13d14c5d2SYehuda Sadeh #include <linux/ceph/ceph_debug.h> 21d3576fdSSage Weil 31d3576fdSSage Weil #include <linux/backing-dev.h> 41d3576fdSSage Weil #include <linux/fs.h> 51d3576fdSSage Weil #include <linux/mm.h> 61d3576fdSSage Weil #include <linux/pagemap.h> 71d3576fdSSage Weil #include <linux/writeback.h> /* generic_writepages */ 85a0e3ad6STejun Heo #include <linux/slab.h> 91d3576fdSSage Weil #include <linux/pagevec.h> 101d3576fdSSage Weil #include <linux/task_io_accounting_ops.h> 111d3576fdSSage Weil 121d3576fdSSage Weil #include "super.h" 133d14c5d2SYehuda Sadeh #include "mds_client.h" 143d14c5d2SYehuda Sadeh #include <linux/ceph/osd_client.h> 151d3576fdSSage Weil 161d3576fdSSage Weil /* 171d3576fdSSage Weil * Ceph address space ops. 181d3576fdSSage Weil * 191d3576fdSSage Weil * There are a few funny things going on here. 201d3576fdSSage Weil * 211d3576fdSSage Weil * The page->private field is used to reference a struct 221d3576fdSSage Weil * ceph_snap_context for _every_ dirty page. This indicates which 231d3576fdSSage Weil * snapshot the page was logically dirtied in, and thus which snap 241d3576fdSSage Weil * context needs to be associated with the osd write during writeback. 251d3576fdSSage Weil * 261d3576fdSSage Weil * Similarly, struct ceph_inode_info maintains a set of counters to 2725985edcSLucas De Marchi * count dirty pages on the inode. In the absence of snapshots, 281d3576fdSSage Weil * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 291d3576fdSSage Weil * 301d3576fdSSage Weil * When a snapshot is taken (that is, when the client receives 311d3576fdSSage Weil * notification that a snapshot was taken), each inode with caps and 321d3576fdSSage Weil * with dirty pages (dirty pages implies there is a cap) gets a new 331d3576fdSSage Weil * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 341d3576fdSSage Weil * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 351d3576fdSSage Weil * moved to capsnap->dirty. (Unless a sync write is currently in 361d3576fdSSage Weil * progress. In that case, the capsnap is said to be "pending", new 371d3576fdSSage Weil * writes cannot start, and the capsnap isn't "finalized" until the 381d3576fdSSage Weil * write completes (or fails) and a final size/mtime for the inode for 391d3576fdSSage Weil * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 401d3576fdSSage Weil * 411d3576fdSSage Weil * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 421d3576fdSSage Weil * we look for the first capsnap in i_cap_snaps and write out pages in 431d3576fdSSage Weil * that snap context _only_. Then we move on to the next capsnap, 441d3576fdSSage Weil * eventually reaching the "live" or "head" context (i.e., pages that 451d3576fdSSage Weil * are not yet snapped) and are writing the most recently dirtied 461d3576fdSSage Weil * pages. 471d3576fdSSage Weil * 481d3576fdSSage Weil * Invalidate and so forth must take care to ensure the dirty page 491d3576fdSSage Weil * accounting is preserved. 501d3576fdSSage Weil */ 511d3576fdSSage Weil 522baba250SYehuda Sadeh #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 532baba250SYehuda Sadeh #define CONGESTION_OFF_THRESH(congestion_kb) \ 542baba250SYehuda Sadeh (CONGESTION_ON_THRESH(congestion_kb) - \ 552baba250SYehuda Sadeh (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 562baba250SYehuda Sadeh 5761600ef8SYan, Zheng static inline struct ceph_snap_context *page_snap_context(struct page *page) 5861600ef8SYan, Zheng { 5961600ef8SYan, Zheng if (PagePrivate(page)) 6061600ef8SYan, Zheng return (void *)page->private; 6161600ef8SYan, Zheng return NULL; 6261600ef8SYan, Zheng } 631d3576fdSSage Weil 641d3576fdSSage Weil /* 651d3576fdSSage Weil * Dirty a page. Optimistically adjust accounting, on the assumption 661d3576fdSSage Weil * that we won't race with invalidate. If we do, readjust. 671d3576fdSSage Weil */ 681d3576fdSSage Weil static int ceph_set_page_dirty(struct page *page) 691d3576fdSSage Weil { 701d3576fdSSage Weil struct address_space *mapping = page->mapping; 711d3576fdSSage Weil struct inode *inode; 721d3576fdSSage Weil struct ceph_inode_info *ci; 731d3576fdSSage Weil int undo = 0; 741d3576fdSSage Weil struct ceph_snap_context *snapc; 751d3576fdSSage Weil 761d3576fdSSage Weil if (unlikely(!mapping)) 771d3576fdSSage Weil return !TestSetPageDirty(page); 781d3576fdSSage Weil 791d3576fdSSage Weil if (TestSetPageDirty(page)) { 801d3576fdSSage Weil dout("%p set_page_dirty %p idx %lu -- already dirty\n", 811d3576fdSSage Weil mapping->host, page, page->index); 821d3576fdSSage Weil return 0; 831d3576fdSSage Weil } 841d3576fdSSage Weil 851d3576fdSSage Weil inode = mapping->host; 861d3576fdSSage Weil ci = ceph_inode(inode); 871d3576fdSSage Weil 881d3576fdSSage Weil /* 891d3576fdSSage Weil * Note that we're grabbing a snapc ref here without holding 901d3576fdSSage Weil * any locks! 911d3576fdSSage Weil */ 921d3576fdSSage Weil snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context); 931d3576fdSSage Weil 941d3576fdSSage Weil /* dirty the head */ 95be655596SSage Weil spin_lock(&ci->i_ceph_lock); 967d8cb26dSSage Weil if (ci->i_head_snapc == NULL) 971d3576fdSSage Weil ci->i_head_snapc = ceph_get_snap_context(snapc); 981d3576fdSSage Weil ++ci->i_wrbuffer_ref_head; 991d3576fdSSage Weil if (ci->i_wrbuffer_ref == 0) 1000444d76aSDave Chinner ihold(inode); 1011d3576fdSSage Weil ++ci->i_wrbuffer_ref; 1021d3576fdSSage Weil dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " 1031d3576fdSSage Weil "snapc %p seq %lld (%d snaps)\n", 1041d3576fdSSage Weil mapping->host, page, page->index, 1051d3576fdSSage Weil ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 1061d3576fdSSage Weil ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 1071d3576fdSSage Weil snapc, snapc->seq, snapc->num_snaps); 108be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 1091d3576fdSSage Weil 1101d3576fdSSage Weil /* now adjust page */ 1111d3576fdSSage Weil spin_lock_irq(&mapping->tree_lock); 1121d3576fdSSage Weil if (page->mapping) { /* Race with truncate? */ 1131d3576fdSSage Weil WARN_ON_ONCE(!PageUptodate(page)); 114679ceaceSMichael Rubin account_page_dirtied(page, page->mapping); 1151d3576fdSSage Weil radix_tree_tag_set(&mapping->page_tree, 1161d3576fdSSage Weil page_index(page), PAGECACHE_TAG_DIRTY); 1171d3576fdSSage Weil 1181d3576fdSSage Weil /* 1191d3576fdSSage Weil * Reference snap context in page->private. Also set 1201d3576fdSSage Weil * PagePrivate so that we get invalidatepage callback. 1211d3576fdSSage Weil */ 1221d3576fdSSage Weil page->private = (unsigned long)snapc; 1231d3576fdSSage Weil SetPagePrivate(page); 1241d3576fdSSage Weil } else { 1251d3576fdSSage Weil dout("ANON set_page_dirty %p (raced truncate?)