1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/backing-dev.h> 5 #include <linux/fs.h> 6 #include <linux/mm.h> 7 #include <linux/pagemap.h> 8 #include <linux/writeback.h> /* generic_writepages */ 9 #include <linux/slab.h> 10 #include <linux/pagevec.h> 11 #include <linux/task_io_accounting_ops.h> 12 #include <linux/signal.h> 13 14 #include "super.h" 15 #include "mds_client.h" 16 #include "cache.h" 17 #include <linux/ceph/osd_client.h> 18 19 /* 20 * Ceph address space ops. 21 * 22 * There are a few funny things going on here. 23 * 24 * The page->private field is used to reference a struct 25 * ceph_snap_context for _every_ dirty page. This indicates which 26 * snapshot the page was logically dirtied in, and thus which snap 27 * context needs to be associated with the osd write during writeback. 28 * 29 * Similarly, struct ceph_inode_info maintains a set of counters to 30 * count dirty pages on the inode. In the absence of snapshots, 31 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 32 * 33 * When a snapshot is taken (that is, when the client receives 34 * notification that a snapshot was taken), each inode with caps and 35 * with dirty pages (dirty pages implies there is a cap) gets a new 36 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 37 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 38 * moved to capsnap->dirty. (Unless a sync write is currently in 39 * progress. In that case, the capsnap is said to be "pending", new 40 * writes cannot start, and the capsnap isn't "finalized" until the 41 * write completes (or fails) and a final size/mtime for the inode for 42 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 43 * 44 * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 45 * we look for the first capsnap in i_cap_snaps and write out pages in 46 * that snap context _only_. Then we move on to the next capsnap, 47 * eventually reaching the "live" or "head" context (i.e., pages that 48 * are not yet snapped) and are writing the most recently dirtied 49 * pages. 50 * 51 * Invalidate and so forth must take care to ensure the dirty page 52 * accounting is preserved. 53 */ 54 55 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 56 #define CONGESTION_OFF_THRESH(congestion_kb) \ 57 (CONGESTION_ON_THRESH(congestion_kb) - \ 58 (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 59 60 static inline struct ceph_snap_context *page_snap_context(struct page *page) 61 { 62 if (PagePrivate(page)) 63 return (void *)page->private; 64 return NULL; 65 } 66 67 /* 68 * Dirty a page. Optimistically adjust accounting, on the assumption 69 * that we won't race with invalidate. If we do, readjust. 70 */ 71 static int ceph_set_page_dirty(struct page *page) 72 { 73 struct address_space *mapping = page->mapping; 74 struct inode *inode; 75 struct ceph_inode_info *ci; 76 struct ceph_snap_context *snapc; 77 int ret; 78 79 if (unlikely(!mapping)) 80 return !TestSetPageDirty(page); 81 82 if (PageDirty(page)) { 83 dout("%p set_page_dirty %p idx %lu -- already dirty\n", 84 mapping->host, page, page->index); 85 BUG_ON(!PagePrivate(page)); 86 return 0; 87 } 88 89 inode = mapping->host; 90 ci = ceph_inode(inode); 91 92 /* dirty the head */ 93 spin_lock(&ci->i_ceph_lock); 94 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference 95 if (__ceph_have_pending_cap_snap(ci)) { 96 struct ceph_cap_snap *capsnap = 97 list_last_entry(&ci->i_cap_snaps, 98 struct ceph_cap_snap, 99 ci_item); 100 snapc = ceph_get_snap_context(capsnap->context); 101 capsnap->dirty_pages++; 102 } else { 103 BUG_ON(!ci->i_head_snapc); 104 snapc = ceph_get_snap_context(ci->i_head_snapc); 105 ++ci->i_wrbuffer_ref_head; 106 } 107 if (ci->i_wrbuffer_ref == 0) 108 ihold(inode); 109 ++ci->i_wrbuffer_ref; 110 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " 111 "snapc %p seq %lld (%d snaps)\n", 112 mapping->host, page, page->index, 113 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 114 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 115 snapc, snapc->seq, snapc->num_snaps); 116 spin_unlock(&ci->i_ceph_lock); 117 118 /* 119 * Reference snap context in page->private. Also set 120 * PagePrivate so that we get invalidatepage callback. 121 */ 122 BUG_ON(PagePrivate(page)); 123 page->private = (unsigned long)snapc; 124 SetPagePrivate(page); 125 126 ret = __set_page_dirty_nobuffers(page); 127 WARN_ON(!PageLocked(page)); 128 WARN_ON(!page->mapping); 129 130 return ret; 131 } 132 133 /* 134 * If we are truncating the full page (i.e. offset == 0), adjust the 135 * dirty page counters appropriately. Only called if there is private 136 * data on the page. 137 */ 138 static void ceph_invalidatepage(struct page *page, unsigned int offset, 139 unsigned int length) 140 { 141 struct inode *inode; 142 struct ceph_inode_info *ci; 143 struct ceph_snap_context *snapc = page_snap_context(page); 144 145 inode = page->mapping->host; 146 ci = ceph_inode(inode); 147 148 if (offset != 0 || length != PAGE_SIZE) { 149 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", 150 inode, page, page->index, offset, length); 151 return; 152 } 153 154 ceph_invalidate_fscache_page(inode, page); 155 156 WARN_ON(!PageLocked(page)); 157 if (!PagePrivate(page)) 158 return; 159 160 ClearPageChecked(page); 161 162 dout("%p invalidatepage %p idx %lu full dirty page\n", 163 inode, page, page->index); 164 165 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 166 ceph_put_snap_context(snapc); 167 page->private = 0; 168 ClearPagePrivate(page); 169 } 170 171 static int ceph_releasepage(struct page *page, gfp_t g) 172 { 173 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host, 174 page, page->index, PageDirty(page) ? "" : "not "); 175 176 /* Can we release the page from the cache? */ 177 if (!ceph_release_fscache_page(page, g)) 178 return 0; 179 180 return !PagePrivate(page); 181 } 182 183 /* 184 * read a single page, without unlocking it. 185 */ 186 static int ceph_do_readpage(struct file *filp, struct page *page) 187 { 188 struct inode *inode = file_inode(filp); 189 struct ceph_inode_info *ci = ceph_inode(inode); 190 struct ceph_osd_client *osdc = 191 &ceph_inode_to_client(inode)->client->osdc; 192 int err = 0; 193 u64 off = page_offset(page); 194 u64 len = PAGE_SIZE; 195 196 if (off >= i_size_read(inode)) { 197 zero_user_segment(page, 0, PAGE_SIZE); 198 SetPageUptodate(page); 199 return 0; 200 } 201 202 if (ci->i_inline_version != CEPH_INLINE_NONE) { 203 /* 204 * Uptodate inline data should have been added 205 * into page cache while getting Fcr caps. 206 */ 207 if (off == 0) 208 return -EINVAL; 209 zero_user_segment(page, 0, PAGE_SIZE); 210 SetPageUptodate(page); 211 return 0; 212 } 213 214 err = ceph_readpage_from_fscache(inode, page); 215 if (err == 0) 216 return -EINPROGRESS; 217 218 dout("readpage inode %p file %p page %p index %lu\n", 219 inode, filp, page, page->index); 220 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, 221 off, &len, 222 ci->i_truncate_seq, ci->i_truncate_size, 223 &page, 1, 0); 224 if (err == -ENOENT) 225 err = 0; 226 if (err < 0) { 227 SetPageError(page); 228 ceph_fscache_readpage_cancel(inode, page); 229 goto out; 230 } 231 if (err < PAGE_SIZE) 232 /* zero fill remainder of page */ 233 zero_user_segment(page, err, PAGE_SIZE); 234 else 235 flush_dcache_page(page); 236 237 