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 #include <linux/iversion.h> 14 15 #include "super.h" 16 #include "mds_client.h" 17 #include "cache.h" 18 #include <linux/ceph/osd_client.h> 19 #include <linux/ceph/striper.h> 20 21 /* 22 * Ceph address space ops. 23 * 24 * There are a few funny things going on here. 25 * 26 * The page->private field is used to reference a struct 27 * ceph_snap_context for _every_ dirty page. This indicates which 28 * snapshot the page was logically dirtied in, and thus which snap 29 * context needs to be associated with the osd write during writeback. 30 * 31 * Similarly, struct ceph_inode_info maintains a set of counters to 32 * count dirty pages on the inode. In the absence of snapshots, 33 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 34 * 35 * When a snapshot is taken (that is, when the client receives 36 * notification that a snapshot was taken), each inode with caps and 37 * with dirty pages (dirty pages implies there is a cap) gets a new 38 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 39 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 40 * moved to capsnap->dirty. (Unless a sync write is currently in 41 * progress. In that case, the capsnap is said to be "pending", new 42 * writes cannot start, and the capsnap isn't "finalized" until the 43 * write completes (or fails) and a final size/mtime for the inode for 44 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 45 * 46 * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 47 * we look for the first capsnap in i_cap_snaps and write out pages in 48 * that snap context _only_. Then we move on to the next capsnap, 49 * eventually reaching the "live" or "head" context (i.e., pages that 50 * are not yet snapped) and are writing the most recently dirtied 51 * pages. 52 * 53 * Invalidate and so forth must take care to ensure the dirty page 54 * accounting is preserved. 55 */ 56 57 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 58 #define CONGESTION_OFF_THRESH(congestion_kb) \ 59 (CONGESTION_ON_THRESH(congestion_kb) - \ 60 (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 61 62 static inline struct ceph_snap_context *page_snap_context(struct page *page) 63 { 64 if (PagePrivate(page)) 65 return (void *)page->private; 66 return NULL; 67 } 68 69 /* 70 * Dirty a page. Optimistically adjust accounting, on the assumption 71 * that we won't race with invalidate. If we do, readjust. 72 */ 73 static int ceph_set_page_dirty(struct page *page) 74 { 75 struct address_space *mapping = page->mapping; 76 struct inode *inode; 77 struct ceph_inode_info *ci; 78 struct ceph_snap_context *snapc; 79 int ret; 80 81 if (unlikely(!mapping)) 82 return !TestSetPageDirty(page); 83 84 if (PageDirty(page)) { 85 dout("%p set_page_dirty %p idx %lu -- already dirty\n", 86 mapping->host, page, page->index); 87 BUG_ON(!PagePrivate(page)); 88 return 0; 89 } 90 91 inode = mapping->host; 92 ci = ceph_inode(inode); 93 94 /* dirty the head */ 95 spin_lock(&ci->i_ceph_lock); 96 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference 97 if (__ceph_have_pending_cap_snap(ci)) { 98 struct ceph_cap_snap *capsnap = 99 list_last_entry(&ci->i_cap_snaps, 100 struct ceph_cap_snap, 101 ci_item); 102 snapc = ceph_get_snap_context(capsnap->context); 103 capsnap->dirty_pages++; 104 } else { 105 BUG_ON(!ci->i_head_snapc); 106 snapc = ceph_get_snap_context(ci->i_head_snapc); 107 ++ci->i_wrbuffer_ref_head; 108 } 109 if (ci->i_wrbuffer_ref == 0) 110 ihold(inode); 111 ++ci->i_wrbuffer_ref; 112 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " 113 "snapc %p seq %lld (%d snaps)\n", 114 mapping->host, page, page->index, 115 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 116 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 117 snapc, snapc->seq, snapc->num_snaps); 118 spin_unlock(&ci->i_ceph_lock); 119 120 /* 121 * Reference snap context in page->private. Also set 122 * PagePrivate so that we get invalidatepage callback. 123 */ 124 BUG_ON(PagePrivate(page)); 125 page->private = (unsigned long)snapc; 126 SetPagePrivate(page); 127 128 ret = __set_page_dirty_nobuffers(page); 129 WARN_ON(!PageLocked(page)); 130 WARN_ON(!page->mapping); 131 132 return ret; 133 } 134 135 /* 136 * If we are truncating the full page (i.e. offset == 0), adjust the 137 * dirty page counters appropriately. Only called if there is private 138 * data on the page. 139 */ 140 static void ceph_invalidatepage(struct page *page, unsigned int offset, 141 unsigned int length) 142 { 143 struct inode *inode; 144 struct ceph_inode_info *ci; 145 struct ceph_snap_context *snapc = page_snap_context(page); 146 147 inode = page->mapping->host; 148 ci = ceph_inode(inode); 149 150 if (offset != 0 || length != PAGE_SIZE) { 151 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", 152 inode, page, page->index, offset, length); 153 return; 154 } 155 156 ceph_invalidate_fscache_page(inode, page); 157 158 WARN_ON(!PageLocked(page)); 159 if (!PagePrivate(page)) 160 return; 161 162 ClearPageChecked(page); 163 164 dout("%p invalidatepage %p idx %lu full dirty page\n", 165 inode, page, page->index); 166 167 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 168 ceph_put_snap_context(snapc); 169 page->private = 0; 170 ClearPagePrivate(page); 171 } 172 173 static int ceph_releasepage(struct page *page, gfp_t g) 174 { 175 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host, 176 page, page->index, PageDirty(page) ? "" : "not "); 177 178 /* Can we release the page from the cache? */ 