]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/ceph/addr.c
ceph: fix mds cluster availability check
[mirror_ubuntu-zesty-kernel.git] / fs / ceph / addr.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
1d3576fd
SW
2
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 */
5a0e3ad6 8#include <linux/slab.h>
1d3576fd
SW
9#include <linux/pagevec.h>
10#include <linux/task_io_accounting_ops.h>
11
12#include "super.h"
3d14c5d2 13#include "mds_client.h"
99ccbd22 14#include "cache.h"
3d14c5d2 15#include <linux/ceph/osd_client.h>
1d3576fd
SW
16
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
25985edc 28 * count dirty pages on the inode. In the absence of snapshots,
1d3576fd
SW
29 * 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
2baba250
YS
53#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
61600ef8
YZ
58static 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}
1d3576fd
SW
64
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;
1d3576fd 74 struct ceph_snap_context *snapc;
7d6e1f54 75 int ret;
1d3576fd
SW
76
77 if (unlikely(!mapping))
78 return !TestSetPageDirty(page);
79
7d6e1f54 80 if (PageDirty(page)) {
1d3576fd
SW
81 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
82 mapping->host, page, page->index);
7d6e1f54 83 BUG_ON(!PagePrivate(page));
1d3576fd
SW
84 return 0;
85 }
86
87 inode = mapping->host;
88 ci = ceph_inode(inode);
89
1d3576fd 90 /* dirty the head */
be655596 91 spin_lock(&ci->i_ceph_lock);
5dda377c
YZ
92 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
93 if (__ceph_have_pending_cap_snap(ci)) {
94 struct ceph_cap_snap *capsnap =
95 list_last_entry(&ci->i_cap_snaps,
96 struct ceph_cap_snap,
97 ci_item);
98 snapc = ceph_get_snap_context(capsnap->context);
99 capsnap->dirty_pages++;
100 } else {
101 BUG_ON(!ci->i_head_snapc);
102 snapc = ceph_get_snap_context(ci->i_head_snapc);
103 ++ci->i_wrbuffer_ref_head;
104 }
1d3576fd 105 if (ci->i_wrbuffer_ref == 0)
0444d76a 106 ihold(inode);
1d3576fd
SW
107 ++ci->i_wrbuffer_ref;
108 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
109 "snapc %p seq %lld (%d snaps)\n",
110 mapping->host, page, page->index,
111 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
112 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
113 snapc, snapc->seq, snapc->num_snaps);
be655596 114 spin_unlock(&ci->i_ceph_lock);
1d3576fd 115
7d6e1f54
SZ
116 /*
117 * Reference snap context in page->private. Also set
118 * PagePrivate so that we get invalidatepage callback.
119 */
120 BUG_ON(PagePrivate(page));
121 page->private = (unsigned long)snapc;
122 SetPagePrivate(page);
1d3576fd 123
7d6e1f54
SZ
124 ret = __set_page_dirty_nobuffers(page);
125 WARN_ON(!PageLocked(page));
126 WARN_ON(!page->mapping);
1d3576fd 127
7d6e1f54 128 return ret;
1d3576fd
SW
129}
130
131/*
132 * If we are truncating the full page (i.e. offset == 0), adjust the
133 * dirty page counters appropriately. Only called if there is private
134 * data on the page.
135 */
d47992f8
LC
136static void ceph_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length)
1d3576fd 138{
4ce1e9ad 139 struct inode *inode;
1d3576fd 140 struct ceph_inode_info *ci;
61600ef8 141 struct ceph_snap_context *snapc = page_snap_context(page);
1d3576fd 142
4ce1e9ad 143 inode = page->mapping->host;
b150f5c1
MT
144 ci = ceph_inode(inode);
145
09cbfeaf 146 if (offset != 0 || length != PAGE_SIZE) {
b150f5c1
MT
147 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
148 inode, page, page->index, offset, length);
149 return;
150 }
4ce1e9ad 151
99ccbd22
MT
152 ceph_invalidate_fscache_page(inode, page);
153
154 if (!PagePrivate(page))
155 return;
156
1d3576fd
SW
157 /*
158 * We can get non-dirty pages here due to races between
159 * set_page_dirty and truncate_complete_page; just spit out a
160 * warning, in case we end up with accounting problems later.
161 */
162 if (!PageDirty(page))
163 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
164
b150f5c1 165 ClearPageChecked(page);
1d3576fd 166
b150f5c1
MT
167 dout("%p invalidatepage %p idx %lu full dirty page\n",
168 inode, page, page->index);
169
170 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
171 ceph_put_snap_context(snapc);
172 page->private = 0;
173 ClearPagePrivate(page);
1d3576fd
SW
174}
175
1d3576fd
SW
176static int ceph_releasepage(struct page *page, gfp_t g)
177{
e55f1a18
N
178 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
179 page, page->index, PageDirty(page) ? "" : "not ");
99ccbd22
MT
180
181 /* Can we release the page from the cache? */
182 if (!ceph_release_fscache_page(page, g))
183 return 0;
184
185 return !PagePrivate(page);
1d3576fd
SW
186}
187
188/*
189 * read a single page, without unlocking it.
190 */
191static int readpage_nounlock(struct file *filp, struct page *page)
192{
496ad9aa 193 struct inode *inode = file_inode(filp);
1d3576fd 194 struct ceph_inode_info *ci = ceph_inode(inode);
99ccbd22 195 struct ceph_osd_client *osdc =
3d14c5d2 196 &ceph_inode_to_client(inode)->client->osdc;
1d3576fd 197 int err = 0;
83701246 198 u64 off = page_offset(page);
09cbfeaf 199 u64 len = PAGE_SIZE;
1d3576fd 200
83701246 201 if (off >= i_size_read(inode)) {
09cbfeaf 202 zero_user_segment(page, 0, PAGE_SIZE);
83701246
YZ
203 SetPageUptodate(page);
204 return 0;
205 }
99ccbd22 206
fcc02d2a
YZ
207 if (ci->i_inline_version != CEPH_INLINE_NONE) {
208 /*
209 * Uptodate inline data should have been added
210 * into page cache while getting Fcr caps.
211 */
212 if (off == 0)
213 return -EINVAL;
09cbfeaf 214 zero_user_segment(page, 0, PAGE_SIZE);
fcc02d2a
YZ
215 SetPageUptodate(page);
216 return 0;
217 }
83701246
YZ
218
219 err = ceph_readpage_from_fscache(inode, page);
99ccbd22
MT
220 if (err == 0)
221 goto out;
222
1d3576fd
SW
223 dout("readpage inode %p file %p page %p index %lu\n",
224 inode, filp, page, page->index);
225 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
83701246 226 off, &len,
1d3576fd 227 ci->i_truncate_seq, ci->i_truncate_size,
b7495fc2 228 &page, 1, 0);
1d3576fd
SW
229 if (err == -ENOENT)
230 err = 0;
231 if (err < 0) {
232 SetPageError(page);
18302805 233 ceph_fscache_readpage_cancel(inode, page);
1d3576fd 234 goto out;
1d3576fd 235 }
09cbfeaf 236 if (err < PAGE_SIZE)
23cd573b 237 /* zero fill remainder of page */
09cbfeaf 238 zero_user_segment(page, err, PAGE_SIZE);
23cd573b
ZZ
239 else
240 flush_dcache_page(page);
1d3576fd 241
23cd573b
ZZ
242 SetPageUptodate(page);
243 ceph_readpage_to_fscache(inode, page);
99ccbd22 244
1d3576fd
SW
245out:
246 return err < 0 ? err : 0;
247}
248
249static int ceph_readpage(struct file *filp, struct page *page)
250{
251 int r = readpage_nounlock(filp, page);
252 unlock_page(page);
253 return r;
254}
255
256/*
7c272194 257 * Finish an async read(ahead) op.
1d3576fd 258 */
85e084fe 259static void finish_read(struct ceph_osd_request *req)
1d3576fd 260{
7c272194 261 struct inode *inode = req->r_inode;
87060c10 262 struct ceph_osd_data *osd_data;
85e084fe
ID
263 int rc = req->r_result <= 0 ? req->r_result : 0;
264 int bytes = req->r_result >= 0 ? req->r_result : 0;
e0c59487 265 int num_pages;
7c272194 266 int i;
1d3576fd 267
7c272194
SW
268 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
269
270 /* unlock all pages, zeroing any data we didn't read */
406e2c9f 271 osd_data = osd_req_op_extent_osd_data(req, 0);
87060c10
AE
272 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
273 num_pages = calc_pages_for((u64)osd_data->alignment,
274 (u64)osd_data->length);
e0c59487 275 for (i = 0; i < num_pages; i++) {
87060c10 276 struct page *page = osd_data->pages[i];
7c272194 277
368e3585
YZ
278 if (rc < 0 && rc != -ENOENT) {
279 ceph_fscache_readpage_cancel(inode, page);
f36132a7 280 goto unlock;
368e3585 281 }
09cbfeaf 282 if (bytes < (int)PAGE_SIZE) {
7c272194
SW
283 /* zero (remainder of) page */
284 int s = bytes < 0 ? 0 : bytes;
09cbfeaf 285 zero_user_segment(page, s, PAGE_SIZE);
1d3576fd 286 }
7c272194
SW
287 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
288 page->index);
289 flush_dcache_page(page);
290 SetPageUptodate(page);
99ccbd22 291 ceph_readpage_to_fscache(inode, page);
f36132a7 292unlock:
7c272194 293 unlock_page(page);
09cbfeaf
KS
294 put_page(page);
295 bytes -= PAGE_SIZE;
1d3576fd 296 }
87060c10 297 kfree(osd_data->pages);
1d3576fd
SW
298}
299
300/*
7c272194
SW
301 * start an async read(ahead) operation. return nr_pages we submitted
302 * a read for on success, or negative error code.