SetPageUptodate(page); 238 ceph_readpage_to_fscache(inode, page); 239 240 out: 241 return err < 0 ? err : 0; 242 } 243 244 static int ceph_readpage(struct file *filp, struct page *page) 245 { 246 int r = ceph_do_readpage(filp, page); 247 if (r != -EINPROGRESS) 248 unlock_page(page); 249 else 250 r = 0; 251 return r; 252 } 253 254 /* 255 * Finish an async read(ahead) op. 256 */ 257 static void finish_read(struct ceph_osd_request *req) 258 { 259 struct inode *inode = req->r_inode; 260 struct ceph_osd_data *osd_data; 261 int rc = req->r_result <= 0 ? req->r_result : 0; 262 int bytes = req->r_result >= 0 ? req->r_result : 0; 263 int num_pages; 264 int i; 265 266 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); 267 268 /* unlock all pages, zeroing any data we didn't read */ 269 osd_data = osd_req_op_extent_osd_data(req, 0); 270 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 271 num_pages = calc_pages_for((u64)osd_data->alignment, 272 (u64)osd_data->length); 273 for (i = 0; i < num_pages; i++) { 274 struct page *page = osd_data->pages[i]; 275 276 if (rc < 0 && rc != -ENOENT) { 277 ceph_fscache_readpage_cancel(inode, page); 278 goto unlock; 279 } 280 if (bytes < (int)PAGE_SIZE) { 281 /* zero (remainder of) page */ 282 int s = bytes < 0 ? 0 : bytes; 283 zero_user_segment(page, s, PAGE_SIZE); 284 } 285 dout("finish_read %p uptodate %p idx %lu\n", inode, page, 286 page->index); 287 flush_dcache_page(page); 288 SetPageUptodate(page); 289 ceph_readpage_to_fscache(inode, page); 290 unlock: 291 unlock_page(page); 292 put_page(page); 293 bytes -= PAGE_SIZE; 294 } 295 kfree(osd_data->pages); 296 } 297 298 /* 299 * start an async read(ahead) operation. return nr_pages we submitted 300 * a read for on success, or negative error code. 301 */ 302 static int start_read(struct inode *inode, struct ceph_rw_context *rw_ctx, 303 struct list_head *page_list, int max) 304 { 305 struct ceph_osd_client *osdc = 306 &ceph_inode_to_client(inode)->client->osdc; 307 struct ceph_inode_info *ci = ceph_inode(inode); 308 struct page *page = list_entry(page_list->prev, struct page, lru); 309 struct ceph_vino vino; 310 struct ceph_osd_request *req; 311 u64 off; 312 u64 len; 313 int i; 314 struct page **pages; 315 pgoff_t next_index; 316 int nr_pages = 0; 317 int got = 0; 318 int ret = 0; 319 320 if (!rw_ctx) { 321 /* caller of readpages does not hold buffer and read caps 322 * (fadvise, madvise and readahead cases) */ 323 int want = CEPH_CAP_FILE_CACHE; 324 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got); 325 if (ret < 0) { 326 dout("start_read %p, error getting cap\n", inode); 327 } else if (!(got & want)) { 328 dout("start_read %p, no cache cap\n", inode); 329 ret = 0; 330 } 331 if (ret <= 0) { 332 if (got) 333 ceph_put_cap_refs(ci, got); 334 while (!list_empty(page_list)) { 335 page = list_entry(page_list->prev, 336 struct page, lru); 337 list_del(&page->lru); 338 put_page(page); 339 } 340 return ret; 341 } 342 } 343 344 off = (u64) page_offset(page); 345 346 /* count pages */ 347 next_index = page->index; 348 list_for_each_entry_reverse(page, page_list, lru) { 349 if (page->index != next_index) 350 break; 351 nr_pages++; 352 next_index++; 353 if (max && nr_pages == max) 354 break; 355 } 356 len = nr_pages << PAGE_SHIFT; 357 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, 358 off, len); 359 vino = ceph_vino(inode); 360 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, 361 0, 1, CEPH_OSD_OP_READ, 362 CEPH_OSD_FLAG_READ, NULL, 363 ci->i_truncate_seq, ci->i_truncate_size, 364 false); 365 if (IS_ERR(req)) { 366 ret = PTR_ERR(req); 367 goto out; 368 } 369 370 /* build page vector */ 371 nr_pages = calc_pages_for(0, len); 372 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL); 373 if (!pages) { 374 ret = -ENOMEM; 375 goto out_put; 376 } 377 for (i = 0; i < nr_pages; ++i) { 378 page = list_entry(page_list->prev, struct page, lru); 379 BUG_ON(PageLocked(page)); 380 list_del(&page->lru); 381 382 dout("start_read %p adding %p idx %lu\n", inode, page, 383 page->index); 384 if (add_to_page_cache_lru(page, &inode->i_data, page->index, 385 GFP_KERNEL)) { 386 ceph_fscache_uncache_page(inode, page); 387 put_page(page); 388 dout("start_read %p add_to_page_cache failed %p\n", 389 inode, page); 390 nr_pages = i; 391 if (nr_pages > 0) { 392 len = nr_pages << PAGE_SHIFT; 393 osd_req_op_extent_update(req, 0, len); 394 break; 395 } 396 goto out_pages; 397 } 398 pages[i] = page; 399 } 400 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); 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); 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); 414 415 return nr_pages; 416 417 out_pages: 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); 423 out_put: 424 ceph_osdc_put_request(req); 425 out: 426 if (got) 427 ceph_put_cap_refs(ci, got); 428 return ret; 429 } 430 431 432 /* 433 * Read multiple pages. Leave pages we don't read + unlock in page_list; 434 * the caller (VM) cleans them up. 435 */ 436 static int ceph_readpages(struct file *file, struct address_space *mapping, 437 struct list_head *page_list, unsigned nr_pages) 438 { 439 struct inode *inode = file_inode(file); 440 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 441 struct ceph_file_info *ci = file->private_data; 442 struct ceph_rw_context *rw_ctx; 443 int rc = 0; 444 int max = 0; 445 446 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) 447 return -EINVAL; 448 449 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list, 450 &nr_pages); 451 452 if (rc == 0) 453 goto out; 454 455 rw_ctx = ceph_find_rw_context(ci); 456 max = fsc->mount_options->rsize >> PAGE_SHIFT; 457 dout("readpages %p file %p ctx %p nr_pages %d max %d\n", 458 inode, file, rw_ctx, nr_pages, max); 459 while (!list_empty(page_list)) { 460 rc = start_read(inode, rw_ctx, page_list, max); 461 if (rc < 0) 462 goto out; 463 } 464 out: 465 ceph_fscache_readpages_cancel(inode, page_list); 466 467 dout("readpages %p file %p ret %d\n", inode, file, rc); 468 return rc; 469 } 470 471 struct ceph_writeback_ctl 472 { 473 loff_t i_size; 474 u64 truncate_size; 475 u32 truncate_seq; 476 bool size_stable; 477 bool head_snapc; 478 }; 479 480 /* 481 * Get ref for the oldest snapc for an inode with dirty data... that is, the 482 * only snap context we are allowed to write back. 483 */ 484 static struct ceph_snap_context * 485 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, 486 struct ceph_snap_context *page_snapc) 487 { 488 struct ceph_inode_info *ci = ceph_inode(inode); 489 struct ceph_snap_context *snapc = NULL; 490 struct ceph_cap_snap *capsnap = NULL; 491 492 spin_lock(&ci->i_ceph_lock); 493 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 494 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 495 capsnap->context, capsnap->dirty_pages); 496 if (!capsnap->dirty_pages) 497 continue; 498 499 /* get i_size, truncate_{seq,size} for page_snapc? */ 500 if (snapc && capsnap->context != page_snapc) 501 continue; 502 503 if (ctl) { 504 if (capsnap->writing) { 505 ctl->i_size = i_size_read(inode); 506 ctl->size_stable = false; 507 } else { 508 ctl->i_size = capsnap->size; 509 ctl->size_stable = true; 510 } 511 ctl->truncate_size = capsnap->truncate_size; 512 ctl->truncate_seq = capsnap->truncate_seq; 513 ctl->head_snapc = false; 514 } 515 516 if (snapc) 