\n", page); 1261d3576fdSSage Weil undo = 1; 1271d3576fdSSage Weil } 1281d3576fdSSage Weil 1291d3576fdSSage Weil spin_unlock_irq(&mapping->tree_lock); 1301d3576fdSSage Weil 1311d3576fdSSage Weil if (undo) 1321d3576fdSSage Weil /* whoops, we failed to dirty the page */ 1331d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 1341d3576fdSSage Weil 1351d3576fdSSage Weil __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 1361d3576fdSSage Weil 1371d3576fdSSage Weil BUG_ON(!PageDirty(page)); 1381d3576fdSSage Weil return 1; 1391d3576fdSSage Weil } 1401d3576fdSSage Weil 1411d3576fdSSage Weil /* 1421d3576fdSSage Weil * If we are truncating the full page (i.e. offset == 0), adjust the 1431d3576fdSSage Weil * dirty page counters appropriately. Only called if there is private 1441d3576fdSSage Weil * data on the page. 1451d3576fdSSage Weil */ 1461d3576fdSSage Weil static void ceph_invalidatepage(struct page *page, unsigned long offset) 1471d3576fdSSage Weil { 1484ce1e9adSAlexander Beregalov struct inode *inode; 1491d3576fdSSage Weil struct ceph_inode_info *ci; 15061600ef8SYan, Zheng struct ceph_snap_context *snapc = page_snap_context(page); 1511d3576fdSSage Weil 1521d3576fdSSage Weil BUG_ON(!PageLocked(page)); 1531d3576fdSSage Weil BUG_ON(!PagePrivate(page)); 1541d3576fdSSage Weil BUG_ON(!page->mapping); 1551d3576fdSSage Weil 1564ce1e9adSAlexander Beregalov inode = page->mapping->host; 1574ce1e9adSAlexander Beregalov 1581d3576fdSSage Weil /* 1591d3576fdSSage Weil * We can get non-dirty pages here due to races between 1601d3576fdSSage Weil * set_page_dirty and truncate_complete_page; just spit out a 1611d3576fdSSage Weil * warning, in case we end up with accounting problems later. 1621d3576fdSSage Weil */ 1631d3576fdSSage Weil if (!PageDirty(page)) 1641d3576fdSSage Weil pr_err("%p invalidatepage %p page not dirty\n", inode, page); 1651d3576fdSSage Weil 1661d3576fdSSage Weil if (offset == 0) 1671d3576fdSSage Weil ClearPageChecked(page); 1681d3576fdSSage Weil 1691d3576fdSSage Weil ci = ceph_inode(inode); 1701d3576fdSSage Weil if (offset == 0) { 1711d3576fdSSage Weil dout("%p invalidatepage %p idx %lu full dirty page %lu\n", 1721d3576fdSSage Weil inode, page, page->index, offset); 1731d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 1741d3576fdSSage Weil ceph_put_snap_context(snapc); 1751d3576fdSSage Weil page->private = 0; 1761d3576fdSSage Weil ClearPagePrivate(page); 1771d3576fdSSage Weil } else { 1781d3576fdSSage Weil dout("%p invalidatepage %p idx %lu partial dirty page\n", 1791d3576fdSSage Weil inode, page, page->index); 1801d3576fdSSage Weil } 1811d3576fdSSage Weil } 1821d3576fdSSage Weil 1831d3576fdSSage Weil /* just a sanity check */ 1841d3576fdSSage Weil static int ceph_releasepage(struct page *page, gfp_t g) 1851d3576fdSSage Weil { 1861d3576fdSSage Weil struct inode *inode = page->mapping ? page->mapping->host : NULL; 1871d3576fdSSage Weil dout("%p releasepage %p idx %lu\n", inode, page, page->index); 1881d3576fdSSage Weil WARN_ON(PageDirty(page)); 1891d3576fdSSage Weil WARN_ON(PagePrivate(page)); 1901d3576fdSSage Weil return 0; 1911d3576fdSSage Weil } 1921d3576fdSSage Weil 1931d3576fdSSage Weil /* 1941d3576fdSSage Weil * read a single page, without unlocking it. 1951d3576fdSSage Weil */ 1961d3576fdSSage Weil static int readpage_nounlock(struct file *filp, struct page *page) 1971d3576fdSSage Weil { 1981d3576fdSSage Weil struct inode *inode = filp->f_dentry->d_inode; 1991d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2003d14c5d2SYehuda Sadeh struct ceph_osd_client *osdc = 2013d14c5d2SYehuda Sadeh &ceph_inode_to_client(inode)->client->osdc; 2021d3576fdSSage Weil int err = 0; 2031d3576fdSSage Weil u64 len = PAGE_CACHE_SIZE; 2041d3576fdSSage Weil 2051d3576fdSSage Weil dout("readpage inode %p file %p page %p index %lu\n", 2061d3576fdSSage Weil inode, filp, page, page->index); 2071d3576fdSSage Weil err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, 2086285bc23SAlex Elder (u64) page_offset(page), &len, 2091d3576fdSSage Weil ci->i_truncate_seq, ci->i_truncate_size, 210b7495fc2SSage Weil &page, 1, 0); 2111d3576fdSSage Weil if (err == -ENOENT) 2121d3576fdSSage Weil err = 0; 2131d3576fdSSage Weil if (err < 0) { 2141d3576fdSSage Weil SetPageError(page); 2151d3576fdSSage Weil goto out; 2161d3576fdSSage Weil } else if (err < PAGE_CACHE_SIZE) { 2171d3576fdSSage Weil /* zero fill remainder of page */ 2181d3576fdSSage Weil zero_user_segment(page, err, PAGE_CACHE_SIZE); 2191d3576fdSSage Weil } 2201d3576fdSSage Weil SetPageUptodate(page); 2211d3576fdSSage Weil 2221d3576fdSSage Weil out: 2231d3576fdSSage Weil return err < 0 ? err : 0; 2241d3576fdSSage Weil } 2251d3576fdSSage Weil 2261d3576fdSSage Weil static int ceph_readpage(struct file *filp, struct page *page) 2271d3576fdSSage Weil { 2281d3576fdSSage Weil int r = readpage_nounlock(filp, page); 2291d3576fdSSage Weil unlock_page(page); 2301d3576fdSSage Weil return r; 2311d3576fdSSage Weil } 2321d3576fdSSage Weil 2331d3576fdSSage Weil /* 2347c272194SSage Weil * Finish an async read(ahead) op. 2351d3576fdSSage Weil */ 2367c272194SSage Weil static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg) 2371d3576fdSSage Weil { 2387c272194SSage Weil struct inode *inode = req->r_inode; 2397c272194SSage Weil struct ceph_osd_reply_head *replyhead; 2407c272194SSage Weil int rc, bytes; 2417c272194SSage Weil int i; 2427c272194SSage Weil 2437c272194SSage Weil /* parse reply */ 2447c272194SSage Weil replyhead = msg->front.iov_base; 2457c272194SSage Weil WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); 2467c272194SSage Weil rc = le32_to_cpu(replyhead->result); 2477c272194SSage Weil bytes = le32_to_cpu(msg->hdr.data_len); 2487c272194SSage Weil 2497c272194SSage Weil dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); 2507c272194SSage Weil 2517c272194SSage Weil /* unlock all pages, zeroing any data we didn't read */ 2527c272194SSage Weil for (i = 0; i < req->r_num_pages; i++, bytes -= PAGE_CACHE_SIZE) { 2537c272194SSage Weil struct page *page = req->r_pages[i]; 2547c272194SSage Weil 2557c272194SSage Weil if (bytes < (int)PAGE_CACHE_SIZE) { 2567c272194SSage Weil /* zero (remainder of) page */ 2577c272194SSage Weil int s = bytes < 0 ? 0 : bytes; 2587c272194SSage Weil zero_user_segment(page, s, PAGE_CACHE_SIZE); 2597c272194SSage Weil } 2607c272194SSage Weil dout("finish_read %p uptodate %p idx %lu\n", inode, page, 2617c272194SSage Weil page->index); 2627c272194SSage Weil flush_dcache_page(page); 2637c272194SSage Weil SetPageUptodate(page); 2647c272194SSage Weil unlock_page(page); 2657c272194SSage Weil page_cache_release(page); 2667c272194SSage Weil } 2677c272194SSage Weil kfree(req->r_pages); 2687c272194SSage Weil } 2697c272194SSage Weil 2707c272194SSage Weil /* 2717c272194SSage Weil * start an async read(ahead) operation. return nr_pages we submitted 2727c272194SSage Weil * a read for on success, or negative error code. 2737c272194SSage Weil */ 2740d66a487SSage Weil static int start_read(struct inode *inode, struct list_head *page_list, int max) 2757c272194SSage Weil { 2767c272194SSage Weil struct ceph_osd_client *osdc = 2777c272194SSage Weil &ceph_inode_to_client(inode)->client->osdc; 2787c272194SSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 2797c272194SSage Weil struct page *page = list_entry(page_list->prev, struct page, lru); 2807c272194SSage Weil struct ceph_osd_request *req; 2817c272194SSage Weil u64 off; 2827c272194SSage Weil u64 len; 2837c272194SSage Weil int i; 2841d3576fdSSage Weil struct page **pages; 2857c272194SSage Weil pgoff_t next_index; 2867c272194SSage Weil int nr_pages = 0; 2877c272194SSage Weil int ret; 2887c272194SSage Weil 2896285bc23SAlex Elder off = (u64) page_offset(page); 2907c272194SSage Weil 2917c272194SSage Weil /* count pages */ 2927c272194SSage Weil next_index = page->index; 2937c272194SSage Weil list_for_each_entry_reverse(page, page_list, lru) { 2947c272194SSage Weil if (page->index != next_index) 2957c272194SSage Weil break; 2967c272194SSage Weil nr_pages++; 2977c272194SSage Weil next_index++; 2980d66a487SSage Weil if (max && nr_pages == max) 2990d66a487SSage