blob: 4a3f55f27ab4d524fc1950935494cee94fe6e8e2 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil1d3576f2009-10-06 11:31:09 -07002
3#include <linux/backing-dev.h>
4#include <linux/fs.h>
5#include <linux/mm.h>
6#include <linux/pagemap.h>
7#include <linux/writeback.h> /* generic_writepages */
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09008#include <linux/slab.h>
Sage Weil1d3576f2009-10-06 11:31:09 -07009#include <linux/pagevec.h>
10#include <linux/task_io_accounting_ops.h>
11
12#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070013#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040014#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070015#include <linux/ceph/osd_client.h>
Sage Weil1d3576f2009-10-06 11:31:09 -070016
17/*
18 * Ceph address space ops.
19 *
20 * There are a few funny things going on here.
21 *
22 * The page->private field is used to reference a struct
23 * ceph_snap_context for _every_ dirty page. This indicates which
24 * snapshot the page was logically dirtied in, and thus which snap
25 * context needs to be associated with the osd write during writeback.
26 *
27 * Similarly, struct ceph_inode_info maintains a set of counters to
Lucas De Marchi25985ed2011-03-30 22:57:33 -030028 * count dirty pages on the inode. In the absence of snapshots,
Sage Weil1d3576f2009-10-06 11:31:09 -070029 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
30 *
31 * When a snapshot is taken (that is, when the client receives
32 * notification that a snapshot was taken), each inode with caps and
33 * with dirty pages (dirty pages implies there is a cap) gets a new
34 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
35 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
36 * moved to capsnap->dirty. (Unless a sync write is currently in
37 * progress. In that case, the capsnap is said to be "pending", new
38 * writes cannot start, and the capsnap isn't "finalized" until the
39 * write completes (or fails) and a final size/mtime for the inode for
40 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
41 *
42 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
43 * we look for the first capsnap in i_cap_snaps and write out pages in
44 * that snap context _only_. Then we move on to the next capsnap,
45 * eventually reaching the "live" or "head" context (i.e., pages that
46 * are not yet snapped) and are writing the most recently dirtied
47 * pages.
48 *
49 * Invalidate and so forth must take care to ensure the dirty page
50 * accounting is preserved.
51 */
52
Yehuda Sadeh2baba252009-12-18 13:51:57 -080053#define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
54#define CONGESTION_OFF_THRESH(congestion_kb) \
55 (CONGESTION_ON_THRESH(congestion_kb) - \
56 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
57
Yan, Zheng61600ef2012-05-28 14:44:30 +080058static inline struct ceph_snap_context *page_snap_context(struct page *page)
59{
60 if (PagePrivate(page))
61 return (void *)page->private;
62 return NULL;
63}
Sage Weil1d3576f2009-10-06 11:31:09 -070064
65/*
66 * Dirty a page. Optimistically adjust accounting, on the assumption
67 * that we won't race with invalidate. If we do, readjust.
68 */
69static int ceph_set_page_dirty(struct page *page)
70{
71 struct address_space *mapping = page->mapping;
72 struct inode *inode;
73 struct ceph_inode_info *ci;
Sage Weil1d3576f2009-10-06 11:31:09 -070074 struct ceph_snap_context *snapc;
Sha Zhengju7d6e1f52013-08-21 16:27:34 +080075 int ret;
Sage Weil1d3576f2009-10-06 11:31:09 -070076
77 if (unlikely(!mapping))
78 return !TestSetPageDirty(page);
79
Sha Zhengju7d6e1f52013-08-21 16:27:34 +080080 if (PageDirty(page)) {
Sage Weil1d3576f2009-10-06 11:31:09 -070081 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
82 mapping->host, page, page->index);
Sha Zhengju7d6e1f52013-08-21 16:27:34 +080083 BUG_ON(!PagePrivate(page));
Sage Weil1d3576f2009-10-06 11:31:09 -070084 return 0;
85 }
86
87 inode = mapping->host;
88 ci = ceph_inode(inode);
89
90 /*
91 * Note that we're grabbing a snapc ref here without holding
92 * any locks!
93 */
94 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
95
96 /* dirty the head */
Sage Weilbe655592011-11-30 09:47:09 -080097 spin_lock(&ci->i_ceph_lock);
Sage Weil7d8cb262010-08-24 08:44:16 -070098 if (ci->i_head_snapc == NULL)
Sage Weil1d3576f2009-10-06 11:31:09 -070099 ci->i_head_snapc = ceph_get_snap_context(snapc);
100 ++ci->i_wrbuffer_ref_head;
101 if (ci->i_wrbuffer_ref == 0)
Dave Chinner0444d762011-03-29 18:08:50 +1100102 ihold(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700103 ++ci->i_wrbuffer_ref;
104 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
105 "snapc %p seq %lld (%d snaps)\n",
106 mapping->host, page, page->index,
107 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
108 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
109 snapc, snapc->seq, snapc->num_snaps);
Sage Weilbe655592011-11-30 09:47:09 -0800110 spin_unlock(&ci->i_ceph_lock);
Sage Weil1d3576f2009-10-06 11:31:09 -0700111
Sha Zhengju7d6e1f52013-08-21 16:27:34 +0800112 /*
113 * Reference snap context in page->private. Also set
114 * PagePrivate so that we get invalidatepage callback.
115 */
116 BUG_ON(PagePrivate(page));
117 page->private = (unsigned long)snapc;
118 SetPagePrivate(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700119
Sha Zhengju7d6e1f52013-08-21 16:27:34 +0800120 ret = __set_page_dirty_nobuffers(page);
121 WARN_ON(!PageLocked(page));
122 WARN_ON(!page->mapping);
Sage Weil1d3576f2009-10-06 11:31:09 -0700123
Sha Zhengju7d6e1f52013-08-21 16:27:34 +0800124 return ret;
Sage Weil1d3576f2009-10-06 11:31:09 -0700125}
126
127/*
128 * If we are truncating the full page (i.e. offset == 0), adjust the
129 * dirty page counters appropriately. Only called if there is private
130 * data on the page.
131 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400132static void ceph_invalidatepage(struct page *page, unsigned int offset,
133 unsigned int length)
Sage Weil1d3576f2009-10-06 11:31:09 -0700134{
Alexander Beregalov4ce1e9a2010-02-22 17:17:44 +0300135 struct inode *inode;
Sage Weil1d3576f2009-10-06 11:31:09 -0700136 struct ceph_inode_info *ci;
Yan, Zheng61600ef2012-05-28 14:44:30 +0800137 struct ceph_snap_context *snapc = page_snap_context(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700138
Alexander Beregalov4ce1e9a2010-02-22 17:17:44 +0300139 inode = page->mapping->host;
Milosz Tanskib150f5c12013-08-09 12:59:55 -0400140 ci = ceph_inode(inode);
141
142 if (offset != 0 || length != PAGE_CACHE_SIZE) {
143 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
144 inode, page, page->index, offset, length);
145 return;
146 }
Alexander Beregalov4ce1e9a2010-02-22 17:17:44 +0300147
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400148 ceph_invalidate_fscache_page(inode, page);
149
150 if (!PagePrivate(page))
151 return;
152
Sage Weil1d3576f2009-10-06 11:31:09 -0700153 /*
154 * We can get non-dirty pages here due to races between
155 * set_page_dirty and truncate_complete_page; just spit out a
156 * warning, in case we end up with accounting problems later.