517 break; 518 519 snapc = ceph_get_snap_context(capsnap->context); 520 if (!page_snapc || 521 page_snapc == snapc || 522 page_snapc->seq > snapc->seq) 523 break; 524 } 525 if (!snapc && ci->i_wrbuffer_ref_head) { 526 snapc = ceph_get_snap_context(ci->i_head_snapc); 527 dout(" head snapc %p has %d dirty pages\n", 528 snapc, ci->i_wrbuffer_ref_head); 529 if (ctl) { 530 ctl->i_size = i_size_read(inode); 531 ctl->truncate_size = ci->i_truncate_size; 532 ctl->truncate_seq = ci->i_truncate_seq; 533 ctl->size_stable = false; 534 ctl->head_snapc = true; 535 } 536 } 537 spin_unlock(&ci->i_ceph_lock); 538 return snapc; 539 } 540 541 static u64 get_writepages_data_length(struct inode *inode, 542 struct page *page, u64 start) 543 { 544 struct ceph_inode_info *ci = ceph_inode(inode); 545 struct ceph_snap_context *snapc = page_snap_context(page); 546 struct ceph_cap_snap *capsnap = NULL; 547 u64 end = i_size_read(inode); 548 549 if (snapc != ci->i_head_snapc) { 550 bool found = false; 551 spin_lock(&ci->i_ceph_lock); 552 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 553 if (capsnap->context == snapc) { 554 if (!capsnap->writing) 555 end = capsnap->size; 556 found = true; 557 break; 558 } 559 } 560 spin_unlock(&ci->i_ceph_lock); 561 WARN_ON(!found); 562 } 563 if (end > page_offset(page) + PAGE_SIZE) 564 end = page_offset(page) + PAGE_SIZE; 565 return end > start ? end - start : 0; 566 } 567 568 /* 569 * Write a single page, but leave the page locked. 570 * 571 * If we get a write error, set the page error bit, but still adjust the 572 * dirty page accounting (i.e., page is no longer dirty). 573 */ 574 static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 575 { 576 struct inode *inode; 577 struct ceph_inode_info *ci; 578 struct ceph_fs_client *fsc; 579 struct ceph_snap_context *snapc, *oldest; 580 loff_t page_off = page_offset(page); 581 int err, len = PAGE_SIZE; 582 struct ceph_writeback_ctl ceph_wbc; 583 584 dout("writepage %p idx %lu\n", page, page->index); 585 586 inode = page->mapping->host; 587 ci = ceph_inode(inode); 588 fsc = ceph_inode_to_client(inode); 589 590 /* verify this is a writeable snap context */ 591 snapc = page_snap_context(page); 592 if (!snapc) { 593 dout("writepage %p page %p not dirty?\n", inode, page); 594 return 0; 595 } 596 oldest = get_oldest_context(inode, &ceph_wbc, snapc); 597 if (snapc->seq > oldest->seq) { 598 dout("writepage %p page %p snapc %p not writeable - noop\n", 599 inode, page, snapc); 600 /* we should only noop if called by kswapd */ 601 WARN_ON(!(current->flags & PF_MEMALLOC)); 602 ceph_put_snap_context(oldest); 603 redirty_page_for_writepage(wbc, page); 604 return 0; 605 } 606 ceph_put_snap_context(oldest); 607 608 /* is this a partial page at end of file? */ 609 if (page_off >= ceph_wbc.i_size) { 610 dout("%p page eof %llu\n", page, ceph_wbc.i_size); 611 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); 612 return 0; 613 } 614 615 if (ceph_wbc.i_size < page_off + len) 616 len = ceph_wbc.i_size - page_off; 617 618 dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n", 619 inode, page, page->index, page_off, len, snapc, snapc->seq); 620 621 if (atomic_long_inc_return(&fsc->writeback_count) > 622 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 623 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 624 625 set_page_writeback(page); 626 err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode), 627 &ci->i_layout, snapc, page_off, len, 628 ceph_wbc.truncate_seq, 629 ceph_wbc.truncate_size, 630 &inode->i_mtime, &page, 1); 631 if (err < 0) { 632 struct writeback_control tmp_wbc; 633 if (!wbc) 634 wbc = &tmp_wbc; 635 if (err == -ERESTARTSYS) { 636 /* killed by SIGKILL */ 637 dout("writepage interrupted page %p\n", page); 638 redirty_page_for_writepage(wbc, page); 639 end_page_writeback(page); 640 return err; 641 } 642 dout("writepage setting page/mapping error %d %p\n", 643 err, page); 644 SetPageError(page); 645 mapping_set_error(&inode->i_data, err); 646 wbc->pages_skipped++; 647 } else { 648 dout("writepage cleaned page %p\n", page); 649 err = 0; /* vfs expects us to return 0 */ 650 } 651 page->private = 0; 652 ClearPagePrivate(page); 653 end_page_writeback(page); 654 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 655 ceph_put_snap_context(snapc); /* page's reference */ 656 657 if (atomic_long_dec_return(&fsc->writeback_count) < 658 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) 659 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 660 661 return err; 662 } 663 664 static int ceph_writepage(struct page *page, struct writeback_control *wbc) 665 { 666 int err; 667 struct inode *inode = page->mapping->host; 668 BUG_ON(!inode); 669 ihold(inode); 670 err = writepage_nounlock(page, wbc); 671 if (err == -ERESTARTSYS) { 672 /* direct memory reclaimer was killed by SIGKILL. return 0 673 * to prevent caller from setting mapping/page error */ 674 err = 0; 675 } 676 unlock_page(page); 677 iput(inode); 678 return err; 679 } 680 681 /* 682 * lame release_pages helper. release_pages() isn't exported to 683 * modules. 684 */ 685 static void ceph_release_pages(struct page **pages, int num) 686 { 687 struct pagevec pvec; 688 int i; 689 690 pagevec_init(&pvec); 691 for (i = 0; i < num; i++) { 692 if (pagevec_add(&pvec, pages[i]) == 0) 693 pagevec_release(&pvec); 694 } 695 pagevec_release(&pvec); 696 } 697 698 /* 699 * async writeback completion handler. 700 * 701 * If we get an error, set the mapping error bit, but not the individual 702 * page error bits. 703 */ 704 static void writepages_finish(struct ceph_osd_request *req) 705 { 706 struct inode *inode = req->r_inode; 707 struct ceph_inode_info *ci = ceph_inode(inode); 708 struct ceph_osd_data *osd_data; 709 struct page *page; 710 int num_pages, total_pages = 0; 711 int i, j; 712 int rc = req->r_result; 713 struct ceph_snap_context *snapc = req->r_snapc; 714 struct address_space *mapping = inode->i_mapping; 715 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 716 bool remove_page; 717 718 dout("writepages_finish %p rc %d\n", inode, rc); 719 if (rc < 0) { 720 mapping_set_error(mapping, rc); 721 ceph_set_error_write(ci); 722 } else { 723 ceph_clear_error_write(ci); 724 } 725 726 /* 727 * We lost the cache cap, need to truncate the page before 728 * it is unlocked, otherwise we'd truncate it later in the 729 * page truncation thread, possibly losing some data that 730 * raced its way in 731 */ 732 remove_page = !