Weil break; 3007c272194SSage Weil } 3017c272194SSage Weil len = nr_pages << PAGE_CACHE_SHIFT; 3027c272194SSage Weil dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, 3037c272194SSage Weil off, len); 3047c272194SSage Weil 3057c272194SSage Weil req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), 3067c272194SSage Weil off, &len, 3077c272194SSage Weil CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, 3087c272194SSage Weil NULL, 0, 3097c272194SSage Weil ci->i_truncate_seq, ci->i_truncate_size, 3107c272194SSage Weil NULL, false, 1, 0); 3116816282dSSage Weil if (IS_ERR(req)) 3126816282dSSage Weil return PTR_ERR(req); 3131d3576fdSSage Weil 3141d3576fdSSage Weil /* build page vector */ 3157c272194SSage Weil nr_pages = len >> PAGE_CACHE_SHIFT; 3167c272194SSage Weil pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS); 3177c272194SSage Weil ret = -ENOMEM; 3181d3576fdSSage Weil if (!pages) 3197c272194SSage Weil goto out; 3207c272194SSage Weil for (i = 0; i < nr_pages; ++i) { 3217c272194SSage Weil page = list_entry(page_list->prev, struct page, lru); 3227c272194SSage Weil BUG_ON(PageLocked(page)); 3237c272194SSage Weil list_del(&page->lru); 3241d3576fdSSage Weil 3257c272194SSage Weil dout("start_read %p adding %p idx %lu\n", inode, page, 3267c272194SSage Weil page->index); 3277c272194SSage Weil if (add_to_page_cache_lru(page, &inode->i_data, page->index, 3287c272194SSage Weil GFP_NOFS)) { 3297c272194SSage Weil page_cache_release(page); 3307c272194SSage Weil dout("start_read %p add_to_page_cache failed %p\n", 3317c272194SSage Weil inode, page); 3327c272194SSage Weil nr_pages = i; 3337c272194SSage Weil goto out_pages; 3341d3576fdSSage Weil } 3357c272194SSage Weil pages[i] = page; 3361d3576fdSSage Weil } 3377c272194SSage Weil req->r_pages = pages; 3387c272194SSage Weil req->r_num_pages = nr_pages; 3397c272194SSage Weil req->r_callback = finish_read; 3407c272194SSage Weil req->r_inode = inode; 3417c272194SSage Weil 3427c272194SSage Weil dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); 3437c272194SSage Weil ret = ceph_osdc_start_request(osdc, req, false); 3447c272194SSage Weil if (ret < 0) 3457c272194SSage Weil goto out_pages; 3467c272194SSage Weil ceph_osdc_put_request(req); 3477c272194SSage Weil return nr_pages; 3487c272194SSage Weil 3497c272194SSage Weil out_pages: 3507c272194SSage Weil ceph_release_page_vector(pages, nr_pages); 3517c272194SSage Weil out: 3527c272194SSage Weil ceph_osdc_put_request(req); 3537c272194SSage Weil return ret; 3541d3576fdSSage Weil } 3551d3576fdSSage Weil 3567c272194SSage Weil 3571d3576fdSSage Weil /* 3581d3576fdSSage Weil * Read multiple pages. Leave pages we don't read + unlock in page_list; 3591d3576fdSSage Weil * the caller (VM) cleans them up. 3601d3576fdSSage Weil */ 3611d3576fdSSage Weil static int ceph_readpages(struct file *file, struct address_space *mapping, 3621d3576fdSSage Weil struct list_head *page_list, unsigned nr_pages) 3631d3576fdSSage Weil { 3641d3576fdSSage Weil struct inode *inode = file->f_dentry->d_inode; 3650d66a487SSage Weil struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 3661d3576fdSSage Weil int rc = 0; 3670d66a487SSage Weil int max = 0; 3681d3576fdSSage Weil 3690d66a487SSage Weil if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE) 3700d66a487SSage Weil max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1) 3710d66a487SSage Weil >> PAGE_SHIFT; 3720d66a487SSage Weil 3730d66a487SSage Weil dout("readpages %p file %p nr_pages %d max %d\n", inode, file, nr_pages, 3740d66a487SSage Weil max); 3757c272194SSage Weil while (!list_empty(page_list)) { 3760d66a487SSage Weil rc = start_read(inode, page_list, max); 3771d3576fdSSage Weil if (rc < 0) 3781d3576fdSSage Weil goto out; 3797c272194SSage Weil BUG_ON(rc == 0); 3801d3576fdSSage Weil } 3811d3576fdSSage Weil out: 3827c272194SSage Weil dout("readpages %p file %p ret %d\n", inode, file, rc); 3831d3576fdSSage Weil return rc; 3841d3576fdSSage Weil } 3851d3576fdSSage Weil 3861d3576fdSSage Weil /* 3871d3576fdSSage Weil * Get ref for the oldest snapc for an inode with dirty data... that is, the 3881d3576fdSSage Weil * only snap context we are allowed to write back. 3891d3576fdSSage Weil */ 3906298a337SSage Weil static struct ceph_snap_context *get_oldest_context(struct inode *inode, 3911d3576fdSSage Weil u64 *snap_size) 3921d3576fdSSage Weil { 3931d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 3941d3576fdSSage Weil struct ceph_snap_context *snapc = NULL; 3951d3576fdSSage Weil struct ceph_cap_snap *capsnap = NULL; 3961d3576fdSSage Weil 397be655596SSage Weil spin_lock(&ci->i_ceph_lock); 3981d3576fdSSage Weil list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 3991d3576fdSSage Weil dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 4001d3576fdSSage Weil capsnap->context, capsnap->dirty_pages); 4011d3576fdSSage Weil if (capsnap->dirty_pages) { 4021d3576fdSSage Weil snapc = ceph_get_snap_context(capsnap->context); 4031d3576fdSSage Weil if (snap_size) 4041d3576fdSSage Weil *snap_size = capsnap->size; 4051d3576fdSSage Weil break; 4061d3576fdSSage Weil } 4071d3576fdSSage Weil } 4087d8cb26dSSage Weil if (!snapc && ci->i_wrbuffer_ref_head) { 40980e755feSSage Weil snapc = ceph_get_snap_context(ci->i_head_snapc); 4101d3576fdSSage Weil dout(" head snapc %p has %d dirty pages\n", 4111d3576fdSSage Weil snapc, ci->i_wrbuffer_ref_head); 4121d3576fdSSage Weil } 413be655596SSage Weil spin_unlock(&ci->i_ceph_lock); 4141d3576fdSSage Weil return snapc; 4151d3576fdSSage Weil } 4161d3576fdSSage Weil 4171d3576fdSSage Weil /* 4181d3576fdSSage Weil * Write a single page, but leave the page locked. 4191d3576fdSSage Weil * 4201d3576fdSSage Weil * If we get a write error, set the page error bit, but still adjust the 4211d3576fdSSage Weil * dirty page accounting (i.e., page is no longer dirty). 4221d3576fdSSage Weil */ 4231d3576fdSSage Weil static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 4241d3576fdSSage Weil { 4251d3576fdSSage Weil struct inode *inode; 4261d3576fdSSage Weil struct ceph_inode_info *ci; 4273d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc; 4281d3576fdSSage Weil struct ceph_osd_client *osdc; 4296285bc23SAlex Elder loff_t page_off = page_offset(page); 4301d3576fdSSage Weil int len = PAGE_CACHE_SIZE; 4311d3576fdSSage Weil loff_t i_size; 4321d3576fdSSage Weil int err = 0; 4336298a337SSage Weil struct ceph_snap_context *snapc, *oldest; 4341d3576fdSSage Weil u64 snap_size = 0; 4352baba250SYehuda Sadeh long writeback_stat; 4361d3576fdSSage Weil 4371d3576fdSSage Weil dout("writepage %p idx %lu\n", page, page->index); 4381d3576fdSSage Weil 4391d3576fdSSage Weil if (!page->mapping || !page->mapping->host) { 4401d3576fdSSage Weil dout("writepage %p - no mapping\n", page); 4411d3576fdSSage Weil return -EFAULT; 4421d3576fdSSage Weil } 4431d3576fdSSage Weil inode = page->mapping->host; 4441d3576fdSSage Weil ci = ceph_inode(inode); 4453d14c5d2SYehuda Sadeh fsc = ceph_inode_to_client(inode); 4463d14c5d2SYehuda Sadeh osdc = &fsc->client->osdc; 4471d3576fdSSage Weil 4481d3576fdSSage Weil /* verify this is a writeable snap context */ 44961600ef8SYan, Zheng snapc = page_snap_context(page); 4501d3576fdSSage Weil if (snapc == NULL) { 4511d3576fdSSage Weil dout("writepage %p page %p not dirty?