1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 23d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 31d3576fdSSage Weil 41d3576fdSSage Weil #include <linux/backing-dev.h> 51d3576fdSSage Weil #include <linux/fs.h> 61d3576fdSSage Weil #include <linux/mm.h> 7d7bdba1cSDavid Howells #include <linux/swap.h> 81d3576fdSSage Weil #include <linux/pagemap.h> 95a0e3ad6STejun Heo #include <linux/slab.h> 101d3576fdSSage Weil #include <linux/pagevec.h> 111d3576fdSSage Weil #include <linux/task_io_accounting_ops.h> 12f361bf4aSIngo Molnar #include <linux/signal.h> 135c308356SJeff Layton #include <linux/iversion.h> 1497e27aaaSXiubo Li #include <linux/ktime.h> 15f0702876SJeff Layton #include <linux/netfs.h> 161d3576fdSSage Weil 171d3576fdSSage Weil #include "super.h" 183d14c5d2SYehuda Sadeh #include "mds_client.h" 1999ccbd22SMilosz Tanski #include "cache.h" 2097e27aaaSXiubo Li #include "metric.h" 213d14c5d2SYehuda Sadeh #include <linux/ceph/osd_client.h> 2208c1ac50SIlya Dryomov #include <linux/ceph/striper.h> 231d3576fdSSage Weil 241d3576fdSSage Weil /* 251d3576fdSSage Weil * Ceph address space ops. 261d3576fdSSage Weil * 271d3576fdSSage Weil * There are a few funny things going on here. 281d3576fdSSage Weil * 291d3576fdSSage Weil * The page->private field is used to reference a struct 301d3576fdSSage Weil * ceph_snap_context for _every_ dirty page. This indicates which 311d3576fdSSage Weil * snapshot the page was logically dirtied in, and thus which snap 321d3576fdSSage Weil * context needs to be associated with the osd write during writeback. 331d3576fdSSage Weil * 341d3576fdSSage Weil * Similarly, struct ceph_inode_info maintains a set of counters to 3525985edcSLucas De Marchi * count dirty pages on the inode. In the absence of snapshots, 361d3576fdSSage Weil * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 371d3576fdSSage Weil * 381d3576fdSSage Weil * When a snapshot is taken (that is, when the client receives 391d3576fdSSage Weil * notification that a snapshot was taken), each inode with caps and 401d3576fdSSage Weil * with dirty pages (dirty pages implies there is a cap) gets a new 411d3576fdSSage Weil * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 421d3576fdSSage Weil * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 431d3576fdSSage Weil * moved to capsnap->dirty. (Unless a sync write is currently in 441d3576fdSSage Weil * progress. In that case, the capsnap is said to be "pending", new 451d3576fdSSage Weil * writes cannot start, and the capsnap isn't "finalized" until the 461d3576fdSSage Weil * write completes (or fails) and a final size/mtime for the inode for 471d3576fdSSage Weil * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 481d3576fdSSage Weil * 491d3576fdSSage Weil * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 501d3576fdSSage Weil * we look for the first capsnap in i_cap_snaps and write out pages in 511d3576fdSSage Weil * that snap context _only_. Then we move on to the next capsnap, 521d3576fdSSage Weil * eventually reaching the "live" or "head" context (i.e., pages that 531d3576fdSSage Weil * are not yet snapped) and are writing the most recently dirtied 541d3576fdSSage Weil * pages. 551d3576fdSSage Weil * 561d3576fdSSage Weil * Invalidate and so forth must take care to ensure the dirty page 571d3576fdSSage Weil * accounting is preserved. 581d3576fdSSage Weil */ 591d3576fdSSage Weil 602baba250SYehuda Sadeh #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 612baba250SYehuda Sadeh #define CONGESTION_OFF_THRESH(congestion_kb) \ 622baba250SYehuda Sadeh (CONGESTION_ON_THRESH(congestion_kb) - \ 632baba250SYehuda Sadeh (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 642baba250SYehuda Sadeh 65d801327dSJeff Layton static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, 66fac47b43SXiubo Li struct folio **foliop, void **_fsdata); 67d801327dSJeff Layton 6861600ef8SYan, Zheng static inline struct ceph_snap_context *page_snap_context(struct page *page) 6961600ef8SYan, Zheng { 7061600ef8SYan, Zheng if (PagePrivate(page)) 7161600ef8SYan, Zheng return (void *)page->private; 7261600ef8SYan, Zheng return NULL; 7361600ef8SYan, Zheng } 741d3576fdSSage Weil 751d3576fdSSage Weil /* 761d3576fdSSage Weil * Dirty a page. Optimistically adjust accounting, on the assumption 771d3576fdSSage Weil * that we won't race with invalidate. If we do, readjust. 781d3576fdSSage Weil */ 798fb72b4aSMatthew Wilcox (Oracle) static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio) 801d3576fdSSage Weil { 811d3576fdSSage Weil struct inode *inode; 821d3576fdSSage Weil struct ceph_inode_info *ci; 831d3576fdSSage Weil struct ceph_snap_context *snapc; 841d3576fdSSage Weil 858fb72b4aSMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) { 868fb72b4aSMatthew Wilcox (Oracle) dout("%p dirty_folio %p idx %lu -- already dirty\n", 878fb72b4aSMatthew Wilcox (Oracle) mapping->host, folio, folio->index); 88642d51fbSXiubo Li VM_BUG_ON_FOLIO(!folio_test_private(folio), folio); 898fb72b4aSMatthew Wilcox (Oracle) return false; 901d3576fdSSage Weil } 911d3576fdSSage Weil 921d3576fdSSage Weil inode = mapping->host; 931d3576fdSSage Weil ci = ceph_inode(inode); 941d3576fdSSage Weil 951d3576fdSSage Weil /* dirty the head */ 96be655596SSage Weil spin_lock(&ci->i_ceph_lock); 975dda377cSYan, Zheng BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference 985dda377cSYan, Zheng if (__ceph_have_pending_cap_snap(ci)) { 995dda377cSYan, Zheng struct ceph_cap_snap *capsnap = 1005dda377cSYan, Zheng list_last_entry(&ci->i_cap_snaps, 1015dda377cSYan, Zheng struct ceph_cap_snap, 1025dda377cSYan, Zheng ci_item); 1035dda377cSYan, Zheng snapc = ceph_get_snap_context(capsnap->context); 1045dda377cSYan, Zheng capsnap->dirty_pages++; 1055dda377cSYan, Zheng } else { 1065dda377cSYan, Zheng BUG_ON(!ci->i_head_snapc); 1075dda377cSYan, Zheng snapc = ceph_get_snap_context(ci->i_head_snapc); 1081d3576fdSSage Weil ++ci->i_wrbuffer_ref_head; 1095dda377cSYan, Zheng } 1101d3576fdSSage Weil if (ci->i_wrbuffer_ref == 0) 1110444d76aSDave Chinner ihold(inode); 1121d3576fdSSage Weil ++ci->i_wrbuffer_ref; 1138fb72b4aSMatthew Wilcox (Oracle) dout("%p dirty_folio %p idx %lu head %d/%d -> %d/%d " 1141d3576fdSSage Weil "snapc %p seq %lld (%d snaps)\n", 1158fb72b4aSMatthew Wilcox (Oracle) mapping->host, folio, folio->index, 1161d3576fdSSage Weil ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 1171d3576fdSSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 1181d3576fdSSage Weil snapc, snapc->seq, snapc->num_snaps); 119be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1201d3576fdSSage Weil 1211d3576fdSSage Weil /* 1228fb72b4aSMatthew Wilcox (Oracle) * Reference snap context in folio->private. Also set 1239872f4deSMatthew Wilcox (Oracle) * PagePrivate so that we get invalidate_folio callback. 1241d3576fdSSage Weil */ 125020bc44aSJeff Layton VM_WARN_ON_FOLIO(folio->private, folio); 1268fb72b4aSMatthew Wilcox (Oracle) folio_attach_private(folio, snapc); 1271d3576fdSSage Weil 1288fb72b4aSMatthew Wilcox (Oracle) return ceph_fscache_dirty_folio(mapping, folio); 1291d3576fdSSage Weil } 1301d3576fdSSage Weil 1311d3576fdSSage Weil /* 1329872f4deSMatthew Wilcox (Oracle) * If we are truncating the full folio (i.e. offset == 0), adjust the 1339872f4deSMatthew Wilcox (Oracle) * dirty folio counters appropriately. Only called if there is private 1349872f4deSMatthew Wilcox (Oracle) * data on the folio. 1351d3576fdSSage Weil */ 1369872f4deSMatthew Wilcox (Oracle) static void ceph_invalidate_folio(struct folio *folio, size_t offset, 1379872f4deSMatthew Wilcox (Oracle) size_t length) 1381d3576fdSSage Weil { 1394ce1e9adSAlexander Beregalov struct inode *inode; 1401d3576fdSSage Weil struct ceph_inode_info *ci; 141379fc7faSJeff Layton struct ceph_snap_context *snapc; 1421d3576fdSSage Weil 1439872f4deSMatthew Wilcox (Oracle) inode = folio->mapping->host; 144b150f5c1SMilosz Tanski ci = ceph_inode(inode); 145b150f5c1SMilosz Tanski 1469872f4deSMatthew Wilcox (Oracle) if (offset != 0 || length != folio_size(folio)) { 1479872f4deSMatthew Wilcox (Oracle) dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n", 1489872f4deSMatthew Wilcox (Oracle) inode, folio->index, offset, length); 149b150f5c1SMilosz Tanski return; 150b150f5c1SMilosz Tanski } 1514ce1e9adSAlexander Beregalov 1529872f4deSMatthew Wilcox (Oracle) WARN_ON(!folio_test_locked(folio)); 153642d51fbSXiubo Li if (folio_test_private(folio)) { 1549872f4deSMatthew Wilcox (Oracle) dout("%p invalidate_folio idx %lu full dirty page\n", 1559872f4deSMatthew Wilcox (Oracle) inode, folio->index); 156b150f5c1SMilosz Tanski 1579872f4deSMatthew Wilcox (Oracle) snapc = folio_detach_private(folio); 1581d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 1591d3576fdSSage Weil ceph_put_snap_context(snapc); 1601d3576fdSSage Weil } 1611d3576fdSSage Weil 1629872f4deSMatthew Wilcox (Oracle) folio_wait_fscache(folio); 163400e1286SJeff Layton } 164400e1286SJeff Layton 1655e414655SMatthew Wilcox (Oracle) static bool ceph_release_folio(struct folio *folio, gfp_t gfp) 1661d3576fdSSage Weil { 1675e414655SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 168400e1286SJeff Layton 1695e414655SMatthew Wilcox (Oracle) dout("%llx:%llx release_folio idx %lu (%sdirty)\n", 1705e414655SMatthew Wilcox (Oracle) ceph_vinop(inode), 1715e414655SMatthew Wilcox (Oracle) folio->index, folio_test_dirty(folio) ? "" : "not "); 172400e1286SJeff Layton 1735e414655SMatthew Wilcox (Oracle) if (folio_test_private(folio)) 1745e414655SMatthew Wilcox (Oracle) return false; 17599ccbd22SMilosz Tanski 1765e414655SMatthew Wilcox (Oracle) if (folio_test_fscache(folio)) { 177d7bdba1cSDavid Howells if (current_is_kswapd() || !(gfp & __GFP_FS)) 1785e414655SMatthew Wilcox (Oracle) return false; 1795e414655SMatthew Wilcox (Oracle) folio_wait_fscache(folio); 1807c46b318SJeff Layton } 181400e1286SJeff Layton ceph_fscache_note_page_release(inode); 1825e414655SMatthew Wilcox (Oracle) return true; 1831d3576fdSSage Weil } 1841d3576fdSSage Weil 1856a19114bSDavid Howells static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq) 186f0702876SJeff Layton { 187a25cedb4SJeff Layton struct inode *inode = rreq->inode; 188f0702876SJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 189f0702876SJeff Layton struct ceph_file_layout *lo = &ci->i_layout; 190f0702876SJeff Layton u32 blockoff; 191f0702876SJeff Layton u64 blockno; 192f0702876SJeff Layton 193f0702876SJeff Layton /* Expand the start downward */ 194f0702876SJeff Layton blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff); 195f0702876SJeff Layton rreq->start = blockno * lo->stripe_unit; 196f0702876SJeff Layton rreq->len += blockoff; 197f0702876SJeff Layton 198f0702876SJeff Layton /* Now, round up the length to the next block */ 199f0702876SJeff Layton rreq->len = roundup(rreq->len, lo->stripe_unit); 200f0702876SJeff Layton } 201f0702876SJeff Layton 2026a19114bSDavid Howells static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq) 203f0702876SJeff Layton { 204a25cedb4SJeff Layton struct inode *inode = subreq->rreq->inode; 205f0702876SJeff Layton struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 206f0702876SJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 207f0702876SJeff Layton u64 objno, objoff; 208f0702876SJeff Layton u32 xlen; 209f0702876SJeff Layton 210f0702876SJeff Layton /* Truncate the extent at the end of the current block */ 211f0702876SJeff Layton ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len, 212f0702876SJeff Layton &objno, &objoff, &xlen); 213f0702876SJeff Layton subreq->len = min(xlen, fsc->mount_options->rsize); 214f0702876SJeff Layton return true; 215f0702876SJeff Layton } 216f0702876SJeff Layton 217f0702876SJeff Layton static void finish_netfs_read(struct ceph_osd_request *req) 218f0702876SJeff Layton { 219f0702876SJeff Layton struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode); 220f0702876SJeff Layton struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); 2216a19114bSDavid Howells struct netfs_io_subrequest *subreq = req->r_priv; 222f0702876SJeff Layton int num_pages; 223f0702876SJeff Layton int err = req->r_result; 224f0702876SJeff Layton 2258ae99ae2SXiubo Li ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency, 226903f4fecSXiubo Li req->r_end_latency, osd_data->length, err); 227f0702876SJeff Layton 228f0702876SJeff Layton dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result, 229f0702876SJeff Layton subreq->len, i_size_read(req->r_inode)); 230f0702876SJeff Layton 231f0702876SJeff Layton /* no object means success