1a4ce40a9SAlex Elder 23d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 33d14c5d2SYehuda Sadeh 43d14c5d2SYehuda Sadeh #include <linux/module.h> 53d14c5d2SYehuda Sadeh #include <linux/err.h> 63d14c5d2SYehuda Sadeh #include <linux/highmem.h> 73d14c5d2SYehuda Sadeh #include <linux/mm.h> 83d14c5d2SYehuda Sadeh #include <linux/pagemap.h> 93d14c5d2SYehuda Sadeh #include <linux/slab.h> 103d14c5d2SYehuda Sadeh #include <linux/uaccess.h> 113d14c5d2SYehuda Sadeh #ifdef CONFIG_BLOCK 123d14c5d2SYehuda Sadeh #include <linux/bio.h> 133d14c5d2SYehuda Sadeh #endif 143d14c5d2SYehuda Sadeh 153d14c5d2SYehuda Sadeh #include <linux/ceph/libceph.h> 163d14c5d2SYehuda Sadeh #include <linux/ceph/osd_client.h> 173d14c5d2SYehuda Sadeh #include <linux/ceph/messenger.h> 183d14c5d2SYehuda Sadeh #include <linux/ceph/decode.h> 193d14c5d2SYehuda Sadeh #include <linux/ceph/auth.h> 203d14c5d2SYehuda Sadeh #include <linux/ceph/pagelist.h> 213d14c5d2SYehuda Sadeh 223d14c5d2SYehuda Sadeh #define OSD_OPREPLY_FRONT_LEN 512 233d14c5d2SYehuda Sadeh 245522ae0bSAlex Elder static struct kmem_cache *ceph_osd_request_cache; 255522ae0bSAlex Elder 263d14c5d2SYehuda Sadeh static const struct ceph_connection_operations osd_con_ops; 273d14c5d2SYehuda Sadeh 283d14c5d2SYehuda Sadeh /* 293d14c5d2SYehuda Sadeh * Implement client access to distributed object storage cluster. 303d14c5d2SYehuda Sadeh * 313d14c5d2SYehuda Sadeh * All data objects are stored within a cluster/cloud of OSDs, or 323d14c5d2SYehuda Sadeh * "object storage devices." (Note that Ceph OSDs have _nothing_ to 333d14c5d2SYehuda Sadeh * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply 343d14c5d2SYehuda Sadeh * remote daemons serving up and coordinating consistent and safe 353d14c5d2SYehuda Sadeh * access to storage. 363d14c5d2SYehuda Sadeh * 373d14c5d2SYehuda Sadeh * Cluster membership and the mapping of data objects onto storage devices 383d14c5d2SYehuda Sadeh * are described by the osd map. 393d14c5d2SYehuda Sadeh * 403d14c5d2SYehuda Sadeh * We keep track of pending OSD requests (read, write), resubmit 413d14c5d2SYehuda Sadeh * requests to different OSDs when the cluster topology/data layout 423d14c5d2SYehuda Sadeh * change, or retry the affected requests when the communications 433d14c5d2SYehuda Sadeh * channel with an OSD is reset. 443d14c5d2SYehuda Sadeh */ 453d14c5d2SYehuda Sadeh 465aea3dcdSIlya Dryomov static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req); 475aea3dcdSIlya Dryomov static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req); 48922dab61SIlya Dryomov static void link_linger(struct ceph_osd *osd, 49922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq); 50922dab61SIlya Dryomov static void unlink_linger(struct ceph_osd *osd, 51922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq); 525aea3dcdSIlya Dryomov 535aea3dcdSIlya Dryomov #if 1 545aea3dcdSIlya Dryomov static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem) 555aea3dcdSIlya Dryomov { 565aea3dcdSIlya Dryomov bool wrlocked = true; 575aea3dcdSIlya Dryomov 585aea3dcdSIlya Dryomov if (unlikely(down_read_trylock(sem))) { 595aea3dcdSIlya Dryomov wrlocked = false; 605aea3dcdSIlya Dryomov up_read(sem); 615aea3dcdSIlya Dryomov } 625aea3dcdSIlya Dryomov 635aea3dcdSIlya Dryomov return wrlocked; 645aea3dcdSIlya Dryomov } 655aea3dcdSIlya Dryomov static inline void verify_osdc_locked(struct ceph_osd_client *osdc) 665aea3dcdSIlya Dryomov { 675aea3dcdSIlya Dryomov WARN_ON(!rwsem_is_locked(&osdc->lock)); 685aea3dcdSIlya Dryomov } 695aea3dcdSIlya Dryomov static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) 705aea3dcdSIlya Dryomov { 715aea3dcdSIlya Dryomov WARN_ON(!rwsem_is_wrlocked(&osdc->lock)); 725aea3dcdSIlya Dryomov } 735aea3dcdSIlya Dryomov static inline void verify_osd_locked(struct ceph_osd *osd) 745aea3dcdSIlya Dryomov { 755aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 765aea3dcdSIlya Dryomov 775aea3dcdSIlya Dryomov WARN_ON(!(mutex_is_locked(&osd->lock) && 785aea3dcdSIlya Dryomov rwsem_is_locked(&osdc->lock)) && 795aea3dcdSIlya Dryomov !rwsem_is_wrlocked(&osdc->lock)); 805aea3dcdSIlya Dryomov } 81922dab61SIlya Dryomov static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) 82922dab61SIlya Dryomov { 83922dab61SIlya Dryomov WARN_ON(!mutex_is_locked(&lreq->lock)); 84922dab61SIlya Dryomov } 855aea3dcdSIlya Dryomov #else 865aea3dcdSIlya Dryomov static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { } 875aea3dcdSIlya Dryomov static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { } 885aea3dcdSIlya Dryomov static inline void verify_osd_locked(struct ceph_osd *osd) { } 89922dab61SIlya Dryomov static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { } 905aea3dcdSIlya Dryomov #endif 915aea3dcdSIlya Dryomov 923d14c5d2SYehuda Sadeh /* 933d14c5d2SYehuda Sadeh * calculate the mapping of a file extent onto an object, and fill out the 943d14c5d2SYehuda Sadeh * request accordingly. shorten extent as necessary if it crosses an 953d14c5d2SYehuda Sadeh * object boundary. 963d14c5d2SYehuda Sadeh * 973d14c5d2SYehuda Sadeh * fill osd op in request message. 983d14c5d2SYehuda Sadeh */ 99dbe0fc41SAlex Elder static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, 100a19dadfbSAlex Elder u64 *objnum, u64 *objoff, u64 *objlen) 1013d14c5d2SYehuda Sadeh { 10260e56f13SAlex Elder u64 orig_len = *plen; 103d63b77f4SSage Weil int r; 1043d14c5d2SYehuda Sadeh 10560e56f13SAlex Elder /* object extent? */ 10675d1c941SAlex Elder r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, 10775d1c941SAlex Elder objoff, objlen); 108d63b77f4SSage Weil if (r < 0) 109d63b77f4SSage Weil return r; 11075d1c941SAlex Elder if (*objlen < orig_len) { 11175d1c941SAlex Elder *plen = *objlen; 11260e56f13SAlex Elder dout(" skipping last %llu, final file extent %llu~%llu\n", 11360e56f13SAlex Elder orig_len - *plen, off, *plen); 11460e56f13SAlex Elder } 11560e56f13SAlex Elder 11675d1c941SAlex Elder dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); 1173d14c5d2SYehuda Sadeh 1183ff5f385SAlex Elder return 0; 1193d14c5d2SYehuda Sadeh } 1203d14c5d2SYehuda Sadeh 121c54d47bfSAlex Elder static void ceph_osd_data_init(struct ceph_osd_data *osd_data) 122c54d47bfSAlex Elder { 123c54d47bfSAlex Elder memset(osd_data, 0, sizeof (*osd_data)); 124c54d47bfSAlex Elder osd_data->type = CEPH_OSD_DATA_TYPE_NONE; 125c54d47bfSAlex Elder } 126c54d47bfSAlex Elder 127a4ce40a9SAlex Elder static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, 12843bfe5deSAlex Elder struct page **pages, u64 length, u32 alignment, 12943bfe5deSAlex Elder bool pages_from_pool, bool own_pages) 13043bfe5deSAlex Elder { 13143bfe5deSAlex Elder osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; 13243bfe5deSAlex Elder osd_data->pages = pages; 13343bfe5deSAlex Elder osd_data->length = length; 13443bfe5deSAlex Elder osd_data->alignment = alignment; 13543bfe5deSAlex Elder osd_data->pages_from_pool = pages_from_pool; 13643bfe5deSAlex Elder osd_data->own_pages = own_pages; 13743bfe5deSAlex Elder } 13843bfe5deSAlex Elder 139a4ce40a9SAlex Elder static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, 14043bfe5deSAlex Elder struct ceph_pagelist *pagelist) 14143bfe5deSAlex Elder { 14243bfe5deSAlex Elder osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; 14343bfe5deSAlex Elder osd_data->pagelist = pagelist; 14443bfe5deSAlex Elder } 14543bfe5deSAlex Elder 14643bfe5deSAlex Elder #ifdef CONFIG_BLOCK 147a4ce40a9SAlex Elder static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, 14843bfe5deSAlex Elder struct bio *bio, size_t bio_length) 14943bfe5deSAlex Elder { 15043bfe5deSAlex Elder osd_data->type = CEPH_OSD_DATA_TYPE_BIO; 15143bfe5deSAlex Elder osd_data->bio = bio; 15243bfe5deSAlex Elder osd_data->bio_length = bio_length; 15343bfe5deSAlex Elder } 15443bfe5deSAlex Elder #endif /* CONFIG_BLOCK */ 15543bfe5deSAlex Elder 156863c7eb5SAlex Elder #define osd_req_op_data(oreq, whch, typ, fld) \ 157863c7eb5SAlex Elder ({ \ 1588a703a38SIoana Ciornei struct ceph_osd_request *__oreq = (oreq); \ 1598a703a38SIoana Ciornei unsigned int __whch = (whch); \ 1608a703a38SIoana Ciornei BUG_ON(__whch >= __oreq->r_num_ops); \ 1618a703a38SIoana Ciornei &__oreq->r_ops[__whch].typ.fld; \ 162863c7eb5SAlex Elder }) 163863c7eb5SAlex Elder 16449719778SAlex Elder static struct ceph_osd_data * 16549719778SAlex Elder osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) 16649719778SAlex Elder { 16749719778SAlex Elder BUG_ON(which >= osd_req->r_num_ops); 16849719778SAlex Elder 16949719778SAlex Elder return &osd_req->r_ops[which].raw_data_in; 17049719778SAlex Elder } 17149719778SAlex Elder 172a4ce40a9SAlex Elder struct ceph_osd_data * 173a4ce40a9SAlex Elder osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, 174406e2c9fSAlex Elder unsigned int which) 175a4ce40a9SAlex Elder { 176863c7eb5SAlex Elder return osd_req_op_data(osd_req, which, extent, osd_data); 177a4ce40a9SAlex Elder } 178a4ce40a9SAlex Elder EXPORT_SYMBOL(osd_req_op_extent_osd_data); 179a4ce40a9SAlex Elder 18049719778SAlex Elder void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, 18149719778SAlex Elder unsigned int which, struct page **pages, 18249719778SAlex Elder u64 length, u32 alignment, 18349719778SAlex Elder bool pages_from_pool, bool own_pages) 18449719778SAlex Elder { 18549719778SAlex Elder struct ceph_osd_data *osd_data; 18649719778SAlex Elder 18749719778SAlex Elder osd_data = osd_req_op_raw_data_in(osd_req, which); 18849719778SAlex Elder ceph_osd_data_pages_init(osd_data, pages, length, alignment, 18949719778SAlex Elder pages_from_pool, own_pages); 19049719778SAlex Elder } 19149719778SAlex Elder EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); 19249719778SAlex Elder 193a4ce40a9SAlex Elder void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, 194406e2c9fSAlex Elder unsigned int which, struct page **pages, 195406e2c9fSAlex Elder u64 length, u32 alignment, 196a4ce40a9SAlex Elder bool pages_from_pool, bool own_pages) 197a4ce40a9SAlex Elder { 198a4ce40a9SAlex Elder struct ceph_osd_data *osd_data; 199a4ce40a9SAlex Elder 200863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 201a4ce40a9SAlex Elder ceph_osd_data_pages_init(osd_data, pages, length, alignment, 202a4ce40a9SAlex Elder pages_from_pool, own_pages); 203a4ce40a9SAlex Elder } 204a4ce40a9SAlex Elder EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); 205a4ce40a9SAlex Elder 206a4ce40a9SAlex Elder void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, 207406e2c9fSAlex Elder unsigned int which, struct ceph_pagelist *pagelist) 208a4ce40a9SAlex Elder { 209a4ce40a9SAlex Elder struct ceph_osd_data *osd_data; 210a4ce40a9SAlex Elder 211863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 212a4ce40a9SAlex Elder ceph_osd_data_pagelist_init(osd_data, pagelist); 213a4ce40a9SAlex Elder } 214a4ce40a9SAlex Elder EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); 215a4ce40a9SAlex Elder 216a4ce40a9SAlex Elder #ifdef CONFIG_BLOCK 217a4ce40a9SAlex Elder void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, 218406e2c9fSAlex Elder unsigned int which, struct bio *bio, size_t bio_length) 219a4ce40a9SAlex Elder { 220a4ce40a9SAlex Elder struct ceph_osd_data *osd_data; 221863c7eb5SAlex Elder 222863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, extent, osd_data); 223a4ce40a9SAlex Elder ceph_osd_data_bio_init(osd_data, bio, bio_length); 224a4ce40a9SAlex Elder } 225a4ce40a9SAlex Elder EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); 226a4ce40a9SAlex Elder #endif /* CONFIG_BLOCK */ 227a4ce40a9SAlex Elder 228a4ce40a9SAlex Elder static void osd_req_op_cls_request_info_pagelist( 229a4ce40a9SAlex Elder struct ceph_osd_request *osd_req, 230a4ce40a9SAlex Elder unsigned int which, struct ceph_pagelist *pagelist) 231a4ce40a9SAlex Elder { 232a4ce40a9SAlex Elder struct ceph_osd_data *osd_data; 233a4ce40a9SAlex Elder 234863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, cls, request_info); 235a4ce40a9SAlex Elder ceph_osd_data_pagelist_init(osd_data, pagelist); 236a4ce40a9SAlex Elder } 237a4ce40a9SAlex Elder 23804017e29SAlex Elder void osd_req_op_cls_request_data_pagelist( 23904017e29SAlex Elder struct ceph_osd_request *osd_req, 24004017e29SAlex Elder unsigned int which, struct ceph_pagelist *pagelist) 24104017e29SAlex Elder { 24204017e29SAlex Elder struct ceph_osd_data *osd_data; 24304017e29SAlex Elder 244863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, cls, request_data); 24504017e29SAlex Elder ceph_osd_data_pagelist_init(osd_data, pagelist); 246bb873b53SIlya Dryomov osd_req->r_ops[which].cls.indata_len += pagelist->length; 247bb873b53SIlya Dryomov osd_req->r_ops[which].indata_len += pagelist->length; 24804017e29SAlex Elder } 24904017e29SAlex Elder EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); 25004017e29SAlex Elder 2516c57b554SAlex Elder void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, 2526c57b554SAlex Elder unsigned int which, struct page **pages, u64 length, 2536c57b554SAlex Elder u32 alignment, bool pages_from_pool, bool own_pages) 2546c57b554SAlex Elder { 2556c57b554SAlex Elder struct ceph_osd_data *osd_data; 2566c57b554SAlex Elder 2576c57b554SAlex Elder osd_data = osd_req_op_data(osd_req, which, cls, request_data); 2586c57b554SAlex Elder ceph_osd_data_pages_init(osd_data, pages, length, alignment, 2596c57b554SAlex Elder pages_from_pool, own_pages); 260bb873b53SIlya Dryomov osd_req->r_ops[which].cls.indata_len += length; 261bb873b53SIlya Dryomov osd_req->r_ops[which].indata_len += length; 2626c57b554SAlex Elder } 2636c57b554SAlex Elder EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); 2646c57b554SAlex Elder 265a4ce40a9SAlex Elder void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, 266a4ce40a9SAlex Elder unsigned int which, struct page **pages, u64 length, 267a4ce40a9SAlex Elder u32 alignment, bool pages_from_pool, bool own_pages) 268a4ce40a9SAlex Elder { 269a4ce40a9SAlex Elder struct ceph_osd_data *osd_data; 270a4ce40a9SAlex Elder 271863c7eb5SAlex Elder osd_data = osd_req_op_data(osd_req, which, cls, response_data); 272a4ce40a9SAlex Elder ceph_osd_data_pages_init(osd_data, pages, length, alignment, 273a4ce40a9SAlex Elder pages_from_pool, own_pages); 274a4ce40a9SAlex Elder } 275a4ce40a9SAlex Elder EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); 276a4ce40a9SAlex Elder 27723c08a9cSAlex Elder static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) 27823c08a9cSAlex Elder { 27923c08a9cSAlex Elder switch (osd_data->type) { 28023c08a9cSAlex Elder case CEPH_OSD_DATA_TYPE_NONE: 28123c08a9cSAlex Elder return 0; 28223c08a9cSAlex Elder case CEPH_OSD_DATA_TYPE_PAGES: 28323c08a9cSAlex Elder return osd_data->length; 28423c08a9cSAlex Elder case CEPH_OSD_DATA_TYPE_PAGELIST: 28523c08a9cSAlex Elder return (u64)osd_data->pagelist->length; 28623c08a9cSAlex Elder #ifdef CONFIG_BLOCK 28723c08a9cSAlex Elder case CEPH_OSD_DATA_TYPE_BIO: 28823c08a9cSAlex Elder return (u64)osd_data->bio_length; 28923c08a9cSAlex Elder #endif /* CONFIG_BLOCK */ 29023c08a9cSAlex Elder default: 29123c08a9cSAlex Elder WARN(true, "unrecognized data type %d\n", (int)osd_data->type); 29223c08a9cSAlex Elder return 0; 29323c08a9cSAlex Elder } 29423c08a9cSAlex Elder } 29523c08a9cSAlex Elder 296c54d47bfSAlex Elder static void ceph_osd_data_release(struct ceph_osd_data *osd_data) 297c54d47bfSAlex Elder { 2985476492fSAlex Elder if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { 299c54d47bfSAlex Elder int num_pages; 300c54d47bfSAlex Elder 301c54d47bfSAlex Elder num_pages = calc_pages_for((u64)osd_data->alignment, 302c54d47bfSAlex Elder (u64)osd_data->length); 303c54d47bfSAlex Elder ceph_release_page_vector(osd_data->pages, num_pages); 304c54d47bfSAlex Elder } 3055476492fSAlex Elder ceph_osd_data_init(osd_data); 3065476492fSAlex Elder } 3075476492fSAlex Elder 3085476492fSAlex Elder static void osd_req_op_data_release(struct ceph_osd_request *osd_req, 3095476492fSAlex Elder unsigned int which) 3105476492fSAlex Elder { 3115476492fSAlex Elder struct ceph_osd_req_op *op; 3125476492fSAlex Elder 3135476492fSAlex Elder BUG_ON(which >= osd_req->r_num_ops); 3145476492fSAlex Elder op = &osd_req->r_ops[which]; 3155476492fSAlex Elder 3165476492fSAlex Elder switch (op->op) { 3175476492fSAlex Elder case CEPH_OSD_OP_READ: 3185476492fSAlex Elder case CEPH_OSD_OP_WRITE: 319e30b7577SIlya Dryomov case CEPH_OSD_OP_WRITEFULL: 3205476492fSAlex Elder ceph_osd_data_release(&op->extent.osd_data); 3215476492fSAlex Elder break; 3225476492fSAlex Elder case CEPH_OSD_OP_CALL: 3235476492fSAlex Elder ceph_osd_data_release(&op->cls.request_info); 32404017e29SAlex Elder ceph_osd_data_release(&op->cls.request_data); 3255476492fSAlex Elder ceph_osd_data_release(&op->cls.response_data); 3265476492fSAlex Elder break; 327d74b50beSYan, Zheng case CEPH_OSD_OP_SETXATTR: 328d74b50beSYan, Zheng case CEPH_OSD_OP_CMPXATTR: 329d74b50beSYan, Zheng ceph_osd_data_release(&op->xattr.osd_data); 330d74b50beSYan, Zheng break; 33166ba609fSYan, Zheng case CEPH_OSD_OP_STAT: 33266ba609fSYan, Zheng ceph_osd_data_release(&op->raw_data_in); 33366ba609fSYan, Zheng break; 334922dab61SIlya Dryomov case CEPH_OSD_OP_NOTIFY_ACK: 335922dab61SIlya Dryomov ceph_osd_data_release(&op->notify_ack.request_data); 336922dab61SIlya Dryomov break; 33719079203SIlya Dryomov case CEPH_OSD_OP_NOTIFY: 33819079203SIlya Dryomov ceph_osd_data_release(&op->notify.request_data); 33919079203SIlya Dryomov ceph_osd_data_release(&op->notify.response_data); 34019079203SIlya Dryomov break; 341a4ed38d7SDouglas Fuller case CEPH_OSD_OP_LIST_WATCHERS: 342a4ed38d7SDouglas Fuller ceph_osd_data_release(&op->list_watchers.response_data); 343a4ed38d7SDouglas Fuller break; 3445476492fSAlex Elder default: 3455476492fSAlex Elder break; 3465476492fSAlex Elder } 347c54d47bfSAlex Elder } 348c54d47bfSAlex Elder 3493d14c5d2SYehuda Sadeh /* 35063244fa1SIlya Dryomov * Assumes @t is zero-initialized. 35163244fa1SIlya Dryomov */ 35263244fa1SIlya Dryomov static void target_init(struct ceph_osd_request_target *t) 35363244fa1SIlya Dryomov { 35463244fa1SIlya Dryomov ceph_oid_init(&t->base_oid); 35563244fa1SIlya Dryomov ceph_oloc_init(&t->base_oloc); 35663244fa1SIlya Dryomov ceph_oid_init(&t->target_oid); 35763244fa1SIlya Dryomov ceph_oloc_init(&t->target_oloc); 35863244fa1SIlya Dryomov 35963244fa1SIlya Dryomov ceph_osds_init(&t->acting); 36063244fa1SIlya Dryomov ceph_osds_init(&t->up); 36163244fa1SIlya Dryomov t->size = -1; 36263244fa1SIlya Dryomov t->min_size = -1; 36363244fa1SIlya Dryomov 36463244fa1SIlya Dryomov t->osd = CEPH_HOMELESS_OSD; 36563244fa1SIlya Dryomov } 36663244fa1SIlya Dryomov 367922dab61SIlya Dryomov static void target_copy(struct ceph_osd_request_target *dest, 368922dab61SIlya Dryomov const struct ceph_osd_request_target *src) 369922dab61SIlya Dryomov { 370922dab61SIlya Dryomov ceph_oid_copy(&dest->base_oid, &src->base_oid); 371922dab61SIlya Dryomov ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); 372922dab61SIlya Dryomov ceph_oid_copy(&dest->target_oid, &src->target_oid); 373922dab61SIlya Dryomov ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); 374922dab61SIlya Dryomov 375922dab61SIlya Dryomov dest->pgid = src->pgid; /* struct */ 376922dab61SIlya Dryomov dest->pg_num = src->pg_num; 377922dab61SIlya Dryomov dest->pg_num_mask = src->pg_num_mask; 378922dab61SIlya Dryomov ceph_osds_copy(&dest->acting, &src->acting); 379922dab61SIlya Dryomov ceph_osds_copy(&dest->up, &src->up); 380922dab61SIlya Dryomov dest->size = src->size; 381922dab61SIlya Dryomov dest->min_size = src->min_size; 382922dab61SIlya Dryomov dest->sort_bitwise = src->sort_bitwise; 383922dab61SIlya Dryomov 384922dab61SIlya Dryomov dest->flags = src->flags; 385922dab61SIlya Dryomov dest->paused = src->paused; 386922dab61SIlya Dryomov 387922dab61SIlya Dryomov dest->osd = src->osd; 388922dab61SIlya Dryomov } 389922dab61SIlya Dryomov 39063244fa1SIlya Dryomov static void target_destroy(struct ceph_osd_request_target *t) 39163244fa1SIlya Dryomov { 39263244fa1SIlya Dryomov ceph_oid_destroy(&t->base_oid); 39330c156d9SYan, Zheng ceph_oloc_destroy(&t->base_oloc); 39463244fa1SIlya Dryomov ceph_oid_destroy(&t->target_oid); 39530c156d9SYan, Zheng ceph_oloc_destroy(&t->target_oloc); 39663244fa1SIlya Dryomov } 39763244fa1SIlya Dryomov 39863244fa1SIlya Dryomov /* 3993d14c5d2SYehuda Sadeh * requests 4003d14c5d2SYehuda Sadeh */ 4013540bfdbSIlya Dryomov static void request_release_checks(struct ceph_osd_request *req) 4023540bfdbSIlya Dryomov { 4033540bfdbSIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&req->r_node)); 4044609245eSIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); 4053540bfdbSIlya Dryomov WARN_ON(!list_empty(&req->r_unsafe_item)); 4063540bfdbSIlya Dryomov WARN_ON(req->r_osd); 4073540bfdbSIlya Dryomov } 4083540bfdbSIlya Dryomov 4099e94af20SIlya Dryomov static void ceph_osdc_release_request(struct kref *kref) 4103d14c5d2SYehuda Sadeh { 4119e94af20SIlya Dryomov struct ceph_osd_request *req = container_of(kref, 4129e94af20SIlya Dryomov struct ceph_osd_request, r_kref); 4135476492fSAlex Elder unsigned int which; 4143d14c5d2SYehuda Sadeh 4159e94af20SIlya Dryomov dout("%s %p (r_request %p r_reply %p)\n", __func__, req, 4169e94af20SIlya Dryomov req->r_request, req->r_reply); 4173540bfdbSIlya Dryomov request_release_checks(req); 4189e94af20SIlya Dryomov 4193d14c5d2SYehuda Sadeh if (req->r_request) 4203d14c5d2SYehuda Sadeh ceph_msg_put(req->r_request); 4215aea3dcdSIlya Dryomov if (req->r_reply) 422ab8cb34aSAlex Elder ceph_msg_put(req->r_reply); 4230fff87ecSAlex Elder 4245476492fSAlex Elder for (which = 0; which < req->r_num_ops; which++) 4255476492fSAlex Elder osd_req_op_data_release(req, which); 4260fff87ecSAlex Elder 427a66dd383SIlya Dryomov target_destroy(&req->r_t); 4283d14c5d2SYehuda Sadeh ceph_put_snap_context(req->r_snapc); 429d30291b9SIlya Dryomov 4303d14c5d2SYehuda Sadeh if (req->r_mempool) 4313d14c5d2SYehuda Sadeh mempool_free(req, req->r_osdc->req_mempool); 4323f1af42aSIlya Dryomov else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) 4335522ae0bSAlex Elder kmem_cache_free(ceph_osd_request_cache, req); 4343f1af42aSIlya Dryomov else 4353f1af42aSIlya Dryomov kfree(req); 4363d14c5d2SYehuda Sadeh } 4379e94af20SIlya Dryomov 4389e94af20SIlya Dryomov void ceph_osdc_get_request(struct ceph_osd_request *req) 4399e94af20SIlya Dryomov { 4409e94af20SIlya Dryomov dout("%s %p (was %d)\n", __func__, req, 4419e94af20SIlya Dryomov atomic_read(&req->r_kref.refcount)); 4429e94af20SIlya Dryomov kref_get(&req->r_kref); 4439e94af20SIlya Dryomov } 4449e94af20SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_get_request); 4459e94af20SIlya Dryomov 4469e94af20SIlya Dryomov void ceph_osdc_put_request(struct ceph_osd_request *req) 4479e94af20SIlya Dryomov { 4483ed97d63SIlya Dryomov if (req) { 4499e94af20SIlya Dryomov dout("%s %p (was %d)\n", __func__, req, 4509e94af20SIlya Dryomov atomic_read(&req->r_kref.refcount)); 4519e94af20SIlya Dryomov kref_put(&req->r_kref, ceph_osdc_release_request); 4529e94af20SIlya Dryomov } 4533ed97d63SIlya Dryomov } 4549e94af20SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_put_request); 4553d14c5d2SYehuda Sadeh 4563540bfdbSIlya Dryomov static void request_init(struct ceph_osd_request *req) 4573540bfdbSIlya Dryomov { 4583540bfdbSIlya Dryomov /* req only, each op is zeroed in _osd_req_op_init() */ 4593540bfdbSIlya Dryomov memset(req, 0, sizeof(*req)); 4603540bfdbSIlya Dryomov 4613540bfdbSIlya Dryomov kref_init(&req->r_kref); 4623540bfdbSIlya Dryomov init_completion(&req->r_completion); 4633540bfdbSIlya Dryomov init_completion(&req->r_safe_completion); 4643540bfdbSIlya Dryomov RB_CLEAR_NODE(&req->r_node); 4654609245eSIlya Dryomov RB_CLEAR_NODE(&req->r_mc_node); 4663540bfdbSIlya Dryomov INIT_LIST_HEAD(&req->r_unsafe_item); 4673540bfdbSIlya Dryomov 4683540bfdbSIlya Dryomov target_init(&req->r_t); 4693540bfdbSIlya Dryomov } 4703540bfdbSIlya Dryomov 471922dab61SIlya Dryomov /* 472922dab61SIlya Dryomov * This is ugly, but it allows us to reuse linger registration and ping 473922dab61SIlya Dryomov * requests, keeping the structure of the code around send_linger{_ping}() 474922dab61SIlya Dryomov * reasonable. Setting up a min_nr=2 mempool for each linger request 475922dab61SIlya Dryomov * and dealing with copying ops (this blasts req only, watch op remains 476922dab61SIlya Dryomov * intact) isn't any better. 477922dab61SIlya Dryomov */ 478922dab61SIlya Dryomov static void request_reinit(struct ceph_osd_request *req) 479922dab61SIlya Dryomov { 480922dab61SIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 481922dab61SIlya Dryomov bool mempool = req->r_mempool; 482922dab61SIlya Dryomov unsigned int num_ops = req->r_num_ops; 483922dab61SIlya Dryomov u64 snapid = req->r_snapid; 484922dab61SIlya Dryomov struct ceph_snap_context *snapc = req->r_snapc; 485922dab61SIlya Dryomov bool linger = req->r_linger; 486922dab61SIlya Dryomov struct ceph_msg *request_msg = req->r_request; 487922dab61SIlya Dryomov struct ceph_msg *reply_msg = req->r_reply; 488922dab61SIlya Dryomov 489922dab61SIlya Dryomov dout("%s req %p\n", __func__, req); 490922dab61SIlya Dryomov WARN_ON(atomic_read(&req->r_kref.refcount) != 1); 491922dab61SIlya Dryomov request_release_checks(req); 492922dab61SIlya Dryomov 493922dab61SIlya Dryomov WARN_ON(atomic_read(&request_msg->kref.refcount) != 1); 494922dab61SIlya Dryomov WARN_ON(atomic_read(&reply_msg->kref.refcount) != 1); 495922dab61SIlya Dryomov target_destroy(&req->r_t); 496922dab61SIlya Dryomov 497922dab61SIlya Dryomov request_init(req); 498922dab61SIlya Dryomov req->r_osdc = osdc; 499922dab61SIlya Dryomov req->r_mempool = mempool; 500922dab61SIlya Dryomov req->r_num_ops = num_ops; 501922dab61SIlya Dryomov req->r_snapid = snapid; 502922dab61SIlya Dryomov req->r_snapc = snapc; 503922dab61SIlya Dryomov req->r_linger = linger; 504922dab61SIlya Dryomov req->r_request = request_msg; 505922dab61SIlya Dryomov req->r_reply = reply_msg; 506922dab61SIlya Dryomov } 507922dab61SIlya Dryomov 5083d14c5d2SYehuda Sadeh struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, 5093d14c5d2SYehuda Sadeh struct ceph_snap_context *snapc, 5101b83bef2SSage Weil unsigned int num_ops, 5113d14c5d2SYehuda Sadeh bool use_mempool, 51254a54007SAlex Elder gfp_t gfp_flags) 5133d14c5d2SYehuda Sadeh { 5143d14c5d2SYehuda Sadeh struct ceph_osd_request *req; 5153d14c5d2SYehuda Sadeh 5163d14c5d2SYehuda Sadeh if (use_mempool) { 5173f1af42aSIlya Dryomov BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); 5183d14c5d2SYehuda Sadeh req = mempool_alloc(osdc->req_mempool, gfp_flags); 5193f1af42aSIlya Dryomov } else if (num_ops <= CEPH_OSD_SLAB_OPS) { 5203f1af42aSIlya Dryomov req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); 5213d14c5d2SYehuda Sadeh } else { 5223f1af42aSIlya Dryomov BUG_ON(num_ops > CEPH_OSD_MAX_OPS); 5233f1af42aSIlya Dryomov req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), 5243f1af42aSIlya Dryomov gfp_flags); 5253d14c5d2SYehuda Sadeh } 5263f1af42aSIlya Dryomov if (unlikely(!req)) 5273d14c5d2SYehuda Sadeh return NULL; 5283d14c5d2SYehuda Sadeh 5293540bfdbSIlya Dryomov request_init(req); 5303d14c5d2SYehuda Sadeh req->r_osdc = osdc; 5313d14c5d2SYehuda Sadeh req->r_mempool = use_mempool; 53279528734SAlex Elder req->r_num_ops = num_ops; 53384127282SIlya Dryomov req->r_snapid = CEPH_NOSNAP; 53484127282SIlya Dryomov req->r_snapc = ceph_get_snap_context(snapc); 5353d14c5d2SYehuda Sadeh 53613d1ad16SIlya Dryomov dout("%s req %p\n", __func__, req); 53713d1ad16SIlya Dryomov return req; 5383f1af42aSIlya Dryomov } 53913d1ad16SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_alloc_request); 5403f1af42aSIlya Dryomov 54130c156d9SYan, Zheng static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc) 54230c156d9SYan, Zheng { 54330c156d9SYan, Zheng return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); 54430c156d9SYan, Zheng } 54530c156d9SYan, Zheng 54613d1ad16SIlya Dryomov int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) 54713d1ad16SIlya Dryomov { 54813d1ad16SIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 54913d1ad16SIlya Dryomov struct ceph_msg *msg; 55013d1ad16SIlya Dryomov int msg_size; 5513d14c5d2SYehuda Sadeh 552d30291b9SIlya Dryomov WARN_ON(ceph_oid_empty(&req->r_base_oid)); 55330c156d9SYan, Zheng WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); 554d30291b9SIlya Dryomov 55513d1ad16SIlya Dryomov /* create request message */ 556ae458f5aSIlya Dryomov msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */ 557ae458f5aSIlya Dryomov msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */ 55830c156d9SYan, Zheng msg_size += CEPH_ENCODING_START_BLK_LEN + 55930c156d9SYan, Zheng ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ 560ae458f5aSIlya Dryomov msg_size += 1 + 8 + 4 + 4; /* pgid */ 56113d1ad16SIlya Dryomov msg_size += 4 + req->r_base_oid.name_len; /* oid */ 56213d1ad16SIlya Dryomov msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); 563ae458f5aSIlya Dryomov msg_size += 8; /* snapid */ 564ae458f5aSIlya Dryomov msg_size += 8; /* snap_seq */ 56513d1ad16SIlya Dryomov msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); 566ae458f5aSIlya Dryomov msg_size += 4; /* retry_attempt */ 567ae458f5aSIlya Dryomov 56813d1ad16SIlya Dryomov if (req->r_mempool) 5693d14c5d2SYehuda Sadeh msg = ceph_msgpool_get(&osdc->msgpool_op, 0); 5703d14c5d2SYehuda Sadeh else 57113d1ad16SIlya Dryomov msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); 57213d1ad16SIlya Dryomov if (!msg) 57313d1ad16SIlya Dryomov return -ENOMEM; 5743d14c5d2SYehuda Sadeh 5753d14c5d2SYehuda Sadeh memset(msg->front.iov_base, 0, msg->front.iov_len); 5763d14c5d2SYehuda Sadeh req->r_request = msg; 5773d14c5d2SYehuda Sadeh 57813d1ad16SIlya Dryomov /* create reply message */ 57913d1ad16SIlya Dryomov msg_size = OSD_OPREPLY_FRONT_LEN; 580711da55dSIlya Dryomov msg_size += req->r_base_oid.name_len; 581711da55dSIlya Dryomov msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); 58213d1ad16SIlya Dryomov 58313d1ad16SIlya Dryomov if (req->r_mempool) 58413d1ad16SIlya Dryomov msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); 58513d1ad16SIlya Dryomov else 58613d1ad16SIlya Dryomov msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); 58713d1ad16SIlya Dryomov if (!msg) 58813d1ad16SIlya Dryomov return -ENOMEM; 58913d1ad16SIlya Dryomov 59013d1ad16SIlya Dryomov req->r_reply = msg; 59113d1ad16SIlya Dryomov 59213d1ad16SIlya Dryomov return 0; 59313d1ad16SIlya Dryomov } 59413d1ad16SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_alloc_messages); 5953d14c5d2SYehuda Sadeh 596a8dd0a37SAlex Elder static bool osd_req_opcode_valid(u16 opcode) 597a8dd0a37SAlex Elder { 598a8dd0a37SAlex Elder switch (opcode) { 59970b5bfa3SIlya Dryomov #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; 60070b5bfa3SIlya Dryomov __CEPH_FORALL_OSD_OPS(GENERATE_CASE) 60170b5bfa3SIlya Dryomov #undef GENERATE_CASE 602a8dd0a37SAlex Elder default: 603a8dd0a37SAlex Elder return false; 604a8dd0a37SAlex Elder } 605a8dd0a37SAlex Elder } 606a8dd0a37SAlex Elder 60733803f33SAlex Elder /* 60833803f33SAlex Elder * This is an osd op init function for opcodes that have no data or 60933803f33SAlex Elder * other information associated with them. It also serves as a 61033803f33SAlex Elder * common init routine for all the other init functions, below. 