\n", inode, page); 4521d3576fdSSage Weil goto out; 4531d3576fdSSage Weil } 4546298a337SSage Weil oldest = get_oldest_context(inode, &snap_size); 4556298a337SSage Weil if (snapc->seq > oldest->seq) { 4561d3576fdSSage Weil dout("writepage %p page %p snapc %p not writeable - noop\n", 45761600ef8SYan, Zheng inode, page, snapc); 4581d3576fdSSage Weil /* we should only noop if called by kswapd */ 4591d3576fdSSage Weil WARN_ON((current->flags & PF_MEMALLOC) == 0); 4606298a337SSage Weil ceph_put_snap_context(oldest); 4611d3576fdSSage Weil goto out; 4621d3576fdSSage Weil } 4636298a337SSage Weil ceph_put_snap_context(oldest); 4641d3576fdSSage Weil 4651d3576fdSSage Weil /* is this a partial page at end of file? */ 4661d3576fdSSage Weil if (snap_size) 4671d3576fdSSage Weil i_size = snap_size; 4681d3576fdSSage Weil else 4691d3576fdSSage Weil i_size = i_size_read(inode); 4701d3576fdSSage Weil if (i_size < page_off + len) 4711d3576fdSSage Weil len = i_size - page_off; 4721d3576fdSSage Weil 473ae00d4f3SSage Weil dout("writepage %p page %p index %lu on %llu~%u snapc %p\n", 474ae00d4f3SSage Weil inode, page, page->index, page_off, len, snapc); 4751d3576fdSSage Weil 4763d14c5d2SYehuda Sadeh writeback_stat = atomic_long_inc_return(&fsc->writeback_count); 4772baba250SYehuda Sadeh if (writeback_stat > 4783d14c5d2SYehuda Sadeh CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 4793d14c5d2SYehuda Sadeh set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC); 4802baba250SYehuda Sadeh 4811d3576fdSSage Weil set_page_writeback(page); 4821d3576fdSSage Weil err = ceph_osdc_writepages(osdc, ceph_vino(inode), 4831d3576fdSSage Weil &ci->i_layout, snapc, 4841d3576fdSSage Weil page_off, len, 4851d3576fdSSage Weil ci->i_truncate_seq, ci->i_truncate_size, 4861d3576fdSSage Weil &inode->i_mtime, 4871d3576fdSSage Weil &page, 1, 0, 0, true); 4881d3576fdSSage Weil if (err < 0) { 4891d3576fdSSage Weil dout("writepage setting page/mapping error %d %p\n", err, page); 4901d3576fdSSage Weil SetPageError(page); 4911d3576fdSSage Weil mapping_set_error(&inode->i_data, err); 4921d3576fdSSage Weil if (wbc) 4931d3576fdSSage Weil wbc->pages_skipped++; 4941d3576fdSSage Weil } else { 4951d3576fdSSage Weil dout("writepage cleaned page %p\n", page); 4961d3576fdSSage Weil err = 0; /* vfs expects us to return 0 */ 4971d3576fdSSage Weil } 4981d3576fdSSage Weil page->private = 0; 4991d3576fdSSage Weil ClearPagePrivate(page); 5001d3576fdSSage Weil end_page_writeback(page); 5011d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 5026298a337SSage Weil ceph_put_snap_context(snapc); /* page's reference */ 5031d3576fdSSage Weil out: 5041d3576fdSSage Weil return err; 5051d3576fdSSage Weil } 5061d3576fdSSage Weil 5071d3576fdSSage Weil static int ceph_writepage(struct page *page, struct writeback_control *wbc) 5081d3576fdSSage Weil { 509dbd646a8SYehuda Sadeh int err; 510dbd646a8SYehuda Sadeh struct inode *inode = page->mapping->host; 511dbd646a8SYehuda Sadeh BUG_ON(!inode); 51270b666c3SSage Weil ihold(inode); 513dbd646a8SYehuda Sadeh err = writepage_nounlock(page, wbc); 5141d3576fdSSage Weil unlock_page(page); 515dbd646a8SYehuda Sadeh iput(inode); 5161d3576fdSSage Weil return err; 5171d3576fdSSage Weil } 5181d3576fdSSage Weil 5191d3576fdSSage Weil 5201d3576fdSSage Weil /* 5211d3576fdSSage Weil * lame release_pages helper. release_pages() isn't exported to 5221d3576fdSSage Weil * modules. 5231d3576fdSSage Weil */ 5241d3576fdSSage Weil static void ceph_release_pages(struct page **pages, int num) 5251d3576fdSSage Weil { 5261d3576fdSSage Weil struct pagevec pvec; 5271d3576fdSSage Weil int i; 5281d3576fdSSage Weil 5291d3576fdSSage Weil pagevec_init(&pvec, 0); 5301d3576fdSSage Weil for (i = 0; i < num; i++) { 5311d3576fdSSage Weil if (pagevec_add(&pvec, pages[i]) == 0) 5321d3576fdSSage Weil pagevec_release(&pvec); 5331d3576fdSSage Weil } 5341d3576fdSSage Weil pagevec_release(&pvec); 5351d3576fdSSage Weil } 5361d3576fdSSage Weil 5371d3576fdSSage Weil 5381d3576fdSSage Weil /* 5391d3576fdSSage Weil * async writeback completion handler. 5401d3576fdSSage Weil * 5411d3576fdSSage Weil * If we get an error, set the mapping error bit, but not the individual 5421d3576fdSSage Weil * page error bits. 5431d3576fdSSage Weil */ 5441d3576fdSSage Weil static void writepages_finish(struct ceph_osd_request *req, 5451d3576fdSSage Weil struct ceph_msg *msg) 5461d3576fdSSage Weil { 5471d3576fdSSage Weil struct inode *inode = req->r_inode; 5481d3576fdSSage Weil struct ceph_osd_reply_head *replyhead; 5491d3576fdSSage Weil struct ceph_osd_op *op; 5501d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 5511d3576fdSSage Weil unsigned wrote; 5521d3576fdSSage Weil struct page *page; 5531d3576fdSSage Weil int i; 5541d3576fdSSage Weil struct ceph_snap_context *snapc = req->r_snapc; 5551d3576fdSSage Weil struct address_space *mapping = inode->i_mapping; 5561d3576fdSSage Weil __s32 rc = -EIO; 5571d3576fdSSage Weil u64 bytes = 0; 5583d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 5592baba250SYehuda Sadeh long writeback_stat; 5607ff899daSSage Weil unsigned issued = ceph_caps_issued(ci); 5611d3576fdSSage Weil 5621d3576fdSSage Weil /* parse reply */ 5631d3576fdSSage Weil replyhead = msg->front.iov_base; 5641d3576fdSSage Weil WARN_ON(le32_to_cpu(replyhead->num_ops) == 0); 5651d3576fdSSage Weil op = (void *)(replyhead + 1); 5661d3576fdSSage Weil rc = le32_to_cpu(replyhead->result); 5671d3576fdSSage Weil bytes = le64_to_cpu(op->extent.length); 5681d3576fdSSage Weil 5691d3576fdSSage Weil if (rc >= 0) { 57079788c69SSage Weil /* 57179788c69SSage Weil * Assume we wrote the pages we originally sent. The 57279788c69SSage Weil * osd might reply with fewer pages if our writeback 57379788c69SSage Weil * raced with a truncation and was adjusted at the osd, 57479788c69SSage Weil * so don't believe the reply. 57579788c69SSage Weil */ 57679788c69SSage Weil wrote = req->r_num_pages; 5771d3576fdSSage Weil } else { 5781d3576fdSSage Weil wrote = 0; 5791d3576fdSSage Weil mapping_set_error(mapping, rc); 5801d3576fdSSage Weil } 5811d3576fdSSage Weil dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n", 5821d3576fdSSage Weil inode, rc, bytes, wrote); 5831d3576fdSSage Weil 5841d3576fdSSage Weil /* clean all pages */ 5851d3576fdSSage Weil for (i = 0; i < req->r_num_pages; i++) { 5861d3576fdSSage Weil page = req->r_pages[i]; 5871d3576fdSSage Weil BUG_ON(!page); 5881d3576fdSSage Weil WARN_ON(!PageUptodate(page)); 5891d3576fdSSage Weil 5902baba250SYehuda Sadeh writeback_stat = 5913d14c5d2SYehuda Sadeh atomic_long_dec_return(&fsc->writeback_count); 5922baba250SYehuda Sadeh if (writeback_stat < 5933d14c5d2SYehuda Sadeh CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) 5943d14c5d2SYehuda Sadeh clear_bdi_congested(&fsc->backing_dev_info, 5952baba250SYehuda Sadeh BLK_RW_ASYNC); 5962baba250SYehuda Sadeh 59761600ef8SYan, Zheng ceph_put_snap_context(page_snap_context(page)); 5981d3576fdSSage Weil page->private = 0; 5991d3576fdSSage Weil ClearPagePrivate(page); 6001d3576fdSSage Weil dout("unlocking %d %p\n", i, page); 6011d3576fdSSage Weil end_page_writeback(page); 602e63dc5c7SYehuda Sadeh 603e63dc5c7SYehuda Sadeh /* 604e63dc5c7SYehuda Sadeh * We lost the cache cap, need to truncate the page before 605e63dc5c7SYehuda Sadeh * it is unlocked, otherwise we'd truncate it later in the 606e63dc5c7SYehuda Sadeh * page truncation thread, possibly losing some data that 607e63dc5c7SYehuda Sadeh * raced its way in 608e63dc5c7SYehuda Sadeh */ 6092962507cSSage Weil if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) 