil last++; 2196a8599bd8SSage Weil if (had & CEPH_CAP_FILE_CACHE) 2197a8599bd8SSage Weil if (--ci->i_rdcache_ref == 0) 2198a8599bd8SSage Weil last++; 2199a8599bd8SSage Weil if (had & CEPH_CAP_FILE_BUFFER) { 2200d3d0720dSHenry C Chang if (--ci->i_wb_ref == 0) { 2201a8599bd8SSage Weil last++; 2202a8599bd8SSage Weil put++; 2203a8599bd8SSage Weil } 2204d3d0720dSHenry C Chang dout("put_cap_refs %p wb %d -> %d (?)\n", 2205d3d0720dSHenry C Chang inode, ci->i_wb_ref+1, ci->i_wb_ref); 2206a8599bd8SSage Weil } 2207a8599bd8SSage Weil if (had & CEPH_CAP_FILE_WR) 2208a8599bd8SSage Weil if (--ci->i_wr_ref == 0) { 2209a8599bd8SSage Weil last++; 2210a8599bd8SSage Weil if (!list_empty(&ci->i_cap_snaps)) { 2211a8599bd8SSage Weil capsnap = list_first_entry(&ci->i_cap_snaps, 2212a8599bd8SSage Weil struct ceph_cap_snap, 2213a8599bd8SSage Weil ci_item); 2214a8599bd8SSage Weil if (capsnap->writing) { 2215a8599bd8SSage Weil capsnap->writing = 0; 2216a8599bd8SSage Weil flushsnaps = 2217a8599bd8SSage Weil __ceph_finish_cap_snap(ci, 2218a8599bd8SSage Weil capsnap); 2219a8599bd8SSage Weil wake = 1; 2220a8599bd8SSage Weil } 2221a8599bd8SSage Weil } 2222a8599bd8SSage Weil } 2223be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2224a8599bd8SSage Weil 2225819ccbfaSSage Weil dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), 2226819ccbfaSSage Weil last ? " last" : "", put ? " put" : ""); 2227a8599bd8SSage Weil 2228a8599bd8SSage Weil if (last && !flushsnaps) 2229a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 2230a8599bd8SSage Weil else if (flushsnaps) 2231a8599bd8SSage Weil ceph_flush_snaps(ci); 2232a8599bd8SSage Weil if (wake) 223303066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2234a8599bd8SSage Weil if (put) 2235a8599bd8SSage Weil iput(inode); 2236a8599bd8SSage Weil } 2237a8599bd8SSage Weil 2238a8599bd8SSage Weil /* 2239a8599bd8SSage Weil * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap 2240a8599bd8SSage Weil * context. Adjust per-snap dirty page accounting as appropriate. 2241a8599bd8SSage Weil * Once all dirty data for a cap_snap is flushed, flush snapped file 2242a8599bd8SSage Weil * metadata back to the MDS. If we dropped the last ref, call 2243a8599bd8SSage Weil * ceph_check_caps. 2244a8599bd8SSage Weil */ 2245a8599bd8SSage Weil void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, 2246a8599bd8SSage Weil struct ceph_snap_context *snapc) 2247a8599bd8SSage Weil { 2248a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2249a8599bd8SSage Weil int last = 0; 2250819ccbfaSSage Weil int complete_capsnap = 0; 2251819ccbfaSSage Weil int drop_capsnap = 0; 2252a8599bd8SSage Weil int found = 0; 2253a8599bd8SSage Weil struct ceph_cap_snap *capsnap = NULL; 2254a8599bd8SSage Weil 2255be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2256a8599bd8SSage Weil ci->i_wrbuffer_ref -= nr; 2257a8599bd8SSage Weil last = !ci->i_wrbuffer_ref; 2258a8599bd8SSage Weil 2259a8599bd8SSage Weil if (ci->i_head_snapc == snapc) { 2260a8599bd8SSage Weil ci->i_wrbuffer_ref_head -= nr; 22617d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0 && 22627d8cb26dSSage Weil ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) { 22637d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 2264a8599bd8SSage Weil ceph_put_snap_context(ci->i_head_snapc); 2265a8599bd8SSage Weil ci->i_head_snapc = NULL; 2266a8599bd8SSage Weil } 2267a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", 2268a8599bd8SSage Weil inode, 2269a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, 2270a8599bd8SSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 2271a8599bd8SSage Weil last ? " LAST" : ""); 2272a8599bd8SSage Weil } else { 2273a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2274a8599bd8SSage Weil if (capsnap->context == snapc) { 2275a8599bd8SSage Weil found = 1; 2276a8599bd8SSage Weil break; 2277a8599bd8SSage Weil } 2278a8599bd8SSage Weil } 2279a8599bd8SSage Weil BUG_ON(!found); 2280819ccbfaSSage Weil capsnap->dirty_pages -= nr; 2281819ccbfaSSage Weil if (capsnap->dirty_pages == 0) { 2282819ccbfaSSage Weil complete_capsnap = 1; 2283819ccbfaSSage Weil if (capsnap->dirty == 0) 2284819ccbfaSSage Weil /* cap writeback completed before we created 2285819ccbfaSSage Weil * the cap_snap; no FLUSHSNAP is needed */ 2286819ccbfaSSage Weil drop_capsnap = 1; 2287819ccbfaSSage Weil } 2288a8599bd8SSage Weil dout("put_wrbuffer_cap_refs on %p cap_snap %p " 2289819ccbfaSSage Weil " snap %lld %d/%d -> %d/%d %s%s%s\n", 2290a8599bd8SSage Weil inode, capsnap, capsnap->context->seq, 2291a8599bd8SSage Weil ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, 2292a8599bd8SSage Weil ci->i_wrbuffer_ref, capsnap->dirty_pages, 2293a8599bd8SSage Weil last ? " (wrbuffer last)" : "", 2294819ccbfaSSage Weil complete_capsnap ? " (complete capsnap)" : "", 2295819ccbfaSSage Weil drop_capsnap ? " (drop capsnap)" : ""); 2296819ccbfaSSage Weil if (drop_capsnap) { 2297819ccbfaSSage Weil ceph_put_snap_context(capsnap->context); 2298819ccbfaSSage Weil list_del(&capsnap->ci_item); 2299819ccbfaSSage Weil list_del(&capsnap->flushing_item); 2300819ccbfaSSage Weil ceph_put_cap_snap(capsnap); 2301819ccbfaSSage Weil } 2302a8599bd8SSage Weil } 2303a8599bd8SSage Weil 2304be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2305a8599bd8SSage Weil 2306a8599bd8SSage Weil if (last) { 2307a8599bd8SSage Weil ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 2308a8599bd8SSage Weil iput(inode); 2309819ccbfaSSage Weil } else if (complete_capsnap) { 2310a8599bd8SSage Weil ceph_flush_snaps(ci); 231103066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2312a8599bd8SSage Weil } 2313819ccbfaSSage Weil if (drop_capsnap) 2314819ccbfaSSage Weil iput(inode); 2315a8599bd8SSage