but no data */ 232f0702876SJeff Layton if (err == -ENOENT) 233f0702876SJeff Layton err = 0; 234f0702876SJeff Layton else if (err == -EBLOCKLISTED) 235f0702876SJeff Layton fsc->blocklisted = true; 236f0702876SJeff Layton 237f0702876SJeff Layton if (err >= 0 && err < subreq->len) 238f0702876SJeff Layton __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags); 239f0702876SJeff Layton 2407467b044SJeff Layton netfs_subreq_terminated(subreq, err, false); 241f0702876SJeff Layton 242f0702876SJeff Layton num_pages = calc_pages_for(osd_data->alignment, osd_data->length); 243f0702876SJeff Layton ceph_put_page_vector(osd_data->pages, num_pages, false); 244f0702876SJeff Layton iput(req->r_inode); 245f0702876SJeff Layton } 246f0702876SJeff Layton 2476a19114bSDavid Howells static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq) 2485b19f1ebSDavid Howells { 2496a19114bSDavid Howells struct netfs_io_request *rreq = subreq->rreq; 2505b19f1ebSDavid Howells struct inode *inode = rreq->inode; 2515b19f1ebSDavid Howells struct ceph_mds_reply_info_parsed *rinfo; 2525b19f1ebSDavid Howells struct ceph_mds_reply_info_in *iinfo; 2535b19f1ebSDavid Howells struct ceph_mds_request *req; 2545b19f1ebSDavid Howells struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 2555b19f1ebSDavid Howells struct ceph_inode_info *ci = ceph_inode(inode); 2565b19f1ebSDavid Howells struct iov_iter iter; 2575b19f1ebSDavid Howells ssize_t err = 0; 2585b19f1ebSDavid Howells size_t len; 2595eed80fbSXiubo Li int mode; 2605b19f1ebSDavid Howells 2615b19f1ebSDavid Howells __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags); 262f18a3785SDavid Howells __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags); 2635b19f1ebSDavid Howells 2645b19f1ebSDavid Howells if (subreq->start >= inode->i_size) 2655b19f1ebSDavid Howells goto out; 2665b19f1ebSDavid Howells 2675b19f1ebSDavid Howells /* We need to fetch the inline data. */ 2685eed80fbSXiubo Li mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA); 2695eed80fbSXiubo Li req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2705b19f1ebSDavid Howells if (IS_ERR(req)) { 2715b19f1ebSDavid Howells err = PTR_ERR(req); 2725b19f1ebSDavid Howells goto out; 2735b19f1ebSDavid Howells } 2745b19f1ebSDavid Howells req->r_ino1 = ci->i_vino; 2755b19f1ebSDavid Howells req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA); 2765b19f1ebSDavid Howells req->r_num_caps = 2; 2775b19f1ebSDavid Howells 2785b19f1ebSDavid Howells err = ceph_mdsc_do_request(mdsc, NULL, req); 2795b19f1ebSDavid Howells if (err < 0) 2805b19f1ebSDavid Howells goto out; 2815b19f1ebSDavid Howells 2825b19f1ebSDavid Howells rinfo = &req->r_reply_info; 2835b19f1ebSDavid Howells iinfo = &rinfo->targeti; 2845b19f1ebSDavid Howells if (iinfo->inline_version == CEPH_INLINE_NONE) { 2855b19f1ebSDavid Howells /* The data got uninlined */ 2865b19f1ebSDavid Howells ceph_mdsc_put_request(req); 2875b19f1ebSDavid Howells return false; 2885b19f1ebSDavid Howells } 2895b19f1ebSDavid Howells 2905b19f1ebSDavid Howells len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len); 2915b19f1ebSDavid Howells iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len); 2925b19f1ebSDavid Howells err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter); 2935b19f1ebSDavid Howells if (err == 0) 2945b19f1ebSDavid Howells err = -EFAULT; 2955b19f1ebSDavid Howells 2965b19f1ebSDavid Howells ceph_mdsc_put_request(req); 2975b19f1ebSDavid Howells out: 2985b19f1ebSDavid Howells netfs_subreq_terminated(subreq, err, false); 2995b19f1ebSDavid Howells return true; 3005b19f1ebSDavid Howells } 3015b19f1ebSDavid Howells 302f18a3785SDavid Howells static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq) 303f0702876SJeff Layton { 3046a19114bSDavid Howells struct netfs_io_request *rreq = subreq->rreq; 305a25cedb4SJeff Layton struct inode *inode = rreq->inode; 306f0702876SJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 307f0702876SJeff Layton struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 308f0702876SJeff Layton struct ceph_osd_request *req; 309f0702876SJeff Layton struct ceph_vino vino = ceph_vino(inode); 310f0702876SJeff Layton struct iov_iter iter; 311f0702876SJeff Layton struct page **pages; 312f0702876SJeff Layton size_t page_off; 313f0702876SJeff Layton int err = 0; 314f0702876SJeff Layton u64 len = subreq->len; 315f0702876SJeff Layton 31648490776SXiubo Li if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq)) 3175b19f1ebSDavid Howells return; 3185b19f1ebSDavid Howells 319f0702876SJeff Layton req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len, 320f0702876SJeff Layton 0, 1, CEPH_OSD_OP_READ, 321f0702876SJeff Layton CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica, 322f0702876SJeff Layton NULL, ci->i_truncate_seq, ci->i_truncate_size, false); 323f0702876SJeff Layton if (IS_ERR(req)) { 324f0702876SJeff Layton err = PTR_ERR(req); 325f0702876SJeff Layton req = NULL; 326f0702876SJeff Layton goto out; 327f0702876SJeff Layton } 328f0702876SJeff Layton 329f0702876SJeff Layton dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len); 330f0702876SJeff Layton iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len); 331b5358992SAl Viro err = iov_iter_get_pages_alloc2(&iter, &pages, len, &page_off); 332f0702876SJeff Layton if (err < 0) { 333f0702876SJeff Layton dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err); 334f0702876SJeff Layton goto out; 335f0702876SJeff Layton } 336f0702876SJeff Layton 337f0702876SJeff Layton /* should always give us a page-aligned read */ 338f0702876SJeff Layton WARN_ON_ONCE(page_off); 339f0702876SJeff Layton len = err; 340a8af0d68SJeff Layton err = 0; 341f0702876SJeff Layton 342f0702876SJeff Layton osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); 343f0702876SJeff Layton req->r_callback = finish_netfs_read; 344f0702876SJeff Layton req->r_priv = subreq; 345f0702876SJeff Layton req->r_inode = inode; 346f0702876SJeff Layton ihold(inode); 347f0702876SJeff Layton 348a8af0d68SJeff Layton ceph_osdc_start_request(req->r_osdc, req); 349f0702876SJeff Layton out: 350f0702876SJeff Layton ceph_osdc_put_request(req); 351f0702876SJeff Layton if (err) 352f0702876SJeff Layton netfs_subreq_terminated(subreq, err, false); 353f0702876SJeff Layton dout("%s: result %d\n", __func__, err); 354f0702876SJeff Layton } 355f0702876SJeff Layton 356a5c9dc44SDavid Howells static int ceph_init_request(struct netfs_io_request *rreq, struct file *file) 357a5c9dc44SDavid Howells { 358a5c9dc44SDavid Howells struct inode *inode = rreq->inode; 359a5c9dc44SDavid Howells int got = 0, want = CEPH_CAP_FILE_CACHE; 360a5c9dc44SDavid Howells int ret = 0; 361a5c9dc44SDavid Howells 362a5c9dc44SDavid Howells if (rreq->origin != NETFS_READAHEAD) 363a5c9dc44SDavid Howells return 0; 364a5c9dc44SDavid Howells 365a5c9dc44SDavid Howells if (file) { 366a5c9dc44SDavid Howells struct ceph_rw_context *rw_ctx; 367a5c9dc44SDavid Howells struct ceph_file_info *fi = file->private_data; 368a5c9dc44SDavid Howells 369a5c9dc44SDavid Howells rw_ctx = ceph_find_rw_context(fi); 370a5c9dc44SDavid Howells if (rw_ctx) 371a5c9dc44SDavid Howells return 0; 372a5c9dc44SDavid Howells } 373a5c9dc44SDavid Howells 374a5c9dc44SDavid Howells /* 375a5c9dc44SDavid Howells * readahead callers do not necessarily hold Fcb caps 376a5c9dc44SDavid Howells * (e.g. fadvise, madvise). 377a5c9dc44SDavid Howells */ 378a5c9dc44SDavid Howells ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got); 379a5c9dc44SDavid Howells if (ret < 0) { 380a5c9dc44SDavid Howells dout("start_read %p, error getting cap\n", inode); 381a5c9dc44SDavid Howells return ret; 382a5c9dc44SDavid Howells } 383a5c9dc44SDavid Howells 384a5c9dc44SDavid Howells if (!(got & want)) { 385a5c9dc44SDavid Howells dout("start_read %p, no cache cap\n", inode); 386a5c9dc44SDavid Howells return -EACCES; 387a5c9dc44SDavid Howells } 388a5c9dc44SDavid Howells if (ret == 0) 389a5c9dc44SDavid Howells return -EACCES; 390a5c9dc44SDavid Howells 391a5c9dc44SDavid Howells rreq->netfs_priv = (void *)(uintptr_t)got; 392a5c9dc44SDavid Howells return 0; 393a5c9dc44SDavid Howells } 394a5c9dc44SDavid Howells 39540a81101SDavid Howells static void ceph_netfs_free_request(struct netfs_io_request *rreq) 39649870056SJeff Layton { 39740a81101SDavid Howells struct ceph_inode_info *ci = ceph_inode(rreq->inode); 39840a81101SDavid Howells int got = (uintptr_t)rreq->netfs_priv; 39949870056SJeff Layton 40049870056SJeff Layton if (got) 40149870056SJeff Layton ceph_put_cap_refs(ci, got); 40249870056SJeff Layton } 40349870056SJeff Layton 404bc899ee1SDavid Howells const struct netfs_request_ops ceph_netfs_ops = { 405a5c9dc44SDavid Howells .init_request = ceph_init_request, 40640a81101SDavid Howells .free_request = ceph_netfs_free_request, 407f0702876SJeff Layton .begin_cache_operation = ceph_begin_cache_operation, 408f18a3785SDavid Howells .issue_read = ceph_netfs_issue_read, 409f0702876SJeff Layton .expand_readahead = ceph_netfs_expand_readahead, 410f0702876SJeff Layton .clamp_length = ceph_netfs_clamp_length, 411d801327dSJeff Layton .check_write_begin = ceph_netfs_check_write_begin, 412f0702876SJeff Layton }; 413f0702876SJeff Layton 4141702e797SJeff Layton #ifdef CONFIG_CEPH_FSCACHE 4151702e797SJeff Layton static void ceph_set_page_fscache(struct page *page) 4161702e797SJeff Layton { 4171702e797SJeff Layton set_page_fscache(page); 4181702e797SJeff Layton } 4191702e797SJeff Layton 4201702e797SJeff Layton static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async) 4211702e797SJeff Layton { 4221702e797SJeff Layton struct inode *inode = priv; 4231702e797SJeff Layton 4241702e797SJeff Layton if (IS_ERR_VALUE(error) && error != -ENOBUFS) 4251702e797SJeff Layton ceph_fscache_invalidate(inode, false); 4261702e797SJeff Layton } 4271702e797SJeff Layton 4281702e797SJeff Layton static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) 4291702e797SJeff Layton { 4301702e797SJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 4311702e797SJeff Layton struct fscache_cookie *cookie = ceph_fscache_cookie(ci); 4321702e797SJeff Layton 4331702e797SJeff Layton fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode), 4341702e797SJeff Layton ceph_fscache_write_terminated, inode, caching); 4351702e797SJeff Layton } 4361702e797SJeff Layton #else 4371702e797SJeff Layton static inline void ceph_set_page_fscache(struct page *page) 4381702e797SJeff Layton { 4391702e797SJeff Layton } 4401702e797SJeff Layton 4411702e797SJeff Layton static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) 4421702e797SJeff Layton { 4431702e797SJeff Layton } 4441702e797SJeff Layton #endif /* CONFIG_CEPH_FSCACHE */ 4451702e797SJeff Layton 4461f934b00SYan, Zheng struct ceph_writeback_ctl 4471f934b00SYan, Zheng { 4481f934b00SYan, Zheng loff_t i_size; 4491f934b00SYan, Zheng u64 truncate_size; 4501f934b00SYan, Zheng u32 truncate_seq; 4511f934b00SYan, Zheng bool size_stable; 4522a2d927eSYan, Zheng bool head_snapc; 4531f934b00SYan, Zheng }; 4541f934b00SYan, Zheng 4551d3576fdSSage Weil /* 4561d3576fdSSage Weil * Get ref for the oldest snapc for an inode with dirty data... that is, the 4571d3576fdSSage Weil * only snap context we are allowed to write back. 4581d3576fdSSage Weil */ 4591f934b00SYan, Zheng static struct ceph_snap_context * 46005455e11SYan, Zheng get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, 46105455e11SYan, Zheng struct ceph_snap_context *page_snapc) 4621d3576fdSSage Weil { 4631d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 4641d3576fdSSage Weil struct ceph_snap_context *snapc = NULL; 4651d3576fdSSage Weil struct ceph_cap_snap *capsnap = NULL; 4661d3576fdSSage Weil 467be655596SSage Weil spin_lock(&ci->i_ceph_lock); 4681d3576fdSSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 4691d3576fdSSage Weil dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 4701d3576fdSSage Weil capsnap->context, capsnap->dirty_pages); 47105455e11SYan, Zheng if (!capsnap->dirty_pages) 47205455e11SYan, Zheng continue; 47305455e11SYan, Zheng 47405455e11SYan, Zheng /* get i_size, truncate_{seq,size} for page_snapc? */ 47505455e11SYan, Zheng if (snapc && capsnap->context != page_snapc) 47605455e11SYan, Zheng continue; 47705455e11SYan, Zheng 4781f934b00SYan, Zheng if (ctl) { 4791f934b00SYan, Zheng if (capsnap->writing) { 4801f934b00SYan, Zheng ctl->i_size = i_size_read(inode); 4811f934b00SYan, Zheng ctl->size_stable = false; 4821f934b00SYan, Zheng } else { 4831f934b00SYan, Zheng ctl->i_size = capsnap->size; 4841f934b00SYan, Zheng ctl->size_stable = true; 4851f934b00SYan, Zheng } 4861f934b00SYan, Zheng ctl->truncate_size = capsnap->truncate_size; 4871f934b00SYan, Zheng ctl->truncate_seq = capsnap->truncate_seq; 4882a2d927eSYan, Zheng ctl->head_snapc = false; 4891f934b00SYan, Zheng } 49005455e11SYan, Zheng 49105455e11SYan, Zheng if (snapc) 4921d3576fdSSage Weil break; 49305455e11SYan, Zheng 49405455e11SYan, Zheng snapc = ceph_get_snap_context(capsnap->context); 49505455e11SYan, Zheng if (!page_snapc || 49605455e11SYan, Zheng page_snapc == snapc || 49705455e11SYan, Zheng page_snapc->seq > snapc->seq) 49805455e11SYan, Zheng break; 4991d3576fdSSage Weil } 5007d8cb26dSSage Weil if (!snapc && ci->i_wrbuffer_ref_head) { 50180e755feSSage Weil snapc = ceph_get_snap_context(ci->i_head_snapc); 5021d3576fdSSage Weil dout(" head snapc %p has %d dirty pages\n", 5031d3576fdSSage Weil snapc, ci->i_wrbuffer_ref_head); 5041f934b00SYan, Zheng if (ctl) { 5051f934b00SYan, Zheng ctl->i_size = i_size_read(inode); 5061f934b00SYan, Zheng ctl->truncate_size = ci->i_truncate_size; 5071f934b00SYan, Zheng ctl->truncate_seq = ci->i_truncate_seq; 5081f934b00SYan, Zheng ctl->size_stable = false; 5092a2d927eSYan, Zheng ctl->head_snapc = true; 5101f934b00SYan, Zheng } 5111d3576fdSSage Weil } 512be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 5131d3576fdSSage Weil return snapc; 5141d3576fdSSage Weil } 5151d3576fdSSage Weil 5161f934b00SYan, Zheng static u64 get_writepages_data_length(struct inode *inode, 5171f934b00SYan, Zheng struct page *page, u64 start) 5181f934b00SYan, Zheng { 5191f934b00SYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 5201f934b00SYan, Zheng struct ceph_snap_context *snapc = page_snap_context(page); 5211f934b00SYan, Zheng struct ceph_cap_snap *capsnap = NULL; 5221f934b00SYan, Zheng u64 end = i_size_read(inode); 5231f934b00SYan, Zheng 5241f934b00SYan, Zheng if (snapc != ci->i_head_snapc) { 5251f934b00SYan, Zheng bool found = false; 5261f934b00SYan, Zheng spin_lock(&ci->i_ceph_lock); 5271f934b00SYan, Zheng list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 5281f934b00SYan, Zheng if (capsnap->context == snapc) { 5291f934b00SYan, Zheng if (!capsnap->writing) 5301f934b00SYan, Zheng end = capsnap->size; 5311f934b00SYan, Zheng found = true; 5321f934b00SYan, Zheng break; 5331f934b00SYan, Zheng } 5341f934b00SYan, Zheng } 5351f934b00SYan, Zheng spin_unlock(&ci->i_ceph_lock); 5361f934b00SYan, Zheng WARN_ON(!found); 5371f934b00SYan, Zheng } 5388ff2d290SJeff Layton if (end > page_offset(page) + thp_size(page)) 5398ff2d290SJeff Layton end = page_offset(page) + thp_size(page); 5401f934b00SYan, Zheng return end > start ? end - start : 0; 5411f934b00SYan, Zheng } 5421f934b00SYan, Zheng 5431d3576fdSSage Weil /* 5441d3576fdSSage Weil * Write a single page, but leave the page locked. 5451d3576fdSSage Weil * 546b72b13ebSJeff Layton * If we get a write error, mark the mapping for error, but still adjust the 5471d3576fdSSage Weil * dirty page accounting (i.e., page is no longer dirty). 5481d3576fdSSage Weil */ 5491d3576fdSSage Weil static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 5501d3576fdSSage Weil { 551a628304eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 5526390987fSJeff Layton struct inode *inode = page->mapping->host; 5536390987fSJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 5546390987fSJeff Layton struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 5556298a337SSage Weil struct ceph_snap_context *snapc, *oldest; 556fc2744aaSYan, Zheng loff_t page_off = page_offset(page); 5576390987fSJeff Layton int err; 5588ff2d290SJeff Layton loff_t len = thp_size(page); 5591f934b00SYan, Zheng struct ceph_writeback_ctl ceph_wbc; 5606390987fSJeff Layton struct ceph_osd_client *osdc = &fsc->client->osdc; 5616390987fSJeff Layton struct ceph_osd_request *req; 5621702e797SJeff Layton bool caching = ceph_is_cache_enabled(inode); 5631d3576fdSSage Weil 5641d3576fdSSage Weil dout("writepage %p idx %lu\n", page, page->index); 5651d3576fdSSage Weil 5661d3576fdSSage Weil /* verify this is a writeable snap context */ 56761600ef8SYan, Zheng snapc = page_snap_context(page); 568d37b1d99SMarkus Elfring if (!snapc) { 5691d3576fdSSage Weil dout("writepage %p page %p not dirty?\n", inode, page); 57043986881SYan, Zheng return 0; 5711d3576fdSSage Weil } 57205455e11SYan, Zheng oldest = get_oldest_context(inode, &ceph_wbc, snapc); 5736298a337SSage Weil if (snapc->seq > oldest->seq) { 5741d3576fdSSage Weil dout("writepage %p page %p snapc %p not writeable - noop\n", 57561600ef8SYan, Zheng inode, page, snapc); 5761d3576fdSSage Weil /* we should only noop if called by kswapd */ 577fa71fefbSYan, Zheng WARN_ON(!(current->flags & PF_MEMALLOC)); 5786298a337SSage Weil ceph_put_snap_context(oldest); 579fa71fefbSYan, Zheng redirty_page_for_writepage(wbc, page); 58043986881SYan, Zheng return 0; 5811d3576fdSSage Weil } 5826298a337SSage Weil ceph_put_snap_context(oldest); 5831d3576fdSSage Weil 5841d3576fdSSage Weil /* is this a partial page at end of file? */ 5851f934b00SYan, Zheng if (page_off >= ceph_wbc.i_size) { 586a628304eSMatthew Wilcox (Oracle) dout("folio at %lu beyond eof %llu\n", folio->index, 587a628304eSMatthew Wilcox (Oracle) ceph_wbc.i_size); 588a628304eSMatthew Wilcox (Oracle) folio_invalidate(folio, 0, folio_size(folio)); 58943986881SYan, Zheng return 0; 590fc2744aaSYan, Zheng } 59143986881SYan, Zheng 5921f934b00SYan, Zheng if (ceph_wbc.i_size < page_off + len) 5931f934b00SYan, Zheng len = ceph_wbc.i_size - page_off; 5941d3576fdSSage Weil 5956390987fSJeff Layton dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n", 5961c0a9c2dSYan, Zheng inode, page, page->index, page_off, len, snapc, snapc->seq); 5971d3576fdSSage Weil 598314c4737SYan, Zheng if (atomic_long_inc_return(&fsc->writeback_count) > 5993d14c5d2SYehuda Sadeh CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 600503d4fa6SNeilBrown fsc->write_congested = true; 6012baba250SYehuda Sadeh 6026390987fSJeff Layton req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1, 6036390987fSJeff Layton CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc, 6046390987fSJeff Layton ceph_wbc.truncate_seq, ceph_wbc.truncate_size, 6056390987fSJeff Layton true); 6063459bd0cSXiubo Li if (IS_ERR(req)) { 6073459bd0cSXiubo Li redirty_page_for_writepage(wbc, page); 6086390987fSJeff Layton return PTR_ERR(req); 6093459bd0cSXiubo Li } 6101702e797SJeff Layton 6111702e797SJeff Layton set_page_writeback(page); 6121702e797SJeff Layton if (caching) 6131702e797SJeff Layton ceph_set_page_fscache(page); 6141702e797SJeff Layton ceph_fscache_write_to_cache(inode, page_off, len, caching); 6156390987fSJeff Layton 6166390987fSJeff Layton /* it may be a short write due to an object boundary */ 6178ff2d290SJeff Layton WARN_ON_ONCE(len > thp_size(page)); 6186390987fSJeff Layton osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false); 6196390987fSJeff Layton dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len); 6206390987fSJeff Layton 6216390987fSJeff Layton req->r_mtime = inode->i_mtime; 622a8af0d68SJeff Layton ceph_osdc_start_request(osdc, req); 6236390987fSJeff Layton err = ceph_osdc_wait_request(osdc, req); 6246390987fSJeff Layton 6258ae99ae2SXiubo Li ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 626903f4fecSXiubo Li req->r_end_latency, len, err); 6276390987fSJeff Layton 6286390987fSJeff Layton ceph_osdc_put_request(req); 6296390987fSJeff Layton if (err == 0) 6306390987fSJeff Layton err = len; 6316390987fSJeff Layton 6321d3576fdSSage Weil if (err < 0) { 633ad15ec06SYan, Zheng struct writeback_control tmp_wbc; 634ad15ec06SYan, Zheng if (!wbc) 635ad15ec06SYan, Zheng wbc = &tmp_wbc; 636ad15ec06SYan, Zheng if (err == -ERESTARTSYS) { 637ad15ec06SYan, Zheng /* killed by SIGKILL */ 638ad15ec06SYan, Zheng dout("writepage interrupted page %p\n", page); 639ad15ec06SYan, Zheng redirty_page_for_writepage(wbc, page); 640ad15ec06SYan, Zheng end_page_writeback(page); 64143986881SYan, Zheng return err; 642ad15ec06SYan, Zheng } 6430b98acd6SIlya Dryomov if (err == -EBLOCKLISTED) 6440b98acd6SIlya Dryomov fsc->blocklisted = true; 645ad15ec06SYan, Zheng dout("writepage setting page/mapping error %d %p\n", 646ad15ec06SYan, Zheng err, page); 6471d3576fdSSage Weil mapping_set_error(&inode->i_data, err); 6481d3576fdSSage Weil wbc->pages_skipped++; 6491d3576fdSSage Weil } else { 6501d3576fdSSage Weil dout("writepage cleaned page %p\n", page); 6511d3576fdSSage Weil err = 0; /* vfs expects us to return 0 */ 6521d3576fdSSage Weil } 653379fc7faSJeff Layton oldest = detach_page_private(page); 654379fc7faSJeff Layton WARN_ON_ONCE(oldest != snapc); 6551d3576fdSSage Weil end_page_writeback(page); 6561d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 6576298a337SSage Weil ceph_put_snap_context(snapc); /* page's reference */ 658314c4737SYan, Zheng 659314c4737SYan, Zheng if (atomic_long_dec_return(&fsc->writeback_count) < 660314c4737SYan, Zheng CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) 661503d4fa6SNeilBrown fsc->write_congested = false; 662314c4737SYan, Zheng 6631d3576fdSSage Weil return err; 6641d3576fdSSage Weil } 6651d3576fdSSage Weil 6661d3576fdSSage Weil static int ceph_writepage(struct page *page, struct writeback_control *wbc) 6671d3576fdSSage Weil { 668dbd646a8SYehuda Sadeh int err; 669dbd646a8SYehuda Sadeh struct inode *inode = page->mapping->host; 670dbd646a8SYehuda Sadeh BUG_ON(!inode); 67170b666c3SSage Weil ihold(inode); 6721702e797SJeff Layton 673503d4fa6SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 674503d4fa6SNeilBrown ceph_inode_to_client(inode)->write_congested) 675503d4fa6SNeilBrown return AOP_WRITEPAGE_ACTIVATE; 676503d4fa6SNeilBrown 6771702e797SJeff Layton wait_on_page_fscache(page); 6781702e797SJeff Layton 679dbd646a8SYehuda Sadeh err = writepage_nounlock(page, wbc); 680ad15ec06SYan, Zheng if (err == -ERESTARTSYS) { 681ad15ec06SYan, Zheng /* direct memory reclaimer was killed by SIGKILL. return 0 682ad15ec06SYan, Zheng * to prevent caller from setting mapping/page error */ 683ad15ec06SYan, Zheng err = 0; 684ad15ec06SYan, Zheng } 6851d3576fdSSage Weil unlock_page(page); 686dbd646a8SYehuda Sadeh iput(inode); 6871d3576fdSSage Weil return err; 6881d3576fdSSage Weil } 6891d3576fdSSage Weil 6901d3576fdSSage Weil /* 6911d3576fdSSage Weil * async writeback completion handler. 6921d3576fdSSage Weil * 6931d3576fdSSage Weil * If we get an error, set the mapping error bit, but not the individual 6941d3576fdSSage Weil * page error bits. 6951d3576fdSSage Weil */ 69685e084feSIlya Dryomov static void writepages_finish(struct ceph_osd_request *req) 6971d3576fdSSage Weil { 6981d3576fdSSage Weil struct inode *inode = req->r_inode; 6991d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 70087060c10SAlex Elder struct ceph_osd_data *osd_data; 7011d3576fdSSage Weil struct page *page; 7025b64640cSYan, Zheng int num_pages, total_pages = 0; 7035b64640cSYan, Zheng int i, j; 7045b64640cSYan, Zheng int rc = req->r_result; 7051d3576fdSSage Weil struct ceph_snap_context *snapc = req->r_snapc; 7061d3576fdSSage Weil struct address_space *mapping = inode->i_mapping; 7073d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 708903f4fecSXiubo Li unsigned int len = 0; 7095b64640cSYan, Zheng bool remove_page; 7101d3576fdSSage Weil 7115b64640cSYan, Zheng dout("writepages_finish %p rc %d\n", inode, rc); 71226544c62SJeff Layton if (rc < 0) { 7131d3576fdSSage Weil mapping_set_error(mapping, rc); 71426544c62SJeff Layton ceph_set_error_write(ci); 7150b98acd6SIlya Dryomov if (rc == -EBLOCKLISTED) 7160b98acd6SIlya Dryomov fsc->blocklisted = true; 71726544c62SJeff Layton } else { 71826544c62SJeff Layton ceph_clear_error_write(ci); 71926544c62SJeff Layton } 720e63dc5c7SYehuda Sadeh 721e63dc5c7SYehuda Sadeh /* 722e63dc5c7SYehuda Sadeh * We lost the cache cap, need to truncate the page before 723e63dc5c7SYehuda Sadeh * it is unlocked, otherwise we'd truncate it later in the 724e63dc5c7SYehuda Sadeh * page truncation thread, possibly losing some data that 725e63dc5c7SYehuda Sadeh * raced its way in 726e63dc5c7SYehuda Sadeh */ 7275b64640cSYan, Zheng remove_page = !