179 if (!ceph_release_fscache_page(page, g)) 180 return 0; 181 182 return !PagePrivate(page); 183 } 184 185 /* 186 * read a single page, without unlocking it. 187 */ 188 static int ceph_do_readpage(struct file *filp, struct page *page) 189 { 190 struct inode *inode = file_inode(filp); 191 struct ceph_inode_info *ci = ceph_inode(inode); 192 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 193 int err = 0; 194 u64 off = page_offset(page); 195 u64 len = PAGE_SIZE; 196 197 if (off >= i_size_read(inode)) { 198 zero_user_segment(page, 0, PAGE_SIZE); 199 SetPageUptodate(page); 200 return 0; 201 } 202 203 if (ci->i_inline_version != CEPH_INLINE_NONE) { 204 /* 205 * Uptodate inline data should have been added 206 * into page cache while getting Fcr caps. 207 */ 208 if (off == 0) 209 return -EINVAL; 210 zero_user_segment(page, 0, PAGE_SIZE); 211 SetPageUptodate(page); 212 return 0; 213 } 214 215 err = ceph_readpage_from_fscache(inode, page); 216 if (err == 0) 217 return -EINPROGRESS; 218 219 dout("readpage inode %p file %p page %p index %lu\n", 220 inode, filp, page, page->index); 221 err = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), 222 &ci->i_layout, off, &len, 223 ci->i_truncate_seq, ci->i_truncate_size, 224 &page, 1, 0); 225 if (err == -ENOENT) 226 err = 0; 227 if (err < 0) { 228 SetPageError(page); 229 ceph_fscache_readpage_cancel(inode, page); 230 if (err == -EBLACKLISTED) 231 fsc->blacklisted = true; 232 goto out; 233 } 234 if (err < PAGE_SIZE) 235 /* zero fill remainder of page */ 236 zero_user_segment(page, err, PAGE_SIZE); 237 else 238 flush_dcache_page(page); 239 240 SetPageUptodate(page); 241 ceph_readpage_to_fscache(inode, page); 242 243 out: 244 return err < 0 ? err : 0; 245 } 246 247 static int ceph_readpage(struct file *filp, struct page *page) 248 { 249 int r = ceph_do_readpage(filp, page); 250 if (r != -EINPROGRESS) 251 unlock_page(page); 252 else 253 r = 0; 254 return r; 255 } 256 257 /* 258 * Finish an async read(ahead) op. 259 */ 260 static void finish_read(struct ceph_osd_request *req) 261 { 262 struct inode *inode = req->r_inode; 263 struct ceph_osd_data *osd_data; 264 int rc = req->r_result <= 0 ? req->r_result : 0; 265 int bytes = req->r_result >= 0 ? req->r_result : 0; 266 int num_pages; 267 int i; 268 269 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); 270 if (rc == -EBLACKLISTED) 271 ceph_inode_to_client(inode)->blacklisted = true; 272 273 /* unlock all pages, zeroing any data we didn't read */ 274 osd_data = osd_req_op_extent_osd_data(req, 0); 275 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 276 num_pages = calc_pages_for((u64)osd_data->alignment, 277 (u64)osd_data->length); 278 for (i = 0; i < num_pages; i++) { 279 struct page *page = osd_data->pages[i]; 280 281 if (rc < 0 && rc != -ENOENT) { 282 ceph_fscache_readpage_cancel(inode, page); 283 goto unlock; 284 } 285 if (bytes < (int)PAGE_SIZE) { 286 /* zero (remainder of) page */ 287 int s = bytes < 0 ? 0 : bytes; 288 zero_user_segment(page, s, PAGE_SIZE); 289 } 290 dout("finish_read %p uptodate %p idx %lu\n", inode, page, 291 page->index); 292 flush_dcache_page(page); 293 SetPageUptodate(page); 294 ceph_readpage_to_fscache(inode, page); 295 unlock: 296 unlock_page(page); 297 put_page(page); 298 bytes -= PAGE_SIZE; 299 } 300 kfree(osd_data->pages); 301 } 302 303 /* 304 * start an async read(ahead) operation. return nr_pages we submitted 305 * a read for on success, or negative error code. 306 */ 307 static int start_read(struct inode *inode, struct ceph_rw_context *rw_ctx, 308 struct list_head *page_list, int max) 309 { 310 struct ceph_osd_client *osdc = 311 &ceph_inode_to_client(inode)->client->osdc; 312 struct ceph_inode_info *ci = ceph_inode(inode); 313 struct page *page = lru_to_page(page_list); 314 struct ceph_vino vino; 315 struct ceph_osd_request *req; 316 u64 off; 317 u64 len; 318 int i; 319 struct page **pages; 320 pgoff_t next_index; 321 int nr_pages = 0; 322 int got = 0; 323 int ret = 0; 324 325 if (!rw_ctx) { 326 /* caller of readpages does not hold buffer and read caps 327 * (fadvise, madvise and readahead cases) */ 328 int want = CEPH_CAP_FILE_CACHE; 329 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, 330 true, &got); 331 if (ret < 0) { 332 dout("start_read %p, error getting cap\n", inode); 333 } else if (!(got & want)) { 334 dout("start_read %p, no cache cap\n", inode); 335 ret = 0; 336 } 337 if (ret <= 0) { 338 if (got) 339 ceph_put_cap_refs(ci, got); 340 while (!list_empty(page_list)) { 341 page = lru_to_page(page_list); 342 list_del(&page->lru); 343 put_page(page); 344 } 345 return ret; 346 } 347 } 348 349 off = (u64) page_offset(page); 350 351 /* count pages */ 352 next_index = page->index; 353 list_for_each_entry_reverse(page, page_list, lru) { 354 if (page->index != next_index) 355 break; 356 nr_pages++; 357 next_index++; 358 if (max && nr_pages == max) 359 break; 360 } 361 len = nr_pages << PAGE_SHIFT; 362 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, 363 off, len); 364 vino = ceph_vino(inode); 365 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, 366 0, 1, CEPH_OSD_OP_READ, 367 CEPH_OSD_FLAG_READ, NULL, 368 ci->i_truncate_seq, ci->i_truncate_size, 369 false); 370 if (IS_ERR(req)) { 371 ret = PTR_ERR(req); 372 goto out; 373 } 374 375 /* build page vector */ 376 nr_pages = calc_pages_for(0, len); 377 pages = kmalloc_array(nr_pages, sizeof(*pages), GFP_KERNEL); 378 if (!pages) { 379 ret = -ENOMEM; 380 goto out_put; 381 } 382 for (i = 0; i < nr_pages; ++i) { 383 page = list_entry(page_list->prev, struct page, lru); 384 BUG_ON(PageLocked(page)); 385 list_del(&page->lru); 386 387 dout("start_read %p adding %p idx %lu\n", inode, page, 388 page->index); 389 if (add_to_page_cache_lru(page, &inode->i_data, page->index, 390 GFP_KERNEL)) { 391 ceph_fscache_uncache_page(inode, page); 392 put_page(page); 393 dout("start_read %p add_to_page_cache failed %p\n", 394 inode, page); 395 nr_pages = i; 396 if (nr_pages > 0) { 397 len = nr_pages << PAGE_SHIFT; 398 osd_req_op_extent_update(req, 0, len); 399 break; 400 } 401 goto out_pages; 402 } 403 pages[i] = page; 404 } 405 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); 406 req->r_callback = finish_read; 407 req->r_inode = inode; 408 409 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); 410 ret = ceph_osdc_start_request(osdc, req, false); 411 if (ret < 0) 412 goto out_pages; 413 ceph_osdc_put_request(req); 414 415 /* After adding locked pages to page cache, the inode holds cache cap. 416 * So we can drop our cap refs. */ 417 if (got) 418 ceph_put_cap_refs(ci, got); 419 420 return nr_pages; 421 422 out_pages: 423 for (i = 0; i < nr_pages; ++i) { 424 ceph_fscache_readpage_cancel(inode, pages[i]); 425 unlock_page(pages[i]); 426 } 427 ceph_put_page_vector(pages, nr_pages, false); 428 out_put: 429 ceph_osdc_put_request(req); 430 out: 431 if (got) 432 ceph_put_cap_refs(ci, got); 433 return ret; 434 } 435 436 437 /* 438 * Read multiple pages. Leave pages we don't read + unlock in page_list; 439 * the caller (VM) cleans them up. 440 */ 441 static int ceph_readpages(struct file *file, struct address_space *mapping, 442 struct list_head *page_list, unsigned nr_pages) 443 { 444 struct inode *inode = file_inode(file); 445 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 446 struct ceph_file_info *fi = file->private_data; 447 struct ceph_rw_context *rw_ctx; 448 int rc = 0; 449 int max = 0; 450 451 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) 452 return -EINVAL; 453 454 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list, 455 &nr_pages); 456 457 if (rc == 0) 458 goto out; 459 460 rw_ctx = ceph_find_rw_context(fi); 461 max = fsc->mount_options->rsize >> PAGE_SHIFT; 462 dout("readpages %p file %p ctx %p nr_pages %d max %d\n", 463 inode, file, rw_ctx, nr_pages, max); 464 while (!list_empty(page_list)) { 465 rc = start_read(inode, rw_ctx, page_list, max); 466 if (rc < 0) 467 goto out; 468 } 469 out: 470 ceph_fscache_readpages_cancel(inode, page_list); 471 472 dout("readpages %p file %p ret %d\n", inode, file, rc); 473 return rc; 474 } 475 476 struct ceph_writeback_ctl 477 { 478 loff_t i_size; 479 u64 truncate_size; 480 u32 truncate_seq; 481 bool size_stable; 482 bool head_snapc; 483 }; 484 485 /* 486 * Get ref for the oldest snapc for an inode with dirty data... that is, the 487 * only snap context we are allowed to write back. 488 */ 489 static struct ceph_snap_context * 490 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, 491 struct ceph_snap_context *page_snapc) 492 { 493 struct ceph_inode_info *ci = ceph_inode(inode); 494 struct ceph_snap_context *snapc = NULL; 495 struct ceph_cap_snap *capsnap = NULL; 496 497 spin_lock(&ci->i_ceph_lock); 498 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 499 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 500 capsnap->context, capsnap->dirty_pages); 501 if (!capsnap->dirty_pages) 502 continue; 503 504 /* get i_size, truncate_{seq,size} for page_snapc? */ 505 if (snapc && capsnap->context != page_snapc) 506 continue; 507 508 if (ctl) { 509 if (capsnap->writing) { 510 ctl->i_size = i_size_read(inode); 511 ctl->size_stable = false; 512 } else { 513 ctl->i_size = capsnap->size; 514 ctl->size_stable = true; 515 } 516 ctl->truncate_size = capsnap->truncate_size; 517 ctl->truncate_seq = capsnap->truncate_seq; 518 ctl->head_snapc = false; 519 } 520 521 if (snapc) 522 break; 523 524 snapc = ceph_get_snap_context(capsnap->context); 525 if (!page_snapc || 526 page_snapc == snapc || 527 page_snapc->seq > snapc->seq) 528 break; 529 } 530 if (!snapc && ci->i_wrbuffer_ref_head) { 531 snapc = ceph_get_snap_context(ci->i_head_snapc); 532 dout(" head snapc %p has %d dirty pages\n", 533 snapc, ci->i_wrbuffer_ref_head); 534 if (ctl) { 535 ctl->i_size = i_size_read(inode); 536 ctl->truncate_size = ci->i_truncate_size; 537 ctl->truncate_seq = ci->i_truncate_seq; 538 ctl->size_stable = false; 539 ctl->head_snapc = true; 540 } 541 } 542 spin_unlock(&ci->i_ceph_lock); 543 return snapc; 544 } 545 546 static u64 get_writepages_data_length(struct inode *inode, 547 struct page *page, u64 start) 548 { 549 struct ceph_inode_info *ci = ceph_inode(inode); 550 struct ceph_snap_context *snapc = page_snap_context(page); 551 struct ceph_cap_snap *capsnap = NULL; 552 u64 end = i_size_read(inode); 553 554 if (snapc != ci->i_head_snapc) { 555 bool found = false; 556 spin_lock(&ci->i_ceph_lock); 557 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 558 if (capsnap->context == snapc) { 559 if (!capsnap->writing) 560 end = capsnap->size; 561 found = true; 562 break; 563 } 564 } 565 spin_unlock(&ci->i_ceph_lock); 566 WARN_ON(!found); 567 } 568 if (end > page_offset(page) + PAGE_SIZE) 569 end = page_offset(page) + PAGE_SIZE; 570 return end > start ? end - start : 0; 571 } 572 573 /* 574 * Write a single page, but leave the page locked. 575 * 576 * If we get a write error, mark the mapping for error, but still adjust the 577 * dirty page accounting (i.e., page is no longer dirty). 