Weil } 2316a8599bd8SSage Weil 2317a8599bd8SSage Weil /* 2318a8599bd8SSage Weil * Handle a cap GRANT message from the MDS. (Note that a GRANT may 2319a8599bd8SSage Weil * actually be a revocation if it specifies a smaller cap set.) 2320a8599bd8SSage Weil * 2321be655596SSage Weil * caller holds s_mutex and i_ceph_lock, we drop both. 232215637c8bSSage Weil * 2323a8599bd8SSage Weil * return value: 2324a8599bd8SSage Weil * 0 - ok 2325a8599bd8SSage Weil * 1 - check_caps on auth cap only (writeback) 2326a8599bd8SSage Weil * 2 - check_caps (ack revoke) 2327a8599bd8SSage Weil */ 232815637c8bSSage Weil static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant, 2329a8599bd8SSage Weil struct ceph_mds_session *session, 2330a8599bd8SSage Weil struct ceph_cap *cap, 2331a8599bd8SSage Weil struct ceph_buffer *xattr_buf) 2332be655596SSage Weil __releases(ci->i_ceph_lock) 2333a8599bd8SSage Weil { 2334a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2335a8599bd8SSage Weil int mds = session->s_mds; 23362f56f56aSSage Weil int seq = le32_to_cpu(grant->seq); 2337a8599bd8SSage Weil int newcaps = le32_to_cpu(grant->caps); 2338a8599bd8SSage Weil int issued, implemented, used, wanted, dirty; 2339a8599bd8SSage Weil u64 size = le64_to_cpu(grant->size); 2340a8599bd8SSage Weil u64 max_size = le64_to_cpu(grant->max_size); 2341a8599bd8SSage Weil struct timespec mtime, atime, ctime; 234215637c8bSSage Weil int check_caps = 0; 2343a8599bd8SSage Weil int wake = 0; 2344a8599bd8SSage Weil int writeback = 0; 2345a8599bd8SSage Weil int revoked_rdcache = 0; 23463c6f6b79SSage Weil int queue_invalidate = 0; 2347a8599bd8SSage Weil 23482f56f56aSSage Weil dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", 23492f56f56aSSage Weil inode, cap, mds, seq, ceph_cap_string(newcaps)); 2350a8599bd8SSage Weil dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, 2351a8599bd8SSage Weil inode->i_size); 2352a8599bd8SSage Weil 2353a8599bd8SSage Weil /* 2354a8599bd8SSage Weil * If CACHE is being revoked, and we have no dirty buffers, 2355a8599bd8SSage Weil * try to invalidate (once). (If there are dirty buffers, we 2356a8599bd8SSage Weil * will invalidate _after_ writeback.) 2357a8599bd8SSage Weil */ 23583b454c49SSage Weil if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && 23593b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 2360bcd2cbd1SYehuda Sadeh !ci->i_wrbuffer_ref) { 23615ecad6fdSSage Weil if (try_nonblocking_invalidate(inode) == 0) { 23625ecad6fdSSage Weil revoked_rdcache = 1; 23635ecad6fdSSage Weil } else { 2364a8599bd8SSage Weil /* there were locked pages.. invalidate later 2365a8599bd8SSage Weil in a separate thread. */ 2366a8599bd8SSage Weil if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 23673c6f6b79SSage Weil queue_invalidate = 1; 2368a8599bd8SSage Weil ci->i_rdcache_revoking = ci->i_rdcache_gen; 2369a8599bd8SSage Weil } 2370a8599bd8SSage Weil } 2371a8599bd8SSage Weil } 2372a8599bd8SSage Weil 2373a8599bd8SSage Weil /* side effects now are allowed */ 2374a8599bd8SSage Weil 2375a8599bd8SSage Weil issued = __ceph_caps_issued(ci, &implemented); 2376a8599bd8SSage Weil issued |= implemented | __ceph_caps_dirty(ci); 2377a8599bd8SSage Weil 2378685f9a5dSSage Weil cap->cap_gen = session->s_cap_gen; 2379a8599bd8SSage Weil 2380a8599bd8SSage Weil __check_cap_issue(ci, cap, newcaps); 2381a8599bd8SSage Weil 2382a8599bd8SSage Weil if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 2383a8599bd8SSage Weil inode->i_mode = le32_to_cpu(grant->mode); 238405cb11c1SEric W. Biederman inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid)); 238505cb11c1SEric W. Biederman inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid)); 2386a8599bd8SSage Weil dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 2387bd2bae6aSEric W. Biederman from_kuid(&init_user_ns, inode->i_uid), 2388bd2bae6aSEric W. Biederman from_kgid(&init_user_ns, inode->i_gid)); 2389a8599bd8SSage Weil } 2390a8599bd8SSage Weil 2391a8599bd8SSage Weil if ((issued & CEPH_CAP_LINK_EXCL) == 0) 2392bfe86848SMiklos Szeredi set_nlink(inode, le32_to_cpu(grant->nlink)); 2393a8599bd8SSage Weil 2394a8599bd8SSage Weil if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) { 2395a8599bd8SSage Weil int len = le32_to_cpu(grant->xattr_len); 2396a8599bd8SSage Weil u64 version = le64_to_cpu(grant->xattr_version); 2397a8599bd8SSage Weil 2398a8599bd8SSage Weil if (version > ci->i_xattrs.version) { 2399a8599bd8SSage Weil dout(" got new xattrs v%llu on %p len %d\n", 2400a8599bd8SSage Weil version, inode, len); 2401a8599bd8SSage Weil if (ci->i_xattrs.blob) 2402a8599bd8SSage Weil ceph_buffer_put(ci->i_xattrs.blob); 2403a8599bd8SSage Weil ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); 2404a8599bd8SSage Weil ci->i_xattrs.version = version; 2405a8599bd8SSage Weil } 2406a8599bd8SSage Weil } 2407a8599bd8SSage Weil 2408a8599bd8SSage Weil /* size/ctime/mtime/atime? */ 2409a8599bd8SSage Weil ceph_fill_file_size(inode, issued, 2410a8599bd8SSage Weil le32_to_cpu(grant->truncate_seq), 2411a8599bd8SSage Weil le64_to_cpu(grant->truncate_size), size); 2412a8599bd8SSage Weil ceph_decode_timespec(&mtime, &grant->mtime); 2413a8599bd8SSage Weil ceph_decode_timespec(&atime, &grant->atime); 2414a8599bd8SSage Weil ceph_decode_timespec(&ctime, &grant->ctime); 2415a8599bd8SSage Weil ceph_fill_file_time(inode, issued, 2416a8599bd8SSage Weil le32_to_cpu(grant->time_warp_seq), &ctime, &mtime, 2417a8599bd8SSage Weil &atime); 2418a8599bd8SSage Weil 2419a8599bd8SSage Weil /* max size increase? */ 24205e62ad30SYan, Zheng if (ci->i_auth_cap == cap && max_size != ci->i_max_size) { 2421a8599bd8SSage Weil dout("max_size %lld -> %llu\n", ci->i_max_size, max_size); 2422a8599bd8SSage Weil ci->i_max_size = max_size; 2423a8599bd8SSage Weil if (max_size >= ci->i_wanted_max_size) { 2424a8599bd8SSage Weil ci->i_wanted_max_size = 0; /* reset */ 2425a8599bd8SSage Weil ci->i_requested_max_size = 0; 2426a8599bd8SSage Weil } 2427a8599bd8SSage Weil wake = 1; 2428a8599bd8SSage Weil } 2429a8599bd8SSage Weil 2430a8599bd8SSage Weil /* check cap bits */ 2431a8599bd8SSage Weil wanted = __ceph_caps_wanted(ci); 2432a8599bd8SSage Weil used = __ceph_caps_used(ci); 2433a8599bd8SSage Weil dirty = __ceph_caps_dirty(ci); 2434a8599bd8SSage Weil dout(" my wanted = %s, used = %s, dirty %s\n", 2435a8599bd8SSage Weil ceph_cap_string(wanted), 2436a8599bd8SSage Weil ceph_cap_string(used), 2437a8599bd8SSage Weil ceph_cap_string(dirty)); 2438a8599bd8SSage Weil if (wanted != le32_to_cpu(grant->wanted)) { 2439a8599bd8SSage Weil dout("mds wanted %s -> %s\n", 2440a8599bd8SSage Weil ceph_cap_string(le32_to_cpu(grant->wanted)), 2441a8599bd8SSage Weil ceph_cap_string(wanted)); 2442390306c3SYan, Zheng /* imported cap may not have correct mds_wanted */ 2443390306c3SYan, Zheng if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) 2444390306c3SYan, Zheng check_caps = 1; 2445a8599bd8SSage Weil } 2446a8599bd8SSage Weil 2447a8599bd8SSage Weil cap->seq = seq; 2448a8599bd8SSage Weil 2449a8599bd8SSage Weil /* file layout may have changed */ 2450a8599bd8SSage Weil ci->i_layout = grant->layout; 2451a8599bd8SSage Weil 2452a8599bd8SSage Weil /* revocation, grant, or no-op? */ 2453a8599bd8SSage Weil if (cap->issued & ~newcaps) { 24543b454c49SSage Weil int revoking = cap->issued & ~newcaps; 24553b454c49SSage Weil 24563b454c49SSage Weil dout("revocation: %s -> %s (revoking %s)\n", 24573b454c49SSage Weil ceph_cap_string(cap->issued), 24583b454c49SSage Weil ceph_cap_string(newcaps), 24593b454c49SSage Weil ceph_cap_string(revoking)); 24600eb6cd49SSage Weil if (revoking & used & CEPH_CAP_FILE_BUFFER) 24613b454c49SSage Weil writeback = 1; /* initiate writeback; will delay ack */ 24623b454c49SSage Weil else if (revoking == CEPH_CAP_FILE_CACHE && 24633b454c49SSage Weil (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && 24643b454c49SSage Weil queue_invalidate) 24653b454c49SSage Weil ; /* do nothing yet, invalidation will be queued */ 24663b454c49SSage Weil else if (cap == ci->i_auth_cap) 24673b454c49SSage Weil check_caps = 1; /* check auth cap only */ 24683b454c49SSage Weil else 24693b454c49SSage Weil check_caps = 2; /* check all caps */ 2470a8599bd8SSage Weil cap->issued = newcaps; 2471978097c9SSage Weil cap->implemented |= newcaps; 2472a8599bd8SSage Weil } else if (cap->issued == newcaps) { 2473a8599bd8SSage Weil dout("caps unchanged: %s -> %s\n", 2474a8599bd8SSage Weil ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); 2475a8599bd8SSage Weil } else { 2476a8599bd8SSage Weil dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), 2477a8599bd8SSage Weil ceph_cap_string(newcaps)); 2478a8599bd8SSage Weil cap->issued = newcaps; 2479a8599bd8SSage Weil cap->implemented |= newcaps; /* add bits only, to 2480a8599bd8SSage Weil * avoid stepping on a 2481a8599bd8SSage Weil * pending revocation */ 2482a8599bd8SSage Weil wake = 1; 2483a8599bd8SSage Weil } 2484978097c9SSage Weil BUG_ON(cap->issued & ~cap->implemented); 2485a8599bd8SSage Weil 2486be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 24873c6f6b79SSage Weil if (writeback) 2488a8599bd8SSage Weil /* 2489a8599bd8SSage Weil * queue inode for writeback: we can't actually call 2490a8599bd8SSage Weil * filemap_write_and_wait, etc. from message handler 2491a8599bd8SSage Weil * context. 2492a8599bd8SSage Weil */ 24933c6f6b79SSage Weil ceph_queue_writeback(inode); 24943c6f6b79SSage Weil if (queue_invalidate) 24953c6f6b79SSage Weil ceph_queue_invalidate(inode); 2496a8599bd8SSage Weil if (wake) 249703066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 249815637c8bSSage Weil 249915637c8bSSage Weil if (check_caps == 1) 250015637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, 250115637c8bSSage Weil session); 250215637c8bSSage Weil else if (check_caps == 2) 250315637c8bSSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); 250415637c8bSSage Weil else 250515637c8bSSage Weil mutex_unlock(&session->s_mutex); 2506a8599bd8SSage Weil } 2507a8599bd8SSage Weil 2508a8599bd8SSage Weil /* 2509a8599bd8SSage Weil * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the 2510a8599bd8SSage Weil * MDS has been safely committed. 2511a8599bd8SSage Weil */ 25126df058c0SSage Weil static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, 2513a8599bd8SSage Weil struct ceph_mds_caps *m, 2514a8599bd8SSage Weil struct ceph_mds_session *session, 2515a8599bd8SSage Weil struct ceph_cap *cap) 2516be655596SSage Weil __releases(ci->i_ceph_lock) 2517a8599bd8SSage Weil { 2518a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 25193d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 2520a8599bd8SSage Weil unsigned seq = le32_to_cpu(m->seq); 2521a8599bd8SSage Weil int dirty = le32_to_cpu(m->dirty); 2522a8599bd8SSage Weil int cleaned = 0; 2523afcdaea3SSage Weil int drop = 0; 2524a8599bd8SSage Weil int i; 2525a8599bd8SSage Weil 2526a8599bd8SSage Weil for (i = 0; i < CEPH_CAP_BITS; i++) 2527a8599bd8SSage Weil if ((dirty & (1 << i)) && 2528a8599bd8SSage Weil flush_tid == ci->i_cap_flush_tid[i]) 2529a8599bd8SSage Weil cleaned |= 1 << i; 2530a8599bd8SSage Weil 2531a8599bd8SSage Weil dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," 2532a8599bd8SSage Weil " flushing %s -> %s\n", 2533a8599bd8SSage Weil inode, session->s_mds, seq, ceph_cap_string(dirty), 2534a8599bd8SSage Weil ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), 2535a8599bd8SSage Weil ceph_cap_string(ci->i_flushing_caps & ~cleaned)); 2536a8599bd8SSage