1d3576fd 303 */
0d66a487 304static int start_read(struct inode *inode, struct list_head *page_list, int max)
1d3576fd 305{
3d14c5d2
YS
306 struct ceph_osd_client *osdc =
307 &ceph_inode_to_client(inode)->client->osdc;
7c272194
SW
308 struct ceph_inode_info *ci = ceph_inode(inode);
309 struct page *page = list_entry(page_list->prev, struct page, lru);
acead002 310 struct ceph_vino vino;
7c272194
SW
311 struct ceph_osd_request *req;
312 u64 off;
1d3576fd 313 u64 len;
7c272194
SW
314 int i;
315 struct page **pages;
316 pgoff_t next_index;
317 int nr_pages = 0;
2b1ac852
YZ
318 int got = 0;
319 int ret = 0;
320
321 if (!current->journal_info) {
322 /* caller of readpages does not hold buffer and read caps
323 * (fadvise, madvise and readahead cases) */
324 int want = CEPH_CAP_FILE_CACHE;
325 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
326 if (ret < 0) {
327 dout("start_read %p, error getting cap\n", inode);
328 } else if (!(got & want)) {
329 dout("start_read %p, no cache cap\n", inode);
330 ret = 0;
331 }
332 if (ret <= 0) {
333 if (got)
334 ceph_put_cap_refs(ci, got);
335 while (!list_empty(page_list)) {
336 page = list_entry(page_list->prev,
337 struct page, lru);
338 list_del(&page->lru);
339 put_page(page);
340 }
341 return ret;
342 }
343 }
1d3576fd 344
6285bc23 345 off = (u64) page_offset(page);
1d3576fd 346
7c272194
SW
347 /* count pages */
348 next_index = page->index;
349 list_for_each_entry_reverse(page, page_list, lru) {
350 if (page->index != next_index)
351 break;
352 nr_pages++;
353 next_index++;
0d66a487
SW
354 if (max && nr_pages == max)
355 break;
7c272194 356 }
09cbfeaf 357 len = nr_pages << PAGE_SHIFT;
7c272194
SW
358 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
359 off, len);
acead002
AE
360 vino = ceph_vino(inode);
361 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
715e4cd4 362 0, 1, CEPH_OSD_OP_READ,
acead002 363 CEPH_OSD_FLAG_READ, NULL,
7c272194 364 ci->i_truncate_seq, ci->i_truncate_size,
acead002 365 false);
2b1ac852
YZ
366 if (IS_ERR(req)) {
367 ret = PTR_ERR(req);
368 goto out;
369 }
1d3576fd 370
7c272194 371 /* build page vector */
cf7b7e14 372 nr_pages = calc_pages_for(0, len);
687265e5 373 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
2b1ac852
YZ
374 if (!pages) {
375 ret = -ENOMEM;
376 goto out_put;
377 }
7c272194
SW
378 for (i = 0; i < nr_pages; ++i) {
379 page = list_entry(page_list->prev, struct page, lru);
380 BUG_ON(PageLocked(page));
1d3576fd 381 list_del(&page->lru);
99ccbd22 382
7c272194
SW
383 dout("start_read %p adding %p idx %lu\n", inode, page,
384 page->index);
385 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
687265e5 386 GFP_KERNEL)) {
d4d3aa38 387 ceph_fscache_uncache_page(inode, page);
09cbfeaf 388 put_page(page);
7c272194 389 dout("start_read %p add_to_page_cache failed %p\n",
1d3576fd 390 inode, page);
7c272194 391 nr_pages = i;
1afe4785
YZ
392 if (nr_pages > 0) {
393 len = nr_pages << PAGE_SHIFT;
394 break;
395 }
7c272194 396 goto out_pages;
1d3576fd 397 }
7c272194 398 pages[i] = page;
1d3576fd 399 }
406e2c9f 400 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
7c272194
SW
401 req->r_callback = finish_read;
402 req->r_inode = inode;
403
404 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
405 ret = ceph_osdc_start_request(osdc, req, false);
406 if (ret < 0)
407 goto out_pages;
408 ceph_osdc_put_request(req);
2b1ac852
YZ
409
410 /* After adding locked pages to page cache, the inode holds cache cap.
411 * So we can drop our cap refs. */
412 if (got)
413 ceph_put_cap_refs(ci, got);
414
7c272194
SW
415 return nr_pages;
416
417out_pages:
1afe4785
YZ
418 for (i = 0; i < nr_pages; ++i) {
419 ceph_fscache_readpage_cancel(inode, pages[i]);
420 unlock_page(pages[i]);
421 }
422 ceph_put_page_vector(pages, nr_pages, false);
2b1ac852 423out_put:
7c272194 424 ceph_osdc_put_request(req);
2b1ac852
YZ
425out:
426 if (got)
427 ceph_put_cap_refs(ci, got);
7c272194
SW
428 return ret;
429}
1d3576fd 430
7c272194
SW
431
432/*
433 * Read multiple pages. Leave pages we don't read + unlock in page_list;
434 * the caller (VM) cleans them up.
435 */
436static int ceph_readpages(struct file *file, struct address_space *mapping,
437 struct list_head *page_list, unsigned nr_pages)
438{
496ad9aa 439 struct inode *inode = file_inode(file);
0d66a487 440 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
7c272194 441 int rc = 0;
0d66a487
SW
442 int max = 0;
443
83701246
YZ
444 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
445 return -EINVAL;
446
99ccbd22
MT
447 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
448 &nr_pages);
449
450 if (rc == 0)
451 goto out;
452
09cbfeaf
KS
453 if (fsc->mount_options->rsize >= PAGE_SIZE)
454 max = (fsc->mount_options->rsize + PAGE_SIZE - 1)
0d66a487 455 >> PAGE_SHIFT;
7c272194 456
2794a82a
AE
457 dout("readpages %p file %p nr_pages %d max %d\n", inode,
458 file, nr_pages,
0d66a487 459 max);
7c272194 460 while (!list_empty(page_list)) {
0d66a487 461 rc = start_read(inode, page_list, max);
7c272194
SW
462 if (rc < 0)
463 goto out;
7c272194 464 }
1d3576fd 465out:
76be778b
MT
466 ceph_fscache_readpages_cancel(inode, page_list);
467
7c272194 468 dout("readpages %p file %p ret %d\n", inode, file, rc);
1d3576fd
SW
469 return rc;
470}
471
472/*
473 * Get ref for the oldest snapc for an inode with dirty data... that is, the
474 * only snap context we are allowed to write back.
1d3576fd 475 */
6298a337 476static struct ceph_snap_context *get_oldest_context(struct inode *inode,
5f743e45
YZ
477 loff_t *snap_size,
478 u64 *truncate_size,
479 u32 *truncate_seq)
1d3576fd
SW
480{
481 struct ceph_inode_info *ci = ceph_inode(inode);
482 struct ceph_snap_context *snapc = NULL;
483 struct ceph_cap_snap *capsnap = NULL;
484
be655596 485 spin_lock(&ci->i_ceph_lock);
1d3576fd
SW
486 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
487 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
488 capsnap->context, capsnap->dirty_pages);
489 if (capsnap->dirty_pages) {
490 snapc = ceph_get_snap_context(capsnap->context);
491 if (snap_size)
492 *snap_size = capsnap->size;
5f743e45
YZ
493 if (truncate_size)
494 *truncate_size = capsnap->truncate_size;
495 if (truncate_seq)
496 *truncate_seq = capsnap->truncate_seq;
1d3576fd
SW
497 break;
498 }
499 }
7d8cb26d 500 if (!snapc && ci->i_wrbuffer_ref_head) {
80e755fe 501 snapc = ceph_get_snap_context(ci->i_head_snapc);
1d3576fd
SW
502 dout(" head snapc %p has %d dirty pages\n",
503 snapc, ci->i_wrbuffer_ref_head);
5f743e45
YZ
504 if (truncate_size)
505 *truncate_size = capsnap->truncate_size;
506 if (truncate_seq)
507 *truncate_seq = capsnap->truncate_seq;
1d3576fd 508 }
be655596 509 spin_unlock(&ci->i_ceph_lock);
1d3576fd
SW
510 return snapc;
511}
512
513/*
514 * Write a single page, but leave the page locked.
515 *
516 * If we get a write error, set the page error bit, but still adjust the
517 * dirty page accounting (i.e., page is no longer dirty).