578 */ 579 static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 580 { 581 struct inode *inode; 582 struct ceph_inode_info *ci; 583 struct ceph_fs_client *fsc; 584 struct ceph_snap_context *snapc, *oldest; 585 loff_t page_off = page_offset(page); 586 int err, len = PAGE_SIZE; 587 struct ceph_writeback_ctl ceph_wbc; 588 589 dout("writepage %p idx %lu\n", page, page->index); 590 591 inode = page->mapping->host; 592 ci = ceph_inode(inode); 593 fsc = ceph_inode_to_client(inode); 594 595 /* verify this is a writeable snap context */ 596 snapc = page_snap_context(page); 597 if (!snapc) { 598 dout("writepage %p page %p not dirty?\n", inode, page); 599 return 0; 600 } 601 oldest = get_oldest_context(inode, &ceph_wbc, snapc); 602 if (snapc->seq > oldest->seq) { 603 dout("writepage %p page %p snapc %p not writeable - noop\n", 604 inode, page, snapc); 605 /* we should only noop if called by kswapd */ 606 WARN_ON(!(current->flags & PF_MEMALLOC)); 607 ceph_put_snap_context(oldest); 608 redirty_page_for_writepage(wbc, page); 609 return 0; 610 } 611 ceph_put_snap_context(oldest); 612 613 /* is this a partial page at end of file? */ 614 if (page_off >= ceph_wbc.i_size) { 615 dout("%p page eof %llu\n", page, ceph_wbc.i_size); 616 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); 617 return 0; 618 } 619 620 if (ceph_wbc.i_size < page_off + len) 621 len = ceph_wbc.i_size - page_off; 622 623 dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n", 624 inode, page, page->index, page_off, len, snapc, snapc->seq); 625 626 if (atomic_long_inc_return(&fsc->writeback_count) > 627 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 628 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 629 630 set_page_writeback(page); 631 err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode), 632 &ci->i_layout, snapc, page_off, len, 633 ceph_wbc.truncate_seq, 634 ceph_wbc.truncate_size, 635 &inode->i_mtime, &page, 1); 636 if (err < 0) { 637 struct writeback_control tmp_wbc; 638 if (!wbc) 639 wbc = &tmp_wbc; 640 if (err == -ERESTARTSYS) { 641 /* killed by SIGKILL */ 642 dout("writepage interrupted page %p\n", page); 643 redirty_page_for_writepage(wbc, page); 644 end_page_writeback(page); 645 return err; 646 } 647 if (err == -EBLACKLISTED) 648 fsc->blacklisted = true; 649 dout("writepage setting page/mapping error %d %p\n", 650 err, page); 651 mapping_set_error(&inode->i_data, err); 652 wbc->pages_skipped++; 653 } else { 654 dout("writepage cleaned page %p\n", page); 655 err = 0; /* vfs expects us to return 0 */ 656 } 657 page->private = 0; 658 ClearPagePrivate(page); 659 end_page_writeback(page); 660 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 661 ceph_put_snap_context(snapc); /* page's reference */ 662 663 if (atomic_long_dec_return(&fsc->writeback_count) < 664 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) 665 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 666 667 return err; 668 } 669 670 static int ceph_writepage(struct page *page, struct writeback_control *wbc) 671 { 672 int err; 673 struct inode *inode = page->mapping->host; 674 BUG_ON(!inode); 675 ihold(inode); 676 err = writepage_nounlock(page, wbc); 677 if (err == -ERESTARTSYS) { 678 /* direct memory reclaimer was killed by SIGKILL. return 0 679 * to prevent caller from setting mapping/page error */ 680 err = 0; 681 } 682 unlock_page(page); 683 iput(inode); 684 return err; 685 } 686 687 /* 688 * lame release_pages helper. release_pages() isn't exported to 689 * modules. 690 */ 691 static void ceph_release_pages(struct page **pages, int num) 692 { 693 struct pagevec pvec; 694 int i; 695 696 pagevec_init(&pvec); 697 for (i = 0; i < num; i++) { 698 if (pagevec_add(&pvec, pages[i]) == 0) 699 pagevec_release(&pvec); 700 } 701 pagevec_release(&pvec); 702 } 703 704 /* 705 * async writeback completion handler. 706 * 707 * If we get an error, set the mapping error bit, but not the individual 708 * page error bits. 709 */ 710 static void writepages_finish(struct ceph_osd_request *req) 711 { 712 struct inode *inode = req->r_inode; 713 struct ceph_inode_info *ci = ceph_inode(inode); 714 struct ceph_osd_data *osd_data; 715 struct page *page; 716 int num_pages, total_pages = 0; 717 int i, j; 718 int rc = req->r_result; 719 struct ceph_snap_context *snapc = req->r_snapc; 720 struct address_space *mapping = inode->i_mapping; 721 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 722 bool remove_page; 723 724 dout("writepages_finish %p rc %d\n", inode, rc); 725 if (rc < 0) { 726 mapping_set_error(mapping, rc); 727 ceph_set_error_write(ci); 728 if (rc == -EBLACKLISTED) 729 fsc->blacklisted = true; 730 } else { 731 ceph_clear_error_write(ci); 732 } 733 734 /* 735 * We lost the cache cap, need to truncate the page before 736 * it is unlocked, otherwise we'd truncate it later in the 737 * page truncation thread, possibly losing some data that 738 * raced its way in 739 */ 740 remove_page = !