157 */
158 if (!PageDirty(page))
159 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
160
Milosz Tanskib150f5c12013-08-09 12:59:55 -0400161 ClearPageChecked(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700162
Milosz Tanskib150f5c12013-08-09 12:59:55 -0400163 dout("%p invalidatepage %p idx %lu full dirty page\n",
164 inode, page, page->index);
165
166 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
167 ceph_put_snap_context(snapc);
168 page->private = 0;
169 ClearPagePrivate(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700170}
171
Sage Weil1d3576f2009-10-06 11:31:09 -0700172static int ceph_releasepage(struct page *page, gfp_t g)
173{
174 struct inode *inode = page->mapping ? page->mapping->host : NULL;
175 dout("%p releasepage %p idx %lu\n", inode, page, page->index);
176 WARN_ON(PageDirty(page));
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400177
178 /* Can we release the page from the cache? */
179 if (!ceph_release_fscache_page(page, g))
180 return 0;
181
182 return !PagePrivate(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700183}
184
185/*
186 * read a single page, without unlocking it.
187 */
188static int readpage_nounlock(struct file *filp, struct page *page)
189{
Al Viro496ad9a2013-01-23 17:07:38 -0500190 struct inode *inode = file_inode(filp);
Sage Weil1d3576f2009-10-06 11:31:09 -0700191 struct ceph_inode_info *ci = ceph_inode(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400192 struct ceph_osd_client *osdc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700193 &ceph_inode_to_client(inode)->client->osdc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700194 int err = 0;
195 u64 len = PAGE_CACHE_SIZE;
196
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400197 err = ceph_readpage_from_fscache(inode, page);
198
199 if (err == 0)
200 goto out;
201
Sage Weil1d3576f2009-10-06 11:31:09 -0700202 dout("readpage inode %p file %p page %p index %lu\n",
203 inode, filp, page, page->index);
204 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
Alex Elder6285bc22012-10-02 10:25:51 -0500205 (u64) page_offset(page), &len,
Sage Weil1d3576f2009-10-06 11:31:09 -0700206 ci->i_truncate_seq, ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800207 &page, 1, 0);
Sage Weil1d3576f2009-10-06 11:31:09 -0700208 if (err == -ENOENT)
209 err = 0;
210 if (err < 0) {
211 SetPageError(page);
Li Wang18302802013-12-19 06:03:49 -0800212 ceph_fscache_readpage_cancel(inode, page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700213 goto out;
Sage Weil1d3576f2009-10-06 11:31:09 -0700214 }
Zhang Zhen23cd5732014-05-28 15:09:55 +0800215 if (err < PAGE_CACHE_SIZE)
216 /* zero fill remainder of page */
217 zero_user_segment(page, err, PAGE_CACHE_SIZE);
218 else
219 flush_dcache_page(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700220
Zhang Zhen23cd5732014-05-28 15:09:55 +0800221 SetPageUptodate(page);
222 ceph_readpage_to_fscache(inode, page);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400223
Sage Weil1d3576f2009-10-06 11:31:09 -0700224out:
225 return err < 0 ? err : 0;
226}
227
228static int ceph_readpage(struct file *filp, struct page *page)
229{
230 int r = readpage_nounlock(filp, page);
231 unlock_page(page);
232 return r;
233}
234
235/*
Sage Weil7c272192011-08-03 09:58:09 -0700236 * Finish an async read(ahead) op.
Sage Weil1d3576f2009-10-06 11:31:09 -0700237 */
Sage Weil7c272192011-08-03 09:58:09 -0700238static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg)
Sage Weil1d3576f2009-10-06 11:31:09 -0700239{
Sage Weil7c272192011-08-03 09:58:09 -0700240 struct inode *inode = req->r_inode;
Alex Elder87060c12013-04-03 01:28:58 -0500241 struct ceph_osd_data *osd_data;
Sage Weil1b83bef2013-02-25 16:11:12 -0800242 int rc = req->r_result;
243 int bytes = le32_to_cpu(msg->hdr.data_len);
Alex Eldere0c59482013-03-07 15:38:25 -0600244 int num_pages;
Sage Weil7c272192011-08-03 09:58:09 -0700245 int i;
246
Sage Weil7c272192011-08-03 09:58:09 -0700247 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
248
249 /* unlock all pages, zeroing any data we didn't read */
Alex Elder406e2c92013-04-15 14:50:36 -0500250 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder87060c12013-04-03 01:28:58 -0500251 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
252 num_pages = calc_pages_for((u64)osd_data->alignment,
253 (u64)osd_data->length);
Alex Eldere0c59482013-03-07 15:38:25 -0600254 for (i = 0; i < num_pages; i++) {
Alex Elder87060c12013-04-03 01:28:58 -0500255 struct page *page = osd_data->pages[i];
Sage Weil7c272192011-08-03 09:58:09 -0700256
Li Wangf36132a2013-11-27 22:28:13 +0800257 if (rc < 0)
258 goto unlock;
Sage Weil7c272192011-08-03 09:58:09 -0700259 if (bytes < (int)PAGE_CACHE_SIZE) {
260 /* zero (remainder of) page */
261 int s = bytes < 0 ? 0 : bytes;
262 zero_user_segment(page, s, PAGE_CACHE_SIZE);
263 }
264 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
265 page->index);
266 flush_dcache_page(page);
267 SetPageUptodate(page);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400268 ceph_readpage_to_fscache(inode, page);
Li Wangf36132a2013-11-27 22:28:13 +0800269unlock:
Sage Weil7c272192011-08-03 09:58:09 -0700270 unlock_page(page);
271 page_cache_release(page);
Alex Elder0fff87e2013-02-14 12:16:43 -0600272 bytes -= PAGE_CACHE_SIZE;
Sage Weil7c272192011-08-03 09:58:09 -0700273 }
Alex Elder87060c12013-04-03 01:28:58 -0500274 kfree(osd_data->pages);
Sage Weil7c272192011-08-03 09:58:09 -0700275}
276
David Zafman8884d532012-12-03 19:14:05 -0800277static void ceph_unlock_page_vector(struct page **pages, int num_pages)
278{
279 int i;
280
281 for (i = 0; i < num_pages; i++)
282 unlock_page(pages[i]);
283}
284
Sage Weil7c272192011-08-03 09:58:09 -0700285/*
286 * start an async read(ahead) operation. return nr_pages we submitted
287 * a read for on success, or negative error code.