518 */
519static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
520{
521 struct inode *inode;
522 struct ceph_inode_info *ci;
3d14c5d2 523 struct ceph_fs_client *fsc;
1d3576fd 524 struct ceph_osd_client *osdc;
6298a337 525 struct ceph_snap_context *snapc, *oldest;
fc2744aa 526 loff_t page_off = page_offset(page);
e1966b49 527 loff_t snap_size = -1;
2baba250 528 long writeback_stat;
e1966b49 529 u64 truncate_size;
fc2744aa 530 u32 truncate_seq;
09cbfeaf 531 int err = 0, len = PAGE_SIZE;
1d3576fd
SW
532
533 dout("writepage %p idx %lu\n", page, page->index);
534
535 if (!page->mapping || !page->mapping->host) {
536 dout("writepage %p - no mapping\n", page);
537 return -EFAULT;
538 }
539 inode = page->mapping->host;
540 ci = ceph_inode(inode);
3d14c5d2
YS
541 fsc = ceph_inode_to_client(inode);
542 osdc = &fsc->client->osdc;
1d3576fd
SW
543
544 /* verify this is a writeable snap context */
61600ef8 545 snapc = page_snap_context(page);
1d3576fd
SW
546 if (snapc == NULL) {
547 dout("writepage %p page %p not dirty?\n", inode, page);
548 goto out;
549 }
5f743e45
YZ
550 oldest = get_oldest_context(inode, &snap_size,
551 &truncate_size, &truncate_seq);
6298a337 552 if (snapc->seq > oldest->seq) {
1d3576fd 553 dout("writepage %p page %p snapc %p not writeable - noop\n",
61600ef8 554 inode, page, snapc);
1d3576fd
SW
555 /* we should only noop if called by kswapd */
556 WARN_ON((current->flags & PF_MEMALLOC) == 0);
6298a337 557 ceph_put_snap_context(oldest);
1d3576fd
SW
558 goto out;
559 }
6298a337 560 ceph_put_snap_context(oldest);
1d3576fd 561
e1966b49 562 if (snap_size == -1)
fc2744aa 563 snap_size = i_size_read(inode);
fc2744aa 564
1d3576fd 565 /* is this a partial page at end of file? */
fc2744aa
YZ
566 if (page_off >= snap_size) {
567 dout("%p page eof %llu\n", page, snap_size);
568 goto out;
569 }
570 if (snap_size < page_off + len)
571 len = snap_size - page_off;
1d3576fd 572
ae00d4f3
SW
573 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
574 inode, page, page->index, page_off, len, snapc);
1d3576fd 575
3d14c5d2 576 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
2baba250 577 if (writeback_stat >
3d14c5d2
YS
578 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
579 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
2baba250 580
1d3576fd
SW
581 set_page_writeback(page);
582 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
583 &ci->i_layout, snapc,
584 page_off, len,
fc2744aa 585 truncate_seq, truncate_size,
24808826 586 &inode->i_mtime, &page, 1);
1d3576fd 587 if (err < 0) {
ad15ec06
YZ
588 struct writeback_control tmp_wbc;
589 if (!wbc)
590 wbc = &tmp_wbc;
591 if (err == -ERESTARTSYS) {
592 /* killed by SIGKILL */
593 dout("writepage interrupted page %p\n", page);
594 redirty_page_for_writepage(wbc, page);
595 end_page_writeback(page);
596 goto out;
597 }
598 dout("writepage setting page/mapping error %d %p\n",
599 err, page);
1d3576fd
SW
600 SetPageError(page);
601 mapping_set_error(&inode->i_data, err);
ad15ec06 602 wbc->pages_skipped++;
1d3576fd
SW
603 } else {
604 dout("writepage cleaned page %p\n", page);
605 err = 0; /* vfs expects us to return 0 */
606 }
607 page->private = 0;
608 ClearPagePrivate(page);
609 end_page_writeback(page);
610 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
6298a337 611 ceph_put_snap_context(snapc); /* page's reference */
1d3576fd
SW
612out:
613 return err;
614}
615
616static int ceph_writepage(struct page *page, struct writeback_control *wbc)
617{
dbd646a8
YS
618 int err;
619 struct inode *inode = page->mapping->host;
620 BUG_ON(!inode);
70b666c3 621 ihold(inode);
dbd646a8 622 err = writepage_nounlock(page, wbc);
ad15ec06
YZ
623 if (err == -ERESTARTSYS) {
624 /* direct memory reclaimer was killed by SIGKILL. return 0
625 * to prevent caller from setting mapping/page error */
626 err = 0;
627 }
1d3576fd 628 unlock_page(page);
dbd646a8 629 iput(inode);
1d3576fd
SW
630 return err;
631}
632
1d3576fd
SW
633/*
634 * lame release_pages helper. release_pages() isn't exported to
635 * modules.
636 */
637static void ceph_release_pages(struct page **pages, int num)
638{
639 struct pagevec pvec;
640 int i;
641
642 pagevec_init(&pvec, 0);
643 for (i = 0; i < num; i++) {
644 if (pagevec_add(&pvec, pages[i]) == 0)
645 pagevec_release(&pvec);
646 }
647 pagevec_release(&pvec);
648}
649
1d3576fd
SW
650/*
651 * async writeback completion handler.
652 *
653 * If we get an error, set the mapping error bit, but not the individual
654 * page error bits.
655 */
85e084fe 656static void writepages_finish(struct ceph_osd_request *req)
1d3576fd
SW
657{
658 struct inode *inode = req->r_inode;
1d3576fd 659 struct ceph_inode_info *ci = ceph_inode(inode);
87060c10 660 struct ceph_osd_data *osd_data;
1d3576fd 661 struct page *page;
5b64640c
YZ
662 int num_pages, total_pages = 0;
663 int i, j;
664 int rc = req->r_result;
1d3576fd
SW
665 struct ceph_snap_context *snapc = req->r_snapc;
666 struct address_space *mapping = inode->i_mapping;
3d14c5d2 667 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
5b64640c 668 bool remove_page;
1d3576fd 669
5b64640c
YZ
670 dout("writepages_finish %p rc %d\n", inode, rc);
671 if (rc < 0)
1d3576fd 672 mapping_set_error(mapping, rc);
5b64640c
YZ
673
674 /*
675 * We lost the cache cap, need to truncate the page before
676 * it is unlocked, otherwise we'd truncate it later in the
677 * page truncation thread, possibly losing some data that
678 * raced its way in
679 */
680 remove_page = !(ceph_caps_issued(ci) &
681 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
1d3576fd
SW
682
683 /* clean all pages */
5b64640c
YZ
684 for (i = 0; i < req->r_num_ops; i++) {
685 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
686 break;
e63dc5c7 687
5b64640c
YZ
688 osd_data = osd_req_op_extent_osd_data(req, i);
689 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
690 num_pages = calc_pages_for((u64)osd_data->alignment,
691 (u64)osd_data->length);
692 total_pages += num_pages;
693 for (j = 0; j < num_pages; j++) {
694 page = osd_data->pages[j];
695 BUG_ON(!page);
696 WARN_ON(!PageUptodate(page));
697
698 if (atomic_long_dec_return(&fsc->writeback_count) <
699 CONGESTION_OFF_THRESH(
700 fsc->mount_options->congestion_kb))
701 clear_bdi_congested(&fsc->backing_dev_info,
702 BLK_RW_ASYNC);
703
b109eec6
YZ
704 if (rc < 0)
705 SetPageError(page);
706
5b64640c
YZ
707 ceph_put_snap_context(page_snap_context(page));
708 page->private = 0;
709 ClearPagePrivate(page);
710 dout("unlocking %p\n", page);
711 end_page_writeback(page);
712
713 if (remove_page)
714 generic_error_remove_page(inode->i_mapping,
715 page);
716
717 unlock_page(page);
718 }
719 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
720 inode, osd_data->length, rc >= 0 ? num_pages : 0);
e63dc5c7 721
5b64640c 722 ceph_release_pages(osd_data->pages, num_pages);
1d3576fd 723 }
1d3576fd 724
5b64640c
YZ
725 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
726
727 osd_data = osd_req_op_extent_osd_data(req, 0);
87060c10
AE
728 if (osd_data->pages_from_pool)
729 mempool_free(osd_data->pages,
640ef79d 730 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
1d3576fd 731 else
87060c10 732 kfree(osd_data->pages);
1d3576fd
SW
733 ceph_osdc_put_request(req);
734}
735
1d3576fd
SW
736/*
737 * initiate async writeback
738 */
739static int ceph_writepages_start(struct address_space *mapping,
740 struct writeback_control *wbc)
741{
742 struct inode *inode = mapping->host;
1d3576fd 743 struct ceph_inode_info *ci = ceph_inode(inode);
fc2744aa
YZ
744 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
745 struct ceph_vino vino = ceph_vino(inode);
1d3576fd
SW
746 pgoff_t index, start, end;
747 int range_whole = 0;
748 int should_loop = 1;
749 pgoff_t max_pages = 0, max_pages_ever = 0;
80e755fe 750 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
1d3576fd
SW
751 struct pagevec pvec;
752 int done = 0;
753 int rc = 0;
754 unsigned wsize = 1 << inode->i_blkbits;
755 struct ceph_osd_request *req = NULL;
021b77be 756 int do_sync = 0;
e1966b49
YZ
757 loff_t snap_size, i_size;
758 u64 truncate_size;
fc2744aa 759 u32 truncate_seq;
1d3576fd
SW
760
761 /*
762 * Include a 'sync' in the OSD request if this is a data
763 * integrity write (e.g., O_SYNC write or fsync()), or if our
764 * cap is being revoked.