(ceph_caps_issued(ci) & 7285b64640cSYan, Zheng (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); 7295b64640cSYan, Zheng 7305b64640cSYan, Zheng /* clean all pages */ 7315b64640cSYan, Zheng for (i = 0; i < req->r_num_ops; i++) { 732642d51fbSXiubo Li if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) { 733642d51fbSXiubo Li pr_warn("%s incorrect op %d req %p index %d tid %llu\n", 734642d51fbSXiubo Li __func__, req->r_ops[i].op, req, i, req->r_tid); 7355b64640cSYan, Zheng break; 736642d51fbSXiubo Li } 7375b64640cSYan, Zheng 7385b64640cSYan, Zheng osd_data = osd_req_op_extent_osd_data(req, i); 7395b64640cSYan, Zheng BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 740903f4fecSXiubo Li len += osd_data->length; 7415b64640cSYan, Zheng num_pages = calc_pages_for((u64)osd_data->alignment, 7425b64640cSYan, Zheng (u64)osd_data->length); 7435b64640cSYan, Zheng total_pages += num_pages; 7445b64640cSYan, Zheng for (j = 0; j < num_pages; j++) { 7455b64640cSYan, Zheng page = osd_data->pages[j]; 7465b64640cSYan, Zheng BUG_ON(!page); 7475b64640cSYan, Zheng WARN_ON(!PageUptodate(page)); 7485b64640cSYan, Zheng 7495b64640cSYan, Zheng if (atomic_long_dec_return(&fsc->writeback_count) < 7505b64640cSYan, Zheng CONGESTION_OFF_THRESH( 7515b64640cSYan, Zheng fsc->mount_options->congestion_kb)) 752503d4fa6SNeilBrown fsc->write_congested = false; 7535b64640cSYan, Zheng 754379fc7faSJeff Layton ceph_put_snap_context(detach_page_private(page)); 7555b64640cSYan, Zheng end_page_writeback(page); 756379fc7faSJeff Layton dout("unlocking %p\n", page); 7575b64640cSYan, Zheng 7585b64640cSYan, Zheng if (remove_page) 7595b64640cSYan, Zheng generic_error_remove_page(inode->i_mapping, 7605b64640cSYan, Zheng page); 761e63dc5c7SYehuda Sadeh 7621d3576fdSSage Weil unlock_page(page); 7631d3576fdSSage Weil } 7645b64640cSYan, Zheng dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", 7655b64640cSYan, Zheng inode, osd_data->length, rc >= 0 ? num_pages : 0); 7661d3576fdSSage Weil 76796ac9158SJohn Hubbard release_pages(osd_data->pages, num_pages); 7685b64640cSYan, Zheng } 7695b64640cSYan, Zheng 770903f4fecSXiubo Li ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 771903f4fecSXiubo Li req->r_end_latency, len, rc); 772903f4fecSXiubo Li 7735b64640cSYan, Zheng ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); 7745b64640cSYan, Zheng 7755b64640cSYan, Zheng osd_data = osd_req_op_extent_osd_data(req, 0); 77687060c10SAlex Elder if (osd_data->pages_from_pool) 777a0102bdaSJeff Layton mempool_free(osd_data->pages, ceph_wb_pagevec_pool); 7781d3576fdSSage Weil else 77987060c10SAlex Elder kfree(osd_data->pages); 7801d3576fdSSage Weil ceph_osdc_put_request(req); 7811d3576fdSSage Weil } 7821d3576fdSSage Weil 7831d3576fdSSage Weil /* 7841d3576fdSSage Weil * initiate async writeback 7851d3576fdSSage Weil */ 7861d3576fdSSage Weil static int ceph_writepages_start(struct address_space *mapping, 7871d3576fdSSage Weil struct writeback_control *wbc) 7881d3576fdSSage Weil { 7891d3576fdSSage Weil struct inode *inode = mapping->host; 7901d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 791fc2744aaSYan, Zheng struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 792fc2744aaSYan, Zheng struct ceph_vino vino = ceph_vino(inode); 7932a2d927eSYan, Zheng pgoff_t index, start_index, end = -1; 79480e755feSSage Weil struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 7951d3576fdSSage Weil struct pagevec pvec; 7961d3576fdSSage Weil int rc = 0; 79793407472SFabian Frederick unsigned int wsize = i_blocksize(inode); 7981d3576fdSSage Weil struct ceph_osd_request *req = NULL; 7991f934b00SYan, Zheng struct ceph_writeback_ctl ceph_wbc; 800590e9d98SYan, Zheng bool should_loop, range_whole = false; 801af9cc401SYan, Zheng bool done = false; 8021702e797SJeff Layton bool caching = ceph_is_cache_enabled(inode); 8031d3576fdSSage Weil 804503d4fa6SNeilBrown if (wbc->sync_mode == WB_SYNC_NONE && 805503d4fa6SNeilBrown fsc->write_congested) 806503d4fa6SNeilBrown return 0; 807503d4fa6SNeilBrown 8083fb99d48SYanhu Cao dout("writepages_start %p (mode=%s)\n", inode, 8091d3576fdSSage Weil wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 8101d3576fdSSage Weil (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 8111d3576fdSSage Weil 8125d6451b1SJeff Layton if (ceph_inode_is_shutdown(inode)) { 8136c93df5dSYan, Zheng if (ci->i_wrbuffer_ref > 0) { 8146c93df5dSYan, Zheng pr_warn_ratelimited( 8156c93df5dSYan, Zheng "writepage_start %p %lld forced umount\n", 8166c93df5dSYan, Zheng inode, ceph_ino(inode)); 8176c93df5dSYan, Zheng } 818a341d4dfSYan, Zheng mapping_set_error(mapping, -EIO); 8191d3576fdSSage Weil return -EIO; /* we're in a forced umount, don't write! */ 8201d3576fdSSage Weil } 82195cca2b4SYan, Zheng if (fsc->mount_options->wsize < wsize) 8223d14c5d2SYehuda Sadeh wsize = fsc->mount_options->wsize; 8231d3576fdSSage Weil 82486679820SMel Gorman pagevec_init(&pvec); 8251d3576fdSSage Weil 826590e9d98SYan, Zheng start_index = wbc->range_cyclic ? mapping->writeback_index : 0; 827590e9d98SYan, Zheng index = start_index; 8281d3576fdSSage Weil 8291d3576fdSSage Weil retry: 8301d3576fdSSage Weil /* find oldest snap context with dirty data */ 83105455e11SYan, Zheng snapc = get_oldest_context(inode, &ceph_wbc, NULL); 8321d3576fdSSage Weil if (!snapc) { 8331d3576fdSSage Weil /* hmm, why does writepages get called when there 8341d3576fdSSage Weil is no dirty data? */ 8351d3576fdSSage Weil dout(" no snap context with dirty data?\n"); 8361d3576fdSSage Weil goto out; 8371d3576fdSSage Weil } 8381d3576fdSSage Weil dout(" oldest snapc is %p seq %lld (%d snaps)\n", 8391d3576fdSSage Weil snapc, snapc->seq, snapc->num_snaps); 840fc2744aaSYan, Zheng 8412a2d927eSYan, Zheng should_loop = false; 8422a2d927eSYan, Zheng if (ceph_wbc.head_snapc && snapc != last_snapc) { 8432a2d927eSYan, Zheng /* where to start/end? */ 8442a2d927eSYan, Zheng if (wbc->range_cyclic) { 8452a2d927eSYan, Zheng index = start_index; 8462a2d927eSYan, Zheng end = -1; 8472a2d927eSYan, Zheng if (index > 0) 8482a2d927eSYan, Zheng should_loop = true; 8492a2d927eSYan, Zheng dout(" cyclic, start at %lu\n", index); 8502a2d927eSYan, Zheng } else { 8512a2d927eSYan, Zheng index = wbc->range_start >> PAGE_SHIFT; 8522a2d927eSYan, Zheng end = wbc->range_end >> PAGE_SHIFT; 8532a2d927eSYan, Zheng if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 8542a2d927eSYan, Zheng range_whole = true; 8552a2d927eSYan, Zheng dout(" not cyclic, %lu to %lu\n", index, end); 8561d3576fdSSage Weil } 8572a2d927eSYan, Zheng } else if (!ceph_wbc.head_snapc) { 8582a2d927eSYan, Zheng /* Do not respect wbc->range_{start,end}. Dirty pages 8592a2d927eSYan, Zheng * in that range can be associated with newer snapc. 8602a2d927eSYan, Zheng * They are not writeable until we write all dirty pages 8612a2d927eSYan, Zheng * associated with 'snapc' get written */ 8621582af2eSYan, Zheng if (index > 0) 8632a2d927eSYan, Zheng should_loop = true; 8642a2d927eSYan, Zheng dout(" non-head snapc, range whole\n"); 8652a2d927eSYan, Zheng } 8662a2d927eSYan, Zheng 8672a2d927eSYan, Zheng ceph_put_snap_context(last_snapc); 8681d3576fdSSage Weil last_snapc = snapc; 8691d3576fdSSage Weil 870af9cc401SYan, Zheng while (!done && index <= end) { 8715b64640cSYan, Zheng int num_ops = 0, op_idx; 8720e5ecac7SYan, Zheng unsigned i, pvec_pages, max_pages, locked_pages = 0; 8735b64640cSYan, Zheng struct page **pages = NULL, **data_pages; 8741d3576fdSSage Weil struct page *page; 8750e5ecac7SYan, Zheng pgoff_t strip_unit_end = 0; 8765b64640cSYan, Zheng u64 offset = 0, len = 0; 877a0102bdaSJeff Layton bool from_pool = false; 8781d3576fdSSage Weil 8790e5ecac7SYan, Zheng max_pages = wsize >> PAGE_SHIFT; 8801d3576fdSSage Weil 8811d3576fdSSage Weil get_more_pages: 8822e169296SJeff Layton pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, 8832e169296SJeff Layton end, PAGECACHE_TAG_DIRTY); 8840ed75fc8SJan Kara dout("pagevec_lookup_range_tag got %d\n", pvec_pages); 8851d3576fdSSage Weil if (!pvec_pages && !locked_pages) 8861d3576fdSSage Weil break; 8871d3576fdSSage Weil for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 8881d3576fdSSage Weil page = pvec.pages[i]; 8891d3576fdSSage Weil dout("? %p idx %lu\n", page, page->index); 8901d3576fdSSage Weil if (locked_pages == 0) 8911d3576fdSSage Weil lock_page(page); /* first page */ 8921d3576fdSSage Weil else if (!trylock_page(page)) 8931d3576fdSSage Weil break; 8941d3576fdSSage Weil 8951d3576fdSSage Weil /* only dirty pages, or our accounting breaks */ 8961d3576fdSSage Weil if (unlikely(!PageDirty(page)) || 8971d3576fdSSage Weil unlikely(page->mapping != mapping)) { 8981d3576fdSSage Weil dout("!dirty or !mapping %p\n", page); 8991d3576fdSSage Weil unlock_page(page); 9000713e5f2SYan, Zheng continue; 9011d3576fdSSage Weil } 902af9cc401SYan, Zheng /* only if matching snap context */ 903af9cc401SYan, Zheng pgsnapc = page_snap_context(page); 904af9cc401SYan, Zheng if (pgsnapc != snapc) { 905af9cc401SYan, Zheng dout("page snapc %p %lld != oldest %p %lld\n", 906af9cc401SYan, Zheng pgsnapc, pgsnapc->seq, snapc, snapc->seq); 9071582af2eSYan, Zheng if (!should_loop && 9081582af2eSYan, Zheng !ceph_wbc.head_snapc && 9091582af2eSYan, Zheng wbc->sync_mode != WB_SYNC_NONE) 9101582af2eSYan, Zheng should_loop = true; 9111d3576fdSSage Weil unlock_page(page); 912af9cc401SYan, Zheng continue; 9131d3576fdSSage Weil } 9141f934b00SYan, Zheng if (page_offset(page) >= ceph_wbc.i_size) { 915a628304eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 916a628304eSMatthew Wilcox (Oracle) 917a628304eSMatthew Wilcox (Oracle) dout("folio at %lu beyond eof %llu\n", 918a628304eSMatthew Wilcox (Oracle) folio->index, ceph_wbc.i_size); 919c95f1c5fSErqi Chen if ((ceph_wbc.size_stable || 920a628304eSMatthew Wilcox (Oracle) folio_pos(folio) >= i_size_read(inode)) && 921a628304eSMatthew Wilcox (Oracle) folio_clear_dirty_for_io(folio)) 922a628304eSMatthew Wilcox (Oracle) folio_invalidate(folio, 0, 923a628304eSMatthew Wilcox (Oracle) folio_size(folio)); 924a628304eSMatthew Wilcox (Oracle) folio_unlock(folio); 925af9cc401SYan, Zheng continue; 926af9cc401SYan, Zheng } 927af9cc401SYan, Zheng if (strip_unit_end && (page->index > strip_unit_end)) { 928af9cc401SYan, Zheng dout("end of strip unit %p\n", page); 9291d3576fdSSage Weil unlock_page(page); 9301d3576fdSSage Weil break; 9311d3576fdSSage Weil } 9321702e797SJeff Layton if (PageWriteback(page) || PageFsCache(page)) { 9330713e5f2SYan, Zheng if (wbc->sync_mode == WB_SYNC_NONE) { 9341d3576fdSSage Weil dout("%p under writeback\n", page); 9351d3576fdSSage Weil unlock_page(page); 9360713e5f2SYan, Zheng continue; 9370713e5f2SYan, Zheng } 9380713e5f2SYan, Zheng dout("waiting on writeback %p\n", page); 9390713e5f2SYan, Zheng wait_on_page_writeback(page); 9401702e797SJeff Layton wait_on_page_fscache(page); 9411d3576fdSSage Weil } 9421d3576fdSSage Weil 9431d3576fdSSage Weil if (!clear_page_dirty_for_io(page)) { 9441d3576fdSSage Weil dout("%p !clear_page_dirty_for_io\n", page); 9451d3576fdSSage Weil unlock_page(page); 9460713e5f2SYan, Zheng continue; 9471d3576fdSSage Weil } 9481d3576fdSSage Weil 949e5975c7cSAlex Elder /* 950e5975c7cSAlex Elder * We have something to write. If this is 951e5975c7cSAlex Elder * the first locked page this time through, 9525b64640cSYan, Zheng * calculate max possinle write size and 9535b64640cSYan, Zheng * allocate a page array 954e5975c7cSAlex Elder */ 9551d3576fdSSage Weil if (locked_pages == 0) { 9565b64640cSYan, Zheng u64 objnum; 9575b64640cSYan, Zheng u64 objoff; 958dccbf080SIlya Dryomov u32 xlen; 9595b64640cSYan, Zheng 9601d3576fdSSage Weil /* prepare async write request */ 9616285bc23SAlex Elder offset = (u64)page_offset(page); 962dccbf080SIlya Dryomov ceph_calc_file_object_mapping(&ci->i_layout, 963dccbf080SIlya Dryomov offset, wsize, 9645b64640cSYan, Zheng &objnum, &objoff, 965dccbf080SIlya Dryomov &xlen); 966dccbf080SIlya Dryomov len = xlen; 9678c71897bSHenry C Chang 9683fb99d48SYanhu Cao num_ops = 1; 9695b64640cSYan, Zheng strip_unit_end = page->index + 97009cbfeafSKirill A. Shutemov ((len - 1) >> PAGE_SHIFT); 971715e4cd4SYan, Zheng 9725b64640cSYan, Zheng BUG_ON(pages); 97388486957SAlex Elder max_pages = calc_pages_for(0, (u64)len); 9746da2ec56SKees Cook pages = kmalloc_array(max_pages, 9756da2ec56SKees Cook sizeof(*pages), 976fc2744aaSYan, Zheng GFP_NOFS); 97788486957SAlex Elder if (!pages) { 978a0102bdaSJeff Layton from_pool = true; 979a0102bdaSJeff Layton pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); 980e5975c7cSAlex Elder BUG_ON(!pages); 98188486957SAlex Elder } 9825b64640cSYan, Zheng 9835b64640cSYan, Zheng len = 0; 9845b64640cSYan, Zheng } else if (page->index != 98509cbfeafSKirill A. Shutemov (offset + len) >> PAGE_SHIFT) { 986a0102bdaSJeff Layton if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS : 9875b64640cSYan, Zheng CEPH_OSD_MAX_OPS)) { 9885b64640cSYan, Zheng redirty_page_for_writepage(wbc, page); 9895b64640cSYan, Zheng unlock_page(page); 9905b64640cSYan, Zheng break; 9915b64640cSYan, Zheng } 9925b64640cSYan, Zheng 9935b64640cSYan, Zheng num_ops++; 