288 */
Sage Weil0d66a482011-08-04 08:21:30 -0700289static int start_read(struct inode *inode, struct list_head *page_list, int max)
Sage Weil7c272192011-08-03 09:58:09 -0700290{
291 struct ceph_osd_client *osdc =
292 &ceph_inode_to_client(inode)->client->osdc;
293 struct ceph_inode_info *ci = ceph_inode(inode);
294 struct page *page = list_entry(page_list->prev, struct page, lru);
Alex Elderacead002013-03-14 14:09:05 -0500295 struct ceph_vino vino;
Sage Weil7c272192011-08-03 09:58:09 -0700296 struct ceph_osd_request *req;
297 u64 off;
298 u64 len;
299 int i;
Sage Weil1d3576f2009-10-06 11:31:09 -0700300 struct page **pages;
Sage Weil7c272192011-08-03 09:58:09 -0700301 pgoff_t next_index;
302 int nr_pages = 0;
303 int ret;
304
Alex Elder6285bc22012-10-02 10:25:51 -0500305 off = (u64) page_offset(page);
Sage Weil7c272192011-08-03 09:58:09 -0700306
307 /* count pages */
308 next_index = page->index;
309 list_for_each_entry_reverse(page, page_list, lru) {
310 if (page->index != next_index)
311 break;
312 nr_pages++;
313 next_index++;
Sage Weil0d66a482011-08-04 08:21:30 -0700314 if (max && nr_pages == max)
315 break;
Sage Weil7c272192011-08-03 09:58:09 -0700316 }
317 len = nr_pages << PAGE_CACHE_SHIFT;
318 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
319 off, len);
Alex Elderacead002013-03-14 14:09:05 -0500320 vino = ceph_vino(inode);
321 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800322 0, 1, CEPH_OSD_OP_READ,
Alex Elderacead002013-03-14 14:09:05 -0500323 CEPH_OSD_FLAG_READ, NULL,
Sage Weil7c272192011-08-03 09:58:09 -0700324 ci->i_truncate_seq, ci->i_truncate_size,
Alex Elderacead002013-03-14 14:09:05 -0500325 false);
Sage Weil68162822012-09-24 21:01:02 -0700326 if (IS_ERR(req))
327 return PTR_ERR(req);
Sage Weil1d3576f2009-10-06 11:31:09 -0700328
329 /* build page vector */
Alex Eldercf7b7e12013-03-01 18:00:15 -0600330 nr_pages = calc_pages_for(0, len);
Sage Weil7c272192011-08-03 09:58:09 -0700331 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
332 ret = -ENOMEM;
Sage Weil1d3576f2009-10-06 11:31:09 -0700333 if (!pages)
Sage Weil7c272192011-08-03 09:58:09 -0700334 goto out;
335 for (i = 0; i < nr_pages; ++i) {
336 page = list_entry(page_list->prev, struct page, lru);
337 BUG_ON(PageLocked(page));
338 list_del(&page->lru);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400339
Sage Weil7c272192011-08-03 09:58:09 -0700340 dout("start_read %p adding %p idx %lu\n", inode, page,
341 page->index);
342 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
343 GFP_NOFS)) {
Milosz Tanskid4d3aa32013-09-03 19:11:17 -0400344 ceph_fscache_uncache_page(inode, page);
Sage Weil7c272192011-08-03 09:58:09 -0700345 page_cache_release(page);
346 dout("start_read %p add_to_page_cache failed %p\n",
347 inode, page);
348 nr_pages = i;
349 goto out_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700350 }
Sage Weil7c272192011-08-03 09:58:09 -0700351 pages[i] = page;
Sage Weil1d3576f2009-10-06 11:31:09 -0700352 }
Alex Elder406e2c92013-04-15 14:50:36 -0500353 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
Sage Weil7c272192011-08-03 09:58:09 -0700354 req->r_callback = finish_read;
355 req->r_inode = inode;
356
Alex Elder79528732013-04-03 21:32:51 -0500357 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -0500358
Sage Weil7c272192011-08-03 09:58:09 -0700359 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
360 ret = ceph_osdc_start_request(osdc, req, false);
361 if (ret < 0)
362 goto out_pages;
363 ceph_osdc_put_request(req);
364 return nr_pages;
365
366out_pages:
David Zafman8884d532012-12-03 19:14:05 -0800367 ceph_unlock_page_vector(pages, nr_pages);
Sage Weil7c272192011-08-03 09:58:09 -0700368 ceph_release_page_vector(pages, nr_pages);
Sage Weil7c272192011-08-03 09:58:09 -0700369out:
370 ceph_osdc_put_request(req);
371 return ret;
Sage Weil1d3576f2009-10-06 11:31:09 -0700372}
373
Sage Weil7c272192011-08-03 09:58:09 -0700374
Sage Weil1d3576f2009-10-06 11:31:09 -0700375/*
376 * Read multiple pages. Leave pages we don't read + unlock in page_list;
377 * the caller (VM) cleans them up.
378 */
379static int ceph_readpages(struct file *file, struct address_space *mapping,
380 struct list_head *page_list, unsigned nr_pages)
381{
Al Viro496ad9a2013-01-23 17:07:38 -0500382 struct inode *inode = file_inode(file);
Sage Weil0d66a482011-08-04 08:21:30 -0700383 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700384 int rc = 0;
Sage Weil0d66a482011-08-04 08:21:30 -0700385 int max = 0;
Sage Weil1d3576f2009-10-06 11:31:09 -0700386
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400387 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
388 &nr_pages);
389
390 if (rc == 0)
391 goto out;
392
Sage Weil0d66a482011-08-04 08:21:30 -0700393 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
394 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
395 >> PAGE_SHIFT;
396
Alex Elder2794a822013-02-14 12:16:43 -0600397 dout("readpages %p file %p nr_pages %d max %d\n", inode,
398 file, nr_pages,
Sage Weil0d66a482011-08-04 08:21:30 -0700399 max);
Sage Weil7c272192011-08-03 09:58:09 -0700400 while (!list_empty(page_list)) {
Sage Weil0d66a482011-08-04 08:21:30 -0700401 rc = start_read(inode, page_list, max);
Sage Weil7c272192011-08-03 09:58:09 -0700402 if (rc < 0)
403 goto out;
404 BUG_ON(rc == 0);
Sage Weil1d3576f2009-10-06 11:31:09 -0700405 }
Sage Weil1d3576f2009-10-06 11:31:09 -0700406out:
Milosz Tanski76be7782013-08-21 17:30:27 -0400407 ceph_fscache_readpages_cancel(inode, page_list);
408
Sage Weil7c272192011-08-03 09:58:09 -0700409 dout("readpages %p file %p ret %d\n", inode, file, rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700410 return rc;
411}
412
413/*
414 * Get ref for the oldest snapc for an inode with dirty data... that is, the
415 * only snap context we are allowed to write back.
Sage Weil1d3576f2009-10-06 11:31:09 -0700416 */
Sage Weil6298a332010-03-31 22:01:38 -0700417static struct ceph_snap_context *get_oldest_context(struct inode *inode,
418 u64 *snap_size)
Sage Weil1d3576f2009-10-06 11:31:09 -0700419{
420 struct ceph_inode_info *ci = ceph_inode(inode);
421 struct ceph_snap_context *snapc = NULL;
422 struct ceph_cap_snap *capsnap = NULL;
423
Sage Weilbe655592011-11-30 09:47:09 -0800424 spin_lock(&ci->i_ceph_lock);
Sage Weil1d3576f2009-10-06 11:31:09 -0700425 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
426 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
427 capsnap->context, capsnap->dirty_pages);
428 if (capsnap->dirty_pages) {
429 snapc = ceph_get_snap_context(capsnap->context);
430 if (snap_size)
431 *snap_size = capsnap->size;
432 break;
433 }
434 }
Sage Weil7d8cb262010-08-24 08:44:16 -0700435 if (!snapc && ci->i_wrbuffer_ref_head) {
Sage Weil80e755f2010-03-31 21:52:10 -0700436 snapc = ceph_get_snap_context(ci->i_head_snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700437 dout(" head snapc %p has %d dirty pages\n",
438 snapc, ci->i_wrbuffer_ref_head);
439 }
Sage Weilbe655592011-11-30 09:47:09 -0800440 spin_unlock(&ci->i_ceph_lock);
Sage Weil1d3576f2009-10-06 11:31:09 -0700441 return snapc;
442}
443
444/*
445 * Write a single page, but leave the page locked.
446 *
447 * If we get a write error, set the page error bit, but still adjust the
448 * dirty page accounting (i.e., page is no longer dirty).