(ceph_caps_issued(ci) & 741 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); 742 743 /* clean all pages */ 744 for (i = 0; i < req->r_num_ops; i++) { 745 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) 746 break; 747 748 osd_data = osd_req_op_extent_osd_data(req, i); 749 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 750 num_pages = calc_pages_for((u64)osd_data->alignment, 751 (u64)osd_data->length); 752 total_pages += num_pages; 753 for (j = 0; j < num_pages; j++) { 754 page = osd_data->pages[j]; 755 BUG_ON(!page); 756 WARN_ON(!PageUptodate(page)); 757 758 if (atomic_long_dec_return(&fsc->writeback_count) < 759 CONGESTION_OFF_THRESH( 760 fsc->mount_options->congestion_kb)) 761 clear_bdi_congested(inode_to_bdi(inode), 762 BLK_RW_ASYNC); 763 764 ceph_put_snap_context(page_snap_context(page)); 765 page->private = 0; 766 ClearPagePrivate(page); 767 dout("unlocking %p\n", page); 768 end_page_writeback(page); 769 770 if (remove_page) 771 generic_error_remove_page(inode->i_mapping, 772 page); 773 774 unlock_page(page); 775 } 776 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", 777 inode, osd_data->length, rc >= 0 ? num_pages : 0); 778 779 ceph_release_pages(osd_data->pages, num_pages); 780 } 781 782 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); 783 784 osd_data = osd_req_op_extent_osd_data(req, 0); 785 if (osd_data->pages_from_pool) 786 mempool_free(osd_data->pages, 787 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); 788 else 789 kfree(osd_data->pages); 790 ceph_osdc_put_request(req); 791 } 792 793 /* 794 * initiate async writeback 795 */ 796 static int ceph_writepages_start(struct address_space *mapping, 797 struct writeback_control *wbc) 798 { 799 struct inode *inode = mapping->host; 800 struct ceph_inode_info *ci = ceph_inode(inode); 801 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 802 struct ceph_vino vino = ceph_vino(inode); 803 pgoff_t index, start_index, end = -1; 804 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 805 struct pagevec pvec; 806 int rc = 0; 807 unsigned int wsize = i_blocksize(inode); 808 struct ceph_osd_request *req = NULL; 809 struct ceph_writeback_ctl ceph_wbc; 810 bool should_loop, range_whole = false; 811 bool done = false; 812 813 dout("writepages_start %p (mode=%s)\n", inode, 814 wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 815 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 816 817 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 818 if (ci->i_wrbuffer_ref > 0) { 819 pr_warn_ratelimited( 820 "writepage_start %p %lld forced umount\n", 821 inode, ceph_ino(inode)); 822 } 823 mapping_set_error(mapping, -EIO); 824 return -EIO; /* we're in a forced umount, don't write! */ 825 } 826 if (fsc->mount_options->wsize < wsize) 827 wsize = fsc->mount_options->wsize; 828 829 pagevec_init(&pvec); 830 831 start_index = wbc->range_cyclic ? mapping->writeback_index : 0; 832 index = start_index; 833 834 retry: 835 /* find oldest snap context with dirty data */ 836 snapc = get_oldest_context(inode, &ceph_wbc, NULL); 837 if (!snapc) { 838 /* hmm, why does writepages get called when there 839 is no dirty data? */ 840 dout(" no snap context with dirty data?\n"); 841 goto out; 842 } 843 dout(" oldest snapc is %p seq %lld (%d snaps)\n", 844 snapc, snapc->seq, snapc->num_snaps); 845 846 should_loop = false; 847 if (ceph_wbc.head_snapc && snapc != last_snapc) { 848 /* where to start/end? */ 849 if (wbc->range_cyclic) { 850 index = start_index; 851 end = -1; 852 if (index > 0) 853 should_loop = true; 854 dout(" cyclic, start at %lu\n", index); 855 } else { 856 index = wbc->range_start >> PAGE_SHIFT; 857 end = wbc->range_end >> PAGE_SHIFT; 858 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 859 range_whole = true; 860 dout(" not cyclic, %lu to %lu\n", index, end); 861 } 862 } else if (!ceph_wbc.head_snapc) { 863 /* Do not respect wbc->range_{start,end}. Dirty pages 864 * in that range can be associated with newer snapc. 865 * They are not writeable until we write all dirty pages 866 * associated with 'snapc' get written */ 867 if (index > 0) 868 should_loop = true; 869 dout(" non-head snapc, range whole\n"); 870 } 871 872 ceph_put_snap_context(last_snapc); 873 last_snapc = snapc; 874 875 while (!done && index <= end) { 876 int num_ops = 0, op_idx; 877 unsigned i, pvec_pages, max_pages, locked_pages = 0; 878 struct page **pages = NULL, **data_pages; 879 mempool_t *pool = NULL; /* Becomes non-null if mempool used */ 880 struct page *page; 881 pgoff_t strip_unit_end = 0; 882 u64 offset = 0, len = 0; 883 884 max_pages = wsize >> PAGE_SHIFT; 885 886 get_more_pages: 887 pvec_pages = pagevec_lookup_range_nr_tag(&pvec, mapping, &index, 888 end, PAGECACHE_TAG_DIRTY, 889 max_pages - locked_pages); 890 dout("pagevec_lookup_range_tag got %d\n", pvec_pages); 891 if (!pvec_pages && !locked_pages) 892 break; 893 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 894 page = pvec.pages[i]; 895 dout("? %p idx %lu\n", page, page->index); 896 if (locked_pages == 0) 897 lock_page(page); /* first page */ 898 else if (!trylock_page(page)) 899 break; 900 901 /* only dirty pages, or our accounting breaks */ 902 if (unlikely(!PageDirty(page)) || 903 unlikely(page->mapping != mapping)) { 904 dout("!dirty or !mapping %p\n", page); 905 unlock_page(page); 906 continue; 907 } 908 /* only if matching snap context */ 909 pgsnapc = page_snap_context(page); 910 if (pgsnapc != snapc) { 911 dout("page snapc %p %lld != oldest %p %lld\n", 912 pgsnapc, pgsnapc->seq, snapc, snapc->seq); 913 if (!should_loop && 914 !ceph_wbc.head_snapc && 915 wbc->sync_mode != WB_SYNC_NONE) 916 should_loop = true; 917 unlock_page(page); 918 continue; 919 } 920 if (page_offset(page) >= ceph_wbc.i_size) { 921 dout("%p page eof %llu\n", 922 page, ceph_wbc.i_size); 923 if ((ceph_wbc.size_stable || 924 page_offset(page) >= i_size_read(inode)) && 925 clear_page_dirty_for_io(page)) 926 mapping->a_ops->invalidatepage(page, 927 0, PAGE_SIZE); 928 unlock_page(page); 929 continue; 930 } 931 if (strip_unit_end && (page->index > strip_unit_end)) { 932 dout("end of strip unit %p\n", page); 933 unlock_page(page); 934 