1 #include <linux/ceph/ceph_debug.h> 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 */ 8 #include <linux/slab.h> 9 #include <linux/pagevec.h> 10 #include <linux/task_io_accounting_ops.h> 11 #include <linux/signal.h> 12 13 #include "super.h" 14 #include "mds_client.h" 15 #include "cache.h" 16 #include <linux/ceph/osd_client.h> 17 18 /* 19 * Ceph address space ops. 20 * 21 * There are a few funny things going on here. 22 * 23 * The page->private field is used to reference a struct 24 * ceph_snap_context for _every_ dirty page. This indicates which 25 * snapshot the page was logically dirtied in, and thus which snap 26 * context needs to be associated with the osd write during writeback. 27 * 28 * Similarly, struct ceph_inode_info maintains a set of counters to 29 * count dirty pages on the inode. In the absence of snapshots, 30 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 31 * 32 * When a snapshot is taken (that is, when the client receives 33 * notification that a snapshot was taken), each inode with caps and 34 * with dirty pages (dirty pages implies there is a cap) gets a new 35 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 36 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 37 * moved to capsnap->dirty. (Unless a sync write is currently in 38 * progress. In that case, the capsnap is said to be "pending", new 39 * writes cannot start, and the capsnap isn't "finalized" until the 40 * write completes (or fails) and a final size/mtime for the inode for 41 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 42 * 43 * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 44 * we look for the first capsnap in i_cap_snaps and write out pages in 45 * that snap context _only_. Then we move on to the next capsnap, 46 * eventually reaching the "live" or "head" context (i.e., pages that 47 * are not yet snapped) and are writing the most recently dirtied 48 * pages. 49 * 50 * Invalidate and so forth must take care to ensure the dirty page 51 * accounting is preserved. 52 */ 53 54 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 55 #define CONGESTION_OFF_THRESH(congestion_kb) \ 56 (CONGESTION_ON_THRESH(congestion_kb) - \ 57 (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 58 59 static inline struct ceph_snap_context *page_snap_context(struct page *page) 60 { 61 if (PagePrivate(page)) 62 return (void *)page->private; 63 return NULL; 64 } 65 66 /* 67 * Dirty a page. Optimistically adjust accounting, on the assumption 68 * that we won't race with invalidate. If we do, readjust. 69 */ 70 static int ceph_set_page_dirty(struct page *page) 71 { 72 struct address_space *mapping = page->mapping; 73 struct inode *inode; 74 struct ceph_inode_info *ci; 75 struct ceph_snap_context *snapc; 76 int ret; 77 78 if (unlikely(!mapping)) 79 return !TestSetPageDirty(page); 80 81 if (PageDirty(page)) { 82 dout("%p set_page_dirty %p idx %lu -- already dirty\n", 83 mapping->host, page, page->index); 84 BUG_ON(!PagePrivate(page)); 85 return 0; 86 } 87 88 inode = mapping->host; 89 ci = ceph_inode(inode); 90 91 /* dirty the head */ 92 spin_lock(&ci->i_ceph_lock); 93 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference 94 if (__ceph_have_pending_cap_snap(ci)) { 95 struct ceph_cap_snap *capsnap = 96 list_last_entry(&ci->i_cap_snaps, 97 struct ceph_cap_snap, 98 ci_item); 99 snapc = ceph_get_snap_context(capsnap->context); 100 capsnap->dirty_pages++; 101 } else { 102 BUG_ON(!ci->i_head_snapc); 103 snapc = ceph_get_snap_context(ci->i_head_snapc); 104 ++ci->i_wrbuffer_ref_head; 105 } 106 if (ci->i_wrbuffer_ref == 0) 107 ihold(inode); 108 ++ci->i_wrbuffer_ref; 109 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " 110 "snapc %p seq %lld (%d snaps)\n", 111 mapping->host, page, page->index, 112 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 113 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 114 snapc, snapc->seq, snapc->num_snaps); 115 spin_unlock(&ci->i_ceph_lock); 116 117 /* 118 * Reference snap context in page->private. Also set 119 * PagePrivate so that we get invalidatepage callback. 120 */ 121 BUG_ON(PagePrivate(page)); 122 page->private = (unsigned long)snapc; 123 SetPagePrivate(page); 124 125 ret = __set_page_dirty_nobuffers(page); 126 WARN_ON(!PageLocked(page)); 127 WARN_ON(!page->mapping); 128 129 return ret; 130 } 131 132 /* 133 * If we are truncating the full page (i.e. offset == 0), adjust the 134 * dirty page counters appropriately. Only called if there is private 135 * data on the page. 136 */ 137 static void ceph_invalidatepage(struct page *page, unsigned int offset, 138 unsigned int length) 139 { 140 struct inode *inode; 141 struct ceph_inode_info *ci; 142 struct ceph_snap_context *snapc = page_snap_context(page); 143 144 inode = page->mapping->host; 145 ci = ceph_inode(inode); 146 147 if (offset != 0 || length != PAGE_SIZE) { 148 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", 149 inode, page, page->index, offset, length); 150 return; 151 } 152 153 ceph_invalidate_fscache_page(inode, page); 154 155 WARN_ON(!PageLocked(page)); 156 if (!PagePrivate(page)) 157 return; 158 159 ClearPageChecked(page); 160 161 dout("%p invalidatepage %p idx %lu full dirty page\n", 162 inode, page, page->index); 163 164 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 165 ceph_put_snap_context(snapc); 166 page->private = 0; 167 ClearPagePrivate(page); 168 } 169 170 static int ceph_releasepage(struct page *page, gfp_t g) 171 { 172 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host, 173 page, page->index, PageDirty(page) ? "" : "not "); 174 175 /* Can we release the page from the cache? */ 176 if (!ceph_release_fscache_page(page, g)) 177 return 0; 178 179 return !PagePrivate(page); 180 } 181 182 /* 183 * read a single page, without unlocking it. 184 */ 185 static int ceph_do_readpage(struct file *filp, struct page *page) 186 { 187 struct inode *inode = file_inode(filp); 188 struct ceph_inode_info *ci = ceph_inode(inode); 189 struct ceph_osd_client *osdc = 190 &ceph_inode_to_client(inode)->client->osdc; 191 int err = 0; 192 u64 off = page_offset(page); 193 u64 len = PAGE_SIZE; 194 195 if (off >= i_size_read(inode)) { 196 zero_user_segment(page, 0, PAGE_SIZE); 197 SetPageUptodate(page); 198 return 0; 199 } 200 201 if (ci->i_inline_version != CEPH_INLINE_NONE) { 202 /* 203 * Uptodate inline data should have been added 204 * into page cache while getting Fcr caps. 205 */ 206 if (off == 0) 207 return -EINVAL; 208 zero_user_segment(page, 0, PAGE_SIZE); 209 SetPageUptodate(page); 210 return 0; 211 } 212 213 err = ceph_readpage_from_fscache(inode, page); 214 if (err == 0) 215 return -EINPROGRESS; 216 217 dout("readpage inode %p file %p page %p index %lu\n", 218 inode, filp, page, page->index); 219 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, 220 off, &len, 221 ci->i_truncate_seq, ci->i_truncate_size, 222 &page, 1, 0); 223 if (err == -ENOENT) 224 err = 0; 225 if (err < 0) { 226 SetPageError(page); 227 ceph_fscache_readpage_cancel(inode, page); 228 goto out; 229 } 230 if (err < PAGE_SIZE) 231 /* zero fill remainder of page */ 232 zero_user_segment(page, err, PAGE_SIZE); 233 else 234 flush_dcache_page(page); 235 236 SetPageUptodate(page); 237 ceph_readpage_to_fscache(inode, page); 238 239 out: 240 return