(ceph_caps_issued(ci) & 733 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); 734 735 /* clean all pages */ 736 for (i = 0; i < req->r_num_ops; i++) { 737 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) 738 break; 739 740 osd_data = osd_req_op_extent_osd_data(req, i); 741 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 742 num_pages = calc_pages_for((u64)osd_data->alignment, 743 (u64)osd_data->length); 744 total_pages += num_pages; 745 for (j = 0; j < num_pages; j++) { 746 page = osd_data->pages[j]; 747 BUG_ON(!page); 748 WARN_ON(!PageUptodate(page)); 749 750 if (atomic_long_dec_return(&fsc->writeback_count) < 751 CONGESTION_OFF_THRESH( 752 fsc->mount_options->congestion_kb)) 753 clear_bdi_congested(inode_to_bdi(inode), 754 BLK_RW_ASYNC); 755 756 ceph_put_snap_context(page_snap_context(page)); 757 page->private = 0; 758 ClearPagePrivate(page); 759 dout("unlocking %p\n", page); 760 end_page_writeback(page); 761 762 if (remove_page) 763 generic_error_remove_page(inode->i_mapping, 764 page); 765 766 unlock_page(page); 767 } 768 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", 769 inode, osd_data->length, rc >= 0 ? num_pages : 0); 770 771 ceph_release_pages(osd_data->pages, num_pages); 772 } 773 774 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); 775 776 osd_data = osd_req_op_extent_osd_data(req, 0); 777 if (osd_data->pages_from_pool) 778 mempool_free(osd_data->pages, 779 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); 780 else 781 kfree(osd_data->pages); 782 ceph_osdc_put_request(req); 783 } 784 785 /* 786 * initiate async writeback 787 */ 788 static int ceph_writepages_start(struct address_space *mapping, 789 struct writeback_control *wbc) 790 { 791 struct inode *inode = mapping->host; 792 struct ceph_inode_info *ci = ceph_inode(inode); 793 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 794 struct ceph_vino vino = ceph_vino(inode); 795 pgoff_t index, start_index, end = -1; 796 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 797 struct pagevec pvec; 798 int rc = 0; 799 unsigned int wsize = i_blocksize(inode); 800 struct ceph_osd_request *req = NULL; 801 struct ceph_writeback_ctl ceph_wbc; 802 bool should_loop, range_whole = false; 803 bool stop, done = false; 804 805 dout("writepages_start %p (mode=%s)\n", inode, 806 wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 807 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 808 809 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 810 if (ci->i_wrbuffer_ref > 0) { 811 pr_warn_ratelimited( 812 "writepage_start %p %lld forced umount\n", 813 inode, ceph_ino(inode)); 814 } 815 mapping_set_error(mapping, -EIO); 816 return -EIO; /* we're in a forced umount, don't write! */ 817 } 818 if (fsc->mount_options->wsize < wsize) 819 wsize = fsc->mount_options->wsize; 820 821 pagevec_init(&pvec); 822 823 start_index = wbc->range_cyclic ? mapping->writeback_index : 0; 824 index = start_index; 825 826 retry: 827 /* find oldest snap context with dirty data */ 828 snapc = get_oldest_context(inode, &ceph_wbc, NULL); 829 if (!snapc) { 830 /* hmm, why does writepages get called when there 831 is no dirty data? */ 832 dout(" no snap context with dirty data?\n"); 833 goto out; 834 } 835 dout(" oldest snapc is %p seq %lld (%d snaps)\n", 836 snapc, snapc->seq, snapc->num_snaps); 837 838 should_loop = false; 839 if (ceph_wbc.head_snapc && snapc != last_snapc) { 840 /* where to start/end? */ 841 if (wbc->range_cyclic) { 842 index = start_index; 843 end = -1; 844 if (index > 0) 845 should_loop = true; 846 dout(" cyclic, start at %lu\n", index); 847 } else { 848 index = wbc->range_start >> PAGE_SHIFT; 849 end = wbc->range_end >> PAGE_SHIFT; 850 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 851 range_whole = true; 852 dout(" not cyclic, %lu to %lu\n", index, end); 853 } 854 } else if (!ceph_wbc.head_snapc) { 855 /* Do not respect wbc->range_{start,end}. Dirty pages 856 * in that range can be associated with newer snapc. 857 * They are not writeable until we write all dirty pages 858 * associated with 'snapc' get written */ 859 if (index > 0 || wbc->sync_mode != WB_SYNC_NONE) 860 should_loop = true; 861 dout(" non-head snapc, range whole\n"); 862 } 863 864 ceph_put_snap_context(last_snapc); 865 last_snapc = snapc; 866 867 stop = false; 868 while (!stop && index <= end) { 869 int num_ops = 0, op_idx; 870 unsigned i, pvec_pages, max_pages, locked_pages = 0; 871 struct page **pages = NULL, **data_pages; 872 mempool_t *pool = NULL; /* Becomes non-null if mempool used */ 873 struct page *page; 874 pgoff_t strip_unit_end = 0; 875 u64 offset = 0, len = 0; 876 877 max_pages = wsize >> PAGE_SHIFT; 878 879 get_more_pages: 880 pvec_pages = pagevec_lookup_range_nr_tag(&pvec, mapping, &index, 881 end, PAGECACHE_TAG_DIRTY, 882 max_pages - locked_pages); 883 dout("pagevec_lookup_range_tag got %d\n", pvec_pages); 884 if (!pvec_pages && !locked_pages) 885 break; 886 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 887 page = pvec.pages[i]; 888 dout("? %p idx %lu\n", page, page->index); 889 if (locked_pages == 0) 890 lock_page(page); /* first page */ 891 else if (!trylock_page(page)) 892 break; 893 894 /* only dirty pages, or our accounting breaks */ 895 if (unlikely(!PageDirty(page)) || 896 unlikely(page->mapping != mapping)) { 897 dout("!dirty or !mapping %p\n", page); 898 unlock_page(page); 899 continue; 900 } 901 if (strip_unit_end && (page->index > strip_unit_end)) { 902 dout("end of strip unit %p\n", page); 903 unlock_page(page); 904 break; 905 } 906 if (page_offset(page) >= ceph_wbc.i_size) { 907 dout("%p page eof %llu\n", 908 page, ceph_wbc.i_size); 909 /* not done if range_cyclic */ 910 stop = true; 911 unlock_page(page); 912 break; 913 } 914 if (PageWriteback(page)) { 915 if (wbc->sync_mode == WB_SYNC_NONE) { 916 dout("%p under writeback\n", page); 917 unlock_page(page); 918 continue; 919 } 920 dout("waiting on writeback %p\n", page); 921 wait_on_page_writeback(page); 922 } 923 924 /* only if matching snap context */ 925 pgsnapc = page_snap_context(page); 926 if (pgsnapc != snapc) { 927 dout("page snapc %p %lld != oldest %p %lld\n", 928 pgsnapc, pgsnapc->seq, snapc, snapc->seq); 929 unlock_page(page); 930 continue; 931 } 932 933 if (!clear_page_dirty_for_io(page)) { 934 dout("%p !clear_page_dirty_for_io\n", page); 935 unlock_page(page); 936 continue; 937 } 938 939 /* 940 * We have something to write. If this is 941 * the first locked page this time through, 942 * calculate max possinle write size and 943 * allocate a page array 944 */ 945 if (locked_pages == 0) { 946 u64 objnum; 947 u64 objoff; 948 u32 xlen; 949 950 /* prepare async write request */ 951 offset = (u64)page_offset(page); 952 ceph_calc_file_object_mapping(&ci->i_layout, 953 offset, wsize, 954 &objnum, &objoff, 955 &xlen); 956 len = xlen; 957 958 num_ops = 1; 959 strip_unit_end = page->index + 960 ((len - 1) >> PAGE_SHIFT); 961 962 BUG_ON(pages); 963 max_pages = calc_pages_for(0, (u64)len); 964 pages = kmalloc(max_pages * sizeof (*pages), 965 GFP_NOFS); 966 if (!pages) { 967 pool = fsc->wb_pagevec_pool; 968 pages = mempool_alloc(pool, GFP_NOFS); 969 BUG_ON(!pages); 970 } 971 972 len = 0; 973 } else if (page->index != 974 (offset + len) >> PAGE_SHIFT) { 975 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS : 976 CEPH_OSD_MAX_OPS)) { 977 redirty_page_for_writepage(wbc, page); 978 unlock_page(page); 979 break; 980 } 981 982 num_ops++; 983 offset = (u64)page_offset(page); 984 len = 0; 985 } 986 987 /* note position of first page in pvec */ 988 dout("%p will write page %p idx %lu\n", 989 inode, page, page->index); 990 991 if (atomic_long_inc_return(&fsc->writeback_count) > 992 CONGESTION_ON_THRESH( 993 fsc->mount_options->congestion_kb)) { 994 set_bdi_congested(inode_to_bdi(inode), 995 BLK_RW_ASYNC); 996 } 997 998 999 pages[locked_pages++] = page; 1000 pvec.pages[i] = NULL; 1001 1002 len += PAGE_SIZE; 1003 } 1004 1005 /* did we get anything? */ 1006 if (!locked_pages) 1007 goto release_pvec_pages; 1008 if (i) { 1009 unsigned j, n = 0; 1010 /* shift unused page to beginning of pvec */ 1011 for (j = 0; j < pvec_pages; j++) { 1012 if (!pvec.pages[j]) 