765 */
c62988ec 766 if ((wbc->sync_mode == WB_SYNC_ALL) ||
767 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
1d3576fd
SW
768 do_sync = 1;
769 dout("writepages_start %p dosync=%d (mode=%s)\n",
770 inode, do_sync,
771 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
772 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
773
48fec5d0 774 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
6c93df5d
YZ
775 if (ci->i_wrbuffer_ref > 0) {
776 pr_warn_ratelimited(
777 "writepage_start %p %lld forced umount\n",
778 inode, ceph_ino(inode));
779 }
a341d4df 780 mapping_set_error(mapping, -EIO);
1d3576fd
SW
781 return -EIO; /* we're in a forced umount, don't write! */
782 }
3d14c5d2
YS
783 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
784 wsize = fsc->mount_options->wsize;
09cbfeaf
KS
785 if (wsize < PAGE_SIZE)
786 wsize = PAGE_SIZE;
787 max_pages_ever = wsize >> PAGE_SHIFT;
1d3576fd
SW
788
789 pagevec_init(&pvec, 0);
790
1d3576fd
SW
791 /* where to start/end? */
792 if (wbc->range_cyclic) {
793 start = mapping->writeback_index; /* Start from prev offset */
794 end = -1;
795 dout(" cyclic, start at %lu\n", start);
796 } else {
09cbfeaf
KS
797 start = wbc->range_start >> PAGE_SHIFT;
798 end = wbc->range_end >> PAGE_SHIFT;
1d3576fd
SW
799 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
800 range_whole = 1;
801 should_loop = 0;
802 dout(" not cyclic, %lu to %lu\n", start, end);
803 }
804 index = start;
805
806retry:
807 /* find oldest snap context with dirty data */
808 ceph_put_snap_context(snapc);
e1966b49 809 snap_size = -1;
5f743e45
YZ
810 snapc = get_oldest_context(inode, &snap_size,
811 &truncate_size, &truncate_seq);
1d3576fd
SW
812 if (!snapc) {
813 /* hmm, why does writepages get called when there
814 is no dirty data? */
815 dout(" no snap context with dirty data?\n");
816 goto out;
817 }
818 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
819 snapc, snapc->seq, snapc->num_snaps);
fc2744aa 820
e1966b49 821 i_size = i_size_read(inode);
fc2744aa 822
1d3576fd
SW
823 if (last_snapc && snapc != last_snapc) {
824 /* if we switched to a newer snapc, restart our scan at the
825 * start of the original file range. */
826 dout(" snapc differs from last pass, restarting at %lu\n",
827 index);
828 index = start;
829 }
830 last_snapc = snapc;
831
832 while (!done && index <= end) {
833 unsigned i;
834 int first;
5b64640c
YZ
835 pgoff_t strip_unit_end = 0;
836 int num_ops = 0, op_idx;
837 int pvec_pages, locked_pages = 0;
838 struct page **pages = NULL, **data_pages;
e5975c7c 839 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
1d3576fd
SW
840 struct page *page;
841 int want;
5b64640c 842 u64 offset = 0, len = 0;
1d3576fd 843
1d3576fd
SW
844 max_pages = max_pages_ever;
845
846get_more_pages:
847 first = -1;
848 want = min(end - index,
849 min((pgoff_t)PAGEVEC_SIZE,
850 max_pages - (pgoff_t)locked_pages) - 1)
851 + 1;
852 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
853 PAGECACHE_TAG_DIRTY,
854 want);
855 dout("pagevec_lookup_tag got %d\n", pvec_pages);
856 if (!pvec_pages && !locked_pages)
857 break;
858 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
859 page = pvec.pages[i];
860 dout("? %p idx %lu\n", page, page->index);
861 if (locked_pages == 0)
862 lock_page(page); /* first page */
863 else if (!trylock_page(page))
864 break;
865
866 /* only dirty pages, or our accounting breaks */
867 if (unlikely(!PageDirty(page)) ||
868 unlikely(page->mapping != mapping)) {
869 dout("!dirty or !mapping %p\n", page);
870 unlock_page(page);
871 break;
872 }
873 if (!wbc->range_cyclic && page->index > end) {
874 dout("end of range %p\n", page);
875 done = 1;
876 unlock_page(page);
877 break;
878 }
5b64640c
YZ
879 if (strip_unit_end && (page->index > strip_unit_end)) {
880 dout("end of strip unit %p\n", page);
1d3576fd
SW
881 unlock_page(page);
882 break;
883 }
884 if (wbc->sync_mode != WB_SYNC_NONE) {
885 dout("waiting on writeback %p\n", page);
886 wait_on_page_writeback(page);
887 }
e1966b49
YZ
888 if (page_offset(page) >=
889 (snap_size == -1 ? i_size : snap_size)) {
890 dout("%p page eof %llu\n", page,
891 (snap_size == -1 ? i_size : snap_size));
1d3576fd
SW
892 done = 1;
893 unlock_page(page);
894 break;
895 }
896 if (PageWriteback(page)) {
897 dout("%p under writeback\n", page);
898 unlock_page(page);
899 break;
900 }
901
902 /* only if matching snap context */
61600ef8 903 pgsnapc = page_snap_context(page);
80e755fe
SW
904 if (pgsnapc->seq > snapc->seq) {
905 dout("page snapc %p %lld > oldest %p %lld\n",
906 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
1d3576fd
SW
907 unlock_page(page);
908 if (!locked_pages)
909 continue; /* keep looking for snap */
910 break;
911 }
912
913 if (!clear_page_dirty_for_io(page)) {
914 dout("%p !clear_page_dirty_for_io\n", page);
915 unlock_page(page);
916 break;
917 }
918
e5975c7c
AE
919 /*
920 * We have something to write. If this is
921 * the first locked page this time through,
5b64640c
YZ
922 * calculate max possinle write size and
923 * allocate a page array
e5975c7c 924 */
1d3576fd 925 if (locked_pages == 0) {
5b64640c
YZ
926 u64 objnum;
927 u64 objoff;
928
1d3576fd 929 /* prepare async write request */
e5975c7c 930 offset = (u64)page_offset(page);
1d3576fd 931 len = wsize;
5b64640c
YZ
932
933 rc = ceph_calc_file_object_mapping(&ci->i_layout,
934 offset, len,
935 &objnum, &objoff,
936 &len);
937 if (rc < 0) {
8c71897b
HC
938 unlock_page(page);
939 break;
940 }
941
5b64640c
YZ
942 num_ops = 1 + do_sync;
943 strip_unit_end = page->index +
09cbfeaf 944 ((len - 1) >> PAGE_SHIFT);
88486957 945
5b64640c 946 BUG_ON(pages);
88486957 947 max_pages = calc_pages_for(0, (u64)len);
fc2744aa
YZ
948 pages = kmalloc(max_pages * sizeof (*pages),
949 GFP_NOFS);
88486957
AE
950 if (!pages) {
951 pool = fsc->wb_pagevec_pool;
88486957 952 pages = mempool_alloc(pool, GFP_NOFS);
e5975c7c 953 BUG_ON(!pages);
88486957 954 }
5b64640c
YZ
955
956 len = 0;
957 } else if (page->index !=
09cbfeaf 958 (offset + len) >> PAGE_SHIFT) {
5b64640c
YZ
959 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
960 CEPH_OSD_MAX_OPS)) {
961 redirty_page_for_writepage(wbc, page);
962 unlock_page(page);
963 break;
964 }
965
966 num_ops++;
967 offset = (u64)page_offset(page);
968 len = 0;
1d3576fd
SW
969 }
970
971 /* note position of first page in pvec */
972 if (first < 0)
973 first = i;
974 dout("%p will write page %p idx %lu\n",
975 inode, page, page->index);
2baba250 976
5b64640c
YZ
977 if (atomic_long_inc_return(&fsc->writeback_count) >
978 CONGESTION_ON_THRESH(
3d14c5d2
YS
979 fsc->mount_options->congestion_kb)) {
980 set_bdi_congested(&fsc->backing_dev_info,
213c99ee 981 BLK_RW_ASYNC);
2baba250
YS
982 }
983
e5975c7c 984 pages[locked_pages] = page;
1d3576fd 985 locked_pages++;
09cbfeaf 986 len += PAGE_SIZE;
1d3576fd
SW
987 }
988
989 /* did we get anything? */