610e63dc5c7SYehuda Sadeh generic_error_remove_page(inode->i_mapping, page); 611e63dc5c7SYehuda Sadeh 6121d3576fdSSage Weil unlock_page(page); 6131d3576fdSSage Weil } 6141d3576fdSSage Weil dout("%p wrote+cleaned %d pages\n", inode, wrote); 6151d3576fdSSage Weil ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc); 6161d3576fdSSage Weil 6171d3576fdSSage Weil ceph_release_pages(req->r_pages, req->r_num_pages); 6181d3576fdSSage Weil if (req->r_pages_from_pool) 6191d3576fdSSage Weil mempool_free(req->r_pages, 620640ef79dSCheng Renquan ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); 6211d3576fdSSage Weil else 6221d3576fdSSage Weil kfree(req->r_pages); 6231d3576fdSSage Weil ceph_osdc_put_request(req); 6241d3576fdSSage Weil } 6251d3576fdSSage Weil 6261d3576fdSSage Weil /* 6271d3576fdSSage Weil * allocate a page vec, either directly, or if necessary, via a the 6281d3576fdSSage Weil * mempool. we avoid the mempool if we can because req->r_num_pages 6291d3576fdSSage Weil * may be less than the maximum write size. 6301d3576fdSSage Weil */ 6313d14c5d2SYehuda Sadeh static void alloc_page_vec(struct ceph_fs_client *fsc, 6321d3576fdSSage Weil struct ceph_osd_request *req) 6331d3576fdSSage Weil { 6341d3576fdSSage Weil req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages, 6351d3576fdSSage Weil GFP_NOFS); 6361d3576fdSSage Weil if (!req->r_pages) { 6373d14c5d2SYehuda Sadeh req->r_pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS); 6381d3576fdSSage Weil req->r_pages_from_pool = 1; 6391d3576fdSSage Weil WARN_ON(!req->r_pages); 6401d3576fdSSage Weil } 6411d3576fdSSage Weil } 6421d3576fdSSage Weil 6431d3576fdSSage Weil /* 6441d3576fdSSage Weil * initiate async writeback 6451d3576fdSSage Weil */ 6461d3576fdSSage Weil static int ceph_writepages_start(struct address_space *mapping, 6471d3576fdSSage Weil struct writeback_control *wbc) 6481d3576fdSSage Weil { 6491d3576fdSSage Weil struct inode *inode = mapping->host; 6501d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 6513d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc; 6521d3576fdSSage Weil pgoff_t index, start, end; 6531d3576fdSSage Weil int range_whole = 0; 6541d3576fdSSage Weil int should_loop = 1; 6551d3576fdSSage Weil pgoff_t max_pages = 0, max_pages_ever = 0; 65680e755feSSage Weil struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 6571d3576fdSSage Weil struct pagevec pvec; 6581d3576fdSSage Weil int done = 0; 6591d3576fdSSage Weil int rc = 0; 6601d3576fdSSage Weil unsigned wsize = 1 << inode->i_blkbits; 6611d3576fdSSage Weil struct ceph_osd_request *req = NULL; 6621d3576fdSSage Weil int do_sync; 6631d3576fdSSage Weil u64 snap_size = 0; 6641d3576fdSSage Weil 6651d3576fdSSage Weil /* 6661d3576fdSSage Weil * Include a 'sync' in the OSD request if this is a data 6671d3576fdSSage Weil * integrity write (e.g., O_SYNC write or fsync()), or if our 6681d3576fdSSage Weil * cap is being revoked. 6691d3576fdSSage Weil */ 6701d3576fdSSage Weil do_sync = wbc->sync_mode == WB_SYNC_ALL; 6711d3576fdSSage Weil if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER)) 6721d3576fdSSage Weil do_sync = 1; 6731d3576fdSSage Weil dout("writepages_start %p dosync=%d (mode=%s)\n", 6741d3576fdSSage Weil inode, do_sync, 6751d3576fdSSage Weil wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 6761d3576fdSSage Weil (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 6771d3576fdSSage Weil 6783d14c5d2SYehuda Sadeh fsc = ceph_inode_to_client(inode); 6793d14c5d2SYehuda Sadeh if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) { 6801d3576fdSSage Weil pr_warning("writepage_start %p on forced umount\n", inode); 6811d3576fdSSage Weil return -EIO; /* we're in a forced umount, don't write! */ 6821d3576fdSSage Weil } 6833d14c5d2SYehuda Sadeh if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize) 6843d14c5d2SYehuda Sadeh wsize = fsc->mount_options->wsize; 6851d3576fdSSage Weil if (wsize < PAGE_CACHE_SIZE) 6861d3576fdSSage Weil wsize = PAGE_CACHE_SIZE; 6871d3576fdSSage Weil max_pages_ever = wsize >> PAGE_CACHE_SHIFT; 6881d3576fdSSage Weil 6891d3576fdSSage Weil pagevec_init(&pvec, 0); 6901d3576fdSSage Weil 6911d3576fdSSage Weil /* where to start/end? */ 6921d3576fdSSage Weil if (wbc->range_cyclic) { 6931d3576fdSSage Weil start = mapping->writeback_index; /* Start from prev offset */ 6941d3576fdSSage Weil end = -1; 6951d3576fdSSage Weil dout(" cyclic, start at %lu\n", start); 6961d3576fdSSage Weil } else { 6971d3576fdSSage Weil start = wbc->range_start >> PAGE_CACHE_SHIFT; 6981d3576fdSSage Weil end = wbc->range_end >> PAGE_CACHE_SHIFT; 6991d3576fdSSage Weil if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 7001d3576fdSSage Weil range_whole = 1; 7011d3576fdSSage Weil should_loop = 0; 7021d3576fdSSage Weil dout(" not cyclic, %lu to %lu\n", start, end); 7031d3576fdSSage Weil } 7041d3576fdSSage Weil index = start; 7051d3576fdSSage Weil 7061d3576fdSSage Weil retry: 7071d3576fdSSage Weil /* find oldest snap context with dirty data */ 7081d3576fdSSage Weil ceph_put_snap_context(snapc); 7091d3576fdSSage Weil snapc = get_oldest_context(inode, &snap_size); 7101d3576fdSSage Weil if (!snapc) { 7111d3576fdSSage Weil /* hmm, why does writepages get called when there 7121d3576fdSSage Weil is no dirty data? */ 7131d3576fdSSage Weil dout(" no snap context with dirty data?\n"); 7141d3576fdSSage Weil goto out; 7151d3576fdSSage Weil } 7161d3576fdSSage Weil dout(" oldest snapc is %p seq %lld (%d snaps)\n", 7171d3576fdSSage Weil snapc, snapc->seq, snapc->num_snaps); 7181d3576fdSSage Weil if (last_snapc && snapc != last_snapc) { 7191d3576fdSSage Weil /* if we switched to a newer snapc, restart our scan at the 7201d3576fdSSage Weil * start of the original file range. */ 7211d3576fdSSage Weil dout(" snapc differs from last pass, restarting at %lu\n", 7221d3576fdSSage Weil index); 7231d3576fdSSage Weil index = start; 7241d3576fdSSage Weil } 7251d3576fdSSage Weil last_snapc = snapc; 7261d3576fdSSage Weil 7271d3576fdSSage Weil while (!done && index <= end) { 7281d3576fdSSage Weil unsigned i; 7291d3576fdSSage Weil int first; 7301d3576fdSSage Weil pgoff_t next; 7311d3576fdSSage Weil int pvec_pages, locked_pages; 7321d3576fdSSage Weil struct page *page; 7331d3576fdSSage Weil int want; 7341d3576fdSSage Weil u64 offset, len; 7351d3576fdSSage Weil struct ceph_osd_request_head *reqhead; 7361d3576fdSSage Weil struct ceph_osd_op *op; 7372baba250SYehuda Sadeh long writeback_stat; 7381d3576fdSSage Weil 7391d3576fdSSage Weil next = 0; 7401d3576fdSSage Weil locked_pages = 0; 7411d3576fdSSage Weil max_pages = max_pages_ever; 7421d3576fdSSage Weil 7431d3576fdSSage Weil get_more_pages: 7441d3576fdSSage Weil first = -1; 7451d3576fdSSage Weil want = min(end - index, 7461d3576fdSSage Weil min((pgoff_t)PAGEVEC_SIZE, 7471d3576fdSSage Weil max_pages - (pgoff_t)locked_pages) - 1) 7481d3576fdSSage Weil + 1; 7491d3576fdSSage Weil pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index, 7501d3576fdSSage Weil PAGECACHE_TAG_DIRTY, 7511d3576fdSSage Weil want); 7521d3576fdSSage Weil dout("pagevec_lookup_tag got %d\n", pvec_pages); 7531d3576fdSSage Weil if (!pvec_pages && !locked_pages) 7541d3576fdSSage Weil break; 7551d3576fdSSage Weil for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 7561d3576fdSSage Weil page = pvec.pages[i]; 7571d3576fdSSage Weil dout("? %p idx %lu\n", page, page->index); 7581d3576fdSSage Weil if (locked_pages == 0) 7591d3576fdSSage