break; 935 } 936 if (PageWriteback(page)) { 937 if (wbc->sync_mode == WB_SYNC_NONE) { 938 dout("%p under writeback\n", page); 939 unlock_page(page); 940 continue; 941 } 942 dout("waiting on writeback %p\n", page); 943 wait_on_page_writeback(page); 944 } 945 946 if (!clear_page_dirty_for_io(page)) { 947 dout("%p !clear_page_dirty_for_io\n", page); 948 unlock_page(page); 949 continue; 950 } 951 952 /* 953 * We have something to write. If this is 954 * the first locked page this time through, 955 * calculate max possinle write size and 956 * allocate a page array 957 */ 958 if (locked_pages == 0) { 959 u64 objnum; 960 u64 objoff; 961 u32 xlen; 962 963 /* prepare async write request */ 964 offset = (u64)page_offset(page); 965 ceph_calc_file_object_mapping(&ci->i_layout, 966 offset, wsize, 967 &objnum, &objoff, 968 &xlen); 969 len = xlen; 970 971 num_ops = 1; 972 strip_unit_end = page->index + 973 ((len - 1) >> PAGE_SHIFT); 974 975 BUG_ON(pages); 976 max_pages = calc_pages_for(0, (u64)len); 977 pages = kmalloc_array(max_pages, 978 sizeof(*pages), 979 GFP_NOFS); 980 if (!pages) { 981 pool = fsc->wb_pagevec_pool; 982 pages = mempool_alloc(pool, GFP_NOFS); 983 BUG_ON(!pages); 984 } 985 986 len = 0; 987 } else if (page->index != 988 (offset + len) >> PAGE_SHIFT) { 989 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS : 990 CEPH_OSD_MAX_OPS)) { 991 redirty_page_for_writepage(wbc, page); 992 unlock_page(page); 993 break; 994 } 995 996 num_ops++; 997 offset = (u64)page_offset(page); 998 len = 0; 999 } 1000 1001 /* note position of first page in pvec */ 1002 dout("%p will write page %p idx %lu\n", 1003 inode, page, page->index); 1004 1005 if (atomic_long_inc_return(&fsc->writeback_count) > 1006 CONGESTION_ON_THRESH( 1007 fsc->mount_options->congestion_kb)) { 1008 set_bdi_congested(inode_to_bdi(inode), 1009 BLK_RW_ASYNC); 1010 } 1011 1012 1013 pages[locked_pages++] = page; 1014 pvec.pages[i] = NULL; 1015 1016 len += PAGE_SIZE; 1017 } 1018 1019 /* did we get anything? */ 1020 if (!locked_pages) 1021 goto release_pvec_pages; 1022 if (i) { 1023 unsigned j, n = 0; 1024 /* shift unused page to beginning of pvec */ 1025 for (j = 0; j < pvec_pages; j++) { 1026 if (!pvec.pages[j]) 1027 continue; 1028 if (n < j) 1029 pvec.pages[n] = pvec.pages[j]; 1030 n++; 1031 } 1032 pvec.nr = n; 1033 1034 if (pvec_pages && i == pvec_pages && 1035 locked_pages < max_pages) { 1036 dout("reached end pvec, trying for more\n"); 1037 pagevec_release(&pvec); 1038 goto get_more_pages; 1039 } 1040 } 1041 1042 new_request: 1043 offset = page_offset(pages[0]); 1044 len = wsize; 1045 1046 req = ceph_osdc_new_request(&fsc->client->osdc, 1047 &ci->i_layout, vino, 1048 offset, &len, 0, num_ops, 1049 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1050 snapc, ceph_wbc.truncate_seq, 1051 ceph_wbc.truncate_size, false); 1052 if (IS_ERR(req)) { 1053 req = ceph_osdc_new_request(&fsc->client->osdc, 1054 &ci->i_layout, vino, 1055 offset, &len, 0, 1056 min(num_ops, 1057 CEPH_OSD_SLAB_OPS), 1058 CEPH_OSD_OP_WRITE, 1059 CEPH_OSD_FLAG_WRITE, 1060 snapc, ceph_wbc.truncate_seq, 1061 ceph_wbc.truncate_size, true); 1062 BUG_ON(IS_ERR(req)); 1063 } 1064 BUG_ON(len < page_offset(pages[locked_pages - 1]) + 1065 PAGE_SIZE - offset); 1066 1067 req->r_callback = writepages_finish; 1068 req->r_inode = inode; 1069 1070 /* Format the osd request message and submit the write */ 1071 len = 0; 1072 data_pages = pages; 1073 op_idx = 0; 1074 for (i = 0; i < locked_pages; i++) { 1075 u64 cur_offset = page_offset(pages[i]); 1076 if (offset + len != cur_offset) { 1077 if (op_idx + 1 == req->r_num_ops) 1078 break; 1079 osd_req_op_extent_dup_last(req, op_idx, 1080 cur_offset - offset); 1081 dout("writepages got pages at %llu~%llu\n", 1082 offset, len); 1083 osd_req_op_extent_osd_data_pages(req, op_idx, 1084 data_pages, len, 0, 1085 !!pool, false); 1086 osd_req_op_extent_update(req, op_idx, len); 1087 1088 len = 0; 1089 offset = cur_offset; 1090 data_pages = pages + i; 1091 op_idx++; 1092 } 1093 1094 set_page_writeback(pages[i]); 1095 len += PAGE_SIZE; 1096 } 1097 1098 if (ceph_wbc.size_stable) { 1099 len = min(len, ceph_wbc.i_size - offset); 1100 } else if (i == locked_pages) { 1101 /* writepages_finish() clears writeback pages 1102 * according to the data length, so make sure 1103 * data length covers all locked pages */ 1104 u64 min_len = len + 1 - PAGE_SIZE; 1105 len = get_writepages_data_length(inode, pages[i - 1], 1106 offset); 1107 len = max(len, min_len); 1108 } 1109 dout("writepages got pages at %llu~%llu\n", offset, len); 1110 1111 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, 1112 0, !!pool, false); 1113 osd_req_op_extent_update(req, op_idx, len); 1114 1115 BUG_ON(op_idx + 1 != req->r_num_ops); 1116 1117 pool = NULL; 1118 if (i < locked_pages) { 1119 BUG_ON(num_ops <= req->r_num_ops); 1120 num_ops -= req->r_num_ops; 1121 locked_pages -= i; 1122 1123 /* allocate new pages array for next request */ 1124 data_pages = pages; 1125 pages = kmalloc_array(locked_pages, sizeof(*pages), 1126 GFP_NOFS); 1127 if (!pages) { 1128 pool = fsc->wb_pagevec_pool; 1129 pages = mempool_alloc(pool, GFP_NOFS); 1130 BUG_ON(!pages); 1131 } 1132 memcpy(pages, data_pages + i, 1133 locked_pages * sizeof(*pages)); 1134 memset(data_pages + i, 0, 1135 locked_pages * sizeof(*pages)); 1136 } else { 1137 BUG_ON(num_ops != req->r_num_ops); 1138 index = pages[i - 1]->index + 1; 1139 /* request message now owns the pages array */ 1140 pages = NULL; 1141 } 1142 1143 req->r_mtime = inode->i_mtime; 1144 