9945b64640cSYan, Zheng offset = (u64)page_offset(page); 9955b64640cSYan, Zheng len = 0; 9961d3576fdSSage Weil } 9971d3576fdSSage Weil 9981d3576fdSSage Weil /* note position of first page in pvec */ 9991d3576fdSSage Weil dout("%p will write page %p idx %lu\n", 10001d3576fdSSage Weil inode, page, page->index); 10012baba250SYehuda Sadeh 10025b64640cSYan, Zheng if (atomic_long_inc_return(&fsc->writeback_count) > 10035b64640cSYan, Zheng CONGESTION_ON_THRESH( 1004503d4fa6SNeilBrown fsc->mount_options->congestion_kb)) 1005503d4fa6SNeilBrown fsc->write_congested = true; 10060713e5f2SYan, Zheng 10070713e5f2SYan, Zheng pages[locked_pages++] = page; 10080713e5f2SYan, Zheng pvec.pages[i] = NULL; 10090713e5f2SYan, Zheng 10108ff2d290SJeff Layton len += thp_size(page); 10111d3576fdSSage Weil } 10121d3576fdSSage Weil 10131d3576fdSSage Weil /* did we get anything? */ 10141d3576fdSSage Weil if (!locked_pages) 10151d3576fdSSage Weil goto release_pvec_pages; 10161d3576fdSSage Weil if (i) { 10170713e5f2SYan, Zheng unsigned j, n = 0; 10180713e5f2SYan, Zheng /* shift unused page to beginning of pvec */ 10190713e5f2SYan, Zheng for (j = 0; j < pvec_pages; j++) { 10200713e5f2SYan, Zheng if (!pvec.pages[j]) 10210713e5f2SYan, Zheng continue; 10220713e5f2SYan, Zheng if (n < j) 10230713e5f2SYan, Zheng pvec.pages[n] = pvec.pages[j]; 10240713e5f2SYan, Zheng n++; 10250713e5f2SYan, Zheng } 10260713e5f2SYan, Zheng pvec.nr = n; 10271d3576fdSSage Weil 10281d3576fdSSage Weil if (pvec_pages && i == pvec_pages && 10291d3576fdSSage Weil locked_pages < max_pages) { 10301d3576fdSSage Weil dout("reached end pvec, trying for more\n"); 10310713e5f2SYan, Zheng pagevec_release(&pvec); 10321d3576fdSSage Weil goto get_more_pages; 10331d3576fdSSage Weil } 10341d3576fdSSage Weil } 10351d3576fdSSage Weil 10365b64640cSYan, Zheng new_request: 1037e5975c7cSAlex Elder offset = page_offset(pages[0]); 10385b64640cSYan, Zheng len = wsize; 10395b64640cSYan, Zheng 10405b64640cSYan, Zheng req = ceph_osdc_new_request(&fsc->client->osdc, 10415b64640cSYan, Zheng &ci->i_layout, vino, 10425b64640cSYan, Zheng offset, &len, 0, num_ops, 10431f934b00SYan, Zheng CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 10441f934b00SYan, Zheng snapc, ceph_wbc.truncate_seq, 10451f934b00SYan, Zheng ceph_wbc.truncate_size, false); 10465b64640cSYan, Zheng if (IS_ERR(req)) { 10475b64640cSYan, Zheng req = ceph_osdc_new_request(&fsc->client->osdc, 10485b64640cSYan, Zheng &ci->i_layout, vino, 10495b64640cSYan, Zheng offset, &len, 0, 10505b64640cSYan, Zheng min(num_ops, 10515b64640cSYan, Zheng CEPH_OSD_SLAB_OPS), 10525b64640cSYan, Zheng CEPH_OSD_OP_WRITE, 105354ea0046SIlya Dryomov CEPH_OSD_FLAG_WRITE, 10541f934b00SYan, Zheng snapc, ceph_wbc.truncate_seq, 10551f934b00SYan, Zheng ceph_wbc.truncate_size, true); 10565b64640cSYan, Zheng BUG_ON(IS_ERR(req)); 10575b64640cSYan, Zheng } 10585b64640cSYan, Zheng BUG_ON(len < page_offset(pages[locked_pages - 1]) + 10598ff2d290SJeff Layton thp_size(page) - offset); 10605b64640cSYan, Zheng 10615b64640cSYan, Zheng req->r_callback = writepages_finish; 10625b64640cSYan, Zheng req->r_inode = inode; 10635b64640cSYan, Zheng 10645b64640cSYan, Zheng /* Format the osd request message and submit the write */ 10655b64640cSYan, Zheng len = 0; 10665b64640cSYan, Zheng data_pages = pages; 10675b64640cSYan, Zheng op_idx = 0; 10685b64640cSYan, Zheng for (i = 0; i < locked_pages; i++) { 10695b64640cSYan, Zheng u64 cur_offset = page_offset(pages[i]); 10701702e797SJeff Layton /* 10711702e797SJeff Layton * Discontinuity in page range? Ceph can handle that by just passing 10721702e797SJeff Layton * multiple extents in the write op. 10731702e797SJeff Layton */ 10745b64640cSYan, Zheng if (offset + len != cur_offset) { 10751702e797SJeff Layton /* If it's full, stop here */ 10763fb99d48SYanhu Cao if (op_idx + 1 == req->r_num_ops) 10775b64640cSYan, Zheng break; 10781702e797SJeff Layton 10791702e797SJeff Layton /* Kick off an fscache write with what we have so far. */ 10801702e797SJeff Layton ceph_fscache_write_to_cache(inode, offset, len, caching); 10811702e797SJeff Layton 10821702e797SJeff Layton /* Start a new extent */ 10835b64640cSYan, Zheng osd_req_op_extent_dup_last(req, op_idx, 10845b64640cSYan, Zheng cur_offset - offset); 10855b64640cSYan, Zheng dout("writepages got pages at %llu~%llu\n", 10865b64640cSYan, Zheng offset, len); 10875b64640cSYan, Zheng osd_req_op_extent_osd_data_pages(req, op_idx, 10885b64640cSYan, Zheng data_pages, len, 0, 1089a0102bdaSJeff Layton from_pool, false); 10905b64640cSYan, Zheng osd_req_op_extent_update(req, op_idx, len); 10915b64640cSYan, Zheng 10925b64640cSYan, Zheng len = 0; 10935b64640cSYan, Zheng offset = cur_offset; 10945b64640cSYan, Zheng data_pages = pages + i; 10955b64640cSYan, Zheng op_idx++; 10965b64640cSYan, Zheng } 10975b64640cSYan, Zheng 10985b64640cSYan, Zheng set_page_writeback(pages[i]); 10991702e797SJeff Layton if (caching) 11001702e797SJeff Layton ceph_set_page_fscache(pages[i]); 11018ff2d290SJeff Layton len += thp_size(page); 11025b64640cSYan, Zheng } 11031702e797SJeff Layton ceph_fscache_write_to_cache(inode, offset, len, caching); 11045b64640cSYan, Zheng 11051f934b00SYan, Zheng if (ceph_wbc.size_stable) { 11061f934b00SYan, Zheng len = min(len, ceph_wbc.i_size - offset); 11075b64640cSYan, Zheng } else if (i == locked_pages) { 1108e1966b49SYan, Zheng /* writepages_finish() clears writeback pages 1109e1966b49SYan, Zheng * according to the data length, so make sure 1110e1966b49SYan, Zheng * data length covers all locked pages */ 11118ff2d290SJeff Layton u64 min_len = len + 1 - thp_size(page); 11121f934b00SYan, Zheng len = get_writepages_data_length(inode, pages[i - 1], 11131f934b00SYan, Zheng offset); 11145b64640cSYan, Zheng len = max(len, min_len); 1115e1966b49SYan, Zheng } 11165b64640cSYan, Zheng dout("writepages got pages at %llu~%llu\n", offset, len); 11171d3576fdSSage Weil 11185b64640cSYan, Zheng osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, 1119a0102bdaSJeff Layton 0, from_pool, false); 11205b64640cSYan, Zheng osd_req_op_extent_update(req, op_idx, len); 1121e5975c7cSAlex Elder 11225b64640cSYan, Zheng BUG_ON(op_idx + 1 != req->r_num_ops); 11235b64640cSYan, Zheng 1124a0102bdaSJeff Layton from_pool = false; 11255b64640cSYan, Zheng if (i < locked_pages) { 11265b64640cSYan, Zheng BUG_ON(num_ops <= req->r_num_ops); 11275b64640cSYan, Zheng num_ops -= req->r_num_ops; 11285b64640cSYan, Zheng locked_pages -= i; 1129e5975c7cSAlex Elder 11305b64640cSYan, Zheng /* allocate new pages array for next request */ 11315b64640cSYan, Zheng data_pages = pages; 11326da2ec56SKees Cook pages = kmalloc_array(locked_pages, sizeof(*pages), 11335b64640cSYan, Zheng GFP_NOFS); 11345b64640cSYan, Zheng if (!pages) { 1135a0102bdaSJeff Layton from_pool = true; 1136a0102bdaSJeff Layton pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); 11375b64640cSYan, Zheng BUG_ON(!pages); 11385b64640cSYan, Zheng } 11395b64640cSYan, Zheng memcpy(pages, data_pages + i, 11405b64640cSYan, Zheng locked_pages * sizeof(*pages)); 11415b64640cSYan, Zheng memset(data_pages + i, 0, 11425b64640cSYan, Zheng locked_pages * sizeof(*pages)); 11435b64640cSYan, Zheng } else { 11445b64640cSYan, Zheng BUG_ON(num_ops != req->r_num_ops); 11455b64640cSYan, Zheng index = pages[i - 1]->index + 1; 11465b64640cSYan, Zheng /* request message now owns the pages array */ 11475b64640cSYan, Zheng pages = NULL; 11485b64640cSYan, Zheng } 1149e5975c7cSAlex Elder 1150fac02ddfSArnd Bergmann req->r_mtime = inode->i_mtime; 1151a8af0d68SJeff Layton ceph_osdc_start_request(&fsc->client->osdc, req); 11521d3576fdSSage Weil req = NULL; 11531d3576fdSSage Weil 11545b64640cSYan, Zheng wbc->nr_to_write -= i; 11555b64640cSYan, Zheng if (pages) 11565b64640cSYan, Zheng goto new_request; 11575b64640cSYan, Zheng 11582a2d927eSYan, Zheng /* 11592a2d927eSYan, Zheng * We stop writing back only if we are not doing 11602a2d927eSYan, Zheng * integrity sync. In case of integrity sync we have to 11612a2d927eSYan, Zheng * keep going until we have written all the pages 11622a2d927eSYan, Zheng * we tagged for writeback prior to entering this loop. 11632a2d927eSYan, Zheng */ 11642a2d927eSYan, Zheng if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) 1165af9cc401SYan, Zheng done = true; 11661d3576fdSSage Weil 11671d3576fdSSage Weil release_pvec_pages: 11681d3576fdSSage Weil dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 11691d3576fdSSage Weil pvec.nr ? pvec.pages[0] : NULL); 11701d3576fdSSage Weil pagevec_release(&pvec); 11711d3576fdSSage Weil } 11721d3576fdSSage Weil 11731d3576fdSSage Weil if (should_loop && !done) { 11741d3576fdSSage Weil /* more to do; loop back to beginning of file */ 11751d3576fdSSage Weil dout("writepages looping back to beginning of file\n"); 11762a2d927eSYan, Zheng end = start_index - 1; /* OK even when start_index == 0 */ 1177f275635eSYan, Zheng 1178f275635eSYan, Zheng /* to write dirty pages associated with next snapc, 1179f275635eSYan, Zheng * we need to wait until current writes complete */ 1180f275635eSYan, Zheng if (wbc->sync_mode != WB_SYNC_NONE && 1181f275635eSYan, Zheng start_index == 0 && /* all dirty pages were checked */ 1182f275635eSYan, Zheng !ceph_wbc.head_snapc) { 1183f275635eSYan, Zheng struct page *page; 1184f275635eSYan, Zheng unsigned i, nr; 1185f275635eSYan, Zheng index = 0; 1186f275635eSYan, Zheng while ((index <= end) && 1187f275635eSYan, Zheng (nr = pagevec_lookup_tag(&pvec, mapping, &index, 118867fd707fSJan Kara PAGECACHE_TAG_WRITEBACK))) { 1189f275635eSYan, Zheng for (i = 0; i < nr; i++) { 1190f275635eSYan, Zheng page = pvec.pages[i]; 1191f275635eSYan, Zheng if (page_snap_context(page) != snapc) 1192f275635eSYan, Zheng continue; 1193f275635eSYan, Zheng wait_on_page_writeback(page); 1194f275635eSYan, Zheng } 1195f275635eSYan, Zheng pagevec_release(&pvec); 1196f275635eSYan, Zheng cond_resched(); 1197f275635eSYan, Zheng } 1198f275635eSYan, Zheng } 1199f275635eSYan, Zheng 12002a2d927eSYan, Zheng start_index = 0; 12011d3576fdSSage Weil index = 0; 12021d3576fdSSage Weil goto retry; 12031d3576fdSSage Weil } 12041d3576fdSSage Weil 12051d3576fdSSage Weil if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 12061d3576fdSSage Weil mapping->writeback_index = index; 12071d3576fdSSage Weil 12081d3576fdSSage Weil out: 12091d3576fdSSage Weil ceph_osdc_put_request(req); 12102a2d927eSYan, Zheng ceph_put_snap_context(last_snapc); 12112a2d927eSYan, Zheng dout("writepages dend - startone, rc = %d\n", rc); 12121d3576fdSSage Weil return rc; 12131d3576fdSSage Weil } 12141d3576fdSSage Weil 12151d3576fdSSage Weil 12161d3576fdSSage Weil 12171d3576fdSSage Weil /* 12181d3576fdSSage Weil * See if a given @snapc is either writeable, or already written. 12191d3576fdSSage Weil */ 12201d3576fdSSage Weil static int context_is_writeable_or_written(struct inode *inode, 12211d3576fdSSage Weil struct ceph_snap_context *snapc) 12221d3576fdSSage Weil { 122305455e11SYan, Zheng struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); 12246298a337SSage Weil int ret = !oldest || snapc->seq <= oldest->seq; 12256298a337SSage Weil 12266298a337SSage Weil ceph_put_snap_context(oldest); 12276298a337SSage Weil return ret; 12281d3576fdSSage Weil } 12291d3576fdSSage Weil 123018d620f0SJeff Layton /** 123118d620f0SJeff Layton * ceph_find_incompatible - find an incompatible context and return it 123218d620f0SJeff Layton * @page: page being dirtied 123318d620f0SJeff Layton * 123418d620f0SJeff Layton * We are only allowed to write into/dirty a page if the page is 123518d620f0SJeff Layton * clean, or already dirty within the same snap context. Returns a 123618d620f0SJeff Layton * conflicting context if there is one, NULL if there isn't, or a 123718d620f0SJeff Layton * negative error code on other errors. 