Weil 2537a8599bd8SSage Weil if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned)) 2538a8599bd8SSage Weil goto out; 2539a8599bd8SSage Weil 2540a8599bd8SSage Weil ci->i_flushing_caps &= ~cleaned; 2541a8599bd8SSage Weil 2542a8599bd8SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2543a8599bd8SSage Weil if (ci->i_flushing_caps == 0) { 2544a8599bd8SSage Weil list_del_init(&ci->i_flushing_item); 2545a8599bd8SSage Weil if (!list_empty(&session->s_cap_flushing)) 2546a8599bd8SSage Weil dout(" mds%d still flushing cap on %p\n", 2547a8599bd8SSage Weil session->s_mds, 2548a8599bd8SSage Weil &list_entry(session->s_cap_flushing.next, 2549a8599bd8SSage Weil struct ceph_inode_info, 2550a8599bd8SSage Weil i_flushing_item)->vfs_inode); 2551a8599bd8SSage Weil mdsc->num_cap_flushing--; 255203066f23SYehuda Sadeh wake_up_all(&mdsc->cap_flushing_wq); 2553a8599bd8SSage Weil dout(" inode %p now !flushing\n", inode); 2554afcdaea3SSage Weil 2555afcdaea3SSage Weil if (ci->i_dirty_caps == 0) { 2556a8599bd8SSage Weil dout(" inode %p now clean\n", inode); 2557afcdaea3SSage Weil BUG_ON(!list_empty(&ci->i_dirty_item)); 2558afcdaea3SSage Weil drop = 1; 25597d8cb26dSSage Weil if (ci->i_wrbuffer_ref_head == 0) { 25607d8cb26dSSage Weil BUG_ON(!ci->i_head_snapc); 25617d8cb26dSSage Weil ceph_put_snap_context(ci->i_head_snapc); 25627d8cb26dSSage Weil ci->i_head_snapc = NULL; 25637d8cb26dSSage Weil } 256476e3b390SSage Weil } else { 256576e3b390SSage Weil BUG_ON(list_empty(&ci->i_dirty_item)); 2566afcdaea3SSage Weil } 2567a8599bd8SSage Weil } 2568a8599bd8SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 256903066f23SYehuda Sadeh wake_up_all(&ci->i_cap_wq); 2570a8599bd8SSage Weil 2571a8599bd8SSage Weil out: 2572be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2573afcdaea3SSage Weil if (drop) 2574a8599bd8SSage Weil iput(inode); 2575a8599bd8SSage Weil } 2576a8599bd8SSage Weil 2577a8599bd8SSage Weil /* 2578a8599bd8SSage Weil * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can 2579a8599bd8SSage Weil * throw away our cap_snap. 2580a8599bd8SSage Weil * 2581a8599bd8SSage Weil * Caller hold s_mutex. 2582a8599bd8SSage Weil */ 25836df058c0SSage Weil static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, 2584a8599bd8SSage Weil struct ceph_mds_caps *m, 2585a8599bd8SSage Weil struct ceph_mds_session *session) 2586a8599bd8SSage Weil { 2587a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2588a8599bd8SSage Weil u64 follows = le64_to_cpu(m->snap_follows); 2589a8599bd8SSage Weil struct ceph_cap_snap *capsnap; 2590a8599bd8SSage Weil int drop = 0; 2591a8599bd8SSage Weil 2592a8599bd8SSage Weil dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", 2593a8599bd8SSage Weil inode, ci, session->s_mds, follows); 2594a8599bd8SSage Weil 2595be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2596a8599bd8SSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 2597a8599bd8SSage Weil if (capsnap->follows == follows) { 2598a8599bd8SSage Weil if (capsnap->flush_tid != flush_tid) { 2599a8599bd8SSage Weil dout(" cap_snap %p follows %lld tid %lld !=" 2600a8599bd8SSage Weil " %lld\n", capsnap, follows, 2601a8599bd8SSage Weil flush_tid, capsnap->flush_tid); 2602a8599bd8SSage Weil break; 2603a8599bd8SSage Weil } 2604a8599bd8SSage Weil WARN_ON(capsnap->dirty_pages || capsnap->writing); 2605819ccbfaSSage Weil dout(" removing %p cap_snap %p follows %lld\n", 2606819ccbfaSSage Weil inode, capsnap, follows); 2607a8599bd8SSage Weil ceph_put_snap_context(capsnap->context); 2608a8599bd8SSage Weil list_del(&capsnap->ci_item); 2609a8599bd8SSage Weil list_del(&capsnap->flushing_item); 2610a8599bd8SSage Weil ceph_put_cap_snap(capsnap); 2611a8599bd8SSage Weil drop = 1; 2612a8599bd8SSage Weil break; 2613a8599bd8SSage Weil } else { 2614a8599bd8SSage Weil dout(" skipping cap_snap %p follows %lld\n", 2615a8599bd8SSage Weil capsnap, capsnap->follows); 2616a8599bd8SSage Weil } 2617a8599bd8SSage Weil } 2618be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2619a8599bd8SSage Weil if (drop) 2620a8599bd8SSage Weil iput(inode); 2621a8599bd8SSage Weil } 2622a8599bd8SSage Weil 2623a8599bd8SSage Weil /* 2624a8599bd8SSage Weil * Handle TRUNC from MDS, indicating file truncation. 2625a8599bd8SSage Weil * 2626a8599bd8SSage Weil * caller hold s_mutex. 2627a8599bd8SSage Weil */ 2628a8599bd8SSage Weil static void handle_cap_trunc(struct inode *inode, 2629a8599bd8SSage Weil struct ceph_mds_caps *trunc, 2630a8599bd8SSage Weil struct ceph_mds_session *session) 2631be655596SSage Weil __releases(ci->i_ceph_lock) 2632a8599bd8SSage Weil { 2633a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2634a8599bd8SSage Weil int mds = session->s_mds; 2635a8599bd8SSage Weil int seq = le32_to_cpu(trunc->seq); 2636a8599bd8SSage Weil u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); 2637a8599bd8SSage Weil u64 truncate_size = le64_to_cpu(trunc->truncate_size); 2638a8599bd8SSage Weil u64 size = le64_to_cpu(trunc->size); 2639a8599bd8SSage Weil int implemented = 0; 2640a8599bd8SSage Weil int dirty = __ceph_caps_dirty(ci); 2641a8599bd8SSage Weil int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); 2642a8599bd8SSage Weil int queue_trunc = 0; 2643a8599bd8SSage Weil 2644a8599bd8SSage Weil issued |= implemented | dirty; 2645a8599bd8SSage Weil 2646a8599bd8SSage Weil dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", 2647a8599bd8SSage Weil inode, mds, seq, truncate_size, truncate_seq); 2648a8599bd8SSage Weil queue_trunc = ceph_fill_file_size(inode, issued, 2649a8599bd8SSage Weil truncate_seq, truncate_size, size); 2650be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2651a8599bd8SSage Weil 2652a8599bd8SSage Weil if (queue_trunc) 26533c6f6b79SSage Weil ceph_queue_vmtruncate(inode); 2654a8599bd8SSage Weil } 2655a8599bd8SSage Weil 2656a8599bd8SSage Weil /* 2657a8599bd8SSage Weil * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a 2658a8599bd8SSage Weil * different one. If we are the most recent migration we've seen (as 2659a8599bd8SSage Weil * indicated by mseq), make note of the migrating cap bits for the 2660a8599bd8SSage Weil * duration (until we see the corresponding IMPORT). 