61133803f33SAlex Elder */ 612c99d2d4aSAlex Elder static struct ceph_osd_req_op * 61349719778SAlex Elder _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, 614144cba14SYan, Zheng u16 opcode, u32 flags) 61533803f33SAlex Elder { 616c99d2d4aSAlex Elder struct ceph_osd_req_op *op; 617c99d2d4aSAlex Elder 618c99d2d4aSAlex Elder BUG_ON(which >= osd_req->r_num_ops); 61933803f33SAlex Elder BUG_ON(!osd_req_opcode_valid(opcode)); 62033803f33SAlex Elder 621c99d2d4aSAlex Elder op = &osd_req->r_ops[which]; 62233803f33SAlex Elder memset(op, 0, sizeof (*op)); 62333803f33SAlex Elder op->op = opcode; 624144cba14SYan, Zheng op->flags = flags; 625c99d2d4aSAlex Elder 626c99d2d4aSAlex Elder return op; 62733803f33SAlex Elder } 62833803f33SAlex Elder 62949719778SAlex Elder void osd_req_op_init(struct ceph_osd_request *osd_req, 630144cba14SYan, Zheng unsigned int which, u16 opcode, u32 flags) 63149719778SAlex Elder { 632144cba14SYan, Zheng (void)_osd_req_op_init(osd_req, which, opcode, flags); 63349719778SAlex Elder } 63449719778SAlex Elder EXPORT_SYMBOL(osd_req_op_init); 63549719778SAlex Elder 636c99d2d4aSAlex Elder void osd_req_op_extent_init(struct ceph_osd_request *osd_req, 637c99d2d4aSAlex Elder unsigned int which, u16 opcode, 63833803f33SAlex Elder u64 offset, u64 length, 63933803f33SAlex Elder u64 truncate_size, u32 truncate_seq) 64033803f33SAlex Elder { 641144cba14SYan, Zheng struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 642144cba14SYan, Zheng opcode, 0); 64333803f33SAlex Elder size_t payload_len = 0; 64433803f33SAlex Elder 645ad7a60deSLi Wang BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 646e30b7577SIlya Dryomov opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && 647e30b7577SIlya Dryomov opcode != CEPH_OSD_OP_TRUNCATE); 64833803f33SAlex Elder 64933803f33SAlex Elder op->extent.offset = offset; 65033803f33SAlex Elder op->extent.length = length; 65133803f33SAlex Elder op->extent.truncate_size = truncate_size; 65233803f33SAlex Elder op->extent.truncate_seq = truncate_seq; 653e30b7577SIlya Dryomov if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) 65433803f33SAlex Elder payload_len += length; 65533803f33SAlex Elder 656de2aa102SIlya Dryomov op->indata_len = payload_len; 65733803f33SAlex Elder } 65833803f33SAlex Elder EXPORT_SYMBOL(osd_req_op_extent_init); 65933803f33SAlex Elder 660c99d2d4aSAlex Elder void osd_req_op_extent_update(struct ceph_osd_request *osd_req, 661c99d2d4aSAlex Elder unsigned int which, u64 length) 662e5975c7cSAlex Elder { 663c99d2d4aSAlex Elder struct ceph_osd_req_op *op; 664c99d2d4aSAlex Elder u64 previous; 665c99d2d4aSAlex Elder 666c99d2d4aSAlex Elder BUG_ON(which >= osd_req->r_num_ops); 667c99d2d4aSAlex Elder op = &osd_req->r_ops[which]; 668c99d2d4aSAlex Elder previous = op->extent.length; 669e5975c7cSAlex Elder 670e5975c7cSAlex Elder if (length == previous) 671e5975c7cSAlex Elder return; /* Nothing to do */ 672e5975c7cSAlex Elder BUG_ON(length > previous); 673e5975c7cSAlex Elder 674e5975c7cSAlex Elder op->extent.length = length; 675de2aa102SIlya Dryomov op->indata_len -= previous - length; 676e5975c7cSAlex Elder } 677e5975c7cSAlex Elder EXPORT_SYMBOL(osd_req_op_extent_update); 678e5975c7cSAlex Elder 6792c63f49aSYan, Zheng void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, 6802c63f49aSYan, Zheng unsigned int which, u64 offset_inc) 6812c63f49aSYan, Zheng { 6822c63f49aSYan, Zheng struct ceph_osd_req_op *op, *prev_op; 6832c63f49aSYan, Zheng 6842c63f49aSYan, Zheng BUG_ON(which + 1 >= osd_req->r_num_ops); 6852c63f49aSYan, Zheng 6862c63f49aSYan, Zheng prev_op = &osd_req->r_ops[which]; 6872c63f49aSYan, Zheng op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); 6882c63f49aSYan, Zheng /* dup previous one */ 6892c63f49aSYan, Zheng op->indata_len = prev_op->indata_len; 6902c63f49aSYan, Zheng op->outdata_len = prev_op->outdata_len; 6912c63f49aSYan, Zheng op->extent = prev_op->extent; 6922c63f49aSYan, Zheng /* adjust offset */ 6932c63f49aSYan, Zheng op->extent.offset += offset_inc; 6942c63f49aSYan, Zheng op->extent.length -= offset_inc; 6952c63f49aSYan, Zheng 6962c63f49aSYan, Zheng if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) 6972c63f49aSYan, Zheng op->indata_len -= offset_inc; 6982c63f49aSYan, Zheng } 6992c63f49aSYan, Zheng EXPORT_SYMBOL(osd_req_op_extent_dup_last); 7002c63f49aSYan, Zheng 701c99d2d4aSAlex Elder void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, 70204017e29SAlex Elder u16 opcode, const char *class, const char *method) 70333803f33SAlex Elder { 704144cba14SYan, Zheng struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 705144cba14SYan, Zheng opcode, 0); 7065f562df5SAlex Elder struct ceph_pagelist *pagelist; 70733803f33SAlex Elder size_t payload_len = 0; 70833803f33SAlex Elder size_t size; 70933803f33SAlex Elder 71033803f33SAlex Elder BUG_ON(opcode != CEPH_OSD_OP_CALL); 71133803f33SAlex Elder 7125f562df5SAlex Elder pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); 7135f562df5SAlex Elder BUG_ON(!pagelist); 7145f562df5SAlex Elder ceph_pagelist_init(pagelist); 7155f562df5SAlex Elder 71633803f33SAlex Elder op->cls.class_name = class; 71733803f33SAlex Elder size = strlen(class); 71833803f33SAlex Elder BUG_ON(size > (size_t) U8_MAX); 71933803f33SAlex Elder op->cls.class_len = size; 7205f562df5SAlex Elder ceph_pagelist_append(pagelist, class, size); 72133803f33SAlex Elder payload_len += size; 72233803f33SAlex Elder 72333803f33SAlex Elder op->cls.method_name = method; 72433803f33SAlex Elder size = strlen(method); 72533803f33SAlex Elder BUG_ON(size > (size_t) U8_MAX); 72633803f33SAlex Elder op->cls.method_len = size; 7275f562df5SAlex Elder ceph_pagelist_append(pagelist, method, size); 72833803f33SAlex Elder payload_len += size; 72933803f33SAlex Elder 730a4ce40a9SAlex Elder osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); 7315f562df5SAlex Elder 732de2aa102SIlya Dryomov op->indata_len = payload_len; 73333803f33SAlex Elder } 73433803f33SAlex Elder EXPORT_SYMBOL(osd_req_op_cls_init); 7358c042b0dSAlex Elder 736d74b50beSYan, Zheng int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, 737d74b50beSYan, Zheng u16 opcode, const char *name, const void *value, 738d74b50beSYan, Zheng size_t size, u8 cmp_op, u8 cmp_mode) 739d74b50beSYan, Zheng { 740144cba14SYan, Zheng struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 741144cba14SYan, Zheng opcode, 0); 742d74b50beSYan, Zheng struct ceph_pagelist *pagelist; 743d74b50beSYan, Zheng size_t payload_len; 744d74b50beSYan, Zheng 745d74b50beSYan, Zheng BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); 746d74b50beSYan, Zheng 747d74b50beSYan, Zheng pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); 748d74b50beSYan, Zheng if (!pagelist) 749d74b50beSYan, Zheng return -ENOMEM; 750d74b50beSYan, Zheng 751d74b50beSYan, Zheng ceph_pagelist_init(pagelist); 752d74b50beSYan, Zheng 753d74b50beSYan, Zheng payload_len = strlen(name); 754d74b50beSYan, Zheng op->xattr.name_len = payload_len; 755d74b50beSYan, Zheng ceph_pagelist_append(pagelist, name, payload_len); 756d74b50beSYan, Zheng 757d74b50beSYan, Zheng op->xattr.value_len = size; 758d74b50beSYan, Zheng ceph_pagelist_append(pagelist, value, size); 759d74b50beSYan, Zheng payload_len += size; 760d74b50beSYan, Zheng 761d74b50beSYan, Zheng op->xattr.cmp_op = cmp_op; 762d74b50beSYan, Zheng op->xattr.cmp_mode = cmp_mode; 763d74b50beSYan, Zheng 764d74b50beSYan, Zheng ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); 765de2aa102SIlya Dryomov op->indata_len = payload_len; 766d74b50beSYan, Zheng return 0; 767d74b50beSYan, Zheng } 768d74b50beSYan, Zheng EXPORT_SYMBOL(osd_req_op_xattr_init); 769d74b50beSYan, Zheng 770922dab61SIlya Dryomov /* 771922dab61SIlya Dryomov * @watch_opcode: CEPH_OSD_WATCH_OP_* 772922dab61SIlya Dryomov */ 773922dab61SIlya Dryomov static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, 774922dab61SIlya Dryomov u64 cookie, u8 watch_opcode) 77533803f33SAlex Elder { 776922dab61SIlya Dryomov struct ceph_osd_req_op *op; 77733803f33SAlex Elder 778922dab61SIlya Dryomov op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); 77933803f33SAlex Elder op->watch.cookie = cookie; 780922dab61SIlya Dryomov op->watch.op = watch_opcode; 781922dab61SIlya Dryomov op->watch.gen = 0; 78233803f33SAlex Elder } 78333803f33SAlex Elder 784c647b8a8SIlya Dryomov void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, 785c647b8a8SIlya Dryomov unsigned int which, 786c647b8a8SIlya Dryomov u64 expected_object_size, 787c647b8a8SIlya Dryomov u64 expected_write_size) 788c647b8a8SIlya Dryomov { 789c647b8a8SIlya Dryomov struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, 790144cba14SYan, Zheng CEPH_OSD_OP_SETALLOCHINT, 791144cba14SYan, Zheng 0); 792c647b8a8SIlya Dryomov 793c647b8a8SIlya Dryomov op->alloc_hint.expected_object_size = expected_object_size; 794c647b8a8SIlya Dryomov op->alloc_hint.expected_write_size = expected_write_size; 795c647b8a8SIlya Dryomov 796c647b8a8SIlya Dryomov /* 797c647b8a8SIlya Dryomov * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed 798c647b8a8SIlya Dryomov * not worth a feature bit. Set FAILOK per-op flag to make 799c647b8a8SIlya Dryomov * sure older osds don't trip over an unsupported opcode. 800c647b8a8SIlya Dryomov */ 801c647b8a8SIlya Dryomov op->flags |= CEPH_OSD_OP_FLAG_FAILOK; 802c647b8a8SIlya Dryomov } 803c647b8a8SIlya Dryomov EXPORT_SYMBOL(osd_req_op_alloc_hint_init); 804c647b8a8SIlya Dryomov 80590af3602SAlex Elder static void ceph_osdc_msg_data_add(struct ceph_msg *msg, 806ec9123c5SAlex Elder struct ceph_osd_data *osd_data) 807ec9123c5SAlex Elder { 808ec9123c5SAlex Elder u64 length = ceph_osd_data_length(osd_data); 809ec9123c5SAlex Elder 810ec9123c5SAlex Elder if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { 811ec9123c5SAlex Elder BUG_ON(length > (u64) SIZE_MAX); 812ec9123c5SAlex Elder if (length) 81390af3602SAlex Elder ceph_msg_data_add_pages(msg, osd_data->pages, 814ec9123c5SAlex Elder length, osd_data->alignment); 815ec9123c5SAlex Elder } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { 816ec9123c5SAlex Elder BUG_ON(!length); 81790af3602SAlex Elder ceph_msg_data_add_pagelist(msg, osd_data->pagelist); 818ec9123c5SAlex Elder #ifdef CONFIG_BLOCK 819ec9123c5SAlex Elder } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { 82090af3602SAlex Elder ceph_msg_data_add_bio(msg, osd_data->bio, length); 821ec9123c5SAlex Elder #endif 822ec9123c5SAlex Elder } else { 823ec9123c5SAlex Elder BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); 824ec9123c5SAlex Elder } 825ec9123c5SAlex Elder } 826ec9123c5SAlex Elder 827bb873b53SIlya Dryomov static u32 osd_req_encode_op(struct ceph_osd_op *dst, 828bb873b53SIlya Dryomov const struct ceph_osd_req_op *src) 8293d14c5d2SYehuda Sadeh { 830a8dd0a37SAlex Elder if (WARN_ON(!osd_req_opcode_valid(src->op))) { 831a8dd0a37SAlex Elder pr_err("unrecognized osd opcode %d\n", src->op); 832a8dd0a37SAlex Elder 833a8dd0a37SAlex Elder return 0; 834a8dd0a37SAlex Elder } 8353d14c5d2SYehuda Sadeh 836065a68f9SAlex Elder switch (src->op) { 837fbfab539SAlex Elder case CEPH_OSD_OP_STAT: 838fbfab539SAlex Elder break; 8393d14c5d2SYehuda Sadeh case CEPH_OSD_OP_READ: 8403d14c5d2SYehuda Sadeh case CEPH_OSD_OP_WRITE: 841e30b7577SIlya Dryomov case CEPH_OSD_OP_WRITEFULL: 842ad7a60deSLi Wang case CEPH_OSD_OP_ZERO: 843ad7a60deSLi Wang case CEPH_OSD_OP_TRUNCATE: 844175face2SAlex Elder dst->extent.offset = cpu_to_le64(src->extent.offset); 845175face2SAlex Elder dst->extent.length = cpu_to_le64(src->extent.length); 8463d14c5d2SYehuda Sadeh dst->extent.truncate_size = 8473d14c5d2SYehuda Sadeh cpu_to_le64(src->extent.truncate_size); 8483d14c5d2SYehuda Sadeh dst->extent.truncate_seq = 8493d14c5d2SYehuda Sadeh cpu_to_le32(src->extent.truncate_seq); 8503d14c5d2SYehuda Sadeh break; 8513d14c5d2SYehuda Sadeh case CEPH_OSD_OP_CALL: 8523d14c5d2SYehuda Sadeh dst->cls.class_len = src->cls.class_len; 8533d14c5d2SYehuda Sadeh dst->cls.method_len = src->cls.method_len; 854bb873b53SIlya Dryomov dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); 8553d14c5d2SYehuda Sadeh break; 8563d14c5d2SYehuda Sadeh case CEPH_OSD_OP_STARTSYNC: 8573d14c5d2SYehuda Sadeh break; 858a40c4f10SYehuda Sadeh case CEPH_OSD_OP_WATCH: 859a40c4f10SYehuda Sadeh dst->watch.cookie = cpu_to_le64(src->watch.cookie); 860922dab61SIlya Dryomov dst->watch.ver = cpu_to_le64(0); 861922dab61SIlya Dryomov dst->watch.op = src->watch.op; 862922dab61SIlya Dryomov dst->watch.gen = cpu_to_le32(src->watch.gen); 863922dab61SIlya Dryomov break; 864922dab61SIlya Dryomov case CEPH_OSD_OP_NOTIFY_ACK: 865a40c4f10SYehuda Sadeh break; 86619079203SIlya Dryomov case CEPH_OSD_OP_NOTIFY: 86719079203SIlya Dryomov dst->notify.cookie = cpu_to_le64(src->notify.cookie); 86819079203SIlya Dryomov break; 869a4ed38d7SDouglas Fuller case CEPH_OSD_OP_LIST_WATCHERS: 870a4ed38d7SDouglas Fuller break; 871c647b8a8SIlya Dryomov case CEPH_OSD_OP_SETALLOCHINT: 872c647b8a8SIlya Dryomov dst->alloc_hint.expected_object_size = 873c647b8a8SIlya Dryomov cpu_to_le64(src->alloc_hint.expected_object_size); 874c647b8a8SIlya Dryomov dst->alloc_hint.expected_write_size = 875c647b8a8SIlya Dryomov cpu_to_le64(src->alloc_hint.expected_write_size); 876c647b8a8SIlya Dryomov break; 877d74b50beSYan, Zheng case CEPH_OSD_OP_SETXATTR: 878d74b50beSYan, Zheng case CEPH_OSD_OP_CMPXATTR: 879d74b50beSYan, Zheng dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); 880d74b50beSYan, Zheng dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); 881d74b50beSYan, Zheng dst->xattr.cmp_op = src->xattr.cmp_op; 882d74b50beSYan, Zheng dst->xattr.cmp_mode = src->xattr.cmp_mode; 883d74b50beSYan, Zheng break; 884864e9197SYan, Zheng case CEPH_OSD_OP_CREATE: 885864e9197SYan, Zheng case CEPH_OSD_OP_DELETE: 886864e9197SYan, Zheng break; 8873d14c5d2SYehuda Sadeh default: 8884c46459cSAlex Elder pr_err("unsupported osd opcode %s\n", 8898f63ca2dSAlex Elder ceph_osd_op_name(src->op)); 8904c46459cSAlex Elder WARN_ON(1); 891a8dd0a37SAlex Elder 892a8dd0a37SAlex Elder return 0; 8933d14c5d2SYehuda Sadeh } 8947b25bf5fSIlya Dryomov 895a8dd0a37SAlex Elder dst->op = cpu_to_le16(src->op); 8967b25bf5fSIlya Dryomov dst->flags = cpu_to_le32(src->flags); 897de2aa102SIlya Dryomov dst->payload_len = cpu_to_le32(src->indata_len); 898175face2SAlex Elder 899bb873b53SIlya Dryomov return src->indata_len; 9003d14c5d2SYehuda Sadeh } 9013d14c5d2SYehuda Sadeh 9023d14c5d2SYehuda Sadeh /* 9033d14c5d2SYehuda Sadeh * build new request AND message, calculate layout, and adjust file 9043d14c5d2SYehuda Sadeh * extent as needed. 9053d14c5d2SYehuda Sadeh * 9063d14c5d2SYehuda Sadeh * if the file was recently truncated, we include information about its 9073d14c5d2SYehuda Sadeh * old and new size so that the object can be updated appropriately. (we 9083d14c5d2SYehuda Sadeh * avoid synchronously deleting truncated objects because it's slow.) 9093d14c5d2SYehuda Sadeh * 9103d14c5d2SYehuda Sadeh * if @do_sync, include a 'startsync' command so that the osd will flush 9113d14c5d2SYehuda Sadeh * data quickly. 9123d14c5d2SYehuda Sadeh */ 9133d14c5d2SYehuda Sadeh struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, 9143d14c5d2SYehuda Sadeh struct ceph_file_layout *layout, 9153d14c5d2SYehuda Sadeh struct ceph_vino vino, 916715e4cd4SYan, Zheng u64 off, u64 *plen, 917715e4cd4SYan, Zheng unsigned int which, int num_ops, 9183d14c5d2SYehuda Sadeh int opcode, int flags, 9193d14c5d2SYehuda Sadeh struct ceph_snap_context *snapc, 9203d14c5d2SYehuda Sadeh u32 truncate_seq, 9213d14c5d2SYehuda Sadeh u64 truncate_size, 922153e5167SAlex Elder bool use_mempool) 9233d14c5d2SYehuda Sadeh { 9243d14c5d2SYehuda Sadeh struct ceph_osd_request *req; 92575d1c941SAlex Elder u64 objnum = 0; 92675d1c941SAlex Elder u64 objoff = 0; 92775d1c941SAlex Elder u64 objlen = 0; 9286816282dSSage Weil int r; 9293d14c5d2SYehuda Sadeh 930ad7a60deSLi Wang BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && 931864e9197SYan, Zheng opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && 932864e9197SYan, Zheng opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); 9333d14c5d2SYehuda Sadeh 934acead002SAlex Elder req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, 935ae7ca4a3SAlex Elder GFP_NOFS); 93613d1ad16SIlya Dryomov if (!req) { 93713d1ad16SIlya Dryomov r = -ENOMEM; 93813d1ad16SIlya Dryomov goto fail; 93913d1ad16SIlya Dryomov } 94079528734SAlex Elder 9413d14c5d2SYehuda Sadeh /* calculate max write size */ 942a19dadfbSAlex Elder r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); 94313d1ad16SIlya Dryomov if (r) 94413d1ad16SIlya Dryomov goto fail; 945a19dadfbSAlex Elder 946864e9197SYan, Zheng if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { 947144cba14SYan, Zheng osd_req_op_init(req, which, opcode, 0); 948864e9197SYan, Zheng } else { 9497627151eSYan, Zheng u32 object_size = layout->object_size; 950864e9197SYan, Zheng u32 object_base = off - objoff; 951ccca4e37SYan, Zheng if (!(truncate_seq == 1 && truncate_size == -1ULL)) { 952d18d1e28SAlex Elder if (truncate_size <= object_base) { 953d18d1e28SAlex Elder truncate_size = 0; 954d18d1e28SAlex Elder } else { 955d18d1e28SAlex Elder truncate_size -= object_base; 956d18d1e28SAlex Elder if (truncate_size > object_size) 957d18d1e28SAlex Elder truncate_size = object_size; 958d18d1e28SAlex Elder } 959ccca4e37SYan, Zheng } 960715e4cd4SYan, Zheng osd_req_op_extent_init(req, which, opcode, objoff, objlen, 961b0270324SAlex Elder truncate_size, truncate_seq); 962864e9197SYan, Zheng } 963d18d1e28SAlex Elder 964bb873b53SIlya Dryomov req->r_flags = flags; 9657627151eSYan, Zheng req->r_base_oloc.pool = layout->pool_id; 96630c156d9SYan, Zheng req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); 967d30291b9SIlya Dryomov ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); 968dbe0fc41SAlex Elder 969bb873b53SIlya Dryomov req->r_snapid = vino.snap; 970bb873b53SIlya Dryomov if (flags & CEPH_OSD_FLAG_WRITE) 971bb873b53SIlya Dryomov req->r_data_offset = off; 972bb873b53SIlya Dryomov 97313d1ad16SIlya Dryomov r = ceph_osdc_alloc_messages(req, GFP_NOFS); 97413d1ad16SIlya Dryomov if (r) 97513d1ad16SIlya Dryomov goto fail; 97613d1ad16SIlya Dryomov 9773d14c5d2SYehuda Sadeh return req; 97813d1ad16SIlya Dryomov 97913d1ad16SIlya Dryomov fail: 98013d1ad16SIlya Dryomov ceph_osdc_put_request(req); 98113d1ad16SIlya Dryomov return ERR_PTR(r); 9823d14c5d2SYehuda Sadeh } 9833d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_new_request); 9843d14c5d2SYehuda Sadeh 9853d14c5d2SYehuda Sadeh /* 9863d14c5d2SYehuda Sadeh * We keep osd requests in an rbtree, sorted by ->r_tid. 9873d14c5d2SYehuda Sadeh */ 988fcd00b68SIlya Dryomov DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) 9894609245eSIlya Dryomov DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) 9903d14c5d2SYehuda Sadeh 9910247a0cfSIlya Dryomov static bool osd_homeless(struct ceph_osd *osd) 9920247a0cfSIlya Dryomov { 9930247a0cfSIlya Dryomov return osd->o_osd == CEPH_HOMELESS_OSD; 9940247a0cfSIlya Dryomov } 9950247a0cfSIlya Dryomov 9965aea3dcdSIlya Dryomov static bool osd_registered(struct ceph_osd *osd) 9973d14c5d2SYehuda Sadeh { 9985aea3dcdSIlya Dryomov verify_osdc_locked(osd->o_osdc); 9993d14c5d2SYehuda Sadeh 10005aea3dcdSIlya Dryomov return !RB_EMPTY_NODE(&osd->o_node); 10013d14c5d2SYehuda Sadeh } 10023d14c5d2SYehuda Sadeh 10033d14c5d2SYehuda Sadeh /* 10040247a0cfSIlya Dryomov * Assumes @osd is zero-initialized. 10050247a0cfSIlya Dryomov */ 10060247a0cfSIlya Dryomov static void osd_init(struct ceph_osd *osd) 10070247a0cfSIlya Dryomov { 10080247a0cfSIlya Dryomov atomic_set(&osd->o_ref, 1); 10090247a0cfSIlya Dryomov RB_CLEAR_NODE(&osd->o_node); 10105aea3dcdSIlya Dryomov osd->o_requests = RB_ROOT; 1011922dab61SIlya Dryomov osd->o_linger_requests = RB_ROOT; 10120247a0cfSIlya Dryomov INIT_LIST_HEAD(&osd->o_osd_lru); 10130247a0cfSIlya Dryomov INIT_LIST_HEAD(&osd->o_keepalive_item); 10140247a0cfSIlya Dryomov osd->o_incarnation = 1; 10155aea3dcdSIlya Dryomov mutex_init(&osd->lock); 10160247a0cfSIlya Dryomov } 10170247a0cfSIlya Dryomov 10180247a0cfSIlya Dryomov static void osd_cleanup(struct ceph_osd *osd) 10190247a0cfSIlya Dryomov { 10200247a0cfSIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); 10215aea3dcdSIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 1022922dab61SIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 10230247a0cfSIlya Dryomov WARN_ON(!list_empty(&osd->o_osd_lru)); 10240247a0cfSIlya Dryomov WARN_ON(!list_empty(&osd->o_keepalive_item)); 10250247a0cfSIlya Dryomov 10260247a0cfSIlya Dryomov if (osd->o_auth.authorizer) { 10270247a0cfSIlya Dryomov WARN_ON(osd_homeless(osd)); 10280247a0cfSIlya Dryomov ceph_auth_destroy_authorizer(osd->o_auth.authorizer); 10290247a0cfSIlya Dryomov } 10300247a0cfSIlya Dryomov } 10310247a0cfSIlya Dryomov 10320247a0cfSIlya Dryomov /* 10333d14c5d2SYehuda Sadeh * Track open sessions with osds. 10343d14c5d2SYehuda Sadeh */ 1035e10006f8SAlex Elder static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) 10363d14c5d2SYehuda Sadeh { 10373d14c5d2SYehuda Sadeh struct ceph_osd *osd; 10383d14c5d2SYehuda Sadeh 10390247a0cfSIlya Dryomov WARN_ON(onum == CEPH_HOMELESS_OSD); 10400247a0cfSIlya Dryomov 10417a28f59bSIlya Dryomov osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); 10420247a0cfSIlya Dryomov osd_init(osd); 10433d14c5d2SYehuda Sadeh osd->o_osdc = osdc; 1044e10006f8SAlex Elder osd->o_osd = onum; 10453d14c5d2SYehuda Sadeh 1046b7a9e5ddSSage Weil ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); 10473d14c5d2SYehuda Sadeh 10483d14c5d2SYehuda Sadeh return osd; 10493d14c5d2SYehuda Sadeh } 10503d14c5d2SYehuda Sadeh 10513d14c5d2SYehuda Sadeh static struct ceph_osd *get_osd(struct ceph_osd *osd) 10523d14c5d2SYehuda Sadeh { 10533d14c5d2SYehuda Sadeh if (atomic_inc_not_zero(&osd->o_ref)) { 10543d14c5d2SYehuda Sadeh dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, 10553d14c5d2SYehuda Sadeh atomic_read(&osd->o_ref)); 10563d14c5d2SYehuda Sadeh return osd; 10573d14c5d2SYehuda Sadeh } else { 10583d14c5d2SYehuda Sadeh dout("get_osd %p FAIL\n", osd); 10593d14c5d2SYehuda Sadeh return NULL; 10603d14c5d2SYehuda Sadeh } 10613d14c5d2SYehuda Sadeh } 10623d14c5d2SYehuda Sadeh 10633d14c5d2SYehuda Sadeh static void put_osd(struct ceph_osd *osd) 10643d14c5d2SYehuda Sadeh { 10653d14c5d2SYehuda Sadeh dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), 10663d14c5d2SYehuda Sadeh atomic_read(&osd->o_ref) - 1); 1067b28ec2f3SIlya Dryomov if (atomic_dec_and_test(&osd->o_ref)) { 10680247a0cfSIlya Dryomov osd_cleanup(osd); 10693d14c5d2SYehuda Sadeh kfree(osd); 10703d14c5d2SYehuda Sadeh } 10713d14c5d2SYehuda Sadeh } 10723d14c5d2SYehuda Sadeh 1073fcd00b68SIlya Dryomov DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) 1074fcd00b68SIlya Dryomov 10759dd2845cSIlya Dryomov static void __move_osd_to_lru(struct ceph_osd *osd) 10763d14c5d2SYehuda Sadeh { 10779dd2845cSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 10789dd2845cSIlya Dryomov 10799dd2845cSIlya Dryomov dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 10803d14c5d2SYehuda Sadeh BUG_ON(!list_empty(&osd->o_osd_lru)); 1081bbf37ec3SIlya Dryomov 10829dd2845cSIlya Dryomov spin_lock(&osdc->osd_lru_lock); 10833d14c5d2SYehuda Sadeh list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); 10849dd2845cSIlya Dryomov spin_unlock(&osdc->osd_lru_lock); 10859dd2845cSIlya Dryomov 1086a319bf56SIlya Dryomov osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; 10873d14c5d2SYehuda Sadeh } 10883d14c5d2SYehuda Sadeh 10899dd2845cSIlya Dryomov static void maybe_move_osd_to_lru(struct ceph_osd *osd) 1090bbf37ec3SIlya Dryomov { 10915aea3dcdSIlya Dryomov if (RB_EMPTY_ROOT(&osd->o_requests) && 1092922dab61SIlya Dryomov RB_EMPTY_ROOT(&osd->o_linger_requests)) 10939dd2845cSIlya Dryomov __move_osd_to_lru(osd); 1094bbf37ec3SIlya Dryomov } 1095bbf37ec3SIlya Dryomov 10963d14c5d2SYehuda Sadeh static void __remove_osd_from_lru(struct ceph_osd *osd) 10973d14c5d2SYehuda Sadeh { 10989dd2845cSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 10999dd2845cSIlya Dryomov 11009dd2845cSIlya Dryomov dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 11019dd2845cSIlya Dryomov 11029dd2845cSIlya Dryomov spin_lock(&osdc->osd_lru_lock); 11033d14c5d2SYehuda Sadeh if (!list_empty(&osd->o_osd_lru)) 11043d14c5d2SYehuda Sadeh list_del_init(&osd->o_osd_lru); 11059dd2845cSIlya Dryomov spin_unlock(&osdc->osd_lru_lock); 11063d14c5d2SYehuda Sadeh } 11073d14c5d2SYehuda Sadeh 11083d14c5d2SYehuda Sadeh /* 11095aea3dcdSIlya Dryomov * Close the connection and assign any leftover requests to the 11105aea3dcdSIlya Dryomov * homeless session. 