err < 0 ? err : 0; 241 } 242 243 static int ceph_readpage(struct file *filp, struct page *page) 244 { 245 int r = ceph_do_readpage(filp, page); 246 if (r != -EINPROGRESS) 247 unlock_page(page); 248 else 249 r = 0; 250 return r; 251 } 252 253 /* 254 * Finish an async read(ahead) op. 255 */ 256 static void finish_read(struct ceph_osd_request *req) 257 { 258 struct inode *inode = req->r_inode; 259 struct ceph_osd_data *osd_data; 260 int rc = req->r_result <= 0 ? req->r_result : 0; 261 int bytes = req->r_result >= 0 ? req->r_result : 0; 262 int num_pages; 263 int i; 264 265 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); 266 267 /* unlock all pages, zeroing any data we didn't read */ 268 osd_data = osd_req_op_extent_osd_data(req, 0); 269 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 270 num_pages = calc_pages_for((u64)osd_data->alignment, 271 (u64)osd_data->length); 272 for (i = 0; i < num_pages; i++) { 273 struct page *page = osd_data->pages[i]; 274 275 if (rc < 0 && rc != -ENOENT) { 276 ceph_fscache_readpage_cancel(inode, page); 277 goto unlock; 278 } 279 if (bytes < (int)PAGE_SIZE) { 280 /* zero (remainder of) page */ 281 int s = bytes < 0 ? 0 : bytes; 282 zero_user_segment(page, s, PAGE_SIZE); 283 } 284 dout("finish_read %p uptodate %p idx %lu\n", inode, page, 285 page->index); 286 flush_dcache_page(page); 287 SetPageUptodate(page); 288 ceph_readpage_to_fscache(inode, page); 289 unlock: 290 unlock_page(page); 291 put_page(page); 292 bytes -= PAGE_SIZE; 293 } 294 kfree(osd_data->pages); 295 } 296 297 /* 298 * start an async read(ahead) operation. return nr_pages we submitted 299 * a read for on success, or negative error code. 300 */ 301 static int start_read(struct inode *inode, struct list_head *page_list, int max) 302 { 303 struct ceph_osd_client *osdc = 304 &ceph_inode_to_client(inode)->client->osdc; 305 struct ceph_inode_info *ci = ceph_inode(inode); 306 struct page *page = list_entry(page_list->prev, struct page, lru); 307 struct ceph_vino vino; 308 struct ceph_osd_request *req; 309 u64 off; 310 u64 len; 311 int i; 312 struct page **pages; 313 pgoff_t next_index; 314 int nr_pages = 0; 315 int got = 0; 316 int ret = 0; 317 318 if (!current->journal_info) { 319 /* caller of readpages does not hold buffer and read caps 320 * (fadvise, madvise and readahead cases) */ 321 int want = CEPH_CAP_FILE_CACHE; 322 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got); 323 if (ret < 0) { 324 dout("start_read %p, error getting cap\n", inode); 325 } else if (!(got & want)) { 326 dout("start_read %p, no cache cap\n", inode); 327 ret = 0; 328 } 329 if (ret <= 0) { 330 if (got) 331 ceph_put_cap_refs(ci, got); 332 while (!list_empty(page_list)) { 333 page = list_entry(page_list->prev, 334 struct page, lru); 335 list_del(&page->lru); 336 put_page(page); 337 } 338 return ret; 339 } 340 } 341 342 off = (u64) page_offset(page); 343 344 /* count pages */ 345 next_index = page->index; 346 list_for_each_entry_reverse(page, page_list, lru) { 347 if (page->index != next_index) 348 break; 349 nr_pages++; 350 next_index++; 351 if (max && nr_pages == max) 352 break; 353 } 354 len = nr_pages << PAGE_SHIFT; 355 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, 356 off, len); 357 vino = ceph_vino(inode); 358 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, 359 0, 1, CEPH_OSD_OP_READ, 360 CEPH_OSD_FLAG_READ, NULL, 361 ci->i_truncate_seq, ci->i_truncate_size, 362 false); 363 if (IS_ERR(req)) { 364 ret = PTR_ERR(req); 365 goto out; 366 } 367 368 /* build page vector */ 369 nr_pages = calc_pages_for(0, len); 370 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL); 371 if (!pages) { 372 ret = -ENOMEM; 373 goto out_put; 374 } 375 for (i = 0; i < nr_pages; ++i) { 376 page = list_entry(page_list->prev, struct page, lru); 377 BUG_ON(PageLocked(page)); 378 list_del(&page->lru); 379 380 dout("start_read %p adding %p idx %lu\n", inode, page, 381 page->index); 382 if (add_to_page_cache_lru(page, &inode->i_data, page->index, 383 GFP_KERNEL)) { 384 ceph_fscache_uncache_page(inode, page); 385 put_page(page); 386 dout("start_read %p add_to_page_cache failed %p\n", 387 inode, page); 388 nr_pages = i; 389 if (nr_pages > 0) { 390 len = nr_pages << PAGE_SHIFT; 391 osd_req_op_extent_update(req, 0, len); 392 break; 393 } 394 goto out_pages; 395 } 396 pages[i] = page; 397 } 398 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); 399 req->r_callback = finish_read; 400 req->r_inode = inode; 401 402 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); 403 ret = ceph_osdc_start_request(osdc, req, false); 404 if (ret < 0) 405 goto out_pages; 406 ceph_osdc_put_request(req); 407 408 /* After adding locked pages to page cache, the inode holds cache cap. 409 * So we can drop our cap refs. */ 410 if (got) 411 ceph_put_cap_refs(ci, got); 412 413 return nr_pages; 414 415 out_pages: 416 for (i = 0; i < nr_pages; ++i) { 417 ceph_fscache_readpage_cancel(inode, pages[i]); 418 unlock_page(pages[i]); 419 } 420 ceph_put_page_vector(pages, nr_pages, false); 421 out_put: 422 ceph_osdc_put_request(req); 423 out: 424 if (got) 425 ceph_put_cap_refs(ci, got); 426 return ret; 427 } 428 429 430 /* 431 * Read multiple pages. Leave pages we don't read + unlock in page_list; 432 * the caller (VM) cleans them up. 433 */ 434 static int ceph_readpages(struct file *file, struct address_space *mapping, 435 struct list_head *page_list, unsigned nr_pages) 436 { 437 struct inode *inode = file_inode(file); 438 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 439 int rc = 0; 440 int max = 0; 441 442 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) 443 return -EINVAL; 444 445 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list, 446 &nr_pages); 447 448 if (rc == 0) 449 goto out; 450 451 max = fsc->mount_options->rsize >> PAGE_SHIFT; 452 dout("readpages %p file %p nr_pages %d max %d\n", 453 inode, file, nr_pages, max); 454 while (!list_empty(page_list)) { 455 rc = start_read(inode, page_list, max); 456 if (rc < 0) 457 goto out; 458 } 459 out: 460 ceph_fscache_readpages_cancel(inode, page_list); 461 462 dout("readpages %p file %p ret %d\n", inode, file, rc); 463 return rc; 464 } 465 466 struct ceph_writeback_ctl 467 { 468 loff_t i_size; 469 u64 truncate_size; 470 u32 truncate_seq; 471 bool size_stable; 472 bool head_snapc; 473 }; 474 475 /* 476 * Get ref for the oldest snapc for an inode with dirty data... that is, the 477 * only snap context we are allowed to write back. 478 */ 479 static struct ceph_snap_context * 480 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, 481 struct ceph_snap_context *page_snapc) 482 { 483 struct ceph_inode_info *ci = ceph_inode(inode); 484 struct ceph_snap_context *snapc = NULL; 485 struct ceph_cap_snap *capsnap = NULL; 486 487 spin_lock(&ci->i_ceph_lock); 488 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 489 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 490 capsnap->context, capsnap->dirty_pages); 491 if (!capsnap->dirty_pages) 492 continue; 493 494 /* get i_size, truncate_{seq,size} for page_snapc? */ 495 if (snapc && capsnap->context != page_snapc) 496 continue; 497 498 if (ctl) { 499 if (capsnap->writing) { 500 ctl->i_size = i_size_read(inode); 501 ctl->size_stable = false; 502 } else { 503 ctl->i_size = capsnap->size; 