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); 1145 BUG_ON(rc); 1146 req = NULL; 1147 1148 wbc->nr_to_write -= i; 1149 if (pages) 1150 goto new_request; 1151 1152 /* 1153 * We stop writing back only if we are not doing 1154 * integrity sync. In case of integrity sync we have to 1155 * keep going until we have written all the pages 1156 * we tagged for writeback prior to entering this loop. 1157 */ 1158 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) 1159 done = true; 1160 1161 release_pvec_pages: 1162 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 1163 pvec.nr ? pvec.pages[0] : NULL); 1164 pagevec_release(&pvec); 1165 } 1166 1167 if (should_loop && !done) { 1168 /* more to do; loop back to beginning of file */ 1169 dout("writepages looping back to beginning of file\n"); 1170 end = start_index - 1; /* OK even when start_index == 0 */ 1171 1172 /* to write dirty pages associated with next snapc, 1173 * we need to wait until current writes complete */ 1174 if (wbc->sync_mode != WB_SYNC_NONE && 1175 start_index == 0 && /* all dirty pages were checked */ 1176 !ceph_wbc.head_snapc) { 1177 struct page *page; 1178 unsigned i, nr; 1179 index = 0; 1180 while ((index <= end) && 1181 (nr = pagevec_lookup_tag(&pvec, mapping, &index, 1182 PAGECACHE_TAG_WRITEBACK))) { 1183 for (i = 0; i < nr; i++) { 1184 page = pvec.pages[i]; 1185 if (page_snap_context(page) != snapc) 1186 continue; 1187 wait_on_page_writeback(page); 1188 } 1189 pagevec_release(&pvec); 1190 cond_resched(); 1191 } 1192 } 1193 1194 start_index = 0; 1195 index = 0; 1196 goto retry; 1197 } 1198 1199 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1200 mapping->writeback_index = index; 1201 1202 out: 1203 ceph_osdc_put_request(req); 1204 ceph_put_snap_context(last_snapc); 1205 dout("writepages dend - startone, rc = %d\n", rc); 1206 return rc; 1207 } 1208 1209 1210 1211 /* 1212 * See if a given @snapc is either writeable, or already written. 1213 */ 1214 static int context_is_writeable_or_written(struct inode *inode, 1215 struct ceph_snap_context *snapc) 1216 { 1217 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); 1218 int ret = !oldest || snapc->seq <= oldest->seq; 1219 1220 ceph_put_snap_context(oldest); 1221 return ret; 1222 } 1223 1224 /* 1225 * We are only allowed to write into/dirty the page if the page is 1226 * clean, or already dirty within the same snap context. 1227 * 1228 * called with page locked. 1229 * return success with page locked, 1230 * or any failure (incl -EAGAIN) with page unlocked. 1231 */ 1232 static int ceph_update_writeable_page(struct file *file, 1233 loff_t pos, unsigned len, 1234 struct page *page) 1235 { 1236 struct inode *inode = file_inode(file); 1237 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1238 struct ceph_inode_info *ci = ceph_inode(inode); 1239 loff_t page_off = pos & PAGE_MASK; 1240 int pos_in_page = pos & ~PAGE_MASK; 1241 int end_in_page = pos_in_page + len; 1242 loff_t i_size; 1243 int r; 1244 struct ceph_snap_context *snapc, *oldest; 1245 1246 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { 1247 dout(" page %p forced umount\n", page); 1248 unlock_page(page); 1249 return -EIO; 1250 } 1251 1252 retry_locked: 1253 /* writepages currently holds page lock, but if we change that later, */ 1254 wait_on_page_writeback(page); 1255 1256 snapc = page_snap_context(page); 1257 if (snapc && snapc != ci->i_head_snapc) { 1258 /* 1259 * this page is already dirty in another (older) snap 1260 * context! is it writeable now? 1261 */ 1262 oldest = get_oldest_context(inode, NULL, NULL); 1263 if (snapc->seq > oldest->seq) { 1264 ceph_put_snap_context(oldest); 1265 dout(" page %p snapc %p not current or oldest\n", 1266 page, snapc); 1267 /* 1268 * queue for writeback, and wait for snapc to 1269 * be writeable or written 1270 */ 1271 snapc = ceph_get_snap_context(snapc); 1272 unlock_page(page); 1273 ceph_queue_writeback(inode); 1274 r = wait_event_killable(ci->i_cap_wq, 1275 context_is_writeable_or_written(inode, snapc)); 1276 ceph_put_snap_context(snapc); 1277 if (r == -ERESTARTSYS) 1278 return r; 1279 return -EAGAIN; 1280 } 1281 ceph_put_snap_context(oldest); 1282 1283 /* yay, writeable, do it now (without dropping page lock) */ 1284 dout(" page %p snapc %p not current, but oldest\n", 1285 page, snapc); 1286 if (!clear_page_dirty_for_io(page)) 1287 goto retry_locked; 1288 r = writepage_nounlock(page, NULL); 1289 if (r < 0) 1290 goto fail_unlock; 1291 goto retry_locked; 1292 } 1293 1294 if (PageUptodate(page)) { 1295 dout(" page %p already uptodate\n", page); 1296 return 0; 1297 } 1298 1299 /* full page? */ 1300 if (pos_in_page == 0 && len == PAGE_SIZE) 1301 return 0; 1302 1303 /* past end of file? */ 1304 i_size = i_size_read(inode); 1305 1306 if (page_off >= i_size || 1307 (pos_in_page == 0 && (pos+len) >= i_size && 1308 end_in_page - pos_in_page != PAGE_SIZE)) { 1309 dout(" zeroing %p 0 - %d and %d - %d\n", 1310 page, pos_in_page, end_in_page, (int)PAGE_SIZE); 1311 zero_user_segments(page, 1312 0, pos_in_page, 1313 end_in_page, PAGE_SIZE); 1314 return 0; 1315 } 1316 1317 /* we need to read it. */ 1318 r = ceph_do_readpage(file, page); 1319 if (r < 0) { 1320 if (r == -EINPROGRESS) 1321 return -EAGAIN; 1322 goto fail_unlock; 1323 } 1324 goto retry_locked; 1325 fail_unlock: 1326 unlock_page(page); 1327 return r; 1328 } 1329 1330 /* 1331 * We are only allowed to write into/dirty the page if the page is 1332 * clean, or already dirty within the same snap context. 