11115aea3dcdSIlya Dryomov */ 11125aea3dcdSIlya Dryomov static void close_osd(struct ceph_osd *osd) 11135aea3dcdSIlya Dryomov { 11145aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 11155aea3dcdSIlya Dryomov struct rb_node *n; 11165aea3dcdSIlya Dryomov 11175aea3dcdSIlya Dryomov verify_osdc_wrlocked(osdc); 11185aea3dcdSIlya Dryomov dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 11195aea3dcdSIlya Dryomov 11205aea3dcdSIlya Dryomov ceph_con_close(&osd->o_con); 11215aea3dcdSIlya Dryomov 11225aea3dcdSIlya Dryomov for (n = rb_first(&osd->o_requests); n; ) { 11235aea3dcdSIlya Dryomov struct ceph_osd_request *req = 11245aea3dcdSIlya Dryomov rb_entry(n, struct ceph_osd_request, r_node); 11255aea3dcdSIlya Dryomov 11265aea3dcdSIlya Dryomov n = rb_next(n); /* unlink_request() */ 11275aea3dcdSIlya Dryomov 11285aea3dcdSIlya Dryomov dout(" reassigning req %p tid %llu\n", req, req->r_tid); 11295aea3dcdSIlya Dryomov unlink_request(osd, req); 11305aea3dcdSIlya Dryomov link_request(&osdc->homeless_osd, req); 11315aea3dcdSIlya Dryomov } 1132922dab61SIlya Dryomov for (n = rb_first(&osd->o_linger_requests); n; ) { 1133922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = 1134922dab61SIlya Dryomov rb_entry(n, struct ceph_osd_linger_request, node); 1135922dab61SIlya Dryomov 1136922dab61SIlya Dryomov n = rb_next(n); /* unlink_linger() */ 1137922dab61SIlya Dryomov 1138922dab61SIlya Dryomov dout(" reassigning lreq %p linger_id %llu\n", lreq, 1139922dab61SIlya Dryomov lreq->linger_id); 1140922dab61SIlya Dryomov unlink_linger(osd, lreq); 1141922dab61SIlya Dryomov link_linger(&osdc->homeless_osd, lreq); 1142922dab61SIlya Dryomov } 11435aea3dcdSIlya Dryomov 11445aea3dcdSIlya Dryomov __remove_osd_from_lru(osd); 11455aea3dcdSIlya Dryomov erase_osd(&osdc->osds, osd); 11465aea3dcdSIlya Dryomov put_osd(osd); 11475aea3dcdSIlya Dryomov } 11485aea3dcdSIlya Dryomov 11495aea3dcdSIlya Dryomov /* 11503d14c5d2SYehuda Sadeh * reset osd connect 11513d14c5d2SYehuda Sadeh */ 11525aea3dcdSIlya Dryomov static int reopen_osd(struct ceph_osd *osd) 11533d14c5d2SYehuda Sadeh { 1154c3acb181SAlex Elder struct ceph_entity_addr *peer_addr; 11553d14c5d2SYehuda Sadeh 11565aea3dcdSIlya Dryomov dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 11575aea3dcdSIlya Dryomov 11585aea3dcdSIlya Dryomov if (RB_EMPTY_ROOT(&osd->o_requests) && 1159922dab61SIlya Dryomov RB_EMPTY_ROOT(&osd->o_linger_requests)) { 11605aea3dcdSIlya Dryomov close_osd(osd); 1161c3acb181SAlex Elder return -ENODEV; 1162c3acb181SAlex Elder } 1163c3acb181SAlex Elder 11645aea3dcdSIlya Dryomov peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; 1165c3acb181SAlex Elder if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && 11663d14c5d2SYehuda Sadeh !ceph_con_opened(&osd->o_con)) { 11675aea3dcdSIlya Dryomov struct rb_node *n; 1168c3acb181SAlex Elder 11693d14c5d2SYehuda Sadeh dout("osd addr hasn't changed and connection never opened, " 11700b4af2e8SIlya Dryomov "letting msgr retry\n"); 11713d14c5d2SYehuda Sadeh /* touch each r_stamp for handle_timeout()'s benfit */ 11725aea3dcdSIlya Dryomov for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { 11735aea3dcdSIlya Dryomov struct ceph_osd_request *req = 11745aea3dcdSIlya Dryomov rb_entry(n, struct ceph_osd_request, r_node); 11753d14c5d2SYehuda Sadeh req->r_stamp = jiffies; 11765aea3dcdSIlya Dryomov } 1177c3acb181SAlex Elder 1178c3acb181SAlex Elder return -EAGAIN; 11793d14c5d2SYehuda Sadeh } 1180c3acb181SAlex Elder 1181c3acb181SAlex Elder ceph_con_close(&osd->o_con); 1182c3acb181SAlex Elder ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); 1183c3acb181SAlex Elder osd->o_incarnation++; 1184c3acb181SAlex Elder 1185c3acb181SAlex Elder return 0; 11863d14c5d2SYehuda Sadeh } 11873d14c5d2SYehuda Sadeh 11885aea3dcdSIlya Dryomov static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, 11895aea3dcdSIlya Dryomov bool wrlocked) 11903d14c5d2SYehuda Sadeh { 11915aea3dcdSIlya Dryomov struct ceph_osd *osd; 11925aea3dcdSIlya Dryomov 11935aea3dcdSIlya Dryomov if (wrlocked) 11945aea3dcdSIlya Dryomov verify_osdc_wrlocked(osdc); 11955aea3dcdSIlya Dryomov else 11965aea3dcdSIlya Dryomov verify_osdc_locked(osdc); 11975aea3dcdSIlya Dryomov 11985aea3dcdSIlya Dryomov if (o != CEPH_HOMELESS_OSD) 11995aea3dcdSIlya Dryomov osd = lookup_osd(&osdc->osds, o); 12005aea3dcdSIlya Dryomov else 12015aea3dcdSIlya Dryomov osd = &osdc->homeless_osd; 12025aea3dcdSIlya Dryomov if (!osd) { 12035aea3dcdSIlya Dryomov if (!wrlocked) 12045aea3dcdSIlya Dryomov return ERR_PTR(-EAGAIN); 12055aea3dcdSIlya Dryomov 12065aea3dcdSIlya Dryomov osd = create_osd(osdc, o); 12075aea3dcdSIlya Dryomov insert_osd(&osdc->osds, osd); 12085aea3dcdSIlya Dryomov ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, 12095aea3dcdSIlya Dryomov &osdc->osdmap->osd_addr[osd->o_osd]); 12105aea3dcdSIlya Dryomov } 12115aea3dcdSIlya Dryomov 12125aea3dcdSIlya Dryomov dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); 12135aea3dcdSIlya Dryomov return osd; 1214a40c4f10SYehuda Sadeh } 1215a40c4f10SYehuda Sadeh 12163d14c5d2SYehuda Sadeh /* 12175aea3dcdSIlya Dryomov * Create request <-> OSD session relation. 12185aea3dcdSIlya Dryomov * 12195aea3dcdSIlya Dryomov * @req has to be assigned a tid, @osd may be homeless. 12203d14c5d2SYehuda Sadeh */ 12215aea3dcdSIlya Dryomov static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) 12223d14c5d2SYehuda Sadeh { 12235aea3dcdSIlya Dryomov verify_osd_locked(osd); 12245aea3dcdSIlya Dryomov WARN_ON(!req->r_tid || req->r_osd); 12255aea3dcdSIlya Dryomov dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 122635f9f8a0SSage Weil req, req->r_tid); 12275aea3dcdSIlya Dryomov 12285aea3dcdSIlya Dryomov if (!osd_homeless(osd)) 12295aea3dcdSIlya Dryomov __remove_osd_from_lru(osd); 12305aea3dcdSIlya Dryomov else 12315aea3dcdSIlya Dryomov atomic_inc(&osd->o_osdc->num_homeless); 12325aea3dcdSIlya Dryomov 12335aea3dcdSIlya Dryomov get_osd(osd); 12345aea3dcdSIlya Dryomov insert_request(&osd->o_requests, req); 12355aea3dcdSIlya Dryomov req->r_osd = osd; 123635f9f8a0SSage Weil } 123735f9f8a0SSage Weil 12385aea3dcdSIlya Dryomov static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) 12393d14c5d2SYehuda Sadeh { 12405aea3dcdSIlya Dryomov verify_osd_locked(osd); 12415aea3dcdSIlya Dryomov WARN_ON(req->r_osd != osd); 12425aea3dcdSIlya Dryomov dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, 12435aea3dcdSIlya Dryomov req, req->r_tid); 12445aea3dcdSIlya Dryomov 12455aea3dcdSIlya Dryomov req->r_osd = NULL; 12465aea3dcdSIlya Dryomov erase_request(&osd->o_requests, req); 12475aea3dcdSIlya Dryomov put_osd(osd); 12485aea3dcdSIlya Dryomov 12495aea3dcdSIlya Dryomov if (!osd_homeless(osd)) 12505aea3dcdSIlya Dryomov maybe_move_osd_to_lru(osd); 12515aea3dcdSIlya Dryomov else 12525aea3dcdSIlya Dryomov atomic_dec(&osd->o_osdc->num_homeless); 12533d14c5d2SYehuda Sadeh } 12543d14c5d2SYehuda Sadeh 125563244fa1SIlya Dryomov static bool __pool_full(struct ceph_pg_pool_info *pi) 125663244fa1SIlya Dryomov { 125763244fa1SIlya Dryomov return pi->flags & CEPH_POOL_FLAG_FULL; 125863244fa1SIlya Dryomov } 125963244fa1SIlya Dryomov 126042c1b124SIlya Dryomov static bool have_pool_full(struct ceph_osd_client *osdc) 126142c1b124SIlya Dryomov { 126242c1b124SIlya Dryomov struct rb_node *n; 126342c1b124SIlya Dryomov 126442c1b124SIlya Dryomov for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 126542c1b124SIlya Dryomov struct ceph_pg_pool_info *pi = 126642c1b124SIlya Dryomov rb_entry(n, struct ceph_pg_pool_info, node); 126742c1b124SIlya Dryomov 126842c1b124SIlya Dryomov if (__pool_full(pi)) 126942c1b124SIlya Dryomov return true; 127042c1b124SIlya Dryomov } 127142c1b124SIlya Dryomov 127242c1b124SIlya Dryomov return false; 127342c1b124SIlya Dryomov } 127442c1b124SIlya Dryomov 12755aea3dcdSIlya Dryomov static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) 12765aea3dcdSIlya Dryomov { 12775aea3dcdSIlya Dryomov struct ceph_pg_pool_info *pi; 12785aea3dcdSIlya Dryomov 12795aea3dcdSIlya Dryomov pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 12805aea3dcdSIlya Dryomov if (!pi) 12815aea3dcdSIlya Dryomov return false; 12825aea3dcdSIlya Dryomov 12835aea3dcdSIlya Dryomov return __pool_full(pi); 12845aea3dcdSIlya Dryomov } 12855aea3dcdSIlya Dryomov 12863d14c5d2SYehuda Sadeh /* 1287d29adb34SJosh Durgin * Returns whether a request should be blocked from being sent 1288d29adb34SJosh Durgin * based on the current osdmap and osd_client settings. 1289d29adb34SJosh Durgin */ 129063244fa1SIlya Dryomov static bool target_should_be_paused(struct ceph_osd_client *osdc, 129163244fa1SIlya Dryomov const struct ceph_osd_request_target *t, 129263244fa1SIlya Dryomov struct ceph_pg_pool_info *pi) 129363244fa1SIlya Dryomov { 1294b7ec35b3SIlya Dryomov bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 1295b7ec35b3SIlya Dryomov bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 1296b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 129763244fa1SIlya Dryomov __pool_full(pi); 129863244fa1SIlya Dryomov 129963244fa1SIlya Dryomov WARN_ON(pi->id != t->base_oloc.pool); 130063244fa1SIlya Dryomov return (t->flags & CEPH_OSD_FLAG_READ && pauserd) || 130163244fa1SIlya Dryomov (t->flags & CEPH_OSD_FLAG_WRITE && pausewr); 130263244fa1SIlya Dryomov } 130363244fa1SIlya Dryomov 130463244fa1SIlya Dryomov enum calc_target_result { 130563244fa1SIlya Dryomov CALC_TARGET_NO_ACTION = 0, 130663244fa1SIlya Dryomov CALC_TARGET_NEED_RESEND, 130763244fa1SIlya Dryomov CALC_TARGET_POOL_DNE, 130863244fa1SIlya Dryomov }; 130963244fa1SIlya Dryomov 131063244fa1SIlya Dryomov static enum calc_target_result calc_target(struct ceph_osd_client *osdc, 131163244fa1SIlya Dryomov struct ceph_osd_request_target *t, 131263244fa1SIlya Dryomov u32 *last_force_resend, 131363244fa1SIlya Dryomov bool any_change) 131463244fa1SIlya Dryomov { 131563244fa1SIlya Dryomov struct ceph_pg_pool_info *pi; 131663244fa1SIlya Dryomov struct ceph_pg pgid, last_pgid; 131763244fa1SIlya Dryomov struct ceph_osds up, acting; 131863244fa1SIlya Dryomov bool force_resend = false; 131963244fa1SIlya Dryomov bool need_check_tiering = false; 132063244fa1SIlya Dryomov bool need_resend = false; 1321b7ec35b3SIlya Dryomov bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); 132263244fa1SIlya Dryomov enum calc_target_result ct_res; 132363244fa1SIlya Dryomov int ret; 132463244fa1SIlya Dryomov 132563244fa1SIlya Dryomov pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); 132663244fa1SIlya Dryomov if (!pi) { 132763244fa1SIlya Dryomov t->osd = CEPH_HOMELESS_OSD; 132863244fa1SIlya Dryomov ct_res = CALC_TARGET_POOL_DNE; 132963244fa1SIlya Dryomov goto out; 133063244fa1SIlya Dryomov } 133163244fa1SIlya Dryomov 133263244fa1SIlya Dryomov if (osdc->osdmap->epoch == pi->last_force_request_resend) { 133363244fa1SIlya Dryomov if (last_force_resend && 133463244fa1SIlya Dryomov *last_force_resend < pi->last_force_request_resend) { 133563244fa1SIlya Dryomov *last_force_resend = pi->last_force_request_resend; 133663244fa1SIlya Dryomov force_resend = true; 133763244fa1SIlya Dryomov } else if (!last_force_resend) { 133863244fa1SIlya Dryomov force_resend = true; 133963244fa1SIlya Dryomov } 134063244fa1SIlya Dryomov } 134163244fa1SIlya Dryomov if (ceph_oid_empty(&t->target_oid) || force_resend) { 134263244fa1SIlya Dryomov ceph_oid_copy(&t->target_oid, &t->base_oid); 134363244fa1SIlya Dryomov need_check_tiering = true; 134463244fa1SIlya Dryomov } 134563244fa1SIlya Dryomov if (ceph_oloc_empty(&t->target_oloc) || force_resend) { 134663244fa1SIlya Dryomov ceph_oloc_copy(&t->target_oloc, &t->base_oloc); 134763244fa1SIlya Dryomov need_check_tiering = true; 134863244fa1SIlya Dryomov } 134963244fa1SIlya Dryomov 135063244fa1SIlya Dryomov if (need_check_tiering && 135163244fa1SIlya Dryomov (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { 135263244fa1SIlya Dryomov if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) 135363244fa1SIlya Dryomov t->target_oloc.pool = pi->read_tier; 135463244fa1SIlya Dryomov if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) 135563244fa1SIlya Dryomov t->target_oloc.pool = pi->write_tier; 135663244fa1SIlya Dryomov } 135763244fa1SIlya Dryomov 135863244fa1SIlya Dryomov ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid, 135963244fa1SIlya Dryomov &t->target_oloc, &pgid); 136063244fa1SIlya Dryomov if (ret) { 136163244fa1SIlya Dryomov WARN_ON(ret != -ENOENT); 136263244fa1SIlya Dryomov t->osd = CEPH_HOMELESS_OSD; 136363244fa1SIlya Dryomov ct_res = CALC_TARGET_POOL_DNE; 136463244fa1SIlya Dryomov goto out; 136563244fa1SIlya Dryomov } 136663244fa1SIlya Dryomov last_pgid.pool = pgid.pool; 136763244fa1SIlya Dryomov last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); 136863244fa1SIlya Dryomov 136963244fa1SIlya Dryomov ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting); 137063244fa1SIlya Dryomov if (any_change && 137163244fa1SIlya Dryomov ceph_is_new_interval(&t->acting, 137263244fa1SIlya Dryomov &acting, 137363244fa1SIlya Dryomov &t->up, 137463244fa1SIlya Dryomov &up, 137563244fa1SIlya Dryomov t->size, 137663244fa1SIlya Dryomov pi->size, 137763244fa1SIlya Dryomov t->min_size, 137863244fa1SIlya Dryomov pi->min_size, 137963244fa1SIlya Dryomov t->pg_num, 138063244fa1SIlya Dryomov pi->pg_num, 138163244fa1SIlya Dryomov t->sort_bitwise, 138263244fa1SIlya Dryomov sort_bitwise, 138363244fa1SIlya Dryomov &last_pgid)) 138463244fa1SIlya Dryomov force_resend = true; 138563244fa1SIlya Dryomov 138663244fa1SIlya Dryomov if (t->paused && !target_should_be_paused(osdc, t, pi)) { 138763244fa1SIlya Dryomov t->paused = false; 138863244fa1SIlya Dryomov need_resend = true; 138963244fa1SIlya Dryomov } 139063244fa1SIlya Dryomov 139163244fa1SIlya Dryomov if (ceph_pg_compare(&t->pgid, &pgid) || 139263244fa1SIlya Dryomov ceph_osds_changed(&t->acting, &acting, any_change) || 139363244fa1SIlya Dryomov force_resend) { 139463244fa1SIlya Dryomov t->pgid = pgid; /* struct */ 139563244fa1SIlya Dryomov ceph_osds_copy(&t->acting, &acting); 139663244fa1SIlya Dryomov ceph_osds_copy(&t->up, &up); 139763244fa1SIlya Dryomov t->size = pi->size; 139863244fa1SIlya Dryomov t->min_size = pi->min_size; 139963244fa1SIlya Dryomov t->pg_num = pi->pg_num; 140063244fa1SIlya Dryomov t->pg_num_mask = pi->pg_num_mask; 140163244fa1SIlya Dryomov t->sort_bitwise = sort_bitwise; 140263244fa1SIlya Dryomov 140363244fa1SIlya Dryomov t->osd = acting.primary; 140463244fa1SIlya Dryomov need_resend = true; 140563244fa1SIlya Dryomov } 140663244fa1SIlya Dryomov 140763244fa1SIlya Dryomov ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION; 140863244fa1SIlya Dryomov out: 140963244fa1SIlya Dryomov dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); 141063244fa1SIlya Dryomov return ct_res; 141163244fa1SIlya Dryomov } 141263244fa1SIlya Dryomov 1413bb873b53SIlya Dryomov static void setup_request_data(struct ceph_osd_request *req, 1414bb873b53SIlya Dryomov struct ceph_msg *msg) 14153d14c5d2SYehuda Sadeh { 1416bb873b53SIlya Dryomov u32 data_len = 0; 1417bb873b53SIlya Dryomov int i; 14183d14c5d2SYehuda Sadeh 1419bb873b53SIlya Dryomov if (!list_empty(&msg->data)) 1420bb873b53SIlya Dryomov return; 14213d14c5d2SYehuda Sadeh 1422bb873b53SIlya Dryomov WARN_ON(msg->data_length); 1423bb873b53SIlya Dryomov for (i = 0; i < req->r_num_ops; i++) { 1424bb873b53SIlya Dryomov struct ceph_osd_req_op *op = &req->r_ops[i]; 1425bb873b53SIlya Dryomov 1426bb873b53SIlya Dryomov switch (op->op) { 1427bb873b53SIlya Dryomov /* request */ 1428bb873b53SIlya Dryomov case CEPH_OSD_OP_WRITE: 1429bb873b53SIlya Dryomov case CEPH_OSD_OP_WRITEFULL: 1430bb873b53SIlya Dryomov WARN_ON(op->indata_len != op->extent.length); 1431bb873b53SIlya Dryomov ceph_osdc_msg_data_add(msg, &op->extent.osd_data); 1432bb873b53SIlya Dryomov break; 1433bb873b53SIlya Dryomov case CEPH_OSD_OP_SETXATTR: 1434bb873b53SIlya Dryomov case CEPH_OSD_OP_CMPXATTR: 1435bb873b53SIlya Dryomov WARN_ON(op->indata_len != op->xattr.name_len + 1436bb873b53SIlya Dryomov op->xattr.value_len); 1437bb873b53SIlya Dryomov ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); 1438bb873b53SIlya Dryomov break; 1439922dab61SIlya Dryomov case CEPH_OSD_OP_NOTIFY_ACK: 1440922dab61SIlya Dryomov ceph_osdc_msg_data_add(msg, 1441922dab61SIlya Dryomov &op->notify_ack.request_data); 1442922dab61SIlya Dryomov break; 1443bb873b53SIlya Dryomov 1444bb873b53SIlya Dryomov /* reply */ 1445bb873b53SIlya Dryomov case CEPH_OSD_OP_STAT: 1446bb873b53SIlya Dryomov ceph_osdc_msg_data_add(req->r_reply, 1447bb873b53SIlya Dryomov &op->raw_data_in); 1448bb873b53SIlya Dryomov break; 1449bb873b53SIlya Dryomov case CEPH_OSD_OP_READ: 1450bb873b53SIlya Dryomov ceph_osdc_msg_data_add(req->r_reply, 1451bb873b53SIlya Dryomov &op->extent.osd_data); 1452bb873b53SIlya Dryomov break; 1453a4ed38d7SDouglas Fuller case CEPH_OSD_OP_LIST_WATCHERS: 1454a4ed38d7SDouglas Fuller ceph_osdc_msg_data_add(req->r_reply, 1455a4ed38d7SDouglas Fuller &op->list_watchers.response_data); 1456a4ed38d7SDouglas Fuller break; 1457bb873b53SIlya Dryomov 1458bb873b53SIlya Dryomov /* both */ 1459bb873b53SIlya Dryomov case CEPH_OSD_OP_CALL: 1460bb873b53SIlya Dryomov WARN_ON(op->indata_len != op->cls.class_len + 1461bb873b53SIlya Dryomov op->cls.method_len + 1462bb873b53SIlya Dryomov op->cls.indata_len); 1463bb873b53SIlya Dryomov ceph_osdc_msg_data_add(msg, &op->cls.request_info); 1464bb873b53SIlya Dryomov /* optional, can be NONE */ 1465bb873b53SIlya Dryomov ceph_osdc_msg_data_add(msg, &op->cls.request_data); 1466bb873b53SIlya Dryomov /* optional, can be NONE */ 1467bb873b53SIlya Dryomov ceph_osdc_msg_data_add(req->r_reply, 1468bb873b53SIlya Dryomov &op->cls.response_data); 1469bb873b53SIlya Dryomov break; 147019079203SIlya Dryomov case CEPH_OSD_OP_NOTIFY: 147119079203SIlya Dryomov ceph_osdc_msg_data_add(msg, 147219079203SIlya Dryomov &op->notify.request_data); 147319079203SIlya Dryomov ceph_osdc_msg_data_add(req->r_reply, 147419079203SIlya Dryomov &op->notify.response_data); 147519079203SIlya Dryomov break; 1476bb873b53SIlya Dryomov } 1477bb873b53SIlya Dryomov 1478bb873b53SIlya Dryomov data_len += op->indata_len; 1479bb873b53SIlya Dryomov } 1480bb873b53SIlya Dryomov 1481bb873b53SIlya Dryomov WARN_ON(data_len != msg->data_length); 1482bb873b53SIlya Dryomov } 1483bb873b53SIlya Dryomov 1484bb873b53SIlya Dryomov static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg) 1485bb873b53SIlya Dryomov { 1486bb873b53SIlya Dryomov void *p = msg->front.iov_base; 1487bb873b53SIlya Dryomov void *const end = p + msg->front_alloc_len; 1488bb873b53SIlya Dryomov u32 data_len = 0; 1489bb873b53SIlya Dryomov int i; 1490bb873b53SIlya Dryomov 1491bb873b53SIlya Dryomov if (req->r_flags & CEPH_OSD_FLAG_WRITE) { 1492bb873b53SIlya Dryomov /* snapshots aren't writeable */ 1493bb873b53SIlya Dryomov WARN_ON(req->r_snapid != CEPH_NOSNAP); 1494bb873b53SIlya Dryomov } else { 1495bb873b53SIlya Dryomov WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || 1496bb873b53SIlya Dryomov req->r_data_offset || req->r_snapc); 1497bb873b53SIlya Dryomov } 1498bb873b53SIlya Dryomov 1499bb873b53SIlya Dryomov setup_request_data(req, msg); 1500bb873b53SIlya Dryomov 1501bb873b53SIlya Dryomov ceph_encode_32(&p, 1); /* client_inc, always 1 */ 1502bb873b53SIlya Dryomov ceph_encode_32(&p, req->r_osdc->osdmap->epoch); 1503bb873b53SIlya Dryomov ceph_encode_32(&p, req->r_flags); 1504bb873b53SIlya Dryomov ceph_encode_timespec(p, &req->r_mtime); 1505bb873b53SIlya Dryomov p += sizeof(struct ceph_timespec); 1506bb873b53SIlya Dryomov /* aka reassert_version */ 1507bb873b53SIlya Dryomov memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version)); 1508bb873b53SIlya Dryomov p += sizeof(req->r_replay_version); 1509bb873b53SIlya Dryomov 1510bb873b53SIlya Dryomov /* oloc */ 151130c156d9SYan, Zheng ceph_start_encoding(&p, 5, 4, 151230c156d9SYan, Zheng ceph_oloc_encoding_size(&req->r_t.target_oloc)); 1513bb873b53SIlya Dryomov ceph_encode_64(&p, req->r_t.target_oloc.pool); 1514bb873b53SIlya Dryomov ceph_encode_32(&p, -1); /* preferred */ 1515bb873b53SIlya Dryomov ceph_encode_32(&p, 0); /* key len */ 151630c156d9SYan, Zheng if (req->r_t.target_oloc.pool_ns) 151730c156d9SYan, Zheng ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str, 151830c156d9SYan, Zheng req->r_t.target_oloc.pool_ns->len); 151930c156d9SYan, Zheng else 152030c156d9SYan, Zheng ceph_encode_32(&p, 0); 1521bb873b53SIlya Dryomov 1522bb873b53SIlya Dryomov /* pgid */ 1523bb873b53SIlya Dryomov ceph_encode_8(&p, 1); 1524a66dd383SIlya Dryomov ceph_encode_64(&p, req->r_t.pgid.pool); 1525a66dd383SIlya Dryomov ceph_encode_32(&p, req->r_t.pgid.seed); 1526bb873b53SIlya Dryomov ceph_encode_32(&p, -1); /* preferred */ 15272169aea6SSage Weil 1528bb873b53SIlya Dryomov /* oid */ 1529bb873b53SIlya Dryomov ceph_encode_32(&p, req->r_t.target_oid.name_len); 1530bb873b53SIlya Dryomov memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len); 1531bb873b53SIlya Dryomov p += req->r_t.target_oid.name_len; 1532bb873b53SIlya Dryomov 1533bb873b53SIlya Dryomov /* ops, can imply data */ 1534bb873b53SIlya Dryomov ceph_encode_16(&p, req->r_num_ops); 1535bb873b53SIlya Dryomov for (i = 0; i < req->r_num_ops; i++) { 1536bb873b53SIlya Dryomov data_len += osd_req_encode_op(p, &req->r_ops[i]); 1537bb873b53SIlya Dryomov p += sizeof(struct ceph_osd_op); 1538bb873b53SIlya Dryomov } 1539bb873b53SIlya Dryomov 1540bb873b53SIlya Dryomov ceph_encode_64(&p, req->r_snapid); /* snapid */ 1541bb873b53SIlya Dryomov if (req->r_snapc) { 1542bb873b53SIlya Dryomov ceph_encode_64(&p, req->r_snapc->seq); 1543bb873b53SIlya Dryomov ceph_encode_32(&p, req->r_snapc->num_snaps); 1544bb873b53SIlya Dryomov for (i = 0; i < req->r_snapc->num_snaps; i++) 1545bb873b53SIlya Dryomov ceph_encode_64(&p, req->r_snapc->snaps[i]); 1546bb873b53SIlya Dryomov } else { 1547bb873b53SIlya Dryomov ceph_encode_64(&p, 0); /* snap_seq */ 1548bb873b53SIlya Dryomov ceph_encode_32(&p, 0); /* snaps len */ 1549bb873b53SIlya Dryomov } 1550bb873b53SIlya Dryomov 1551bb873b53SIlya Dryomov ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ 1552bb873b53SIlya Dryomov 1553bb873b53SIlya Dryomov BUG_ON(p > end); 1554bb873b53SIlya Dryomov msg->front.iov_len = p - msg->front.iov_base; 1555bb873b53SIlya Dryomov msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ 1556bb873b53SIlya Dryomov msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 1557bb873b53SIlya Dryomov msg->hdr.data_len = cpu_to_le32(data_len); 1558bb873b53SIlya Dryomov /* 1559bb873b53SIlya Dryomov * The header "data_off" is a hint to the receiver allowing it 1560bb873b53SIlya Dryomov * to align received data into its buffers such that there's no 1561bb873b53SIlya Dryomov * need to re-copy it before writing it to disk (direct I/O). 1562bb873b53SIlya Dryomov */ 1563bb873b53SIlya Dryomov msg->hdr.data_off = cpu_to_le16(req->r_data_offset); 1564bb873b53SIlya Dryomov 15654a3262b1SIlya Dryomov dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__, 15664a3262b1SIlya Dryomov req, req->r_t.target_oid.name, req->r_t.target_oid.name_len, 15674a3262b1SIlya Dryomov msg->front.iov_len, data_len); 1568bb873b53SIlya Dryomov } 1569bb873b53SIlya Dryomov 1570bb873b53SIlya Dryomov /* 1571bb873b53SIlya Dryomov * @req has to be assigned a tid and registered. 1572bb873b53SIlya Dryomov */ 1573bb873b53SIlya Dryomov static void send_request(struct ceph_osd_request *req) 1574bb873b53SIlya Dryomov { 1575bb873b53SIlya Dryomov struct ceph_osd *osd = req->r_osd; 1576bb873b53SIlya Dryomov 15775aea3dcdSIlya Dryomov verify_osd_locked(osd); 1578bb873b53SIlya Dryomov WARN_ON(osd->o_osd != req->r_t.osd); 1579bb873b53SIlya Dryomov 15805aea3dcdSIlya Dryomov /* 15815aea3dcdSIlya Dryomov * We may have a previously queued request message hanging 15825aea3dcdSIlya Dryomov * around. Cancel it to avoid corrupting the msgr. 15835aea3dcdSIlya Dryomov */ 15845aea3dcdSIlya Dryomov if (req->r_sent) 15855aea3dcdSIlya Dryomov ceph_msg_revoke(req->r_request); 15865aea3dcdSIlya Dryomov 1587bb873b53SIlya Dryomov req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; 1588bb873b53SIlya Dryomov if (req->r_attempts) 1589bb873b53SIlya Dryomov req->r_flags |= CEPH_OSD_FLAG_RETRY; 1590bb873b53SIlya Dryomov else 1591bb873b53SIlya Dryomov WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); 1592bb873b53SIlya Dryomov 1593bb873b53SIlya Dryomov encode_request(req, req->r_request); 1594bb873b53SIlya Dryomov 1595bb873b53SIlya Dryomov dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n", 1596bb873b53SIlya Dryomov __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, 1597bb873b53SIlya Dryomov req->r_t.osd, req->r_flags, req->r_attempts); 1598bb873b53SIlya Dryomov 1599bb873b53SIlya Dryomov req->r_t.paused = false; 16003d14c5d2SYehuda Sadeh req->r_stamp = jiffies; 1601bb873b53SIlya Dryomov req->r_attempts++; 16023d14c5d2SYehuda Sadeh 1603bb873b53SIlya Dryomov req->r_sent = osd->o_incarnation; 1604bb873b53SIlya Dryomov req->r_request->hdr.tid = cpu_to_le64(req->r_tid); 1605bb873b53SIlya Dryomov ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); 16063d14c5d2SYehuda Sadeh } 16073d14c5d2SYehuda Sadeh 160842c1b124SIlya Dryomov static void maybe_request_map(struct ceph_osd_client *osdc) 160942c1b124SIlya Dryomov { 161042c1b124SIlya Dryomov bool continuous = false; 161142c1b124SIlya Dryomov 16125aea3dcdSIlya Dryomov verify_osdc_locked(osdc); 161342c1b124SIlya Dryomov WARN_ON(!osdc->osdmap->epoch); 161442c1b124SIlya Dryomov 1615b7ec35b3SIlya Dryomov if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 1616b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || 1617b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 161842c1b124SIlya Dryomov dout("%s osdc %p continuous\n", __func__, osdc); 161942c1b124SIlya Dryomov continuous = true; 162042c1b124SIlya Dryomov } else { 162142c1b124SIlya Dryomov dout("%s osdc %p onetime\n", __func__, osdc); 162242c1b124SIlya Dryomov } 162342c1b124SIlya Dryomov 162442c1b124SIlya Dryomov if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 162542c1b124SIlya Dryomov osdc->osdmap->epoch + 1, continuous)) 162642c1b124SIlya Dryomov ceph_monc_renew_subs(&osdc->client->monc); 162742c1b124SIlya Dryomov } 162842c1b124SIlya Dryomov 16294609245eSIlya Dryomov static void send_map_check(struct ceph_osd_request *req); 16304609245eSIlya Dryomov 16315aea3dcdSIlya Dryomov static void __submit_request(struct ceph_osd_request *req, bool wrlocked) 16320bbfdfe8SIlya Dryomov { 16335aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 16345aea3dcdSIlya Dryomov struct ceph_osd *osd; 16354609245eSIlya Dryomov enum calc_target_result ct_res; 16365aea3dcdSIlya Dryomov bool need_send = false; 16375aea3dcdSIlya Dryomov bool promoted = false; 16380bbfdfe8SIlya Dryomov 16395aea3dcdSIlya Dryomov WARN_ON(req->r_tid || req->r_got_reply); 16405aea3dcdSIlya Dryomov dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); 16415aea3dcdSIlya Dryomov 16425aea3dcdSIlya Dryomov again: 16434609245eSIlya Dryomov ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false); 16444609245eSIlya Dryomov if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) 16454609245eSIlya Dryomov goto promote; 16464609245eSIlya Dryomov 16475aea3dcdSIlya Dryomov osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); 16485aea3dcdSIlya Dryomov if (IS_ERR(osd)) { 16495aea3dcdSIlya Dryomov WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); 16505aea3dcdSIlya Dryomov goto promote; 16515aea3dcdSIlya Dryomov } 16525aea3dcdSIlya Dryomov 16535aea3dcdSIlya Dryomov if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 1654b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { 16555aea3dcdSIlya Dryomov dout("req %p pausewr\n", req); 16565aea3dcdSIlya Dryomov req->r_t.paused = true; 16575aea3dcdSIlya Dryomov maybe_request_map(osdc); 16585aea3dcdSIlya Dryomov } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && 1659b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 16605aea3dcdSIlya Dryomov dout("req %p pauserd\n", req); 16615aea3dcdSIlya Dryomov req->r_t.paused = true; 16625aea3dcdSIlya Dryomov maybe_request_map(osdc); 16635aea3dcdSIlya Dryomov } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && 16645aea3dcdSIlya Dryomov !