990 if (!locked_pages)
991 goto release_pvec_pages;
992 if (i) {
993 int j;
994 BUG_ON(!locked_pages || first < 0);
995
996 if (pvec_pages && i == pvec_pages &&
997 locked_pages < max_pages) {
998 dout("reached end pvec, trying for more\n");
999 pagevec_reinit(&pvec);
1000 goto get_more_pages;
1001 }
1002
1003 /* shift unused pages over in the pvec... we
1004 * will need to release them below. */
1005 for (j = i; j < pvec_pages; j++) {
5b64640c 1006 dout(" pvec leftover page %p\n", pvec.pages[j]);
1d3576fd
SW
1007 pvec.pages[j-i+first] = pvec.pages[j];
1008 }
1009 pvec.nr -= i-first;
1010 }
1011
5b64640c 1012new_request:
e5975c7c 1013 offset = page_offset(pages[0]);
5b64640c
YZ
1014 len = wsize;
1015
1016 req = ceph_osdc_new_request(&fsc->client->osdc,
1017 &ci->i_layout, vino,
1018 offset, &len, 0, num_ops,
1019 CEPH_OSD_OP_WRITE,
1020 CEPH_OSD_FLAG_WRITE |
1021 CEPH_OSD_FLAG_ONDISK,
1022 snapc, truncate_seq,
1023 truncate_size, false);
1024 if (IS_ERR(req)) {
1025 req = ceph_osdc_new_request(&fsc->client->osdc,
1026 &ci->i_layout, vino,
1027 offset, &len, 0,
1028 min(num_ops,
1029 CEPH_OSD_SLAB_OPS),
1030 CEPH_OSD_OP_WRITE,
1031 CEPH_OSD_FLAG_WRITE |
1032 CEPH_OSD_FLAG_ONDISK,
1033 snapc, truncate_seq,
1034 truncate_size, true);
1035 BUG_ON(IS_ERR(req));
e1966b49 1036 }
5b64640c 1037 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
09cbfeaf 1038 PAGE_SIZE - offset);
5b64640c
YZ
1039
1040 req->r_callback = writepages_finish;
1041 req->r_inode = inode;
1d3576fd 1042
5b64640c
YZ
1043 /* Format the osd request message and submit the write */
1044 len = 0;
1045 data_pages = pages;
1046 op_idx = 0;
1047 for (i = 0; i < locked_pages; i++) {
1048 u64 cur_offset = page_offset(pages[i]);
1049 if (offset + len != cur_offset) {
1050 if (op_idx + do_sync + 1 == req->r_num_ops)
1051 break;
1052 osd_req_op_extent_dup_last(req, op_idx,
1053 cur_offset - offset);
1054 dout("writepages got pages at %llu~%llu\n",
1055 offset, len);
1056 osd_req_op_extent_osd_data_pages(req, op_idx,
1057 data_pages, len, 0,
a4ce40a9 1058 !!pool, false);
5b64640c 1059 osd_req_op_extent_update(req, op_idx, len);
e5975c7c 1060
5b64640c
YZ
1061 len = 0;
1062 offset = cur_offset;
1063 data_pages = pages + i;
1064 op_idx++;
1065 }
1066
1067 set_page_writeback(pages[i]);
09cbfeaf 1068 len += PAGE_SIZE;
5b64640c
YZ
1069 }
1070
1071 if (snap_size != -1) {
1072 len = min(len, snap_size - offset);
1073 } else if (i == locked_pages) {
1074 /* writepages_finish() clears writeback pages
1075 * according to the data length, so make sure
1076 * data length covers all locked pages */
09cbfeaf 1077 u64 min_len = len + 1 - PAGE_SIZE;
5b64640c
YZ
1078 len = min(len, (u64)i_size_read(inode) - offset);
1079 len = max(len, min_len);
1080 }
1081 dout("writepages got pages at %llu~%llu\n", offset, len);
e5975c7c 1082
5b64640c
YZ
1083 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1084 0, !!pool, false);
1085 osd_req_op_extent_update(req, op_idx, len);
e5975c7c 1086
5b64640c
YZ
1087 if (do_sync) {
1088 op_idx++;
1089 osd_req_op_init(req, op_idx, CEPH_OSD_OP_STARTSYNC, 0);
1090 }
1091 BUG_ON(op_idx + 1 != req->r_num_ops);
1092
1093 pool = NULL;
1094 if (i < locked_pages) {
1095 BUG_ON(num_ops <= req->r_num_ops);
1096 num_ops -= req->r_num_ops;
1097 num_ops += do_sync;
1098 locked_pages -= i;
1099
1100 /* allocate new pages array for next request */
1101 data_pages = pages;
1102 pages = kmalloc(locked_pages * sizeof (*pages),
1103 GFP_NOFS);
1104 if (!pages) {
1105 pool = fsc->wb_pagevec_pool;
1106 pages = mempool_alloc(pool, GFP_NOFS);
1107 BUG_ON(!pages);
1108 }
1109 memcpy(pages, data_pages + i,
1110 locked_pages * sizeof(*pages));
1111 memset(data_pages + i, 0,
1112 locked_pages * sizeof(*pages));
1113 } else {
1114 BUG_ON(num_ops != req->r_num_ops);
1115 index = pages[i - 1]->index + 1;
1116 /* request message now owns the pages array */
1117 pages = NULL;
1118 }
e5975c7c 1119
bb873b53 1120 req->r_mtime = inode->i_mtime;
9d6fcb08
SW
1121 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1122 BUG_ON(rc);
1d3576fd
SW
1123 req = NULL;
1124
5b64640c
YZ
1125 wbc->nr_to_write -= i;
1126 if (pages)
1127 goto new_request;
1128
1d3576fd
SW
1129 if (wbc->nr_to_write <= 0)
1130 done = 1;
1131
1132release_pvec_pages:
1133 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1134 pvec.nr ? pvec.pages[0] : NULL);
1135 pagevec_release(&pvec);
1136
1137 if (locked_pages && !done)
1138 goto retry;
1139 }
1140
1141 if (should_loop && !done) {
1142 /* more to do; loop back to beginning of file */
1143 dout("writepages looping back to beginning of file\n");
1144 should_loop = 0;
1145 index = 0;
1146 goto retry;
1147 }
1148
1149 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1150 mapping->writeback_index = index;
1151
1152out:
3ed97d63 1153 ceph_osdc_put_request(req);
1d3576fd
SW
1154 ceph_put_snap_context(snapc);
1155 dout("writepages done, rc = %d\n", rc);
1d3576fd
SW
1156 return rc;
1157}
1158
1159
1160
1161/*
1162 * See if a given @snapc is either writeable, or already written.
1163 */
1164static int context_is_writeable_or_written(struct inode *inode,
1165 struct ceph_snap_context *snapc)
1166{
5f743e45
YZ
1167 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL,
1168 NULL, NULL);
6298a337
SW
1169 int ret = !oldest || snapc->seq <= oldest->seq;
1170
1171 ceph_put_snap_context(oldest);
1172 return ret;
1d3576fd
SW
1173}
1174
1175/*
1176 * We are only allowed to write into/dirty the page if the page is
1177 * clean, or already dirty within the same snap context.
8f883c24
SW
1178 *
1179 * called with page locked.
1180 * return success with page locked,
1181 * or any failure (incl -EAGAIN) with page unlocked.
1d3576fd 1182 */
4af6b225
YS
1183static int ceph_update_writeable_page(struct file *file,
1184 loff_t pos, unsigned len,
1185 struct page *page)
1d3576fd 1186{
496ad9aa 1187 struct inode *inode = file_inode(file);
6c93df5d 1188 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1d3576fd 1189 struct ceph_inode_info *ci = ceph_inode(inode);
09cbfeaf
KS
1190 loff_t page_off = pos & PAGE_MASK;
1191 int pos_in_page = pos & ~PAGE_MASK;
1d3576fd
SW
1192 int end_in_page = pos_in_page + len;
1193 loff_t i_size;
1d3576fd 1194 int r;
80e755fe 1195 struct ceph_snap_context *snapc, *oldest;
1d3576fd 1196
6c93df5d
YZ
1197 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1198 dout(" page %p forced umount\n", page);
1199 unlock_page(page);
1200 return -EIO;
1201 }
1202
1d3576fd
SW
1203retry_locked:
1204 /* writepages currently holds page lock, but if we change that later, */
1205 wait_on_page_writeback(page);
1206
61600ef8 1207 snapc = page_snap_context(page);
80e755fe 1208 if (snapc && snapc != ci->i_head_snapc) {
1d3576fd
SW
1209 /*
1210 * this page is already dirty in another (older) snap
1211 * context! is it writeable now?