504 ctl->size_stable = true; 505 } 506 ctl->truncate_size = capsnap->truncate_size; 507 ctl->truncate_seq = capsnap->truncate_seq; 508 ctl->head_snapc = false; 509 } 510 511 if (snapc) 512 break; 513 514 snapc = ceph_get_snap_context(capsnap->context); 515 if (!page_snapc || 516 page_snapc == snapc || 517 page_snapc->seq > snapc->seq) 518 break; 519 } 520 if (!snapc && ci->i_wrbuffer_ref_head) { 521 snapc = ceph_get_snap_context(ci->i_head_snapc); 522 dout(" head snapc %p has %d dirty pages\n", 523 snapc, ci->i_wrbuffer_ref_head); 524 if (ctl) { 525 ctl->i_size = i_size_read(inode); 526 ctl->truncate_size = ci->i_truncate_size; 527 ctl->truncate_seq = ci->i_truncate_seq; 528 ctl->size_stable = false; 529 ctl->head_snapc = true; 530 } 531 } 532 spin_unlock(&ci->i_ceph_lock); 533 return snapc; 534 } 535 536 static u64 get_writepages_data_length(struct inode *inode, 537 struct page *page, u64 start) 538 { 539 struct ceph_inode_info *ci = ceph_inode(inode); 540 struct ceph_snap_context *snapc = page_snap_context(page); 541 struct ceph_cap_snap *capsnap = NULL; 542 u64 end = i_size_read(inode); 543 544 if (snapc != ci->i_head_snapc) { 545 bool found = false; 546 spin_lock(&ci->i_ceph_lock); 547 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 548 if (capsnap->context == snapc) { 549 if (!capsnap->writing) 550 end = capsnap->size; 551 found = true; 552 break; 553 } 554 } 555 spin_unlock(&ci->i_ceph_lock); 556 WARN_ON(!found); 557 } 558 if (end > page_offset(page) + PAGE_SIZE) 559 end = page_offset(page) + PAGE_SIZE; 560 return end > start ? end - start : 0; 561 } 562 563 /* 564 * Write a single page, but leave the page locked. 565 * 566 * If we get a write error, set the page error bit, but still adjust the 567 * dirty page accounting (i.e., page is no longer dirty). 568 */ 569 static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 570 { 571 struct inode *inode; 572 struct ceph_inode_info *ci; 573 struct ceph_fs_client *fsc; 574 struct ceph_snap_context *snapc, *oldest; 575 loff_t page_off = page_offset(page); 576 long writeback_stat; 577 int err, len = PAGE_SIZE; 578 struct ceph_writeback_ctl ceph_wbc; 579 580 dout("writepage %p idx %lu\n", page, page->index); 581 582 inode = page->mapping->host; 583 ci = ceph_inode(inode); 584 fsc = ceph_inode_to_client(inode); 585 586 /* verify this is a writeable snap context */ 587 snapc = page_snap_context(page); 588 if (!snapc) { 589 dout("writepage %p page %p not dirty?\n", inode, page); 590 return 0; 591 } 592 oldest = get_oldest_context(inode, &ceph_wbc, snapc); 593 if (snapc->seq > oldest->seq) { 594 dout("writepage %p page %p snapc %p not writeable - noop\n", 595 inode, page, snapc); 596 /* we should only noop if called by kswapd */ 597 WARN_ON(!(current->flags & PF_MEMALLOC)); 598 ceph_put_snap_context(oldest); 599 redirty_page_for_writepage(wbc, page); 600 return 0; 601 } 602 ceph_put_snap_context(oldest); 603 604 /* is this a partial page at end of file? */ 605 if (page_off >= ceph_wbc.i_size) { 606 dout("%p page eof %llu\n", page, ceph_wbc.i_size); 607 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); 608 return 0; 609 } 610 611 if (ceph_wbc.i_size < page_off + len) 612 len = ceph_wbc.i_size - page_off; 613 614 dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n", 615 inode, page, page->index, page_off, len, snapc, snapc->seq); 616 617 writeback_stat = atomic_long_inc_return(&fsc->writeback_count); 618 if (writeback_stat > 619 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 620 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 621 622 set_page_writeback(page); 623 err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode), 624 &ci->i_layout, snapc, page_off, len, 625 ceph_wbc.truncate_seq, 626 ceph_wbc.truncate_size, 627 &inode->i_mtime, &page, 1); 628 if (err < 0) { 629 struct writeback_control tmp_wbc; 630 if (!wbc) 631 wbc = &tmp_wbc; 632 if (err == -ERESTARTSYS) { 633 /* killed by SIGKILL */ 634 dout("writepage interrupted page %p\n", page); 635 redirty_page_for_writepage(wbc, page); 636 end_page_writeback(page); 637 return err; 638 } 639 dout("writepage setting page/mapping error %d %p\n", 640 err, page); 641 SetPageError(page); 642 mapping_set_error(&inode->i_data, err); 643 wbc->pages_skipped++; 644 } else { 645 dout("writepage cleaned page %p\n", page); 646 err = 0; /* vfs expects us to return 0 */ 647 } 648 page->private = 0; 649 ClearPagePrivate(page); 650 end_page_writeback(page); 651 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 652 ceph_put_snap_context(snapc); /* page's reference */ 653 return err; 654 } 655 656 static int ceph_writepage(struct page *page, struct writeback_control *wbc) 657 { 658 int err; 659 struct inode *inode = page->mapping->host; 660 BUG_ON(!inode); 661 ihold(inode); 662 err = writepage_nounlock(page, wbc); 663 if (err == -ERESTARTSYS) { 664 /* direct memory reclaimer was killed by SIGKILL. return 0 665 * to prevent caller from setting mapping/page error */ 666 err = 0; 667 } 668 unlock_page(page); 669 iput(inode); 670 return err; 671 } 672 673 /* 674 * lame release_pages helper. release_pages() isn't exported to 675 * modules. 676 */ 677 static void ceph_release_pages(struct page **pages, int num) 678 { 679 struct pagevec pvec; 680 int i; 681 682 pagevec_init(&pvec, 0); 683 for (i = 0; i < num; i++) { 684 if (pagevec_add(&pvec, pages[i]) == 0) 685 pagevec_release(&pvec); 686 } 687 pagevec_release(&pvec); 688 } 689 690 /* 691 * async writeback completion handler. 692 * 693 * If we get an error, set the mapping error bit, but not the individual 694 * page error bits. 695 */ 696 static void writepages_finish(struct ceph_osd_request *req) 697 { 698 struct inode *inode = req->r_inode; 699 struct ceph_inode_info *ci = ceph_inode(inode); 700 struct ceph_osd_data *osd_data; 701 struct page *page; 702 int num_pages, total_pages = 0; 703 int i, j; 704 int rc = req->r_result; 705 struct ceph_snap_context *snapc = req->r_snapc; 706 struct address_space *mapping = inode->i_mapping; 707 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 708 bool remove_page; 709 710 dout("writepages_finish %p rc %d\n", inode, rc); 711 if (rc < 0) { 712 mapping_set_error(mapping, rc); 713 ceph_set_error_write(ci); 714 } else { 715 ceph_clear_error_write(ci); 716 } 717 718 /* 719 * We lost the cache cap, need to truncate the page before 720 * it is unlocked, otherwise we'd truncate it later in the 721 * page truncation thread, possibly losing some data that 722 * raced its way in 723 */ 724 remove_page = !