2661a8599bd8SSage Weil * 2662a8599bd8SSage Weil * caller holds s_mutex 2663a8599bd8SSage Weil */ 2664a8599bd8SSage Weil static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, 2665154f42c2SSage Weil struct ceph_mds_session *session, 2666154f42c2SSage Weil int *open_target_sessions) 2667a8599bd8SSage Weil { 2668db354052SSage Weil struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 2669a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2670a8599bd8SSage Weil int mds = session->s_mds; 2671a8599bd8SSage Weil unsigned mseq = le32_to_cpu(ex->migrate_seq); 2672a8599bd8SSage Weil struct ceph_cap *cap = NULL, *t; 2673a8599bd8SSage Weil struct rb_node *p; 2674a8599bd8SSage Weil int remember = 1; 2675a8599bd8SSage Weil 2676a8599bd8SSage Weil dout("handle_cap_export inode %p ci %p mds%d mseq %d\n", 2677a8599bd8SSage Weil inode, ci, mds, mseq); 2678a8599bd8SSage Weil 2679be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2680a8599bd8SSage Weil 2681a8599bd8SSage Weil /* make sure we haven't seen a higher mseq */ 2682a8599bd8SSage Weil for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { 2683a8599bd8SSage Weil t = rb_entry(p, struct ceph_cap, ci_node); 2684a8599bd8SSage Weil if (ceph_seq_cmp(t->mseq, mseq) > 0) { 2685a8599bd8SSage Weil dout(" higher mseq on cap from mds%d\n", 2686a8599bd8SSage Weil t->session->s_mds); 2687a8599bd8SSage Weil remember = 0; 2688a8599bd8SSage Weil } 2689a8599bd8SSage Weil if (t->session->s_mds == mds) 2690a8599bd8SSage Weil cap = t; 2691a8599bd8SSage Weil } 2692a8599bd8SSage Weil 2693a8599bd8SSage Weil if (cap) { 2694a8599bd8SSage Weil if (remember) { 2695a8599bd8SSage Weil /* make note */ 2696a8599bd8SSage Weil ci->i_cap_exporting_mds = mds; 2697a8599bd8SSage Weil ci->i_cap_exporting_mseq = mseq; 2698a8599bd8SSage Weil ci->i_cap_exporting_issued = cap->issued; 2699154f42c2SSage Weil 2700154f42c2SSage Weil /* 2701154f42c2SSage Weil * make sure we have open sessions with all possible 2702154f42c2SSage Weil * export targets, so that we get the matching IMPORT 2703154f42c2SSage Weil */ 2704154f42c2SSage Weil *open_target_sessions = 1; 2705db354052SSage Weil 2706db354052SSage Weil /* 2707db354052SSage Weil * we can't flush dirty caps that we've seen the 2708db354052SSage Weil * EXPORT but no IMPORT for 2709db354052SSage Weil */ 2710db354052SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2711db354052SSage Weil if (!list_empty(&ci->i_dirty_item)) { 2712db354052SSage Weil dout(" moving %p to cap_dirty_migrating\n", 2713db354052SSage Weil inode); 2714db354052SSage Weil list_move(&ci->i_dirty_item, 2715db354052SSage Weil &mdsc->cap_dirty_migrating); 2716db354052SSage Weil } 2717db354052SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2718a8599bd8SSage Weil } 27197c1332b8SSage Weil __ceph_remove_cap(cap); 2720a8599bd8SSage Weil } 27214ea0043aSSage Weil /* else, we already released it */ 2722a8599bd8SSage Weil 2723be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2724a8599bd8SSage Weil } 2725a8599bd8SSage Weil 2726a8599bd8SSage Weil /* 2727a8599bd8SSage Weil * Handle cap IMPORT. If there are temp bits from an older EXPORT, 2728a8599bd8SSage Weil * clean them up. 2729a8599bd8SSage Weil * 2730a8599bd8SSage Weil * caller holds s_mutex. 2731a8599bd8SSage Weil */ 2732a8599bd8SSage Weil static void handle_cap_import(struct ceph_mds_client *mdsc, 2733a8599bd8SSage Weil struct inode *inode, struct ceph_mds_caps *im, 2734a8599bd8SSage Weil struct ceph_mds_session *session, 2735a8599bd8SSage Weil void *snaptrace, int snaptrace_len) 2736a8599bd8SSage Weil { 2737a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2738a8599bd8SSage Weil int mds = session->s_mds; 2739a8599bd8SSage Weil unsigned issued = le32_to_cpu(im->caps); 2740a8599bd8SSage Weil unsigned wanted = le32_to_cpu(im->wanted); 2741a8599bd8SSage Weil unsigned seq = le32_to_cpu(im->seq); 2742a8599bd8SSage Weil unsigned mseq = le32_to_cpu(im->migrate_seq); 2743a8599bd8SSage Weil u64 realmino = le64_to_cpu(im->realm); 2744a8599bd8SSage Weil u64 cap_id = le64_to_cpu(im->cap_id); 2745a8599bd8SSage Weil 2746a8599bd8SSage Weil if (ci->i_cap_exporting_mds >= 0 && 2747a8599bd8SSage Weil ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) { 2748a8599bd8SSage Weil dout("handle_cap_import inode %p ci %p mds%d mseq %d" 2749a8599bd8SSage Weil " - cleared exporting from mds%d\n", 2750a8599bd8SSage Weil inode, ci, mds, mseq, 2751a8599bd8SSage Weil ci->i_cap_exporting_mds); 2752a8599bd8SSage Weil ci->i_cap_exporting_issued = 0; 2753a8599bd8SSage Weil ci->i_cap_exporting_mseq = 0; 2754a8599bd8SSage Weil ci->i_cap_exporting_mds = -1; 2755db354052SSage Weil 2756db354052SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2757db354052SSage Weil if (!list_empty(&ci->i_dirty_item)) { 2758db354052SSage Weil dout(" moving %p back to cap_dirty\n", inode); 2759db354052SSage Weil list_move(&ci->i_dirty_item, &mdsc->cap_dirty); 2760db354052SSage Weil } 2761db354052SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2762a8599bd8SSage Weil } else { 2763a8599bd8SSage Weil dout("handle_cap_import inode %p ci %p mds%d mseq %d\n", 2764a8599bd8SSage Weil inode, ci, mds, mseq); 2765a8599bd8SSage Weil } 2766a8599bd8SSage Weil 2767a8599bd8SSage Weil down_write(&mdsc->snap_rwsem); 2768a8599bd8SSage Weil ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len, 2769a8599bd8SSage Weil false); 2770a8599bd8SSage Weil downgrade_write(&mdsc->snap_rwsem); 2771a8599bd8SSage Weil ceph_add_cap(inode, session, cap_id, -1, 2772a8599bd8SSage Weil issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH, 2773a8599bd8SSage Weil NULL /* no caps context */); 2774088b3f5eSSage Weil kick_flushing_inode_caps(mdsc, session, inode); 2775a8599bd8SSage Weil up_read(&mdsc->snap_rwsem); 2776feb4cc9bSSage Weil 2777feb4cc9bSSage Weil /* make sure we re-request max_size, if necessary */ 2778be655596SSage Weil spin_lock(&ci->i_ceph_lock); 27790e5e1774SYan, Zheng ci->i_wanted_max_size = 0; /* reset */ 2780feb4cc9bSSage Weil ci->i_requested_max_size = 0; 2781be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2782a8599bd8SSage Weil } 2783a8599bd8SSage Weil 2784a8599bd8SSage Weil /* 2785a8599bd8SSage Weil * Handle a caps message from the MDS. 2786a8599bd8SSage Weil * 2787a8599bd8SSage Weil * Identify the appropriate session, inode, and call the right handler 2788a8599bd8SSage Weil * based on the cap op. 2789a8599bd8SSage Weil */ 2790a8599bd8SSage Weil void ceph_handle_caps(struct ceph_mds_session *session, 2791a8599bd8SSage Weil struct ceph_msg *msg) 2792a8599bd8SSage Weil { 2793a8599bd8SSage Weil struct ceph_mds_client *mdsc = session->s_mdsc; 27943d14c5d2SYehuda Sadeh struct super_block *sb = mdsc->fsc->sb; 2795a8599bd8SSage Weil struct inode *inode; 2796be655596SSage Weil struct ceph_inode_info *ci; 2797a8599bd8SSage Weil struct ceph_cap *cap; 2798a8599bd8SSage Weil struct ceph_mds_caps *h; 27992600d2ddSSage Weil int mds = session->s_mds; 2800a8599bd8SSage Weil int op; 28013d7ded4dSSage Weil u32 seq, mseq; 2802a8599bd8SSage Weil struct ceph_vino vino; 2803a8599bd8SSage Weil u64 cap_id; 2804a8599bd8SSage Weil u64 size, max_size; 28056df058c0SSage Weil u64 tid; 280670edb55bSSage Weil void *snaptrace; 2807ce1fbc8dSSage Weil size_t snaptrace_len; 2808ce1fbc8dSSage Weil void *flock; 2809ce1fbc8dSSage Weil u32 flock_len; 2810154f42c2SSage Weil int open_target_sessions = 0; 2811a8599bd8SSage Weil 2812a8599bd8SSage Weil dout("handle_caps from mds%d\n", mds); 2813a8599bd8SSage Weil 2814a8599bd8SSage Weil /* decode */ 28156df058c0SSage Weil tid = le64_to_cpu(msg->hdr.tid); 2816a8599bd8SSage Weil if (msg->front.iov_len < sizeof(*h)) 2817a8599bd8SSage Weil goto bad; 2818a8599bd8SSage Weil h = msg->front.iov_base; 2819a8599bd8SSage Weil op = le32_to_cpu(h->op); 2820a8599bd8SSage Weil vino.ino = le64_to_cpu(h->ino); 2821a8599bd8SSage Weil vino.snap = CEPH_NOSNAP; 2822a8599bd8SSage Weil cap_id = le64_to_cpu(h->cap_id); 2823a8599bd8SSage Weil seq = le32_to_cpu(h->seq); 28243d7ded4dSSage Weil mseq = le32_to_cpu(h->migrate_seq); 2825a8599bd8SSage Weil size = le64_to_cpu(h->size); 2826a8599bd8SSage Weil max_size = le64_to_cpu(h->max_size); 2827a8599bd8SSage Weil 2828ce1fbc8dSSage Weil snaptrace = h + 1; 2829ce1fbc8dSSage Weil snaptrace_len = le32_to_cpu(h->snap_trace_len); 2830ce1fbc8dSSage Weil 2831ce1fbc8dSSage Weil if (le16_to_cpu(msg->hdr.version) >= 2) { 2832ce1fbc8dSSage Weil void *p, *end; 2833ce1fbc8dSSage Weil 2834ce1fbc8dSSage Weil p = snaptrace + snaptrace_len; 2835ce1fbc8dSSage Weil end = msg->front.iov_base + msg->front.iov_len; 2836ce1fbc8dSSage Weil ceph_decode_32_safe(&p, end, flock_len, bad); 2837ce1fbc8dSSage Weil flock = p; 2838ce1fbc8dSSage Weil } else { 2839ce1fbc8dSSage Weil flock = NULL; 2840ce1fbc8dSSage Weil flock_len = 0; 2841ce1fbc8dSSage Weil } 2842ce1fbc8dSSage Weil 2843a8599bd8SSage Weil mutex_lock(&session->s_mutex); 2844a8599bd8SSage Weil session->s_seq++; 2845a8599bd8SSage Weil dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, 2846a8599bd8SSage Weil (unsigned)seq); 2847a8599bd8SSage Weil 284866f58691SYan, Zheng if (op == CEPH_CAP_OP_IMPORT) 284966f58691SYan, Zheng ceph_add_cap_releases(mdsc, session); 285066f58691SYan, Zheng 2851a8599bd8SSage Weil /* lookup ino */ 2852a8599bd8SSage Weil inode = ceph_find_inode(sb, vino); 2853be655596SSage Weil ci = ceph_inode(inode); 2854a8599bd8SSage Weil dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, 2855a8599bd8SSage Weil vino.snap, inode); 2856a8599bd8SSage Weil if (!inode) { 2857a8599bd8SSage Weil dout(" i don't have ino %llx\n", vino.ino); 28583d7ded4dSSage Weil 28593d7ded4dSSage Weil if (op == CEPH_CAP_OP_IMPORT) 28603d7ded4dSSage Weil __queue_cap_release(session, vino.ino, cap_id, 28613d7ded4dSSage Weil mseq, seq); 286221b559deSGreg Farnum goto flush_cap_releases; 2863a8599bd8SSage Weil } 2864a8599bd8SSage Weil 2865a8599bd8SSage Weil /* these will work even if we don't have a cap yet */ 2866a8599bd8SSage Weil switch (op) { 2867a8599bd8SSage Weil case CEPH_CAP_OP_FLUSHSNAP_ACK: 28686df058c0SSage Weil handle_cap_flushsnap_ack(inode, tid, h, session); 2869a8599bd8SSage Weil goto done; 2870a8599bd8SSage Weil 2871a8599bd8SSage Weil case CEPH_CAP_OP_EXPORT: 2872154f42c2SSage Weil handle_cap_export(inode, h, session, &open_target_sessions); 2873a8599bd8SSage Weil goto done; 2874a8599bd8SSage Weil 2875a8599bd8SSage Weil case CEPH_CAP_OP_IMPORT: 2876a8599bd8SSage Weil handle_cap_import(mdsc, inode, h, session, 2877ce1fbc8dSSage Weil snaptrace, snaptrace_len); 2878a8599bd8SSage Weil } 2879a8599bd8SSage Weil 2880a8599bd8SSage Weil /* the rest require a cap */ 2881be655596SSage Weil spin_lock(&ci->i_ceph_lock); 2882a8599bd8SSage Weil cap = __get_cap_for_mds(ceph_inode(inode), mds); 2883a8599bd8SSage Weil if (!cap) { 28849dbd412fSSage Weil dout(" no cap on %p ino %llx.