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | 16655aea3dcdSIlya Dryomov CEPH_OSD_FLAG_FULL_FORCE)) && 1666b7ec35b3SIlya Dryomov (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 16675aea3dcdSIlya Dryomov pool_full(osdc, req->r_t.base_oloc.pool))) { 16685aea3dcdSIlya Dryomov dout("req %p full/pool_full\n", req); 16695aea3dcdSIlya Dryomov pr_warn_ratelimited("FULL or reached pool quota\n"); 16705aea3dcdSIlya Dryomov req->r_t.paused = true; 16715aea3dcdSIlya Dryomov maybe_request_map(osdc); 16725aea3dcdSIlya Dryomov } else if (!osd_homeless(osd)) { 16735aea3dcdSIlya Dryomov need_send = true; 16740bbfdfe8SIlya Dryomov } else { 16755aea3dcdSIlya Dryomov maybe_request_map(osdc); 16760bbfdfe8SIlya Dryomov } 16770bbfdfe8SIlya Dryomov 16785aea3dcdSIlya Dryomov mutex_lock(&osd->lock); 16795aea3dcdSIlya Dryomov /* 16805aea3dcdSIlya Dryomov * Assign the tid atomically with send_request() to protect 16815aea3dcdSIlya Dryomov * multiple writes to the same object from racing with each 16825aea3dcdSIlya Dryomov * other, resulting in out of order ops on the OSDs. 16835aea3dcdSIlya Dryomov */ 16845aea3dcdSIlya Dryomov req->r_tid = atomic64_inc_return(&osdc->last_tid); 16855aea3dcdSIlya Dryomov link_request(osd, req); 16865aea3dcdSIlya Dryomov if (need_send) 16875aea3dcdSIlya Dryomov send_request(req); 16885aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 16895aea3dcdSIlya Dryomov 16904609245eSIlya Dryomov if (ct_res == CALC_TARGET_POOL_DNE) 16914609245eSIlya Dryomov send_map_check(req); 16924609245eSIlya Dryomov 16935aea3dcdSIlya Dryomov if (promoted) 16945aea3dcdSIlya Dryomov downgrade_write(&osdc->lock); 16955aea3dcdSIlya Dryomov return; 16965aea3dcdSIlya Dryomov 16975aea3dcdSIlya Dryomov promote: 16985aea3dcdSIlya Dryomov up_read(&osdc->lock); 16995aea3dcdSIlya Dryomov down_write(&osdc->lock); 17005aea3dcdSIlya Dryomov wrlocked = true; 17015aea3dcdSIlya Dryomov promoted = true; 17025aea3dcdSIlya Dryomov goto again; 17030bbfdfe8SIlya Dryomov } 17040bbfdfe8SIlya Dryomov 17055aea3dcdSIlya Dryomov static void account_request(struct ceph_osd_request *req) 17065aea3dcdSIlya Dryomov { 17075aea3dcdSIlya Dryomov unsigned int mask = CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK; 17085aea3dcdSIlya Dryomov 17095aea3dcdSIlya Dryomov if (req->r_flags & CEPH_OSD_FLAG_READ) { 17105aea3dcdSIlya Dryomov WARN_ON(req->r_flags & mask); 17115aea3dcdSIlya Dryomov req->r_flags |= CEPH_OSD_FLAG_ACK; 17125aea3dcdSIlya Dryomov } else if (req->r_flags & CEPH_OSD_FLAG_WRITE) 17135aea3dcdSIlya Dryomov WARN_ON(!(req->r_flags & mask)); 17145aea3dcdSIlya Dryomov else 17155aea3dcdSIlya Dryomov WARN_ON(1); 17165aea3dcdSIlya Dryomov 17175aea3dcdSIlya Dryomov WARN_ON(req->r_unsafe_callback && (req->r_flags & mask) != mask); 17185aea3dcdSIlya Dryomov atomic_inc(&req->r_osdc->num_requests); 17195aea3dcdSIlya Dryomov } 17205aea3dcdSIlya Dryomov 17215aea3dcdSIlya Dryomov static void submit_request(struct ceph_osd_request *req, bool wrlocked) 17225aea3dcdSIlya Dryomov { 17235aea3dcdSIlya Dryomov ceph_osdc_get_request(req); 17245aea3dcdSIlya Dryomov account_request(req); 17255aea3dcdSIlya Dryomov __submit_request(req, wrlocked); 17265aea3dcdSIlya Dryomov } 17275aea3dcdSIlya Dryomov 17285aea3dcdSIlya Dryomov static void __finish_request(struct ceph_osd_request *req) 17295aea3dcdSIlya Dryomov { 17305aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 17315aea3dcdSIlya Dryomov struct ceph_osd *osd = req->r_osd; 17325aea3dcdSIlya Dryomov 17335aea3dcdSIlya Dryomov verify_osd_locked(osd); 17345aea3dcdSIlya Dryomov dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 17355aea3dcdSIlya Dryomov 17364609245eSIlya Dryomov WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); 17375aea3dcdSIlya Dryomov unlink_request(osd, req); 17385aea3dcdSIlya Dryomov atomic_dec(&osdc->num_requests); 17395aea3dcdSIlya Dryomov 17405aea3dcdSIlya Dryomov /* 17415aea3dcdSIlya Dryomov * If an OSD has failed or returned and a request has been sent 17425aea3dcdSIlya Dryomov * twice, it's possible to get a reply and end up here while the 17435aea3dcdSIlya Dryomov * request message is queued for delivery. We will ignore the 17445aea3dcdSIlya Dryomov * reply, so not a big deal, but better to try and catch it. 17455aea3dcdSIlya Dryomov */ 17465aea3dcdSIlya Dryomov ceph_msg_revoke(req->r_request); 17475aea3dcdSIlya Dryomov ceph_msg_revoke_incoming(req->r_reply); 17485aea3dcdSIlya Dryomov } 17495aea3dcdSIlya Dryomov 17505aea3dcdSIlya Dryomov static void finish_request(struct ceph_osd_request *req) 17515aea3dcdSIlya Dryomov { 17525aea3dcdSIlya Dryomov __finish_request(req); 17535aea3dcdSIlya Dryomov ceph_osdc_put_request(req); 17540bbfdfe8SIlya Dryomov } 17550bbfdfe8SIlya Dryomov 1756fe5da05eSIlya Dryomov static void __complete_request(struct ceph_osd_request *req) 1757fe5da05eSIlya Dryomov { 1758fe5da05eSIlya Dryomov if (req->r_callback) 1759fe5da05eSIlya Dryomov req->r_callback(req); 1760fe5da05eSIlya Dryomov else 1761fe5da05eSIlya Dryomov complete_all(&req->r_completion); 1762fe5da05eSIlya Dryomov } 1763fe5da05eSIlya Dryomov 17644609245eSIlya Dryomov /* 17654609245eSIlya Dryomov * Note that this is open-coded in handle_reply(), which has to deal 17664609245eSIlya Dryomov * with ack vs commit, dup acks, etc. 17674609245eSIlya Dryomov */ 17684609245eSIlya Dryomov static void complete_request(struct ceph_osd_request *req, int err) 17694609245eSIlya Dryomov { 17704609245eSIlya Dryomov dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); 17714609245eSIlya Dryomov 17724609245eSIlya Dryomov req->r_result = err; 17734609245eSIlya Dryomov __finish_request(req); 17744609245eSIlya Dryomov __complete_request(req); 17754609245eSIlya Dryomov complete_all(&req->r_safe_completion); 17764609245eSIlya Dryomov ceph_osdc_put_request(req); 17774609245eSIlya Dryomov } 17784609245eSIlya Dryomov 17794609245eSIlya Dryomov static void cancel_map_check(struct ceph_osd_request *req) 17804609245eSIlya Dryomov { 17814609245eSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 17824609245eSIlya Dryomov struct ceph_osd_request *lookup_req; 17834609245eSIlya Dryomov 17844609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 17854609245eSIlya Dryomov 17864609245eSIlya Dryomov lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 17874609245eSIlya Dryomov if (!lookup_req) 17884609245eSIlya Dryomov return; 17894609245eSIlya Dryomov 17904609245eSIlya Dryomov WARN_ON(lookup_req != req); 17914609245eSIlya Dryomov erase_request_mc(&osdc->map_checks, req); 17924609245eSIlya Dryomov ceph_osdc_put_request(req); 17934609245eSIlya Dryomov } 17944609245eSIlya Dryomov 17955aea3dcdSIlya Dryomov static void cancel_request(struct ceph_osd_request *req) 17965aea3dcdSIlya Dryomov { 17975aea3dcdSIlya Dryomov dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 17985aea3dcdSIlya Dryomov 17994609245eSIlya Dryomov cancel_map_check(req); 18005aea3dcdSIlya Dryomov finish_request(req); 18015aea3dcdSIlya Dryomov } 18025aea3dcdSIlya Dryomov 18034609245eSIlya Dryomov static void check_pool_dne(struct ceph_osd_request *req) 18044609245eSIlya Dryomov { 18054609245eSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 18064609245eSIlya Dryomov struct ceph_osdmap *map = osdc->osdmap; 18074609245eSIlya Dryomov 18084609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 18094609245eSIlya Dryomov WARN_ON(!map->epoch); 18104609245eSIlya Dryomov 18114609245eSIlya Dryomov if (req->r_attempts) { 18124609245eSIlya Dryomov /* 18134609245eSIlya Dryomov * We sent a request earlier, which means that 18144609245eSIlya Dryomov * previously the pool existed, and now it does not 18154609245eSIlya Dryomov * (i.e., it was deleted). 18164609245eSIlya Dryomov */ 18174609245eSIlya Dryomov req->r_map_dne_bound = map->epoch; 18184609245eSIlya Dryomov dout("%s req %p tid %llu pool disappeared\n", __func__, req, 18194609245eSIlya Dryomov req->r_tid); 18204609245eSIlya Dryomov } else { 18214609245eSIlya Dryomov dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, 18224609245eSIlya Dryomov req, req->r_tid, req->r_map_dne_bound, map->epoch); 18234609245eSIlya Dryomov } 18244609245eSIlya Dryomov 18254609245eSIlya Dryomov if (req->r_map_dne_bound) { 18264609245eSIlya Dryomov if (map->epoch >= req->r_map_dne_bound) { 18274609245eSIlya Dryomov /* we had a new enough map */ 18284609245eSIlya Dryomov pr_info_ratelimited("tid %llu pool does not exist\n", 18294609245eSIlya Dryomov req->r_tid); 18304609245eSIlya Dryomov complete_request(req, -ENOENT); 18314609245eSIlya Dryomov } 18324609245eSIlya Dryomov } else { 18334609245eSIlya Dryomov send_map_check(req); 18344609245eSIlya Dryomov } 18354609245eSIlya Dryomov } 18364609245eSIlya Dryomov 18374609245eSIlya Dryomov static void map_check_cb(struct ceph_mon_generic_request *greq) 18384609245eSIlya Dryomov { 18394609245eSIlya Dryomov struct ceph_osd_client *osdc = &greq->monc->client->osdc; 18404609245eSIlya Dryomov struct ceph_osd_request *req; 18414609245eSIlya Dryomov u64 tid = greq->private_data; 18424609245eSIlya Dryomov 18434609245eSIlya Dryomov WARN_ON(greq->result || !greq->u.newest); 18444609245eSIlya Dryomov 18454609245eSIlya Dryomov down_write(&osdc->lock); 18464609245eSIlya Dryomov req = lookup_request_mc(&osdc->map_checks, tid); 18474609245eSIlya Dryomov if (!req) { 18484609245eSIlya Dryomov dout("%s tid %llu dne\n", __func__, tid); 18494609245eSIlya Dryomov goto out_unlock; 18504609245eSIlya Dryomov } 18514609245eSIlya Dryomov 18524609245eSIlya Dryomov dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, 18534609245eSIlya Dryomov req, req->r_tid, req->r_map_dne_bound, greq->u.newest); 18544609245eSIlya Dryomov if (!req->r_map_dne_bound) 18554609245eSIlya Dryomov req->r_map_dne_bound = greq->u.newest; 18564609245eSIlya Dryomov erase_request_mc(&osdc->map_checks, req); 18574609245eSIlya Dryomov check_pool_dne(req); 18584609245eSIlya Dryomov 18594609245eSIlya Dryomov ceph_osdc_put_request(req); 18604609245eSIlya Dryomov out_unlock: 18614609245eSIlya Dryomov up_write(&osdc->lock); 18624609245eSIlya Dryomov } 18634609245eSIlya Dryomov 18644609245eSIlya Dryomov static void send_map_check(struct ceph_osd_request *req) 18654609245eSIlya Dryomov { 18664609245eSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 18674609245eSIlya Dryomov struct ceph_osd_request *lookup_req; 18684609245eSIlya Dryomov int ret; 18694609245eSIlya Dryomov 18704609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 18714609245eSIlya Dryomov 18724609245eSIlya Dryomov lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); 18734609245eSIlya Dryomov if (lookup_req) { 18744609245eSIlya Dryomov WARN_ON(lookup_req != req); 18754609245eSIlya Dryomov return; 18764609245eSIlya Dryomov } 18774609245eSIlya Dryomov 18784609245eSIlya Dryomov ceph_osdc_get_request(req); 18794609245eSIlya Dryomov insert_request_mc(&osdc->map_checks, req); 18804609245eSIlya Dryomov ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 18814609245eSIlya Dryomov map_check_cb, req->r_tid); 18824609245eSIlya Dryomov WARN_ON(ret); 18834609245eSIlya Dryomov } 18844609245eSIlya Dryomov 18850bbfdfe8SIlya Dryomov /* 1886922dab61SIlya Dryomov * lingering requests, watch/notify v2 infrastructure 1887922dab61SIlya Dryomov */ 1888922dab61SIlya Dryomov static void linger_release(struct kref *kref) 1889922dab61SIlya Dryomov { 1890922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = 1891922dab61SIlya Dryomov container_of(kref, struct ceph_osd_linger_request, kref); 1892922dab61SIlya Dryomov 1893922dab61SIlya Dryomov dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, 1894922dab61SIlya Dryomov lreq->reg_req, lreq->ping_req); 1895922dab61SIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&lreq->node)); 1896922dab61SIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); 18974609245eSIlya Dryomov WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); 1898922dab61SIlya Dryomov WARN_ON(!list_empty(&lreq->scan_item)); 1899b07d3c4bSIlya Dryomov WARN_ON(!list_empty(&lreq->pending_lworks)); 1900922dab61SIlya Dryomov WARN_ON(lreq->osd); 1901922dab61SIlya Dryomov 1902922dab61SIlya Dryomov if (lreq->reg_req) 1903922dab61SIlya Dryomov ceph_osdc_put_request(lreq->reg_req); 1904922dab61SIlya Dryomov if (lreq->ping_req) 1905922dab61SIlya Dryomov ceph_osdc_put_request(lreq->ping_req); 1906922dab61SIlya Dryomov target_destroy(&lreq->t); 1907922dab61SIlya Dryomov kfree(lreq); 1908922dab61SIlya Dryomov } 1909922dab61SIlya Dryomov 1910922dab61SIlya Dryomov static void linger_put(struct ceph_osd_linger_request *lreq) 1911922dab61SIlya Dryomov { 1912922dab61SIlya Dryomov if (lreq) 1913922dab61SIlya Dryomov kref_put(&lreq->kref, linger_release); 1914922dab61SIlya Dryomov } 1915922dab61SIlya Dryomov 1916922dab61SIlya Dryomov static struct ceph_osd_linger_request * 1917922dab61SIlya Dryomov linger_get(struct ceph_osd_linger_request *lreq) 1918922dab61SIlya Dryomov { 1919922dab61SIlya Dryomov kref_get(&lreq->kref); 1920922dab61SIlya Dryomov return lreq; 1921922dab61SIlya Dryomov } 1922922dab61SIlya Dryomov 1923922dab61SIlya Dryomov static struct ceph_osd_linger_request * 1924922dab61SIlya Dryomov linger_alloc(struct ceph_osd_client *osdc) 1925922dab61SIlya Dryomov { 1926922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq; 1927922dab61SIlya Dryomov 1928922dab61SIlya Dryomov lreq = kzalloc(sizeof(*lreq), GFP_NOIO); 1929922dab61SIlya Dryomov if (!lreq) 1930922dab61SIlya Dryomov return NULL; 1931922dab61SIlya Dryomov 1932922dab61SIlya Dryomov kref_init(&lreq->kref); 1933922dab61SIlya Dryomov mutex_init(&lreq->lock); 1934922dab61SIlya Dryomov RB_CLEAR_NODE(&lreq->node); 1935922dab61SIlya Dryomov RB_CLEAR_NODE(&lreq->osdc_node); 19364609245eSIlya Dryomov RB_CLEAR_NODE(&lreq->mc_node); 1937922dab61SIlya Dryomov INIT_LIST_HEAD(&lreq->scan_item); 1938b07d3c4bSIlya Dryomov INIT_LIST_HEAD(&lreq->pending_lworks); 1939922dab61SIlya Dryomov init_completion(&lreq->reg_commit_wait); 194019079203SIlya Dryomov init_completion(&lreq->notify_finish_wait); 1941922dab61SIlya Dryomov 1942922dab61SIlya Dryomov lreq->osdc = osdc; 1943922dab61SIlya Dryomov target_init(&lreq->t); 1944922dab61SIlya Dryomov 1945922dab61SIlya Dryomov dout("%s lreq %p\n", __func__, lreq); 1946922dab61SIlya Dryomov return lreq; 1947922dab61SIlya Dryomov } 1948922dab61SIlya Dryomov 1949922dab61SIlya Dryomov DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) 1950922dab61SIlya Dryomov DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) 19514609245eSIlya Dryomov DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) 1952922dab61SIlya Dryomov 1953922dab61SIlya Dryomov /* 1954922dab61SIlya Dryomov * Create linger request <-> OSD session relation. 1955922dab61SIlya Dryomov * 1956922dab61SIlya Dryomov * @lreq has to be registered, @osd may be homeless. 1957922dab61SIlya Dryomov */ 1958922dab61SIlya Dryomov static void link_linger(struct ceph_osd *osd, 1959922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq) 1960922dab61SIlya Dryomov { 1961922dab61SIlya Dryomov verify_osd_locked(osd); 1962922dab61SIlya Dryomov WARN_ON(!lreq->linger_id || lreq->osd); 1963922dab61SIlya Dryomov dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 1964922dab61SIlya Dryomov osd->o_osd, lreq, lreq->linger_id); 1965922dab61SIlya Dryomov 1966922dab61SIlya Dryomov if (!osd_homeless(osd)) 1967922dab61SIlya Dryomov __remove_osd_from_lru(osd); 1968922dab61SIlya Dryomov else 1969922dab61SIlya Dryomov atomic_inc(&osd->o_osdc->num_homeless); 1970922dab61SIlya Dryomov 1971922dab61SIlya Dryomov get_osd(osd); 1972922dab61SIlya Dryomov insert_linger(&osd->o_linger_requests, lreq); 1973922dab61SIlya Dryomov lreq->osd = osd; 1974922dab61SIlya Dryomov } 1975922dab61SIlya Dryomov 1976922dab61SIlya Dryomov static void unlink_linger(struct ceph_osd *osd, 1977922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq) 1978922dab61SIlya Dryomov { 1979922dab61SIlya Dryomov verify_osd_locked(osd); 1980922dab61SIlya Dryomov WARN_ON(lreq->osd != osd); 1981922dab61SIlya Dryomov dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, 1982922dab61SIlya Dryomov osd->o_osd, lreq, lreq->linger_id); 1983922dab61SIlya Dryomov 1984922dab61SIlya Dryomov lreq->osd = NULL; 1985922dab61SIlya Dryomov erase_linger(&osd->o_linger_requests, lreq); 1986922dab61SIlya Dryomov put_osd(osd); 1987922dab61SIlya Dryomov 1988922dab61SIlya Dryomov if (!osd_homeless(osd)) 1989922dab61SIlya Dryomov maybe_move_osd_to_lru(osd); 1990922dab61SIlya Dryomov else 1991922dab61SIlya Dryomov atomic_dec(&osd->o_osdc->num_homeless); 1992922dab61SIlya Dryomov } 1993922dab61SIlya Dryomov 1994922dab61SIlya Dryomov static bool __linger_registered(struct ceph_osd_linger_request *lreq) 1995922dab61SIlya Dryomov { 1996922dab61SIlya Dryomov verify_osdc_locked(lreq->osdc); 1997922dab61SIlya Dryomov 1998922dab61SIlya Dryomov return !RB_EMPTY_NODE(&lreq->osdc_node); 1999922dab61SIlya Dryomov } 2000922dab61SIlya Dryomov 2001922dab61SIlya Dryomov static bool linger_registered(struct ceph_osd_linger_request *lreq) 2002922dab61SIlya Dryomov { 2003922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2004922dab61SIlya Dryomov bool registered; 2005922dab61SIlya Dryomov 2006922dab61SIlya Dryomov down_read(&osdc->lock); 2007922dab61SIlya Dryomov registered = __linger_registered(lreq); 2008922dab61SIlya Dryomov up_read(&osdc->lock); 2009922dab61SIlya Dryomov 2010922dab61SIlya Dryomov return registered; 2011922dab61SIlya Dryomov } 2012922dab61SIlya Dryomov 2013922dab61SIlya Dryomov static void linger_register(struct ceph_osd_linger_request *lreq) 2014922dab61SIlya Dryomov { 2015922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2016922dab61SIlya Dryomov 2017922dab61SIlya Dryomov verify_osdc_wrlocked(osdc); 2018922dab61SIlya Dryomov WARN_ON(lreq->linger_id); 2019922dab61SIlya Dryomov 2020922dab61SIlya Dryomov linger_get(lreq); 2021922dab61SIlya Dryomov lreq->linger_id = ++osdc->last_linger_id; 2022922dab61SIlya Dryomov insert_linger_osdc(&osdc->linger_requests, lreq); 2023922dab61SIlya Dryomov } 2024922dab61SIlya Dryomov 2025922dab61SIlya Dryomov static void linger_unregister(struct ceph_osd_linger_request *lreq) 2026922dab61SIlya Dryomov { 2027922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2028922dab61SIlya Dryomov 2029922dab61SIlya Dryomov verify_osdc_wrlocked(osdc); 2030922dab61SIlya Dryomov 2031922dab61SIlya Dryomov erase_linger_osdc(&osdc->linger_requests, lreq); 2032922dab61SIlya Dryomov linger_put(lreq); 2033922dab61SIlya Dryomov } 2034922dab61SIlya Dryomov 2035922dab61SIlya Dryomov static void cancel_linger_request(struct ceph_osd_request *req) 2036922dab61SIlya Dryomov { 2037922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = req->r_priv; 2038922dab61SIlya Dryomov 2039922dab61SIlya Dryomov WARN_ON(!req->r_linger); 2040922dab61SIlya Dryomov cancel_request(req); 2041922dab61SIlya Dryomov linger_put(lreq); 2042922dab61SIlya Dryomov } 2043922dab61SIlya Dryomov 2044922dab61SIlya Dryomov struct linger_work { 2045922dab61SIlya Dryomov struct work_struct work; 2046922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq; 2047b07d3c4bSIlya Dryomov struct list_head pending_item; 2048b07d3c4bSIlya Dryomov unsigned long queued_stamp; 2049922dab61SIlya Dryomov 2050922dab61SIlya Dryomov union { 2051922dab61SIlya Dryomov struct { 2052922dab61SIlya Dryomov u64 notify_id; 2053922dab61SIlya Dryomov u64 notifier_id; 2054922dab61SIlya Dryomov void *payload; /* points into @msg front */ 2055922dab61SIlya Dryomov size_t payload_len; 2056922dab61SIlya Dryomov 2057922dab61SIlya Dryomov struct ceph_msg *msg; /* for ceph_msg_put() */ 2058922dab61SIlya Dryomov } notify; 2059922dab61SIlya Dryomov struct { 2060922dab61SIlya Dryomov int err; 2061922dab61SIlya Dryomov } error; 2062922dab61SIlya Dryomov }; 2063922dab61SIlya Dryomov }; 2064922dab61SIlya Dryomov 2065922dab61SIlya Dryomov static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, 2066922dab61SIlya Dryomov work_func_t workfn) 2067922dab61SIlya Dryomov { 2068922dab61SIlya Dryomov struct linger_work *lwork; 2069922dab61SIlya Dryomov 2070922dab61SIlya Dryomov lwork = kzalloc(sizeof(*lwork), GFP_NOIO); 2071922dab61SIlya Dryomov if (!lwork) 2072922dab61SIlya Dryomov return NULL; 2073922dab61SIlya Dryomov 2074922dab61SIlya Dryomov INIT_WORK(&lwork->work, workfn); 2075b07d3c4bSIlya Dryomov INIT_LIST_HEAD(&lwork->pending_item); 2076922dab61SIlya Dryomov lwork->lreq = linger_get(lreq); 2077922dab61SIlya Dryomov 2078922dab61SIlya Dryomov return lwork; 2079922dab61SIlya Dryomov } 2080922dab61SIlya Dryomov 2081922dab61SIlya Dryomov static void lwork_free(struct linger_work *lwork) 2082922dab61SIlya Dryomov { 2083922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = lwork->lreq; 2084922dab61SIlya Dryomov 2085b07d3c4bSIlya Dryomov mutex_lock(&lreq->lock); 2086b07d3c4bSIlya Dryomov list_del(&lwork->pending_item); 2087b07d3c4bSIlya Dryomov mutex_unlock(&lreq->lock); 2088b07d3c4bSIlya Dryomov 2089922dab61SIlya Dryomov linger_put(lreq); 2090922dab61SIlya Dryomov kfree(lwork); 2091922dab61SIlya Dryomov } 2092922dab61SIlya Dryomov 2093922dab61SIlya Dryomov static void lwork_queue(struct linger_work *lwork) 2094922dab61SIlya Dryomov { 2095922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = lwork->lreq; 2096922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2097922dab61SIlya Dryomov 2098922dab61SIlya Dryomov verify_lreq_locked(lreq); 2099b07d3c4bSIlya Dryomov WARN_ON(!list_empty(&lwork->pending_item)); 2100b07d3c4bSIlya Dryomov 2101b07d3c4bSIlya Dryomov lwork->queued_stamp = jiffies; 2102b07d3c4bSIlya Dryomov list_add_tail(&lwork->pending_item, &lreq->pending_lworks); 2103922dab61SIlya Dryomov queue_work(osdc->notify_wq, &lwork->work); 2104922dab61SIlya Dryomov } 2105922dab61SIlya Dryomov 2106922dab61SIlya Dryomov static void do_watch_notify(struct work_struct *w) 2107922dab61SIlya Dryomov { 2108922dab61SIlya Dryomov struct linger_work *lwork = container_of(w, struct linger_work, work); 2109922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = lwork->lreq; 2110922dab61SIlya Dryomov 2111922dab61SIlya Dryomov if (!linger_registered(lreq)) { 2112922dab61SIlya Dryomov dout("%s lreq %p not registered\n", __func__, lreq); 2113922dab61SIlya Dryomov goto out; 2114922dab61SIlya Dryomov } 2115922dab61SIlya Dryomov 211619079203SIlya Dryomov WARN_ON(!lreq->is_watch); 2117922dab61SIlya Dryomov dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", 2118922dab61SIlya Dryomov __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, 2119922dab61SIlya Dryomov lwork->notify.payload_len); 2120922dab61SIlya Dryomov lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, 2121922dab61SIlya Dryomov lwork->notify.notifier_id, lwork->notify.payload, 2122922dab61SIlya Dryomov lwork->notify.payload_len); 2123922dab61SIlya Dryomov 2124922dab61SIlya Dryomov out: 2125922dab61SIlya Dryomov ceph_msg_put(lwork->notify.msg); 2126922dab61SIlya Dryomov lwork_free(lwork); 2127922dab61SIlya Dryomov } 2128922dab61SIlya Dryomov 2129922dab61SIlya Dryomov static void do_watch_error(struct work_struct *w) 2130922dab61SIlya Dryomov { 2131922dab61SIlya Dryomov struct linger_work *lwork = container_of(w, struct linger_work, work); 2132922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = lwork->lreq; 2133922dab61SIlya Dryomov 2134922dab61SIlya Dryomov if (!linger_registered(lreq)) { 2135922dab61SIlya Dryomov dout("%s lreq %p not registered\n", __func__, lreq); 2136922dab61SIlya Dryomov goto out; 2137922dab61SIlya Dryomov } 2138922dab61SIlya Dryomov 2139922dab61SIlya Dryomov dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); 2140922dab61SIlya Dryomov lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); 2141922dab61SIlya Dryomov 2142922dab61SIlya Dryomov out: 2143922dab61SIlya Dryomov lwork_free(lwork); 2144922dab61SIlya Dryomov } 2145922dab61SIlya Dryomov 2146922dab61SIlya Dryomov static void queue_watch_error(struct ceph_osd_linger_request *lreq) 2147922dab61SIlya Dryomov { 2148922dab61SIlya Dryomov struct linger_work *lwork; 2149922dab61SIlya Dryomov 2150922dab61SIlya Dryomov lwork = lwork_alloc(lreq, do_watch_error); 2151922dab61SIlya Dryomov if (!lwork) { 2152922dab61SIlya Dryomov pr_err("failed to allocate error-lwork\n"); 2153922dab61SIlya Dryomov return; 2154922dab61SIlya Dryomov } 2155922dab61SIlya Dryomov 2156922dab61SIlya Dryomov lwork->error.err = lreq->last_error; 2157922dab61SIlya Dryomov lwork_queue(lwork); 2158922dab61SIlya Dryomov } 2159922dab61SIlya Dryomov 2160922dab61SIlya Dryomov static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, 2161922dab61SIlya Dryomov int result) 2162922dab61SIlya Dryomov { 2163922dab61SIlya Dryomov if (!completion_done(&lreq->reg_commit_wait)) { 2164922dab61SIlya Dryomov lreq->reg_commit_error = (result <= 0 ? result : 0); 2165922dab61SIlya Dryomov complete_all(&lreq->reg_commit_wait); 2166922dab61SIlya Dryomov } 2167922dab61SIlya Dryomov } 2168922dab61SIlya Dryomov 2169922dab61SIlya Dryomov static void linger_commit_cb(struct ceph_osd_request *req) 2170922dab61SIlya Dryomov { 2171922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = req->r_priv; 2172922dab61SIlya Dryomov 2173922dab61SIlya Dryomov mutex_lock(&lreq->lock); 2174922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, 2175922dab61SIlya Dryomov lreq->linger_id, req->r_result); 2176922dab61SIlya Dryomov WARN_ON(!