blob: 13413d7440d6896fc883a2625bec8f1d53458656 [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;
Yan, Zheng83701242014-11-14 22:36:18 +0800195 u64 off = page_offset(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700196 u64 len = PAGE_CACHE_SIZE;
197
Yan, Zheng83701242014-11-14 22:36:18 +0800198 if (off >= i_size_read(inode)) {
199 zero_user_segment(page, err, PAGE_CACHE_SIZE);
200 SetPageUptodate(page);
201 return 0;
202 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400203
Yan, Zheng83701242014-11-14 22:36:18 +0800204 /*
205 * Uptodate inline data should have been added into page cache
206 * while getting Fcr caps.
207 */
208 if (ci->i_inline_version != CEPH_INLINE_NONE)
209 return -EINVAL;
210
211 err = ceph_readpage_from_fscache(inode, page);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400212 if (err == 0)
213 goto out;
214
Sage Weil1d3576f2009-10-06 11:31:09 -0700215 dout("readpage inode %p file %p page %p index %lu\n",
216 inode, filp, page, page->index);
217 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
Yan, Zheng83701242014-11-14 22:36:18 +0800218 off, &len,
Sage Weil1d3576f2009-10-06 11:31:09 -0700219 ci->i_truncate_seq, ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800220 &page, 1, 0);
Sage Weil1d3576f2009-10-06 11:31:09 -0700221 if (err == -ENOENT)
222 err = 0;
223 if (err < 0) {
224 SetPageError(page);
Li Wang18302802013-12-19 06:03:49 -0800225 ceph_fscache_readpage_cancel(inode, page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700226 goto out;
Sage Weil1d3576f2009-10-06 11:31:09 -0700227 }
Zhang Zhen23cd5732014-05-28 15:09:55 +0800228 if (err < PAGE_CACHE_SIZE)
229 /* zero fill remainder of page */
230 zero_user_segment(page, err, PAGE_CACHE_SIZE);
231 else
232 flush_dcache_page(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700233
Zhang Zhen23cd5732014-05-28 15:09:55 +0800234 SetPageUptodate(page);
235 ceph_readpage_to_fscache(inode, page);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400236
Sage Weil1d3576f2009-10-06 11:31:09 -0700237out:
238 return err < 0 ? err : 0;
239}
240
241static int ceph_readpage(struct file *filp, struct page *page)
242{
243 int r = readpage_nounlock(filp, page);
244 unlock_page(page);
245 return r;
246}
247
248/*
Sage Weil7c272192011-08-03 09:58:09 -0700249 * Finish an async read(ahead) op.
Sage Weil1d3576f2009-10-06 11:31:09 -0700250 */
Sage Weil7c272192011-08-03 09:58:09 -0700251static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg)
Sage Weil1d3576f2009-10-06 11:31:09 -0700252{
Sage Weil7c272192011-08-03 09:58:09 -0700253 struct inode *inode = req->r_inode;
Alex Elder87060c12013-04-03 01:28:58 -0500254 struct ceph_osd_data *osd_data;
Sage Weil1b83bef2013-02-25 16:11:12 -0800255 int rc = req->r_result;
256 int bytes = le32_to_cpu(msg->hdr.data_len);
Alex Eldere0c59482013-03-07 15:38:25 -0600257 int num_pages;
Sage Weil7c272192011-08-03 09:58:09 -0700258 int i;
259
Sage Weil7c272192011-08-03 09:58:09 -0700260 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
261
262 /* unlock all pages, zeroing any data we didn't read */
Alex Elder406e2c92013-04-15 14:50:36 -0500263 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder87060c12013-04-03 01:28:58 -0500264 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
265 num_pages = calc_pages_for((u64)osd_data->alignment,
266 (u64)osd_data->length);
Alex Eldere0c59482013-03-07 15:38:25 -0600267 for (i = 0; i < num_pages; i++) {
Alex Elder87060c12013-04-03 01:28:58 -0500268 struct page *page = osd_data->pages[i];
Sage Weil7c272192011-08-03 09:58:09 -0700269
Li Wangf36132a2013-11-27 22:28:13 +0800270 if (rc < 0)
271 goto unlock;
Sage Weil7c272192011-08-03 09:58:09 -0700272 if (bytes < (int)PAGE_CACHE_SIZE) {
273 /* zero (remainder of) page */
274 int s = bytes < 0 ? 0 : bytes;
275 zero_user_segment(page, s, PAGE_CACHE_SIZE);
276 }
277 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
278 page->index);
279 flush_dcache_page(page);
280 SetPageUptodate(page);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400281 ceph_readpage_to_fscache(inode, page);
Li Wangf36132a2013-11-27 22:28:13 +0800282unlock:
Sage Weil7c272192011-08-03 09:58:09 -0700283 unlock_page(page);
284 page_cache_release(page);
Alex Elder0fff87e2013-02-14 12:16:43 -0600285 bytes -= PAGE_CACHE_SIZE;
Sage Weil7c272192011-08-03 09:58:09 -0700286 }
Alex Elder87060c12013-04-03 01:28:58 -0500287 kfree(osd_data->pages);
Sage Weil7c272192011-08-03 09:58:09 -0700288}
289
David Zafman8884d532012-12-03 19:14:05 -0800290static void ceph_unlock_page_vector(struct page **pages, int num_pages)
291{
292 int i;
293
294 for (i = 0; i < num_pages; i++)
295 unlock_page(pages[i]);
296}
297
Sage Weil7c272192011-08-03 09:58:09 -0700298/*
299 * start an async read(ahead) operation. return nr_pages we submitted
300 * a read for on success, or negative error code.