(ceph_caps_issued(ci) & 725 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); 726 727 /* clean all pages */ 728 for (i = 0; i < req->r_num_ops; i++) { 729 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) 730 break; 731 732 osd_data = osd_req_op_extent_osd_data(req, i); 733 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 734 num_pages = calc_pages_for((u64)osd_data->alignment, 735 (u64)osd_data->length); 736 total_pages += num_pages; 737 for (j = 0; j < num_pages; j++) { 738 page = osd_data->pages[j]; 739 BUG_ON(!page); 740 WARN_ON(!PageUptodate(page)); 741 742 if (atomic_long_dec_return(&fsc->writeback_count) < 743 CONGESTION_OFF_THRESH( 744 fsc->mount_options->congestion_kb)) 745 clear_bdi_congested(inode_to_bdi(inode), 746 BLK_RW_ASYNC); 747 748 ceph_put_snap_context(page_snap_context(page)); 749 page->private = 0; 750 ClearPagePrivate(page); 751 dout("unlocking %p\n", page); 752 end_page_writeback(page); 753 754 if (remove_page) 755 generic_error_remove_page(inode->i_mapping, 756 page); 757 758 unlock_page(page); 759 } 760 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", 761 inode, osd_data->length, rc >= 0 ? num_pages : 0); 762 763 ceph_release_pages(osd_data->pages, num_pages); 764 } 765 766 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); 767 768 osd_data = osd_req_op_extent_osd_data(req, 0); 769 if (osd_data->pages_from_pool) 770 mempool_free(osd_data->pages, 771 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); 772 else 773 kfree(osd_data->pages); 774 ceph_osdc_put_request(req); 775 } 776 777 /* 778 * initiate async writeback 779 */ 780 static int ceph_writepages_start(struct address_space *mapping, 781 struct writeback_control *wbc) 782 { 783 struct inode *inode = mapping->host; 784 struct ceph_inode_info *ci = ceph_inode(inode); 785 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 786 struct ceph_vino vino = ceph_vino(inode); 787 pgoff_t index, start_index, end = -1; 788 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 789 struct pagevec pvec; 790 int rc = 0; 791 unsigned int wsize = i_blocksize(inode); 792 struct ceph_osd_request *req = NULL; 793 struct ceph_writeback_ctl ceph_wbc; 794 bool should_loop, range_whole = false; 795 bool stop, done = false; 796 797 dout("writepages_start %p (mode=%s)\n", inode, 798 wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 799 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 800 801 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 802 if (ci->i_wrbuffer_ref > 0) { 803 pr_warn_ratelimited( 804 "writepage_start %p %lld forced umount\n", 805 inode, ceph_ino(inode)); 806 } 807 mapping_set_error(mapping, -EIO); 808 return -EIO; /* we're in a forced umount, don't write! */ 809 } 810 if (fsc->mount_options->wsize < wsize) 811 wsize = fsc->mount_options->wsize; 812 813 pagevec_init(&pvec, 0); 814 815 start_index = wbc->range_cyclic ? mapping->writeback_index : 0; 816 index = start_index; 817 818 retry: 819 /* find oldest snap context with dirty data */ 820 snapc = get_oldest_context(inode, &ceph_wbc, NULL); 821 if (!snapc) { 822 /* hmm, why does writepages get called when there 823 is no dirty data? */ 824 dout(" no snap context with dirty data?\n"); 825 goto out; 826 } 827 dout(" oldest snapc is %p seq %lld (%d snaps)\n", 828 snapc, snapc->seq, snapc->num_snaps); 829 830 should_loop = false; 831 if (ceph_wbc.head_snapc && snapc != last_snapc) { 832 /* where to start/end? */ 833 if (wbc->range_cyclic) { 834 index = start_index; 835 end = -1; 836 if (index > 0) 837 should_loop = true; 838 dout(" cyclic, start at %lu\n", index); 839 } else { 840 index = wbc->range_start >> PAGE_SHIFT; 841 end = wbc->range_end >> PAGE_SHIFT; 842 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 843 range_whole = true; 844 dout(" not cyclic, %lu to %lu\n", index, end); 845 } 846 } else if (!ceph_wbc.head_snapc) { 847 /* Do not respect wbc->range_{start,end}. Dirty pages 848 * in that range can be associated with newer snapc. 849 * They are not writeable until we write all dirty pages 850 * associated with 'snapc' get written */ 851 if (index > 0 || wbc->sync_mode != WB_SYNC_NONE) 852 should_loop = true; 853 dout(" non-head snapc, range whole\n"); 854 } 855 856 ceph_put_snap_context(last_snapc); 857 last_snapc = snapc; 858 859 stop = false; 860 while (!stop && index <= end) { 861 int num_ops = 0, op_idx; 862 unsigned i, pvec_pages, max_pages, locked_pages = 0; 863 struct page **pages = NULL, **data_pages; 864 mempool_t *pool = NULL; /* Becomes non-null if mempool used */ 865 struct page *page; 866 pgoff_t strip_unit_end = 0; 867 u64 offset = 0, len = 0; 868 869 max_pages = wsize >> PAGE_SHIFT; 870 871 get_more_pages: 872 pvec_pages = min_t(unsigned, PAGEVEC_SIZE, 873 max_pages - locked_pages); 874 if (end - index < (u64)(pvec_pages - 1)) 875 pvec_pages = (unsigned)(end - index) + 1; 876 877 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index, 878 PAGECACHE_TAG_DIRTY, 879 pvec_pages); 880 dout("pagevec_lookup_tag got %d\n", pvec_pages); 881 if (!pvec_pages && !locked_pages) 882 break; 883 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 884 page = pvec.pages[i]; 885 dout("? %p idx %lu\n", page, page->index); 886 if (locked_pages == 0) 887 lock_page(page); /* first page */ 888 else if (!trylock_page(page)) 889 break; 890 891 /* only dirty pages, or our accounting breaks */ 892 if (unlikely(!PageDirty(page)) || 893 unlikely(page->mapping != mapping)) { 894 dout("!dirty or !mapping %p\n", page); 895 unlock_page(page); 896 continue; 897 } 898 if (page->index > end) { 899 dout("end of range %p\n", page); 900 /* can't be range_cyclic (1st pass) because 901 * end == -1 in that case. */ 902 stop = true; 903 if (ceph_wbc.head_snapc) 904 done = true; 905 unlock_page(page); 906 break; 907 } 908 if (strip_unit_end && (page->index > strip_unit_end)) { 909 dout("end of strip unit %p\n", page); 910 unlock_page(page); 911 break; 912 } 913 if (page_offset(page) >= ceph_wbc.i_size) { 914 dout("%p page eof %llu\n", 915 page, ceph_wbc.i_size); 916 /* not done if range_cyclic */ 917 stop = true; 918 unlock_page(page); 919 break; 920 } 921 if (PageWriteback(page)) { 922 if (wbc->sync_mode == WB_SYNC_NONE) { 923 dout("%p under writeback\n", page); 924 unlock_page(page); 925 continue; 926 } 927 dout("waiting on writeback %p\n", page); 928 wait_on_page_writeback(page); 929 } 930 931 /* only if matching snap context */ 932 pgsnapc = page_snap_context(page); 933 if (pgsnapc != snapc) { 934 dout("page snapc %p %lld != oldest %p %lld\n", 935 pgsnapc, pgsnapc->seq, snapc, snapc->seq); 936 unlock_page(page); 937 continue; 938 } 939 940 if (!clear_page_dirty_for_io(page)) { 941 dout("%p !clear_page_dirty_for_io\n", page); 942 unlock_page(page); 943 continue; 944 } 945 946 /* 947 * We have something to write. If this is 948 * the first locked page this time through, 949 * calculate max possinle write size and 950 * allocate a page array 951 */ 952 if (locked_pages == 0) { 953 u64 objnum; 954 u64 objoff; 955 956 /* prepare async write request */ 957 offset = (u64)page_offset(page); 958 len = wsize; 959 960 rc = ceph_calc_file_object_mapping(&ci->i_layout, 961 offset, len, 962 &objnum, &objoff, 963 &len); 964 if (rc < 0) { 965 unlock_page(page); 966 break; 967 } 968 969 num_ops = 1; 970 strip_unit_end = page->index + 971 ((len - 1) >> PAGE_SHIFT); 972 973 BUG_ON(pages); 974 max_pages = calc_pages_for(0, (u64)len); 975 pages = kmalloc(max_pages * sizeof (*pages), 976 GFP_NOFS); 977 if (!pages) { 978 pool = fsc->wb_pagevec_pool; 979 pages = mempool_alloc(pool, GFP_NOFS); 980 BUG_ON(!pages); 981 } 982 983 len = 0; 984 } else if (page->index != 985 (offset + len) >> PAGE_SHIFT) { 986 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS : 987 CEPH_OSD_MAX_OPS)) { 988 redirty_page_for_writepage(wbc, page); 989 unlock_page(page); 990 break; 991 } 992 993 num_ops++; 994 offset = (u64)page_offset(page); 995 len = 0; 996 } 997 998 /* note position of first page in pvec */ 999 dout("%p will write page %p idx %lu\n", 1000 inode, page, page->index); 1001 1002 if (atomic_long_inc_return(&fsc->writeback_count) > 1003 CONGESTION_ON_THRESH( 1004 fsc->mount_options->congestion_kb)) { 1005 set_bdi_congested(inode_to_bdi(inode), 1006 BLK_RW_ASYNC); 1007 } 1008 1009 1010 pages[locked_pages++] = page; 1011 pvec.pages[i] = NULL; 1012 1013 len += PAGE_SIZE; 1014 } 1015 1016 /* did we get anything? */ 1017 if (!locked_pages) 1018 goto release_pvec_pages; 1019 if (i) { 1020 unsigned j, n = 0; 1021 /* shift unused page to beginning of pvec */ 1022 for (j = 0; j < pvec_pages; j++) { 1023 if (!pvec.pages[j]) 1024 continue; 1025 if (n < j) 1026 pvec.pages[n] = pvec.pages[j]; 1027 n++; 1028 } 1029 pvec.nr = n; 1030 1031 if (pvec_pages && i == pvec_pages && 1032 locked_pages < max_pages) { 1033 dout("reached end pvec, trying for more\n"); 1034 pagevec_release(&pvec); 1035 goto get_more_pages; 1036 } 1037 } 1038 1039 new_request: 1040 offset = page_offset(pages[0]); 1041 len = wsize; 1042 1043 req = ceph_osdc_new_request(&fsc->client->osdc, 1044 &ci->i_layout, vino, 1045 offset, &len, 0, num_ops, 1046 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1047 snapc, ceph_wbc.truncate_seq, 1048 ceph_wbc.truncate_size, false); 1049 if (IS_ERR(req)) { 1050 req = ceph_osdc_new_request(&fsc->client->osdc, 1051 &ci->i_layout, vino, 1052 offset, &len, 0, 1053 min(num_ops, 1054 CEPH_OSD_SLAB_OPS), 1055 CEPH_OSD_OP_WRITE, 1056 CEPH_OSD_FLAG_WRITE, 1057 snapc, ceph_wbc.truncate_seq, 1058 ceph_wbc.truncate_size, true); 1059 BUG_ON(IS_ERR(req)); 1060 } 1061 BUG_ON(len < page_offset(pages[locked_pages - 1]) + 1062 PAGE_SIZE - offset); 1063 1064 req->r_callback = writepages_finish; 1065 req->r_inode = inode; 1066 1067 /* Format the osd request message and submit the write */ 1068 len = 0; 1069 data_pages = pages; 1070 op_idx = 0; 1071 for (i = 0; i < locked_pages; i++) { 1072 u64 cur_offset = page_offset(pages[i]); 1073 if (offset + len != cur_offset) { 1074 if (op_idx + 1 == req->r_num_ops) 1075 break; 1076 osd_req_op_extent_dup_last(req, op_idx, 1077 cur_offset - offset); 1078 dout("writepages got pages at %llu~%llu\n", 1079 offset, len); 1080 osd_req_op_extent_osd_data_pages(req, op_idx, 1081 data_pages, len, 0, 1082 !!pool, false); 1083 osd_req_op_extent_update(req, op_idx, len); 1084 1085 len = 0; 1086 offset = cur_offset; 1087 data_pages = pages + i; 1088 op_idx++; 1089 } 1090 1091 set_page_writeback(pages[i]); 1092 len += PAGE_SIZE; 1093 } 1094 1095 if (ceph_wbc.size_stable) { 1096 len = min(len, ceph_wbc.i_size - offset); 1097 } else if (i == locked_pages) { 1098 /* writepages_finish() clears writeback pages 1099 * according to the data length, so make sure 1100 * data length covers all locked pages */ 1101 u64 min_len = len + 1 - PAGE_SIZE; 1102 len = get_writepages_data_length(inode, pages[i - 1], 1103 offset); 1104 len = max(len, min_len); 1105 } 1106 dout("writepages got pages at %llu~%llu\n", offset, len); 1107 1108 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, 1109 0, !!pool, false); 1110 osd_req_op_extent_update(req, op_idx, len); 1111 1112 BUG_ON(op_idx + 1 != req->r_num_ops); 1113 1114 pool = NULL; 1115 if (i < locked_pages) { 1116 BUG_ON(num_ops <= req->r_num_ops); 1117 num_ops -= req->r_num_ops; 1118 locked_pages -= i; 1119 1120 /* allocate new pages array for next request */ 1121 data_pages = pages; 1122 pages = kmalloc(locked_pages * sizeof (*pages), 1123 GFP_NOFS); 1124 if (!pages) { 1125 pool = fsc->wb_pagevec_pool; 1126 pages = mempool_alloc(pool, GFP_NOFS); 1127 BUG_ON(!pages); 1128 } 1129 memcpy(pages, data_pages + i, 1130 locked_pages * sizeof(*pages)); 1131 memset(data_pages + i, 0, 1132 locked_pages * sizeof(*pages)); 1133 } else { 1134 BUG_ON(num_ops != req->r_num_ops); 1135 index = pages[i - 1]->index + 1; 1136 /* request message now owns the pages array */ 1137 pages = NULL; 1138 } 1139 1140 req->r_mtime = inode->i_mtime; 1141 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); 1142 BUG_ON(rc); 1143 req = NULL; 1144 1145 wbc->nr_to_write -= i; 1146 if (pages) 1147 goto new_request; 1148 1149 /* 1150 * We stop writing back only if we are not doing 1151 * integrity sync. In case of integrity sync we have to 1152 * keep going until we have written all the pages 1153 * we tagged for writeback prior to entering this loop. 1154 */ 1155 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) 1156 done = stop = true; 1157 1158 release_pvec_pages: 1159 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 1160 pvec.nr ? pvec.pages[0] : NULL); 1161 pagevec_release(&pvec); 1162 } 1163 1164 if (should_loop && !done) { 1165 /* more to do; loop back to beginning of file */ 1166 dout("writepages looping back to beginning of file\n"); 1167 end = start_index - 1; /* OK even when start_index == 0 */ 1168 start_index = 0; 1169 index = 0; 1170 goto retry; 1171 } 1172 1173 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1174 mapping->writeback_index = index; 1175 1176 out: 1177 ceph_osdc_put_request(req); 1178 ceph_put_snap_context(last_snapc); 1179 dout("writepages dend - startone, rc = %d\n", rc); 1180 return rc; 1181 } 1182 1183 1184 1185 /* 1186 * See if a given @snapc is either writeable, or already written. 1187 */ 1188 static int context_is_writeable_or_written(struct inode *inode, 1189 struct ceph_snap_context *snapc) 1190 { 1191 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); 1192 int ret = !oldest || snapc->seq <= oldest->seq; 1193 1194 ceph_put_snap_context(oldest); 1195 return ret; 1196 } 1197 1198 /* 1199 * We are only allowed to write into/dirty the page if the page is 1200 * clean, or already dirty within the same snap context. 1201 * 1202 * called with page locked. 1203 * return success with page locked, 1204 * or any failure (incl -EAGAIN) with page unlocked. 1205 */ 1206 static int ceph_update_writeable_page(struct file *file, 1207 loff_t pos, unsigned len, 1208 struct page *page) 1209 { 1210 struct inode *inode = file_inode(file); 1211 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1212 struct ceph_inode_info *ci = ceph_inode(inode); 1213 loff_t page_off = pos & PAGE_MASK; 1214 int pos_in_page = pos & ~PAGE_MASK; 1215 int end_in_page = pos_in_page + len; 1216 loff_t i_size; 1217 int r; 1218 struct ceph_snap_context *snapc, *oldest; 1219 1220 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 1221 dout(" page %p forced umount\n", page); 1222 unlock_page(page); 1223 return -EIO; 1224 } 1225 1226 retry_locked: 1227 /* writepages currently holds page lock, but if we change that later, */ 1228 wait_on_page_writeback(page); 1229 1230 snapc = page_snap_context(page); 1231 if (snapc && snapc != ci->i_head_snapc) { 1232 /* 1233 * this page is already dirty in another (older) snap 1234 * context! is it writeable now? 1235 */ 1236 oldest = get_oldest_context(inode, NULL, NULL); 1237 if (snapc->seq > oldest->seq) { 1238 ceph_put_snap_context(oldest); 1239 dout(" page %p snapc %p not current or oldest\n", 1240 page, snapc); 1241 /* 1242 * queue for writeback, and wait for snapc to 1243 * be writeable or written 1244 */ 1245 snapc = ceph_get_snap_context(snapc); 1246 unlock_page(page); 1247 ceph_queue_writeback(inode); 1248 r = wait_event_killable(ci->i_cap_wq, 1249 context_is_writeable_or_written(inode, snapc)); 1250 