1013 continue; 1014 if (n < j) 1015 pvec.pages[n] = pvec.pages[j]; 1016 n++; 1017 } 1018 pvec.nr = n; 1019 1020 if (pvec_pages && i == pvec_pages && 1021 locked_pages < max_pages) { 1022 dout("reached end pvec, trying for more\n"); 1023 pagevec_release(&pvec); 1024 goto get_more_pages; 1025 } 1026 } 1027 1028 new_request: 1029 offset = page_offset(pages[0]); 1030 len = wsize; 1031 1032 req = ceph_osdc_new_request(&fsc->client->osdc, 1033 &ci->i_layout, vino, 1034 offset, &len, 0, num_ops, 1035 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1036 snapc, ceph_wbc.truncate_seq, 1037 ceph_wbc.truncate_size, false); 1038 if (IS_ERR(req)) { 1039 req = ceph_osdc_new_request(&fsc->client->osdc, 1040 &ci->i_layout, vino, 1041 offset, &len, 0, 1042 min(num_ops, 1043 CEPH_OSD_SLAB_OPS), 1044 CEPH_OSD_OP_WRITE, 1045 CEPH_OSD_FLAG_WRITE, 1046 snapc, ceph_wbc.truncate_seq, 1047 ceph_wbc.truncate_size, true); 1048 BUG_ON(IS_ERR(req)); 1049 } 1050 BUG_ON(len < page_offset(pages[locked_pages - 1]) + 1051 PAGE_SIZE - offset); 1052 1053 req->r_callback = writepages_finish; 1054 req->r_inode = inode; 1055 1056 /* Format the osd request message and submit the write */ 1057 len = 0; 1058 data_pages = pages; 1059 op_idx = 0; 1060 for (i = 0; i < locked_pages; i++) { 1061 u64 cur_offset = page_offset(pages[i]); 1062 if (offset + len != cur_offset) { 1063 if (op_idx + 1 == req->r_num_ops) 1064 break; 1065 osd_req_op_extent_dup_last(req, op_idx, 1066 cur_offset - offset); 1067 dout("writepages got pages at %llu~%llu\n", 1068 offset, len); 1069 osd_req_op_extent_osd_data_pages(req, op_idx, 1070 data_pages, len, 0, 1071 !!pool, false); 1072 osd_req_op_extent_update(req, op_idx, len); 1073 1074 len = 0; 1075 offset = cur_offset; 1076 data_pages = pages + i; 1077 op_idx++; 1078 } 1079 1080 set_page_writeback(pages[i]); 1081 len += PAGE_SIZE; 1082 } 1083 1084 if (ceph_wbc.size_stable) { 1085 len = min(len, ceph_wbc.i_size - offset); 1086 } else if (i == locked_pages) { 1087 /* writepages_finish() clears writeback pages 1088 * according to the data length, so make sure 1089 * data length covers all locked pages */ 1090 u64 min_len = len + 1 - PAGE_SIZE; 1091 len = get_writepages_data_length(inode, pages[i - 1], 1092 offset); 1093 len = max(len, min_len); 1094 } 1095 dout("writepages got pages at %llu~%llu\n", offset, len); 1096 1097 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, 1098 0, !!pool, false); 1099 osd_req_op_extent_update(req, op_idx, len); 1100 1101 BUG_ON(op_idx + 1 != req->r_num_ops); 1102 1103 pool = NULL; 1104 if (i < locked_pages) { 1105 BUG_ON(num_ops <= req->r_num_ops); 1106 num_ops -= req->r_num_ops; 1107 locked_pages -= i; 1108 1109 /* allocate new pages array for next request */ 1110 data_pages = pages; 1111 pages = kmalloc(locked_pages * sizeof (*pages), 1112 GFP_NOFS); 1113 if (!pages) { 1114 pool = fsc->wb_pagevec_pool; 1115 pages = mempool_alloc(pool, GFP_NOFS); 1116 BUG_ON(!pages); 1117 } 1118 memcpy(pages, data_pages + i, 1119 locked_pages * sizeof(*pages)); 1120 memset(data_pages + i, 0, 1121 locked_pages * sizeof(*pages)); 1122 } else { 1123 BUG_ON(num_ops != req->r_num_ops); 1124 index = pages[i - 1]->index + 1; 1125 /* request message now owns the pages array */ 1126 pages = NULL; 1127 } 1128 1129 req->r_mtime = inode->i_mtime; 1130 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); 1131 BUG_ON(rc); 1132 req = NULL; 1133 1134 wbc->nr_to_write -= i; 1135 if (pages) 1136 goto new_request; 1137 1138 /* 1139 * We stop writing back only if we are not doing 1140 * integrity sync. In case of integrity sync we have to 1141 * keep going until we have written all the pages 1142 * we tagged for writeback prior to entering this loop. 1143 */ 1144 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) 1145 done = stop = true; 1146 1147 release_pvec_pages: 1148 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 1149 pvec.nr ? pvec.pages[0] : NULL); 1150 pagevec_release(&pvec); 1151 } 1152 1153 if (should_loop && !done) { 1154 /* more to do; loop back to beginning of file */ 1155 dout("writepages looping back to beginning of file\n"); 1156 end = start_index - 1; /* OK even when start_index == 0 */ 1157 1158 /* to write dirty pages associated with next snapc, 1159 * we need to wait until current writes complete */ 1160 if (wbc->sync_mode != WB_SYNC_NONE && 1161 start_index == 0 && /* all dirty pages were checked */ 1162 !ceph_wbc.head_snapc) { 1163 struct page *page; 1164 unsigned i, nr; 1165 index = 0; 1166 while ((index <= end) && 1167 (nr = pagevec_lookup_tag(&pvec, mapping, &index, 1168 PAGECACHE_TAG_WRITEBACK))) { 1169 for (i = 0; i < nr; i++) { 1170 page = pvec.pages[i]; 1171 if (page_snap_context(page) != snapc) 1172 continue; 1173 wait_on_page_writeback(page); 1174 } 1175 pagevec_release(&pvec); 1176 cond_resched(); 1177 } 1178 } 1179 1180 start_index = 0; 1181 index = 0; 1182 goto retry; 1183 } 1184 1185 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1186 mapping->writeback_index = index; 1187 1188 out: 1189 ceph_osdc_put_request(req); 1190 ceph_put_snap_context(last_snapc); 1191 dout("writepages dend - startone, rc = %d\n", rc); 1192 return rc; 1193 } 1194 1195 1196 1197 /* 1198 * See if a given @snapc is either writeable, or already written. 1199 */ 1200 static int context_is_writeable_or_written(struct inode *inode, 1201 struct ceph_snap_context *snapc) 1202 { 1203 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); 1204 int ret = !oldest || snapc->seq <= oldest->seq; 1205 1206 ceph_put_snap_context(oldest); 1207 return ret; 1208 } 1209 1210 /* 1211 * We are only allowed to write into/dirty the page if the page is 1212 * clean, or already dirty within the same snap context. 1213 * 1214 * called with page locked. 1215 * return success with page locked, 1216 * or any failure (incl -EAGAIN) with page unlocked. 1217 */ 1218 static int ceph_update_writeable_page(struct file *file, 1219 loff_t pos, unsigned len, 1220 struct page *page) 1221 { 1222 struct inode *inode = file_inode(file); 1223 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1224 struct ceph_inode_info *ci = ceph_inode(inode); 1225 loff_t page_off = pos & PAGE_MASK; 1226 int pos_in_page = pos & ~PAGE_MASK; 1227 int end_in_page = pos_in_page + len; 1228 loff_t i_size; 1229 int r; 1230 struct ceph_snap_context *snapc, *oldest; 1231 1232 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 1233 dout(" page %p forced umount\n", page); 1234 unlock_page(page); 1235 return -EIO; 1236 } 1237 1238 retry_locked: 1239 /* writepages currently holds page lock, but if we change that later, */ 1240 wait_on_page_writeback(page); 1241 1242 snapc = page_snap_context(page); 1243 if (snapc && snapc != ci->i_head_snapc) { 1244 /* 1245 * this page is already dirty in another (older) snap 1246 * context! is it writeable now? 1247 */ 1248 oldest = get_oldest_context(inode, NULL, NULL); 1249 if (snapc->seq > oldest->seq) { 1250 ceph_put_snap_context(oldest); 1251 dout(" page %p snapc %p not current or oldest\n", 1252 page, snapc); 1253 /* 1254 * queue for writeback, and wait for snapc to 1255 * be writeable or written 1256 */ 1257 snapc = ceph_get_snap_context(snapc); 1258 unlock_page(page); 1259 ceph_queue_writeback(inode); 1260 r = wait_event_killable(ci->i_cap_wq, 1261 context_is_writeable_or_written(inode, snapc)); 