301 */
Sage Weil0d66a482011-08-04 08:21:30 -0700302static int start_read(struct inode *inode, struct list_head *page_list, int max)
Sage Weil7c272192011-08-03 09:58:09 -0700303{
304 struct ceph_osd_client *osdc =
305 &ceph_inode_to_client(inode)->client->osdc;
306 struct ceph_inode_info *ci = ceph_inode(inode);
307 struct page *page = list_entry(page_list->prev, struct page, lru);
Alex Elderacead002013-03-14 14:09:05 -0500308 struct ceph_vino vino;
Sage Weil7c272192011-08-03 09:58:09 -0700309 struct ceph_osd_request *req;
310 u64 off;
311 u64 len;
312 int i;
Sage Weil1d3576f2009-10-06 11:31:09 -0700313 struct page **pages;
Sage Weil7c272192011-08-03 09:58:09 -0700314 pgoff_t next_index;
315 int nr_pages = 0;
316 int ret;
317
Alex Elder6285bc22012-10-02 10:25:51 -0500318 off = (u64) page_offset(page);
Sage Weil7c272192011-08-03 09:58:09 -0700319
320 /* count pages */
321 next_index = page->index;
322 list_for_each_entry_reverse(page, page_list, lru) {
323 if (page->index != next_index)
324 break;
325 nr_pages++;
326 next_index++;
Sage Weil0d66a482011-08-04 08:21:30 -0700327 if (max && nr_pages == max)
328 break;
Sage Weil7c272192011-08-03 09:58:09 -0700329 }
330 len = nr_pages << PAGE_CACHE_SHIFT;
331 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
332 off, len);
Alex Elderacead002013-03-14 14:09:05 -0500333 vino = ceph_vino(inode);
334 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800335 0, 1, CEPH_OSD_OP_READ,
Alex Elderacead002013-03-14 14:09:05 -0500336 CEPH_OSD_FLAG_READ, NULL,
Sage Weil7c272192011-08-03 09:58:09 -0700337 ci->i_truncate_seq, ci->i_truncate_size,
Alex Elderacead002013-03-14 14:09:05 -0500338 false);
Sage Weil68162822012-09-24 21:01:02 -0700339 if (IS_ERR(req))
340 return PTR_ERR(req);
Sage Weil1d3576f2009-10-06 11:31:09 -0700341
342 /* build page vector */
Alex Eldercf7b7e142013-03-01 18:00:15 -0600343 nr_pages = calc_pages_for(0, len);
Sage Weil7c272192011-08-03 09:58:09 -0700344 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
345 ret = -ENOMEM;
Sage Weil1d3576f2009-10-06 11:31:09 -0700346 if (!pages)
Sage Weil7c272192011-08-03 09:58:09 -0700347 goto out;
348 for (i = 0; i < nr_pages; ++i) {
349 page = list_entry(page_list->prev, struct page, lru);
350 BUG_ON(PageLocked(page));
351 list_del(&page->lru);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400352
Sage Weil7c272192011-08-03 09:58:09 -0700353 dout("start_read %p adding %p idx %lu\n", inode, page,
354 page->index);
355 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
356 GFP_NOFS)) {
Milosz Tanskid4d3aa32013-09-03 19:11:17 -0400357 ceph_fscache_uncache_page(inode, page);
Sage Weil7c272192011-08-03 09:58:09 -0700358 page_cache_release(page);
359 dout("start_read %p add_to_page_cache failed %p\n",
360 inode, page);
361 nr_pages = i;
362 goto out_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700363 }
Sage Weil7c272192011-08-03 09:58:09 -0700364 pages[i] = page;
Sage Weil1d3576f2009-10-06 11:31:09 -0700365 }
Alex Elder406e2c92013-04-15 14:50:36 -0500366 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
Sage Weil7c272192011-08-03 09:58:09 -0700367 req->r_callback = finish_read;
368 req->r_inode = inode;
369
Alex Elder79528732013-04-03 21:32:51 -0500370 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -0500371
Sage Weil7c272192011-08-03 09:58:09 -0700372 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
373 ret = ceph_osdc_start_request(osdc, req, false);
374 if (ret < 0)
375 goto out_pages;
376 ceph_osdc_put_request(req);
377 return nr_pages;
378
379out_pages:
David Zafman8884d532012-12-03 19:14:05 -0800380 ceph_unlock_page_vector(pages, nr_pages);
Sage Weil7c272192011-08-03 09:58:09 -0700381 ceph_release_page_vector(pages, nr_pages);
Sage Weil7c272192011-08-03 09:58:09 -0700382out:
383 ceph_osdc_put_request(req);
384 return ret;
Sage Weil1d3576f2009-10-06 11:31:09 -0700385}
386
Sage Weil7c272192011-08-03 09:58:09 -0700387
Sage Weil1d3576f2009-10-06 11:31:09 -0700388/*
389 * Read multiple pages. Leave pages we don't read + unlock in page_list;
390 * the caller (VM) cleans them up.
391 */
392static int ceph_readpages(struct file *file, struct address_space *mapping,
393 struct list_head *page_list, unsigned nr_pages)
394{
Al Viro496ad9a2013-01-23 17:07:38 -0500395 struct inode *inode = file_inode(file);
Sage Weil0d66a482011-08-04 08:21:30 -0700396 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700397 int rc = 0;
Sage Weil0d66a482011-08-04 08:21:30 -0700398 int max = 0;
Sage Weil1d3576f2009-10-06 11:31:09 -0700399
Yan, Zheng83701242014-11-14 22:36:18 +0800400 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
401 return -EINVAL;
402
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400403 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
404 &nr_pages);
405
406 if (rc == 0)
407 goto out;
408
Sage Weil0d66a482011-08-04 08:21:30 -0700409 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
410 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
411 >> PAGE_SHIFT;
412
Alex Elder2794a822013-02-14 12:16:43 -0600413 dout("readpages %p file %p nr_pages %d max %d\n", inode,
414 file, nr_pages,
Sage Weil0d66a482011-08-04 08:21:30 -0700415 max);
Sage Weil7c272192011-08-03 09:58:09 -0700416 while (!list_empty(page_list)) {
Sage Weil0d66a482011-08-04 08:21:30 -0700417 rc = start_read(inode, page_list, max);
Sage Weil7c272192011-08-03 09:58:09 -0700418 if (rc < 0)
419 goto out;
420 BUG_ON(rc == 0);
Sage Weil1d3576f2009-10-06 11:31:09 -0700421 }
Sage Weil1d3576f2009-10-06 11:31:09 -0700422out:
Milosz Tanski76be7782013-08-21 17:30:27 -0400423 ceph_fscache_readpages_cancel(inode, page_list);
424
Sage Weil7c272192011-08-03 09:58:09 -0700425 dout("readpages %p file %p ret %d\n", inode, file, rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700426 return rc;
427}
428
429/*
430 * Get ref for the oldest snapc for an inode with dirty data... that is, the
431 * only snap context we are allowed to write back.
Sage Weil1d3576f2009-10-06 11:31:09 -0700432 */
Sage Weil6298a332010-03-31 22:01:38 -0700433static struct ceph_snap_context *get_oldest_context(struct inode *inode,
434 u64 *snap_size)
Sage Weil1d3576f2009-10-06 11:31:09 -0700435{
436 struct ceph_inode_info *ci = ceph_inode(inode);
437 struct ceph_snap_context *snapc = NULL;
438 struct ceph_cap_snap *capsnap = NULL;
439
Sage Weilbe655592011-11-30 09:47:09 -0800440 spin_lock(&ci->i_ceph_lock);
Sage Weil1d3576f2009-10-06 11:31:09 -0700441 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
442 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
443 capsnap->context, capsnap->dirty_pages);
444 if (capsnap->dirty_pages) {
445 snapc = ceph_get_snap_context(capsnap->context);
446 if (snap_size)
447 *snap_size = capsnap->size;
448 break;
449 }
450 }
Sage Weil7d8cb262010-08-24 08:44:16 -0700451 if (!snapc && ci->i_wrbuffer_ref_head) {
Sage Weil80e755f2010-03-31 21:52:10 -0700452 snapc = ceph_get_snap_context(ci->i_head_snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700453 dout(" head snapc %p has %d dirty pages\n",
454 snapc, ci->i_wrbuffer_ref_head);
455 }
Sage Weilbe655592011-11-30 09:47:09 -0800456 spin_unlock(&ci->i_ceph_lock);
Sage Weil1d3576f2009-10-06 11:31:09 -0700457 return snapc;
458}
459
460/*
461 * Write a single page, but leave the page locked.