ceph_put_snap_context(snapc); 1251 if (r == -ERESTARTSYS) 1252 return r; 1253 return -EAGAIN; 1254 } 1255 ceph_put_snap_context(oldest); 1256 1257 /* yay, writeable, do it now (without dropping page lock) */ 1258 dout(" page %p snapc %p not current, but oldest\n", 1259 page, snapc); 1260 if (!clear_page_dirty_for_io(page)) 1261 goto retry_locked; 1262 r = writepage_nounlock(page, NULL); 1263 if (r < 0) 1264 goto fail_unlock; 1265 goto retry_locked; 1266 } 1267 1268 if (PageUptodate(page)) { 1269 dout(" page %p already uptodate\n", page); 1270 return 0; 1271 } 1272 1273 /* full page? */ 1274 if (pos_in_page == 0 && len == PAGE_SIZE) 1275 return 0; 1276 1277 /* past end of file? */ 1278 i_size = i_size_read(inode); 1279 1280 if (page_off >= i_size || 1281 (pos_in_page == 0 && (pos+len) >= i_size && 1282 end_in_page - pos_in_page != PAGE_SIZE)) { 1283 dout(" zeroing %p 0 - %d and %d - %d\n", 1284 page, pos_in_page, end_in_page, (int)PAGE_SIZE); 1285 zero_user_segments(page, 1286 0, pos_in_page, 1287 end_in_page, PAGE_SIZE); 1288 return 0; 1289 } 1290 1291 /* we need to read it. */ 1292 r = ceph_do_readpage(file, page); 1293 if (r < 0) { 1294 if (r == -EINPROGRESS) 1295 return -EAGAIN; 1296 goto fail_unlock; 1297 } 1298 goto retry_locked; 1299 fail_unlock: 1300 unlock_page(page); 1301 return r; 1302 } 1303 1304 /* 1305 * We are only allowed to write into/dirty the page if the page is 1306 * clean, or already dirty within the same snap context. 1307 */ 1308 static int ceph_write_begin(struct file *file, struct address_space *mapping, 1309 loff_t pos, unsigned len, unsigned flags, 1310 struct page **pagep, void **fsdata) 1311 { 1312 struct inode *inode = file_inode(file); 1313 struct page *page; 1314 pgoff_t index = pos >> PAGE_SHIFT; 1315 int r; 1316 1317 do { 1318 /* get a page */ 1319 page = grab_cache_page_write_begin(mapping, index, 0); 1320 if (!page) 1321 return -ENOMEM; 1322 1323 dout("write_begin file %p inode %p page %p %d~%d\n", file, 1324 inode, page, (int)pos, (int)len); 1325 1326 r = ceph_update_writeable_page(file, pos, len, page); 1327 if (r < 0) 1328 put_page(page); 1329 else 1330 *pagep = page; 1331 } while (r == -EAGAIN); 1332 1333 return r; 1334 } 1335 1336 /* 1337 * we don't do anything in here that simple_write_end doesn't do 1338 * except adjust dirty page accounting 1339 */ 1340 static int ceph_write_end(struct file *file, struct address_space *mapping, 1341 loff_t pos, unsigned len, unsigned copied, 1342 struct page *page, void *fsdata) 1343 { 1344 struct inode *inode = file_inode(file); 1345 bool check_cap = false; 1346 1347 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, 1348 inode, page, (int)pos, (int)copied, (int)len); 1349 1350 /* zero the stale part of the page if we did a short copy */ 1351 if (!PageUptodate(page)) { 1352 if (copied < len) { 1353 copied = 0; 1354 goto out; 1355 } 1356 SetPageUptodate(page); 1357 } 1358 1359 /* did file size increase? */ 1360 if (pos+copied > i_size_read(inode)) 1361 check_cap = ceph_inode_set_size(inode, pos+copied); 1362 1363 set_page_dirty(page); 1364 1365 out: 1366 unlock_page(page); 1367 put_page(page); 1368 1369 if (check_cap) 1370 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); 1371 1372 return copied; 1373 } 1374 1375 /* 1376 * we set .direct_IO to indicate direct io is supported, but since we 1377 * intercept O_DIRECT reads and writes early, this function should 1378 * never get called. 1379 */ 1380 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter) 1381 { 1382 WARN_ON(1); 1383 return -EINVAL; 1384 } 1385 1386 const struct address_space_operations ceph_aops = { 1387 .readpage = ceph_readpage, 1388 .readpages = ceph_readpages, 1389 .writepage = ceph_writepage, 1390 .writepages = ceph_writepages_start, 1391 .write_begin = ceph_write_begin, 1392 .write_end = ceph_write_end, 1393 .set_page_dirty = ceph_set_page_dirty, 1394 .invalidatepage = ceph_invalidatepage, 1395 .releasepage = ceph_releasepage, 1396 .direct_IO = ceph_direct_io, 1397 }; 1398 1399 static void ceph_block_sigs(sigset_t *oldset) 1400 { 1401 sigset_t mask; 1402 siginitsetinv(&mask, sigmask(SIGKILL)); 1403 sigprocmask(SIG_BLOCK, &mask, oldset); 1404 } 1405 1406 static void ceph_restore_sigs(sigset_t *oldset) 1407 { 1408 sigprocmask(SIG_SETMASK, oldset, NULL); 1409 } 1410 1411 /* 1412 * vm ops 1413 */ 1414 static int ceph_filemap_fault(struct vm_fault *vmf) 1415 { 1416 struct vm_area_struct *vma = vmf->vma; 1417 struct inode *inode = file_inode(vma->vm_file); 1418 struct ceph_inode_info *ci = ceph_inode(inode); 1419 struct ceph_file_info *fi = vma->vm_file->private_data; 1420 struct page *pinned_page = NULL; 1421 loff_t off = vmf->pgoff << PAGE_SHIFT; 1422 int want, got, ret; 1423 sigset_t oldset; 1424 1425 ceph_block_sigs(&oldset); 1426 1427 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n", 1428 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE); 1429 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1430 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1431 else 1432 want = CEPH_CAP_FILE_CACHE; 1433 1434 got = 0; 1435 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); 1436 if (ret < 0) 1437 goto out_restore; 1438 1439 dout("filemap_fault %p %llu~%zd got cap refs on %s\n", 1440 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got)); 1441 1442 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || 1443 ci->i_inline_version == CEPH_INLINE_NONE) { 1444 current->journal_info = vma->vm_file; 1445 ret = filemap_fault(vmf); 1446 current->journal_info = NULL; 1447 } else 1448 ret = -EAGAIN; 1449 1450 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n", 1451 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret); 1452 if (pinned_page) 1453 put_page(pinned_page); 1454 ceph_put_cap_refs(ci, got); 1455 1456 if (ret != -EAGAIN) 1457 goto out_restore; 1458 1459 /* read inline data */ 1460 if (off >= PAGE_SIZE) { 1461 /* does not support inline data > PAGE_SIZE */ 1462 ret = VM_FAULT_SIGBUS; 1463 } else { 1464 int ret1; 1465 struct address_space *mapping = inode->i_mapping; 1466 struct page *page = find_or_create_page(mapping, 0, 1467 mapping_gfp_constraint(mapping, 1468 ~__GFP_FS)); 1469 if (!page) { 1470 ret = VM_FAULT_OOM; 1471 goto out_inline; 1472 } 1473 ret1 = __ceph_do_getattr(inode, page, 1474 CEPH_STAT_CAP_INLINE_DATA, true); 1475 if (ret1 < 0 || off >= i_size_read(inode)) { 1476 unlock_page(page); 1477 put_page(page); 1478 if (ret1 < 0) 1479 ret = ret1; 1480 else 1481 ret = VM_FAULT_SIGBUS; 1482 goto out_inline; 1483 } 1484 if (ret1 < PAGE_SIZE) 1485 zero_user_segment(page, ret1, PAGE_SIZE); 1486 else 1487 flush_dcache_page(page); 1488 SetPageUptodate(page); 1489 vmf->page = page; 1490 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; 1491 out_inline: 1492 dout("filemap_fault %p %llu~%zd read inline data ret %d\n", 1493 inode, off, (size_t)PAGE_SIZE, ret); 1494 } 1495 out_restore: 1496 ceph_restore_sigs(&oldset); 1497 if (ret < 0) 1498 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; 1499 1500 return ret; 1501 } 1502 1503 /* 1504 * Reuse write_begin here for simplicity. 