449 */
450static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
451{
452 struct inode *inode;
453 struct ceph_inode_info *ci;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700454 struct ceph_fs_client *fsc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700455 struct ceph_osd_client *osdc;
Sage Weil6298a332010-03-31 22:01:38 -0700456 struct ceph_snap_context *snapc, *oldest;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800457 loff_t page_off = page_offset(page);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800458 long writeback_stat;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800459 u64 truncate_size, snap_size = 0;
460 u32 truncate_seq;
461 int err = 0, len = PAGE_CACHE_SIZE;
Sage Weil1d3576f2009-10-06 11:31:09 -0700462
463 dout("writepage %p idx %lu\n", page, page->index);
464
465 if (!page->mapping || !page->mapping->host) {
466 dout("writepage %p - no mapping\n", page);
467 return -EFAULT;
468 }
469 inode = page->mapping->host;
470 ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700471 fsc = ceph_inode_to_client(inode);
472 osdc = &fsc->client->osdc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700473
474 /* verify this is a writeable snap context */
Yan, Zheng61600ef2012-05-28 14:44:30 +0800475 snapc = page_snap_context(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700476 if (snapc == NULL) {
477 dout("writepage %p page %p not dirty?\n", inode, page);
478 goto out;
479 }
Sage Weil6298a332010-03-31 22:01:38 -0700480 oldest = get_oldest_context(inode, &snap_size);
481 if (snapc->seq > oldest->seq) {
Sage Weil1d3576f2009-10-06 11:31:09 -0700482 dout("writepage %p page %p snapc %p not writeable - noop\n",
Yan, Zheng61600ef2012-05-28 14:44:30 +0800483 inode, page, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700484 /* we should only noop if called by kswapd */
485 WARN_ON((current->flags & PF_MEMALLOC) == 0);
Sage Weil6298a332010-03-31 22:01:38 -0700486 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -0700487 goto out;
488 }
Sage Weil6298a332010-03-31 22:01:38 -0700489 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -0700490
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800491 spin_lock(&ci->i_ceph_lock);
492 truncate_seq = ci->i_truncate_seq;
493 truncate_size = ci->i_truncate_size;
494 if (!snap_size)
495 snap_size = i_size_read(inode);
496 spin_unlock(&ci->i_ceph_lock);
497
Sage Weil1d3576f2009-10-06 11:31:09 -0700498 /* is this a partial page at end of file? */
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800499 if (page_off >= snap_size) {
500 dout("%p page eof %llu\n", page, snap_size);
501 goto out;
502 }
503 if (snap_size < page_off + len)
504 len = snap_size - page_off;
Sage Weil1d3576f2009-10-06 11:31:09 -0700505
Sage Weilae00d4f2010-09-16 16:26:51 -0700506 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
507 inode, page, page->index, page_off, len, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700508
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700509 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800510 if (writeback_stat >
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700511 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
512 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800513
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400514 ceph_readpage_to_fscache(inode, page);
515
Sage Weil1d3576f2009-10-06 11:31:09 -0700516 set_page_writeback(page);
517 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
518 &ci->i_layout, snapc,
519 page_off, len,
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800520 truncate_seq, truncate_size,
Alex Elder24808822013-02-15 11:42:29 -0600521 &inode->i_mtime, &page, 1);
Sage Weil1d3576f2009-10-06 11:31:09 -0700522 if (err < 0) {
523 dout("writepage setting page/mapping error %d %p\n", err, page);
524 SetPageError(page);
525 mapping_set_error(&inode->i_data, err);
526 if (wbc)
527 wbc->pages_skipped++;
528 } else {
529 dout("writepage cleaned page %p\n", page);
530 err = 0; /* vfs expects us to return 0 */
531 }
532 page->private = 0;
533 ClearPagePrivate(page);
534 end_page_writeback(page);
535 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
Sage Weil6298a332010-03-31 22:01:38 -0700536 ceph_put_snap_context(snapc); /* page's reference */
Sage Weil1d3576f2009-10-06 11:31:09 -0700537out:
538 return err;
539}
540
541static int ceph_writepage(struct page *page, struct writeback_control *wbc)
542{
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800543 int err;
544 struct inode *inode = page->mapping->host;
545 BUG_ON(!inode);
Sage Weil70b666c2011-05-27 09:24:26 -0700546 ihold(inode);
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800547 err = writepage_nounlock(page, wbc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700548 unlock_page(page);
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800549 iput(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700550 return err;
551}
552
553
554/*
555 * lame release_pages helper. release_pages() isn't exported to
556 * modules.
557 */
558static void ceph_release_pages(struct page **pages, int num)
559{
560 struct pagevec pvec;
561 int i;
562
563 pagevec_init(&pvec, 0);
564 for (i = 0; i < num; i++) {
565 if (pagevec_add(&pvec, pages[i]) == 0)
566 pagevec_release(&pvec);
567 }
568 pagevec_release(&pvec);
569}
570
Sage Weil1d3576f2009-10-06 11:31:09 -0700571/*
572 * async writeback completion handler.
573 *
574 * If we get an error, set the mapping error bit, but not the individual
575 * page error bits.
576 */
577static void writepages_finish(struct ceph_osd_request *req,
578 struct ceph_msg *msg)
579{
580 struct inode *inode = req->r_inode;
Sage Weil1d3576f2009-10-06 11:31:09 -0700581 struct ceph_inode_info *ci = ceph_inode(inode);
Alex Elder87060c12013-04-03 01:28:58 -0500582 struct ceph_osd_data *osd_data;
Sage Weil1d3576f2009-10-06 11:31:09 -0700583 unsigned wrote;
Sage Weil1d3576f2009-10-06 11:31:09 -0700584 struct page *page;
Alex Eldere0c59482013-03-07 15:38:25 -0600585 int num_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700586 int i;
587 struct ceph_snap_context *snapc = req->r_snapc;
588 struct address_space *mapping = inode->i_mapping;
Sage Weil1b83bef2013-02-25 16:11:12 -0800589 int rc = req->r_result;
Alex Elder79528732013-04-03 21:32:51 -0500590 u64 bytes = req->r_ops[0].extent.length;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700591 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800592 long writeback_stat;
Sage Weil7ff899d2010-04-23 10:25:33 -0700593 unsigned issued = ceph_caps_issued(ci);
Sage Weil1d3576f2009-10-06 11:31:09 -0700594
Alex Elder406e2c92013-04-15 14:50:36 -0500595 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder87060c12013-04-03 01:28:58 -0500596 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
597 num_pages = calc_pages_for((u64)osd_data->alignment,
598 (u64)osd_data->length);
Sage Weil1d3576f2009-10-06 11:31:09 -0700599 if (rc >= 0) {
Sage Weil79788c62010-02-02 16:34:04 -0800600 /*
601 * Assume we wrote the pages we originally sent. The
602 * osd might reply with fewer pages if our writeback
603 * raced with a truncation and was adjusted at the osd,
604 * so don't believe the reply.