462 *
463 * If we get a write error, set the page error bit, but still adjust the
464 * dirty page accounting (i.e., page is no longer dirty).
465 */
466static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
467{
468 struct inode *inode;
469 struct ceph_inode_info *ci;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700470 struct ceph_fs_client *fsc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700471 struct ceph_osd_client *osdc;
Sage Weil6298a332010-03-31 22:01:38 -0700472 struct ceph_snap_context *snapc, *oldest;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800473 loff_t page_off = page_offset(page);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800474 long writeback_stat;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800475 u64 truncate_size, snap_size = 0;
476 u32 truncate_seq;
477 int err = 0, len = PAGE_CACHE_SIZE;
Sage Weil1d3576f2009-10-06 11:31:09 -0700478
479 dout("writepage %p idx %lu\n", page, page->index);
480
481 if (!page->mapping || !page->mapping->host) {
482 dout("writepage %p - no mapping\n", page);
483 return -EFAULT;
484 }
485 inode = page->mapping->host;
486 ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700487 fsc = ceph_inode_to_client(inode);
488 osdc = &fsc->client->osdc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700489
490 /* verify this is a writeable snap context */
Yan, Zheng61600ef2012-05-28 14:44:30 +0800491 snapc = page_snap_context(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700492 if (snapc == NULL) {
493 dout("writepage %p page %p not dirty?\n", inode, page);
494 goto out;
495 }
Sage Weil6298a332010-03-31 22:01:38 -0700496 oldest = get_oldest_context(inode, &snap_size);
497 if (snapc->seq > oldest->seq) {
Sage Weil1d3576f2009-10-06 11:31:09 -0700498 dout("writepage %p page %p snapc %p not writeable - noop\n",
Yan, Zheng61600ef2012-05-28 14:44:30 +0800499 inode, page, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700500 /* we should only noop if called by kswapd */
501 WARN_ON((current->flags & PF_MEMALLOC) == 0);
Sage Weil6298a332010-03-31 22:01:38 -0700502 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -0700503 goto out;
504 }
Sage Weil6298a332010-03-31 22:01:38 -0700505 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -0700506
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800507 spin_lock(&ci->i_ceph_lock);
508 truncate_seq = ci->i_truncate_seq;
509 truncate_size = ci->i_truncate_size;
510 if (!snap_size)
511 snap_size = i_size_read(inode);
512 spin_unlock(&ci->i_ceph_lock);
513
Sage Weil1d3576f2009-10-06 11:31:09 -0700514 /* is this a partial page at end of file? */
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800515 if (page_off >= snap_size) {
516 dout("%p page eof %llu\n", page, snap_size);
517 goto out;
518 }
519 if (snap_size < page_off + len)
520 len = snap_size - page_off;
Sage Weil1d3576f2009-10-06 11:31:09 -0700521
Sage Weilae00d4f2010-09-16 16:26:51 -0700522 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
523 inode, page, page->index, page_off, len, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700524
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700525 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800526 if (writeback_stat >
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700527 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
528 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800529
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400530 ceph_readpage_to_fscache(inode, page);
531
Sage Weil1d3576f2009-10-06 11:31:09 -0700532 set_page_writeback(page);
533 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
534 &ci->i_layout, snapc,
535 page_off, len,
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800536 truncate_seq, truncate_size,
Alex Elder24808822013-02-15 11:42:29 -0600537 &inode->i_mtime, &page, 1);
Sage Weil1d3576f2009-10-06 11:31:09 -0700538 if (err < 0) {
539 dout("writepage setting page/mapping error %d %p\n", err, page);
540 SetPageError(page);
541 mapping_set_error(&inode->i_data, err);
542 if (wbc)
543 wbc->pages_skipped++;
544 } else {
545 dout("writepage cleaned page %p\n", page);
546 err = 0; /* vfs expects us to return 0 */
547 }
548 page->private = 0;
549 ClearPagePrivate(page);
550 end_page_writeback(page);
551 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
Sage Weil6298a332010-03-31 22:01:38 -0700552 ceph_put_snap_context(snapc); /* page's reference */
Sage Weil1d3576f2009-10-06 11:31:09 -0700553out:
554 return err;
555}
556
557static int ceph_writepage(struct page *page, struct writeback_control *wbc)
558{
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800559 int err;
560 struct inode *inode = page->mapping->host;
561 BUG_ON(!inode);
Sage Weil70b666c2011-05-27 09:24:26 -0700562 ihold(inode);
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800563 err = writepage_nounlock(page, wbc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700564 unlock_page(page);
Yehuda Sadehdbd646a2009-12-16 14:51:06 -0800565 iput(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700566 return err;
567}
568
569
570/*
571 * lame release_pages helper. release_pages() isn't exported to
572 * modules.
573 */
574static void ceph_release_pages(struct page **pages, int num)
575{
576 struct pagevec pvec;
577 int i;
578
579 pagevec_init(&pvec, 0);
580 for (i = 0; i < num; i++) {
581 if (pagevec_add(&pvec, pages[i]) == 0)
582 pagevec_release(&pvec);
583 }
584 pagevec_release(&pvec);
585}
586
Sage Weil1d3576f2009-10-06 11:31:09 -0700587/*
588 * async writeback completion handler.
589 *
590 * If we get an error, set the mapping error bit, but not the individual
591 * page error bits.
592 */
593static void writepages_finish(struct ceph_osd_request *req,
594 struct ceph_msg *msg)
595{
596 struct inode *inode = req->r_inode;
Sage Weil1d3576f2009-10-06 11:31:09 -0700597 struct ceph_inode_info *ci = ceph_inode(inode);
Alex Elder87060c12013-04-03 01:28:58 -0500598 struct ceph_osd_data *osd_data;
Sage Weil1d3576f2009-10-06 11:31:09 -0700599 unsigned wrote;
Sage Weil1d3576f2009-10-06 11:31:09 -0700600 struct page *page;
Alex Eldere0c59482013-03-07 15:38:25 -0600601 int num_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700602 int i;
603 struct ceph_snap_context *snapc = req->r_snapc;
604 struct address_space *mapping = inode->i_mapping;
Sage Weil1b83bef2013-02-25 16:11:12 -0800605 int rc = req->r_result;
Alex Elder79528732013-04-03 21:32:51 -0500606 u64 bytes = req->r_ops[0].extent.length;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700607 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800608 long writeback_stat;
Sage Weil7ff899d2010-04-23 10:25:33 -0700609 unsigned issued = ceph_caps_issued(ci);
Sage Weil1d3576f2009-10-06 11:31:09 -0700610
Alex Elder406e2c92013-04-15 14:50:36 -0500611 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder87060c12013-04-03 01:28:58 -0500612 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
613 num_pages = calc_pages_for((u64)osd_data->alignment,
614 (u64)osd_data->length);
Sage Weil1d3576f2009-10-06 11:31:09 -0700615 if (rc >= 0) {
Sage Weil79788c62010-02-02 16:34:04 -0800616 /*
617 * Assume we wrote the pages we originally sent. The
618 * osd might reply with fewer pages if our writeback
619 * raced with a truncation and was adjusted at the osd,
620 * so don't believe the reply.