1262 ceph_put_snap_context(snapc); 1263 if (r == -ERESTARTSYS) 1264 return r; 1265 return -EAGAIN; 1266 } 1267 ceph_put_snap_context(oldest); 1268 1269 /* yay, writeable, do it now (without dropping page lock) */ 1270 dout(" page %p snapc %p not current, but oldest\n", 1271 page, snapc); 1272 if (!clear_page_dirty_for_io(page)) 1273 goto retry_locked; 1274 r = writepage_nounlock(page, NULL); 1275 if (r < 0) 1276 goto fail_unlock; 1277 goto retry_locked; 1278 } 1279 1280 if (PageUptodate(page)) { 1281 dout(" page %p already uptodate\n", page); 1282 return 0; 1283 } 1284 1285 /* full page? */ 1286 if (pos_in_page == 0 && len == PAGE_SIZE) 1287 return 0; 1288 1289 /* past end of file? */ 1290 i_size = i_size_read(inode); 1291 1292 if (page_off >= i_size || 1293 (pos_in_page == 0 && (pos+len) >= i_size && 1294 end_in_page - pos_in_page != PAGE_SIZE)) { 1295 dout(" zeroing %p 0 - %d and %d - %d\n", 1296 page, pos_in_page, end_in_page, (int)PAGE_SIZE); 1297 zero_user_segments(page, 1298 0, pos_in_page, 1299 end_in_page, PAGE_SIZE); 1300 return 0; 1301 } 1302 1303 /* we need to read it. */ 1304 r = ceph_do_readpage(file, page); 1305 if (r < 0) { 1306 if (r == -EINPROGRESS) 1307 return -EAGAIN; 1308 goto fail_unlock; 1309 } 1310 goto retry_locked; 1311 fail_unlock: 1312 unlock_page(page); 1313 return r; 1314 } 1315 1316 /* 1317 * We are only allowed to write into/dirty the page if the page is 1318 * clean, or already dirty within the same snap context. 1319 */ 1320 static int ceph_write_begin(struct file *file, struct address_space *mapping, 1321 loff_t pos, unsigned len, unsigned flags, 1322 struct page **pagep, void **fsdata) 1323 { 1324 struct inode *inode = file_inode(file); 1325 struct page *page; 1326 pgoff_t index = pos >> PAGE_SHIFT; 1327 int r; 1328 1329 do { 1330 /* get a page */ 1331 page = grab_cache_page_write_begin(mapping, index, 0); 1332 if (!page) 1333 return -ENOMEM; 1334 1335 dout("write_begin file %p inode %p page %p %d~%d\n", file, 1336 inode, page, (int)pos, (int)len); 1337 1338 r = ceph_update_writeable_page(file, pos, len, page); 1339 if (r < 0) 1340 put_page(page); 1341 else 1342 *pagep = page; 1343 } while (r == -EAGAIN); 1344 1345 return r; 1346 } 1347 1348 /* 1349 * we don't do anything in here that simple_write_end doesn't do 1350 * except adjust dirty page accounting 1351 */ 1352 static int ceph_write_end(struct file *file, struct address_space *mapping, 1353 loff_t pos, unsigned len, unsigned copied, 1354 struct page *page, void *fsdata) 1355 { 1356 struct inode *inode = file_inode(file); 1357 bool check_cap = false; 1358 1359 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, 1360 inode, page, (int)pos, (int)copied, (int)len); 1361 1362 /* zero the stale part of the page if we did a short copy */ 1363 if (!PageUptodate(page)) { 1364 if (copied < len) { 1365 copied = 0; 1366 goto out; 1367 } 1368 SetPageUptodate(page); 1369 } 1370 1371 /* did file size increase? */ 1372 if (pos+copied > i_size_read(inode)) 1373 check_cap = ceph_inode_set_size(inode, pos+copied); 1374 1375 set_page_dirty(page); 1376 1377 out: 1378 unlock_page(page); 1379 put_page(page); 1380 1381 if (check_cap) 1382 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); 1383 1384 return copied; 1385 } 1386 1387 /* 1388 * we set .direct_IO to indicate direct io is supported, but since we 1389 * intercept O_DIRECT reads and writes early, this function should 1390 * never get called. 1391 */ 1392 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter) 1393 { 1394 WARN_ON(1); 1395 return -EINVAL; 1396 } 1397 1398 const struct address_space_operations ceph_aops = { 1399 .readpage = ceph_readpage, 1400 .readpages = ceph_readpages, 1401 .writepage = ceph_writepage, 1402 .writepages = ceph_writepages_start, 1403 .write_begin = ceph_write_begin, 1404 .write_end = ceph_write_end, 1405 .set_page_dirty = ceph_set_page_dirty, 1406 .invalidatepage = ceph_invalidatepage, 1407 .releasepage = ceph_releasepage, 1408 .direct_IO = ceph_direct_io, 1409 }; 1410 1411 static void ceph_block_sigs(sigset_t *oldset) 1412 { 1413 sigset_t mask; 1414 siginitsetinv(&mask, sigmask(SIGKILL)); 1415 sigprocmask(SIG_BLOCK, &mask, oldset); 1416 } 1417 1418 static void ceph_restore_sigs(sigset_t *oldset) 1419 { 1420 sigprocmask(SIG_SETMASK, oldset, NULL); 1421 } 1422 1423 /* 1424 * vm ops 1425 */ 1426 static int ceph_filemap_fault(struct vm_fault *vmf) 1427 { 1428 struct vm_area_struct *vma = vmf->vma; 1429 struct inode *inode = file_inode(vma->vm_file); 1430 struct ceph_inode_info *ci = ceph_inode(inode); 1431 struct ceph_file_info *fi = vma->vm_file->private_data; 1432 struct page *pinned_page = NULL; 1433 loff_t off = vmf->pgoff << PAGE_SHIFT; 1434 int want, got, ret; 1435 sigset_t oldset; 1436 1437 ceph_block_sigs(&oldset); 1438 1439 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n", 1440 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE); 1441 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1442 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1443 else 1444 want = CEPH_CAP_FILE_CACHE; 1445 1446 got = 0; 1447 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); 1448 if (ret < 0) 1449 goto out_restore; 1450 1451 dout("filemap_fault %p %llu~%zd got cap refs on %s\n", 1452 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got)); 1453 1454 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || 1455 ci->i_inline_version == CEPH_INLINE_NONE) { 1456 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); 1457 ceph_add_rw_context(fi, &rw_ctx); 1458 ret = filemap_fault(vmf); 1459 ceph_del_rw_context(fi, &rw_ctx); 1460 } else 1461 ret = -EAGAIN; 1462 1463 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n", 1464 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret); 1465 if (pinned_page) 1466 put_page(pinned_page); 1467 ceph_put_cap_refs(ci, got); 1468 1469 if (ret != -EAGAIN) 1470 goto out_restore; 1471 1472 /* read inline data */ 1473 if (off >= PAGE_SIZE) { 1474 /* does not support inline data > PAGE_SIZE */ 1475 ret = VM_FAULT_SIGBUS; 1476 } else { 1477 int ret1; 1478 struct address_space *mapping = inode->i_mapping; 1479 struct page *page = find_or_create_page(mapping, 0, 1480 mapping_gfp_constraint(mapping, 1481 ~__GFP_FS)); 1482 if (!page) { 1483 ret = VM_FAULT_OOM; 1484 goto out_inline; 1485 } 1486 ret1 = __ceph_do_getattr(inode, page, 1487 CEPH_STAT_CAP_INLINE_DATA, true); 1488 if (ret1 < 0 || off >= i_size_read(inode)) { 1489 unlock_page(page); 1490 put_page(page); 1491 if (ret1 < 0) 1492 ret = ret1; 1493 else 1494 ret = VM_FAULT_SIGBUS; 1495 goto out_inline; 1496 } 1497 if (ret1 < PAGE_SIZE) 1498 zero_user_segment(page, ret1, PAGE_SIZE); 1499 else 1500 flush_dcache_page(page); 1501 SetPageUptodate(page); 1502 vmf->page = page; 1503 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; 1504 out_inline: 1505 dout("filemap_fault %p %llu~%zd read inline data ret %d\n", 1506 inode, off, (size_t)PAGE_SIZE, ret); 1507 } 1508 out_restore: 1509 ceph_restore_sigs(&oldset); 1510 if (ret < 0) 1511 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; 1512 1513 return ret; 1514 } 1515 1516 /* 1517 * Reuse write_begin here for simplicity. 