605 */
Alex Eldere0c59482013-03-07 15:38:25 -0600606 wrote = num_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700607 } else {
608 wrote = 0;
609 mapping_set_error(mapping, rc);
610 }
611 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
612 inode, rc, bytes, wrote);
613
614 /* clean all pages */
Alex Eldere0c59482013-03-07 15:38:25 -0600615 for (i = 0; i < num_pages; i++) {
Alex Elder87060c12013-04-03 01:28:58 -0500616 page = osd_data->pages[i];
Sage Weil1d3576f2009-10-06 11:31:09 -0700617 BUG_ON(!page);
618 WARN_ON(!PageUptodate(page));
619
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800620 writeback_stat =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700621 atomic_long_dec_return(&fsc->writeback_count);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800622 if (writeback_stat <
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700623 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
624 clear_bdi_congested(&fsc->backing_dev_info,
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800625 BLK_RW_ASYNC);
626
Yan, Zheng61600ef2012-05-28 14:44:30 +0800627 ceph_put_snap_context(page_snap_context(page));
Sage Weil1d3576f2009-10-06 11:31:09 -0700628 page->private = 0;
629 ClearPagePrivate(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700630 dout("unlocking %d %p\n", i, page);
631 end_page_writeback(page);
Yehuda Sadehe63dc5c2010-02-19 00:07:01 +0000632
633 /*
634 * We lost the cache cap, need to truncate the page before
635 * it is unlocked, otherwise we'd truncate it later in the
636 * page truncation thread, possibly losing some data that
637 * raced its way in
638 */
Sage Weil29625072010-05-27 10:40:43 -0700639 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
Yehuda Sadehe63dc5c2010-02-19 00:07:01 +0000640 generic_error_remove_page(inode->i_mapping, page);
641
Sage Weil1d3576f2009-10-06 11:31:09 -0700642 unlock_page(page);
643 }
644 dout("%p wrote+cleaned %d pages\n", inode, wrote);
Alex Eldere0c59482013-03-07 15:38:25 -0600645 ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700646
Alex Elder87060c12013-04-03 01:28:58 -0500647 ceph_release_pages(osd_data->pages, num_pages);
648 if (osd_data->pages_from_pool)
649 mempool_free(osd_data->pages,
Cheng Renquan640ef792010-03-26 17:40:33 +0800650 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
Sage Weil1d3576f2009-10-06 11:31:09 -0700651 else
Alex Elder87060c12013-04-03 01:28:58 -0500652 kfree(osd_data->pages);
Sage Weil1d3576f2009-10-06 11:31:09 -0700653 ceph_osdc_put_request(req);
654}
655
Sage Weil1d3576f2009-10-06 11:31:09 -0700656/*
657 * initiate async writeback
658 */
659static int ceph_writepages_start(struct address_space *mapping,
660 struct writeback_control *wbc)
661{
662 struct inode *inode = mapping->host;
Sage Weil1d3576f2009-10-06 11:31:09 -0700663 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800664 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
665 struct ceph_vino vino = ceph_vino(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700666 pgoff_t index, start, end;
667 int range_whole = 0;
668 int should_loop = 1;
669 pgoff_t max_pages = 0, max_pages_ever = 0;
Sage Weil80e755f2010-03-31 21:52:10 -0700670 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700671 struct pagevec pvec;
672 int done = 0;
673 int rc = 0;
674 unsigned wsize = 1 << inode->i_blkbits;
675 struct ceph_osd_request *req = NULL;
676 int do_sync;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800677 u64 truncate_size, snap_size;
678 u32 truncate_seq;
Sage Weil1d3576f2009-10-06 11:31:09 -0700679
680 /*
681 * Include a 'sync' in the OSD request if this is a data
682 * integrity write (e.g., O_SYNC write or fsync()), or if our
683 * cap is being revoked.
684 */
majianpengc62988e2013-06-19 15:12:06 +0800685 if ((wbc->sync_mode == WB_SYNC_ALL) ||
686 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
Sage Weil1d3576f2009-10-06 11:31:09 -0700687 do_sync = 1;
688 dout("writepages_start %p dosync=%d (mode=%s)\n",
689 inode, do_sync,
690 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
691 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
692
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700693 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
Fabian Frederickf3ae1b92014-06-06 14:35:37 -0700694 pr_warn("writepage_start %p on forced umount\n", inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700695 return -EIO; /* we're in a forced umount, don't write! */
696 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700697 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
698 wsize = fsc->mount_options->wsize;
Sage Weil1d3576f2009-10-06 11:31:09 -0700699 if (wsize < PAGE_CACHE_SIZE)
700 wsize = PAGE_CACHE_SIZE;
701 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
702
703 pagevec_init(&pvec, 0);
704
Sage Weil1d3576f2009-10-06 11:31:09 -0700705 /* where to start/end? */
706 if (wbc->range_cyclic) {
707 start = mapping->writeback_index; /* Start from prev offset */
708 end = -1;
709 dout(" cyclic, start at %lu\n", start);
710 } else {
711 start = wbc->range_start >> PAGE_CACHE_SHIFT;
712 end = wbc->range_end >> PAGE_CACHE_SHIFT;
713 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
714 range_whole = 1;
715 should_loop = 0;
716 dout(" not cyclic, %lu to %lu\n", start, end);
717 }
718 index = start;
719
720retry:
721 /* find oldest snap context with dirty data */
722 ceph_put_snap_context(snapc);
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800723 snap_size = 0;
Sage Weil1d3576f2009-10-06 11:31:09 -0700724 snapc = get_oldest_context(inode, &snap_size);
725 if (!snapc) {
726 /* hmm, why does writepages get called when there
727 is no dirty data? */
728 dout(" no snap context with dirty data?\n");
729 goto out;
730 }
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800731 if (snap_size == 0)
732 snap_size = i_size_read(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700733 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
734 snapc, snapc->seq, snapc->num_snaps);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800735
736 spin_lock(&ci->i_ceph_lock);
737 truncate_seq = ci->i_truncate_seq;
738 truncate_size = ci->i_truncate_size;
739 if (!snap_size)
740 snap_size = i_size_read(inode);
741 spin_unlock(&ci->i_ceph_lock);
742
Sage Weil1d3576f2009-10-06 11:31:09 -0700743 if (last_snapc && snapc != last_snapc) {
744 /* if we switched to a newer snapc, restart our scan at the
745 * start of the original file range. */
746 dout(" snapc differs from last pass, restarting at %lu\n",
747 index);
748 index = start;
749 }
750 last_snapc = snapc;
751
752 while (!done && index <= end) {
753 unsigned i;
754 int first;
755 pgoff_t next;
756 int pvec_pages, locked_pages;
Alex Eldere5975c72013-03-14 14:09:05 -0500757 struct page **pages = NULL;
758 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
Sage Weil1d3576f2009-10-06 11:31:09 -0700759 struct page *page;
760 int want;
761 u64 offset, len;
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800762 long writeback_stat;
Sage Weil1d3576f2009-10-06 11:31:09 -0700763
764 next = 0;
765 locked_pages = 0;
766 max_pages = max_pages_ever;
767
768get_more_pages:
769 first = -1;
770 want = min(end - index,
771 min((pgoff_t)PAGEVEC_SIZE,
772 max_pages - (pgoff_t)locked_pages) - 1)
773 + 1;
774 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
775 PAGECACHE_TAG_DIRTY,
776 want);
777 dout("pagevec_lookup_tag got %d\n", pvec_pages);
778 if (!pvec_pages && !locked_pages)
779 break;
780 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
781 page = pvec.pages[i];
782 dout("? %p idx %lu\n", page, page->index);
783 if (locked_pages == 0)
784 lock_page(page); /* first page */
785 else if (!trylock_page(page))
786 break;
787
788 /* only dirty pages, or our accounting breaks */
789 if (unlikely(!PageDirty(page)) ||
790 unlikely(page->mapping != mapping)) {
791 dout("!dirty or !mapping %p\n", page);
792 unlock_page(page);
793 break;
794 }
795 if (!wbc->range_cyclic && page->index > end) {
796 dout("end of range %p\n", page);
797 done = 1;
798 unlock_page(page);
799 break;
800 }
801 if (next && (page->index != next)) {
802 dout("not consecutive %p\n", page);
803 unlock_page(page);
804 break;
805 }
806 if (wbc->sync_mode != WB_SYNC_NONE) {
807 dout("waiting on writeback %p\n", page);
808 wait_on_page_writeback(page);
809 }
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800810 if (page_offset(page) >= snap_size) {
811 dout("%p page eof %llu\n", page, snap_size);
Sage Weil1d3576f2009-10-06 11:31:09 -0700812 done = 1;
813 unlock_page(page);
814 break;
815 }
816 if (PageWriteback(page)) {
817 dout("%p under writeback\n", page);
818 unlock_page(page);
819 break;
820 }
821
822 /* only if matching snap context */
Yan, Zheng61600ef2012-05-28 14:44:30 +0800823 pgsnapc = page_snap_context(page);
Sage Weil80e755f2010-03-31 21:52:10 -0700824 if (pgsnapc->seq > snapc->seq) {
825 dout("page snapc %p %lld > oldest %p %lld\n",
826 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
Sage Weil1d3576f2009-10-06 11:31:09 -0700827 unlock_page(page);
828 if (!locked_pages)
829 continue; /* keep looking for snap */
830 break;
831 }
832
833 if (!clear_page_dirty_for_io(page)) {
834 dout("%p !clear_page_dirty_for_io\n", page);
835 unlock_page(page);
836 break;
837 }
838
Alex Eldere5975c72013-03-14 14:09:05 -0500839 /*
840 * We have something to write. If this is
841 * the first locked page this time through,
842 * allocate an osd request and a page array
843 * that it will use.