621 */
Alex Eldere0c59482013-03-07 15:38:25 -0600622 wrote = num_pages;
Sage Weil1d3576f2009-10-06 11:31:09 -0700623 } else {
624 wrote = 0;
625 mapping_set_error(mapping, rc);
626 }
627 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
628 inode, rc, bytes, wrote);
629
630 /* clean all pages */
Alex Eldere0c59482013-03-07 15:38:25 -0600631 for (i = 0; i < num_pages; i++) {
Alex Elder87060c12013-04-03 01:28:58 -0500632 page = osd_data->pages[i];
Sage Weil1d3576f2009-10-06 11:31:09 -0700633 BUG_ON(!page);
634 WARN_ON(!PageUptodate(page));
635
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800636 writeback_stat =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700637 atomic_long_dec_return(&fsc->writeback_count);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800638 if (writeback_stat <
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700639 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
640 clear_bdi_congested(&fsc->backing_dev_info,
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800641 BLK_RW_ASYNC);
642
Yan, Zheng61600ef2012-05-28 14:44:30 +0800643 ceph_put_snap_context(page_snap_context(page));
Sage Weil1d3576f2009-10-06 11:31:09 -0700644 page->private = 0;
645 ClearPagePrivate(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700646 dout("unlocking %d %p\n", i, page);
647 end_page_writeback(page);
Yehuda Sadehe63dc5c2010-02-19 00:07:01 +0000648
649 /*
650 * We lost the cache cap, need to truncate the page before
651 * it is unlocked, otherwise we'd truncate it later in the
652 * page truncation thread, possibly losing some data that
653 * raced its way in
654 */
Sage Weil29625072010-05-27 10:40:43 -0700655 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
Yehuda Sadehe63dc5c2010-02-19 00:07:01 +0000656 generic_error_remove_page(inode->i_mapping, page);
657
Sage Weil1d3576f2009-10-06 11:31:09 -0700658 unlock_page(page);
659 }
660 dout("%p wrote+cleaned %d pages\n", inode, wrote);
Alex Eldere0c59482013-03-07 15:38:25 -0600661 ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700662
Alex Elder87060c12013-04-03 01:28:58 -0500663 ceph_release_pages(osd_data->pages, num_pages);
664 if (osd_data->pages_from_pool)
665 mempool_free(osd_data->pages,
Cheng Renquan640ef792010-03-26 17:40:33 +0800666 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
Sage Weil1d3576f2009-10-06 11:31:09 -0700667 else
Alex Elder87060c12013-04-03 01:28:58 -0500668 kfree(osd_data->pages);
Sage Weil1d3576f2009-10-06 11:31:09 -0700669 ceph_osdc_put_request(req);
670}
671
Sage Weil1d3576f2009-10-06 11:31:09 -0700672/*
673 * initiate async writeback
674 */
675static int ceph_writepages_start(struct address_space *mapping,
676 struct writeback_control *wbc)
677{
678 struct inode *inode = mapping->host;
Sage Weil1d3576f2009-10-06 11:31:09 -0700679 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800680 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
681 struct ceph_vino vino = ceph_vino(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700682 pgoff_t index, start, end;
683 int range_whole = 0;
684 int should_loop = 1;
685 pgoff_t max_pages = 0, max_pages_ever = 0;
Sage Weil80e755f2010-03-31 21:52:10 -0700686 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
Sage Weil1d3576f2009-10-06 11:31:09 -0700687 struct pagevec pvec;
688 int done = 0;
689 int rc = 0;
690 unsigned wsize = 1 << inode->i_blkbits;
691 struct ceph_osd_request *req = NULL;
692 int do_sync;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800693 u64 truncate_size, snap_size;
694 u32 truncate_seq;
Sage Weil1d3576f2009-10-06 11:31:09 -0700695
696 /*
697 * Include a 'sync' in the OSD request if this is a data
698 * integrity write (e.g., O_SYNC write or fsync()), or if our
699 * cap is being revoked.
700 */
majianpengc62988e2013-06-19 15:12:06 +0800701 if ((wbc->sync_mode == WB_SYNC_ALL) ||
702 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
Sage Weil1d3576f2009-10-06 11:31:09 -0700703 do_sync = 1;
704 dout("writepages_start %p dosync=%d (mode=%s)\n",
705 inode, do_sync,
706 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
707 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
708
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700709 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
Fabian Frederickf3ae1b92014-06-06 14:35:37 -0700710 pr_warn("writepage_start %p on forced umount\n", inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700711 return -EIO; /* we're in a forced umount, don't write! */
712 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700713 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
714 wsize = fsc->mount_options->wsize;
Sage Weil1d3576f2009-10-06 11:31:09 -0700715 if (wsize < PAGE_CACHE_SIZE)
716 wsize = PAGE_CACHE_SIZE;
717 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
718
719 pagevec_init(&pvec, 0);
720
Sage Weil1d3576f2009-10-06 11:31:09 -0700721 /* where to start/end? */
722 if (wbc->range_cyclic) {
723 start = mapping->writeback_index; /* Start from prev offset */
724 end = -1;
725 dout(" cyclic, start at %lu\n", start);
726 } else {
727 start = wbc->range_start >> PAGE_CACHE_SHIFT;
728 end = wbc->range_end >> PAGE_CACHE_SHIFT;
729 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
730 range_whole = 1;
731 should_loop = 0;
732 dout(" not cyclic, %lu to %lu\n", start, end);
733 }
734 index = start;
735
736retry:
737 /* find oldest snap context with dirty data */
738 ceph_put_snap_context(snapc);
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800739 snap_size = 0;
Sage Weil1d3576f2009-10-06 11:31:09 -0700740 snapc = get_oldest_context(inode, &snap_size);
741 if (!snapc) {
742 /* hmm, why does writepages get called when there
743 is no dirty data? */
744 dout(" no snap context with dirty data?\n");
745 goto out;
746 }
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800747 if (snap_size == 0)
748 snap_size = i_size_read(inode);
Sage Weil1d3576f2009-10-06 11:31:09 -0700749 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
750 snapc, snapc->seq, snapc->num_snaps);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800751
752 spin_lock(&ci->i_ceph_lock);
753 truncate_seq = ci->i_truncate_seq;
754 truncate_size = ci->i_truncate_size;
755 if (!snap_size)
756 snap_size = i_size_read(inode);
757 spin_unlock(&ci->i_ceph_lock);
758
Sage Weil1d3576f2009-10-06 11:31:09 -0700759 if (last_snapc && snapc != last_snapc) {
760 /* if we switched to a newer snapc, restart our scan at the
761 * start of the original file range. */
762 dout(" snapc differs from last pass, restarting at %lu\n",
763 index);
764 index = start;
765 }
766 last_snapc = snapc;
767
768 while (!done && index <= end) {
769 unsigned i;
770 int first;