1212 */
5f743e45 1213 oldest = get_oldest_context(inode, NULL, NULL, NULL);
1d3576fd 1214
80e755fe 1215 if (snapc->seq > oldest->seq) {
6298a337 1216 ceph_put_snap_context(oldest);
1d3576fd 1217 dout(" page %p snapc %p not current or oldest\n",
6298a337 1218 page, snapc);
1d3576fd
SW
1219 /*
1220 * queue for writeback, and wait for snapc to
1221 * be writeable or written
1222 */
6298a337 1223 snapc = ceph_get_snap_context(snapc);
1d3576fd 1224 unlock_page(page);
3c6f6b79 1225 ceph_queue_writeback(inode);
a78bbd4b 1226 r = wait_event_killable(ci->i_cap_wq,
1d3576fd
SW
1227 context_is_writeable_or_written(inode, snapc));
1228 ceph_put_snap_context(snapc);
8f883c24
SW
1229 if (r == -ERESTARTSYS)
1230 return r;
4af6b225 1231 return -EAGAIN;
1d3576fd 1232 }
6298a337 1233 ceph_put_snap_context(oldest);
1d3576fd
SW
1234
1235 /* yay, writeable, do it now (without dropping page lock) */
1236 dout(" page %p snapc %p not current, but oldest\n",
1237 page, snapc);
1238 if (!clear_page_dirty_for_io(page))
1239 goto retry_locked;
1240 r = writepage_nounlock(page, NULL);
1241 if (r < 0)
1242 goto fail_nosnap;
1243 goto retry_locked;
1244 }
1245
1246 if (PageUptodate(page)) {
1247 dout(" page %p already uptodate\n", page);
1248 return 0;
1249 }
1250
1251 /* full page? */
09cbfeaf 1252 if (pos_in_page == 0 && len == PAGE_SIZE)
1d3576fd
SW
1253 return 0;
1254
1255 /* past end of file? */
99c88e69 1256 i_size = i_size_read(inode);
1d3576fd 1257
1d3576fd
SW
1258 if (page_off >= i_size ||
1259 (pos_in_page == 0 && (pos+len) >= i_size &&
09cbfeaf 1260 end_in_page - pos_in_page != PAGE_SIZE)) {
1d3576fd 1261 dout(" zeroing %p 0 - %d and %d - %d\n",
09cbfeaf 1262 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1d3576fd
SW
1263 zero_user_segments(page,
1264 0, pos_in_page,
09cbfeaf 1265 end_in_page, PAGE_SIZE);
1d3576fd
SW
1266 return 0;
1267 }
1268
1269 /* we need to read it. */
1d3576fd
SW
1270 r = readpage_nounlock(file, page);
1271 if (r < 0)
1272 goto fail_nosnap;
1273 goto retry_locked;
1d3576fd
SW
1274fail_nosnap:
1275 unlock_page(page);
1276 return r;
1277}
1278
4af6b225
YS
1279/*
1280 * We are only allowed to write into/dirty the page if the page is
1281 * clean, or already dirty within the same snap context.
1282 */
1283static int ceph_write_begin(struct file *file, struct address_space *mapping,
1284 loff_t pos, unsigned len, unsigned flags,
1285 struct page **pagep, void **fsdata)
1286{
496ad9aa 1287 struct inode *inode = file_inode(file);
4af6b225 1288 struct page *page;
09cbfeaf 1289 pgoff_t index = pos >> PAGE_SHIFT;
7971bd92 1290 int r;
4af6b225
YS
1291
1292 do {
8f883c24 1293 /* get a page */
4af6b225 1294 page = grab_cache_page_write_begin(mapping, index, 0);
7971bd92
SW
1295 if (!page)
1296 return -ENOMEM;
4af6b225
YS
1297
1298 dout("write_begin file %p inode %p page %p %d~%d\n", file,
213c99ee 1299 inode, page, (int)pos, (int)len);
4af6b225
YS
1300
1301 r = ceph_update_writeable_page(file, pos, len, page);
c1d00b2d 1302 if (r < 0)
09cbfeaf 1303 put_page(page);
c1d00b2d
TK
1304 else
1305 *pagep = page;
4af6b225
YS
1306 } while (r == -EAGAIN);
1307
1308 return r;
1309}
1310
1d3576fd
SW
1311/*
1312 * we don't do anything in here that simple_write_end doesn't do
5dda377c 1313 * except adjust dirty page accounting
1d3576fd
SW
1314 */
1315static int ceph_write_end(struct file *file, struct address_space *mapping,
1316 loff_t pos, unsigned len, unsigned copied,
1317 struct page *page, void *fsdata)
1318{
496ad9aa 1319 struct inode *inode = file_inode(file);
1d3576fd
SW
1320 int check_cap = 0;
1321
1322 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1323 inode, page, (int)pos, (int)copied, (int)len);
1324
1325 /* zero the stale part of the page if we did a short copy */
b9de313c
AV
1326 if (!PageUptodate(page)) {
1327 if (copied < len) {
1328 copied = 0;
1329 goto out;
1330 }
1331 SetPageUptodate(page);
1332 }
1d3576fd
SW
1333
1334 /* did file size increase? */
99c88e69 1335 if (pos+copied > i_size_read(inode))
1d3576fd
SW
1336 check_cap = ceph_inode_set_size(inode, pos+copied);
1337
1d3576fd
SW
1338 set_page_dirty(page);
1339
b9de313c 1340out:
1d3576fd 1341 unlock_page(page);
09cbfeaf 1342 put_page(page);
1d3576fd
SW
1343
1344 if (check_cap)
1345 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1346
1347 return copied;
1348}
1349
1350/*
1351 * we set .direct_IO to indicate direct io is supported, but since we
1352 * intercept O_DIRECT reads and writes early, this function should
1353 * never get called.
1354 */
c8b8e32d 1355static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1d3576fd
SW
1356{
1357 WARN_ON(1);
1358 return -EINVAL;
1359}
1360
1361const struct address_space_operations ceph_aops = {
1362 .readpage = ceph_readpage,
1363 .readpages = ceph_readpages,
1364 .writepage = ceph_writepage,
1365 .writepages = ceph_writepages_start,
1366 .write_begin = ceph_write_begin,
1367 .write_end = ceph_write_end,
1368 .set_page_dirty = ceph_set_page_dirty,
1369 .invalidatepage = ceph_invalidatepage,
1370 .releasepage = ceph_releasepage,
1371 .direct_IO = ceph_direct_io,
1372};
1373
4f7e89f6
YZ
1374static void ceph_block_sigs(sigset_t *oldset)
1375{
1376 sigset_t mask;
1377 siginitsetinv(&mask, sigmask(SIGKILL));
1378 sigprocmask(SIG_BLOCK, &mask, oldset);
1379}
1380
1381static void ceph_restore_sigs(sigset_t *oldset)
1382{
1383 sigprocmask(SIG_SETMASK, oldset, NULL);
1384}
1d3576fd
SW
1385
1386/*
1387 * vm ops
1388 */
61f68816
YZ
1389static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1390{
1391 struct inode *inode = file_inode(vma->vm_file);
1392 struct ceph_inode_info *ci = ceph_inode(inode);
1393 struct ceph_file_info *fi = vma->vm_file->private_data;
3738daa6 1394 struct page *pinned_page = NULL;
09cbfeaf 1395 loff_t off = vmf->pgoff << PAGE_SHIFT;
61f68816 1396 int want, got, ret;
4f7e89f6
YZ
1397 sigset_t oldset;
1398
1399 ceph_block_sigs(&oldset);
61f68816
YZ
1400
1401 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
09cbfeaf 1402 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
61f68816
YZ
1403 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1404 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1405 else
1406 want = CEPH_CAP_FILE_CACHE;
4f7e89f6
YZ
1407
1408 got = 0;
1409 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
6ce026e4 1410 if (ret < 0)
4f7e89f6 1411 goto out_restore;
6ce026e4 1412
61f68816 1413 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
09cbfeaf 1414 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
61f68816 1415
83701246 1416 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
2b1ac852
YZ
1417 ci->i_inline_version == CEPH_INLINE_NONE) {
1418 current->journal_info = vma->vm_file;
83701246 1419 ret = filemap_fault(vma, vmf);
2b1ac852
YZ
1420 current->journal_info = NULL;
1421 } else
83701246 1422 ret = -EAGAIN;
61f68816
YZ
1423
1424 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
09cbfeaf 1425 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
3738daa6 1426 if (pinned_page)
09cbfeaf 1427 put_page(pinned_page);
61f68816
YZ
1428 ceph_put_cap_refs(ci, got);
1429
83701246 1430 if (ret != -EAGAIN)
4f7e89f6 1431 goto out_restore;
83701246
YZ
1432
1433 /* read inline data */
09cbfeaf 1434 if (off >= PAGE_SIZE) {
83701246
YZ
1435 /* does not support inline data > PAGE_SIZE */
1436 ret = VM_FAULT_SIGBUS;
1437 } else {
1438 int ret1;
1439 struct address_space *mapping = inode->i_mapping;
1440 struct page *page = find_or_create_page(mapping, 0,
c62d2555
MH
1441 mapping_gfp_constraint(mapping,
1442 ~__GFP_FS));
83701246
YZ
1443 if (!page) {
1444 ret = VM_FAULT_OOM;
4f7e89f6 1445 goto out_inline;
83701246
YZ
1446 }
1447 ret1 = __ceph_do_getattr(inode, page,
1448 CEPH_STAT_CAP_INLINE_DATA, true);
1449 if (ret1 < 0 || off >= i_size_read(inode)) {
1450 unlock_page(page);
09cbfeaf 1451 put_page(page);
6ce026e4
YZ
1452 if (ret1 < 0)
1453 ret = ret1;
1454 else
1455 ret = VM_FAULT_SIGBUS;
4f7e89f6 1456 goto out_inline;
83701246 1457 }
09cbfeaf
KS
1458 if (ret1 < PAGE_SIZE)
1459 zero_user_segment(page, ret1, PAGE_SIZE);
83701246
YZ
1460 else
1461 flush_dcache_page(page);
1462 SetPageUptodate(page);
1463 vmf->page = page;
1464 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
4f7e89f6
YZ
1465out_inline:
1466 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1467 inode, off, (size_t)PAGE_SIZE, ret);
83701246 1468 }
4f7e89f6
YZ
1469out_restore:
1470 ceph_restore_sigs(&oldset);
6ce026e4
YZ
1471 if (ret < 0)
1472 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1473
61f68816
YZ
1474 return ret;
1475}
1d3576fd
SW
1476
1477/*
1478 * Reuse write_begin here for simplicity.