844 */
Sage Weil1d3576f2009-10-06 11:31:09 -0700845 if (locked_pages == 0) {
Alex Eldere5975c72013-03-14 14:09:05 -0500846 BUG_ON(pages);
Sage Weil1d3576f2009-10-06 11:31:09 -0700847 /* prepare async write request */
Alex Eldere5975c72013-03-14 14:09:05 -0500848 offset = (u64)page_offset(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700849 len = wsize;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800850 req = ceph_osdc_new_request(&fsc->client->osdc,
851 &ci->i_layout, vino,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800852 offset, &len, 0,
853 do_sync ? 2 : 1,
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800854 CEPH_OSD_OP_WRITE,
855 CEPH_OSD_FLAG_WRITE |
856 CEPH_OSD_FLAG_ONDISK,
857 snapc, truncate_seq,
858 truncate_size, true);
Sage Weil68162822012-09-24 21:01:02 -0700859 if (IS_ERR(req)) {
860 rc = PTR_ERR(req);
Henry C Chang8c718972011-05-03 09:45:16 +0000861 unlock_page(page);
862 break;
863 }
864
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800865 if (do_sync)
866 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
867
Alex Elder88486952013-03-14 14:09:05 -0500868 req->r_callback = writepages_finish;
869 req->r_inode = inode;
870
871 max_pages = calc_pages_for(0, (u64)len);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800872 pages = kmalloc(max_pages * sizeof (*pages),
873 GFP_NOFS);
Alex Elder88486952013-03-14 14:09:05 -0500874 if (!pages) {
875 pool = fsc->wb_pagevec_pool;
Alex Elder88486952013-03-14 14:09:05 -0500876 pages = mempool_alloc(pool, GFP_NOFS);
Alex Eldere5975c72013-03-14 14:09:05 -0500877 BUG_ON(!pages);
Alex Elder88486952013-03-14 14:09:05 -0500878 }
Sage Weil1d3576f2009-10-06 11:31:09 -0700879 }
880
881 /* note position of first page in pvec */
882 if (first < 0)
883 first = i;
884 dout("%p will write page %p idx %lu\n",
885 inode, page, page->index);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800886
Sage Weil213c99e2010-08-03 10:25:11 -0700887 writeback_stat =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700888 atomic_long_inc_return(&fsc->writeback_count);
Sage Weil213c99e2010-08-03 10:25:11 -0700889 if (writeback_stat > CONGESTION_ON_THRESH(
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700890 fsc->mount_options->congestion_kb)) {
891 set_bdi_congested(&fsc->backing_dev_info,
Sage Weil213c99e2010-08-03 10:25:11 -0700892 BLK_RW_ASYNC);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800893 }
894
Sage Weil1d3576f2009-10-06 11:31:09 -0700895 set_page_writeback(page);
Alex Eldere5975c72013-03-14 14:09:05 -0500896 pages[locked_pages] = page;
Sage Weil1d3576f2009-10-06 11:31:09 -0700897 locked_pages++;
898 next = page->index + 1;
899 }
900
901 /* did we get anything? */
902 if (!locked_pages)
903 goto release_pvec_pages;
904 if (i) {
905 int j;
906 BUG_ON(!locked_pages || first < 0);
907
908 if (pvec_pages && i == pvec_pages &&
909 locked_pages < max_pages) {
910 dout("reached end pvec, trying for more\n");
911 pagevec_reinit(&pvec);
912 goto get_more_pages;
913 }
914
915 /* shift unused pages over in the pvec... we
916 * will need to release them below. */
917 for (j = i; j < pvec_pages; j++) {
918 dout(" pvec leftover page %p\n",
919 pvec.pages[j]);
920 pvec.pages[j-i+first] = pvec.pages[j];
921 }
922 pvec.nr -= i-first;
923 }
924
Alex Eldere5975c72013-03-14 14:09:05 -0500925 /* Format the osd request message and submit the write */
926
927 offset = page_offset(pages[0]);
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800928 len = min(snap_size - offset,
Sage Weil1d3576f2009-10-06 11:31:09 -0700929 (u64)locked_pages << PAGE_CACHE_SHIFT);
930 dout("writepages got %d pages at %llu~%llu\n",
931 locked_pages, offset, len);
932
Alex Elder406e2c92013-04-15 14:50:36 -0500933 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500934 !!pool, false);
Alex Eldere5975c72013-03-14 14:09:05 -0500935
936 pages = NULL; /* request message now owns the pages array */
937 pool = NULL;
938
939 /* Update the write op length in case we changed it */
940
Alex Elderc99d2d42013-04-05 01:27:11 -0500941 osd_req_op_extent_update(req, 0, len);
Alex Eldere5975c72013-03-14 14:09:05 -0500942
943 vino = ceph_vino(inode);
Alex Elder79528732013-04-03 21:32:51 -0500944 ceph_osdc_build_request(req, offset, snapc, vino.snap,
945 &inode->i_mtime);
Sage Weil1d3576f2009-10-06 11:31:09 -0700946
Sage Weil9d6fcb02011-05-12 15:48:16 -0700947 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
948 BUG_ON(rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700949 req = NULL;
950
951 /* continue? */
952 index = next;
953 wbc->nr_to_write -= locked_pages;
954 if (wbc->nr_to_write <= 0)
955 done = 1;
956
957release_pvec_pages:
958 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
959 pvec.nr ? pvec.pages[0] : NULL);
960 pagevec_release(&pvec);
961
962 if (locked_pages && !done)
963 goto retry;
964 }
965
966 if (should_loop && !done) {
967 /* more to do; loop back to beginning of file */
968 dout("writepages looping back to beginning of file\n");
969 should_loop = 0;
970 index = 0;
971 goto retry;
972 }
973
974 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
975 mapping->writeback_index = index;
976
977out:
978 if (req)
979 ceph_osdc_put_request(req);
Sage Weil1d3576f2009-10-06 11:31:09 -0700980 ceph_put_snap_context(snapc);
981 dout("writepages done, rc = %d\n", rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700982 return rc;
983}
984
985
986
987/*
988 * See if a given @snapc is either writeable, or already written.
989 */
990static int context_is_writeable_or_written(struct inode *inode,
991 struct ceph_snap_context *snapc)
992{
993 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
Sage Weil6298a332010-03-31 22:01:38 -0700994 int ret = !oldest || snapc->seq <= oldest->seq;
995
996 ceph_put_snap_context(oldest);
997 return ret;
Sage Weil1d3576f2009-10-06 11:31:09 -0700998}
999
1000/*
1001 * We are only allowed to write into/dirty the page if the page is
1002 * clean, or already dirty within the same snap context.
Sage Weil8f883c22010-03-19 13:27:53 -07001003 *
1004 * called with page locked.
1005 * return success with page locked,
1006 * or any failure (incl -EAGAIN) with page unlocked.