771 pgoff_t next;
772 int pvec_pages, locked_pages;
Alex Eldere5975c72013-03-14 14:09:05 -0500773 struct page **pages = NULL;
774 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
Sage Weil1d3576f2009-10-06 11:31:09 -0700775 struct page *page;
776 int want;
777 u64 offset, len;
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800778 long writeback_stat;
Sage Weil1d3576f2009-10-06 11:31:09 -0700779
780 next = 0;
781 locked_pages = 0;
782 max_pages = max_pages_ever;
783
784get_more_pages:
785 first = -1;
786 want = min(end - index,
787 min((pgoff_t)PAGEVEC_SIZE,
788 max_pages - (pgoff_t)locked_pages) - 1)
789 + 1;
790 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
791 PAGECACHE_TAG_DIRTY,
792 want);
793 dout("pagevec_lookup_tag got %d\n", pvec_pages);
794 if (!pvec_pages && !locked_pages)
795 break;
796 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
797 page = pvec.pages[i];
798 dout("? %p idx %lu\n", page, page->index);
799 if (locked_pages == 0)
800 lock_page(page); /* first page */
801 else if (!trylock_page(page))
802 break;
803
804 /* only dirty pages, or our accounting breaks */
805 if (unlikely(!PageDirty(page)) ||
806 unlikely(page->mapping != mapping)) {
807 dout("!dirty or !mapping %p\n", page);
808 unlock_page(page);
809 break;
810 }
811 if (!wbc->range_cyclic && page->index > end) {
812 dout("end of range %p\n", page);
813 done = 1;
814 unlock_page(page);
815 break;
816 }
817 if (next && (page->index != next)) {
818 dout("not consecutive %p\n", page);
819 unlock_page(page);
820 break;
821 }
822 if (wbc->sync_mode != WB_SYNC_NONE) {
823 dout("waiting on writeback %p\n", page);
824 wait_on_page_writeback(page);
825 }
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800826 if (page_offset(page) >= snap_size) {
827 dout("%p page eof %llu\n", page, snap_size);
Sage Weil1d3576f2009-10-06 11:31:09 -0700828 done = 1;
829 unlock_page(page);
830 break;
831 }
832 if (PageWriteback(page)) {
833 dout("%p under writeback\n", page);
834 unlock_page(page);
835 break;
836 }
837
838 /* only if matching snap context */
Yan, Zheng61600ef2012-05-28 14:44:30 +0800839 pgsnapc = page_snap_context(page);
Sage Weil80e755f2010-03-31 21:52:10 -0700840 if (pgsnapc->seq > snapc->seq) {
841 dout("page snapc %p %lld > oldest %p %lld\n",
842 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
Sage Weil1d3576f2009-10-06 11:31:09 -0700843 unlock_page(page);
844 if (!locked_pages)
845 continue; /* keep looking for snap */
846 break;
847 }
848
849 if (!clear_page_dirty_for_io(page)) {
850 dout("%p !clear_page_dirty_for_io\n", page);
851 unlock_page(page);
852 break;
853 }
854
Alex Eldere5975c72013-03-14 14:09:05 -0500855 /*
856 * We have something to write. If this is
857 * the first locked page this time through,
858 * allocate an osd request and a page array
859 * that it will use.
860 */
Sage Weil1d3576f2009-10-06 11:31:09 -0700861 if (locked_pages == 0) {
Alex Eldere5975c72013-03-14 14:09:05 -0500862 BUG_ON(pages);
Sage Weil1d3576f2009-10-06 11:31:09 -0700863 /* prepare async write request */
Alex Eldere5975c72013-03-14 14:09:05 -0500864 offset = (u64)page_offset(page);
Sage Weil1d3576f2009-10-06 11:31:09 -0700865 len = wsize;
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800866 req = ceph_osdc_new_request(&fsc->client->osdc,
867 &ci->i_layout, vino,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800868 offset, &len, 0,
869 do_sync ? 2 : 1,
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800870 CEPH_OSD_OP_WRITE,
871 CEPH_OSD_FLAG_WRITE |
872 CEPH_OSD_FLAG_ONDISK,
873 snapc, truncate_seq,
874 truncate_size, true);
Sage Weil68162822012-09-24 21:01:02 -0700875 if (IS_ERR(req)) {
876 rc = PTR_ERR(req);
Henry C Chang8c718972011-05-03 09:45:16 +0000877 unlock_page(page);
878 break;
879 }
880
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800881 if (do_sync)
882 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
883
Alex Elder88486952013-03-14 14:09:05 -0500884 req->r_callback = writepages_finish;
885 req->r_inode = inode;
886
887 max_pages = calc_pages_for(0, (u64)len);
Yan, Zhengfc2744a2013-05-31 16:48:29 +0800888 pages = kmalloc(max_pages * sizeof (*pages),
889 GFP_NOFS);
Alex Elder88486952013-03-14 14:09:05 -0500890 if (!pages) {
891 pool = fsc->wb_pagevec_pool;
Alex Elder88486952013-03-14 14:09:05 -0500892 pages = mempool_alloc(pool, GFP_NOFS);
Alex Eldere5975c72013-03-14 14:09:05 -0500893 BUG_ON(!pages);
Alex Elder88486952013-03-14 14:09:05 -0500894 }
Sage Weil1d3576f2009-10-06 11:31:09 -0700895 }
896
897 /* note position of first page in pvec */
898 if (first < 0)
899 first = i;
900 dout("%p will write page %p idx %lu\n",
901 inode, page, page->index);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800902
Sage Weil213c99e2010-08-03 10:25:11 -0700903 writeback_stat =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700904 atomic_long_inc_return(&fsc->writeback_count);
Sage Weil213c99e2010-08-03 10:25:11 -0700905 if (writeback_stat > CONGESTION_ON_THRESH(
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700906 fsc->mount_options->congestion_kb)) {
907 set_bdi_congested(&fsc->backing_dev_info,
Sage Weil213c99e2010-08-03 10:25:11 -0700908 BLK_RW_ASYNC);
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800909 }
910
Sage Weil1d3576f2009-10-06 11:31:09 -0700911 set_page_writeback(page);
Alex Eldere5975c72013-03-14 14:09:05 -0500912 pages[locked_pages] = page;
Sage Weil1d3576f2009-10-06 11:31:09 -0700913 locked_pages++;
914 next = page->index + 1;
915 }
916
917 /* did we get anything? */
918 if (!locked_pages)
919 goto release_pvec_pages;
920 if (i) {
921 int j;
922 BUG_ON(!locked_pages || first < 0);
923
924 if (pvec_pages && i == pvec_pages &&
925 locked_pages < max_pages) {
926 dout("reached end pvec, trying for more\n");
927 pagevec_reinit(&pvec);
928 goto get_more_pages;
929 }
930
931 /* shift unused pages over in the pvec... we
932 * will need to release them below. */
933 for (j = i; j < pvec_pages; j++) {
934 dout(" pvec leftover page %p\n",
935 pvec.pages[j]);
936 pvec.pages[j-i+first] = pvec.pages[j];
937 }
938 pvec.nr -= i-first;
939 }
940
Alex Eldere5975c72013-03-14 14:09:05 -0500941 /* Format the osd request message and submit the write */
942
943 offset = page_offset(pages[0]);
Yan, Zheng1ac0fc82013-04-12 21:45:42 +0800944 len = min(snap_size - offset,
Sage Weil1d3576f2009-10-06 11:31:09 -0700945 (u64)locked_pages << PAGE_CACHE_SHIFT);
946 dout("writepages got %d pages at %llu~%llu\n",
947 locked_pages, offset, len);
948
Alex Elder406e2c92013-04-15 14:50:36 -0500949 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500950 !!pool, false);
Alex Eldere5975c72013-03-14 14:09:05 -0500951
952 pages = NULL; /* request message now owns the pages array */