Weil lock_page(page); /* first page */ 7601d3576fdSSage Weil else if (!trylock_page(page)) 7611d3576fdSSage Weil break; 7621d3576fdSSage Weil 7631d3576fdSSage Weil /* only dirty pages, or our accounting breaks */ 7641d3576fdSSage Weil if (unlikely(!PageDirty(page)) || 7651d3576fdSSage Weil unlikely(page->mapping != mapping)) { 7661d3576fdSSage Weil dout("!dirty or !mapping %p\n", page); 7671d3576fdSSage Weil unlock_page(page); 7681d3576fdSSage Weil break; 7691d3576fdSSage Weil } 7701d3576fdSSage Weil if (!wbc->range_cyclic && page->index > end) { 7711d3576fdSSage Weil dout("end of range %p\n", page); 7721d3576fdSSage Weil done = 1; 7731d3576fdSSage Weil unlock_page(page); 7741d3576fdSSage Weil break; 7751d3576fdSSage Weil } 7761d3576fdSSage Weil if (next && (page->index != next)) { 7771d3576fdSSage Weil dout("not consecutive %p\n", page); 7781d3576fdSSage Weil unlock_page(page); 7791d3576fdSSage Weil break; 7801d3576fdSSage Weil } 7811d3576fdSSage Weil if (wbc->sync_mode != WB_SYNC_NONE) { 7821d3576fdSSage Weil dout("waiting on writeback %p\n", page); 7831d3576fdSSage Weil wait_on_page_writeback(page); 7841d3576fdSSage Weil } 7851d3576fdSSage Weil if ((snap_size && page_offset(page) > snap_size) || 7861d3576fdSSage Weil (!snap_size && 7871d3576fdSSage Weil page_offset(page) > i_size_read(inode))) { 7881d3576fdSSage Weil dout("%p page eof %llu\n", page, snap_size ? 7891d3576fdSSage Weil snap_size : i_size_read(inode)); 7901d3576fdSSage Weil done = 1; 7911d3576fdSSage Weil unlock_page(page); 7921d3576fdSSage Weil break; 7931d3576fdSSage Weil } 7941d3576fdSSage Weil if (PageWriteback(page)) { 7951d3576fdSSage Weil dout("%p under writeback\n", page); 7961d3576fdSSage Weil unlock_page(page); 7971d3576fdSSage Weil break; 7981d3576fdSSage Weil } 7991d3576fdSSage Weil 8001d3576fdSSage Weil /* only if matching snap context */ 80161600ef8SYan, Zheng pgsnapc = page_snap_context(page); 80280e755feSSage Weil if (pgsnapc->seq > snapc->seq) { 80380e755feSSage Weil dout("page snapc %p %lld > oldest %p %lld\n", 80480e755feSSage Weil pgsnapc, pgsnapc->seq, snapc, snapc->seq); 8051d3576fdSSage Weil unlock_page(page); 8061d3576fdSSage Weil if (!locked_pages) 8071d3576fdSSage Weil continue; /* keep looking for snap */ 8081d3576fdSSage Weil break; 8091d3576fdSSage Weil } 8101d3576fdSSage Weil 8111d3576fdSSage Weil if (!clear_page_dirty_for_io(page)) { 8121d3576fdSSage Weil dout("%p !clear_page_dirty_for_io\n", page); 8131d3576fdSSage Weil unlock_page(page); 8141d3576fdSSage Weil break; 8151d3576fdSSage Weil } 8161d3576fdSSage Weil 8171d3576fdSSage Weil /* ok */ 8181d3576fdSSage Weil if (locked_pages == 0) { 8191d3576fdSSage Weil /* prepare async write request */ 8206285bc23SAlex Elder offset = (u64) page_offset(page); 8211d3576fdSSage Weil len = wsize; 8223d14c5d2SYehuda Sadeh req = ceph_osdc_new_request(&fsc->client->osdc, 8231d3576fdSSage Weil &ci->i_layout, 8241d3576fdSSage Weil ceph_vino(inode), 8251d3576fdSSage Weil offset, &len, 8261d3576fdSSage Weil CEPH_OSD_OP_WRITE, 8271d3576fdSSage Weil CEPH_OSD_FLAG_WRITE | 8281d3576fdSSage Weil CEPH_OSD_FLAG_ONDISK, 8291d3576fdSSage Weil snapc, do_sync, 8301d3576fdSSage Weil ci->i_truncate_seq, 8311d3576fdSSage Weil ci->i_truncate_size, 832b7495fc2SSage Weil &inode->i_mtime, true, 1, 0); 8338c71897bSHenry C Chang 8346816282dSSage Weil if (IS_ERR(req)) { 8356816282dSSage Weil rc = PTR_ERR(req); 8368c71897bSHenry C Chang unlock_page(page); 8378c71897bSHenry C Chang break; 8388c71897bSHenry C Chang } 8398c71897bSHenry C Chang 8401d3576fdSSage Weil max_pages = req->r_num_pages; 8411d3576fdSSage Weil 8423d14c5d2SYehuda Sadeh alloc_page_vec(fsc, req); 8431d3576fdSSage Weil req->r_callback = writepages_finish; 8441d3576fdSSage Weil req->r_inode = inode; 8451d3576fdSSage Weil } 8461d3576fdSSage Weil 8471d3576fdSSage Weil /* note position of first page in pvec */ 8481d3576fdSSage Weil if (first < 0) 8491d3576fdSSage Weil first = i; 8501d3576fdSSage Weil dout("%p will write page %p idx %lu\n", 8511d3576fdSSage Weil inode, page, page->index); 8522baba250SYehuda Sadeh 853213c99eeSSage Weil writeback_stat = 8543d14c5d2SYehuda Sadeh atomic_long_inc_return(&fsc->writeback_count); 855213c99eeSSage Weil if (writeback_stat > CONGESTION_ON_THRESH( 8563d14c5d2SYehuda Sadeh fsc->mount_options->congestion_kb)) { 8573d14c5d2SYehuda Sadeh set_bdi_congested(&fsc->backing_dev_info, 858213c99eeSSage Weil BLK_RW_ASYNC); 8592baba250SYehuda Sadeh } 8602baba250SYehuda Sadeh 8611d3576fdSSage Weil set_page_writeback(page); 8621d3576fdSSage Weil req->r_pages[locked_pages] = page; 8631d3576fdSSage Weil locked_pages++; 8641d3576fdSSage Weil next = page->index + 1; 8651d3576fdSSage Weil } 8661d3576fdSSage Weil 8671d3576fdSSage Weil /* did we get anything? */ 8681d3576fdSSage Weil if (!locked_pages) 8691d3576fdSSage Weil goto release_pvec_pages; 8701d3576fdSSage Weil if (i) { 8711d3576fdSSage Weil int j; 8721d3576fdSSage Weil BUG_ON(!locked_pages || first < 0); 8731d3576fdSSage Weil 8741d3576fdSSage Weil if (pvec_pages && i == pvec_pages && 8751d3576fdSSage Weil locked_pages < max_pages) { 8761d3576fdSSage Weil dout("reached end pvec, trying for more\n"); 8771d3576fdSSage Weil pagevec_reinit(&pvec); 8781d3576fdSSage Weil goto get_more_pages; 8791d3576fdSSage Weil } 8801d3576fdSSage Weil 8811d3576fdSSage Weil /* shift unused pages over in the pvec... we 8821d3576fdSSage Weil * will need to release them below. */ 8831d3576fdSSage Weil for (j = i; j < pvec_pages; j++) { 8841d3576fdSSage Weil dout(" pvec leftover page %p\n", 8851d3576fdSSage Weil pvec.pages[j]); 8861d3576fdSSage Weil pvec.pages[j-i+first] = pvec.pages[j]; 8871d3576fdSSage Weil } 8881d3576fdSSage Weil pvec.nr -= i-first; 8891d3576fdSSage Weil } 8901d3576fdSSage Weil 8911d3576fdSSage Weil /* submit the write */ 8921d3576fdSSage Weil offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT; 8931d3576fdSSage Weil len = min((snap_size ? snap_size : i_size_read(inode)) - offset, 8941d3576fdSSage Weil (u64)locked_pages << PAGE_CACHE_SHIFT); 8951d3576fdSSage Weil dout("writepages got %d pages at %llu~%llu\n", 8961d3576fdSSage Weil locked_pages, offset, len); 8971d3576fdSSage Weil 8981d3576fdSSage Weil /* revise final length, page count */ 8991d3576fdSSage Weil req->r_num_pages = locked_pages; 9001d3576fdSSage Weil reqhead = req->r_request->front.iov_base; 9011d3576fdSSage Weil op = (void *)(reqhead + 1); 9021d3576fdSSage Weil op->extent.length = cpu_to_le64(len); 9031d3576fdSSage Weil op->payload_len = cpu_to_le32(len); 9041d3576fdSSage Weil req->r_request->hdr.data_len = cpu_to_le32(len); 9051d3576fdSSage Weil 9069d6fcb08SSage Weil rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); 9079d6fcb08SSage Weil BUG_ON(rc); 9081d3576fdSSage Weil req = NULL; 9091d3576fdSSage Weil 9101d3576fdSSage Weil /* continue? */ 9111d3576fdSSage Weil index = next; 9121d3576fdSSage Weil wbc->nr_to_write -= locked_pages; 9131d3576fdSSage Weil if (wbc->nr_to_write <= 0) 9141d3576fdSSage Weil done = 1; 9151d3576fdSSage Weil 9161d3576fdSSage Weil release_pvec_pages: 9171d3576fdSSage Weil dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 9181d3576fdSSage Weil pvec.nr ? pvec.pages[0] : NULL); 9191d3576fdSSage Weil pagevec_release(&pvec); 9201d3576fdSSage Weil 9211d3576fdSSage Weil if (locked_pages && !done) 9221d3576fdSSage Weil goto retry; 9231d3576fdSSage Weil } 9241d3576fdSSage Weil 9251d3576fdSSage Weil if (should_loop && !done) { 9261d3576fdSSage Weil /* more to do; loop back to beginning of file */ 9271d3576fdSSage Weil dout("writepages looping back to beginning of file\n"); 9281d3576fdSSage Weil should_loop = 0; 9291d3576fdSSage Weil index = 0; 9301d3576fdSSage Weil goto retry; 9311d3576fdSSage Weil } 9321d3576fdSSage Weil 9331d3576fdSSage Weil if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 9341d3576fdSSage Weil mapping->writeback_index = index; 9351d3576fdSSage Weil 9361d3576fdSSage Weil out: 9371d3576fdSSage Weil if (req) 9381d3576fdSSage Weil ceph_osdc_put_request(req); 9391d3576fdSSage Weil ceph_put_snap_context(snapc); 9401d3576fdSSage Weil dout("writepages done, rc = %d\n", rc); 9411d3576fdSSage Weil return rc; 9421d3576fdSSage Weil } 9431d3576fdSSage Weil 9441d3576fdSSage Weil 9451d3576fdSSage Weil 9461d3576fdSSage Weil /* 9471d3576fdSSage Weil * See if a given @snapc is either writeable, or already written. 