__linger_registered(lreq)); 2177922dab61SIlya Dryomov linger_reg_commit_complete(lreq, req->r_result); 2178922dab61SIlya Dryomov lreq->committed = true; 2179922dab61SIlya Dryomov 218019079203SIlya Dryomov if (!lreq->is_watch) { 218119079203SIlya Dryomov struct ceph_osd_data *osd_data = 218219079203SIlya Dryomov osd_req_op_data(req, 0, notify, response_data); 218319079203SIlya Dryomov void *p = page_address(osd_data->pages[0]); 218419079203SIlya Dryomov 218519079203SIlya Dryomov WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || 218619079203SIlya Dryomov osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 218719079203SIlya Dryomov 218819079203SIlya Dryomov /* make note of the notify_id */ 218919079203SIlya Dryomov if (req->r_ops[0].outdata_len >= sizeof(u64)) { 219019079203SIlya Dryomov lreq->notify_id = ceph_decode_64(&p); 219119079203SIlya Dryomov dout("lreq %p notify_id %llu\n", lreq, 219219079203SIlya Dryomov lreq->notify_id); 219319079203SIlya Dryomov } else { 219419079203SIlya Dryomov dout("lreq %p no notify_id\n", lreq); 219519079203SIlya Dryomov } 219619079203SIlya Dryomov } 219719079203SIlya Dryomov 2198922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 2199922dab61SIlya Dryomov linger_put(lreq); 2200922dab61SIlya Dryomov } 2201922dab61SIlya Dryomov 2202922dab61SIlya Dryomov static int normalize_watch_error(int err) 2203922dab61SIlya Dryomov { 2204922dab61SIlya Dryomov /* 2205922dab61SIlya Dryomov * Translate ENOENT -> ENOTCONN so that a delete->disconnection 2206922dab61SIlya Dryomov * notification and a failure to reconnect because we raced with 2207922dab61SIlya Dryomov * the delete appear the same to the user. 2208922dab61SIlya Dryomov */ 2209922dab61SIlya Dryomov if (err == -ENOENT) 2210922dab61SIlya Dryomov err = -ENOTCONN; 2211922dab61SIlya Dryomov 2212922dab61SIlya Dryomov return err; 2213922dab61SIlya Dryomov } 2214922dab61SIlya Dryomov 2215922dab61SIlya Dryomov static void linger_reconnect_cb(struct ceph_osd_request *req) 2216922dab61SIlya Dryomov { 2217922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = req->r_priv; 2218922dab61SIlya Dryomov 2219922dab61SIlya Dryomov mutex_lock(&lreq->lock); 2220922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, 2221922dab61SIlya Dryomov lreq, lreq->linger_id, req->r_result, lreq->last_error); 2222922dab61SIlya Dryomov if (req->r_result < 0) { 2223922dab61SIlya Dryomov if (!lreq->last_error) { 2224922dab61SIlya Dryomov lreq->last_error = normalize_watch_error(req->r_result); 2225922dab61SIlya Dryomov queue_watch_error(lreq); 2226922dab61SIlya Dryomov } 2227922dab61SIlya Dryomov } 2228922dab61SIlya Dryomov 2229922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 2230922dab61SIlya Dryomov linger_put(lreq); 2231922dab61SIlya Dryomov } 2232922dab61SIlya Dryomov 2233922dab61SIlya Dryomov static void send_linger(struct ceph_osd_linger_request *lreq) 2234922dab61SIlya Dryomov { 2235922dab61SIlya Dryomov struct ceph_osd_request *req = lreq->reg_req; 2236922dab61SIlya Dryomov struct ceph_osd_req_op *op = &req->r_ops[0]; 2237922dab61SIlya Dryomov 2238922dab61SIlya Dryomov verify_osdc_wrlocked(req->r_osdc); 2239922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2240922dab61SIlya Dryomov 2241922dab61SIlya Dryomov if (req->r_osd) 2242922dab61SIlya Dryomov cancel_linger_request(req); 2243922dab61SIlya Dryomov 2244922dab61SIlya Dryomov request_reinit(req); 2245922dab61SIlya Dryomov ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 2246922dab61SIlya Dryomov ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 2247922dab61SIlya Dryomov req->r_flags = lreq->t.flags; 2248922dab61SIlya Dryomov req->r_mtime = lreq->mtime; 2249922dab61SIlya Dryomov 2250922dab61SIlya Dryomov mutex_lock(&lreq->lock); 225119079203SIlya Dryomov if (lreq->is_watch && lreq->committed) { 2252922dab61SIlya Dryomov WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2253922dab61SIlya Dryomov op->watch.cookie != lreq->linger_id); 2254922dab61SIlya Dryomov op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT; 2255922dab61SIlya Dryomov op->watch.gen = ++lreq->register_gen; 2256922dab61SIlya Dryomov dout("lreq %p reconnect register_gen %u\n", lreq, 2257922dab61SIlya Dryomov op->watch.gen); 2258922dab61SIlya Dryomov req->r_callback = linger_reconnect_cb; 2259922dab61SIlya Dryomov } else { 226019079203SIlya Dryomov if (!lreq->is_watch) 226119079203SIlya Dryomov lreq->notify_id = 0; 226219079203SIlya Dryomov else 2263922dab61SIlya Dryomov WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH); 2264922dab61SIlya Dryomov dout("lreq %p register\n", lreq); 2265922dab61SIlya Dryomov req->r_callback = linger_commit_cb; 2266922dab61SIlya Dryomov } 2267922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 2268922dab61SIlya Dryomov 2269922dab61SIlya Dryomov req->r_priv = linger_get(lreq); 2270922dab61SIlya Dryomov req->r_linger = true; 2271922dab61SIlya Dryomov 2272922dab61SIlya Dryomov submit_request(req, true); 2273922dab61SIlya Dryomov } 2274922dab61SIlya Dryomov 2275922dab61SIlya Dryomov static void linger_ping_cb(struct ceph_osd_request *req) 2276922dab61SIlya Dryomov { 2277922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = req->r_priv; 2278922dab61SIlya Dryomov 2279922dab61SIlya Dryomov mutex_lock(&lreq->lock); 2280922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", 2281922dab61SIlya Dryomov __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, 2282922dab61SIlya Dryomov lreq->last_error); 2283922dab61SIlya Dryomov if (lreq->register_gen == req->r_ops[0].watch.gen) { 2284b07d3c4bSIlya Dryomov if (!req->r_result) { 2285b07d3c4bSIlya Dryomov lreq->watch_valid_thru = lreq->ping_sent; 2286b07d3c4bSIlya Dryomov } else if (!lreq->last_error) { 2287922dab61SIlya Dryomov lreq->last_error = normalize_watch_error(req->r_result); 2288922dab61SIlya Dryomov queue_watch_error(lreq); 2289922dab61SIlya Dryomov } 2290922dab61SIlya Dryomov } else { 2291922dab61SIlya Dryomov dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, 2292922dab61SIlya Dryomov lreq->register_gen, req->r_ops[0].watch.gen); 2293922dab61SIlya Dryomov } 2294922dab61SIlya Dryomov 2295922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 2296922dab61SIlya Dryomov linger_put(lreq); 2297922dab61SIlya Dryomov } 2298922dab61SIlya Dryomov 2299922dab61SIlya Dryomov static void send_linger_ping(struct ceph_osd_linger_request *lreq) 2300922dab61SIlya Dryomov { 2301922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2302922dab61SIlya Dryomov struct ceph_osd_request *req = lreq->ping_req; 2303922dab61SIlya Dryomov struct ceph_osd_req_op *op = &req->r_ops[0]; 2304922dab61SIlya Dryomov 2305b7ec35b3SIlya Dryomov if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { 2306922dab61SIlya Dryomov dout("%s PAUSERD\n", __func__); 2307922dab61SIlya Dryomov return; 2308922dab61SIlya Dryomov } 2309922dab61SIlya Dryomov 2310922dab61SIlya Dryomov lreq->ping_sent = jiffies; 2311922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", 2312922dab61SIlya Dryomov __func__, lreq, lreq->linger_id, lreq->ping_sent, 2313922dab61SIlya Dryomov lreq->register_gen); 2314922dab61SIlya Dryomov 2315922dab61SIlya Dryomov if (req->r_osd) 2316922dab61SIlya Dryomov cancel_linger_request(req); 2317922dab61SIlya Dryomov 2318922dab61SIlya Dryomov request_reinit(req); 2319922dab61SIlya Dryomov target_copy(&req->r_t, &lreq->t); 2320922dab61SIlya Dryomov 2321922dab61SIlya Dryomov WARN_ON(op->op != CEPH_OSD_OP_WATCH || 2322922dab61SIlya Dryomov op->watch.cookie != lreq->linger_id || 2323922dab61SIlya Dryomov op->watch.op != CEPH_OSD_WATCH_OP_PING); 2324922dab61SIlya Dryomov op->watch.gen = lreq->register_gen; 2325922dab61SIlya Dryomov req->r_callback = linger_ping_cb; 2326922dab61SIlya Dryomov req->r_priv = linger_get(lreq); 2327922dab61SIlya Dryomov req->r_linger = true; 2328922dab61SIlya Dryomov 2329922dab61SIlya Dryomov ceph_osdc_get_request(req); 2330922dab61SIlya Dryomov account_request(req); 2331922dab61SIlya Dryomov req->r_tid = atomic64_inc_return(&osdc->last_tid); 2332922dab61SIlya Dryomov link_request(lreq->osd, req); 2333922dab61SIlya Dryomov send_request(req); 2334922dab61SIlya Dryomov } 2335922dab61SIlya Dryomov 2336922dab61SIlya Dryomov static void linger_submit(struct ceph_osd_linger_request *lreq) 2337922dab61SIlya Dryomov { 2338922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2339922dab61SIlya Dryomov struct ceph_osd *osd; 2340922dab61SIlya Dryomov 2341922dab61SIlya Dryomov calc_target(osdc, &lreq->t, &lreq->last_force_resend, false); 2342922dab61SIlya Dryomov osd = lookup_create_osd(osdc, lreq->t.osd, true); 2343922dab61SIlya Dryomov link_linger(osd, lreq); 2344922dab61SIlya Dryomov 2345922dab61SIlya Dryomov send_linger(lreq); 2346922dab61SIlya Dryomov } 2347922dab61SIlya Dryomov 23484609245eSIlya Dryomov static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) 23494609245eSIlya Dryomov { 23504609245eSIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 23514609245eSIlya Dryomov struct ceph_osd_linger_request *lookup_lreq; 23524609245eSIlya Dryomov 23534609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 23544609245eSIlya Dryomov 23554609245eSIlya Dryomov lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 23564609245eSIlya Dryomov lreq->linger_id); 23574609245eSIlya Dryomov if (!lookup_lreq) 23584609245eSIlya Dryomov return; 23594609245eSIlya Dryomov 23604609245eSIlya Dryomov WARN_ON(lookup_lreq != lreq); 23614609245eSIlya Dryomov erase_linger_mc(&osdc->linger_map_checks, lreq); 23624609245eSIlya Dryomov linger_put(lreq); 23634609245eSIlya Dryomov } 23644609245eSIlya Dryomov 2365922dab61SIlya Dryomov /* 2366922dab61SIlya Dryomov * @lreq has to be both registered and linked. 2367922dab61SIlya Dryomov */ 2368922dab61SIlya Dryomov static void __linger_cancel(struct ceph_osd_linger_request *lreq) 2369922dab61SIlya Dryomov { 237019079203SIlya Dryomov if (lreq->is_watch && lreq->ping_req->r_osd) 2371922dab61SIlya Dryomov cancel_linger_request(lreq->ping_req); 2372922dab61SIlya Dryomov if (lreq->reg_req->r_osd) 2373922dab61SIlya Dryomov cancel_linger_request(lreq->reg_req); 23744609245eSIlya Dryomov cancel_linger_map_check(lreq); 2375922dab61SIlya Dryomov unlink_linger(lreq->osd, lreq); 2376922dab61SIlya Dryomov linger_unregister(lreq); 2377922dab61SIlya Dryomov } 2378922dab61SIlya Dryomov 2379922dab61SIlya Dryomov static void linger_cancel(struct ceph_osd_linger_request *lreq) 2380922dab61SIlya Dryomov { 2381922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2382922dab61SIlya Dryomov 2383922dab61SIlya Dryomov down_write(&osdc->lock); 2384922dab61SIlya Dryomov if (__linger_registered(lreq)) 2385922dab61SIlya Dryomov __linger_cancel(lreq); 2386922dab61SIlya Dryomov up_write(&osdc->lock); 2387922dab61SIlya Dryomov } 2388922dab61SIlya Dryomov 23894609245eSIlya Dryomov static void send_linger_map_check(struct ceph_osd_linger_request *lreq); 23904609245eSIlya Dryomov 23914609245eSIlya Dryomov static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) 23924609245eSIlya Dryomov { 23934609245eSIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 23944609245eSIlya Dryomov struct ceph_osdmap *map = osdc->osdmap; 23954609245eSIlya Dryomov 23964609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 23974609245eSIlya Dryomov WARN_ON(!map->epoch); 23984609245eSIlya Dryomov 23994609245eSIlya Dryomov if (lreq->register_gen) { 24004609245eSIlya Dryomov lreq->map_dne_bound = map->epoch; 24014609245eSIlya Dryomov dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, 24024609245eSIlya Dryomov lreq, lreq->linger_id); 24034609245eSIlya Dryomov } else { 24044609245eSIlya Dryomov dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", 24054609245eSIlya Dryomov __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 24064609245eSIlya Dryomov map->epoch); 24074609245eSIlya Dryomov } 24084609245eSIlya Dryomov 24094609245eSIlya Dryomov if (lreq->map_dne_bound) { 24104609245eSIlya Dryomov if (map->epoch >= lreq->map_dne_bound) { 24114609245eSIlya Dryomov /* we had a new enough map */ 24124609245eSIlya Dryomov pr_info("linger_id %llu pool does not exist\n", 24134609245eSIlya Dryomov lreq->linger_id); 24144609245eSIlya Dryomov linger_reg_commit_complete(lreq, -ENOENT); 24154609245eSIlya Dryomov __linger_cancel(lreq); 24164609245eSIlya Dryomov } 24174609245eSIlya Dryomov } else { 24184609245eSIlya Dryomov send_linger_map_check(lreq); 24194609245eSIlya Dryomov } 24204609245eSIlya Dryomov } 24214609245eSIlya Dryomov 24224609245eSIlya Dryomov static void linger_map_check_cb(struct ceph_mon_generic_request *greq) 24234609245eSIlya Dryomov { 24244609245eSIlya Dryomov struct ceph_osd_client *osdc = &greq->monc->client->osdc; 24254609245eSIlya Dryomov struct ceph_osd_linger_request *lreq; 24264609245eSIlya Dryomov u64 linger_id = greq->private_data; 24274609245eSIlya Dryomov 24284609245eSIlya Dryomov WARN_ON(greq->result || !greq->u.newest); 24294609245eSIlya Dryomov 24304609245eSIlya Dryomov down_write(&osdc->lock); 24314609245eSIlya Dryomov lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); 24324609245eSIlya Dryomov if (!lreq) { 24334609245eSIlya Dryomov dout("%s linger_id %llu dne\n", __func__, linger_id); 24344609245eSIlya Dryomov goto out_unlock; 24354609245eSIlya Dryomov } 24364609245eSIlya Dryomov 24374609245eSIlya Dryomov dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", 24384609245eSIlya Dryomov __func__, lreq, lreq->linger_id, lreq->map_dne_bound, 24394609245eSIlya Dryomov greq->u.newest); 24404609245eSIlya Dryomov if (!lreq->map_dne_bound) 24414609245eSIlya Dryomov lreq->map_dne_bound = greq->u.newest; 24424609245eSIlya Dryomov erase_linger_mc(&osdc->linger_map_checks, lreq); 24434609245eSIlya Dryomov check_linger_pool_dne(lreq); 24444609245eSIlya Dryomov 24454609245eSIlya Dryomov linger_put(lreq); 24464609245eSIlya Dryomov out_unlock: 24474609245eSIlya Dryomov up_write(&osdc->lock); 24484609245eSIlya Dryomov } 24494609245eSIlya Dryomov 24504609245eSIlya Dryomov static void send_linger_map_check(struct ceph_osd_linger_request *lreq) 24514609245eSIlya Dryomov { 24524609245eSIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 24534609245eSIlya Dryomov struct ceph_osd_linger_request *lookup_lreq; 24544609245eSIlya Dryomov int ret; 24554609245eSIlya Dryomov 24564609245eSIlya Dryomov verify_osdc_wrlocked(osdc); 24574609245eSIlya Dryomov 24584609245eSIlya Dryomov lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, 24594609245eSIlya Dryomov lreq->linger_id); 24604609245eSIlya Dryomov if (lookup_lreq) { 24614609245eSIlya Dryomov WARN_ON(lookup_lreq != lreq); 24624609245eSIlya Dryomov return; 24634609245eSIlya Dryomov } 24644609245eSIlya Dryomov 24654609245eSIlya Dryomov linger_get(lreq); 24664609245eSIlya Dryomov insert_linger_mc(&osdc->linger_map_checks, lreq); 24674609245eSIlya Dryomov ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", 24684609245eSIlya Dryomov linger_map_check_cb, lreq->linger_id); 24694609245eSIlya Dryomov WARN_ON(ret); 24704609245eSIlya Dryomov } 24714609245eSIlya Dryomov 2472922dab61SIlya Dryomov static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) 2473922dab61SIlya Dryomov { 2474922dab61SIlya Dryomov int ret; 2475922dab61SIlya Dryomov 2476922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 2477922dab61SIlya Dryomov ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); 2478922dab61SIlya Dryomov return ret ?: lreq->reg_commit_error; 2479922dab61SIlya Dryomov } 2480922dab61SIlya Dryomov 248119079203SIlya Dryomov static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) 248219079203SIlya Dryomov { 248319079203SIlya Dryomov int ret; 248419079203SIlya Dryomov 248519079203SIlya Dryomov dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); 248619079203SIlya Dryomov ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); 248719079203SIlya Dryomov return ret ?: lreq->notify_finish_error; 248819079203SIlya Dryomov } 248919079203SIlya Dryomov 2490922dab61SIlya Dryomov /* 2491fbca9635SIlya Dryomov * Timeout callback, called every N seconds. When 1 or more OSD 2492fbca9635SIlya Dryomov * requests has been active for more than N seconds, we send a keepalive 2493fbca9635SIlya Dryomov * (tag + timestamp) to its OSD to ensure any communications channel 2494fbca9635SIlya Dryomov * reset is detected. 24953d14c5d2SYehuda Sadeh */ 24963d14c5d2SYehuda Sadeh static void handle_timeout(struct work_struct *work) 24973d14c5d2SYehuda Sadeh { 24983d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = 24993d14c5d2SYehuda Sadeh container_of(work, struct ceph_osd_client, timeout_work.work); 2500a319bf56SIlya Dryomov struct ceph_options *opts = osdc->client->options; 25015aea3dcdSIlya Dryomov unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; 25025aea3dcdSIlya Dryomov LIST_HEAD(slow_osds); 25035aea3dcdSIlya Dryomov struct rb_node *n, *p; 25043d14c5d2SYehuda Sadeh 25055aea3dcdSIlya Dryomov dout("%s osdc %p\n", __func__, osdc); 25065aea3dcdSIlya Dryomov down_write(&osdc->lock); 25073d14c5d2SYehuda Sadeh 25083d14c5d2SYehuda Sadeh /* 25093d14c5d2SYehuda Sadeh * ping osds that are a bit slow. this ensures that if there 25103d14c5d2SYehuda Sadeh * is a break in the TCP connection we will notice, and reopen 25113d14c5d2SYehuda Sadeh * a connection with that osd (from the fault callback). 25123d14c5d2SYehuda Sadeh */ 25135aea3dcdSIlya Dryomov for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 25145aea3dcdSIlya Dryomov struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 25155aea3dcdSIlya Dryomov bool found = false; 25163d14c5d2SYehuda Sadeh 25175aea3dcdSIlya Dryomov for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { 25185aea3dcdSIlya Dryomov struct ceph_osd_request *req = 25195aea3dcdSIlya Dryomov rb_entry(p, struct ceph_osd_request, r_node); 25205aea3dcdSIlya Dryomov 25215aea3dcdSIlya Dryomov if (time_before(req->r_stamp, cutoff)) { 25225aea3dcdSIlya Dryomov dout(" req %p tid %llu on osd%d is laggy\n", 25235aea3dcdSIlya Dryomov req, req->r_tid, osd->o_osd); 25245aea3dcdSIlya Dryomov found = true; 25255aea3dcdSIlya Dryomov } 25265aea3dcdSIlya Dryomov } 2527922dab61SIlya Dryomov for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { 2528922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = 2529922dab61SIlya Dryomov rb_entry(p, struct ceph_osd_linger_request, node); 2530922dab61SIlya Dryomov 2531922dab61SIlya Dryomov dout(" lreq %p linger_id %llu is served by osd%d\n", 2532922dab61SIlya Dryomov lreq, lreq->linger_id, osd->o_osd); 2533922dab61SIlya Dryomov found = true; 2534922dab61SIlya Dryomov 2535922dab61SIlya Dryomov mutex_lock(&lreq->lock); 253619079203SIlya Dryomov if (lreq->is_watch && lreq->committed && !lreq->last_error) 2537922dab61SIlya Dryomov send_linger_ping(lreq); 2538922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 2539922dab61SIlya Dryomov } 25405aea3dcdSIlya Dryomov 25415aea3dcdSIlya Dryomov if (found) 25423d14c5d2SYehuda Sadeh list_move_tail(&osd->o_keepalive_item, &slow_osds); 25433d14c5d2SYehuda Sadeh } 25445aea3dcdSIlya Dryomov 25455aea3dcdSIlya Dryomov if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) 25465aea3dcdSIlya Dryomov maybe_request_map(osdc); 25475aea3dcdSIlya Dryomov 25483d14c5d2SYehuda Sadeh while (!list_empty(&slow_osds)) { 25495aea3dcdSIlya Dryomov struct ceph_osd *osd = list_first_entry(&slow_osds, 25505aea3dcdSIlya Dryomov struct ceph_osd, 25513d14c5d2SYehuda Sadeh o_keepalive_item); 25523d14c5d2SYehuda Sadeh list_del_init(&osd->o_keepalive_item); 25533d14c5d2SYehuda Sadeh ceph_con_keepalive(&osd->o_con); 25543d14c5d2SYehuda Sadeh } 25553d14c5d2SYehuda Sadeh 25565aea3dcdSIlya Dryomov up_write(&osdc->lock); 2557fbca9635SIlya Dryomov schedule_delayed_work(&osdc->timeout_work, 2558fbca9635SIlya Dryomov osdc->client->options->osd_keepalive_timeout); 25593d14c5d2SYehuda Sadeh } 25603d14c5d2SYehuda Sadeh 25613d14c5d2SYehuda Sadeh static void handle_osds_timeout(struct work_struct *work) 25623d14c5d2SYehuda Sadeh { 25633d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = 25643d14c5d2SYehuda Sadeh container_of(work, struct ceph_osd_client, 25653d14c5d2SYehuda Sadeh osds_timeout_work.work); 2566a319bf56SIlya Dryomov unsigned long delay = osdc->client->options->osd_idle_ttl / 4; 256742a2c09fSIlya Dryomov struct ceph_osd *osd, *nosd; 25683d14c5d2SYehuda Sadeh 256942a2c09fSIlya Dryomov dout("%s osdc %p\n", __func__, osdc); 25705aea3dcdSIlya Dryomov down_write(&osdc->lock); 257142a2c09fSIlya Dryomov list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { 257242a2c09fSIlya Dryomov if (time_before(jiffies, osd->lru_ttl)) 257342a2c09fSIlya Dryomov break; 257442a2c09fSIlya Dryomov 25755aea3dcdSIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); 2576922dab61SIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); 25775aea3dcdSIlya Dryomov close_osd(osd); 257842a2c09fSIlya Dryomov } 257942a2c09fSIlya Dryomov 25805aea3dcdSIlya Dryomov up_write(&osdc->lock); 25813d14c5d2SYehuda Sadeh schedule_delayed_work(&osdc->osds_timeout_work, 25823d14c5d2SYehuda Sadeh round_jiffies_relative(delay)); 25833d14c5d2SYehuda Sadeh } 25843d14c5d2SYehuda Sadeh 2585205ee118SIlya Dryomov static int ceph_oloc_decode(void **p, void *end, 2586205ee118SIlya Dryomov struct ceph_object_locator *oloc) 2587205ee118SIlya Dryomov { 2588205ee118SIlya Dryomov u8 struct_v, struct_cv; 2589205ee118SIlya Dryomov u32 len; 2590205ee118SIlya Dryomov void *struct_end; 2591205ee118SIlya Dryomov int ret = 0; 2592205ee118SIlya Dryomov 2593205ee118SIlya Dryomov ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 2594205ee118SIlya Dryomov struct_v = ceph_decode_8(p); 2595205ee118SIlya Dryomov struct_cv = ceph_decode_8(p); 2596205ee118SIlya Dryomov if (struct_v < 3) { 2597205ee118SIlya Dryomov pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", 2598205ee118SIlya Dryomov struct_v, struct_cv); 2599205ee118SIlya Dryomov goto e_inval; 2600205ee118SIlya Dryomov } 2601205ee118SIlya Dryomov if (struct_cv > 6) { 2602205ee118SIlya Dryomov pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", 2603205ee118SIlya Dryomov struct_v, struct_cv); 2604205ee118SIlya Dryomov goto e_inval; 2605205ee118SIlya Dryomov } 2606205ee118SIlya Dryomov len = ceph_decode_32(p); 2607205ee118SIlya Dryomov ceph_decode_need(p, end, len, e_inval); 2608205ee118SIlya Dryomov struct_end = *p + len; 2609205ee118SIlya Dryomov 2610205ee118SIlya Dryomov oloc->pool = ceph_decode_64(p); 2611205ee118SIlya Dryomov *p += 4; /* skip preferred */ 2612205ee118SIlya Dryomov 2613205ee118SIlya Dryomov len = ceph_decode_32(p); 2614205ee118SIlya Dryomov if (len > 0) { 2615205ee118SIlya Dryomov pr_warn("ceph_object_locator::key is set\n"); 2616205ee118SIlya Dryomov goto e_inval; 2617205ee118SIlya Dryomov } 2618205ee118SIlya Dryomov 2619205ee118SIlya Dryomov if (struct_v >= 5) { 2620cd08e0a2SYan, Zheng bool changed = false; 2621cd08e0a2SYan, Zheng 2622205ee118SIlya Dryomov len = ceph_decode_32(p); 2623205ee118SIlya Dryomov if (len > 0) { 262430c156d9SYan, Zheng ceph_decode_need(p, end, len, e_inval); 2625cd08e0a2SYan, Zheng if (!oloc->pool_ns || 2626cd08e0a2SYan, Zheng ceph_compare_string(oloc->pool_ns, *p, len)) 2627cd08e0a2SYan, Zheng changed = true; 262830c156d9SYan, Zheng *p += len; 2629cd08e0a2SYan, Zheng } else { 2630cd08e0a2SYan, Zheng if (oloc->pool_ns) 2631cd08e0a2SYan, Zheng changed = true; 2632cd08e0a2SYan, Zheng } 2633cd08e0a2SYan, Zheng if (changed) { 2634cd08e0a2SYan, Zheng /* redirect changes namespace */ 2635cd08e0a2SYan, Zheng pr_warn("ceph_object_locator::nspace is changed\n"); 2636cd08e0a2SYan, Zheng goto e_inval; 2637205ee118SIlya Dryomov } 2638205ee118SIlya Dryomov } 2639205ee118SIlya Dryomov 2640205ee118SIlya Dryomov if (struct_v >= 6) { 2641205ee118SIlya Dryomov s64 hash = ceph_decode_64(p); 2642205ee118SIlya Dryomov if (hash != -1) { 2643205ee118SIlya Dryomov pr_warn("ceph_object_locator::hash is set\n"); 2644205ee118SIlya Dryomov goto e_inval; 2645205ee118SIlya Dryomov } 2646205ee118SIlya Dryomov } 2647205ee118SIlya Dryomov 2648205ee118SIlya Dryomov /* skip the rest */ 2649205ee118SIlya Dryomov *p = struct_end; 2650205ee118SIlya Dryomov out: 2651205ee118SIlya Dryomov return ret; 2652205ee118SIlya Dryomov 2653205ee118SIlya Dryomov e_inval: 2654205ee118SIlya Dryomov ret = -EINVAL; 2655205ee118SIlya Dryomov goto out; 2656205ee118SIlya Dryomov } 2657205ee118SIlya Dryomov 2658205ee118SIlya Dryomov static int ceph_redirect_decode(void **p, void *end, 2659205ee118SIlya Dryomov struct ceph_request_redirect *redir) 2660205ee118SIlya Dryomov { 2661205ee118SIlya Dryomov u8 struct_v, struct_cv; 2662205ee118SIlya Dryomov u32 len; 2663205ee118SIlya Dryomov void *struct_end; 2664205ee118SIlya Dryomov int ret; 2665205ee118SIlya Dryomov 2666205ee118SIlya Dryomov ceph_decode_need(p, end, 1 + 1 + 4, e_inval); 2667205ee118SIlya Dryomov struct_v = ceph_decode_8(p); 2668205ee118SIlya Dryomov struct_cv = ceph_decode_8(p); 2669205ee118SIlya Dryomov if (struct_cv > 1) { 2670205ee118SIlya Dryomov pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", 2671205ee118SIlya Dryomov struct_v, struct_cv); 2672205ee118SIlya Dryomov goto e_inval; 2673205ee118SIlya Dryomov } 2674205ee118SIlya Dryomov len = ceph_decode_32(p); 2675205ee118SIlya Dryomov ceph_decode_need(p, end, len, e_inval); 2676205ee118SIlya Dryomov struct_end = *p + len; 2677205ee118SIlya Dryomov 2678205ee118SIlya Dryomov ret = ceph_oloc_decode(p, end, &redir->oloc); 2679205ee118SIlya Dryomov if (ret) 2680205ee118SIlya Dryomov goto out; 2681205ee118SIlya Dryomov 2682205ee118SIlya Dryomov len = ceph_decode_32(p); 2683205ee118SIlya Dryomov if (len > 0) { 2684205ee118SIlya Dryomov pr_warn("ceph_request_redirect::object_name is set\n"); 2685205ee118SIlya Dryomov goto e_inval; 2686205ee118SIlya Dryomov } 2687205ee118SIlya Dryomov 2688205ee118SIlya Dryomov len = ceph_decode_32(p); 2689205ee118SIlya Dryomov *p += len; /* skip osd_instructions */ 2690205ee118SIlya Dryomov 2691205ee118SIlya Dryomov /* skip the rest */ 2692205ee118SIlya Dryomov *p = struct_end; 2693205ee118SIlya Dryomov out: 2694205ee118SIlya Dryomov return ret; 2695205ee118SIlya Dryomov 2696205ee118SIlya Dryomov e_inval: 2697205ee118SIlya Dryomov ret = -EINVAL; 2698205ee118SIlya Dryomov goto out; 2699205ee118SIlya Dryomov } 2700205ee118SIlya Dryomov 2701fe5da05eSIlya Dryomov struct MOSDOpReply { 2702fe5da05eSIlya Dryomov struct ceph_pg pgid; 2703fe5da05eSIlya Dryomov u64 flags; 2704fe5da05eSIlya Dryomov int result; 2705fe5da05eSIlya Dryomov u32 epoch; 2706fe5da05eSIlya Dryomov int num_ops; 2707fe5da05eSIlya Dryomov u32 outdata_len[CEPH_OSD_MAX_OPS]; 2708fe5da05eSIlya Dryomov s32 rval[CEPH_OSD_MAX_OPS]; 2709fe5da05eSIlya Dryomov int retry_attempt; 2710fe5da05eSIlya Dryomov struct ceph_eversion replay_version; 2711fe5da05eSIlya Dryomov u64 user_version; 2712fe5da05eSIlya Dryomov struct ceph_request_redirect redirect; 2713fe5da05eSIlya Dryomov }; 271425845472SSage Weil 2715fe5da05eSIlya Dryomov static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) 27163d14c5d2SYehuda Sadeh { 2717fe5da05eSIlya Dryomov void *p = msg->front.iov_base; 2718fe5da05eSIlya Dryomov void *const end = p + msg->front.iov_len; 2719fe5da05eSIlya Dryomov u16 version = le16_to_cpu(msg->hdr.version); 2720fe5da05eSIlya Dryomov struct ceph_eversion bad_replay_version; 2721b0b31a8fSIlya Dryomov u8 decode_redir; 2722fe5da05eSIlya Dryomov u32 len; 2723fe5da05eSIlya Dryomov int ret; 2724fe5da05eSIlya Dryomov int i; 27253d14c5d2SYehuda Sadeh 2726fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, len, e_inval); 2727fe5da05eSIlya Dryomov ceph_decode_need(&p, end, len, e_inval); 2728fe5da05eSIlya Dryomov p += len; /* skip oid */ 27291b83bef2SSage Weil 2730fe5da05eSIlya Dryomov ret = ceph_decode_pgid(&p, end, &m->pgid); 2731fe5da05eSIlya Dryomov if (ret) 2732fe5da05eSIlya Dryomov return ret; 27331b83bef2SSage Weil 2734fe5da05eSIlya Dryomov ceph_decode_64_safe(&p, end, m->flags, e_inval); 2735fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, m->result, e_inval); 2736fe5da05eSIlya Dryomov ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); 2737fe5da05eSIlya Dryomov memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); 2738fe5da05eSIlya Dryomov p += sizeof(bad_replay_version); 2739fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, m->epoch, e_inval); 27401b83bef2SSage Weil 2741fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, m->num_ops, e_inval); 2742fe5da05eSIlya Dryomov if (m->num_ops > ARRAY_SIZE(m->outdata_len)) 2743fe5da05eSIlya Dryomov goto e_inval; 27441b83bef2SSage Weil 2745fe5da05eSIlya Dryomov ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), 2746fe5da05eSIlya Dryomov e_inval); 2747fe5da05eSIlya Dryomov for (i = 0; i < m->num_ops; i++) { 27481b83bef2SSage Weil struct ceph_osd_op *op = p; 27491b83bef2SSage Weil 2750fe5da05eSIlya Dryomov m->outdata_len[i] = le32_to_cpu(op->payload_len); 27511b83bef2SSage Weil p += sizeof(*op); 27521b83bef2SSage Weil } 2753fe5da05eSIlya Dryomov 2754fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); 2755fe5da05eSIlya Dryomov for (i = 0; i < m->num_ops; i++) 2756fe5da05eSIlya Dryomov ceph_decode_32_safe(&p, end, m->rval[i], e_inval); 2757fe5da05eSIlya Dryomov 2758fe5da05eSIlya Dryomov if (version >= 5) { 2759fe5da05eSIlya Dryomov ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); 2760fe5da05eSIlya Dryomov memcpy(&m->replay_version, p, sizeof(m->replay_version)); 2761fe5da05eSIlya Dryomov p += sizeof(m->replay_version); 2762fe5da05eSIlya Dryomov ceph_decode_64_safe(&p, end, m->user_version, e_inval); 2763fe5da05eSIlya Dryomov } else { 2764fe5da05eSIlya Dryomov m->replay_version = bad_replay_version; /* struct */ 2765fe5da05eSIlya Dryomov m->user_version = le64_to_cpu(m->replay_version.version); 27661b83bef2SSage Weil } 27671b83bef2SSage Weil 2768fe5da05eSIlya Dryomov if (version >= 6) { 2769fe5da05eSIlya Dryomov if (version >= 7) 2770fe5da05eSIlya Dryomov ceph_decode_8_safe(&p, end, decode_redir, e_inval); 2771b0b31a8fSIlya Dryomov else 2772b0b31a8fSIlya Dryomov decode_redir = 1; 2773b0b31a8fSIlya Dryomov } else { 2774b0b31a8fSIlya Dryomov decode_redir = 0; 2775b0b31a8fSIlya Dryomov } 2776b0b31a8fSIlya Dryomov 2777b0b31a8fSIlya Dryomov if (decode_redir) { 2778fe5da05eSIlya Dryomov ret = ceph_redirect_decode(&p, end, &m->redirect); 2779fe5da05eSIlya Dryomov if (ret) 2780fe5da05eSIlya Dryomov return ret; 2781205ee118SIlya Dryomov } else { 2782fe5da05eSIlya Dryomov ceph_oloc_init(&m->redirect.oloc); 2783205ee118SIlya Dryomov } 2784205ee118SIlya Dryomov 2785fe5da05eSIlya Dryomov return 0; 2786205ee118SIlya Dryomov 2787fe5da05eSIlya Dryomov e_inval: 2788fe5da05eSIlya Dryomov return -EINVAL; 2789fe5da05eSIlya Dryomov } 2790fe5da05eSIlya Dryomov 2791fe5da05eSIlya Dryomov /* 2792fe5da05eSIlya Dryomov * We are done with @req if 2793fe5da05eSIlya Dryomov * - @m is a safe reply, or 2794fe5da05eSIlya Dryomov * - @m is an unsafe reply and we didn't want a safe one 2795fe5da05eSIlya Dryomov */ 2796fe5da05eSIlya Dryomov static bool done_request(const struct ceph_osd_request *req, 2797fe5da05eSIlya Dryomov const struct MOSDOpReply *m) 2798fe5da05eSIlya Dryomov { 2799fe5da05eSIlya Dryomov return (m->result < 0 || 2800fe5da05eSIlya Dryomov (m->flags & CEPH_OSD_FLAG_ONDISK) || 2801fe5da05eSIlya Dryomov !