953 pool = NULL;
954
955 /* Update the write op length in case we changed it */
956
Alex Elderc99d2d42013-04-05 01:27:11 -0500957 osd_req_op_extent_update(req, 0, len);
Alex Eldere5975c72013-03-14 14:09:05 -0500958
959 vino = ceph_vino(inode);
Alex Elder79528732013-04-03 21:32:51 -0500960 ceph_osdc_build_request(req, offset, snapc, vino.snap,
961 &inode->i_mtime);
Sage Weil1d3576f2009-10-06 11:31:09 -0700962
Sage Weil9d6fcb02011-05-12 15:48:16 -0700963 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
964 BUG_ON(rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700965 req = NULL;
966
967 /* continue? */
968 index = next;
969 wbc->nr_to_write -= locked_pages;
970 if (wbc->nr_to_write <= 0)
971 done = 1;
972
973release_pvec_pages:
974 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
975 pvec.nr ? pvec.pages[0] : NULL);
976 pagevec_release(&pvec);
977
978 if (locked_pages && !done)
979 goto retry;
980 }
981
982 if (should_loop && !done) {
983 /* more to do; loop back to beginning of file */
984 dout("writepages looping back to beginning of file\n");
985 should_loop = 0;
986 index = 0;
987 goto retry;
988 }
989
990 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
991 mapping->writeback_index = index;
992
993out:
994 if (req)
995 ceph_osdc_put_request(req);
Sage Weil1d3576f2009-10-06 11:31:09 -0700996 ceph_put_snap_context(snapc);
997 dout("writepages done, rc = %d\n", rc);
Sage Weil1d3576f2009-10-06 11:31:09 -0700998 return rc;
999}
1000
1001
1002
1003/*
1004 * See if a given @snapc is either writeable, or already written.
1005 */
1006static int context_is_writeable_or_written(struct inode *inode,
1007 struct ceph_snap_context *snapc)
1008{
1009 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
Sage Weil6298a332010-03-31 22:01:38 -07001010 int ret = !oldest || snapc->seq <= oldest->seq;
1011
1012 ceph_put_snap_context(oldest);
1013 return ret;
Sage Weil1d3576f2009-10-06 11:31:09 -07001014}
1015
1016/*
1017 * We are only allowed to write into/dirty the page if the page is
1018 * clean, or already dirty within the same snap context.
Sage Weil8f883c22010-03-19 13:27:53 -07001019 *
1020 * called with page locked.
1021 * return success with page locked,
1022 * or any failure (incl -EAGAIN) with page unlocked.
Sage Weil1d3576f2009-10-06 11:31:09 -07001023 */
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001024static int ceph_update_writeable_page(struct file *file,
1025 loff_t pos, unsigned len,
1026 struct page *page)
Sage Weil1d3576f2009-10-06 11:31:09 -07001027{
Al Viro496ad9a2013-01-23 17:07:38 -05001028 struct inode *inode = file_inode(file);
Sage Weil1d3576f2009-10-06 11:31:09 -07001029 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001030 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weil1d3576f2009-10-06 11:31:09 -07001031 loff_t page_off = pos & PAGE_CACHE_MASK;
1032 int pos_in_page = pos & ~PAGE_CACHE_MASK;
1033 int end_in_page = pos_in_page + len;
1034 loff_t i_size;
Sage Weil1d3576f2009-10-06 11:31:09 -07001035 int r;
Sage Weil80e755f2010-03-31 21:52:10 -07001036 struct ceph_snap_context *snapc, *oldest;
Sage Weil1d3576f2009-10-06 11:31:09 -07001037
Sage Weil1d3576f2009-10-06 11:31:09 -07001038retry_locked:
1039 /* writepages currently holds page lock, but if we change that later, */
1040 wait_on_page_writeback(page);
1041
1042 /* check snap context */
1043 BUG_ON(!ci->i_snap_realm);
1044 down_read(&mdsc->snap_rwsem);
1045 BUG_ON(!ci->i_snap_realm->cached_context);
Yan, Zheng61600ef2012-05-28 14:44:30 +08001046 snapc = page_snap_context(page);
Sage Weil80e755f2010-03-31 21:52:10 -07001047 if (snapc && snapc != ci->i_head_snapc) {
Sage Weil1d3576f2009-10-06 11:31:09 -07001048 /*
1049 * this page is already dirty in another (older) snap
1050 * context! is it writeable now?
1051 */
Sage Weil80e755f2010-03-31 21:52:10 -07001052 oldest = get_oldest_context(inode, NULL);
Sage Weil1d3576f2009-10-06 11:31:09 -07001053 up_read(&mdsc->snap_rwsem);
1054
Sage Weil80e755f2010-03-31 21:52:10 -07001055 if (snapc->seq > oldest->seq) {
Sage Weil6298a332010-03-31 22:01:38 -07001056 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -07001057 dout(" page %p snapc %p not current or oldest\n",
Sage Weil6298a332010-03-31 22:01:38 -07001058 page, snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -07001059 /*
1060 * queue for writeback, and wait for snapc to
1061 * be writeable or written
1062 */
Sage Weil6298a332010-03-31 22:01:38 -07001063 snapc = ceph_get_snap_context(snapc);
Sage Weil1d3576f2009-10-06 11:31:09 -07001064 unlock_page(page);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001065 ceph_queue_writeback(inode);
Sage Weil8f883c22010-03-19 13:27:53 -07001066 r = wait_event_interruptible(ci->i_cap_wq,
Sage Weil1d3576f2009-10-06 11:31:09 -07001067 context_is_writeable_or_written(inode, snapc));
1068 ceph_put_snap_context(snapc);
Sage Weil8f883c22010-03-19 13:27:53 -07001069 if (r == -ERESTARTSYS)
1070 return r;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001071 return -EAGAIN;
Sage Weil1d3576f2009-10-06 11:31:09 -07001072 }
Sage Weil6298a332010-03-31 22:01:38 -07001073 ceph_put_snap_context(oldest);
Sage Weil1d3576f2009-10-06 11:31:09 -07001074
1075 /* yay, writeable, do it now (without dropping page lock) */
1076 dout(" page %p snapc %p not current, but oldest\n",
1077 page, snapc);
1078 if (!clear_page_dirty_for_io(page))
1079 goto retry_locked;
1080 r = writepage_nounlock(page, NULL);
1081 if (r < 0)
1082 goto fail_nosnap;
1083 goto retry_locked;
1084 }
1085
1086 if (PageUptodate(page)) {
1087 dout(" page %p already uptodate\n", page);
1088 return 0;
1089 }
1090
1091 /* full page? */
1092 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1093 return 0;
1094
1095 /* past end of file? */
1096 i_size = inode->i_size; /* caller holds i_mutex */
1097
Sage Weil1d3576f2009-10-06 11:31:09 -07001098 if (page_off >= i_size ||
1099 (pos_in_page == 0 && (pos+len) >= i_size &&
1100 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1101 dout(" zeroing %p 0 - %d and %d - %d\n",
1102 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1103 zero_user_segments(page,
1104 0, pos_in_page,
1105 end_in_page, PAGE_CACHE_SIZE);
1106 return 0;
1107 }
1108
1109 /* we need to read it. */
1110 up_read(&mdsc->snap_rwsem);
1111 r = readpage_nounlock(file, page);
1112 if (r < 0)
1113 goto fail_nosnap;
1114 goto retry_locked;
Sage Weil1d3576f2009-10-06 11:31:09 -07001115fail_nosnap:
1116 unlock_page(page);
1117 return r;
1118}
1119
1120/*
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001121 * We are only allowed to write into/dirty the page if the page is
1122 * clean, or already dirty within the same snap context.