123818d620f0SJeff Layton * 123918d620f0SJeff Layton * Must be called with page lock held. 124018d620f0SJeff Layton */ 124118d620f0SJeff Layton static struct ceph_snap_context * 1242d45156bfSJeff Layton ceph_find_incompatible(struct page *page) 124318d620f0SJeff Layton { 1244d45156bfSJeff Layton struct inode *inode = page->mapping->host; 124518d620f0SJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 124618d620f0SJeff Layton 12475d6451b1SJeff Layton if (ceph_inode_is_shutdown(inode)) { 12485d6451b1SJeff Layton dout(" page %p %llx:%llx is shutdown\n", page, 12495d6451b1SJeff Layton ceph_vinop(inode)); 12505d6451b1SJeff Layton return ERR_PTR(-ESTALE); 125118d620f0SJeff Layton } 125218d620f0SJeff Layton 125318d620f0SJeff Layton for (;;) { 125418d620f0SJeff Layton struct ceph_snap_context *snapc, *oldest; 125518d620f0SJeff Layton 125618d620f0SJeff Layton wait_on_page_writeback(page); 125718d620f0SJeff Layton 125818d620f0SJeff Layton snapc = page_snap_context(page); 125918d620f0SJeff Layton if (!snapc || snapc == ci->i_head_snapc) 126018d620f0SJeff Layton break; 126118d620f0SJeff Layton 126218d620f0SJeff Layton /* 126318d620f0SJeff Layton * this page is already dirty in another (older) snap 126418d620f0SJeff Layton * context! is it writeable now? 126518d620f0SJeff Layton */ 126618d620f0SJeff Layton oldest = get_oldest_context(inode, NULL, NULL); 126718d620f0SJeff Layton if (snapc->seq > oldest->seq) { 126818d620f0SJeff Layton /* not writeable -- return it for the caller to deal with */ 126918d620f0SJeff Layton ceph_put_snap_context(oldest); 127018d620f0SJeff Layton dout(" page %p snapc %p not current or oldest\n", page, snapc); 127118d620f0SJeff Layton return ceph_get_snap_context(snapc); 127218d620f0SJeff Layton } 127318d620f0SJeff Layton ceph_put_snap_context(oldest); 127418d620f0SJeff Layton 127518d620f0SJeff Layton /* yay, writeable, do it now (without dropping page lock) */ 127618d620f0SJeff Layton dout(" page %p snapc %p not current, but oldest\n", page, snapc); 127718d620f0SJeff Layton if (clear_page_dirty_for_io(page)) { 127818d620f0SJeff Layton int r = writepage_nounlock(page, NULL); 127918d620f0SJeff Layton if (r < 0) 128018d620f0SJeff Layton return ERR_PTR(r); 128118d620f0SJeff Layton } 128218d620f0SJeff Layton } 128318d620f0SJeff Layton return NULL; 128418d620f0SJeff Layton } 128518d620f0SJeff Layton 1286d801327dSJeff Layton static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, 1287fac47b43SXiubo Li struct folio **foliop, void **_fsdata) 1288d801327dSJeff Layton { 1289d801327dSJeff Layton struct inode *inode = file_inode(file); 1290d801327dSJeff Layton struct ceph_inode_info *ci = ceph_inode(inode); 1291d801327dSJeff Layton struct ceph_snap_context *snapc; 1292d801327dSJeff Layton 1293fac47b43SXiubo Li snapc = ceph_find_incompatible(folio_page(*foliop, 0)); 1294d801327dSJeff Layton if (snapc) { 1295d801327dSJeff Layton int r; 1296d801327dSJeff Layton 1297fac47b43SXiubo Li folio_unlock(*foliop); 1298fac47b43SXiubo Li folio_put(*foliop); 1299fac47b43SXiubo Li *foliop = NULL; 1300d801327dSJeff Layton if (IS_ERR(snapc)) 1301d801327dSJeff Layton return PTR_ERR(snapc); 1302d801327dSJeff Layton 1303d801327dSJeff Layton ceph_queue_writeback(inode); 1304d801327dSJeff Layton r = wait_event_killable(ci->i_cap_wq, 1305d801327dSJeff Layton context_is_writeable_or_written(inode, snapc)); 1306d801327dSJeff Layton ceph_put_snap_context(snapc); 1307d801327dSJeff Layton return r == 0 ? -EAGAIN : r; 1308d801327dSJeff Layton } 1309d801327dSJeff Layton return 0; 1310d801327dSJeff Layton } 1311d801327dSJeff Layton 13121d3576fdSSage Weil /* 13131d3576fdSSage Weil * We are only allowed to write into/dirty the page if the page is 13141d3576fdSSage Weil * clean, or already dirty within the same snap context. 13154af6b225SYehuda Sadeh */ 13164af6b225SYehuda Sadeh static int ceph_write_begin(struct file *file, struct address_space *mapping, 13179d6b0cd7SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 13184af6b225SYehuda Sadeh struct page **pagep, void **fsdata) 13194af6b225SYehuda Sadeh { 1320496ad9aaSAl Viro struct inode *inode = file_inode(file); 1321e81fb419SLinus Torvalds struct ceph_inode_info *ci = ceph_inode(inode); 132278525c74SDavid Howells struct folio *folio = NULL; 1323d801327dSJeff Layton int r; 13244af6b225SYehuda Sadeh 1325e81fb419SLinus Torvalds r = netfs_write_begin(&ci->netfs, file, inode->i_mapping, pos, len, &folio, NULL); 1326c460f4e4SXiubo Li if (r < 0) 1327c460f4e4SXiubo Li return r; 1328c460f4e4SXiubo Li 132978525c74SDavid Howells folio_wait_fscache(folio); 133078525c74SDavid Howells WARN_ON_ONCE(!folio_test_locked(folio)); 133178525c74SDavid Howells *pagep = &folio->page; 1332c460f4e4SXiubo Li return 0; 13334af6b225SYehuda Sadeh } 13344af6b225SYehuda Sadeh 13354af6b225SYehuda Sadeh /* 13361d3576fdSSage Weil * we don't do anything in here that simple_write_end doesn't do 13375dda377cSYan, Zheng * except adjust dirty page accounting 13381d3576fdSSage Weil */ 13391d3576fdSSage Weil static int ceph_write_end(struct file *file, struct address_space *mapping, 13401d3576fdSSage Weil loff_t pos, unsigned len, unsigned copied, 134178525c74SDavid Howells struct page *subpage, void *fsdata) 13421d3576fdSSage Weil { 134378525c74SDavid Howells struct folio *folio = page_folio(subpage); 1344496ad9aaSAl Viro struct inode *inode = file_inode(file); 1345efb0ca76SYan, Zheng bool check_cap = false; 13461d3576fdSSage Weil 134778525c74SDavid Howells dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file, 134878525c74SDavid Howells inode, folio, (int)pos, (int)copied, (int)len); 13491d3576fdSSage Weil 135078525c74SDavid Howells if (!folio_test_uptodate(folio)) { 1351ce3a8732SJeff Layton /* just return that nothing was copied on a short copy */ 1352b9de313cSAl Viro if (copied < len) { 1353b9de313cSAl Viro copied = 0; 1354b9de313cSAl Viro goto out; 1355b9de313cSAl Viro } 135678525c74SDavid Howells folio_mark_uptodate(folio); 1357b9de313cSAl Viro } 13581d3576fdSSage Weil 13591d3576fdSSage Weil /* did file size increase? */ 136099c88e69SYan, Zheng if (pos+copied > i_size_read(inode)) 13611d3576fdSSage Weil check_cap = ceph_inode_set_size(inode, pos+copied); 13621d3576fdSSage Weil 136378525c74SDavid Howells folio_mark_dirty(folio); 13641d3576fdSSage Weil 1365b9de313cSAl Viro out: 136678525c74SDavid Howells folio_unlock(folio); 136778525c74SDavid Howells folio_put(folio); 13681d3576fdSSage Weil 13691d3576fdSSage Weil if (check_cap) 1370*e4b731ccSXiubo Li ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY); 13711d3576fdSSage Weil 13721d3576fdSSage Weil return copied; 13731d3576fdSSage Weil } 13741d3576fdSSage Weil 13751d3576fdSSage Weil const struct address_space_operations ceph_aops = { 13766c62371bSMatthew Wilcox (Oracle) .read_folio = netfs_read_folio, 1377bc899ee1SDavid Howells .readahead = netfs_readahead, 13781d3576fdSSage Weil .writepage = ceph_writepage, 13791d3576fdSSage Weil .writepages = ceph_writepages_start, 13801d3576fdSSage Weil .write_begin = ceph_write_begin, 13811d3576fdSSage Weil .write_end = ceph_write_end, 13828fb72b4aSMatthew Wilcox (Oracle) .dirty_folio = ceph_dirty_folio, 13839872f4deSMatthew Wilcox (Oracle) .invalidate_folio = ceph_invalidate_folio, 13845e414655SMatthew Wilcox (Oracle) .release_folio = ceph_release_folio, 13859c43ff44SJeff Layton .direct_IO = noop_direct_IO, 13861d3576fdSSage Weil }; 13871d3576fdSSage Weil 13884f7e89f6SYan, Zheng static void ceph_block_sigs(sigset_t *oldset) 13894f7e89f6SYan, Zheng { 13904f7e89f6SYan, Zheng sigset_t mask; 13914f7e89f6SYan, Zheng siginitsetinv(&mask, sigmask(SIGKILL)); 13924f7e89f6SYan, Zheng sigprocmask(SIG_BLOCK, &mask, oldset); 13934f7e89f6SYan, Zheng } 13944f7e89f6SYan, Zheng 13954f7e89f6SYan, Zheng static void ceph_restore_sigs(sigset_t *oldset) 13964f7e89f6SYan, Zheng { 13974f7e89f6SYan, Zheng sigprocmask(SIG_SETMASK, oldset, NULL); 13984f7e89f6SYan, Zheng } 13991d3576fdSSage Weil 14001d3576fdSSage Weil /* 14011d3576fdSSage Weil * vm ops 14021d3576fdSSage Weil */ 140324499847SSouptick Joarder static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf) 140461f68816SYan, Zheng { 140511bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 140661f68816SYan, Zheng struct inode *inode = file_inode(vma->vm_file); 140761f68816SYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 140861f68816SYan, Zheng struct ceph_file_info *fi = vma->vm_file->private_data; 1409c403c3a2SMatthew Wilcox (Oracle) loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT; 141024499847SSouptick Joarder int want, got, err; 14114f7e89f6SYan, Zheng sigset_t oldset; 141224499847SSouptick Joarder vm_fault_t ret = VM_FAULT_SIGBUS; 14134f7e89f6SYan, Zheng 14145d6451b1SJeff Layton if (ceph_inode_is_shutdown(inode)) 14155d6451b1SJeff Layton return ret; 14165d6451b1SJeff Layton 14174f7e89f6SYan, Zheng ceph_block_sigs(&oldset); 141861f68816SYan, Zheng 14198ff2d290SJeff Layton dout("filemap_fault %p %llx.%llx %llu trying to get caps\n", 14208ff2d290SJeff Layton inode, ceph_vinop(inode), off); 142161f68816SYan, Zheng if (fi->fmode & CEPH_FILE_MODE_LAZY) 142261f68816SYan, Zheng want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 142361f68816SYan, Zheng else 142461f68816SYan, Zheng want = CEPH_CAP_FILE_CACHE; 14254f7e89f6SYan, Zheng 142661f68816SYan, Zheng got = 0; 1427e72968e1SJeff Layton err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got); 142824499847SSouptick Joarder if (err < 0) 14294f7e89f6SYan, Zheng goto out_restore; 14306ce026e4SYan, Zheng 14318ff2d290SJeff Layton dout("filemap_fault %p %llu got cap refs on %s\n", 14328ff2d290SJeff Layton inode, off, ceph_cap_string(got)); 143361f68816SYan, Zheng 143483701246SYan, Zheng if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || 143548490776SXiubo Li !ceph_has_inline_data(ci)) { 14365d988308SYan, Zheng CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); 14375d988308SYan, Zheng ceph_add_rw_context(fi, &rw_ctx); 143811bac800SDave Jiang ret = filemap_fault(vmf); 14395d988308SYan, Zheng ceph_del_rw_context(fi, &rw_ctx); 14408ff2d290SJeff Layton dout("filemap_fault %p %llu drop cap refs %s ret %x\n", 14418ff2d290SJeff Layton inode, off, ceph_cap_string(got), ret); 14422b1ac852SYan, Zheng } else 144324499847SSouptick Joarder err = -EAGAIN; 144461f68816SYan, Zheng 144561f68816SYan, Zheng ceph_put_cap_refs(ci, got); 144661f68816SYan, Zheng 144724499847SSouptick Joarder if (err != -EAGAIN) 14484f7e89f6SYan, Zheng goto out_restore; 144983701246SYan, Zheng 145083701246SYan, Zheng /* read inline data */ 145109cbfeafSKirill A. Shutemov if (off >= PAGE_SIZE) { 145283701246SYan, Zheng /* does not support inline data > PAGE_SIZE */ 145383701246SYan, Zheng ret = VM_FAULT_SIGBUS; 145483701246SYan, Zheng } else { 145583701246SYan, Zheng struct address_space *mapping = inode->i_mapping; 1456057ba5b2SJan Kara struct page *page; 1457057ba5b2SJan Kara 1458057ba5b2SJan Kara filemap_invalidate_lock_shared(mapping); 1459057ba5b2SJan Kara page = find_or_create_page(mapping, 0, 1460057ba5b2SJan Kara mapping_gfp_constraint(mapping, ~__GFP_FS)); 146183701246SYan, Zheng if (!page) { 146283701246SYan, Zheng ret = VM_FAULT_OOM; 14634f7e89f6SYan, Zheng goto out_inline; 146483701246SYan, Zheng } 146524499847SSouptick Joarder err = __ceph_do_getattr(inode, page, 146683701246SYan, Zheng CEPH_STAT_CAP_INLINE_DATA, true); 146724499847SSouptick Joarder if (err < 0 || off >= i_size_read(inode)) { 146883701246SYan, Zheng unlock_page(page); 146909cbfeafSKirill A. Shutemov put_page(page); 1470c64a2b05SSouptick Joarder ret = vmf_error(err); 14714f7e89f6SYan, Zheng goto out_inline; 147283701246SYan, Zheng } 147324499847SSouptick Joarder if (err < PAGE_SIZE) 147424499847SSouptick Joarder zero_user_segment(page, err, PAGE_SIZE); 147583701246SYan, Zheng else 147683701246SYan, Zheng flush_dcache_page(page); 147783701246SYan, Zheng SetPageUptodate(page); 147883701246SYan, Zheng vmf->page = page; 147983701246SYan, Zheng ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; 14804f7e89f6SYan, Zheng out_inline: 1481057ba5b2SJan Kara filemap_invalidate_unlock_shared(mapping); 14828ff2d290SJeff Layton dout("filemap_fault %p %llu read inline data ret %x\n", 14838ff2d290SJeff Layton inode, off, ret); 14844f7e89f6SYan, Zheng } 14854f7e89f6SYan, Zheng out_restore: 14864f7e89f6SYan, Zheng ceph_restore_sigs(&oldset); 148724499847SSouptick Joarder if (err < 0) 148824499847SSouptick Joarder ret = vmf_error(err); 14896ce026e4SYan, Zheng 149061f68816SYan, Zheng return ret; 149161f68816SYan, Zheng } 14921d3576fdSSage Weil 149324499847SSouptick Joarder static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf) 14941d3576fdSSage Weil { 149511bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 1496496ad9aaSAl Viro struct inode *inode = file_inode(vma->vm_file); 149761f68816SYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 149861f68816SYan, Zheng struct ceph_file_info *fi = vma->vm_file->private_data; 1499f66fd9f0SYan, Zheng struct ceph_cap_flush *prealloc_cf; 150061f68816SYan, Zheng struct page *page = vmf->page; 15016285bc23SAlex Elder loff_t off = page_offset(page); 150261f68816SYan, Zheng loff_t size = i_size_read(inode); 150361f68816SYan, Zheng size_t len; 150424499847SSouptick Joarder int want, got, err; 15054f7e89f6SYan, Zheng sigset_t oldset; 150624499847SSouptick Joarder vm_fault_t ret = VM_FAULT_SIGBUS; 15071d3576fdSSage