(req->r_flags & CEPH_OSD_FLAG_ONDISK)); 2802fe5da05eSIlya Dryomov } 2803fe5da05eSIlya Dryomov 2804fe5da05eSIlya Dryomov /* 2805fe5da05eSIlya Dryomov * handle osd op reply. either call the callback if it is specified, 2806fe5da05eSIlya Dryomov * or do the completion to wake up the waiting thread. 2807fe5da05eSIlya Dryomov * 2808fe5da05eSIlya Dryomov * ->r_unsafe_callback is set? yes no 2809fe5da05eSIlya Dryomov * 2810fe5da05eSIlya Dryomov * first reply is OK (needed r_cb/r_completion, r_cb/r_completion, 2811fe5da05eSIlya Dryomov * any or needed/got safe) r_safe_completion r_safe_completion 2812fe5da05eSIlya Dryomov * 2813fe5da05eSIlya Dryomov * first reply is unsafe r_unsafe_cb(true) (nothing) 2814fe5da05eSIlya Dryomov * 2815fe5da05eSIlya Dryomov * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion, 2816fe5da05eSIlya Dryomov * r_safe_completion r_safe_completion 2817fe5da05eSIlya Dryomov */ 28185aea3dcdSIlya Dryomov static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) 2819fe5da05eSIlya Dryomov { 28205aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 2821fe5da05eSIlya Dryomov struct ceph_osd_request *req; 2822fe5da05eSIlya Dryomov struct MOSDOpReply m; 2823fe5da05eSIlya Dryomov u64 tid = le64_to_cpu(msg->hdr.tid); 2824fe5da05eSIlya Dryomov u32 data_len = 0; 2825fe5da05eSIlya Dryomov bool already_acked; 2826fe5da05eSIlya Dryomov int ret; 2827fe5da05eSIlya Dryomov int i; 2828fe5da05eSIlya Dryomov 2829fe5da05eSIlya Dryomov dout("%s msg %p tid %llu\n", __func__, msg, tid); 2830fe5da05eSIlya Dryomov 28315aea3dcdSIlya Dryomov down_read(&osdc->lock); 28325aea3dcdSIlya Dryomov if (!osd_registered(osd)) { 28335aea3dcdSIlya Dryomov dout("%s osd%d unknown\n", __func__, osd->o_osd); 28345aea3dcdSIlya Dryomov goto out_unlock_osdc; 2835fe5da05eSIlya Dryomov } 28365aea3dcdSIlya Dryomov WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); 28375aea3dcdSIlya Dryomov 28385aea3dcdSIlya Dryomov mutex_lock(&osd->lock); 28395aea3dcdSIlya Dryomov req = lookup_request(&osd->o_requests, tid); 28405aea3dcdSIlya Dryomov if (!req) { 28415aea3dcdSIlya Dryomov dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); 28425aea3dcdSIlya Dryomov goto out_unlock_session; 28435aea3dcdSIlya Dryomov } 2844fe5da05eSIlya Dryomov 2845cd08e0a2SYan, Zheng m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; 2846fe5da05eSIlya Dryomov ret = decode_MOSDOpReply(msg, &m); 2847cd08e0a2SYan, Zheng m.redirect.oloc.pool_ns = NULL; 2848fe5da05eSIlya Dryomov if (ret) { 2849fe5da05eSIlya Dryomov pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", 2850fe5da05eSIlya Dryomov req->r_tid, ret); 2851fe5da05eSIlya Dryomov ceph_msg_dump(msg); 2852fe5da05eSIlya Dryomov goto fail_request; 2853fe5da05eSIlya Dryomov } 2854fe5da05eSIlya Dryomov dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", 2855fe5da05eSIlya Dryomov __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, 2856fe5da05eSIlya Dryomov m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), 2857fe5da05eSIlya Dryomov le64_to_cpu(m.replay_version.version), m.user_version); 2858fe5da05eSIlya Dryomov 2859fe5da05eSIlya Dryomov if (m.retry_attempt >= 0) { 2860fe5da05eSIlya Dryomov if (m.retry_attempt != req->r_attempts - 1) { 2861fe5da05eSIlya Dryomov dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", 2862fe5da05eSIlya Dryomov req, req->r_tid, m.retry_attempt, 2863fe5da05eSIlya Dryomov req->r_attempts - 1); 28645aea3dcdSIlya Dryomov goto out_unlock_session; 2865fe5da05eSIlya Dryomov } 2866fe5da05eSIlya Dryomov } else { 2867fe5da05eSIlya Dryomov WARN_ON(1); /* MOSDOpReply v4 is assumed */ 2868fe5da05eSIlya Dryomov } 2869fe5da05eSIlya Dryomov 2870fe5da05eSIlya Dryomov if (!ceph_oloc_empty(&m.redirect.oloc)) { 2871fe5da05eSIlya Dryomov dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, 2872fe5da05eSIlya Dryomov m.redirect.oloc.pool); 28735aea3dcdSIlya Dryomov unlink_request(osd, req); 28745aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 2875205ee118SIlya Dryomov 287630c156d9SYan, Zheng /* 287730c156d9SYan, Zheng * Not ceph_oloc_copy() - changing pool_ns is not 287830c156d9SYan, Zheng * supported. 287930c156d9SYan, Zheng */ 288030c156d9SYan, Zheng req->r_t.target_oloc.pool = m.redirect.oloc.pool; 28815aea3dcdSIlya Dryomov req->r_flags |= CEPH_OSD_FLAG_REDIRECTED; 28825aea3dcdSIlya Dryomov req->r_tid = 0; 28835aea3dcdSIlya Dryomov __submit_request(req, false); 28845aea3dcdSIlya Dryomov goto out_unlock_osdc; 2885205ee118SIlya Dryomov } 2886205ee118SIlya Dryomov 2887fe5da05eSIlya Dryomov if (m.num_ops != req->r_num_ops) { 2888fe5da05eSIlya Dryomov pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, 2889fe5da05eSIlya Dryomov req->r_num_ops, req->r_tid); 2890fe5da05eSIlya Dryomov goto fail_request; 2891fe5da05eSIlya Dryomov } 2892fe5da05eSIlya Dryomov for (i = 0; i < req->r_num_ops; i++) { 2893fe5da05eSIlya Dryomov dout(" req %p tid %llu op %d rval %d len %u\n", req, 2894fe5da05eSIlya Dryomov req->r_tid, i, m.rval[i], m.outdata_len[i]); 2895fe5da05eSIlya Dryomov req->r_ops[i].rval = m.rval[i]; 2896fe5da05eSIlya Dryomov req->r_ops[i].outdata_len = m.outdata_len[i]; 2897fe5da05eSIlya Dryomov data_len += m.outdata_len[i]; 2898fe5da05eSIlya Dryomov } 2899fe5da05eSIlya Dryomov if (data_len != le32_to_cpu(msg->hdr.data_len)) { 2900fe5da05eSIlya Dryomov pr_err("sum of lens %u != %u for tid %llu\n", data_len, 2901fe5da05eSIlya Dryomov le32_to_cpu(msg->hdr.data_len), req->r_tid); 2902fe5da05eSIlya Dryomov goto fail_request; 2903fe5da05eSIlya Dryomov } 2904fe5da05eSIlya Dryomov dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__, 2905fe5da05eSIlya Dryomov req, req->r_tid, req->r_got_reply, m.result, data_len); 29063d14c5d2SYehuda Sadeh 2907fe5da05eSIlya Dryomov already_acked = req->r_got_reply; 2908fe5da05eSIlya Dryomov if (!already_acked) { 2909fe5da05eSIlya Dryomov req->r_result = m.result ?: data_len; 2910fe5da05eSIlya Dryomov req->r_replay_version = m.replay_version; /* struct */ 2911fe5da05eSIlya Dryomov req->r_got_reply = true; 2912fe5da05eSIlya Dryomov } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) { 2913fe5da05eSIlya Dryomov dout("req %p tid %llu dup ack\n", req, req->r_tid); 29145aea3dcdSIlya Dryomov goto out_unlock_session; 29153d14c5d2SYehuda Sadeh } 29163d14c5d2SYehuda Sadeh 2917fe5da05eSIlya Dryomov if (done_request(req, &m)) { 29185aea3dcdSIlya Dryomov __finish_request(req); 2919fe5da05eSIlya Dryomov if (req->r_linger) { 2920fe5da05eSIlya Dryomov WARN_ON(req->r_unsafe_callback); 2921922dab61SIlya Dryomov dout("req %p tid %llu cb (locked)\n", req, req->r_tid); 2922922dab61SIlya Dryomov __complete_request(req); 2923fe5da05eSIlya Dryomov } 2924fe5da05eSIlya Dryomov } 29253d14c5d2SYehuda Sadeh 29265aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 29275aea3dcdSIlya Dryomov up_read(&osdc->lock); 29283d14c5d2SYehuda Sadeh 2929fe5da05eSIlya Dryomov if (done_request(req, &m)) { 2930fe5da05eSIlya Dryomov if (already_acked && req->r_unsafe_callback) { 2931fe5da05eSIlya Dryomov dout("req %p tid %llu safe-cb\n", req, req->r_tid); 293261c5d6bfSYan, Zheng req->r_unsafe_callback(req, false); 2933922dab61SIlya Dryomov } else if (!req->r_linger) { 2934fe5da05eSIlya Dryomov dout("req %p tid %llu cb\n", req, req->r_tid); 2935fe5da05eSIlya Dryomov __complete_request(req); 293661c5d6bfSYan, Zheng } 2937dc045a91SIlya Dryomov if (m.flags & CEPH_OSD_FLAG_ONDISK) 2938dc045a91SIlya Dryomov complete_all(&req->r_safe_completion); 2939dc045a91SIlya Dryomov ceph_osdc_put_request(req); 2940fe5da05eSIlya Dryomov } else { 2941fe5da05eSIlya Dryomov if (req->r_unsafe_callback) { 2942fe5da05eSIlya Dryomov dout("req %p tid %llu unsafe-cb\n", req, req->r_tid); 2943fe5da05eSIlya Dryomov req->r_unsafe_callback(req, true); 2944fe5da05eSIlya Dryomov } else { 2945fe5da05eSIlya Dryomov WARN_ON(1); 2946fe5da05eSIlya Dryomov } 2947fe5da05eSIlya Dryomov } 29483d14c5d2SYehuda Sadeh 29493d14c5d2SYehuda Sadeh return; 2950fe5da05eSIlya Dryomov 2951fe5da05eSIlya Dryomov fail_request: 29524609245eSIlya Dryomov complete_request(req, -EIO); 29535aea3dcdSIlya Dryomov out_unlock_session: 29545aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 29555aea3dcdSIlya Dryomov out_unlock_osdc: 29565aea3dcdSIlya Dryomov up_read(&osdc->lock); 29573d14c5d2SYehuda Sadeh } 29583d14c5d2SYehuda Sadeh 295942c1b124SIlya Dryomov static void set_pool_was_full(struct ceph_osd_client *osdc) 296042c1b124SIlya Dryomov { 296142c1b124SIlya Dryomov struct rb_node *n; 296242c1b124SIlya Dryomov 296342c1b124SIlya Dryomov for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { 296442c1b124SIlya Dryomov struct ceph_pg_pool_info *pi = 296542c1b124SIlya Dryomov rb_entry(n, struct ceph_pg_pool_info, node); 296642c1b124SIlya Dryomov 296742c1b124SIlya Dryomov pi->was_full = __pool_full(pi); 296842c1b124SIlya Dryomov } 296942c1b124SIlya Dryomov } 297042c1b124SIlya Dryomov 29715aea3dcdSIlya Dryomov static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) 29723d14c5d2SYehuda Sadeh { 29735aea3dcdSIlya Dryomov struct ceph_pg_pool_info *pi; 29743d14c5d2SYehuda Sadeh 29755aea3dcdSIlya Dryomov pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); 29765aea3dcdSIlya Dryomov if (!pi) 29775aea3dcdSIlya Dryomov return false; 29783d14c5d2SYehuda Sadeh 29795aea3dcdSIlya Dryomov return pi->was_full && !__pool_full(pi); 29803d14c5d2SYehuda Sadeh } 29813d14c5d2SYehuda Sadeh 2982922dab61SIlya Dryomov static enum calc_target_result 2983922dab61SIlya Dryomov recalc_linger_target(struct ceph_osd_linger_request *lreq) 2984922dab61SIlya Dryomov { 2985922dab61SIlya Dryomov struct ceph_osd_client *osdc = lreq->osdc; 2986922dab61SIlya Dryomov enum calc_target_result ct_res; 2987922dab61SIlya Dryomov 2988922dab61SIlya Dryomov ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true); 2989922dab61SIlya Dryomov if (ct_res == CALC_TARGET_NEED_RESEND) { 2990922dab61SIlya Dryomov struct ceph_osd *osd; 2991922dab61SIlya Dryomov 2992922dab61SIlya Dryomov osd = lookup_create_osd(osdc, lreq->t.osd, true); 2993922dab61SIlya Dryomov if (osd != lreq->osd) { 2994922dab61SIlya Dryomov unlink_linger(lreq->osd, lreq); 2995922dab61SIlya Dryomov link_linger(osd, lreq); 2996922dab61SIlya Dryomov } 2997922dab61SIlya Dryomov } 2998922dab61SIlya Dryomov 2999922dab61SIlya Dryomov return ct_res; 3000922dab61SIlya Dryomov } 3001922dab61SIlya Dryomov 30023d14c5d2SYehuda Sadeh /* 30035aea3dcdSIlya Dryomov * Requeue requests whose mapping to an OSD has changed. 30043d14c5d2SYehuda Sadeh */ 30055aea3dcdSIlya Dryomov static void scan_requests(struct ceph_osd *osd, 30065aea3dcdSIlya Dryomov bool force_resend, 30075aea3dcdSIlya Dryomov bool cleared_full, 30085aea3dcdSIlya Dryomov bool check_pool_cleared_full, 30095aea3dcdSIlya Dryomov struct rb_root *need_resend, 30105aea3dcdSIlya Dryomov struct list_head *need_resend_linger) 30113d14c5d2SYehuda Sadeh { 30125aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 30135aea3dcdSIlya Dryomov struct rb_node *n; 30145aea3dcdSIlya Dryomov bool force_resend_writes; 30153d14c5d2SYehuda Sadeh 3016922dab61SIlya Dryomov for (n = rb_first(&osd->o_linger_requests); n; ) { 3017922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = 3018922dab61SIlya Dryomov rb_entry(n, struct ceph_osd_linger_request, node); 3019922dab61SIlya Dryomov enum calc_target_result ct_res; 3020922dab61SIlya Dryomov 3021922dab61SIlya Dryomov n = rb_next(n); /* recalc_linger_target() */ 3022922dab61SIlya Dryomov 3023922dab61SIlya Dryomov dout("%s lreq %p linger_id %llu\n", __func__, lreq, 3024922dab61SIlya Dryomov lreq->linger_id); 3025922dab61SIlya Dryomov ct_res = recalc_linger_target(lreq); 3026922dab61SIlya Dryomov switch (ct_res) { 3027922dab61SIlya Dryomov case CALC_TARGET_NO_ACTION: 3028922dab61SIlya Dryomov force_resend_writes = cleared_full || 3029922dab61SIlya Dryomov (check_pool_cleared_full && 3030922dab61SIlya Dryomov pool_cleared_full(osdc, lreq->t.base_oloc.pool)); 3031922dab61SIlya Dryomov if (!force_resend && !force_resend_writes) 3032922dab61SIlya Dryomov break; 3033922dab61SIlya Dryomov 3034922dab61SIlya Dryomov /* fall through */ 3035922dab61SIlya Dryomov case CALC_TARGET_NEED_RESEND: 30364609245eSIlya Dryomov cancel_linger_map_check(lreq); 3037922dab61SIlya Dryomov /* 3038922dab61SIlya Dryomov * scan_requests() for the previous epoch(s) 3039922dab61SIlya Dryomov * may have already added it to the list, since 3040922dab61SIlya Dryomov * it's not unlinked here. 3041922dab61SIlya Dryomov */ 3042922dab61SIlya Dryomov if (list_empty(&lreq->scan_item)) 3043922dab61SIlya Dryomov list_add_tail(&lreq->scan_item, need_resend_linger); 3044922dab61SIlya Dryomov break; 3045922dab61SIlya Dryomov case CALC_TARGET_POOL_DNE: 30464609245eSIlya Dryomov check_linger_pool_dne(lreq); 3047922dab61SIlya Dryomov break; 3048922dab61SIlya Dryomov } 3049922dab61SIlya Dryomov } 3050922dab61SIlya Dryomov 30515aea3dcdSIlya Dryomov for (n = rb_first(&osd->o_requests); n; ) { 30525aea3dcdSIlya Dryomov struct ceph_osd_request *req = 30535aea3dcdSIlya Dryomov rb_entry(n, struct ceph_osd_request, r_node); 30545aea3dcdSIlya Dryomov enum calc_target_result ct_res; 3055ab60b16dSAlex Elder 30564609245eSIlya Dryomov n = rb_next(n); /* unlink_request(), check_pool_dne() */ 3057ab60b16dSAlex Elder 30585aea3dcdSIlya Dryomov dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 30595aea3dcdSIlya Dryomov ct_res = calc_target(osdc, &req->r_t, 30605aea3dcdSIlya Dryomov &req->r_last_force_resend, false); 30615aea3dcdSIlya Dryomov switch (ct_res) { 30625aea3dcdSIlya Dryomov case CALC_TARGET_NO_ACTION: 30635aea3dcdSIlya Dryomov force_resend_writes = cleared_full || 30645aea3dcdSIlya Dryomov (check_pool_cleared_full && 30655aea3dcdSIlya Dryomov pool_cleared_full(osdc, req->r_t.base_oloc.pool)); 30665aea3dcdSIlya Dryomov if (!force_resend && 30675aea3dcdSIlya Dryomov (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || 30685aea3dcdSIlya Dryomov !force_resend_writes)) 30695aea3dcdSIlya Dryomov break; 3070a40c4f10SYehuda Sadeh 30715aea3dcdSIlya Dryomov /* fall through */ 30725aea3dcdSIlya Dryomov case CALC_TARGET_NEED_RESEND: 30734609245eSIlya Dryomov cancel_map_check(req); 30745aea3dcdSIlya Dryomov unlink_request(osd, req); 30755aea3dcdSIlya Dryomov insert_request(need_resend, req); 30765aea3dcdSIlya Dryomov break; 30775aea3dcdSIlya Dryomov case CALC_TARGET_POOL_DNE: 30784609245eSIlya Dryomov check_pool_dne(req); 30795aea3dcdSIlya Dryomov break; 3080a40c4f10SYehuda Sadeh } 30813d14c5d2SYehuda Sadeh } 30823d14c5d2SYehuda Sadeh } 30836f6c7006SSage Weil 308442c1b124SIlya Dryomov static int handle_one_map(struct ceph_osd_client *osdc, 30855aea3dcdSIlya Dryomov void *p, void *end, bool incremental, 30865aea3dcdSIlya Dryomov struct rb_root *need_resend, 30875aea3dcdSIlya Dryomov struct list_head *need_resend_linger) 308842c1b124SIlya Dryomov { 308942c1b124SIlya Dryomov struct ceph_osdmap *newmap; 309042c1b124SIlya Dryomov struct rb_node *n; 309142c1b124SIlya Dryomov bool skipped_map = false; 309242c1b124SIlya Dryomov bool was_full; 309342c1b124SIlya Dryomov 3094b7ec35b3SIlya Dryomov was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 309542c1b124SIlya Dryomov set_pool_was_full(osdc); 309642c1b124SIlya Dryomov 309742c1b124SIlya Dryomov if (incremental) 309842c1b124SIlya Dryomov newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); 309942c1b124SIlya Dryomov else 310042c1b124SIlya Dryomov newmap = ceph_osdmap_decode(&p, end); 310142c1b124SIlya Dryomov if (IS_ERR(newmap)) 310242c1b124SIlya Dryomov return PTR_ERR(newmap); 310342c1b124SIlya Dryomov 310442c1b124SIlya Dryomov if (newmap != osdc->osdmap) { 310542c1b124SIlya Dryomov /* 310642c1b124SIlya Dryomov * Preserve ->was_full before destroying the old map. 310742c1b124SIlya Dryomov * For pools that weren't in the old map, ->was_full 310842c1b124SIlya Dryomov * should be false. 310942c1b124SIlya Dryomov */ 311042c1b124SIlya Dryomov for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { 311142c1b124SIlya Dryomov struct ceph_pg_pool_info *pi = 311242c1b124SIlya Dryomov rb_entry(n, struct ceph_pg_pool_info, node); 311342c1b124SIlya Dryomov struct ceph_pg_pool_info *old_pi; 311442c1b124SIlya Dryomov 311542c1b124SIlya Dryomov old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); 311642c1b124SIlya Dryomov if (old_pi) 311742c1b124SIlya Dryomov pi->was_full = old_pi->was_full; 311842c1b124SIlya Dryomov else 311942c1b124SIlya Dryomov WARN_ON(pi->was_full); 312042c1b124SIlya Dryomov } 312142c1b124SIlya Dryomov 312242c1b124SIlya Dryomov if (osdc->osdmap->epoch && 312342c1b124SIlya Dryomov osdc->osdmap->epoch + 1 < newmap->epoch) { 312442c1b124SIlya Dryomov WARN_ON(incremental); 312542c1b124SIlya Dryomov skipped_map = true; 312642c1b124SIlya Dryomov } 312742c1b124SIlya Dryomov 312842c1b124SIlya Dryomov ceph_osdmap_destroy(osdc->osdmap); 312942c1b124SIlya Dryomov osdc->osdmap = newmap; 313042c1b124SIlya Dryomov } 313142c1b124SIlya Dryomov 3132b7ec35b3SIlya Dryomov was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); 31335aea3dcdSIlya Dryomov scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, 31345aea3dcdSIlya Dryomov need_resend, need_resend_linger); 31355aea3dcdSIlya Dryomov 31365aea3dcdSIlya Dryomov for (n = rb_first(&osdc->osds); n; ) { 31375aea3dcdSIlya Dryomov struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 31385aea3dcdSIlya Dryomov 31395aea3dcdSIlya Dryomov n = rb_next(n); /* close_osd() */ 31405aea3dcdSIlya Dryomov 31415aea3dcdSIlya Dryomov scan_requests(osd, skipped_map, was_full, true, need_resend, 31425aea3dcdSIlya Dryomov need_resend_linger); 31435aea3dcdSIlya Dryomov if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || 31445aea3dcdSIlya Dryomov memcmp(&osd->o_con.peer_addr, 31455aea3dcdSIlya Dryomov ceph_osd_addr(osdc->osdmap, osd->o_osd), 31465aea3dcdSIlya Dryomov sizeof(struct ceph_entity_addr))) 31475aea3dcdSIlya Dryomov close_osd(osd); 31485aea3dcdSIlya Dryomov } 314942c1b124SIlya Dryomov 315042c1b124SIlya Dryomov return 0; 315142c1b124SIlya Dryomov } 31526f6c7006SSage Weil 31535aea3dcdSIlya Dryomov static void kick_requests(struct ceph_osd_client *osdc, 31545aea3dcdSIlya Dryomov struct rb_root *need_resend, 31555aea3dcdSIlya Dryomov struct list_head *need_resend_linger) 31565aea3dcdSIlya Dryomov { 3157922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq, *nlreq; 31585aea3dcdSIlya Dryomov struct rb_node *n; 31595aea3dcdSIlya Dryomov 31605aea3dcdSIlya Dryomov for (n = rb_first(need_resend); n; ) { 31615aea3dcdSIlya Dryomov struct ceph_osd_request *req = 31625aea3dcdSIlya Dryomov rb_entry(n, struct ceph_osd_request, r_node); 31635aea3dcdSIlya Dryomov struct ceph_osd *osd; 31645aea3dcdSIlya Dryomov 31655aea3dcdSIlya Dryomov n = rb_next(n); 31665aea3dcdSIlya Dryomov erase_request(need_resend, req); /* before link_request() */ 31675aea3dcdSIlya Dryomov 31685aea3dcdSIlya Dryomov WARN_ON(req->r_osd); 31695aea3dcdSIlya Dryomov calc_target(osdc, &req->r_t, NULL, false); 31705aea3dcdSIlya Dryomov osd = lookup_create_osd(osdc, req->r_t.osd, true); 31715aea3dcdSIlya Dryomov link_request(osd, req); 31725aea3dcdSIlya Dryomov if (!req->r_linger) { 31735aea3dcdSIlya Dryomov if (!osd_homeless(osd) && !req->r_t.paused) 31745aea3dcdSIlya Dryomov send_request(req); 3175922dab61SIlya Dryomov } else { 3176922dab61SIlya Dryomov cancel_linger_request(req); 31775aea3dcdSIlya Dryomov } 31785aea3dcdSIlya Dryomov } 3179922dab61SIlya Dryomov 3180922dab61SIlya Dryomov list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { 3181922dab61SIlya Dryomov if (!osd_homeless(lreq->osd)) 3182922dab61SIlya Dryomov send_linger(lreq); 3183922dab61SIlya Dryomov 3184922dab61SIlya Dryomov list_del_init(&lreq->scan_item); 3185922dab61SIlya Dryomov } 31865aea3dcdSIlya Dryomov } 31875aea3dcdSIlya Dryomov 31883d14c5d2SYehuda Sadeh /* 31893d14c5d2SYehuda Sadeh * Process updated osd map. 31903d14c5d2SYehuda Sadeh * 31913d14c5d2SYehuda Sadeh * The message contains any number of incremental and full maps, normally 31923d14c5d2SYehuda Sadeh * indicating some sort of topology change in the cluster. Kick requests 31933d14c5d2SYehuda Sadeh * off to different OSDs as needed. 31943d14c5d2SYehuda Sadeh */ 31953d14c5d2SYehuda Sadeh void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) 31963d14c5d2SYehuda Sadeh { 319742c1b124SIlya Dryomov void *p = msg->front.iov_base; 319842c1b124SIlya Dryomov void *const end = p + msg->front.iov_len; 31993d14c5d2SYehuda Sadeh u32 nr_maps, maplen; 32003d14c5d2SYehuda Sadeh u32 epoch; 32013d14c5d2SYehuda Sadeh struct ceph_fsid fsid; 32025aea3dcdSIlya Dryomov struct rb_root need_resend = RB_ROOT; 32035aea3dcdSIlya Dryomov LIST_HEAD(need_resend_linger); 320442c1b124SIlya Dryomov bool handled_incremental = false; 320542c1b124SIlya Dryomov bool was_pauserd, was_pausewr; 320642c1b124SIlya Dryomov bool pauserd, pausewr; 320742c1b124SIlya Dryomov int err; 32083d14c5d2SYehuda Sadeh 320942c1b124SIlya Dryomov dout("%s have %u\n", __func__, osdc->osdmap->epoch); 32105aea3dcdSIlya Dryomov down_write(&osdc->lock); 32113d14c5d2SYehuda Sadeh 32123d14c5d2SYehuda Sadeh /* verify fsid */ 32133d14c5d2SYehuda Sadeh ceph_decode_need(&p, end, sizeof(fsid), bad); 32143d14c5d2SYehuda Sadeh ceph_decode_copy(&p, &fsid, sizeof(fsid)); 32153d14c5d2SYehuda Sadeh if (ceph_check_fsid(osdc->client, &fsid) < 0) 321642c1b124SIlya Dryomov goto bad; 32173d14c5d2SYehuda Sadeh 3218b7ec35b3SIlya Dryomov was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3219b7ec35b3SIlya Dryomov was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3220b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 322142c1b124SIlya Dryomov have_pool_full(osdc); 32229a1ea2dbSJosh Durgin 32233d14c5d2SYehuda Sadeh /* incremental maps */ 32243d14c5d2SYehuda Sadeh ceph_decode_32_safe(&p, end, nr_maps, bad); 32253d14c5d2SYehuda Sadeh dout(" %d inc maps\n", nr_maps); 32263d14c5d2SYehuda Sadeh while (nr_maps > 0) { 32273d14c5d2SYehuda Sadeh ceph_decode_need(&p, end, 2*sizeof(u32), bad); 32283d14c5d2SYehuda Sadeh epoch = ceph_decode_32(&p); 32293d14c5d2SYehuda Sadeh maplen = ceph_decode_32(&p); 32303d14c5d2SYehuda Sadeh ceph_decode_need(&p, end, maplen, bad); 323142c1b124SIlya Dryomov if (osdc->osdmap->epoch && 323242c1b124SIlya Dryomov osdc->osdmap->epoch + 1 == epoch) { 32333d14c5d2SYehuda Sadeh dout("applying incremental map %u len %d\n", 32343d14c5d2SYehuda Sadeh epoch, maplen); 32355aea3dcdSIlya Dryomov err = handle_one_map(osdc, p, p + maplen, true, 32365aea3dcdSIlya Dryomov &need_resend, &need_resend_linger); 323742c1b124SIlya Dryomov if (err) 32383d14c5d2SYehuda Sadeh goto bad; 323942c1b124SIlya Dryomov handled_incremental = true; 32403d14c5d2SYehuda Sadeh } else { 32413d14c5d2SYehuda Sadeh dout("ignoring incremental map %u len %d\n", 32423d14c5d2SYehuda Sadeh epoch, maplen); 32433d14c5d2SYehuda Sadeh } 324442c1b124SIlya Dryomov p += maplen; 32453d14c5d2SYehuda Sadeh nr_maps--; 32463d14c5d2SYehuda Sadeh } 324742c1b124SIlya Dryomov if (handled_incremental) 32483d14c5d2SYehuda Sadeh goto done; 32493d14c5d2SYehuda Sadeh 32503d14c5d2SYehuda Sadeh /* full maps */ 32513d14c5d2SYehuda Sadeh ceph_decode_32_safe(&p, end, nr_maps, bad); 32523d14c5d2SYehuda Sadeh dout(" %d full maps\n", nr_maps); 32533d14c5d2SYehuda Sadeh while (nr_maps) { 32543d14c5d2SYehuda Sadeh ceph_decode_need(&p, end, 2*sizeof(u32), bad); 32553d14c5d2SYehuda Sadeh epoch = ceph_decode_32(&p); 32563d14c5d2SYehuda Sadeh maplen = ceph_decode_32(&p); 32573d14c5d2SYehuda Sadeh ceph_decode_need(&p, end, maplen, bad); 32583d14c5d2SYehuda Sadeh if (nr_maps > 1) { 32593d14c5d2SYehuda Sadeh dout("skipping non-latest full map %u len %d\n", 32603d14c5d2SYehuda Sadeh epoch, maplen); 3261e5253a7bSIlya Dryomov } else if (osdc->osdmap->epoch >= epoch) { 32623d14c5d2SYehuda Sadeh dout("skipping full map %u len %d, " 32633d14c5d2SYehuda Sadeh "older than our %u\n", epoch, maplen, 32643d14c5d2SYehuda Sadeh osdc->osdmap->epoch); 32653d14c5d2SYehuda Sadeh } else { 32663d14c5d2SYehuda Sadeh dout("taking full map %u len %d\n", epoch, maplen); 32675aea3dcdSIlya Dryomov err = handle_one_map(osdc, p, p + maplen, false, 32685aea3dcdSIlya Dryomov &need_resend, &need_resend_linger); 326942c1b124SIlya Dryomov if (err) 32703d14c5d2SYehuda Sadeh goto bad; 32713d14c5d2SYehuda Sadeh } 32723d14c5d2SYehuda Sadeh p += maplen; 32733d14c5d2SYehuda Sadeh nr_maps--; 32743d14c5d2SYehuda Sadeh } 32753d14c5d2SYehuda Sadeh 32763d14c5d2SYehuda Sadeh done: 3277cd634fb6SSage Weil /* 3278cd634fb6SSage Weil * subscribe to subsequent osdmap updates if full to ensure 3279cd634fb6SSage Weil * we find out when we are no longer full and stop returning 3280cd634fb6SSage Weil * ENOSPC. 3281cd634fb6SSage Weil */ 3282b7ec35b3SIlya Dryomov pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); 3283b7ec35b3SIlya Dryomov pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || 3284b7ec35b3SIlya Dryomov ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || 328542c1b124SIlya Dryomov have_pool_full(osdc); 328642c1b124SIlya Dryomov if (was_pauserd || was_pausewr || pauserd || pausewr) 328742c1b124SIlya Dryomov maybe_request_map(osdc); 3288cd634fb6SSage Weil 32895aea3dcdSIlya Dryomov kick_requests(osdc, &need_resend, &need_resend_linger); 329042c1b124SIlya Dryomov 329142c1b124SIlya Dryomov ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, 329242c1b124SIlya Dryomov osdc->osdmap->epoch); 32935aea3dcdSIlya Dryomov up_write(&osdc->lock); 32943d14c5d2SYehuda Sadeh wake_up_all(&osdc->client->auth_wq); 32953d14c5d2SYehuda Sadeh return; 32963d14c5d2SYehuda Sadeh 32973d14c5d2SYehuda Sadeh bad: 32983d14c5d2SYehuda Sadeh pr_err("osdc handle_map corrupt msg\n"); 32993d14c5d2SYehuda Sadeh ceph_msg_dump(msg); 33005aea3dcdSIlya Dryomov up_write(&osdc->lock); 33015aea3dcdSIlya Dryomov } 33025aea3dcdSIlya Dryomov 33035aea3dcdSIlya Dryomov /* 33045aea3dcdSIlya Dryomov * Resubmit requests pending on the given osd. 33055aea3dcdSIlya Dryomov */ 33065aea3dcdSIlya Dryomov static void kick_osd_requests(struct ceph_osd *osd) 33075aea3dcdSIlya Dryomov { 33085aea3dcdSIlya Dryomov struct rb_node *n; 33095aea3dcdSIlya Dryomov 3310922dab61SIlya Dryomov for (n = rb_first(&osd->o_requests); n; ) { 33115aea3dcdSIlya Dryomov struct ceph_osd_request *req = 33125aea3dcdSIlya Dryomov rb_entry(n, struct ceph_osd_request, r_node); 33135aea3dcdSIlya Dryomov 3314922dab61SIlya Dryomov n = rb_next(n); /* cancel_linger_request() */ 3315922dab61SIlya Dryomov 33165aea3dcdSIlya Dryomov if (!req->r_linger) { 33175aea3dcdSIlya Dryomov if (!req->r_t.paused) 33185aea3dcdSIlya Dryomov send_request(req); 3319922dab61SIlya Dryomov } else { 3320922dab61SIlya Dryomov cancel_linger_request(req); 33215aea3dcdSIlya Dryomov } 33225aea3dcdSIlya Dryomov } 3323922dab61SIlya Dryomov for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { 3324922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq = 3325922dab61SIlya Dryomov rb_entry(n, struct ceph_osd_linger_request, node); 3326922dab61SIlya Dryomov 3327922dab61SIlya Dryomov send_linger(lreq); 3328922dab61SIlya Dryomov } 33295aea3dcdSIlya Dryomov } 33305aea3dcdSIlya Dryomov 33315aea3dcdSIlya Dryomov /* 33325aea3dcdSIlya Dryomov * If the osd connection drops, we need to resubmit all requests. 