%llx from mds%d\n", 2885a8599bd8SSage Weil inode, ceph_ino(inode), ceph_snap(inode), mds); 2886be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 288721b559deSGreg Farnum goto flush_cap_releases; 2888a8599bd8SSage Weil } 2889a8599bd8SSage Weil 2890be655596SSage Weil /* note that each of these drops i_ceph_lock for us */ 2891a8599bd8SSage Weil switch (op) { 2892a8599bd8SSage Weil case CEPH_CAP_OP_REVOKE: 2893a8599bd8SSage Weil case CEPH_CAP_OP_GRANT: 28940e5e1774SYan, Zheng case CEPH_CAP_OP_IMPORT: 289515637c8bSSage Weil handle_cap_grant(inode, h, session, cap, msg->middle); 289615637c8bSSage Weil goto done_unlocked; 2897a8599bd8SSage Weil 2898a8599bd8SSage Weil case CEPH_CAP_OP_FLUSH_ACK: 28996df058c0SSage Weil handle_cap_flush_ack(inode, tid, h, session, cap); 2900a8599bd8SSage Weil break; 2901a8599bd8SSage Weil 2902a8599bd8SSage Weil case CEPH_CAP_OP_TRUNC: 2903a8599bd8SSage Weil handle_cap_trunc(inode, h, session); 2904a8599bd8SSage Weil break; 2905a8599bd8SSage Weil 2906a8599bd8SSage Weil default: 2907be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 2908a8599bd8SSage Weil pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, 2909a8599bd8SSage Weil ceph_cap_op_name(op)); 2910a8599bd8SSage Weil } 2911a8599bd8SSage Weil 291221b559deSGreg Farnum goto done; 291321b559deSGreg Farnum 291421b559deSGreg Farnum flush_cap_releases: 291521b559deSGreg Farnum /* 291621b559deSGreg Farnum * send any full release message to try to move things 291721b559deSGreg Farnum * along for the mds (who clearly thinks we still have this 291821b559deSGreg Farnum * cap). 291921b559deSGreg Farnum */ 292021b559deSGreg Farnum ceph_add_cap_releases(mdsc, session); 292121b559deSGreg Farnum ceph_send_cap_releases(mdsc, session); 292221b559deSGreg Farnum 2923a8599bd8SSage Weil done: 2924a8599bd8SSage Weil mutex_unlock(&session->s_mutex); 292515637c8bSSage Weil done_unlocked: 2926a8599bd8SSage Weil if (inode) 2927a8599bd8SSage Weil iput(inode); 2928154f42c2SSage Weil if (open_target_sessions) 2929154f42c2SSage Weil ceph_mdsc_open_export_target_sessions(mdsc, session); 2930a8599bd8SSage Weil return; 2931a8599bd8SSage Weil 2932a8599bd8SSage Weil bad: 2933a8599bd8SSage Weil pr_err("ceph_handle_caps: corrupt message\n"); 29349ec7cab1SSage Weil ceph_msg_dump(msg); 2935a8599bd8SSage Weil return; 2936a8599bd8SSage Weil } 2937a8599bd8SSage Weil 2938a8599bd8SSage Weil /* 2939a8599bd8SSage Weil * Delayed work handler to process end of delayed cap release LRU list. 2940a8599bd8SSage Weil */ 2941afcdaea3SSage Weil void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) 2942a8599bd8SSage Weil { 2943a8599bd8SSage Weil struct ceph_inode_info *ci; 2944a8599bd8SSage Weil int flags = CHECK_CAPS_NODELAY; 2945a8599bd8SSage Weil 2946a8599bd8SSage Weil dout("check_delayed_caps\n"); 2947a8599bd8SSage Weil while (1) { 2948a8599bd8SSage Weil spin_lock(&mdsc->cap_delay_lock); 2949a8599bd8SSage Weil if (list_empty(&mdsc->cap_delay_list)) 2950a8599bd8SSage Weil break; 2951a8599bd8SSage Weil ci = list_first_entry(&mdsc->cap_delay_list, 2952a8599bd8SSage Weil struct ceph_inode_info, 2953a8599bd8SSage Weil i_cap_delay_list); 2954a8599bd8SSage Weil if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && 2955a8599bd8SSage Weil time_before(jiffies, ci->i_hold_caps_max)) 2956a8599bd8SSage Weil break; 2957a8599bd8SSage Weil list_del_init(&ci->i_cap_delay_list); 2958a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 2959a8599bd8SSage Weil dout("check_delayed_caps on %p\n", &ci->vfs_inode); 2960a8599bd8SSage Weil ceph_check_caps(ci, flags, NULL); 2961a8599bd8SSage Weil } 2962a8599bd8SSage Weil spin_unlock(&mdsc->cap_delay_lock); 2963a8599bd8SSage Weil } 2964a8599bd8SSage Weil 2965a8599bd8SSage Weil /* 2966afcdaea3SSage Weil * Flush all dirty caps to the mds 2967afcdaea3SSage Weil */ 2968afcdaea3SSage Weil void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) 2969afcdaea3SSage Weil { 2970db354052SSage Weil struct ceph_inode_info *ci; 2971db354052SSage Weil struct inode *inode; 2972afcdaea3SSage Weil 2973afcdaea3SSage Weil dout("flush_dirty_caps\n"); 2974afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2975db354052SSage Weil while (!list_empty(&mdsc->cap_dirty)) { 2976db354052SSage Weil ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info, 2977afcdaea3SSage Weil i_dirty_item); 297870b666c3SSage Weil inode = &ci->vfs_inode; 297970b666c3SSage Weil ihold(inode); 2980db354052SSage Weil dout("flush_dirty_caps %p\n", inode); 2981afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 298270b666c3SSage Weil ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL); 2983afcdaea3SSage Weil iput(inode); 2984afcdaea3SSage Weil spin_lock(&mdsc->cap_dirty_lock); 2985afcdaea3SSage Weil } 2986afcdaea3SSage Weil spin_unlock(&mdsc->cap_dirty_lock); 2987db354052SSage Weil dout("flush_dirty_caps done\n"); 2988afcdaea3SSage Weil } 2989afcdaea3SSage Weil 2990afcdaea3SSage Weil /* 2991a8599bd8SSage Weil * Drop open file reference. If we were the last open file, 2992a8599bd8SSage Weil * we may need to release capabilities to the MDS (or schedule 2993a8599bd8SSage Weil * their delayed release). 2994a8599bd8SSage Weil */ 2995a8599bd8SSage Weil void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) 2996a8599bd8SSage Weil { 2997a8599bd8SSage Weil struct inode *inode = &ci->vfs_inode; 2998a8599bd8SSage Weil int last = 0; 2999a8599bd8SSage Weil 3000be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3001a8599bd8SSage Weil dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode, 3002a8599bd8SSage Weil ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1); 3003a8599bd8SSage Weil BUG_ON(ci->i_nr_by_mode[fmode] == 0); 3004a8599bd8SSage Weil if (--ci->i_nr_by_mode[fmode] == 0) 3005a8599bd8SSage Weil last++; 3006be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3007a8599bd8SSage Weil 3008a8599bd8SSage Weil if (last && ci->i_vino.snap == CEPH_NOSNAP) 3009a8599bd8SSage Weil ceph_check_caps(ci, 0, NULL); 3010a8599bd8SSage Weil } 3011a8599bd8SSage Weil 3012a8599bd8SSage Weil /* 3013a8599bd8SSage Weil * Helpers for embedding cap and dentry lease releases into mds 3014a8599bd8SSage Weil * requests. 