1518 */ 1519 static int ceph_page_mkwrite(struct vm_fault *vmf) 1520 { 1521 struct vm_area_struct *vma = vmf->vma; 1522 struct inode *inode = file_inode(vma->vm_file); 1523 struct ceph_inode_info *ci = ceph_inode(inode); 1524 struct ceph_file_info *fi = vma->vm_file->private_data; 1525 struct ceph_cap_flush *prealloc_cf; 1526 struct page *page = vmf->page; 1527 loff_t off = page_offset(page); 1528 loff_t size = i_size_read(inode); 1529 size_t len; 1530 int want, got, ret; 1531 sigset_t oldset; 1532 1533 prealloc_cf = ceph_alloc_cap_flush(); 1534 if (!prealloc_cf) 1535 return VM_FAULT_OOM; 1536 1537 ceph_block_sigs(&oldset); 1538 1539 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1540 struct page *locked_page = NULL; 1541 if (off == 0) { 1542 lock_page(page); 1543 locked_page = page; 1544 } 1545 ret = ceph_uninline_data(vma->vm_file, locked_page); 1546 if (locked_page) 1547 unlock_page(locked_page); 1548 if (ret < 0) 1549 goto out_free; 1550 } 1551 1552 if (off + PAGE_SIZE <= size) 1553 len = PAGE_SIZE; 1554 else 1555 len = size & ~PAGE_MASK; 1556 1557 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", 1558 inode, ceph_vinop(inode), off, len, size); 1559 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1560 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 1561 else 1562 want = CEPH_CAP_FILE_BUFFER; 1563 1564 got = 0; 1565 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len, 1566 &got, NULL); 1567 if (ret < 0) 1568 goto out_free; 1569 1570 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", 1571 inode, off, len, ceph_cap_string(got)); 1572 1573 /* Update time before taking page lock */ 1574 file_update_time(vma->vm_file); 1575 1576 do { 1577 lock_page(page); 1578 1579 if ((off > size) || (page->mapping != inode->i_mapping)) { 1580 unlock_page(page); 1581 ret = VM_FAULT_NOPAGE; 1582 break; 1583 } 1584 1585 ret = ceph_update_writeable_page(vma->vm_file, off, len, page); 1586 if (ret >= 0) { 1587 /* success. we'll keep the page locked. */ 1588 set_page_dirty(page); 1589 ret = VM_FAULT_LOCKED; 1590 } 1591 } while (ret == -EAGAIN); 1592 1593 if (ret == VM_FAULT_LOCKED || 1594 ci->i_inline_version != CEPH_INLINE_NONE) { 1595 int dirty; 1596 spin_lock(&ci->i_ceph_lock); 1597 ci->i_inline_version = CEPH_INLINE_NONE; 1598 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1599 &prealloc_cf); 1600 spin_unlock(&ci->i_ceph_lock); 1601 if (dirty) 1602 __mark_inode_dirty(inode, dirty); 1603 } 1604 1605 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n", 1606 inode, off, len, ceph_cap_string(got), ret); 1607 ceph_put_cap_refs(ci, got); 1608 out_free: 1609 ceph_restore_sigs(&oldset); 1610 ceph_free_cap_flush(prealloc_cf); 1611 if (ret < 0) 1612 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; 1613 return ret; 1614 } 1615 1616 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, 1617 char *data, size_t len) 1618 { 1619 struct address_space *mapping = inode->i_mapping; 1620 struct page *page; 1621 1622 if (locked_page) { 1623 page = locked_page; 1624 } else { 1625 if (i_size_read(inode) == 0) 1626 return; 1627 page = find_or_create_page(mapping, 0, 1628 mapping_gfp_constraint(mapping, 1629 ~__GFP_FS)); 1630 if (!page) 1631 return; 1632 if (PageUptodate(page)) { 1633 unlock_page(page); 1634 put_page(page); 1635 return; 1636 } 1637 } 1638 1639 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", 1640 inode, ceph_vinop(inode), len, locked_page); 1641 1642 if (len > 0) { 1643 void *kaddr = kmap_atomic(page); 1644 memcpy(kaddr, data, len); 1645 kunmap_atomic(kaddr); 1646 } 1647 1648 if (page != locked_page) { 1649 if (len < PAGE_SIZE) 1650 zero_user_segment(page, len, PAGE_SIZE); 1651 else 1652 flush_dcache_page(page); 1653 1654 SetPageUptodate(page); 1655 unlock_page(page); 1656 put_page(page); 1657 } 1658 } 1659 1660 int ceph_uninline_data(struct file *filp, struct page *locked_page) 1661 { 1662 struct inode *inode = file_inode(filp); 1663 struct ceph_inode_info *ci = ceph_inode(inode); 1664 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1665 struct ceph_osd_request *req; 1666 struct page *page = NULL; 1667 u64 len, inline_version; 1668 int err = 0; 1669 bool from_pagecache = false; 1670 1671 spin_lock(&ci->i_ceph_lock); 1672 inline_version = ci->i_inline_version; 1673 spin_unlock(&ci->i_ceph_lock); 1674 1675 dout("uninline_data %p %llx.%llx inline_version %llu\n", 1676 inode, ceph_vinop(inode), inline_version); 1677 1678 if (inline_version == 1 || /* initial version, no data */ 1679 inline_version == CEPH_INLINE_NONE) 1680 goto out; 1681 1682 if (locked_page) { 1683 page = locked_page; 1684 WARN_ON(!PageUptodate(page)); 1685 } else if (ceph_caps_issued(ci) & 1686 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { 1687 page = find_get_page(inode->i_mapping, 0); 1688 if (page) { 1689 if (PageUptodate(page)) { 1690 from_pagecache = true; 1691 lock_page(page); 1692 } else { 1693 put_page(page); 1694 page = NULL; 1695 } 1696 } 1697 } 1698 1699 if (page) { 1700 len = i_size_read(inode); 1701 if (len > PAGE_SIZE) 1702 len = PAGE_SIZE; 1703 } else { 1704 page = __page_cache_alloc(GFP_NOFS); 1705 if (!page) { 1706 err = -ENOMEM; 1707 goto out; 1708 } 1709 err = __ceph_do_getattr(inode, page, 1710 CEPH_STAT_CAP_INLINE_DATA, true); 1711 if (err < 0) { 1712 /* no inline data */ 1713 if (err == -ENODATA) 1714 err = 0; 1715 goto out; 1716 } 1717 len = err; 1718 } 1719 1720 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1721 ceph_vino(inode), 0, &len, 0, 1, 1722 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, 1723 NULL, 0, 0, false); 1724 if (IS_ERR(req)) { 1725 err = PTR_ERR(req); 1726 goto out; 1727 } 1728 1729 req->r_mtime = inode->i_mtime; 1730 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1731 if (!err) 1732 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1733 ceph_osdc_put_request(req); 1734 if (err < 0) 1735 goto out; 1736 1737 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1738 ceph_vino(inode), 0, &len, 1, 3, 1739 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1740 NULL, ci->i_truncate_seq, 1741 ci->i_truncate_size, false); 1742 if (IS_ERR(req)) { 1743 err = PTR_ERR(req); 1744 goto out; 1745 } 1746 1747 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); 1748 1749 { 1750 __le64 xattr_buf = cpu_to_le64(inline_version); 1751 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, 1752 "inline_version", &xattr_buf, 1753 sizeof(xattr_buf), 1754 CEPH_OSD_CMPXATTR_OP_GT, 1755 CEPH_OSD_CMPXATTR_MODE_U64); 1756 if (err) 1757 goto out_put; 1758 } 1759 1760 { 1761 char xattr_buf[32]; 1762 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), 1763 "%llu", inline_version); 1764 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, 1765 "inline_version", 1766 xattr_buf, xattr_len, 0, 0); 1767 if (err) 1768 goto out_put; 1769 } 1770 1771 req->r_mtime = inode->i_mtime; 1772 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1773 if (!err) 1774 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1775 out_put: 1776 ceph_osdc_put_request(req); 1777 if (err == -ECANCELED) 1778 err = 0; 1779 out: 1780 if (page && page != locked_page) { 1781 if (from_pagecache) { 1782 unlock_page(page); 1783 put_page(page); 1784 } else 1785 __free_pages(page, 0); 1786 } 1787 1788 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", 1789 inode, ceph_vinop(inode), inline_version, err); 1790 return err; 1791 } 1792 1793 static const struct