33335aea3dcdSIlya Dryomov */ 33345aea3dcdSIlya Dryomov static void osd_fault(struct ceph_connection *con) 33355aea3dcdSIlya Dryomov { 33365aea3dcdSIlya Dryomov struct ceph_osd *osd = con->private; 33375aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 33385aea3dcdSIlya Dryomov 33395aea3dcdSIlya Dryomov dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); 33405aea3dcdSIlya Dryomov 33415aea3dcdSIlya Dryomov down_write(&osdc->lock); 33425aea3dcdSIlya Dryomov if (!osd_registered(osd)) { 33435aea3dcdSIlya Dryomov dout("%s osd%d unknown\n", __func__, osd->o_osd); 33445aea3dcdSIlya Dryomov goto out_unlock; 33455aea3dcdSIlya Dryomov } 33465aea3dcdSIlya Dryomov 33475aea3dcdSIlya Dryomov if (!reopen_osd(osd)) 33485aea3dcdSIlya Dryomov kick_osd_requests(osd); 33495aea3dcdSIlya Dryomov maybe_request_map(osdc); 33505aea3dcdSIlya Dryomov 33515aea3dcdSIlya Dryomov out_unlock: 33525aea3dcdSIlya Dryomov up_write(&osdc->lock); 33533d14c5d2SYehuda Sadeh } 33543d14c5d2SYehuda Sadeh 33553d14c5d2SYehuda Sadeh /* 3356a40c4f10SYehuda Sadeh * Process osd watch notifications 3357a40c4f10SYehuda Sadeh */ 33583c663bbdSAlex Elder static void handle_watch_notify(struct ceph_osd_client *osdc, 33593c663bbdSAlex Elder struct ceph_msg *msg) 3360a40c4f10SYehuda Sadeh { 3361922dab61SIlya Dryomov void *p = msg->front.iov_base; 3362922dab61SIlya Dryomov void *const end = p + msg->front.iov_len; 3363922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq; 3364922dab61SIlya Dryomov struct linger_work *lwork; 3365922dab61SIlya Dryomov u8 proto_ver, opcode; 3366922dab61SIlya Dryomov u64 cookie, notify_id; 3367922dab61SIlya Dryomov u64 notifier_id = 0; 336819079203SIlya Dryomov s32 return_code = 0; 3369922dab61SIlya Dryomov void *payload = NULL; 3370922dab61SIlya Dryomov u32 payload_len = 0; 3371a40c4f10SYehuda Sadeh 3372a40c4f10SYehuda Sadeh ceph_decode_8_safe(&p, end, proto_ver, bad); 3373a40c4f10SYehuda Sadeh ceph_decode_8_safe(&p, end, opcode, bad); 3374a40c4f10SYehuda Sadeh ceph_decode_64_safe(&p, end, cookie, bad); 3375922dab61SIlya Dryomov p += 8; /* skip ver */ 3376a40c4f10SYehuda Sadeh ceph_decode_64_safe(&p, end, notify_id, bad); 3377a40c4f10SYehuda Sadeh 3378922dab61SIlya Dryomov if (proto_ver >= 1) { 3379922dab61SIlya Dryomov ceph_decode_32_safe(&p, end, payload_len, bad); 3380922dab61SIlya Dryomov ceph_decode_need(&p, end, payload_len, bad); 3381922dab61SIlya Dryomov payload = p; 3382922dab61SIlya Dryomov p += payload_len; 3383a40c4f10SYehuda Sadeh } 3384a40c4f10SYehuda Sadeh 3385922dab61SIlya Dryomov if (le16_to_cpu(msg->hdr.version) >= 2) 338619079203SIlya Dryomov ceph_decode_32_safe(&p, end, return_code, bad); 3387922dab61SIlya Dryomov 3388922dab61SIlya Dryomov if (le16_to_cpu(msg->hdr.version) >= 3) 3389922dab61SIlya Dryomov ceph_decode_64_safe(&p, end, notifier_id, bad); 3390922dab61SIlya Dryomov 3391922dab61SIlya Dryomov down_read(&osdc->lock); 3392922dab61SIlya Dryomov lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); 3393922dab61SIlya Dryomov if (!lreq) { 3394922dab61SIlya Dryomov dout("%s opcode %d cookie %llu dne\n", __func__, opcode, 3395922dab61SIlya Dryomov cookie); 3396922dab61SIlya Dryomov goto out_unlock_osdc; 3397922dab61SIlya Dryomov } 3398922dab61SIlya Dryomov 3399922dab61SIlya Dryomov mutex_lock(&lreq->lock); 340019079203SIlya Dryomov dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, 340119079203SIlya Dryomov opcode, cookie, lreq, lreq->is_watch); 3402922dab61SIlya Dryomov if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { 3403922dab61SIlya Dryomov if (!lreq->last_error) { 3404922dab61SIlya Dryomov lreq->last_error = -ENOTCONN; 3405922dab61SIlya Dryomov queue_watch_error(lreq); 3406922dab61SIlya Dryomov } 340719079203SIlya Dryomov } else if (!lreq->is_watch) { 340819079203SIlya Dryomov /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ 340919079203SIlya Dryomov if (lreq->notify_id && lreq->notify_id != notify_id) { 341019079203SIlya Dryomov dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, 341119079203SIlya Dryomov lreq->notify_id, notify_id); 341219079203SIlya Dryomov } else if (!completion_done(&lreq->notify_finish_wait)) { 341319079203SIlya Dryomov struct ceph_msg_data *data = 341419079203SIlya Dryomov list_first_entry_or_null(&msg->data, 341519079203SIlya Dryomov struct ceph_msg_data, 341619079203SIlya Dryomov links); 341719079203SIlya Dryomov 341819079203SIlya Dryomov if (data) { 341919079203SIlya Dryomov if (lreq->preply_pages) { 342019079203SIlya Dryomov WARN_ON(data->type != 342119079203SIlya Dryomov CEPH_MSG_DATA_PAGES); 342219079203SIlya Dryomov *lreq->preply_pages = data->pages; 342319079203SIlya Dryomov *lreq->preply_len = data->length; 342419079203SIlya Dryomov } else { 342519079203SIlya Dryomov ceph_release_page_vector(data->pages, 342619079203SIlya Dryomov calc_pages_for(0, data->length)); 342719079203SIlya Dryomov } 342819079203SIlya Dryomov } 342919079203SIlya Dryomov lreq->notify_finish_error = return_code; 343019079203SIlya Dryomov complete_all(&lreq->notify_finish_wait); 343119079203SIlya Dryomov } 3432922dab61SIlya Dryomov } else { 3433922dab61SIlya Dryomov /* CEPH_WATCH_EVENT_NOTIFY */ 3434922dab61SIlya Dryomov lwork = lwork_alloc(lreq, do_watch_notify); 3435922dab61SIlya Dryomov if (!lwork) { 3436922dab61SIlya Dryomov pr_err("failed to allocate notify-lwork\n"); 3437922dab61SIlya Dryomov goto out_unlock_lreq; 3438922dab61SIlya Dryomov } 3439922dab61SIlya Dryomov 3440922dab61SIlya Dryomov lwork->notify.notify_id = notify_id; 3441922dab61SIlya Dryomov lwork->notify.notifier_id = notifier_id; 3442922dab61SIlya Dryomov lwork->notify.payload = payload; 3443922dab61SIlya Dryomov lwork->notify.payload_len = payload_len; 3444922dab61SIlya Dryomov lwork->notify.msg = ceph_msg_get(msg); 3445922dab61SIlya Dryomov lwork_queue(lwork); 3446922dab61SIlya Dryomov } 3447922dab61SIlya Dryomov 3448922dab61SIlya Dryomov out_unlock_lreq: 3449922dab61SIlya Dryomov mutex_unlock(&lreq->lock); 3450922dab61SIlya Dryomov out_unlock_osdc: 3451922dab61SIlya Dryomov up_read(&osdc->lock); 3452a40c4f10SYehuda Sadeh return; 3453a40c4f10SYehuda Sadeh 3454a40c4f10SYehuda Sadeh bad: 3455a40c4f10SYehuda Sadeh pr_err("osdc handle_watch_notify corrupt msg\n"); 3456a40c4f10SYehuda Sadeh } 3457a40c4f10SYehuda Sadeh 3458a40c4f10SYehuda Sadeh /* 34593d14c5d2SYehuda Sadeh * Register request, send initial attempt. 34603d14c5d2SYehuda Sadeh */ 34613d14c5d2SYehuda Sadeh int ceph_osdc_start_request(struct ceph_osd_client *osdc, 34623d14c5d2SYehuda Sadeh struct ceph_osd_request *req, 34633d14c5d2SYehuda Sadeh bool nofail) 34643d14c5d2SYehuda Sadeh { 34655aea3dcdSIlya Dryomov down_read(&osdc->lock); 34665aea3dcdSIlya Dryomov submit_request(req, false); 34675aea3dcdSIlya Dryomov up_read(&osdc->lock); 34683d14c5d2SYehuda Sadeh 34695aea3dcdSIlya Dryomov return 0; 34703d14c5d2SYehuda Sadeh } 34713d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_start_request); 34723d14c5d2SYehuda Sadeh 34733d14c5d2SYehuda Sadeh /* 3474c9f9b93dSIlya Dryomov * Unregister a registered request. The request is not completed (i.e. 3475c9f9b93dSIlya Dryomov * no callbacks or wakeups) - higher layers are supposed to know what 3476c9f9b93dSIlya Dryomov * they are canceling. 3477c9f9b93dSIlya Dryomov */ 3478c9f9b93dSIlya Dryomov void ceph_osdc_cancel_request(struct ceph_osd_request *req) 3479c9f9b93dSIlya Dryomov { 3480c9f9b93dSIlya Dryomov struct ceph_osd_client *osdc = req->r_osdc; 3481c9f9b93dSIlya Dryomov 34825aea3dcdSIlya Dryomov down_write(&osdc->lock); 34835aea3dcdSIlya Dryomov if (req->r_osd) 34845aea3dcdSIlya Dryomov cancel_request(req); 34855aea3dcdSIlya Dryomov up_write(&osdc->lock); 3486c9f9b93dSIlya Dryomov } 3487c9f9b93dSIlya Dryomov EXPORT_SYMBOL(ceph_osdc_cancel_request); 3488c9f9b93dSIlya Dryomov 3489c9f9b93dSIlya Dryomov /* 349042b06965SIlya Dryomov * @timeout: in jiffies, 0 means "wait forever" 349142b06965SIlya Dryomov */ 349242b06965SIlya Dryomov static int wait_request_timeout(struct ceph_osd_request *req, 349342b06965SIlya Dryomov unsigned long timeout) 349442b06965SIlya Dryomov { 349542b06965SIlya Dryomov long left; 349642b06965SIlya Dryomov 349742b06965SIlya Dryomov dout("%s req %p tid %llu\n", __func__, req, req->r_tid); 34980e76abf2SYan, Zheng left = wait_for_completion_killable_timeout(&req->r_completion, 349942b06965SIlya Dryomov ceph_timeout_jiffies(timeout)); 350042b06965SIlya Dryomov if (left <= 0) { 350142b06965SIlya Dryomov left = left ?: -ETIMEDOUT; 350242b06965SIlya Dryomov ceph_osdc_cancel_request(req); 350342b06965SIlya Dryomov 350442b06965SIlya Dryomov /* kludge - need to to wake ceph_osdc_sync() */ 350542b06965SIlya Dryomov complete_all(&req->r_safe_completion); 350642b06965SIlya Dryomov } else { 350742b06965SIlya Dryomov left = req->r_result; /* completed */ 350842b06965SIlya Dryomov } 350942b06965SIlya Dryomov 351042b06965SIlya Dryomov return left; 351142b06965SIlya Dryomov } 351242b06965SIlya Dryomov 351342b06965SIlya Dryomov /* 35143d14c5d2SYehuda Sadeh * wait for a request to complete 35153d14c5d2SYehuda Sadeh */ 35163d14c5d2SYehuda Sadeh int ceph_osdc_wait_request(struct ceph_osd_client *osdc, 35173d14c5d2SYehuda Sadeh struct ceph_osd_request *req) 35183d14c5d2SYehuda Sadeh { 351942b06965SIlya Dryomov return wait_request_timeout(req, 0); 35203d14c5d2SYehuda Sadeh } 35213d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_wait_request); 35223d14c5d2SYehuda Sadeh 35233d14c5d2SYehuda Sadeh /* 35243d14c5d2SYehuda Sadeh * sync - wait for all in-flight requests to flush. avoid starvation. 35253d14c5d2SYehuda Sadeh */ 35263d14c5d2SYehuda Sadeh void ceph_osdc_sync(struct ceph_osd_client *osdc) 35273d14c5d2SYehuda Sadeh { 35285aea3dcdSIlya Dryomov struct rb_node *n, *p; 35295aea3dcdSIlya Dryomov u64 last_tid = atomic64_read(&osdc->last_tid); 35303d14c5d2SYehuda Sadeh 35315aea3dcdSIlya Dryomov again: 35325aea3dcdSIlya Dryomov down_read(&osdc->lock); 35335aea3dcdSIlya Dryomov for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { 35345aea3dcdSIlya Dryomov struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); 35355aea3dcdSIlya Dryomov 35365aea3dcdSIlya Dryomov mutex_lock(&osd->lock); 35375aea3dcdSIlya Dryomov for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { 35385aea3dcdSIlya Dryomov struct ceph_osd_request *req = 35395aea3dcdSIlya Dryomov rb_entry(p, struct ceph_osd_request, r_node); 35405aea3dcdSIlya Dryomov 35413d14c5d2SYehuda Sadeh if (req->r_tid > last_tid) 35423d14c5d2SYehuda Sadeh break; 35433d14c5d2SYehuda Sadeh 35445aea3dcdSIlya Dryomov if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) 35453d14c5d2SYehuda Sadeh continue; 35463d14c5d2SYehuda Sadeh 35473d14c5d2SYehuda Sadeh ceph_osdc_get_request(req); 35485aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 35495aea3dcdSIlya Dryomov up_read(&osdc->lock); 35505aea3dcdSIlya Dryomov dout("%s waiting on req %p tid %llu last_tid %llu\n", 35515aea3dcdSIlya Dryomov __func__, req, req->r_tid, last_tid); 35523d14c5d2SYehuda Sadeh wait_for_completion(&req->r_safe_completion); 35533d14c5d2SYehuda Sadeh ceph_osdc_put_request(req); 35545aea3dcdSIlya Dryomov goto again; 35553d14c5d2SYehuda Sadeh } 35565aea3dcdSIlya Dryomov 35575aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 35585aea3dcdSIlya Dryomov } 35595aea3dcdSIlya Dryomov 35605aea3dcdSIlya Dryomov up_read(&osdc->lock); 35615aea3dcdSIlya Dryomov dout("%s done last_tid %llu\n", __func__, last_tid); 35623d14c5d2SYehuda Sadeh } 35633d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_sync); 35643d14c5d2SYehuda Sadeh 3565922dab61SIlya Dryomov static struct ceph_osd_request * 3566922dab61SIlya Dryomov alloc_linger_request(struct ceph_osd_linger_request *lreq) 3567922dab61SIlya Dryomov { 3568922dab61SIlya Dryomov struct ceph_osd_request *req; 3569922dab61SIlya Dryomov 3570922dab61SIlya Dryomov req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO); 3571922dab61SIlya Dryomov if (!req) 3572922dab61SIlya Dryomov return NULL; 3573922dab61SIlya Dryomov 3574922dab61SIlya Dryomov ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 3575922dab61SIlya Dryomov ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 3576922dab61SIlya Dryomov 3577922dab61SIlya Dryomov if (ceph_osdc_alloc_messages(req, GFP_NOIO)) { 3578922dab61SIlya Dryomov ceph_osdc_put_request(req); 3579922dab61SIlya Dryomov return NULL; 3580922dab61SIlya Dryomov } 3581922dab61SIlya Dryomov 3582922dab61SIlya Dryomov return req; 3583922dab61SIlya Dryomov } 3584922dab61SIlya Dryomov 3585922dab61SIlya Dryomov /* 3586922dab61SIlya Dryomov * Returns a handle, caller owns a ref. 3587922dab61SIlya Dryomov */ 3588922dab61SIlya Dryomov struct ceph_osd_linger_request * 3589922dab61SIlya Dryomov ceph_osdc_watch(struct ceph_osd_client *osdc, 3590922dab61SIlya Dryomov struct ceph_object_id *oid, 3591922dab61SIlya Dryomov struct ceph_object_locator *oloc, 3592922dab61SIlya Dryomov rados_watchcb2_t wcb, 3593922dab61SIlya Dryomov rados_watcherrcb_t errcb, 3594922dab61SIlya Dryomov void *data) 3595922dab61SIlya Dryomov { 3596922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq; 3597922dab61SIlya Dryomov int ret; 3598922dab61SIlya Dryomov 3599922dab61SIlya Dryomov lreq = linger_alloc(osdc); 3600922dab61SIlya Dryomov if (!lreq) 3601922dab61SIlya Dryomov return ERR_PTR(-ENOMEM); 3602922dab61SIlya Dryomov 360319079203SIlya Dryomov lreq->is_watch = true; 3604922dab61SIlya Dryomov lreq->wcb = wcb; 3605922dab61SIlya Dryomov lreq->errcb = errcb; 3606922dab61SIlya Dryomov lreq->data = data; 3607b07d3c4bSIlya Dryomov lreq->watch_valid_thru = jiffies; 3608922dab61SIlya Dryomov 3609922dab61SIlya Dryomov ceph_oid_copy(&lreq->t.base_oid, oid); 3610922dab61SIlya Dryomov ceph_oloc_copy(&lreq->t.base_oloc, oloc); 3611922dab61SIlya Dryomov lreq->t.flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; 3612922dab61SIlya Dryomov lreq->mtime = CURRENT_TIME; 3613922dab61SIlya Dryomov 3614922dab61SIlya Dryomov lreq->reg_req = alloc_linger_request(lreq); 3615922dab61SIlya Dryomov if (!lreq->reg_req) { 3616922dab61SIlya Dryomov ret = -ENOMEM; 3617922dab61SIlya Dryomov goto err_put_lreq; 3618922dab61SIlya Dryomov } 3619922dab61SIlya Dryomov 3620922dab61SIlya Dryomov lreq->ping_req = alloc_linger_request(lreq); 3621922dab61SIlya Dryomov if (!lreq->ping_req) { 3622922dab61SIlya Dryomov ret = -ENOMEM; 3623922dab61SIlya Dryomov goto err_put_lreq; 3624922dab61SIlya Dryomov } 3625922dab61SIlya Dryomov 3626922dab61SIlya Dryomov down_write(&osdc->lock); 3627922dab61SIlya Dryomov linger_register(lreq); /* before osd_req_op_* */ 3628922dab61SIlya Dryomov osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id, 3629922dab61SIlya Dryomov CEPH_OSD_WATCH_OP_WATCH); 3630922dab61SIlya Dryomov osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id, 3631922dab61SIlya Dryomov CEPH_OSD_WATCH_OP_PING); 3632922dab61SIlya Dryomov linger_submit(lreq); 3633922dab61SIlya Dryomov up_write(&osdc->lock); 3634922dab61SIlya Dryomov 3635922dab61SIlya Dryomov ret = linger_reg_commit_wait(lreq); 3636922dab61SIlya Dryomov if (ret) { 3637922dab61SIlya Dryomov linger_cancel(lreq); 3638922dab61SIlya Dryomov goto err_put_lreq; 3639922dab61SIlya Dryomov } 3640922dab61SIlya Dryomov 3641922dab61SIlya Dryomov return lreq; 3642922dab61SIlya Dryomov 3643922dab61SIlya Dryomov err_put_lreq: 3644922dab61SIlya Dryomov linger_put(lreq); 3645922dab61SIlya Dryomov return ERR_PTR(ret); 3646922dab61SIlya Dryomov } 3647922dab61SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_watch); 3648922dab61SIlya Dryomov 3649922dab61SIlya Dryomov /* 3650922dab61SIlya Dryomov * Releases a ref. 3651922dab61SIlya Dryomov * 3652922dab61SIlya Dryomov * Times out after mount_timeout to preserve rbd unmap behaviour 3653922dab61SIlya Dryomov * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap 3654922dab61SIlya Dryomov * with mount_timeout"). 3655922dab61SIlya Dryomov */ 3656922dab61SIlya Dryomov int ceph_osdc_unwatch(struct ceph_osd_client *osdc, 3657922dab61SIlya Dryomov struct ceph_osd_linger_request *lreq) 3658922dab61SIlya Dryomov { 3659922dab61SIlya Dryomov struct ceph_options *opts = osdc->client->options; 3660922dab61SIlya Dryomov struct ceph_osd_request *req; 3661922dab61SIlya Dryomov int ret; 3662922dab61SIlya Dryomov 3663922dab61SIlya Dryomov req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3664922dab61SIlya Dryomov if (!req) 3665922dab61SIlya Dryomov return -ENOMEM; 3666922dab61SIlya Dryomov 3667922dab61SIlya Dryomov ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); 3668922dab61SIlya Dryomov ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); 3669922dab61SIlya Dryomov req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK; 3670922dab61SIlya Dryomov req->r_mtime = CURRENT_TIME; 3671922dab61SIlya Dryomov osd_req_op_watch_init(req, 0, lreq->linger_id, 3672922dab61SIlya Dryomov CEPH_OSD_WATCH_OP_UNWATCH); 3673922dab61SIlya Dryomov 3674922dab61SIlya Dryomov ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3675922dab61SIlya Dryomov if (ret) 3676922dab61SIlya Dryomov goto out_put_req; 3677922dab61SIlya Dryomov 3678922dab61SIlya Dryomov ceph_osdc_start_request(osdc, req, false); 3679922dab61SIlya Dryomov linger_cancel(lreq); 3680922dab61SIlya Dryomov linger_put(lreq); 3681922dab61SIlya Dryomov ret = wait_request_timeout(req, opts->mount_timeout); 3682922dab61SIlya Dryomov 3683922dab61SIlya Dryomov out_put_req: 3684922dab61SIlya Dryomov ceph_osdc_put_request(req); 3685922dab61SIlya Dryomov return ret; 3686922dab61SIlya Dryomov } 3687922dab61SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_unwatch); 3688922dab61SIlya Dryomov 3689922dab61SIlya Dryomov static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, 3690922dab61SIlya Dryomov u64 notify_id, u64 cookie, void *payload, 3691922dab61SIlya Dryomov size_t payload_len) 3692922dab61SIlya Dryomov { 3693922dab61SIlya Dryomov struct ceph_osd_req_op *op; 3694922dab61SIlya Dryomov struct ceph_pagelist *pl; 3695922dab61SIlya Dryomov int ret; 3696922dab61SIlya Dryomov 3697922dab61SIlya Dryomov op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); 3698922dab61SIlya Dryomov 3699922dab61SIlya Dryomov pl = kmalloc(sizeof(*pl), GFP_NOIO); 3700922dab61SIlya Dryomov if (!pl) 3701922dab61SIlya Dryomov return -ENOMEM; 3702922dab61SIlya Dryomov 3703922dab61SIlya Dryomov ceph_pagelist_init(pl); 3704922dab61SIlya Dryomov ret = ceph_pagelist_encode_64(pl, notify_id); 3705922dab61SIlya Dryomov ret |= ceph_pagelist_encode_64(pl, cookie); 3706922dab61SIlya Dryomov if (payload) { 3707922dab61SIlya Dryomov ret |= ceph_pagelist_encode_32(pl, payload_len); 3708922dab61SIlya Dryomov ret |= ceph_pagelist_append(pl, payload, payload_len); 3709922dab61SIlya Dryomov } else { 3710922dab61SIlya Dryomov ret |= ceph_pagelist_encode_32(pl, 0); 3711922dab61SIlya Dryomov } 3712922dab61SIlya Dryomov if (ret) { 3713922dab61SIlya Dryomov ceph_pagelist_release(pl); 3714922dab61SIlya Dryomov return -ENOMEM; 3715922dab61SIlya Dryomov } 3716922dab61SIlya Dryomov 3717922dab61SIlya Dryomov ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); 3718922dab61SIlya Dryomov op->indata_len = pl->length; 3719922dab61SIlya Dryomov return 0; 3720922dab61SIlya Dryomov } 3721922dab61SIlya Dryomov 3722922dab61SIlya Dryomov int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, 3723922dab61SIlya Dryomov struct ceph_object_id *oid, 3724922dab61SIlya Dryomov struct ceph_object_locator *oloc, 3725922dab61SIlya Dryomov u64 notify_id, 3726922dab61SIlya Dryomov u64 cookie, 3727922dab61SIlya Dryomov void *payload, 3728922dab61SIlya Dryomov size_t payload_len) 3729922dab61SIlya Dryomov { 3730922dab61SIlya Dryomov struct ceph_osd_request *req; 3731922dab61SIlya Dryomov int ret; 3732922dab61SIlya Dryomov 3733922dab61SIlya Dryomov req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3734922dab61SIlya Dryomov if (!req) 3735922dab61SIlya Dryomov return -ENOMEM; 3736922dab61SIlya Dryomov 3737922dab61SIlya Dryomov ceph_oid_copy(&req->r_base_oid, oid); 3738922dab61SIlya Dryomov ceph_oloc_copy(&req->r_base_oloc, oloc); 3739922dab61SIlya Dryomov req->r_flags = CEPH_OSD_FLAG_READ; 3740922dab61SIlya Dryomov 3741922dab61SIlya Dryomov ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3742922dab61SIlya Dryomov if (ret) 3743922dab61SIlya Dryomov goto out_put_req; 3744922dab61SIlya Dryomov 3745922dab61SIlya Dryomov ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, 3746922dab61SIlya Dryomov payload_len); 3747922dab61SIlya Dryomov if (ret) 3748922dab61SIlya Dryomov goto out_put_req; 3749922dab61SIlya Dryomov 3750922dab61SIlya Dryomov ceph_osdc_start_request(osdc, req, false); 3751922dab61SIlya Dryomov ret = ceph_osdc_wait_request(osdc, req); 3752922dab61SIlya Dryomov 3753922dab61SIlya Dryomov out_put_req: 3754922dab61SIlya Dryomov ceph_osdc_put_request(req); 3755922dab61SIlya Dryomov return ret; 3756922dab61SIlya Dryomov } 3757922dab61SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_notify_ack); 3758922dab61SIlya Dryomov 375919079203SIlya Dryomov static int osd_req_op_notify_init(struct ceph_osd_request *req, int which, 376019079203SIlya Dryomov u64 cookie, u32 prot_ver, u32 timeout, 376119079203SIlya Dryomov void *payload, size_t payload_len) 376219079203SIlya Dryomov { 376319079203SIlya Dryomov struct ceph_osd_req_op *op; 376419079203SIlya Dryomov struct ceph_pagelist *pl; 376519079203SIlya Dryomov int ret; 376619079203SIlya Dryomov 376719079203SIlya Dryomov op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); 376819079203SIlya Dryomov op->notify.cookie = cookie; 376919079203SIlya Dryomov 377019079203SIlya Dryomov pl = kmalloc(sizeof(*pl), GFP_NOIO); 377119079203SIlya Dryomov if (!pl) 377219079203SIlya Dryomov return -ENOMEM; 377319079203SIlya Dryomov 377419079203SIlya Dryomov ceph_pagelist_init(pl); 377519079203SIlya Dryomov ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */ 377619079203SIlya Dryomov ret |= ceph_pagelist_encode_32(pl, timeout); 377719079203SIlya Dryomov ret |= ceph_pagelist_encode_32(pl, payload_len); 377819079203SIlya Dryomov ret |= ceph_pagelist_append(pl, payload, payload_len); 377919079203SIlya Dryomov if (ret) { 378019079203SIlya Dryomov ceph_pagelist_release(pl); 378119079203SIlya Dryomov return -ENOMEM; 378219079203SIlya Dryomov } 378319079203SIlya Dryomov 378419079203SIlya Dryomov ceph_osd_data_pagelist_init(&op->notify.request_data, pl); 378519079203SIlya Dryomov op->indata_len = pl->length; 378619079203SIlya Dryomov return 0; 378719079203SIlya Dryomov } 378819079203SIlya Dryomov 378919079203SIlya Dryomov /* 379019079203SIlya Dryomov * @timeout: in seconds 379119079203SIlya Dryomov * 379219079203SIlya Dryomov * @preply_{pages,len} are initialized both on success and error. 379319079203SIlya Dryomov * The caller is responsible for: 379419079203SIlya Dryomov * 379519079203SIlya Dryomov * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) 379619079203SIlya Dryomov */ 379719079203SIlya Dryomov int ceph_osdc_notify(struct ceph_osd_client *osdc, 379819079203SIlya Dryomov struct ceph_object_id *oid, 379919079203SIlya Dryomov struct ceph_object_locator *oloc, 380019079203SIlya Dryomov void *payload, 380119079203SIlya Dryomov size_t payload_len, 380219079203SIlya Dryomov u32 timeout, 380319079203SIlya Dryomov struct page ***preply_pages, 380419079203SIlya Dryomov size_t *preply_len) 380519079203SIlya Dryomov { 380619079203SIlya Dryomov struct ceph_osd_linger_request *lreq; 380719079203SIlya Dryomov struct page **pages; 380819079203SIlya Dryomov int ret; 380919079203SIlya Dryomov 381019079203SIlya Dryomov WARN_ON(!timeout); 381119079203SIlya Dryomov if (preply_pages) { 381219079203SIlya Dryomov *preply_pages = NULL; 381319079203SIlya Dryomov *preply_len = 0; 381419079203SIlya Dryomov } 381519079203SIlya Dryomov 381619079203SIlya Dryomov lreq = linger_alloc(osdc); 381719079203SIlya Dryomov if (!lreq) 381819079203SIlya Dryomov return -ENOMEM; 381919079203SIlya Dryomov 382019079203SIlya Dryomov lreq->preply_pages = preply_pages; 382119079203SIlya Dryomov lreq->preply_len = preply_len; 382219079203SIlya Dryomov 382319079203SIlya Dryomov ceph_oid_copy(&lreq->t.base_oid, oid); 382419079203SIlya Dryomov ceph_oloc_copy(&lreq->t.base_oloc, oloc); 382519079203SIlya Dryomov lreq->t.flags = CEPH_OSD_FLAG_READ; 382619079203SIlya Dryomov 382719079203SIlya Dryomov lreq->reg_req = alloc_linger_request(lreq); 382819079203SIlya Dryomov if (!lreq->reg_req) { 382919079203SIlya Dryomov ret = -ENOMEM; 383019079203SIlya Dryomov goto out_put_lreq; 383119079203SIlya Dryomov } 383219079203SIlya Dryomov 383319079203SIlya Dryomov /* for notify_id */ 383419079203SIlya Dryomov pages = ceph_alloc_page_vector(1, GFP_NOIO); 383519079203SIlya Dryomov if (IS_ERR(pages)) { 383619079203SIlya Dryomov ret = PTR_ERR(pages); 383719079203SIlya Dryomov goto out_put_lreq; 383819079203SIlya Dryomov } 383919079203SIlya Dryomov 384019079203SIlya Dryomov down_write(&osdc->lock); 384119079203SIlya Dryomov linger_register(lreq); /* before osd_req_op_* */ 384219079203SIlya Dryomov ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1, 384319079203SIlya Dryomov timeout, payload, payload_len); 384419079203SIlya Dryomov if (ret) { 384519079203SIlya Dryomov linger_unregister(lreq); 384619079203SIlya Dryomov up_write(&osdc->lock); 384719079203SIlya Dryomov ceph_release_page_vector(pages, 1); 384819079203SIlya Dryomov goto out_put_lreq; 384919079203SIlya Dryomov } 385019079203SIlya Dryomov ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify, 385119079203SIlya Dryomov response_data), 385219079203SIlya Dryomov pages, PAGE_SIZE, 0, false, true); 385319079203SIlya Dryomov linger_submit(lreq); 385419079203SIlya Dryomov up_write(&osdc->lock); 385519079203SIlya Dryomov 385619079203SIlya Dryomov ret = linger_reg_commit_wait(lreq); 385719079203SIlya Dryomov if (!ret) 385819079203SIlya Dryomov ret = linger_notify_finish_wait(lreq); 385919079203SIlya Dryomov else 386019079203SIlya Dryomov dout("lreq %p failed to initiate notify %d\n", lreq, ret); 386119079203SIlya Dryomov 386219079203SIlya Dryomov linger_cancel(lreq); 386319079203SIlya Dryomov out_put_lreq: 386419079203SIlya Dryomov linger_put(lreq); 386519079203SIlya Dryomov return ret; 386619079203SIlya Dryomov } 386719079203SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_notify); 386819079203SIlya Dryomov 38693d14c5d2SYehuda Sadeh /* 3870b07d3c4bSIlya Dryomov * Return the number of milliseconds since the watch was last 3871b07d3c4bSIlya Dryomov * confirmed, or an error. If there is an error, the watch is no 3872b07d3c4bSIlya Dryomov * longer valid, and should be destroyed with ceph_osdc_unwatch(). 3873b07d3c4bSIlya Dryomov */ 3874b07d3c4bSIlya Dryomov int ceph_osdc_watch_check(struct ceph_osd_client *osdc, 3875b07d3c4bSIlya Dryomov struct ceph_osd_linger_request *lreq) 3876b07d3c4bSIlya Dryomov { 3877b07d3c4bSIlya Dryomov unsigned long stamp, age; 3878b07d3c4bSIlya Dryomov int ret; 3879b07d3c4bSIlya Dryomov 3880b07d3c4bSIlya Dryomov down_read(&osdc->lock); 3881b07d3c4bSIlya Dryomov mutex_lock(&lreq->lock); 3882b07d3c4bSIlya Dryomov stamp = lreq->watch_valid_thru; 3883b07d3c4bSIlya Dryomov if (!list_empty(&lreq->pending_lworks)) { 3884b07d3c4bSIlya Dryomov struct linger_work *lwork = 3885b07d3c4bSIlya Dryomov list_first_entry(&lreq->pending_lworks, 3886b07d3c4bSIlya Dryomov struct linger_work, 3887b07d3c4bSIlya Dryomov pending_item); 3888b07d3c4bSIlya Dryomov 3889b07d3c4bSIlya Dryomov if (time_before(lwork->queued_stamp, stamp)) 3890b07d3c4bSIlya Dryomov stamp = lwork->queued_stamp; 3891b07d3c4bSIlya Dryomov } 3892b07d3c4bSIlya Dryomov age = jiffies - stamp; 3893b07d3c4bSIlya Dryomov dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, 3894b07d3c4bSIlya Dryomov lreq, lreq->linger_id, age, lreq->last_error); 3895b07d3c4bSIlya Dryomov /* we are truncating to msecs, so return a safe upper bound */ 3896b07d3c4bSIlya Dryomov ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); 3897b07d3c4bSIlya Dryomov 3898b07d3c4bSIlya Dryomov mutex_unlock(&lreq->lock); 3899b07d3c4bSIlya Dryomov up_read(&osdc->lock); 3900b07d3c4bSIlya Dryomov return ret; 3901b07d3c4bSIlya Dryomov } 3902b07d3c4bSIlya Dryomov 3903a4ed38d7SDouglas Fuller static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) 3904a4ed38d7SDouglas Fuller { 3905a4ed38d7SDouglas Fuller u8 struct_v; 3906a4ed38d7SDouglas Fuller u32 struct_len; 3907a4ed38d7SDouglas Fuller int ret; 3908a4ed38d7SDouglas Fuller 3909a4ed38d7SDouglas Fuller ret = ceph_start_decoding(p, end, 2, "watch_item_t", 3910a4ed38d7SDouglas Fuller &struct_v, &struct_len); 3911a4ed38d7SDouglas Fuller if (ret) 3912a4ed38d7SDouglas Fuller return ret; 3913a4ed38d7SDouglas Fuller 3914a4ed38d7SDouglas Fuller ceph_decode_copy(p, &item->name, sizeof(item->name)); 3915a4ed38d7SDouglas Fuller item->cookie = ceph_decode_64(p); 3916a4ed38d7SDouglas Fuller *p += 4; /* skip timeout_seconds */ 3917a4ed38d7SDouglas Fuller if (struct_v >= 2) { 3918a4ed38d7SDouglas Fuller ceph_decode_copy(p, &item->addr, sizeof(item->addr)); 3919a4ed38d7SDouglas Fuller ceph_decode_addr(&item->addr); 3920a4ed38d7SDouglas Fuller } 3921a4ed38d7SDouglas Fuller 3922a4ed38d7SDouglas Fuller dout("%s %s%llu cookie %llu addr %s\n", __func__, 3923a4ed38d7SDouglas Fuller ENTITY_NAME(item->name), item->cookie, 3924a4ed38d7SDouglas Fuller ceph_pr_addr(&item->addr.in_addr)); 3925a4ed38d7SDouglas Fuller return 0; 3926a4ed38d7SDouglas Fuller } 3927a4ed38d7SDouglas Fuller 3928a4ed38d7SDouglas Fuller static int decode_watchers(void **p, void *end, 3929a4ed38d7SDouglas Fuller struct ceph_watch_item **watchers, 3930a4ed38d7SDouglas Fuller u32 *num_watchers) 3931a4ed38d7SDouglas Fuller { 3932a4ed38d7SDouglas Fuller u8 struct_v; 3933a4ed38d7SDouglas Fuller u32 struct_len; 3934a4ed38d7SDouglas Fuller int i; 3935a4ed38d7SDouglas Fuller int ret; 3936a4ed38d7SDouglas Fuller 3937a4ed38d7SDouglas Fuller ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", 3938a4ed38d7SDouglas Fuller &struct_v, &struct_len); 3939a4ed38d7SDouglas Fuller if (ret) 3940a4ed38d7SDouglas Fuller return ret; 3941a4ed38d7SDouglas Fuller 3942a4ed38d7SDouglas Fuller *num_watchers = ceph_decode_32(p); 3943a4ed38d7SDouglas Fuller *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); 3944a4ed38d7SDouglas Fuller if (!