1479 */
1480static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1481{
496ad9aa 1482 struct inode *inode = file_inode(vma->vm_file);
61f68816
YZ
1483 struct ceph_inode_info *ci = ceph_inode(inode);
1484 struct ceph_file_info *fi = vma->vm_file->private_data;
f66fd9f0 1485 struct ceph_cap_flush *prealloc_cf;
61f68816 1486 struct page *page = vmf->page;
6285bc23 1487 loff_t off = page_offset(page);
61f68816
YZ
1488 loff_t size = i_size_read(inode);
1489 size_t len;
1490 int want, got, ret;
4f7e89f6 1491 sigset_t oldset;
3ca9c3bd 1492
f66fd9f0
YZ
1493 prealloc_cf = ceph_alloc_cap_flush();
1494 if (!prealloc_cf)
6ce026e4 1495 return VM_FAULT_OOM;
f66fd9f0 1496
4f7e89f6 1497 ceph_block_sigs(&oldset);
f66fd9f0 1498
28127bdd
YZ
1499 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1500 struct page *locked_page = NULL;
1501 if (off == 0) {
1502 lock_page(page);
1503 locked_page = page;
1504 }
1505 ret = ceph_uninline_data(vma->vm_file, locked_page);
1506 if (locked_page)
1507 unlock_page(locked_page);
6ce026e4 1508 if (ret < 0)
f66fd9f0 1509 goto out_free;
28127bdd
YZ
1510 }
1511
09cbfeaf
KS
1512 if (off + PAGE_SIZE <= size)
1513 len = PAGE_SIZE;
1d3576fd 1514 else
09cbfeaf 1515 len = size & ~PAGE_MASK;
1d3576fd 1516
61f68816
YZ
1517 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1518 inode, ceph_vinop(inode), off, len, size);
1519 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1520 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1521 else
1522 want = CEPH_CAP_FILE_BUFFER;
4f7e89f6
YZ
1523
1524 got = 0;
1525 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1526 &got, NULL);
6ce026e4 1527 if (ret < 0)
4f7e89f6 1528 goto out_free;
6ce026e4 1529
61f68816
YZ
1530 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1531 inode, off, len, ceph_cap_string(got));
1532
1533 /* Update time before taking page lock */
1534 file_update_time(vma->vm_file);
4af6b225 1535
f0b33df5
YZ
1536 do {
1537 lock_page(page);
4af6b225 1538
f0b33df5
YZ
1539 if ((off > size) || (page->mapping != inode->i_mapping)) {
1540 unlock_page(page);
1541 ret = VM_FAULT_NOPAGE;
1542 break;
1543 }
1544
1545 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1546 if (ret >= 0) {
1547 /* success. we'll keep the page locked. */
1548 set_page_dirty(page);
1549 ret = VM_FAULT_LOCKED;
1550 }
1551 } while (ret == -EAGAIN);
4af6b225 1552
28127bdd
YZ
1553 if (ret == VM_FAULT_LOCKED ||
1554 ci->i_inline_version != CEPH_INLINE_NONE) {
61f68816
YZ
1555 int dirty;
1556 spin_lock(&ci->i_ceph_lock);
28127bdd 1557 ci->i_inline_version = CEPH_INLINE_NONE;
f66fd9f0
YZ
1558 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1559 &prealloc_cf);
61f68816
YZ
1560 spin_unlock(&ci->i_ceph_lock);
1561 if (dirty)
1562 __mark_inode_dirty(inode, dirty);
1563 }
1564
1565 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1566 inode, off, len, ceph_cap_string(got), ret);
1567 ceph_put_cap_refs(ci, got);
f66fd9f0 1568out_free:
4f7e89f6 1569 ceph_restore_sigs(&oldset);
f66fd9f0 1570 ceph_free_cap_flush(prealloc_cf);
6ce026e4
YZ
1571 if (ret < 0)
1572 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1d3576fd
SW
1573 return ret;
1574}
1575
31c542a1
YZ
1576void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1577 char *data, size_t len)
1578{
1579 struct address_space *mapping = inode->i_mapping;
1580 struct page *page;
1581
1582 if (locked_page) {
1583 page = locked_page;
1584 } else {
1585 if (i_size_read(inode) == 0)
1586 return;
1587 page = find_or_create_page(mapping, 0,
c62d2555
MH
1588 mapping_gfp_constraint(mapping,
1589 ~__GFP_FS));
31c542a1
YZ
1590 if (!page)
1591 return;
1592 if (PageUptodate(page)) {
1593 unlock_page(page);
09cbfeaf 1594 put_page(page);
31c542a1
YZ
1595 return;
1596 }
1597 }
1598
0668ff52 1599 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
31c542a1
YZ
1600 inode, ceph_vinop(inode), len, locked_page);
1601
1602 if (len > 0) {
1603 void *kaddr = kmap_atomic(page);
1604 memcpy(kaddr, data, len);
1605 kunmap_atomic(kaddr);
1606 }
1607
1608 if (page != locked_page) {
09cbfeaf
KS
1609 if (len < PAGE_SIZE)
1610 zero_user_segment(page, len, PAGE_SIZE);
31c542a1
YZ
1611 else
1612 flush_dcache_page(page);
1613
1614 SetPageUptodate(page);
1615 unlock_page(page);
09cbfeaf 1616 put_page(page);
31c542a1
YZ
1617 }
1618}
1619
28127bdd
YZ
1620int ceph_uninline_data(struct file *filp, struct page *locked_page)
1621{
1622 struct inode *inode = file_inode(filp);
1623 struct ceph_inode_info *ci = ceph_inode(inode);
1624 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1625 struct ceph_osd_request *req;
1626 struct page *page = NULL;
1627 u64 len, inline_version;
1628 int err = 0;
1629 bool from_pagecache = false;
1630
1631 spin_lock(&ci->i_ceph_lock);
1632 inline_version = ci->i_inline_version;
1633 spin_unlock(&ci->i_ceph_lock);
1634
1635 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1636 inode, ceph_vinop(inode), inline_version);
1637
1638 if (inline_version == 1 || /* initial version, no data */
1639 inline_version == CEPH_INLINE_NONE)
1640 goto out;
1641
1642 if (locked_page) {
1643 page = locked_page;
1644 WARN_ON(!PageUptodate(page));
1645 } else if (ceph_caps_issued(ci) &
1646 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1647 page = find_get_page(inode->i_mapping, 0);
1648 if (page) {
1649 if (PageUptodate(page)) {
1650 from_pagecache = true;
1651 lock_page(page);
1652 } else {
09cbfeaf 1653 put_page(page);
28127bdd
YZ
1654 page = NULL;
1655 }
1656 }
1657 }
1658
1659 if (page) {
1660 len = i_size_read(inode);
09cbfeaf
KS
1661 if (len > PAGE_SIZE)
1662 len = PAGE_SIZE;
28127bdd
YZ
1663 } else {
1664 page = __page_cache_alloc(GFP_NOFS);
1665 if (!page) {
1666 err = -ENOMEM;
1667 goto out;
1668 }
1669 err = __ceph_do_getattr(inode, page,
1670 CEPH_STAT_CAP_INLINE_DATA, true);
1671 if (err < 0) {
1672 /* no inline data */
1673 if (err == -ENODATA)
1674 err = 0;
1675 goto out;
1676 }
1677 len = err;
1678 }
1679
1680 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1681 ceph_vino(inode), 0, &len, 0, 1,
1682 CEPH_OSD_OP_CREATE,
1683 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
34b759b4 1684 NULL, 0, 0, false);
28127bdd
YZ
1685 if (IS_ERR(req)) {
1686 err = PTR_ERR(req);
1687 goto out;
1688 }
1689
bb873b53 1690 req->r_mtime = inode->i_mtime;
28127bdd
YZ
1691 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1692 if (!err)
1693 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1694 ceph_osdc_put_request(req);
1695 if (err < 0)
1696 goto out;
1697
1698 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1699 ceph_vino(inode), 0, &len, 1, 3,
1700 CEPH_OSD_OP_WRITE,
1701 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
34b759b4
ID
1702 NULL, ci->i_truncate_seq,
1703 ci->i_truncate_size, false);
28127bdd
YZ
1704 if (IS_ERR(req)) {
1705 err = PTR_ERR(req);
1706 goto out;
1707 }
1708
1709 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1710
ec137c10
YZ
1711 {
1712 __le64 xattr_buf = cpu_to_le64(inline_version);
1713 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1714 "inline_version", &xattr_buf,
1715 sizeof(xattr_buf),
1716 CEPH_OSD_CMPXATTR_OP_GT,
1717 CEPH_OSD_CMPXATTR_MODE_U64);
1718 if (err)
1719 goto out_put;
1720 }
1721
1722 {
1723 char xattr_buf[32];
1724 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1725 "%llu", inline_version);
1726 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1727 "inline_version",
1728 xattr_buf, xattr_len, 0, 0);
1729 if (err)
1730 goto out_put;
1731 }
28127bdd 1732
bb873b53 1733 req->r_mtime = inode->i_mtime;
28127bdd
YZ
1734 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1735 if (!err)
1736 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1737out_put:
1738 ceph_osdc_put_request(req);
1739 if (err == -ECANCELED)
1740 err = 0;
1741out:
1742 if (page && page != locked_page) {
1743 if (from_pagecache) {
1744 unlock_page(page);
09cbfeaf 1745 put_page(page);
28127bdd
YZ
1746 } else
1747 __free_pages(page, 0);
1748 }
1749