Weil 15085d6451b1SJeff Layton if (ceph_inode_is_shutdown(inode)) 15095d6451b1SJeff Layton return ret; 15105d6451b1SJeff Layton 1511f66fd9f0SYan, Zheng prealloc_cf = ceph_alloc_cap_flush(); 1512f66fd9f0SYan, Zheng if (!prealloc_cf) 15136ce026e4SYan, Zheng return VM_FAULT_OOM; 1514f66fd9f0SYan, Zheng 1515249c1df5SJeff Layton sb_start_pagefault(inode->i_sb); 15164f7e89f6SYan, Zheng ceph_block_sigs(&oldset); 15171d3576fdSSage Weil 15188ff2d290SJeff Layton if (off + thp_size(page) <= size) 15198ff2d290SJeff Layton len = thp_size(page); 15201d3576fdSSage Weil else 15218ff2d290SJeff Layton len = offset_in_thp(page, size); 15221d3576fdSSage Weil 152361f68816SYan, Zheng dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", 152461f68816SYan, Zheng inode, ceph_vinop(inode), off, len, size); 152561f68816SYan, Zheng if (fi->fmode & CEPH_FILE_MODE_LAZY) 152661f68816SYan, Zheng want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 152761f68816SYan, Zheng else 152861f68816SYan, Zheng want = CEPH_CAP_FILE_BUFFER; 15294f7e89f6SYan, Zheng 153061f68816SYan, Zheng got = 0; 1531e72968e1SJeff Layton err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got); 153224499847SSouptick Joarder if (err < 0) 1533f66fd9f0SYan, Zheng goto out_free; 15346ce026e4SYan, Zheng 153561f68816SYan, Zheng dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", 153661f68816SYan, Zheng inode, off, len, ceph_cap_string(got)); 153761f68816SYan, Zheng 153861f68816SYan, Zheng /* Update time before taking page lock */ 153961f68816SYan, Zheng file_update_time(vma->vm_file); 15405c308356SJeff Layton inode_inc_iversion_raw(inode); 15414af6b225SYehuda Sadeh 1542f0b33df5SYan, Zheng do { 1543d45156bfSJeff Layton struct ceph_snap_context *snapc; 1544d45156bfSJeff Layton 15454af6b225SYehuda Sadeh lock_page(page); 15464af6b225SYehuda Sadeh 1547cb03c143SAndreas Gruenbacher if (page_mkwrite_check_truncate(page, inode) < 0) { 1548f9cac5acSYan, Zheng unlock_page(page); 15496ce026e4SYan, Zheng ret = VM_FAULT_NOPAGE; 1550f0b33df5SYan, Zheng break; 1551f9cac5acSYan, Zheng } 15524af6b225SYehuda Sadeh 1553d45156bfSJeff Layton snapc = ceph_find_incompatible(page); 1554d45156bfSJeff Layton if (!snapc) { 15554af6b225SYehuda Sadeh /* success. we'll keep the page locked. */ 15561d3576fdSSage Weil set_page_dirty(page); 15571d3576fdSSage Weil ret = VM_FAULT_LOCKED; 1558d45156bfSJeff Layton break; 15591d3576fdSSage Weil } 1560d45156bfSJeff Layton 1561d45156bfSJeff Layton unlock_page(page); 1562d45156bfSJeff Layton 1563d45156bfSJeff Layton if (IS_ERR(snapc)) { 1564d45156bfSJeff Layton ret = VM_FAULT_SIGBUS; 1565d45156bfSJeff Layton break; 1566d45156bfSJeff Layton } 1567d45156bfSJeff Layton 1568d45156bfSJeff Layton ceph_queue_writeback(inode); 1569d45156bfSJeff Layton err = wait_event_killable(ci->i_cap_wq, 1570d45156bfSJeff Layton context_is_writeable_or_written(inode, snapc)); 1571d45156bfSJeff Layton ceph_put_snap_context(snapc); 1572d45156bfSJeff Layton } while (err == 0); 1573f0b33df5SYan, Zheng 1574083db6fdSDavid Howells if (ret == VM_FAULT_LOCKED) { 157561f68816SYan, Zheng int dirty; 157661f68816SYan, Zheng spin_lock(&ci->i_ceph_lock); 1577f66fd9f0SYan, Zheng dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1578f66fd9f0SYan, Zheng &prealloc_cf); 157961f68816SYan, Zheng spin_unlock(&ci->i_ceph_lock); 158061f68816SYan, Zheng if (dirty) 158161f68816SYan, Zheng __mark_inode_dirty(inode, dirty); 158261f68816SYan, Zheng } 158361f68816SYan, Zheng 158424499847SSouptick Joarder dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n", 158561f68816SYan, Zheng inode, off, len, ceph_cap_string(got), ret); 1586a8810cdcSJeff Layton ceph_put_cap_refs_async(ci, got); 1587f66fd9f0SYan, Zheng out_free: 15884f7e89f6SYan, Zheng ceph_restore_sigs(&oldset); 1589249c1df5SJeff Layton sb_end_pagefault(inode->i_sb); 1590f66fd9f0SYan, Zheng ceph_free_cap_flush(prealloc_cf); 159124499847SSouptick Joarder if (err < 0) 159224499847SSouptick Joarder ret = vmf_error(err); 15931d3576fdSSage Weil return ret; 15941d3576fdSSage Weil } 15951d3576fdSSage Weil 159631c542a1SYan, Zheng void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, 159731c542a1SYan, Zheng char *data, size_t len) 159831c542a1SYan, Zheng { 159931c542a1SYan, Zheng struct address_space *mapping = inode->i_mapping; 160031c542a1SYan, Zheng struct page *page; 160131c542a1SYan, Zheng 160231c542a1SYan, Zheng if (locked_page) { 160331c542a1SYan, Zheng page = locked_page; 160431c542a1SYan, Zheng } else { 160531c542a1SYan, Zheng if (i_size_read(inode) == 0) 160631c542a1SYan, Zheng return; 160731c542a1SYan, Zheng page = find_or_create_page(mapping, 0, 1608c62d2555SMichal Hocko mapping_gfp_constraint(mapping, 1609c62d2555SMichal Hocko ~__GFP_FS)); 161031c542a1SYan, Zheng if (!page) 161131c542a1SYan, Zheng return; 161231c542a1SYan, Zheng if (PageUptodate(page)) { 161331c542a1SYan, Zheng unlock_page(page); 161409cbfeafSKirill A. Shutemov put_page(page); 161531c542a1SYan, Zheng return; 161631c542a1SYan, Zheng } 161731c542a1SYan, Zheng } 161831c542a1SYan, Zheng 16190668ff52SIlya Dryomov dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", 162031c542a1SYan, Zheng inode, ceph_vinop(inode), len, locked_page); 162131c542a1SYan, Zheng 162231c542a1SYan, Zheng if (len > 0) { 162331c542a1SYan, Zheng void *kaddr = kmap_atomic(page); 162431c542a1SYan, Zheng memcpy(kaddr, data, len); 162531c542a1SYan, Zheng kunmap_atomic(kaddr); 162631c542a1SYan, Zheng } 162731c542a1SYan, Zheng 162831c542a1SYan, Zheng if (page != locked_page) { 162909cbfeafSKirill A. Shutemov if (len < PAGE_SIZE) 163009cbfeafSKirill A. Shutemov zero_user_segment(page, len, PAGE_SIZE); 163131c542a1SYan, Zheng else 163231c542a1SYan, Zheng flush_dcache_page(page); 163331c542a1SYan, Zheng 163431c542a1SYan, Zheng SetPageUptodate(page); 163531c542a1SYan, Zheng unlock_page(page); 163609cbfeafSKirill A. Shutemov put_page(page); 163731c542a1SYan, Zheng } 163831c542a1SYan, Zheng } 163931c542a1SYan, Zheng 1640083db6fdSDavid Howells int ceph_uninline_data(struct file *file) 164128127bddSYan, Zheng { 1642083db6fdSDavid Howells struct inode *inode = file_inode(file); 164328127bddSYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 164428127bddSYan, Zheng struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1645825978fdSXiubo Li struct ceph_osd_request *req = NULL; 1646083db6fdSDavid Howells struct ceph_cap_flush *prealloc_cf; 1647083db6fdSDavid Howells struct folio *folio = NULL; 1648c38af982SDan Carpenter u64 inline_version = CEPH_INLINE_NONE; 1649083db6fdSDavid Howells struct page *pages[1]; 165028127bddSYan, Zheng int err = 0; 1651c38af982SDan Carpenter u64 len; 1652083db6fdSDavid Howells 165328127bddSYan, Zheng spin_lock(&ci->i_ceph_lock); 165428127bddSYan, Zheng inline_version = ci->i_inline_version; 165528127bddSYan, Zheng spin_unlock(&ci->i_ceph_lock); 165628127bddSYan, Zheng 165728127bddSYan, Zheng dout("uninline_data %p %llx.%llx inline_version %llu\n", 165828127bddSYan, Zheng inode, ceph_vinop(inode), inline_version); 165928127bddSYan, Zheng 1660825978fdSXiubo Li if (inline_version == CEPH_INLINE_NONE) 1661825978fdSXiubo Li return 0; 1662825978fdSXiubo Li 1663825978fdSXiubo Li prealloc_cf = ceph_alloc_cap_flush(); 1664825978fdSXiubo Li if (!prealloc_cf) 1665825978fdSXiubo Li return -ENOMEM; 1666825978fdSXiubo Li 1667825978fdSXiubo Li if (inline_version == 1) /* initial version, no data */ 1668825978fdSXiubo Li goto out_uninline; 1669825978fdSXiubo Li 1670825978fdSXiubo Li folio = read_mapping_folio(inode->i_mapping, 0, file); 1671825978fdSXiubo Li if (IS_ERR(folio)) { 1672825978fdSXiubo Li err = PTR_ERR(folio); 1673825978fdSXiubo Li goto out; 1674825978fdSXiubo Li } 1675825978fdSXiubo Li 1676825978fdSXiubo Li folio_lock(folio); 167728127bddSYan, Zheng 167828127bddSYan, Zheng len = i_size_read(inode); 1679083db6fdSDavid Howells if (len > folio_size(folio)) 1680083db6fdSDavid Howells len = folio_size(folio); 168128127bddSYan, Zheng 168228127bddSYan, Zheng req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 168328127bddSYan, Zheng ceph_vino(inode), 0, &len, 0, 1, 168454ea0046SIlya Dryomov CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, 168534b759b4SIlya Dryomov NULL, 0, 0, false); 168628127bddSYan, Zheng if (IS_ERR(req)) { 168728127bddSYan, Zheng err = PTR_ERR(req); 1688083db6fdSDavid Howells goto out_unlock; 168928127bddSYan, Zheng } 169028127bddSYan, Zheng 1691fac02ddfSArnd Bergmann req->r_mtime = inode->i_mtime; 1692a8af0d68SJeff Layton ceph_osdc_start_request(&fsc->client->osdc, req); 169328127bddSYan, Zheng err = ceph_osdc_wait_request(&fsc->client->osdc, req); 169428127bddSYan, Zheng ceph_osdc_put_request(req); 169528127bddSYan, Zheng if (err < 0) 1696083db6fdSDavid Howells goto out_unlock; 169728127bddSYan, Zheng 169828127bddSYan, Zheng req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 169928127bddSYan, Zheng ceph_vino(inode), 0, &len, 1, 3, 170054ea0046SIlya Dryomov CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 170134b759b4SIlya Dryomov NULL, ci->i_truncate_seq, 170234b759b4SIlya Dryomov ci->i_truncate_size, false); 170328127bddSYan, Zheng if (IS_ERR(req)) { 170428127bddSYan, Zheng err = PTR_ERR(req); 1705083db6fdSDavid Howells goto out_unlock; 170628127bddSYan, Zheng } 170728127bddSYan, Zheng 1708083db6fdSDavid Howells pages[0] = folio_page(folio, 0); 1709083db6fdSDavid Howells osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false); 171028127bddSYan, Zheng 1711ec137c10SYan, Zheng { 1712ec137c10SYan, Zheng __le64 xattr_buf = cpu_to_le64(inline_version); 171328127bddSYan, Zheng err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, 1714ec137c10SYan, Zheng "inline_version", &xattr_buf, 1715ec137c10SYan, Zheng sizeof(xattr_buf), 171628127bddSYan, Zheng CEPH_OSD_CMPXATTR_OP_GT, 171728127bddSYan, Zheng CEPH_OSD_CMPXATTR_MODE_U64); 171828127bddSYan, Zheng if (err) 1719083db6fdSDavid Howells goto out_put_req; 1720ec137c10SYan, Zheng } 172128127bddSYan, Zheng 1722ec137c10SYan, Zheng { 1723ec137c10SYan, Zheng char xattr_buf[32]; 1724ec137c10SYan, Zheng int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), 1725ec137c10SYan, Zheng "%llu", inline_version); 172628127bddSYan, Zheng err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, 1727ec137c10SYan, Zheng "inline_version", 1728ec137c10SYan, Zheng xattr_buf, xattr_len, 0, 0); 172928127bddSYan, Zheng if (err) 1730083db6fdSDavid Howells goto out_put_req; 1731ec137c10SYan, Zheng } 173228127bddSYan, Zheng 1733fac02ddfSArnd Bergmann req->r_mtime = inode->i_mtime; 1734a8af0d68SJeff Layton ceph_osdc_start_request(&fsc->client->osdc, req); 173528127bddSYan, Zheng err = ceph_osdc_wait_request(&fsc->client->osdc, req); 173697e27aaaSXiubo Li 17378ae99ae2SXiubo Li ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 1738903f4fecSXiubo Li req->r_end_latency, len, err); 173997e27aaaSXiubo Li 1740825978fdSXiubo Li out_uninline: 1741083db6fdSDavid Howells if (!err) { 1742083db6fdSDavid Howells int dirty; 1743083db6fdSDavid Howells 1744083db6fdSDavid Howells /* Set to CAP_INLINE_NONE and dirty the caps */ 1745083db6fdSDavid Howells down_read(&fsc->mdsc->snap_rwsem); 1746083db6fdSDavid Howells spin_lock(&ci->i_ceph_lock); 1747083db6fdSDavid Howells ci->i_inline_version = CEPH_INLINE_NONE; 1748083db6fdSDavid Howells dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf); 1749083db6fdSDavid Howells spin_unlock(&ci->i_ceph_lock); 1750083db6fdSDavid Howells up_read(&fsc->mdsc->snap_rwsem); 1751083db6fdSDavid Howells if (dirty) 1752083db6fdSDavid Howells __mark_inode_dirty(inode, dirty); 1753083db6fdSDavid Howells } 1754083db6fdSDavid Howells out_put_req: 175528127bddSYan, Zheng ceph_osdc_put_request(req); 175628127bddSYan, Zheng if (err == -ECANCELED) 175728127bddSYan, Zheng err = 0; 1758083db6fdSDavid Howells out_unlock: 1759825978fdSXiubo Li if (folio) { 1760083db6fdSDavid Howells folio_unlock(folio); 1761083db6fdSDavid Howells folio_put(folio); 1762825978fdSXiubo Li } 176328127bddSYan, Zheng out: 1764083db6fdSDavid Howells ceph_free_cap_flush(prealloc_cf); 176528127bddSYan, Zheng dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", 176628127bddSYan, Zheng inode, ceph_vinop(inode), inline_version, err); 176728127bddSYan, Zheng return err; 176828127bddSYan, Zheng } 176928127bddSYan, Zheng 17707cbea8dcSKirill A. Shutemov static const struct vm_operations_struct ceph_vmops = { 177161f68816SYan, Zheng .fault = ceph_filemap_fault, 17721d3576fdSSage Weil .page_mkwrite = ceph_page_mkwrite, 17731d3576fdSSage Weil }; 17741d3576fdSSage Weil 17751d3576fdSSage Weil int ceph_mmap(struct