*watchers) 3945a4ed38d7SDouglas Fuller return -ENOMEM; 3946a4ed38d7SDouglas Fuller 3947a4ed38d7SDouglas Fuller for (i = 0; i < *num_watchers; i++) { 3948a4ed38d7SDouglas Fuller ret = decode_watcher(p, end, *watchers + i); 3949a4ed38d7SDouglas Fuller if (ret) { 3950a4ed38d7SDouglas Fuller kfree(*watchers); 3951a4ed38d7SDouglas Fuller return ret; 3952a4ed38d7SDouglas Fuller } 3953a4ed38d7SDouglas Fuller } 3954a4ed38d7SDouglas Fuller 3955a4ed38d7SDouglas Fuller return 0; 3956a4ed38d7SDouglas Fuller } 3957a4ed38d7SDouglas Fuller 3958a4ed38d7SDouglas Fuller /* 3959a4ed38d7SDouglas Fuller * On success, the caller is responsible for: 3960a4ed38d7SDouglas Fuller * 3961a4ed38d7SDouglas Fuller * kfree(watchers); 3962a4ed38d7SDouglas Fuller */ 3963a4ed38d7SDouglas Fuller int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, 3964a4ed38d7SDouglas Fuller struct ceph_object_id *oid, 3965a4ed38d7SDouglas Fuller struct ceph_object_locator *oloc, 3966a4ed38d7SDouglas Fuller struct ceph_watch_item **watchers, 3967a4ed38d7SDouglas Fuller u32 *num_watchers) 3968a4ed38d7SDouglas Fuller { 3969a4ed38d7SDouglas Fuller struct ceph_osd_request *req; 3970a4ed38d7SDouglas Fuller struct page **pages; 3971a4ed38d7SDouglas Fuller int ret; 3972a4ed38d7SDouglas Fuller 3973a4ed38d7SDouglas Fuller req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); 3974a4ed38d7SDouglas Fuller if (!req) 3975a4ed38d7SDouglas Fuller return -ENOMEM; 3976a4ed38d7SDouglas Fuller 3977a4ed38d7SDouglas Fuller ceph_oid_copy(&req->r_base_oid, oid); 3978a4ed38d7SDouglas Fuller ceph_oloc_copy(&req->r_base_oloc, oloc); 3979a4ed38d7SDouglas Fuller req->r_flags = CEPH_OSD_FLAG_READ; 3980a4ed38d7SDouglas Fuller 3981a4ed38d7SDouglas Fuller ret = ceph_osdc_alloc_messages(req, GFP_NOIO); 3982a4ed38d7SDouglas Fuller if (ret) 3983a4ed38d7SDouglas Fuller goto out_put_req; 3984a4ed38d7SDouglas Fuller 3985a4ed38d7SDouglas Fuller pages = ceph_alloc_page_vector(1, GFP_NOIO); 3986a4ed38d7SDouglas Fuller if (IS_ERR(pages)) { 3987a4ed38d7SDouglas Fuller ret = PTR_ERR(pages); 3988a4ed38d7SDouglas Fuller goto out_put_req; 3989a4ed38d7SDouglas Fuller } 3990a4ed38d7SDouglas Fuller 3991a4ed38d7SDouglas Fuller osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); 3992a4ed38d7SDouglas Fuller ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, 3993a4ed38d7SDouglas Fuller response_data), 3994a4ed38d7SDouglas Fuller pages, PAGE_SIZE, 0, false, true); 3995a4ed38d7SDouglas Fuller 3996a4ed38d7SDouglas Fuller ceph_osdc_start_request(osdc, req, false); 3997a4ed38d7SDouglas Fuller ret = ceph_osdc_wait_request(osdc, req); 3998a4ed38d7SDouglas Fuller if (ret >= 0) { 3999a4ed38d7SDouglas Fuller void *p = page_address(pages[0]); 4000a4ed38d7SDouglas Fuller void *const end = p + req->r_ops[0].outdata_len; 4001a4ed38d7SDouglas Fuller 4002a4ed38d7SDouglas Fuller ret = decode_watchers(&p, end, watchers, num_watchers); 4003a4ed38d7SDouglas Fuller } 4004a4ed38d7SDouglas Fuller 4005a4ed38d7SDouglas Fuller out_put_req: 4006a4ed38d7SDouglas Fuller ceph_osdc_put_request(req); 4007a4ed38d7SDouglas Fuller return ret; 4008a4ed38d7SDouglas Fuller } 4009a4ed38d7SDouglas Fuller EXPORT_SYMBOL(ceph_osdc_list_watchers); 4010a4ed38d7SDouglas Fuller 4011b07d3c4bSIlya Dryomov /* 4012dd935f44SJosh Durgin * Call all pending notify callbacks - for use after a watch is 4013dd935f44SJosh Durgin * unregistered, to make sure no more callbacks for it will be invoked 4014dd935f44SJosh Durgin */ 4015f6479449Sstephen hemminger void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) 4016dd935f44SJosh Durgin { 4017dd935f44SJosh Durgin flush_workqueue(osdc->notify_wq); 4018dd935f44SJosh Durgin } 4019dd935f44SJosh Durgin EXPORT_SYMBOL(ceph_osdc_flush_notifies); 4020dd935f44SJosh Durgin 40217cca78c9SIlya Dryomov void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) 40227cca78c9SIlya Dryomov { 40237cca78c9SIlya Dryomov down_read(&osdc->lock); 40247cca78c9SIlya Dryomov maybe_request_map(osdc); 40257cca78c9SIlya Dryomov up_read(&osdc->lock); 40267cca78c9SIlya Dryomov } 40277cca78c9SIlya Dryomov EXPORT_SYMBOL(ceph_osdc_maybe_request_map); 4028dd935f44SJosh Durgin 4029dd935f44SJosh Durgin /* 40303d14c5d2SYehuda Sadeh * init, shutdown 40313d14c5d2SYehuda Sadeh */ 40323d14c5d2SYehuda Sadeh int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) 40333d14c5d2SYehuda Sadeh { 40343d14c5d2SYehuda Sadeh int err; 40353d14c5d2SYehuda Sadeh 40363d14c5d2SYehuda Sadeh dout("init\n"); 40373d14c5d2SYehuda Sadeh osdc->client = client; 40385aea3dcdSIlya Dryomov init_rwsem(&osdc->lock); 40393d14c5d2SYehuda Sadeh osdc->osds = RB_ROOT; 40403d14c5d2SYehuda Sadeh INIT_LIST_HEAD(&osdc->osd_lru); 40419dd2845cSIlya Dryomov spin_lock_init(&osdc->osd_lru_lock); 40425aea3dcdSIlya Dryomov osd_init(&osdc->homeless_osd); 40435aea3dcdSIlya Dryomov osdc->homeless_osd.o_osdc = osdc; 40445aea3dcdSIlya Dryomov osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; 4045922dab61SIlya Dryomov osdc->linger_requests = RB_ROOT; 40464609245eSIlya Dryomov osdc->map_checks = RB_ROOT; 40474609245eSIlya Dryomov osdc->linger_map_checks = RB_ROOT; 40483d14c5d2SYehuda Sadeh INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); 40493d14c5d2SYehuda Sadeh INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); 40503d14c5d2SYehuda Sadeh 40513d14c5d2SYehuda Sadeh err = -ENOMEM; 4052e5253a7bSIlya Dryomov osdc->osdmap = ceph_osdmap_alloc(); 4053e5253a7bSIlya Dryomov if (!osdc->osdmap) 4054e5253a7bSIlya Dryomov goto out; 4055e5253a7bSIlya Dryomov 40569e767adbSIlya Dryomov osdc->req_mempool = mempool_create_slab_pool(10, 40579e767adbSIlya Dryomov ceph_osd_request_cache); 40583d14c5d2SYehuda Sadeh if (!osdc->req_mempool) 4059e5253a7bSIlya Dryomov goto out_map; 40603d14c5d2SYehuda Sadeh 4061d50b409fSSage Weil err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, 4062711da55dSIlya Dryomov PAGE_SIZE, 10, true, "osd_op"); 40633d14c5d2SYehuda Sadeh if (err < 0) 40643d14c5d2SYehuda Sadeh goto out_mempool; 4065d50b409fSSage Weil err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, 4066711da55dSIlya Dryomov PAGE_SIZE, 10, true, "osd_op_reply"); 40673d14c5d2SYehuda Sadeh if (err < 0) 40683d14c5d2SYehuda Sadeh goto out_msgpool; 4069a40c4f10SYehuda Sadeh 4070dbcae088SDan Carpenter err = -ENOMEM; 4071a40c4f10SYehuda Sadeh osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); 4072dbcae088SDan Carpenter if (!osdc->notify_wq) 4073c172ec5cSIlya Dryomov goto out_msgpool_reply; 4074c172ec5cSIlya Dryomov 4075fbca9635SIlya Dryomov schedule_delayed_work(&osdc->timeout_work, 4076fbca9635SIlya Dryomov osdc->client->options->osd_keepalive_timeout); 4077b37ee1b9SIlya Dryomov schedule_delayed_work(&osdc->osds_timeout_work, 4078b37ee1b9SIlya Dryomov round_jiffies_relative(osdc->client->options->osd_idle_ttl)); 4079b37ee1b9SIlya Dryomov 40803d14c5d2SYehuda Sadeh return 0; 40813d14c5d2SYehuda Sadeh 4082c172ec5cSIlya Dryomov out_msgpool_reply: 4083c172ec5cSIlya Dryomov ceph_msgpool_destroy(&osdc->msgpool_op_reply); 40843d14c5d2SYehuda Sadeh out_msgpool: 40853d14c5d2SYehuda Sadeh ceph_msgpool_destroy(&osdc->msgpool_op); 40863d14c5d2SYehuda Sadeh out_mempool: 40873d14c5d2SYehuda Sadeh mempool_destroy(osdc->req_mempool); 4088e5253a7bSIlya Dryomov out_map: 4089e5253a7bSIlya Dryomov ceph_osdmap_destroy(osdc->osdmap); 40903d14c5d2SYehuda Sadeh out: 40913d14c5d2SYehuda Sadeh return err; 40923d14c5d2SYehuda Sadeh } 40933d14c5d2SYehuda Sadeh 40943d14c5d2SYehuda Sadeh void ceph_osdc_stop(struct ceph_osd_client *osdc) 40953d14c5d2SYehuda Sadeh { 4096a40c4f10SYehuda Sadeh flush_workqueue(osdc->notify_wq); 4097a40c4f10SYehuda Sadeh destroy_workqueue(osdc->notify_wq); 40983d14c5d2SYehuda Sadeh cancel_delayed_work_sync(&osdc->timeout_work); 40993d14c5d2SYehuda Sadeh cancel_delayed_work_sync(&osdc->osds_timeout_work); 410042a2c09fSIlya Dryomov 41015aea3dcdSIlya Dryomov down_write(&osdc->lock); 410242a2c09fSIlya Dryomov while (!RB_EMPTY_ROOT(&osdc->osds)) { 410342a2c09fSIlya Dryomov struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), 410442a2c09fSIlya Dryomov struct ceph_osd, o_node); 41055aea3dcdSIlya Dryomov close_osd(osd); 410642a2c09fSIlya Dryomov } 41075aea3dcdSIlya Dryomov up_write(&osdc->lock); 41085aea3dcdSIlya Dryomov WARN_ON(atomic_read(&osdc->homeless_osd.o_ref) != 1); 41095aea3dcdSIlya Dryomov osd_cleanup(&osdc->homeless_osd); 41105aea3dcdSIlya Dryomov 41115aea3dcdSIlya Dryomov WARN_ON(!list_empty(&osdc->osd_lru)); 4112922dab61SIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); 41134609245eSIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); 41144609245eSIlya Dryomov WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); 41155aea3dcdSIlya Dryomov WARN_ON(atomic_read(&osdc->num_requests)); 41165aea3dcdSIlya Dryomov WARN_ON(atomic_read(&osdc->num_homeless)); 411742a2c09fSIlya Dryomov 41183d14c5d2SYehuda Sadeh ceph_osdmap_destroy(osdc->osdmap); 41193d14c5d2SYehuda Sadeh mempool_destroy(osdc->req_mempool); 41203d14c5d2SYehuda Sadeh ceph_msgpool_destroy(&osdc->msgpool_op); 41213d14c5d2SYehuda Sadeh ceph_msgpool_destroy(&osdc->msgpool_op_reply); 41223d14c5d2SYehuda Sadeh } 41233d14c5d2SYehuda Sadeh 41243d14c5d2SYehuda Sadeh /* 41253d14c5d2SYehuda Sadeh * Read some contiguous pages. If we cross a stripe boundary, shorten 41263d14c5d2SYehuda Sadeh * *plen. Return number of bytes read, or error. 41273d14c5d2SYehuda Sadeh */ 41283d14c5d2SYehuda Sadeh int ceph_osdc_readpages(struct ceph_osd_client *osdc, 41293d14c5d2SYehuda Sadeh struct ceph_vino vino, struct ceph_file_layout *layout, 41303d14c5d2SYehuda Sadeh u64 off, u64 *plen, 41313d14c5d2SYehuda Sadeh u32 truncate_seq, u64 truncate_size, 4132b7495fc2SSage Weil struct page **pages, int num_pages, int page_align) 41333d14c5d2SYehuda Sadeh { 41343d14c5d2SYehuda Sadeh struct ceph_osd_request *req; 41353d14c5d2SYehuda Sadeh int rc = 0; 41363d14c5d2SYehuda Sadeh 41373d14c5d2SYehuda Sadeh dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, 41383d14c5d2SYehuda Sadeh vino.snap, off, *plen); 4139715e4cd4SYan, Zheng req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, 41403d14c5d2SYehuda Sadeh CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, 4141acead002SAlex Elder NULL, truncate_seq, truncate_size, 4142153e5167SAlex Elder false); 41436816282dSSage Weil if (IS_ERR(req)) 41446816282dSSage Weil return PTR_ERR(req); 41453d14c5d2SYehuda Sadeh 41463d14c5d2SYehuda Sadeh /* it may be a short read due to an object boundary */ 4147406e2c9fSAlex Elder osd_req_op_extent_osd_data_pages(req, 0, 4148a4ce40a9SAlex Elder pages, *plen, page_align, false, false); 41493d14c5d2SYehuda Sadeh 4150e0c59487SAlex Elder dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", 415143bfe5deSAlex Elder off, *plen, *plen, page_align); 41523d14c5d2SYehuda Sadeh 41533d14c5d2SYehuda Sadeh rc = ceph_osdc_start_request(osdc, req, false); 41543d14c5d2SYehuda Sadeh if (!rc) 41553d14c5d2SYehuda Sadeh rc = ceph_osdc_wait_request(osdc, req); 41563d14c5d2SYehuda Sadeh 41573d14c5d2SYehuda Sadeh ceph_osdc_put_request(req); 41583d14c5d2SYehuda Sadeh dout("readpages result %d\n", rc); 41593d14c5d2SYehuda Sadeh return rc; 41603d14c5d2SYehuda Sadeh } 41613d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_readpages); 41623d14c5d2SYehuda Sadeh 41633d14c5d2SYehuda Sadeh /* 41643d14c5d2SYehuda Sadeh * do a synchronous write on N pages 41653d14c5d2SYehuda Sadeh */ 41663d14c5d2SYehuda Sadeh int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, 41673d14c5d2SYehuda Sadeh struct ceph_file_layout *layout, 41683d14c5d2SYehuda Sadeh struct ceph_snap_context *snapc, 41693d14c5d2SYehuda Sadeh u64 off, u64 len, 41703d14c5d2SYehuda Sadeh u32 truncate_seq, u64 truncate_size, 41713d14c5d2SYehuda Sadeh struct timespec *mtime, 417224808826SAlex Elder struct page **pages, int num_pages) 41733d14c5d2SYehuda Sadeh { 41743d14c5d2SYehuda Sadeh struct ceph_osd_request *req; 41753d14c5d2SYehuda Sadeh int rc = 0; 4176b7495fc2SSage Weil int page_align = off & ~PAGE_MASK; 41773d14c5d2SYehuda Sadeh 4178715e4cd4SYan, Zheng req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, 41793d14c5d2SYehuda Sadeh CEPH_OSD_OP_WRITE, 418024808826SAlex Elder CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, 4181acead002SAlex Elder snapc, truncate_seq, truncate_size, 4182153e5167SAlex Elder true); 41836816282dSSage Weil if (IS_ERR(req)) 41846816282dSSage Weil return PTR_ERR(req); 41853d14c5d2SYehuda Sadeh 41863d14c5d2SYehuda Sadeh /* it may be a short write due to an object boundary */ 4187406e2c9fSAlex Elder osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, 418843bfe5deSAlex Elder false, false); 418943bfe5deSAlex Elder dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); 41903d14c5d2SYehuda Sadeh 4191bb873b53SIlya Dryomov req->r_mtime = *mtime; 419287f979d3SAlex Elder rc = ceph_osdc_start_request(osdc, req, true); 41933d14c5d2SYehuda Sadeh if (!rc) 41943d14c5d2SYehuda Sadeh rc = ceph_osdc_wait_request(osdc, req); 41953d14c5d2SYehuda Sadeh 41963d14c5d2SYehuda Sadeh ceph_osdc_put_request(req); 41973d14c5d2SYehuda Sadeh if (rc == 0) 41983d14c5d2SYehuda Sadeh rc = len; 41993d14c5d2SYehuda Sadeh dout("writepages result %d\n", rc); 42003d14c5d2SYehuda Sadeh return rc; 42013d14c5d2SYehuda Sadeh } 42023d14c5d2SYehuda Sadeh EXPORT_SYMBOL(ceph_osdc_writepages); 42033d14c5d2SYehuda Sadeh 42045522ae0bSAlex Elder int ceph_osdc_setup(void) 42055522ae0bSAlex Elder { 42063f1af42aSIlya Dryomov size_t size = sizeof(struct ceph_osd_request) + 42073f1af42aSIlya Dryomov CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); 42083f1af42aSIlya Dryomov 42095522ae0bSAlex Elder BUG_ON(ceph_osd_request_cache); 42103f1af42aSIlya Dryomov ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, 42113f1af42aSIlya Dryomov 0, 0, NULL); 42125522ae0bSAlex Elder 42135522ae0bSAlex Elder return ceph_osd_request_cache ? 0 : -ENOMEM; 42145522ae0bSAlex Elder } 42155522ae0bSAlex Elder EXPORT_SYMBOL(ceph_osdc_setup); 42165522ae0bSAlex Elder 42175522ae0bSAlex Elder void ceph_osdc_cleanup(void) 42185522ae0bSAlex Elder { 42195522ae0bSAlex Elder BUG_ON(!ceph_osd_request_cache); 42205522ae0bSAlex Elder kmem_cache_destroy(ceph_osd_request_cache); 42215522ae0bSAlex Elder ceph_osd_request_cache = NULL; 42225522ae0bSAlex Elder } 42235522ae0bSAlex Elder EXPORT_SYMBOL(ceph_osdc_cleanup); 42245522ae0bSAlex Elder 42253d14c5d2SYehuda Sadeh /* 42263d14c5d2SYehuda Sadeh * handle incoming message 42273d14c5d2SYehuda Sadeh */ 42283d14c5d2SYehuda Sadeh static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) 42293d14c5d2SYehuda Sadeh { 42303d14c5d2SYehuda Sadeh struct ceph_osd *osd = con->private; 42315aea3dcdSIlya Dryomov struct ceph_osd_client *osdc = osd->o_osdc; 42323d14c5d2SYehuda Sadeh int type = le16_to_cpu(msg->hdr.type); 42333d14c5d2SYehuda Sadeh 42343d14c5d2SYehuda Sadeh switch (type) { 42353d14c5d2SYehuda Sadeh case CEPH_MSG_OSD_MAP: 42363d14c5d2SYehuda Sadeh ceph_osdc_handle_map(osdc, msg); 42373d14c5d2SYehuda Sadeh break; 42383d14c5d2SYehuda Sadeh case CEPH_MSG_OSD_OPREPLY: 42395aea3dcdSIlya Dryomov handle_reply(osd, msg); 42403d14c5d2SYehuda Sadeh break; 4241a40c4f10SYehuda Sadeh case CEPH_MSG_WATCH_NOTIFY: 4242a40c4f10SYehuda Sadeh handle_watch_notify(osdc, msg); 4243a40c4f10SYehuda Sadeh break; 42443d14c5d2SYehuda Sadeh 42453d14c5d2SYehuda Sadeh default: 42463d14c5d2SYehuda Sadeh pr_err("received unknown message type %d %s\n", type, 42473d14c5d2SYehuda Sadeh ceph_msg_type_name(type)); 42483d14c5d2SYehuda Sadeh } 42495aea3dcdSIlya Dryomov 42503d14c5d2SYehuda Sadeh ceph_msg_put(msg); 42513d14c5d2SYehuda Sadeh } 42523d14c5d2SYehuda Sadeh 42533d14c5d2SYehuda Sadeh /* 4254d15f9d69SIlya Dryomov * Lookup and return message for incoming reply. Don't try to do 4255d15f9d69SIlya Dryomov * anything about a larger than preallocated data portion of the 4256d15f9d69SIlya Dryomov * message at the moment - for now, just skip the message. 42573d14c5d2SYehuda Sadeh */ 42583d14c5d2SYehuda Sadeh static struct ceph_msg *get_reply(struct ceph_connection *con, 42593d14c5d2SYehuda Sadeh struct ceph_msg_header *hdr, 42603d14c5d2SYehuda Sadeh int *skip) 42613d14c5d2SYehuda Sadeh { 42623d14c5d2SYehuda Sadeh struct ceph_osd *osd = con->private; 42633d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = osd->o_osdc; 42645aea3dcdSIlya Dryomov struct ceph_msg *m = NULL; 42653d14c5d2SYehuda Sadeh struct ceph_osd_request *req; 42663f0a4ac5SIlya Dryomov int front_len = le32_to_cpu(hdr->front_len); 42673d14c5d2SYehuda Sadeh int data_len = le32_to_cpu(hdr->data_len); 42685aea3dcdSIlya Dryomov u64 tid = le64_to_cpu(hdr->tid); 42693d14c5d2SYehuda Sadeh 42705aea3dcdSIlya Dryomov down_read(&osdc->lock); 42715aea3dcdSIlya Dryomov if (!osd_registered(osd)) { 42725aea3dcdSIlya Dryomov dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); 42735aea3dcdSIlya Dryomov *skip = 1; 42745aea3dcdSIlya Dryomov goto out_unlock_osdc; 42755aea3dcdSIlya Dryomov } 42765aea3dcdSIlya Dryomov WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); 42775aea3dcdSIlya Dryomov 42785aea3dcdSIlya Dryomov mutex_lock(&osd->lock); 42795aea3dcdSIlya Dryomov req = lookup_request(&osd->o_requests, tid); 42803d14c5d2SYehuda Sadeh if (!req) { 4281cd8140c6SIlya Dryomov dout("%s osd%d tid %llu unknown, skipping\n", __func__, 4282cd8140c6SIlya Dryomov osd->o_osd, tid); 4283d15f9d69SIlya Dryomov *skip = 1; 42845aea3dcdSIlya Dryomov goto out_unlock_session; 42853d14c5d2SYehuda Sadeh } 42863d14c5d2SYehuda Sadeh 42878921d114SAlex Elder ceph_msg_revoke_incoming(req->r_reply); 42883d14c5d2SYehuda Sadeh 4289f2be82b0SIlya Dryomov if (front_len > req->r_reply->front_alloc_len) { 4290d15f9d69SIlya Dryomov pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", 4291d15f9d69SIlya Dryomov __func__, osd->o_osd, req->r_tid, front_len, 4292d15f9d69SIlya Dryomov req->r_reply->front_alloc_len); 42933f0a4ac5SIlya Dryomov m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, 42943f0a4ac5SIlya Dryomov false); 42953d14c5d2SYehuda Sadeh if (!m) 42965aea3dcdSIlya Dryomov goto out_unlock_session; 42973d14c5d2SYehuda Sadeh ceph_msg_put(req->r_reply); 42983d14c5d2SYehuda Sadeh req->r_reply = m; 42993d14c5d2SYehuda Sadeh } 43003d14c5d2SYehuda Sadeh 4301d15f9d69SIlya Dryomov if (data_len > req->r_reply->data_length) { 4302d15f9d69SIlya Dryomov pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", 4303d15f9d69SIlya Dryomov __func__, osd->o_osd, req->r_tid, data_len, 4304d15f9d69SIlya Dryomov req->r_reply->data_length); 43053d14c5d2SYehuda Sadeh m = NULL; 4306d15f9d69SIlya Dryomov *skip = 1; 43075aea3dcdSIlya Dryomov goto out_unlock_session; 43083d14c5d2SYehuda Sadeh } 4309d15f9d69SIlya Dryomov 4310d15f9d69SIlya Dryomov m = ceph_msg_get(req->r_reply); 43113d14c5d2SYehuda Sadeh dout("get_reply tid %lld %p\n", tid, m); 43123d14c5d2SYehuda Sadeh 43135aea3dcdSIlya Dryomov out_unlock_session: 43145aea3dcdSIlya Dryomov mutex_unlock(&osd->lock); 43155aea3dcdSIlya Dryomov out_unlock_osdc: 43165aea3dcdSIlya Dryomov up_read(&osdc->lock); 43173d14c5d2SYehuda Sadeh return m; 43183d14c5d2SYehuda Sadeh } 43193d14c5d2SYehuda Sadeh 432019079203SIlya Dryomov /* 432119079203SIlya Dryomov * TODO: switch to a msg-owned pagelist 432219079203SIlya Dryomov */ 432319079203SIlya Dryomov static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) 432419079203SIlya Dryomov { 432519079203SIlya Dryomov struct ceph_msg *m; 432619079203SIlya Dryomov int type = le16_to_cpu(hdr->type); 432719079203SIlya Dryomov u32 front_len = le32_to_cpu(hdr->front_len); 432819079203SIlya Dryomov u32 data_len = le32_to_cpu(hdr->data_len); 432919079203SIlya Dryomov 433019079203SIlya Dryomov m = ceph_msg_new(type, front_len, GFP_NOIO, false); 433119079203SIlya Dryomov if (!m) 433219079203SIlya Dryomov return NULL; 433319079203SIlya Dryomov 433419079203SIlya Dryomov if (data_len) { 433519079203SIlya Dryomov struct page **pages; 433619079203SIlya Dryomov struct ceph_osd_data osd_data; 433719079203SIlya Dryomov 433819079203SIlya Dryomov pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), 433919079203SIlya Dryomov GFP_NOIO); 4340c22e853aSWei Yongjun if (IS_ERR(pages)) { 434119079203SIlya Dryomov ceph_msg_put(m); 434219079203SIlya Dryomov return NULL; 434319079203SIlya Dryomov } 434419079203SIlya Dryomov 434519079203SIlya Dryomov ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false, 434619079203SIlya Dryomov false); 434719079203SIlya Dryomov ceph_osdc_msg_data_add(m, &osd_data); 434819079203SIlya Dryomov } 434919079203SIlya Dryomov 435019079203SIlya Dryomov return m; 435119079203SIlya Dryomov } 435219079203SIlya Dryomov 43533d14c5d2SYehuda Sadeh static struct ceph_msg *alloc_msg(struct ceph_connection *con, 43543d14c5d2SYehuda Sadeh struct ceph_msg_header *hdr, 43553d14c5d2SYehuda Sadeh int *skip) 43563d14c5d2SYehuda Sadeh { 43573d14c5d2SYehuda Sadeh struct ceph_osd *osd = con->private; 43583d14c5d2SYehuda Sadeh int type = le16_to_cpu(hdr->type); 43593d14c5d2SYehuda Sadeh 43601c20f2d2SAlex Elder *skip = 0; 43613d14c5d2SYehuda Sadeh switch (type) { 43623d14c5d2SYehuda Sadeh case CEPH_MSG_OSD_MAP: 4363a40c4f10SYehuda Sadeh case CEPH_MSG_WATCH_NOTIFY: 436419079203SIlya Dryomov return alloc_msg_with_page_vector(hdr); 43653d14c5d2SYehuda Sadeh case CEPH_MSG_OSD_OPREPLY: 43663d14c5d2SYehuda Sadeh return get_reply(con, hdr, skip); 43673d14c5d2SYehuda Sadeh default: 43685aea3dcdSIlya Dryomov pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, 43695aea3dcdSIlya Dryomov osd->o_osd, type); 43703d14c5d2SYehuda Sadeh *skip = 1; 43713d14c5d2SYehuda Sadeh return NULL; 43723d14c5d2SYehuda Sadeh } 43733d14c5d2SYehuda Sadeh } 43743d14c5d2SYehuda Sadeh 43753d14c5d2SYehuda Sadeh /* 43763d14c5d2SYehuda Sadeh * Wrappers to refcount containing ceph_osd struct 43773d14c5d2SYehuda Sadeh */ 43783d14c5d2SYehuda Sadeh static struct ceph_connection *get_osd_con(struct ceph_connection *con) 43793d14c5d2SYehuda Sadeh { 43803d14c5d2SYehuda Sadeh struct ceph_osd *osd = con->private; 43813d14c5d2SYehuda Sadeh if (get_osd(osd)) 43823d14c5d2SYehuda Sadeh return con; 43833d14c5d2SYehuda Sadeh return NULL; 43843d14c5d2SYehuda Sadeh } 43853d14c5d2SYehuda Sadeh 43863d14c5d2SYehuda Sadeh static void put_osd_con(struct ceph_connection *con) 43873d14c5d2SYehuda Sadeh { 43883d14c5d2SYehuda Sadeh struct ceph_osd *osd = con->private; 43893d14c5d2SYehuda Sadeh put_osd(osd); 43903d14c5d2SYehuda Sadeh } 43913d14c5d2SYehuda Sadeh 43923d14c5d2SYehuda Sadeh /* 43933d14c5d2SYehuda Sadeh * authentication 43943d14c5d2SYehuda Sadeh */ 4395a3530df3SAlex Elder /* 4396a3530df3SAlex Elder * Note: returned pointer is the address of a structure that's 4397a3530df3SAlex Elder * managed separately. Caller must *not* attempt to free it. 4398a3530df3SAlex Elder */ 4399a3530df3SAlex Elder static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, 44008f43fb53SAlex Elder int *proto, int force_new) 44013d14c5d2SYehuda Sadeh { 44023d14c5d2SYehuda Sadeh struct ceph_osd *o = con->private; 44033d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = o->o_osdc; 44043d14c5d2SYehuda Sadeh struct ceph_auth_client *ac = osdc->client->monc.auth; 440574f1869fSAlex Elder struct ceph_auth_handshake *auth = &o->o_auth; 44063d14c5d2SYehuda Sadeh 440774f1869fSAlex Elder if (force_new && auth->authorizer) { 44086c1ea260SIlya Dryomov ceph_auth_destroy_authorizer(auth->authorizer); 440974f1869fSAlex Elder auth->authorizer = NULL; 44103d14c5d2SYehuda Sadeh } 441127859f97SSage Weil if (!auth->authorizer) { 441227859f97SSage Weil int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 4413a3530df3SAlex Elder auth); 44143d14c5d2SYehuda Sadeh if (ret) 4415a3530df3SAlex Elder return ERR_PTR(ret); 441627859f97SSage Weil } else { 441727859f97SSage Weil int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, 44180bed9b5cSSage Weil auth); 44190bed9b5cSSage Weil if (ret) 44200bed9b5cSSage Weil return ERR_PTR(ret); 44213d14c5d2SYehuda Sadeh } 44223d14c5d2SYehuda Sadeh *proto = ac->protocol; 442374f1869fSAlex Elder 4424a3530df3SAlex Elder return auth; 44253d14c5d2SYehuda Sadeh } 44263d14c5d2SYehuda Sadeh 44273d14c5d2SYehuda Sadeh 44283d14c5d2SYehuda Sadeh static int verify_authorizer_reply(struct ceph_connection *con, int len) 44293d14c5d2SYehuda Sadeh { 44303d14c5d2SYehuda Sadeh struct ceph_osd *o = con->private; 44313d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = o->o_osdc; 44323d14c5d2SYehuda Sadeh struct ceph_auth_client *ac = osdc->client->monc.auth; 44333d14c5d2SYehuda Sadeh 443427859f97SSage Weil return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len); 44353d14c5d2SYehuda Sadeh } 44363d14c5d2SYehuda Sadeh 44373d14c5d2SYehuda Sadeh static int invalidate_authorizer(struct ceph_connection *con) 44383d14c5d2SYehuda Sadeh { 44393d14c5d2SYehuda Sadeh struct ceph_osd *o = con->private; 44403d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = o->o_osdc; 44413d14c5d2SYehuda Sadeh struct ceph_auth_client *ac = osdc->client->monc.auth; 44423d14c5d2SYehuda Sadeh 444327859f97SSage Weil ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); 44443d14c5d2SYehuda Sadeh return ceph_monc_validate_auth(&osdc->client->monc); 44453d14c5d2SYehuda Sadeh } 44463d14c5d2SYehuda Sadeh 444779dbd1baSIlya Dryomov static int osd_sign_message(struct ceph_msg *msg) 444833d07337SYan, Zheng { 444979dbd1baSIlya Dryomov struct ceph_osd *o = msg->con->private; 445033d07337SYan, Zheng struct ceph_auth_handshake *auth = &o->o_auth; 445179dbd1baSIlya Dryomov 445233d07337SYan, Zheng return ceph_auth_sign_message(auth, msg); 445333d07337SYan, Zheng } 445433d07337SYan, Zheng 445579dbd1baSIlya Dryomov static int osd_check_message_signature(struct ceph_msg *msg) 445633d07337SYan, Zheng { 445779dbd1baSIlya Dryomov struct ceph_osd *o = msg->con->private; 445833d07337SYan, Zheng struct ceph_auth_handshake *auth = &o->o_auth; 445979dbd1baSIlya Dryomov 446033d07337SYan, Zheng return ceph_auth_check_message_signature(auth, msg); 446133d07337SYan, Zheng } 446233d07337SYan, Zheng 44633d14c5d2SYehuda Sadeh static const struct ceph_connection_operations osd_con_ops = { 44643d14c5d2SYehuda Sadeh .get = get_osd_con, 44653d14c5d2SYehuda Sadeh .put = put_osd_con, 44663d14c5d2SYehuda Sadeh .dispatch = dispatch, 44673d14c5d2SYehuda Sadeh .get_authorizer = get_authorizer, 44683d14c5d2SYehuda Sadeh .verify_authorizer_reply = verify_authorizer_reply, 44693d14c5d2SYehuda Sadeh .invalidate_authorizer = invalidate_authorizer, 44703d14c5d2SYehuda Sadeh .alloc_msg = alloc_msg, 447179dbd1baSIlya Dryomov .sign_message = osd_sign_message, 447279dbd1baSIlya Dryomov .check_message_signature = osd_check_message_signature, 44735aea3dcdSIlya Dryomov .fault = osd_fault, 44743d14c5d2SYehuda Sadeh }; 4475