9481d3576fdSSage Weil */ 9491d3576fdSSage Weil static int context_is_writeable_or_written(struct inode *inode, 9501d3576fdSSage Weil struct ceph_snap_context *snapc) 9511d3576fdSSage Weil { 9521d3576fdSSage Weil struct ceph_snap_context *oldest = get_oldest_context(inode, NULL); 9536298a337SSage Weil int ret = !oldest || snapc->seq <= oldest->seq; 9546298a337SSage Weil 9556298a337SSage Weil ceph_put_snap_context(oldest); 9566298a337SSage Weil return ret; 9571d3576fdSSage Weil } 9581d3576fdSSage Weil 9591d3576fdSSage Weil /* 9601d3576fdSSage Weil * We are only allowed to write into/dirty the page if the page is 9611d3576fdSSage Weil * clean, or already dirty within the same snap context. 9628f883c24SSage Weil * 9638f883c24SSage Weil * called with page locked. 9648f883c24SSage Weil * return success with page locked, 9658f883c24SSage Weil * or any failure (incl -EAGAIN) with page unlocked. 9661d3576fdSSage Weil */ 9674af6b225SYehuda Sadeh static int ceph_update_writeable_page(struct file *file, 9684af6b225SYehuda Sadeh loff_t pos, unsigned len, 9694af6b225SYehuda Sadeh struct page *page) 9701d3576fdSSage Weil { 9711d3576fdSSage Weil struct inode *inode = file->f_dentry->d_inode; 9721d3576fdSSage Weil struct ceph_inode_info *ci = ceph_inode(inode); 9733d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 9741d3576fdSSage Weil loff_t page_off = pos & PAGE_CACHE_MASK; 9751d3576fdSSage Weil int pos_in_page = pos & ~PAGE_CACHE_MASK; 9761d3576fdSSage Weil int end_in_page = pos_in_page + len; 9771d3576fdSSage Weil loff_t i_size; 9781d3576fdSSage Weil int r; 97980e755feSSage Weil struct ceph_snap_context *snapc, *oldest; 9801d3576fdSSage Weil 9811d3576fdSSage Weil retry_locked: 9821d3576fdSSage Weil /* writepages currently holds page lock, but if we change that later, */ 9831d3576fdSSage Weil wait_on_page_writeback(page); 9841d3576fdSSage Weil 9851d3576fdSSage Weil /* check snap context */ 9861d3576fdSSage Weil BUG_ON(!ci->i_snap_realm); 9871d3576fdSSage Weil down_read(&mdsc->snap_rwsem); 9881d3576fdSSage Weil BUG_ON(!ci->i_snap_realm->cached_context); 98961600ef8SYan, Zheng snapc = page_snap_context(page); 99080e755feSSage Weil if (snapc && snapc != ci->i_head_snapc) { 9911d3576fdSSage Weil /* 9921d3576fdSSage Weil * this page is already dirty in another (older) snap 9931d3576fdSSage Weil * context! is it writeable now? 9941d3576fdSSage Weil */ 99580e755feSSage Weil oldest = get_oldest_context(inode, NULL); 9961d3576fdSSage Weil up_read(&mdsc->snap_rwsem); 9971d3576fdSSage Weil 99880e755feSSage Weil if (snapc->seq > oldest->seq) { 9996298a337SSage Weil ceph_put_snap_context(oldest); 10001d3576fdSSage Weil dout(" page %p snapc %p not current or oldest\n", 10016298a337SSage Weil page, snapc); 10021d3576fdSSage Weil /* 10031d3576fdSSage Weil * queue for writeback, and wait for snapc to 10041d3576fdSSage Weil * be writeable or written 10051d3576fdSSage Weil */ 10066298a337SSage Weil snapc = ceph_get_snap_context(snapc); 10071d3576fdSSage Weil unlock_page(page); 10083c6f6b79SSage Weil ceph_queue_writeback(inode); 10098f883c24SSage Weil r = wait_event_interruptible(ci->i_cap_wq, 10101d3576fdSSage Weil context_is_writeable_or_written(inode, snapc)); 10111d3576fdSSage Weil ceph_put_snap_context(snapc); 10128f883c24SSage Weil if (r == -ERESTARTSYS) 10138f883c24SSage Weil return r; 10144af6b225SYehuda Sadeh return -EAGAIN; 10151d3576fdSSage Weil } 10166298a337SSage Weil ceph_put_snap_context(oldest); 10171d3576fdSSage Weil 10181d3576fdSSage Weil /* yay, writeable, do it now (without dropping page lock) */ 10191d3576fdSSage Weil dout(" page %p snapc %p not current, but oldest\n", 10201d3576fdSSage Weil page, snapc); 10211d3576fdSSage Weil if (!clear_page_dirty_for_io(page)) 10221d3576fdSSage Weil goto retry_locked; 10231d3576fdSSage Weil r = writepage_nounlock(page, NULL); 10241d3576fdSSage Weil if (r < 0) 10251d3576fdSSage Weil goto fail_nosnap; 10261d3576fdSSage Weil goto retry_locked; 10271d3576fdSSage Weil } 10281d3576fdSSage Weil 10291d3576fdSSage Weil if (PageUptodate(page)) { 10301d3576fdSSage Weil dout(" page %p already uptodate\n", page); 10311d3576fdSSage Weil return 0; 10321d3576fdSSage Weil } 10331d3576fdSSage Weil 10341d3576fdSSage Weil /* full page? */ 10351d3576fdSSage Weil if (pos_in_page == 0 && len == PAGE_CACHE_SIZE) 10361d3576fdSSage Weil return 0; 10371d3576fdSSage Weil 10381d3576fdSSage Weil /* past end of file? */ 10391d3576fdSSage Weil i_size = inode->i_size; /* caller holds i_mutex */ 10401d3576fdSSage Weil 10411d3576fdSSage Weil if (i_size + len > inode->i_sb->s_maxbytes) { 10421d3576fdSSage Weil /* file is too big */ 10431d3576fdSSage Weil r = -EINVAL; 10441d3576fdSSage Weil goto fail; 10451d3576fdSSage Weil } 10461d3576fdSSage Weil 10471d3576fdSSage Weil if (page_off >= i_size || 10481d3576fdSSage Weil (pos_in_page == 0 && (pos+len) >= i_size && 10491d3576fdSSage Weil end_in_page - pos_in_page != PAGE_CACHE_SIZE)) { 10501d3576fdSSage Weil dout(" zeroing %p 0 - %d and %d - %d\n", 10511d3576fdSSage Weil page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE); 10521d3576fdSSage Weil zero_user_segments(page, 10531d3576fdSSage Weil 0, pos_in_page, 10541d3576fdSSage Weil end_in_page, PAGE_CACHE_SIZE); 10551d3576fdSSage Weil return 0; 10561d3576fdSSage Weil } 10571d3576fdSSage Weil 10581d3576fdSSage Weil /* we need to read it. */ 10591d3576fdSSage Weil up_read(&mdsc->snap_rwsem); 10601d3576fdSSage Weil r = readpage_nounlock(file, page); 10611d3576fdSSage Weil if (r < 0) 10621d3576fdSSage Weil goto fail_nosnap; 10631d3576fdSSage Weil goto retry_locked; 10641d3576fdSSage Weil 10651d3576fdSSage Weil fail: 10661d3576fdSSage Weil up_read(&mdsc->snap_rwsem); 10671d3576fdSSage Weil fail_nosnap: 10681d3576fdSSage Weil unlock_page(page); 10691d3576fdSSage Weil return r; 10701d3576fdSSage Weil } 10711d3576fdSSage Weil 10721d3576fdSSage Weil /* 10734af6b225SYehuda Sadeh * We are only allowed to write into/dirty the page if the page is 10744af6b225SYehuda Sadeh * clean, or already dirty within the same snap context. 