1123 */
1124static int ceph_write_begin(struct file *file, struct address_space *mapping,
1125 loff_t pos, unsigned len, unsigned flags,
1126 struct page **pagep, void **fsdata)
1127{
Al Viro496ad9a2013-01-23 17:07:38 -05001128 struct inode *inode = file_inode(file);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001129 struct page *page;
1130 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Sage Weil7971bd92013-05-01 21:15:58 -07001131 int r;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001132
1133 do {
Sage Weil8f883c22010-03-19 13:27:53 -07001134 /* get a page */
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001135 page = grab_cache_page_write_begin(mapping, index, 0);
Sage Weil7971bd92013-05-01 21:15:58 -07001136 if (!page)
1137 return -ENOMEM;
1138 *pagep = page;
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001139
1140 dout("write_begin file %p inode %p page %p %d~%d\n", file,
Sage Weil213c99e2010-08-03 10:25:11 -07001141 inode, page, (int)pos, (int)len);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001142
1143 r = ceph_update_writeable_page(file, pos, len, page);
1144 } while (r == -EAGAIN);
1145
1146 return r;
1147}
1148
1149/*
Sage Weil1d3576f2009-10-06 11:31:09 -07001150 * we don't do anything in here that simple_write_end doesn't do
1151 * except adjust dirty page accounting and drop read lock on
1152 * mdsc->snap_rwsem.
1153 */
1154static int ceph_write_end(struct file *file, struct address_space *mapping,
1155 loff_t pos, unsigned len, unsigned copied,
1156 struct page *page, void *fsdata)
1157{
Al Viro496ad9a2013-01-23 17:07:38 -05001158 struct inode *inode = file_inode(file);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001159 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1160 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil1d3576f2009-10-06 11:31:09 -07001161 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1162 int check_cap = 0;
1163
1164 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1165 inode, page, (int)pos, (int)copied, (int)len);
1166
1167 /* zero the stale part of the page if we did a short copy */
1168 if (copied < len)
1169 zero_user_segment(page, from+copied, len);
1170
1171 /* did file size increase? */
1172 /* (no need for i_size_read(); we caller holds i_mutex */
1173 if (pos+copied > inode->i_size)
1174 check_cap = ceph_inode_set_size(inode, pos+copied);
1175
1176 if (!PageUptodate(page))
1177 SetPageUptodate(page);
1178
1179 set_page_dirty(page);
1180
1181 unlock_page(page);
1182 up_read(&mdsc->snap_rwsem);
1183 page_cache_release(page);
1184
1185 if (check_cap)
1186 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1187
1188 return copied;
1189}
1190
1191/*
1192 * we set .direct_IO to indicate direct io is supported, but since we
1193 * intercept O_DIRECT reads and writes early, this function should
1194 * never get called.
1195 */
1196static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05001197 struct iov_iter *iter,
1198 loff_t pos)
Sage Weil1d3576f2009-10-06 11:31:09 -07001199{
1200 WARN_ON(1);
1201 return -EINVAL;
1202}
1203
1204const struct address_space_operations ceph_aops = {
1205 .readpage = ceph_readpage,
1206 .readpages = ceph_readpages,
1207 .writepage = ceph_writepage,
1208 .writepages = ceph_writepages_start,
1209 .write_begin = ceph_write_begin,
1210 .write_end = ceph_write_end,
1211 .set_page_dirty = ceph_set_page_dirty,
1212 .invalidatepage = ceph_invalidatepage,
1213 .releasepage = ceph_releasepage,
1214 .direct_IO = ceph_direct_io,
1215};
1216
1217
1218/*
1219 * vm ops
1220 */
Yan, Zheng61f68812013-11-28 14:28:14 +08001221static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1222{
1223 struct inode *inode = file_inode(vma->vm_file);
1224 struct ceph_inode_info *ci = ceph_inode(inode);
1225 struct ceph_file_info *fi = vma->vm_file->private_data;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001226 struct page *pinned_page = NULL;
Yan, Zheng61f68812013-11-28 14:28:14 +08001227 loff_t off = vmf->pgoff << PAGE_CACHE_SHIFT;
1228 int want, got, ret;
1229
1230 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001231 inode, ceph_vinop(inode), off, (size_t)PAGE_CACHE_SIZE);
Yan, Zheng61f68812013-11-28 14:28:14 +08001232 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1233 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1234 else
1235 want = CEPH_CAP_FILE_CACHE;
1236 while (1) {
1237 got = 0;
Yan, Zheng83701242014-11-14 22:36:18 +08001238 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want,
1239 -1, &got, &pinned_page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001240 if (ret == 0)
1241 break;
1242 if (ret != -ERESTARTSYS) {
1243 WARN_ON(1);
1244 return VM_FAULT_SIGBUS;
1245 }
1246 }
1247 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001248 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got));
Yan, Zheng61f68812013-11-28 14:28:14 +08001249
Yan, Zheng83701242014-11-14 22:36:18 +08001250 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1251 ci->i_inline_version == CEPH_INLINE_NONE)
1252 ret = filemap_fault(vma, vmf);
1253 else
1254 ret = -EAGAIN;
Yan, Zheng61f68812013-11-28 14:28:14 +08001255
1256 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
Ilya Dryomov37b52fe2014-01-28 18:29:29 +02001257 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got), ret);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001258 if (pinned_page)
1259 page_cache_release(pinned_page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001260 ceph_put_cap_refs(ci, got);
1261
Yan, Zheng83701242014-11-14 22:36:18 +08001262 if (ret != -EAGAIN)
1263 return ret;
1264
1265 /* read inline data */
1266 if (off >= PAGE_CACHE_SIZE) {
1267 /* does not support inline data > PAGE_SIZE */
1268 ret = VM_FAULT_SIGBUS;
1269 } else {
1270 int ret1;
1271 struct address_space *mapping = inode->i_mapping;
1272 struct page *page = find_or_create_page(mapping, 0,
1273 mapping_gfp_mask(mapping) &
1274 ~__GFP_FS);
1275 if (!page) {
1276 ret = VM_FAULT_OOM;
1277 goto out;
1278 }
1279 ret1 = __ceph_do_getattr(inode, page,
1280 CEPH_STAT_CAP_INLINE_DATA, true);
1281 if (ret1 < 0 || off >= i_size_read(inode)) {
1282 unlock_page(page);
1283 page_cache_release(page);
1284 ret = VM_FAULT_SIGBUS;
1285 goto out;
1286 }
1287 if (ret1 < PAGE_CACHE_SIZE)
1288 zero_user_segment(page, ret1, PAGE_CACHE_SIZE);
1289 else
1290 flush_dcache_page(page);
1291 SetPageUptodate(page);
1292 vmf->page = page;
1293 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1294 }
1295out:
1296 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1297 inode, off, (size_t)PAGE_CACHE_SIZE, ret);
Yan, Zheng61f68812013-11-28 14:28:14 +08001298 return ret;
1299}
Sage Weil1d3576f2009-10-06 11:31:09 -07001300
1301/*
1302 * Reuse write_begin here for simplicity.