Sage Weil1d3576f2009-10-06 11:31:09 -07001007 */
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001008static int ceph_update_writeable_page(struct file *file,
1009 loff_t pos, unsigned len,
1010 struct page *page)
Sage Weil1d3576f2009-10-06 11:31:09 -07001011{
Al Viro496ad9a2013-01-23 17:07:38 -05001012 struct inode *inode = file_inode(file);
Sage Weil1d3576f2009-10-06 11:31:09 -07001013 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001014 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weil1d3576f2009-10-06 11:31:09 -07001015 loff_t page_off = pos & PAGE_CACHE_MASK;
1016 int pos_in_page = pos & ~PAGE_CACHE_MASK;
1017 int end_in_page = pos_in_page + len;
1018 loff_t i_size;
Sage Weil1d3576f2009-10-06 11:31:09 -07001019 int r;
Sage Weil80e755f2010-03-31 21:52:10 -07001020 struct ceph_snap_context *snapc, *oldest;
Sage Weil1d3576f2009-10-06 11:31:09 -07001021
Sage Weil1d3576f2009-10-06 11:31:09 -07001022retry_locked:
1023 /* writepages currently holds page lock, but if we change that later, */
1024 wait_on_page_writeback(page);
1025
1026 /* check snap context */
1027 BUG_ON(!ci->i_snap_realm);
1028 down_read(&mdsc->snap_rwsem);
1029 BUG_ON(!ci->i_snap_realm->cached_context);
Yan, Zheng61600ef2012-05-28 14:44:30 +08001030 snapc = page_snap_context(page);
Sage Weil80e755f2010-03-31 21:52:10 -07001031 if (snapc && snapc != ci->i_head_snapc) {
Sage Weil1d3576f2009-10-06 11:31:09 -07001032 /*
1033 * this page is already dirty in another (older) snap
1034 * context! is it writeable now?
1035 */
Sage Weil80e755f2010-03-31 21:52:10 -07001036 oldest = get_oldest_context(inode, NULL);
Sage Weil1d3576f2009-10-06 11:31:09 -07001037 up_read(&mdsc->snap_rwsem);
1038
Sage Weil80e755f2010-03-31 21:52:10 -07001039 if (snapc->seq > oldest->seq) {
Sage Weil6298a332010-03-31 22:01:38 -07001040 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -07001041 dout(" page %p snapc %p not current or oldest\n",
Sage Weil6298a332010-03-31 22:01:38 -07001042 page, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -07001043 /*
1044 * queue for writeback, and wait for snapc to
1045 * be writeable or written
1046 */
Sage Weil6298a332010-03-31 22:01:38 -07001047 snapc = ceph_get_snap_context(snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -07001048 unlock_page(page);
Sage Weil3c6f6b792010-02-09 15:24:44 -08001049 ceph_queue_writeback(inode);
Sage Weil8f883c22010-03-19 13:27:53 -07001050 r = wait_event_interruptible(ci->i_cap_wq,
Sage Weil1d3576f2009-10-06 11:31:09 -07001051 context_is_writeable_or_written(inode, snapc));
1052 ceph_put_snap_context(snapc);
Sage Weil8f883c22010-03-19 13:27:53 -07001053 if (r == -ERESTARTSYS)
1054 return r;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001055 return -EAGAIN;
Sage Weil1d3576f2009-10-06 11:31:09 -07001056 }
Sage Weil6298a332010-03-31 22:01:38 -07001057 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -07001058
1059 /* yay, writeable, do it now (without dropping page lock) */
1060 dout(" page %p snapc %p not current, but oldest\n",
1061 page, snapc);
1062 if (!clear_page_dirty_for_io(page))
1063 goto retry_locked;
1064 r = writepage_nounlock(page, NULL);
1065 if (r < 0)
1066 goto fail_nosnap;
1067 goto retry_locked;
1068 }
1069
1070 if (PageUptodate(page)) {
1071 dout(" page %p already uptodate\n", page);
1072 return 0;
1073 }
1074
1075 /* full page? */
1076 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1077 return 0;
1078
1079 /* past end of file? */
1080 i_size = inode->i_size; /* caller holds i_mutex */
1081
Sage Weil1d3576f2009-10-06 11:31:09 -07001082 if (page_off >= i_size ||
1083 (pos_in_page == 0 && (pos+len) >= i_size &&
1084 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1085 dout(" zeroing %p 0 - %d and %d - %d\n",
1086 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1087 zero_user_segments(page,
1088 0, pos_in_page,
1089 end_in_page, PAGE_CACHE_SIZE);
1090 return 0;
1091 }
1092
1093 /* we need to read it. */
1094 up_read(&mdsc->snap_rwsem);
1095 r = readpage_nounlock(file, page);
1096 if (r < 0)
1097 goto fail_nosnap;
1098 goto retry_locked;
Sage Weil1d3576f2009-10-06 11:31:09 -07001099fail_nosnap:
1100 unlock_page(page);
1101 return r;
1102}
1103
1104/*
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001105 * We are only allowed to write into/dirty the page if the page is
1106 * clean, or already dirty within the same snap context.
1107 */
1108static int ceph_write_begin(struct file *file, struct address_space *mapping,
1109 loff_t pos, unsigned len, unsigned flags,
1110 struct page **pagep, void **fsdata)
1111{
Al Viro496ad9a2013-01-23 17:07:38 -05001112 struct inode *inode = file_inode(file);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001113 struct page *page;
1114 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Sage Weil7971bd92013-05-01 21:15:58 -07001115 int r;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001116
1117 do {
Sage Weil8f883c22010-03-19 13:27:53 -07001118 /* get a page */
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001119 page = grab_cache_page_write_begin(mapping, index, 0);
Sage Weil7971bd92013-05-01 21:15:58 -07001120 if (!page)
1121 return -ENOMEM;
1122 *pagep = page;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001123
1124 dout("write_begin file %p inode %p page %p %d~%d\n", file,
Sage Weil213c99e2010-08-03 10:25:11 -07001125 inode, page, (int)pos, (int)len);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001126
1127 r = ceph_update_writeable_page(file, pos, len, page);
1128 } while (r == -EAGAIN);
1129
1130 return r;
1131}
1132
1133/*
Sage Weil1d3576f2009-10-06 11:31:09 -07001134 * we don't do anything in here that simple_write_end doesn't do
1135 * except adjust dirty page accounting and drop read lock on
1136 * mdsc->snap_rwsem.
1137 */
1138static int ceph_write_end(struct file *file, struct address_space *mapping,
1139 loff_t pos, unsigned len, unsigned copied,
1140 struct page *page, void *fsdata)
1141{
Al Viro496ad9a2013-01-23 17:07:38 -05001142 struct inode *inode = file_inode(file);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001143 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1144 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil1d3576f2009-10-06 11:31:09 -07001145 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1146 int check_cap = 0;
1147
1148 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1149 inode, page, (int)pos, (int)copied, (int)len);
1150
1151 /* zero the stale part of the page if we did a short copy */
1152 if (copied < len)
1153 zero_user_segment(page, from+copied, len);
1154
1155 /* did file size increase? */
1156 /* (no need for i_size_read(); we caller holds i_mutex */
1157 if (pos+copied > inode->i_size)
1158 check_cap = ceph_inode_set_size(inode, pos+copied);
1159
1160 if (!PageUptodate(page))
1161 SetPageUptodate(page);
1162
1163 set_page_dirty(page);
1164
1165 unlock_page(page);
1166 up_read(&mdsc->snap_rwsem);
1167 page_cache_release(page);
1168
1169 if (check_cap)
1170 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1171
1172 return copied;
1173}
1174
1175/*
1176 * we set .direct_IO to indicate direct io is supported, but since we
1177 * intercept O_DIRECT reads and writes early, this function should
1178 * never get called.