file *file, struct vm_area_struct *vma) 17761d3576fdSSage Weil { 17771d3576fdSSage Weil struct address_space *mapping = file->f_mapping; 17781d3576fdSSage Weil 17797e0a1265SMatthew Wilcox (Oracle) if (!mapping->a_ops->read_folio) 17801d3576fdSSage Weil return -ENOEXEC; 17811d3576fdSSage Weil vma->vm_ops = &ceph_vmops; 17821d3576fdSSage Weil return 0; 17831d3576fdSSage Weil } 178410183a69SYan, Zheng 178510183a69SYan, Zheng enum { 178610183a69SYan, Zheng POOL_READ = 1, 178710183a69SYan, Zheng POOL_WRITE = 2, 178810183a69SYan, Zheng }; 178910183a69SYan, Zheng 1790779fe0fbSYan, Zheng static int __ceph_pool_perm_get(struct ceph_inode_info *ci, 1791779fe0fbSYan, Zheng s64 pool, struct ceph_string *pool_ns) 179210183a69SYan, Zheng { 1793874c8ca1SDavid Howells struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->netfs.inode); 179410183a69SYan, Zheng struct ceph_mds_client *mdsc = fsc->mdsc; 179510183a69SYan, Zheng struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; 179610183a69SYan, Zheng struct rb_node **p, *parent; 179710183a69SYan, Zheng struct ceph_pool_perm *perm; 179810183a69SYan, Zheng struct page **pages; 1799779fe0fbSYan, Zheng size_t pool_ns_len; 180010183a69SYan, Zheng int err = 0, err2 = 0, have = 0; 180110183a69SYan, Zheng 180210183a69SYan, Zheng down_read(&mdsc->pool_perm_rwsem); 180310183a69SYan, Zheng p = &mdsc->pool_perm_tree.rb_node; 180410183a69SYan, Zheng while (*p) { 180510183a69SYan, Zheng perm = rb_entry(*p, struct ceph_pool_perm, node); 180610183a69SYan, Zheng if (pool < perm->pool) 180710183a69SYan, Zheng p = &(*p)->rb_left; 180810183a69SYan, Zheng else if (pool > perm->pool) 180910183a69SYan, Zheng p = &(*p)->rb_right; 181010183a69SYan, Zheng else { 1811779fe0fbSYan, Zheng int ret = ceph_compare_string(pool_ns, 1812779fe0fbSYan, Zheng perm->pool_ns, 1813779fe0fbSYan, Zheng perm->pool_ns_len); 1814779fe0fbSYan, Zheng if (ret < 0) 1815779fe0fbSYan, Zheng p = &(*p)->rb_left; 1816779fe0fbSYan, Zheng else if (ret > 0) 1817779fe0fbSYan, Zheng p = &(*p)->rb_right; 1818779fe0fbSYan, Zheng else { 181910183a69SYan, Zheng have = perm->perm; 182010183a69SYan, Zheng break; 182110183a69SYan, Zheng } 182210183a69SYan, Zheng } 1823779fe0fbSYan, Zheng } 182410183a69SYan, Zheng up_read(&mdsc->pool_perm_rwsem); 182510183a69SYan, Zheng if (*p) 182610183a69SYan, Zheng goto out; 182710183a69SYan, Zheng 1828779fe0fbSYan, Zheng if (pool_ns) 1829779fe0fbSYan, Zheng dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", 1830779fe0fbSYan, Zheng pool, (int)pool_ns->len, pool_ns->str); 1831779fe0fbSYan, Zheng else 18327627151eSYan, Zheng dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); 183310183a69SYan, Zheng 183410183a69SYan, Zheng down_write(&mdsc->pool_perm_rwsem); 1835779fe0fbSYan, Zheng p = &mdsc->pool_perm_tree.rb_node; 183610183a69SYan, Zheng parent = NULL; 183710183a69SYan, Zheng while (*p) { 183810183a69SYan, Zheng parent = *p; 183910183a69SYan, Zheng perm = rb_entry(parent, struct ceph_pool_perm, node); 184010183a69SYan, Zheng if (pool < perm->pool) 184110183a69SYan, Zheng p = &(*p)->rb_left; 184210183a69SYan, Zheng else if (pool > perm->pool) 184310183a69SYan, Zheng p = &(*p)->rb_right; 184410183a69SYan, Zheng else { 1845779fe0fbSYan, Zheng int ret = ceph_compare_string(pool_ns, 1846779fe0fbSYan, Zheng perm->pool_ns, 1847779fe0fbSYan, Zheng perm->pool_ns_len); 1848779fe0fbSYan, Zheng if (ret < 0) 1849779fe0fbSYan, Zheng p = &(*p)->rb_left; 1850779fe0fbSYan, Zheng else if (ret > 0) 1851779fe0fbSYan, Zheng p = &(*p)->rb_right; 1852779fe0fbSYan, Zheng else { 185310183a69SYan, Zheng have = perm->perm; 185410183a69SYan, Zheng break; 185510183a69SYan, Zheng } 185610183a69SYan, Zheng } 1857779fe0fbSYan, Zheng } 185810183a69SYan, Zheng if (*p) { 185910183a69SYan, Zheng up_write(&mdsc->pool_perm_rwsem); 186010183a69SYan, Zheng goto out; 186110183a69SYan, Zheng } 186210183a69SYan, Zheng 186334b759b4SIlya Dryomov rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 186410183a69SYan, Zheng 1, false, GFP_NOFS); 186510183a69SYan, Zheng if (!rd_req) { 186610183a69SYan, Zheng err = -ENOMEM; 186710183a69SYan, Zheng goto out_unlock; 186810183a69SYan, Zheng } 186910183a69SYan, Zheng 187010183a69SYan, Zheng rd_req->r_flags = CEPH_OSD_FLAG_READ; 187110183a69SYan, Zheng osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); 187210183a69SYan, Zheng rd_req->r_base_oloc.pool = pool; 1873779fe0fbSYan, Zheng if (pool_ns) 1874779fe0fbSYan, Zheng rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); 1875d30291b9SIlya Dryomov ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); 187610183a69SYan, Zheng 187713d1ad16SIlya Dryomov err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); 187813d1ad16SIlya Dryomov if (err) 187913d1ad16SIlya Dryomov goto out_unlock; 188010183a69SYan, Zheng 188134b759b4SIlya Dryomov wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 188210183a69SYan, Zheng 1, false, GFP_NOFS); 188310183a69SYan, Zheng if (!wr_req) { 188410183a69SYan, Zheng err = -ENOMEM; 188510183a69SYan, Zheng goto out_unlock; 188610183a69SYan, Zheng } 188710183a69SYan, Zheng 188854ea0046SIlya Dryomov wr_req->r_flags = CEPH_OSD_FLAG_WRITE; 188910183a69SYan, Zheng osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); 189063244fa1SIlya Dryomov ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); 1891d30291b9SIlya Dryomov ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); 189210183a69SYan, Zheng 189313d1ad16SIlya Dryomov err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); 189413d1ad16SIlya Dryomov if (err) 189513d1ad16SIlya Dryomov goto out_unlock; 189610183a69SYan, Zheng 189710183a69SYan, Zheng /* one page should be large enough for STAT data */ 189810183a69SYan, Zheng pages = ceph_alloc_page_vector(1, GFP_KERNEL); 189910183a69SYan, Zheng if (IS_ERR(pages)) { 190010183a69SYan, Zheng err = PTR_ERR(pages); 190110183a69SYan, Zheng goto out_unlock; 190210183a69SYan, Zheng } 190310183a69SYan, Zheng 190410183a69SYan, Zheng osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, 190510183a69SYan, Zheng 0, false, true); 1906a8af0d68SJeff Layton ceph_osdc_start_request(&fsc->client->osdc, rd_req); 190710183a69SYan, Zheng 1908874c8ca1SDavid Howells wr_req->r_mtime = ci->netfs.inode.i_mtime; 1909a8af0d68SJeff Layton ceph_osdc_start_request(&fsc->client->osdc, wr_req); 191010183a69SYan, Zheng 191110183a69SYan, Zheng err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); 191210183a69SYan, Zheng err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); 191310183a69SYan, Zheng 191410183a69SYan, Zheng if (err >= 0 || err == -ENOENT) 191510183a69SYan, Zheng have |= POOL_READ; 1916131d7eb4SYan, Zheng else if (err != -EPERM) { 19170b98acd6SIlya Dryomov if (err == -EBLOCKLISTED) 19180b98acd6SIlya Dryomov fsc->blocklisted = true; 191910183a69SYan, Zheng goto out_unlock; 1920131d7eb4SYan, Zheng } 192110183a69SYan, Zheng 192210183a69SYan, Zheng if (err2 == 0 || err2 == -EEXIST) 192310183a69SYan, Zheng have |= POOL_WRITE; 192410183a69SYan, Zheng else if (err2 != -EPERM) { 19250b98acd6SIlya Dryomov if (err2 == -EBLOCKLISTED) 19260b98acd6SIlya Dryomov fsc->blocklisted = true; 192710183a69SYan, Zheng err = err2; 192810183a69SYan, Zheng goto out_unlock; 192910183a69SYan, Zheng } 193010183a69SYan, Zheng 1931779fe0fbSYan, Zheng pool_ns_len = pool_ns ? pool_ns->len : 0; 1932779fe0fbSYan, Zheng perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); 193310183a69SYan, Zheng if (!perm) { 193410183a69SYan, Zheng err = -ENOMEM; 193510183a69SYan, Zheng goto out_unlock; 193610183a69SYan, Zheng } 193710183a69SYan, Zheng 193810183a69SYan, Zheng perm->pool = pool; 193910183a69SYan, Zheng perm->perm = have; 1940779fe0fbSYan, Zheng perm->pool_ns_len = pool_ns_len; 1941779fe0fbSYan, Zheng if (pool_ns_len > 0) 1942779fe0fbSYan, Zheng memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); 1943779fe0fbSYan, Zheng perm->pool_ns[pool_ns_len] = 0; 1944779fe0fbSYan, Zheng 194510183a69SYan, Zheng rb_link_node(&perm->node, parent, p); 194610183a69SYan, Zheng rb_insert_color(&perm->node, &mdsc->pool_perm_tree); 194710183a69SYan, Zheng err = 0; 194810183a69SYan, Zheng out_unlock: 194910183a69SYan, Zheng up_write(&mdsc->pool_perm_rwsem); 195010183a69SYan, Zheng 195110183a69SYan, Zheng ceph_osdc_put_request(rd_req); 195210183a69SYan, Zheng ceph_osdc_put_request(wr_req); 195310183a69SYan, Zheng out: 195410183a69SYan, Zheng if (!err) 195510183a69SYan, Zheng err = have; 1956779fe0fbSYan, Zheng if (pool_ns) 1957779fe0fbSYan, Zheng dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", 1958779fe0fbSYan, Zheng pool, (int)pool_ns->len, pool_ns->str, err); 1959779fe0fbSYan, Zheng else 19607627151eSYan, Zheng dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); 196110183a69SYan, Zheng return err; 196210183a69SYan, Zheng } 196310183a69SYan, Zheng 19645e3ded1bSYan, Zheng int ceph_pool_perm_check(struct inode *inode, int need) 196510183a69SYan, Zheng { 19665e3ded1bSYan, Zheng struct ceph_inode_info *ci = ceph_inode(inode); 1967779fe0fbSYan, Zheng struct ceph_string *pool_ns; 19685e3ded1bSYan, Zheng s64 pool; 196910183a69SYan, Zheng int ret, flags; 197010183a69SYan, Zheng 1971e9b22501SJeff Layton /* Only need to do this for regular files */ 1972e9b22501SJeff Layton if (!S_ISREG(inode->i_mode)) 1973e9b22501SJeff Layton return 0; 1974e9b22501SJeff Layton 197580e80fbbSYan, Zheng if (ci->i_vino.snap != CEPH_NOSNAP) { 197680e80fbbSYan, Zheng /* 197780e80fbbSYan, Zheng * Pool permission check needs to write to the first object. 197880e80fbbSYan, Zheng * But for snapshot, head of the first object may have alread 197980e80fbbSYan, Zheng * been deleted. Skip check to avoid creating orphan object. 198080e80fbbSYan, Zheng */ 198180e80fbbSYan, Zheng return 0; 198280e80fbbSYan, Zheng } 198380e80fbbSYan, Zheng 19845e3ded1bSYan, Zheng if (ceph_test_mount_opt(ceph_inode_to_client(inode), 198510183a69SYan, Zheng NOPOOLPERM)) 198610183a69SYan, Zheng return 0; 198710183a69SYan, Zheng 198810183a69SYan, Zheng spin_lock(&ci->i_ceph_lock); 198910183a69SYan, Zheng flags = ci->i_ceph_flags; 19907627151eSYan, Zheng pool = ci->i_layout.pool_id; 199110183a69SYan, Zheng spin_unlock(&ci->i_ceph_lock); 199210183a69SYan, Zheng check: 199310183a69SYan, Zheng if (flags & CEPH_I_POOL_PERM) { 199410183a69SYan, Zheng if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { 19957627151eSYan, Zheng dout("ceph_pool_perm_check pool %lld no read perm\n", 199610183a69SYan, Zheng pool); 199710183a69SYan, Zheng return -EPERM; 199810183a69SYan, Zheng } 199910183a69SYan, Zheng if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { 20007627151eSYan, Zheng dout("ceph_pool_perm_check pool %lld no write perm\n", 200110183a69SYan, Zheng pool); 200210183a69SYan, Zheng return -EPERM; 200310183a69SYan, Zheng } 200410183a69SYan, Zheng return 0; 200510183a69SYan, Zheng } 200610183a69SYan, Zheng 2007779fe0fbSYan, Zheng pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); 2008779fe0fbSYan, Zheng ret = __ceph_pool_perm_get(ci, pool, pool_ns); 2009779fe0fbSYan, Zheng ceph_put_string(pool_ns); 201010183a69SYan, Zheng if (ret < 0) 201110183a69SYan, Zheng return ret; 201210183a69SYan, Zheng 201310183a69SYan, Zheng flags = CEPH_I_POOL_PERM; 201410183a69SYan, Zheng if (ret & POOL_READ) 201510183a69SYan, Zheng flags |= CEPH_I_POOL_RD; 201610183a69SYan, Zheng if (ret & POOL_WRITE) 201710183a69SYan, Zheng flags |= CEPH_I_POOL_WR; 201810183a69SYan, Zheng 201910183a69SYan, Zheng spin_lock(&ci->i_ceph_lock); 2020779fe0fbSYan, Zheng if (pool == ci->i_layout.pool_id && 2021779fe0fbSYan, Zheng pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { 2022779fe0fbSYan, Zheng ci->i_ceph_flags |= flags; 202310183a69SYan, Zheng } else { 20247627151eSYan, Zheng pool = ci->i_layout.pool_id; 202510183a69SYan, Zheng flags = ci->i_ceph_flags; 202610183a69SYan, Zheng } 202710183a69SYan, Zheng spin_unlock(&ci->i_ceph_lock); 202810183a69SYan, Zheng goto check; 202910183a69SYan, Zheng } 203010183a69SYan, Zheng 203110183a69SYan, Zheng void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) 203210183a69SYan, Zheng { 203310183a69SYan, Zheng struct ceph_pool_perm *perm; 203410183a69SYan, Zheng struct rb_node *n; 203510183a69SYan, Zheng 203610183a69SYan, Zheng while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { 203710183a69SYan, Zheng n = rb_first(&mdsc->pool_perm_tree); 203810183a69SYan, Zheng perm = rb_entry(n, struct ceph_pool_perm, node); 203910183a69SYan, Zheng rb_erase(n, &mdsc->pool_perm_tree); 204010183a69SYan, Zheng kfree(perm); 204110183a69SYan, Zheng } 204210183a69SYan, Zheng } 2043