1303 */
1304static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1305{
Al Viro496ad9a2013-01-23 17:07:38 -05001306 struct inode *inode = file_inode(vma->vm_file);
Yan, Zheng61f68812013-11-28 14:28:14 +08001307 struct ceph_inode_info *ci = ceph_inode(inode);
1308 struct ceph_file_info *fi = vma->vm_file->private_data;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001309 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng61f68812013-11-28 14:28:14 +08001310 struct page *page = vmf->page;
Alex Elder6285bc22012-10-02 10:25:51 -05001311 loff_t off = page_offset(page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001312 loff_t size = i_size_read(inode);
1313 size_t len;
1314 int want, got, ret;
Sage Weil1d3576f2009-10-06 11:31:09 -07001315
Sage Weil1d3576f2009-10-06 11:31:09 -07001316 if (off + PAGE_CACHE_SIZE <= size)
1317 len = PAGE_CACHE_SIZE;
1318 else
1319 len = size & ~PAGE_CACHE_MASK;
1320
Yan, Zheng61f68812013-11-28 14:28:14 +08001321 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1322 inode, ceph_vinop(inode), off, len, size);
1323 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1324 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1325 else
1326 want = CEPH_CAP_FILE_BUFFER;
1327 while (1) {
1328 got = 0;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001329 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1330 &got, NULL);
Yan, Zheng61f68812013-11-28 14:28:14 +08001331 if (ret == 0)
1332 break;
1333 if (ret != -ERESTARTSYS) {
1334 WARN_ON(1);
1335 return VM_FAULT_SIGBUS;
1336 }
1337 }
1338 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1339 inode, off, len, ceph_cap_string(got));
1340
1341 /* Update time before taking page lock */
1342 file_update_time(vma->vm_file);
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001343
1344 lock_page(page);
1345
1346 ret = VM_FAULT_NOPAGE;
1347 if ((off > size) ||
1348 (page->mapping != inode->i_mapping))
1349 goto out;
1350
1351 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1352 if (ret == 0) {
1353 /* success. we'll keep the page locked. */
Sage Weil1d3576f2009-10-06 11:31:09 -07001354 set_page_dirty(page);
1355 up_read(&mdsc->snap_rwsem);
Sage Weil1d3576f2009-10-06 11:31:09 -07001356 ret = VM_FAULT_LOCKED;
1357 } else {
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001358 if (ret == -ENOMEM)
1359 ret = VM_FAULT_OOM;
1360 else
1361 ret = VM_FAULT_SIGBUS;
Sage Weil1d3576f2009-10-06 11:31:09 -07001362 }
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001363out:
Yan, Zheng61f68812013-11-28 14:28:14 +08001364 if (ret != VM_FAULT_LOCKED) {
Yehuda Sadeh4af6b222010-02-09 11:02:51 -08001365 unlock_page(page);
Yan, Zheng61f68812013-11-28 14:28:14 +08001366 } else {
1367 int dirty;
1368 spin_lock(&ci->i_ceph_lock);
1369 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1370 spin_unlock(&ci->i_ceph_lock);
1371 if (dirty)
1372 __mark_inode_dirty(inode, dirty);
1373 }
1374
1375 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1376 inode, off, len, ceph_cap_string(got), ret);
1377 ceph_put_cap_refs(ci, got);
1378
Sage Weil1d3576f2009-10-06 11:31:09 -07001379 return ret;
1380}
1381
Yan, Zheng31c542a2014-11-14 21:41:55 +08001382void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1383 char *data, size_t len)
1384{
1385 struct address_space *mapping = inode->i_mapping;
1386 struct page *page;
1387
1388 if (locked_page) {
1389 page = locked_page;
1390 } else {
1391 if (i_size_read(inode) == 0)
1392 return;
1393 page = find_or_create_page(mapping, 0,
1394 mapping_gfp_mask(mapping) & ~__GFP_FS);
1395 if (!page)
1396 return;
1397 if (PageUptodate(page)) {
1398 unlock_page(page);
1399 page_cache_release(page);
1400 return;
1401 }
1402 }
1403
1404 dout("fill_inline_data %p %llx.%llx len %lu locked_page %p\n",
1405 inode, ceph_vinop(inode), len, locked_page);
1406
1407 if (len > 0) {
1408 void *kaddr = kmap_atomic(page);
1409 memcpy(kaddr, data, len);
1410 kunmap_atomic(kaddr);
1411 }
1412
1413 if (page != locked_page) {
1414 if (len < PAGE_CACHE_SIZE)
1415 zero_user_segment(page, len, PAGE_CACHE_SIZE);
1416 else
1417 flush_dcache_page(page);
1418
1419 SetPageUptodate(page);
1420 unlock_page(page);
1421 page_cache_release(page);
1422 }
1423}
1424
Sage Weil1d3576f2009-10-06 11:31:09 -07001425static struct vm_operations_struct ceph_vmops = {
Yan, Zheng61f68812013-11-28 14:28:14 +08001426 .fault = ceph_filemap_fault,
Sage Weil1d3576f2009-10-06 11:31:09 -07001427 .page_mkwrite = ceph_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001428 .remap_pages = generic_file_remap_pages,
Sage Weil1d3576f2009-10-06 11:31:09 -07001429};
1430
1431int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1432{
1433 struct address_space *mapping = file->f_mapping;
1434
1435 if (!mapping->a_ops->readpage)
1436 return -ENOEXEC;
1437 file_accessed(file);
1438 vma->vm_ops = &ceph_vmops;
Sage Weil1d3576f2009-10-06 11:31:09 -07001439 return 0;
1440}