1750 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1751 inode, ceph_vinop(inode), inline_version, err);
1752 return err;
1753}
1754
7cbea8dc 1755static const struct vm_operations_struct ceph_vmops = {
61f68816 1756 .fault = ceph_filemap_fault,
1d3576fd
SW
1757 .page_mkwrite = ceph_page_mkwrite,
1758};
1759
1760int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1761{
1762 struct address_space *mapping = file->f_mapping;
1763
1764 if (!mapping->a_ops->readpage)
1765 return -ENOEXEC;
1766 file_accessed(file);
1767 vma->vm_ops = &ceph_vmops;
1d3576fd
SW
1768 return 0;
1769}
10183a69
YZ
1770
1771enum {
1772 POOL_READ = 1,
1773 POOL_WRITE = 2,
1774};
1775
779fe0fb
YZ
1776static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1777 s64 pool, struct ceph_string *pool_ns)
10183a69
YZ
1778{
1779 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1780 struct ceph_mds_client *mdsc = fsc->mdsc;
1781 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1782 struct rb_node **p, *parent;
1783 struct ceph_pool_perm *perm;
1784 struct page **pages;
779fe0fb 1785 size_t pool_ns_len;
10183a69
YZ
1786 int err = 0, err2 = 0, have = 0;
1787
1788 down_read(&mdsc->pool_perm_rwsem);
1789 p = &mdsc->pool_perm_tree.rb_node;
1790 while (*p) {
1791 perm = rb_entry(*p, struct ceph_pool_perm, node);
1792 if (pool < perm->pool)
1793 p = &(*p)->rb_left;
1794 else if (pool > perm->pool)
1795 p = &(*p)->rb_right;
1796 else {
779fe0fb
YZ
1797 int ret = ceph_compare_string(pool_ns,
1798 perm->pool_ns,
1799 perm->pool_ns_len);
1800 if (ret < 0)
1801 p = &(*p)->rb_left;
1802 else if (ret > 0)
1803 p = &(*p)->rb_right;
1804 else {
1805 have = perm->perm;
1806 break;
1807 }
10183a69
YZ
1808 }
1809 }
1810 up_read(&mdsc->pool_perm_rwsem);
1811 if (*p)
1812 goto out;
1813
779fe0fb
YZ
1814 if (pool_ns)
1815 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1816 pool, (int)pool_ns->len, pool_ns->str);
1817 else
1818 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
10183a69
YZ
1819
1820 down_write(&mdsc->pool_perm_rwsem);
779fe0fb 1821 p = &mdsc->pool_perm_tree.rb_node;
10183a69
YZ
1822 parent = NULL;
1823 while (*p) {
1824 parent = *p;
1825 perm = rb_entry(parent, struct ceph_pool_perm, node);
1826 if (pool < perm->pool)
1827 p = &(*p)->rb_left;
1828 else if (pool > perm->pool)
1829 p = &(*p)->rb_right;
1830 else {
779fe0fb
YZ
1831 int ret = ceph_compare_string(pool_ns,
1832 perm->pool_ns,
1833 perm->pool_ns_len);
1834 if (ret < 0)
1835 p = &(*p)->rb_left;
1836 else if (ret > 0)
1837 p = &(*p)->rb_right;
1838 else {
1839 have = perm->perm;
1840 break;
1841 }
10183a69
YZ
1842 }
1843 }
1844 if (*p) {
1845 up_write(&mdsc->pool_perm_rwsem);
1846 goto out;
1847 }
1848
34b759b4 1849 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
10183a69
YZ
1850 1, false, GFP_NOFS);
1851 if (!rd_req) {
1852 err = -ENOMEM;
1853 goto out_unlock;
1854 }
1855
1856 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1857 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1858 rd_req->r_base_oloc.pool = pool;
779fe0fb
YZ
1859 if (pool_ns)
1860 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
d30291b9 1861 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
10183a69 1862
13d1ad16
ID
1863 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1864 if (err)
1865 goto out_unlock;
10183a69 1866
34b759b4 1867 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
10183a69
YZ
1868 1, false, GFP_NOFS);
1869 if (!wr_req) {
1870 err = -ENOMEM;
1871 goto out_unlock;
1872 }
1873
fe5da05e 1874 wr_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ACK;
10183a69 1875 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
63244fa1 1876 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
d30291b9 1877 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
10183a69 1878
13d1ad16
ID
1879 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1880 if (err)
1881 goto out_unlock;
10183a69
YZ
1882
1883 /* one page should be large enough for STAT data */
1884 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1885 if (IS_ERR(pages)) {
1886 err = PTR_ERR(pages);
1887 goto out_unlock;
1888 }
1889
1890 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1891 0, false, true);
10183a69
YZ
1892 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1893
bb873b53 1894 wr_req->r_mtime = ci->vfs_inode.i_mtime;
10183a69
YZ
1895 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1896
1897 if (!err)
1898 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1899 if (!err2)
1900 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1901
1902 if (err >= 0 || err == -ENOENT)
1903 have |= POOL_READ;
1904 else if (err != -EPERM)
1905 goto out_unlock;
1906
1907 if (err2 == 0 || err2 == -EEXIST)
1908 have |= POOL_WRITE;
1909 else if (err2 != -EPERM) {
1910 err = err2;
1911 goto out_unlock;
1912 }
1913
779fe0fb
YZ
1914 pool_ns_len = pool_ns ? pool_ns->len : 0;
1915 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
10183a69
YZ
1916 if (!perm) {
1917 err = -ENOMEM;
1918 goto out_unlock;
1919 }
1920
1921 perm->pool = pool;
1922 perm->perm = have;
779fe0fb
YZ
1923 perm->pool_ns_len = pool_ns_len;
1924 if (pool_ns_len > 0)
1925 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1926 perm->pool_ns[pool_ns_len] = 0;
1927
10183a69
YZ
1928 rb_link_node(&perm->node, parent, p);
1929 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1930 err = 0;
1931out_unlock:
1932 up_write(&mdsc->pool_perm_rwsem);
1933
3ed97d63
ID
1934 ceph_osdc_put_request(rd_req);
1935 ceph_osdc_put_request(wr_req);
10183a69
YZ
1936out:
1937 if (!err)
1938 err = have;
779fe0fb
YZ
1939 if (pool_ns)
1940 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1941 pool, (int)pool_ns->len, pool_ns->str, err);
1942 else
1943 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
10183a69
YZ
1944 return err;
1945}
1946
1947int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1948{
7627151e 1949 s64 pool;
779fe0fb 1950 struct ceph_string *pool_ns;
10183a69
YZ
1951 int ret, flags;
1952
80e80fbb
YZ
1953 if (ci->i_vino.snap != CEPH_NOSNAP) {
1954 /*
1955 * Pool permission check needs to write to the first object.
1956 * But for snapshot, head of the first object may have alread
1957 * been deleted. Skip check to avoid creating orphan object.
1958 */
1959 return 0;
1960 }
1961
10183a69
YZ
1962 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
1963 NOPOOLPERM))
1964 return 0;
1965
1966 spin_lock(&ci->i_ceph_lock);
1967 flags = ci->i_ceph_flags;
7627151e 1968 pool = ci->i_layout.pool_id;
10183a69
YZ
1969 spin_unlock(&ci->i_ceph_lock);
1970check:
1971 if (flags & CEPH_I_POOL_PERM) {
1972 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
7627151e 1973 dout("ceph_pool_perm_check pool %lld no read perm\n",
10183a69
YZ
1974 pool);
1975 return -EPERM;
1976 }
1977 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
7627151e 1978 dout("ceph_pool_perm_check pool %lld no write perm\n",
10183a69
YZ
1979 pool);
1980 return -EPERM;
1981 }
1982 return 0;
1983 }
1984
779fe0fb
YZ
1985 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1986 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1987 ceph_put_string(pool_ns);
10183a69
YZ
1988 if (ret < 0)
1989 return ret;
1990
1991 flags = CEPH_I_POOL_PERM;
1992 if (ret & POOL_READ)
1993 flags |= CEPH_I_POOL_RD;
1994 if (ret & POOL_WRITE)
1995 flags |= CEPH_I_POOL_WR;
1996
1997 spin_lock(&ci->i_ceph_lock);
779fe0fb
YZ
1998 if (pool == ci->i_layout.pool_id &&
1999 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2000 ci->i_ceph_flags |= flags;
10183a69 2001 } else {
7627151e 2002 pool = ci->i_layout.pool_id;
10183a69
YZ
2003 flags = ci->i_ceph_flags;
2004 }
2005 spin_unlock(&ci->i_ceph_lock);
2006 goto check;
2007}
2008
2009void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2010{
2011 struct ceph_pool_perm *perm;
2012 struct rb_node *n;
2013
2014 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2015 n = rb_first(&mdsc->pool_perm_tree);
2016 perm = rb_entry(n, struct ceph_pool_perm, node);
2017 rb_erase(n, &mdsc->pool_perm_tree);
2018 kfree(perm);
2019 }
2020}