10754af6b225SYehuda Sadeh */ 10764af6b225SYehuda Sadeh static int ceph_write_begin(struct file *file, struct address_space *mapping, 10774af6b225SYehuda Sadeh loff_t pos, unsigned len, unsigned flags, 10784af6b225SYehuda Sadeh struct page **pagep, void **fsdata) 10794af6b225SYehuda Sadeh { 10804af6b225SYehuda Sadeh struct inode *inode = file->f_dentry->d_inode; 10814af6b225SYehuda Sadeh struct page *page; 10824af6b225SYehuda Sadeh pgoff_t index = pos >> PAGE_CACHE_SHIFT; 10834af6b225SYehuda Sadeh int r; 10844af6b225SYehuda Sadeh 10854af6b225SYehuda Sadeh do { 10864af6b225SYehuda Sadeh /* get a page */ 10874af6b225SYehuda Sadeh page = grab_cache_page_write_begin(mapping, index, 0); 10884af6b225SYehuda Sadeh if (!page) 10894af6b225SYehuda Sadeh return -ENOMEM; 10904af6b225SYehuda Sadeh *pagep = page; 10914af6b225SYehuda Sadeh 10924af6b225SYehuda Sadeh dout("write_begin file %p inode %p page %p %d~%d\n", file, 10934af6b225SYehuda Sadeh inode, page, (int)pos, (int)len); 10944af6b225SYehuda Sadeh 10954af6b225SYehuda Sadeh r = ceph_update_writeable_page(file, pos, len, page); 10964af6b225SYehuda Sadeh } while (r == -EAGAIN); 10974af6b225SYehuda Sadeh 10984af6b225SYehuda Sadeh return r; 10994af6b225SYehuda Sadeh } 11004af6b225SYehuda Sadeh 11014af6b225SYehuda Sadeh /* 11021d3576fdSSage Weil * we don't do anything in here that simple_write_end doesn't do 11031d3576fdSSage Weil * except adjust dirty page accounting and drop read lock on 11041d3576fdSSage Weil * mdsc->snap_rwsem. 11051d3576fdSSage Weil */ 11061d3576fdSSage Weil static int ceph_write_end(struct file *file, struct address_space *mapping, 11071d3576fdSSage Weil loff_t pos, unsigned len, unsigned copied, 11081d3576fdSSage Weil struct page *page, void *fsdata) 11091d3576fdSSage Weil { 11101d3576fdSSage Weil struct inode *inode = file->f_dentry->d_inode; 11113d14c5d2SYehuda Sadeh struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 11123d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = fsc->mdsc; 11131d3576fdSSage Weil unsigned from = pos & (PAGE_CACHE_SIZE - 1); 11141d3576fdSSage Weil int check_cap = 0; 11151d3576fdSSage Weil 11161d3576fdSSage Weil dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, 11171d3576fdSSage Weil inode, page, (int)pos, (int)copied, (int)len); 11181d3576fdSSage Weil 11191d3576fdSSage Weil /* zero the stale part of the page if we did a short copy */ 11201d3576fdSSage Weil if (copied < len) 11211d3576fdSSage Weil zero_user_segment(page, from+copied, len); 11221d3576fdSSage Weil 11231d3576fdSSage Weil /* did file size increase? */ 11241d3576fdSSage Weil /* (no need for i_size_read(); we caller holds i_mutex */ 11251d3576fdSSage Weil if (pos+copied > inode->i_size) 11261d3576fdSSage Weil check_cap = ceph_inode_set_size(inode, pos+copied); 11271d3576fdSSage Weil 11281d3576fdSSage Weil if (!PageUptodate(page)) 11291d3576fdSSage Weil SetPageUptodate(page); 11301d3576fdSSage Weil 11311d3576fdSSage Weil set_page_dirty(page); 11321d3576fdSSage Weil 11331d3576fdSSage Weil unlock_page(page); 11341d3576fdSSage Weil up_read(&mdsc->snap_rwsem); 11351d3576fdSSage Weil page_cache_release(page); 11361d3576fdSSage Weil 11371d3576fdSSage Weil if (check_cap) 11381d3576fdSSage Weil ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); 11391d3576fdSSage Weil 11401d3576fdSSage Weil return copied; 11411d3576fdSSage Weil } 11421d3576fdSSage Weil 11431d3576fdSSage Weil /* 11441d3576fdSSage Weil * we set .direct_IO to indicate direct io is supported, but since we 11451d3576fdSSage Weil * intercept O_DIRECT reads and writes early, this function should 11461d3576fdSSage Weil * never get called. 11471d3576fdSSage Weil */ 11481d3576fdSSage Weil static ssize_t ceph_direct_io(int rw, struct kiocb *iocb, 11491d3576fdSSage Weil const struct iovec *iov, 11501d3576fdSSage Weil loff_t pos, unsigned long nr_segs) 11511d3576fdSSage Weil { 11521d3576fdSSage Weil WARN_ON(1); 11531d3576fdSSage Weil return -EINVAL; 11541d3576fdSSage Weil } 11551d3576fdSSage Weil 11561d3576fdSSage Weil const struct address_space_operations ceph_aops = { 11571d3576fdSSage Weil .readpage = ceph_readpage, 11581d3576fdSSage Weil .readpages = ceph_readpages, 11591d3576fdSSage Weil .writepage = ceph_writepage, 11601d3576fdSSage Weil .writepages = ceph_writepages_start, 11611d3576fdSSage Weil .write_begin = ceph_write_begin, 11621d3576fdSSage Weil .write_end = ceph_write_end, 11631d3576fdSSage Weil .set_page_dirty = ceph_set_page_dirty, 11641d3576fdSSage Weil .invalidatepage = ceph_invalidatepage, 11651d3576fdSSage Weil .releasepage = ceph_releasepage, 11661d3576fdSSage Weil .direct_IO = ceph_direct_io, 11671d3576fdSSage Weil }; 11681d3576fdSSage Weil 11691d3576fdSSage Weil 11701d3576fdSSage Weil /* 11711d3576fdSSage Weil * vm ops 11721d3576fdSSage Weil */ 11731d3576fdSSage Weil 11741d3576fdSSage Weil /* 11751d3576fdSSage Weil * Reuse write_begin here for simplicity. 11761d3576fdSSage Weil */ 11771d3576fdSSage Weil static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 11781d3576fdSSage Weil { 11791d3576fdSSage Weil struct inode *inode = vma->vm_file->f_dentry->d_inode; 11801d3576fdSSage Weil struct page *page = vmf->page; 11813d14c5d2SYehuda Sadeh struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 11826285bc23SAlex Elder loff_t off = page_offset(page); 11831d3576fdSSage Weil loff_t size, len; 11841d3576fdSSage Weil int ret; 11851d3576fdSSage Weil 11863ca9c3bdSJan Kara /* Update time before taking page lock */ 11873ca9c3bdSJan Kara file_update_time(vma->vm_file); 11883ca9c3bdSJan Kara 11891d3576fdSSage Weil size = i_size_read(inode); 11901d3576fdSSage Weil if (off + PAGE_CACHE_SIZE <= size) 11911d3576fdSSage Weil len = PAGE_CACHE_SIZE; 11921d3576fdSSage Weil else 11931d3576fdSSage Weil len = size & ~PAGE_CACHE_MASK; 11941d3576fdSSage Weil 11951d3576fdSSage Weil dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode, 11961d3576fdSSage Weil off, len, page, page->index); 11974af6b225SYehuda Sadeh 11984af6b225SYehuda Sadeh lock_page(page); 11994af6b225SYehuda Sadeh 12004af6b225SYehuda Sadeh ret = VM_FAULT_NOPAGE; 12014af6b225SYehuda Sadeh if ((off > size) || 12024af6b225SYehuda Sadeh (page->mapping != inode->i_mapping)) 12034af6b225SYehuda Sadeh goto out; 12044af6b225SYehuda Sadeh 12054af6b225SYehuda Sadeh ret = ceph_update_writeable_page(vma->vm_file, off, len, page); 12064af6b225SYehuda Sadeh if (ret == 0) { 12074af6b225SYehuda Sadeh /* success. we'll keep the page locked. */ 12081d3576fdSSage Weil set_page_dirty(page); 12091d3576fdSSage Weil up_read(&mdsc->snap_rwsem); 12101d3576fdSSage Weil ret = VM_FAULT_LOCKED; 12111d3576fdSSage Weil } else { 12124af6b225SYehuda Sadeh if (ret == -ENOMEM) 12134af6b225SYehuda Sadeh ret = VM_FAULT_OOM; 12144af6b225SYehuda Sadeh else 12151d3576fdSSage Weil ret = VM_FAULT_SIGBUS; 12161d3576fdSSage Weil } 12174af6b225SYehuda Sadeh out: 12181d3576fdSSage Weil dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret); 12194af6b225SYehuda Sadeh if (ret != VM_FAULT_LOCKED) 12204af6b225SYehuda Sadeh unlock_page(page); 12211d3576fdSSage Weil return ret; 12221d3576fdSSage Weil } 12231d3576fdSSage Weil 12241d3576fdSSage Weil static struct vm_operations_struct ceph_vmops = { 12251d3576fdSSage Weil .fault = filemap_fault, 12261d3576fdSSage Weil .page_mkwrite = ceph_page_mkwrite, 12270b173bc4SKonstantin Khlebnikov .remap_pages = generic_file_remap_pages, 12281d3576fdSSage Weil }; 12291d3576fdSSage Weil 12301d3576fdSSage Weil int ceph_mmap(struct file *file, struct vm_area_struct *vma) 12311d3576fdSSage Weil { 12321d3576fdSSage Weil struct address_space *mapping = file->f_mapping; 12331d3576fdSSage Weil 12341d3576fdSSage Weil if (!mapping->a_ops->readpage) 12351d3576fdSSage Weil return -ENOEXEC; 12361d3576fdSSage Weil file_accessed(file); 12371d3576fdSSage Weil vma->vm_ops = &ceph_vmops; 12381d3576fdSSage Weil return 0; 12391d3576fdSSage Weil } 1240