1179 */
1180static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05001181 struct iov_iter *iter,
1182 loff_t pos)
Sage Weil1d3576f2009-10-06 11:31:09 -07001183{
1184 WARN_ON(1);
1185 return -EINVAL;
1186}
1187
1188const struct address_space_operations ceph_aops = {
1189 .readpage = ceph_readpage,
1190 .readpages = ceph_readpages,
1191 .writepage = ceph_writepage,
1192 .writepages = ceph_writepages_start,
1193 .write_begin = ceph_write_begin,
1194 .write_end = ceph_write_end,
1195 .set_page_dirty = ceph_set_page_dirty,
1196 .invalidatepage = ceph_invalidatepage,
1197 .releasepage = ceph_releasepage,
1198 .direct_IO = ceph_direct_io,
1199};
1200
1201
1202/*
1203 * vm ops
1204 */
Yan, Zheng61f68812013-11-28 14:28:14 +08001205static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1206{
1207 struct inode *inode = file_inode(vma->vm_file);
1208 struct ceph_inode_info *ci = ceph_inode(inode);
1209 struct ceph_file_info *fi = vma->vm_file->private_data;
1210 loff_t off = vmf->pgoff << PAGE_CACHE_SHIFT;
1211 int want, got, ret;
1212
1213 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001214 inode, ceph_vinop(inode), off, (size_t)PAGE_CACHE_SIZE);
Yan, Zheng61f68812013-11-28 14:28:14 +08001215 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1216 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1217 else
1218 want = CEPH_CAP_FILE_CACHE;
1219 while (1) {
1220 got = 0;
1221 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
1222 if (ret == 0)
1223 break;
1224 if (ret != -ERESTARTSYS) {
1225 WARN_ON(1);
1226 return VM_FAULT_SIGBUS;
1227 }
1228 }
1229 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001230 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got));
Yan, Zheng61f68812013-11-28 14:28:14 +08001231
1232 ret = filemap_fault(vma, vmf);
1233
1234 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001235 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got), ret);
Yan, Zheng61f68812013-11-28 14:28:14 +08001236 ceph_put_cap_refs(ci, got);
1237
1238 return ret;
1239}
Sage Weil1d3576f2009-10-06 11:31:09 -07001240
1241/*
1242 * Reuse write_begin here for simplicity.
1243 */
1244static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1245{
Al Viro496ad9a2013-01-23 17:07:38 -05001246 struct inode *inode = file_inode(vma->vm_file);
Yan, Zheng61f68812013-11-28 14:28:14 +08001247 struct ceph_inode_info *ci = ceph_inode(inode);
1248 struct ceph_file_info *fi = vma->vm_file->private_data;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001249 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng61f68812013-11-28 14:28:14 +08001250 struct page *page = vmf->page;
Alex Elder6285bc22012-10-02 10:25:51 -05001251 loff_t off = page_offset(page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001252 loff_t size = i_size_read(inode);
1253 size_t len;
1254 int want, got, ret;
Sage Weil1d3576f2009-10-06 11:31:09 -07001255
Sage Weil1d3576f2009-10-06 11:31:09 -07001256 if (off + PAGE_CACHE_SIZE <= size)
1257 len = PAGE_CACHE_SIZE;
1258 else
1259 len = size & ~PAGE_CACHE_MASK;
1260
Yan, Zheng61f68812013-11-28 14:28:14 +08001261 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1262 inode, ceph_vinop(inode), off, len, size);
1263 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1264 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1265 else
1266 want = CEPH_CAP_FILE_BUFFER;
1267 while (1) {
1268 got = 0;
1269 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, off + len);
1270 if (ret == 0)
1271 break;
1272 if (ret != -ERESTARTSYS) {
1273 WARN_ON(1);
1274 return VM_FAULT_SIGBUS;
1275 }
1276 }
1277 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1278 inode, off, len, ceph_cap_string(got));
1279
1280 /* Update time before taking page lock */
1281 file_update_time(vma->vm_file);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001282
1283 lock_page(page);
1284
1285 ret = VM_FAULT_NOPAGE;
1286 if ((off > size) ||
1287 (page->mapping != inode->i_mapping))
1288 goto out;
1289
1290 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1291 if (ret == 0) {
1292 /* success. we'll keep the page locked. */
Sage Weil1d3576f2009-10-06 11:31:09 -07001293 set_page_dirty(page);
1294 up_read(&mdsc->snap_rwsem);
Sage Weil1d3576f2009-10-06 11:31:09 -07001295 ret = VM_FAULT_LOCKED;
1296 } else {
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001297 if (ret == -ENOMEM)
1298 ret = VM_FAULT_OOM;
1299 else
1300 ret = VM_FAULT_SIGBUS;
Sage Weil1d3576f2009-10-06 11:31:09 -07001301 }
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001302out:
Yan, Zheng61f68812013-11-28 14:28:14 +08001303 if (ret != VM_FAULT_LOCKED) {
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001304 unlock_page(page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001305 } else {
1306 int dirty;
1307 spin_lock(&ci->i_ceph_lock);
1308 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1309 spin_unlock(&ci->i_ceph_lock);
1310 if (dirty)
1311 __mark_inode_dirty(inode, dirty);
1312 }
1313
1314 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1315 inode, off, len, ceph_cap_string(got), ret);
1316 ceph_put_cap_refs(ci, got);
1317
Sage Weil1d3576f2009-10-06 11:31:09 -07001318 return ret;
1319}
1320
Yan, Zheng31c542a2014-11-14 21:41:55 +08001321void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1322 char *data, size_t len)
1323{
1324 struct address_space *mapping = inode->i_mapping;
1325 struct page *page;
1326
1327 if (locked_page) {
1328 page = locked_page;
1329 } else {
1330 if (i_size_read(inode) == 0)
1331 return;
1332 page = find_or_create_page(mapping, 0,
1333 mapping_gfp_mask(mapping) & ~__GFP_FS);
1334 if (!page)
1335 return;
1336 if (PageUptodate(page)) {
1337 unlock_page(page);
1338 page_cache_release(page);
1339 return;
1340 }
1341 }
1342
1343 dout("fill_inline_data %p %llx.%llx len %lu locked_page %p\n",
1344 inode, ceph_vinop(inode), len, locked_page);
1345
1346 if (len > 0) {
1347 void *kaddr = kmap_atomic(page);
1348 memcpy(kaddr, data, len);
1349 kunmap_atomic(kaddr);
1350 }
1351
1352 if (page != locked_page) {
1353 if (len < PAGE_CACHE_SIZE)
1354 zero_user_segment(page, len, PAGE_CACHE_SIZE);
1355 else
1356 flush_dcache_page(page);
1357
1358 SetPageUptodate(page);
1359 unlock_page(page);
1360 page_cache_release(page);
1361 }
1362}
1363
Sage Weil1d3576f2009-10-06 11:31:09 -07001364static struct vm_operations_struct ceph_vmops = {
Yan, Zheng61f68812013-11-28 14:28:14 +08001365 .fault = ceph_filemap_fault,
Sage Weil1d3576f2009-10-06 11:31:09 -07001366 .page_mkwrite = ceph_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001367 .remap_pages = generic_file_remap_pages,
Sage Weil1d3576f2009-10-06 11:31:09 -07001368};
1369
1370int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1371{
1372 struct address_space *mapping = file->f_mapping;
1373
1374 if (!mapping->a_ops->readpage)
1375 return -ENOEXEC;
1376 file_accessed(file);
1377 vma->vm_ops = &ceph_vmops;
Sage Weil1d3576f2009-10-06 11:31:09 -07001378 return 0;
1379}