vm_operations_struct ceph_vmops = { 1794 .fault = ceph_filemap_fault, 1795 .page_mkwrite = ceph_page_mkwrite, 1796 }; 1797 1798 int ceph_mmap(struct file *file, struct vm_area_struct *vma) 1799 { 1800 struct address_space *mapping = file->f_mapping; 1801 1802 if (!mapping->a_ops->readpage) 1803 return -ENOEXEC; 1804 file_accessed(file); 1805 vma->vm_ops = &ceph_vmops; 1806 return 0; 1807 } 1808 1809 enum { 1810 POOL_READ = 1, 1811 POOL_WRITE = 2, 1812 }; 1813 1814 static int __ceph_pool_perm_get(struct ceph_inode_info *ci, 1815 s64 pool, struct ceph_string *pool_ns) 1816 { 1817 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 1818 struct ceph_mds_client *mdsc = fsc->mdsc; 1819 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; 1820 struct rb_node **p, *parent; 1821 struct ceph_pool_perm *perm; 1822 struct page **pages; 1823 size_t pool_ns_len; 1824 int err = 0, err2 = 0, have = 0; 1825 1826 down_read(&mdsc->pool_perm_rwsem); 1827 p = &mdsc->pool_perm_tree.rb_node; 1828 while (*p) { 1829 perm = rb_entry(*p, struct ceph_pool_perm, node); 1830 if (pool < perm->pool) 1831 p = &(*p)->rb_left; 1832 else if (pool > perm->pool) 1833 p = &(*p)->rb_right; 1834 else { 1835 int ret = ceph_compare_string(pool_ns, 1836 perm->pool_ns, 1837 perm->pool_ns_len); 1838 if (ret < 0) 1839 p = &(*p)->rb_left; 1840 else if (ret > 0) 1841 p = &(*p)->rb_right; 1842 else { 1843 have = perm->perm; 1844 break; 1845 } 1846 } 1847 } 1848 up_read(&mdsc->pool_perm_rwsem); 1849 if (*p) 1850 goto out; 1851 1852 if (pool_ns) 1853 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", 1854 pool, (int)pool_ns->len, pool_ns->str); 1855 else 1856 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); 1857 1858 down_write(&mdsc->pool_perm_rwsem); 1859 p = &mdsc->pool_perm_tree.rb_node; 1860 parent = NULL; 1861 while (*p) { 1862 parent = *p; 1863 perm = rb_entry(parent, struct ceph_pool_perm, node); 1864 if (pool < perm->pool) 1865 p = &(*p)->rb_left; 1866 else if (pool > perm->pool) 1867 p = &(*p)->rb_right; 1868 else { 1869 int ret = ceph_compare_string(pool_ns, 1870 perm->pool_ns, 1871 perm->pool_ns_len); 1872 if (ret < 0) 1873 p = &(*p)->rb_left; 1874 else if (ret > 0) 1875 p = &(*p)->rb_right; 1876 else { 1877 have = perm->perm; 1878 break; 1879 } 1880 } 1881 } 1882 if (*p) { 1883 up_write(&mdsc->pool_perm_rwsem); 1884 goto out; 1885 } 1886 1887 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1888 1, false, GFP_NOFS); 1889 if (!rd_req) { 1890 err = -ENOMEM; 1891 goto out_unlock; 1892 } 1893 1894 rd_req->r_flags = CEPH_OSD_FLAG_READ; 1895 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); 1896 rd_req->r_base_oloc.pool = pool; 1897 if (pool_ns) 1898 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); 1899 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); 1900 1901 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); 1902 if (err) 1903 goto out_unlock; 1904 1905 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1906 1, false, GFP_NOFS); 1907 if (!wr_req) { 1908 err = -ENOMEM; 1909 goto out_unlock; 1910 } 1911 1912 wr_req->r_flags = CEPH_OSD_FLAG_WRITE; 1913 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); 1914 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); 1915 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); 1916 1917 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); 1918 if (err) 1919 goto out_unlock; 1920 1921 /* one page should be large enough for STAT data */ 1922 pages = ceph_alloc_page_vector(1, GFP_KERNEL); 1923 if (IS_ERR(pages)) { 1924 err = PTR_ERR(pages); 1925 goto out_unlock; 1926 } 1927 1928 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, 1929 0, false, true); 1930 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); 1931 1932 wr_req->r_mtime = ci->vfs_inode.i_mtime; 1933 wr_req->r_abort_on_full = true; 1934 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); 1935 1936 if (!err) 1937 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); 1938 if (!err2) 1939 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); 1940 1941 if (err >= 0 || err == -ENOENT) 1942 have |= POOL_READ; 1943 else if (err != -EPERM) 1944 goto out_unlock; 1945 1946 if (err2 == 0 || err2 == -EEXIST) 1947 have |= POOL_WRITE; 1948 else if (err2 != -EPERM) { 1949 err = err2; 1950 goto out_unlock; 1951 } 1952 1953 pool_ns_len = pool_ns ? pool_ns->len : 0; 1954 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); 1955 if (!perm) { 1956 err = -ENOMEM; 1957 goto out_unlock; 1958 } 1959 1960 perm->pool = pool; 1961 perm->perm = have; 1962 perm->pool_ns_len = pool_ns_len; 1963 if (pool_ns_len > 0) 1964 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); 1965 perm->pool_ns[pool_ns_len] = 0; 1966 1967 rb_link_node(&perm->node, parent, p); 1968 rb_insert_color(&perm->node, &mdsc->pool_perm_tree); 1969 err = 0; 1970 out_unlock: 1971 up_write(&mdsc->pool_perm_rwsem); 1972 1973 ceph_osdc_put_request(rd_req); 1974 ceph_osdc_put_request(wr_req); 1975 out: 1976 if (!err) 1977 err = have; 1978 if (pool_ns) 1979 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", 1980 pool, (int)pool_ns->len, pool_ns->str, err); 1981 else 1982 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); 1983 return err; 1984 } 1985 1986 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need) 1987 { 1988 s64 pool; 1989 struct ceph_string *pool_ns; 1990 int ret, flags; 1991 1992 if (ci->i_vino.snap != CEPH_NOSNAP) { 1993 /* 1994 * Pool permission check needs to write to the first object. 1995 * But for snapshot, head of the first object may have alread 1996 * been deleted. Skip check to avoid creating orphan object. 1997 */ 1998 return 0; 1999 } 2000 2001 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode), 2002 NOPOOLPERM)) 2003 return 0; 2004 2005 spin_lock(&ci->i_ceph_lock); 2006 flags = ci->i_ceph_flags; 2007 pool = ci->i_layout.pool_id; 2008 spin_unlock(&ci->i_ceph_lock); 2009 check: 2010 if (flags & CEPH_I_POOL_PERM) { 2011 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { 2012 dout("ceph_pool_perm_check pool %lld no read perm\n", 2013 pool); 2014 return -EPERM; 2015 } 2016 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { 2017 dout("ceph_pool_perm_check pool %lld no write perm\n", 2018 pool); 2019 return -EPERM; 2020 } 2021 return 0; 2022 } 2023 2024 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); 2025 ret = __ceph_pool_perm_get(ci, pool, pool_ns); 2026 ceph_put_string(pool_ns); 2027 if (ret < 0) 2028 return ret; 2029 2030 flags = CEPH_I_POOL_PERM; 2031 if (ret & POOL_READ) 2032 flags |= CEPH_I_POOL_RD; 2033 if (ret & POOL_WRITE) 2034 flags |= CEPH_I_POOL_WR; 2035 2036 spin_lock(&ci->i_ceph_lock); 2037 if (pool == ci->i_layout.pool_id && 2038 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { 2039 ci->i_ceph_flags |= flags; 2040 } else { 2041 pool = ci->i_layout.pool_id; 2042 flags = ci->i_ceph_flags; 2043 } 2044 spin_unlock(&ci->i_ceph_lock); 2045 goto check; 2046 } 2047 2048 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) 2049 { 2050 struct ceph_pool_perm *perm; 2051 struct rb_node *n; 2052 2053 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { 2054 n = rb_first(&mdsc->pool_perm_tree); 2055 perm = rb_entry(n, struct ceph_pool_perm, node); 2056 rb_erase(n, &mdsc->pool_perm_tree); 2057 kfree(perm); 2058 } 2059 } 2060