3015a8599bd8SSage Weil * 3016a8599bd8SSage Weil * @force is used by dentry_release (below) to force inclusion of a 3017a8599bd8SSage Weil * record for the directory inode, even when there aren't any caps to 3018a8599bd8SSage Weil * drop. 3019a8599bd8SSage Weil */ 3020a8599bd8SSage Weil int ceph_encode_inode_release(void **p, struct inode *inode, 3021a8599bd8SSage Weil int mds, int drop, int unless, int force) 3022a8599bd8SSage Weil { 3023a8599bd8SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 3024a8599bd8SSage Weil struct ceph_cap *cap; 3025a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3026ec97f88bSSage Weil int used, dirty; 3027a8599bd8SSage Weil int ret = 0; 3028a8599bd8SSage Weil 3029be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3030916623daSSage Weil used = __ceph_caps_used(ci); 3031ec97f88bSSage Weil dirty = __ceph_caps_dirty(ci); 3032916623daSSage Weil 3033ec97f88bSSage Weil dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", 3034ec97f88bSSage Weil inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), 3035916623daSSage Weil ceph_cap_string(unless)); 3036916623daSSage Weil 3037ec97f88bSSage Weil /* only drop unused, clean caps */ 3038ec97f88bSSage Weil drop &= ~(used | dirty); 3039916623daSSage Weil 3040a8599bd8SSage Weil cap = __get_cap_for_mds(ci, mds); 3041a8599bd8SSage Weil if (cap && __cap_is_valid(cap)) { 3042a8599bd8SSage Weil if (force || 3043a8599bd8SSage Weil ((cap->issued & drop) && 3044a8599bd8SSage Weil (cap->issued & unless) == 0)) { 3045a8599bd8SSage Weil if ((cap->issued & drop) && 3046a8599bd8SSage Weil (cap->issued & unless) == 0) { 3047bb137f84SYan, Zheng int wanted = __ceph_caps_wanted(ci); 3048bb137f84SYan, Zheng if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0) 3049bb137f84SYan, Zheng wanted |= cap->mds_wanted; 3050bb137f84SYan, Zheng dout("encode_inode_release %p cap %p " 3051bb137f84SYan, Zheng "%s -> %s, wanted %s -> %s\n", inode, cap, 3052a8599bd8SSage Weil ceph_cap_string(cap->issued), 3053bb137f84SYan, Zheng ceph_cap_string(cap->issued & ~drop), 3054bb137f84SYan, Zheng ceph_cap_string(cap->mds_wanted), 3055bb137f84SYan, Zheng ceph_cap_string(wanted)); 3056bb137f84SYan, Zheng 3057a8599bd8SSage Weil cap->issued &= ~drop; 3058a8599bd8SSage Weil cap->implemented &= ~drop; 3059bb137f84SYan, Zheng cap->mds_wanted = wanted; 3060a8599bd8SSage Weil } else { 3061a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s" 3062a8599bd8SSage Weil " (force)\n", inode, cap, 3063a8599bd8SSage Weil ceph_cap_string(cap->issued)); 3064a8599bd8SSage Weil } 3065a8599bd8SSage Weil 3066a8599bd8SSage Weil rel->ino = cpu_to_le64(ceph_ino(inode)); 3067a8599bd8SSage Weil rel->cap_id = cpu_to_le64(cap->cap_id); 3068a8599bd8SSage Weil rel->seq = cpu_to_le32(cap->seq); 3069a8599bd8SSage Weil rel->issue_seq = cpu_to_le32(cap->issue_seq), 3070a8599bd8SSage Weil rel->mseq = cpu_to_le32(cap->mseq); 3071a8599bd8SSage Weil rel->caps = cpu_to_le32(cap->issued); 3072a8599bd8SSage Weil rel->wanted = cpu_to_le32(cap->mds_wanted); 3073a8599bd8SSage Weil rel->dname_len = 0; 3074a8599bd8SSage Weil rel->dname_seq = 0; 3075a8599bd8SSage Weil *p += sizeof(*rel); 3076a8599bd8SSage Weil ret = 1; 3077a8599bd8SSage Weil } else { 3078a8599bd8SSage Weil dout("encode_inode_release %p cap %p %s\n", 3079a8599bd8SSage Weil inode, cap, ceph_cap_string(cap->issued)); 3080a8599bd8SSage Weil } 3081a8599bd8SSage Weil } 3082be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 3083a8599bd8SSage Weil return ret; 3084a8599bd8SSage Weil } 3085a8599bd8SSage Weil 3086a8599bd8SSage Weil int ceph_encode_dentry_release(void **p, struct dentry *dentry, 3087a8599bd8SSage Weil int mds, int drop, int unless) 3088a8599bd8SSage Weil { 3089a8599bd8SSage Weil struct inode *dir = dentry->d_parent->d_inode; 3090a8599bd8SSage Weil struct ceph_mds_request_release *rel = *p; 3091a8599bd8SSage Weil struct ceph_dentry_info *di = ceph_dentry(dentry); 3092a8599bd8SSage Weil int force = 0; 3093a8599bd8SSage Weil int ret; 3094a8599bd8SSage Weil 3095a8599bd8SSage Weil /* 3096a8599bd8SSage Weil * force an record for the directory caps if we have a dentry lease. 3097be655596SSage Weil * this is racy (can't take i_ceph_lock and d_lock together), but it 3098a8599bd8SSage Weil * doesn't have to be perfect; the mds will revoke anything we don't 3099a8599bd8SSage Weil * release. 3100a8599bd8SSage Weil */ 3101a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3102a8599bd8SSage Weil if (di->lease_session && di->lease_session->s_mds == mds) 3103a8599bd8SSage Weil force = 1; 3104a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3105a8599bd8SSage Weil 3106a8599bd8SSage Weil ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); 3107a8599bd8SSage Weil 3108a8599bd8SSage Weil spin_lock(&dentry->d_lock); 3109a8599bd8SSage Weil if (ret && di->lease_session && di->lease_session->s_mds == mds) { 3110a8599bd8SSage Weil dout("encode_dentry_release %p mds%d seq %d\n", 3111a8599bd8SSage Weil dentry, mds, (int)di->lease_seq); 3112a8599bd8SSage Weil rel->dname_len = cpu_to_le32(dentry->d_name.len); 3113a8599bd8SSage Weil memcpy(*p, dentry->d_name.name, dentry->d_name.len); 3114a8599bd8SSage Weil *p += dentry->d_name.len; 3115a8599bd8SSage Weil rel->dname_seq = cpu_to_le32(di->lease_seq); 31161dadcce3SSage Weil __ceph_mdsc_drop_dentry_lease(dentry); 3117a8599bd8SSage Weil } 3118a8599bd8SSage Weil spin_unlock(&dentry->d_lock); 3119a8599bd8SSage Weil return ret; 3120a8599bd8SSage Weil } 3121