1505 */ 1506 static int ceph_page_mkwrite(struct vm_fault *vmf) 1507 { 1508 struct vm_area_struct *vma = vmf->vma; 1509 struct inode *inode = file_inode(vma->vm_file); 1510 struct ceph_inode_info *ci = ceph_inode(inode); 1511 struct ceph_file_info *fi = vma->vm_file->private_data; 1512 struct ceph_cap_flush *prealloc_cf; 1513 struct page *page = vmf->page; 1514 loff_t off = page_offset(page); 1515 loff_t size = i_size_read(inode); 1516 size_t len; 1517 int want, got, ret; 1518 sigset_t oldset; 1519 1520 prealloc_cf = ceph_alloc_cap_flush(); 1521 if (!prealloc_cf) 1522 return VM_FAULT_OOM; 1523 1524 ceph_block_sigs(&oldset); 1525 1526 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1527 struct page *locked_page = NULL; 1528 if (off == 0) { 1529 lock_page(page); 1530 locked_page = page; 1531 } 1532 ret = ceph_uninline_data(vma->vm_file, locked_page); 1533 if (locked_page) 1534 unlock_page(locked_page); 1535 if (ret < 0) 1536 goto out_free; 1537 } 1538 1539 if (off + PAGE_SIZE <= size) 1540 len = PAGE_SIZE; 1541 else 1542 len = size & ~PAGE_MASK; 1543 1544 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", 1545 inode, ceph_vinop(inode), off, len, size); 1546 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1547 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 1548 else 1549 want = CEPH_CAP_FILE_BUFFER; 1550 1551 got = 0; 1552 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len, 1553 &got, NULL); 1554 if (ret < 0) 1555 goto out_free; 1556 1557 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", 1558 inode, off, len, ceph_cap_string(got)); 1559 1560 /* Update time before taking page lock */ 1561 file_update_time(vma->vm_file); 1562 1563 do { 1564 lock_page(page); 1565 1566 if ((off > size) || (page->mapping != inode->i_mapping)) { 1567 unlock_page(page); 1568 ret = VM_FAULT_NOPAGE; 1569 break; 1570 } 1571 1572 ret = ceph_update_writeable_page(vma->vm_file, off, len, page); 1573 if (ret >= 0) { 1574 /* success. we'll keep the page locked. */ 1575 set_page_dirty(page); 1576 ret = VM_FAULT_LOCKED; 1577 } 1578 } while (ret == -EAGAIN); 1579 1580 if (ret == VM_FAULT_LOCKED || 1581 ci->i_inline_version != CEPH_INLINE_NONE) { 1582 int dirty; 1583 spin_lock(&ci->i_ceph_lock); 1584 ci->i_inline_version = CEPH_INLINE_NONE; 1585 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1586 &prealloc_cf); 1587 spin_unlock(&ci->i_ceph_lock); 1588 if (dirty) 1589 __mark_inode_dirty(inode, dirty); 1590 } 1591 1592 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n", 1593 inode, off, len, ceph_cap_string(got), ret); 1594 ceph_put_cap_refs(ci, got); 1595 out_free: 1596 ceph_restore_sigs(&oldset); 1597 ceph_free_cap_flush(prealloc_cf); 1598 if (ret < 0) 1599 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; 1600 return ret; 1601 } 1602 1603 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, 1604 char *data, size_t len) 1605 { 1606 struct address_space *mapping = inode->i_mapping; 1607 struct page *page; 1608 1609 if (locked_page) { 1610 page = locked_page; 1611 } else { 1612 if (i_size_read(inode) == 0) 1613 return; 1614 page = find_or_create_page(mapping, 0, 1615 mapping_gfp_constraint(mapping, 1616 ~__GFP_FS)); 1617 if (!page) 1618 return; 1619 if (PageUptodate(page)) { 1620 unlock_page(page); 1621 put_page(page); 1622 return; 1623 } 1624 } 1625 1626 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", 1627 inode, ceph_vinop(inode), len, locked_page); 1628 1629 if (len > 0) { 1630 void *kaddr = kmap_atomic(page); 1631 memcpy(kaddr, data, len); 1632 kunmap_atomic(kaddr); 1633 } 1634 1635 if (page != locked_page) { 1636 if (len < PAGE_SIZE) 1637 zero_user_segment(page, len, PAGE_SIZE); 1638 else 1639 flush_dcache_page(page); 1640 1641 SetPageUptodate(page); 1642 unlock_page(page); 1643 put_page(page); 1644 } 1645 } 1646 1647 int ceph_uninline_data(struct file *filp, struct page *locked_page) 1648 { 1649 struct inode *inode = file_inode(filp); 1650 struct ceph_inode_info *ci = ceph_inode(inode); 1651 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1652 struct ceph_osd_request *req; 1653 struct page *page = NULL; 1654 u64 len, inline_version; 1655 int err = 0; 1656 bool from_pagecache = false; 1657 1658 spin_lock(&ci->i_ceph_lock); 1659 inline_version = ci->i_inline_version; 1660 spin_unlock(&ci->i_ceph_lock); 1661 1662 dout("uninline_data %p %llx.%llx inline_version %llu\n", 1663 inode, ceph_vinop(inode), inline_version); 1664 1665 if (inline_version == 1 || /* initial version, no data */ 1666 inline_version == CEPH_INLINE_NONE) 1667 goto out; 1668 1669 if (locked_page) { 1670 page = locked_page; 1671 WARN_ON(!PageUptodate(page)); 1672 } else if (ceph_caps_issued(ci) & 1673 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { 1674 page = find_get_page(inode->i_mapping, 0); 1675 if (page) { 1676 if (PageUptodate(page)) { 1677 from_pagecache = true; 1678 lock_page(page); 1679 } else { 1680 put_page(page); 1681 page = NULL; 1682 } 1683 } 1684 } 1685 1686 if (page) { 1687 len = i_size_read(inode); 1688 if (len > PAGE_SIZE) 1689 len = PAGE_SIZE; 1690 } else { 1691 page = __page_cache_alloc(GFP_NOFS); 1692 if (!page) { 1693 err = -ENOMEM; 1694 goto out; 1695 } 1696 err = __ceph_do_getattr(inode, page, 1697 CEPH_STAT_CAP_INLINE_DATA, true); 1698 if (err < 0) { 1699 /* no inline data */ 1700 if (err == -ENODATA) 1701 err = 0; 1702 goto out; 1703 } 1704 len = err; 1705 } 1706 1707 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1708 ceph_vino(inode), 0, &len, 0, 1, 1709 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, 1710 NULL, 0, 0, false); 1711 if (IS_ERR(req)) { 1712 err = PTR_ERR(req); 1713 goto out; 1714 } 1715 1716 req->r_mtime = inode->i_mtime; 1717 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1718 if (!err) 1719 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1720 ceph_osdc_put_request(req); 1721 if (err < 0) 1722 goto out; 1723 1724 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1725 ceph_vino(inode), 0, &len, 1, 3, 1726 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1727 NULL, ci->i_truncate_seq, 1728 ci->i_truncate_size, false); 1729 if (IS_ERR(req)) { 1730 err = PTR_ERR(req); 1731 goto out; 1732 } 1733 1734 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); 1735 1736 { 1737 __le64 xattr_buf = cpu_to_le64(inline_version); 1738 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, 1739 "inline_version", &xattr_buf, 1740 sizeof(xattr_buf), 1741 CEPH_OSD_CMPXATTR_OP_GT, 1742 CEPH_OSD_CMPXATTR_MODE_U64); 1743 if (err) 1744 goto out_put; 1745 } 1746 1747 { 1748 char xattr_buf[32]; 1749 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), 1750 "%llu", inline_version); 1751 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, 1752 "inline_version", 1753 xattr_buf, xattr_len, 0, 0); 1754 if (err) 1755 goto out_put; 1756 } 1757 1758 req->r_mtime = inode->i_mtime; 1759 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1760 if (!err) 1761 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1762 out_put: 1763 ceph_osdc_put_request(req); 1764 if (err == -ECANCELED) 1765 err = 0; 1766 out: 1767 if (page && page != locked_page) { 1768 if (from_pagecache) { 1769 unlock_page(page); 1770 put_page(page); 1771 } else 1772 __free_pages(page, 0); 1773 } 1774 1775 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", 1776 inode, ceph_vinop(inode), inline_version, err); 1777 return err; 1778 } 1779 1780 static const struct vm_operations_struct ceph_vmops = { 1781 .fault = ceph_filemap_fault, 1782 .page_mkwrite = ceph_page_mkwrite, 1783 }; 1784 1785 int ceph_mmap(struct file *file, struct vm_area_struct *vma) 1786 { 1787 struct address_space *mapping = file->f_mapping; 1788 1789 if (!mapping->a_ops->readpage) 1790 return -ENOEXEC; 1791 file_accessed(file); 1792 vma->vm_ops = &ceph_vmops; 1793 return 0; 1794 } 1795 1796 enum { 1797 POOL_READ = 1, 1798 POOL_WRITE = 2, 1799 }; 1800 1801 static int __ceph_pool_perm_get(struct ceph_inode_info *ci, 1802 s64 pool, struct ceph_string *pool_ns) 1803 { 1804 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 1805 struct ceph_mds_client *mdsc = fsc->mdsc; 1806 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; 1807 struct rb_node **p, *parent; 1808 struct ceph_pool_perm *perm; 1809 struct page **pages; 1810 size_t pool_ns_len; 1811 int err = 0, err2 = 0, have = 0; 1812 1813 down_read(&mdsc->pool_perm_rwsem); 1814 p = &mdsc->pool_perm_tree.rb_node; 1815 while (*p) { 1816 perm = rb_entry(*p, struct ceph_pool_perm, node); 1817 if (pool < perm->pool) 1818 p = &(*p)->rb_left; 1819 else if (pool > perm->pool) 1820 p = &(*p)->rb_right; 1821 else { 1822 int ret = ceph_compare_string(pool_ns, 1823 perm->pool_ns, 1824 perm->pool_ns_len); 1825 if (ret < 0) 1826 p = &(*p)->rb_left; 1827 else if (ret > 0) 1828 p = &(*p)->rb_right; 1829 else { 1830 have = perm->perm; 1831 break; 1832 } 1833 } 1834 } 1835 up_read(&mdsc->pool_perm_rwsem); 1836 if (*p) 1837 goto out; 1838 1839 if (pool_ns) 1840 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", 1841 pool, (int)pool_ns->len, pool_ns->str); 1842 else 1843 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); 1844 1845 down_write(&mdsc->pool_perm_rwsem); 1846 p = &mdsc->pool_perm_tree.rb_node; 1847 parent = NULL; 1848 while (*p) { 1849 parent = *p; 1850 perm = rb_entry(parent, struct ceph_pool_perm, node); 1851 if (pool < perm->pool) 1852 p = &(*p)->rb_left; 1853 else if (pool > perm->pool) 1854 p = &(*p)->rb_right; 1855 else { 1856 int ret = ceph_compare_string(pool_ns, 1857 perm->pool_ns, 1858 perm->pool_ns_len); 1859 if (ret < 0) 1860 p = &(*p)->rb_left; 1861 else if (ret > 0) 1862 p = &(*p)->rb_right; 1863 else { 1864 have = perm->perm; 1865 break; 1866 } 1867 } 1868 } 1869 if (*p) { 1870 up_write(&mdsc->pool_perm_rwsem); 1871 goto out; 1872 } 1873 1874 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1875 1, false, GFP_NOFS); 1876 if (!rd_req) { 1877 err = -ENOMEM; 1878 goto out_unlock; 1879 } 1880 1881 rd_req->r_flags = CEPH_OSD_FLAG_READ; 1882 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); 1883 rd_req->r_base_oloc.pool = pool; 1884 if (pool_ns) 1885 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); 1886 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); 1887 1888 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); 1889 if (err) 1890 goto out_unlock; 1891 1892 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1893 1, false, GFP_NOFS); 1894 if (!wr_req) { 1895 err = -ENOMEM; 1896 goto out_unlock; 1897 } 1898 1899 wr_req->r_flags = CEPH_OSD_FLAG_WRITE; 1900 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); 1901 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); 1902 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); 1903 1904 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); 1905 if (err) 1906 goto out_unlock; 1907 1908 /* one page should be large enough for STAT data */ 1909 pages = ceph_alloc_page_vector(1, GFP_KERNEL); 1910 if (IS_ERR(pages)) { 1911 err = PTR_ERR(pages); 1912 goto out_unlock; 1913 } 1914 1915 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, 1916 0, false, true); 1917 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); 1918 1919 wr_req->r_mtime = ci->vfs_inode.i_mtime; 1920 wr_req->r_abort_on_full = true; 1921 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); 1922 1923 if (!err) 1924 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); 1925 if (!err2) 1926 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); 1927 1928 if (err >= 0 || err == -ENOENT) 1929 have |= POOL_READ; 1930 else if (err != -EPERM) 1931 goto out_unlock; 1932 1933 if (err2 == 0 || err2 == -EEXIST) 1934 have |= POOL_WRITE; 1935 else if (err2 != -EPERM) { 1936 err = err2; 1937 goto out_unlock; 1938 } 1939 1940 pool_ns_len = pool_ns ? pool_ns->len : 0; 1941 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); 1942 if (!perm) { 1943 err = -ENOMEM; 1944 goto out_unlock; 1945 } 1946 1947 perm->pool = pool; 1948 perm->perm = have; 1949 perm->pool_ns_len = pool_ns_len; 1950 if (pool_ns_len > 0) 1951 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); 1952 perm->pool_ns[pool_ns_len] = 0; 1953 1954 rb_link_node(&perm->node, parent, p); 1955 rb_insert_color(&perm->node, &mdsc->pool_perm_tree); 1956 err = 0; 1957 out_unlock: 1958 up_write(&mdsc->pool_perm_rwsem); 1959 1960 ceph_osdc_put_request(rd_req); 1961 ceph_osdc_put_request(wr_req); 1962 out: 1963 if (!err) 1964 err = have; 1965 if (pool_ns) 1966 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", 1967 pool, (int)pool_ns->len, pool_ns->str, err); 1968 else 1969 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); 1970 return err; 1971 } 1972 1973 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need) 1974 { 1975 s64 pool; 1976 struct ceph_string *pool_ns; 1977 int ret, flags; 1978 1979 if (ci->i_vino.snap != CEPH_NOSNAP) { 1980 /* 1981 * Pool permission check needs to write to the first object. 1982 * But for snapshot, head of the first object may have alread 1983 * been deleted. Skip check to avoid creating orphan object. 1984 */ 1985 return 0; 1986 } 1987 1988 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode), 1989 NOPOOLPERM)) 1990 return 0; 1991 1992 spin_lock(&ci->i_ceph_lock); 1993 flags = ci->i_ceph_flags; 1994 pool = ci->i_layout.pool_id; 1995 spin_unlock(&ci->i_ceph_lock); 1996 check: 1997 if (flags & CEPH_I_POOL_PERM) { 1998 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { 1999 dout("ceph_pool_perm_check pool %lld no read perm\n", 2000 pool); 2001 return -EPERM; 2002 } 2003 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { 2004 dout("ceph_pool_perm_check pool %lld no write perm\n", 2005 pool); 2006 return -EPERM; 2007 } 2008 return 0; 2009 } 2010 2011 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); 2012 ret = __ceph_pool_perm_get(ci, pool, pool_ns); 2013 ceph_put_string(pool_ns); 2014 if (ret < 0) 2015 return ret; 2016 2017 flags = CEPH_I_POOL_PERM; 2018 if (ret & POOL_READ) 2019 flags |= CEPH_I_POOL_RD; 2020 if (ret & POOL_WRITE) 2021 flags |= CEPH_I_POOL_WR; 2022 2023 spin_lock(&ci->i_ceph_lock); 2024 if (pool == ci->i_layout.pool_id && 2025 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { 2026 ci->i_ceph_flags |= flags; 2027 } else { 2028 pool = ci->i_layout.pool_id; 2029 flags = ci->i_ceph_flags; 2030 } 2031 spin_unlock(&ci->i_ceph_lock); 2032 goto check; 2033 } 2034 2035 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) 2036 { 2037 struct ceph_pool_perm *perm; 2038 struct rb_node *n; 2039 2040 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { 2041 n = rb_first(&mdsc->pool_perm_tree); 2042 perm = rb_entry(n, struct ceph_pool_perm, node); 2043 rb_erase(n, &mdsc->pool_perm_tree); 2044 kfree(perm); 2045 } 2046 } 2047