1333 */ 1334 static int ceph_write_begin(struct file *file, struct address_space *mapping, 1335 loff_t pos, unsigned len, unsigned flags, 1336 struct page **pagep, void **fsdata) 1337 { 1338 struct inode *inode = file_inode(file); 1339 struct page *page; 1340 pgoff_t index = pos >> PAGE_SHIFT; 1341 int r; 1342 1343 do { 1344 /* get a page */ 1345 page = grab_cache_page_write_begin(mapping, index, 0); 1346 if (!page) 1347 return -ENOMEM; 1348 1349 dout("write_begin file %p inode %p page %p %d~%d\n", file, 1350 inode, page, (int)pos, (int)len); 1351 1352 r = ceph_update_writeable_page(file, pos, len, page); 1353 if (r < 0) 1354 put_page(page); 1355 else 1356 *pagep = page; 1357 } while (r == -EAGAIN); 1358 1359 return r; 1360 } 1361 1362 /* 1363 * we don't do anything in here that simple_write_end doesn't do 1364 * except adjust dirty page accounting 1365 */ 1366 static int ceph_write_end(struct file *file, struct address_space *mapping, 1367 loff_t pos, unsigned len, unsigned copied, 1368 struct page *page, void *fsdata) 1369 { 1370 struct inode *inode = file_inode(file); 1371 bool check_cap = false; 1372 1373 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, 1374 inode, page, (int)pos, (int)copied, (int)len); 1375 1376 /* zero the stale part of the page if we did a short copy */ 1377 if (!PageUptodate(page)) { 1378 if (copied < len) { 1379 copied = 0; 1380 goto out; 1381 } 1382 SetPageUptodate(page); 1383 } 1384 1385 /* did file size increase? */ 1386 if (pos+copied > i_size_read(inode)) 1387 check_cap = ceph_inode_set_size(inode, pos+copied); 1388 1389 set_page_dirty(page); 1390 1391 out: 1392 unlock_page(page); 1393 put_page(page); 1394 1395 if (check_cap) 1396 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); 1397 1398 return copied; 1399 } 1400 1401 /* 1402 * we set .direct_IO to indicate direct io is supported, but since we 1403 * intercept O_DIRECT reads and writes early, this function should 1404 * never get called. 1405 */ 1406 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter) 1407 { 1408 WARN_ON(1); 1409 return -EINVAL; 1410 } 1411 1412 const struct address_space_operations ceph_aops = { 1413 .readpage = ceph_readpage, 1414 .readpages = ceph_readpages, 1415 .writepage = ceph_writepage, 1416 .writepages = ceph_writepages_start, 1417 .write_begin = ceph_write_begin, 1418 .write_end = ceph_write_end, 1419 .set_page_dirty = ceph_set_page_dirty, 1420 .invalidatepage = ceph_invalidatepage, 1421 .releasepage = ceph_releasepage, 1422 .direct_IO = ceph_direct_io, 1423 }; 1424 1425 static void ceph_block_sigs(sigset_t *oldset) 1426 { 1427 sigset_t mask; 1428 siginitsetinv(&mask, sigmask(SIGKILL)); 1429 sigprocmask(SIG_BLOCK, &mask, oldset); 1430 } 1431 1432 static void ceph_restore_sigs(sigset_t *oldset) 1433 { 1434 sigprocmask(SIG_SETMASK, oldset, NULL); 1435 } 1436 1437 /* 1438 * vm ops 1439 */ 1440 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf) 1441 { 1442 struct vm_area_struct *vma = vmf->vma; 1443 struct inode *inode = file_inode(vma->vm_file); 1444 struct ceph_inode_info *ci = ceph_inode(inode); 1445 struct ceph_file_info *fi = vma->vm_file->private_data; 1446 struct page *pinned_page = NULL; 1447 loff_t off = vmf->pgoff << PAGE_SHIFT; 1448 int want, got, err; 1449 sigset_t oldset; 1450 vm_fault_t ret = VM_FAULT_SIGBUS; 1451 1452 ceph_block_sigs(&oldset); 1453 1454 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n", 1455 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE); 1456 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1457 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1458 else 1459 want = CEPH_CAP_FILE_CACHE; 1460 1461 got = 0; 1462 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, 1463 &got, &pinned_page); 1464 if (err < 0) 1465 goto out_restore; 1466 1467 dout("filemap_fault %p %llu~%zd got cap refs on %s\n", 1468 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got)); 1469 1470 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || 1471 ci->i_inline_version == CEPH_INLINE_NONE) { 1472 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); 1473 ceph_add_rw_context(fi, &rw_ctx); 1474 ret = filemap_fault(vmf); 1475 ceph_del_rw_context(fi, &rw_ctx); 1476 dout("filemap_fault %p %llu~%zd drop cap refs %s ret %x\n", 1477 inode, off, (size_t)PAGE_SIZE, 1478 ceph_cap_string(got), ret); 1479 } else 1480 err = -EAGAIN; 1481 1482 if (pinned_page) 1483 put_page(pinned_page); 1484 ceph_put_cap_refs(ci, got); 1485 1486 if (err != -EAGAIN) 1487 goto out_restore; 1488 1489 /* read inline data */ 1490 if (off >= PAGE_SIZE) { 1491 /* does not support inline data > PAGE_SIZE */ 1492 ret = VM_FAULT_SIGBUS; 1493 } else { 1494 struct address_space *mapping = inode->i_mapping; 1495 struct page *page = find_or_create_page(mapping, 0, 1496 mapping_gfp_constraint(mapping, 1497 ~__GFP_FS)); 1498 if (!page) { 1499 ret = VM_FAULT_OOM; 1500 goto out_inline; 1501 } 1502 err = __ceph_do_getattr(inode, page, 1503 CEPH_STAT_CAP_INLINE_DATA, true); 1504 if (err < 0 || off >= i_size_read(inode)) { 1505 unlock_page(page); 1506 put_page(page); 1507 ret = vmf_error(err); 1508 goto out_inline; 1509 } 1510 if (err < PAGE_SIZE) 1511 zero_user_segment(page, err, PAGE_SIZE); 1512 else 1513 flush_dcache_page(page); 1514 SetPageUptodate(page); 1515 vmf->page = page; 1516 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; 1517 out_inline: 1518 dout("filemap_fault %p %llu~%zd read inline data ret %x\n", 1519 inode, off, (size_t)PAGE_SIZE, ret); 1520 } 1521 out_restore: 1522 ceph_restore_sigs(&oldset); 1523 if (err < 0) 1524 ret = vmf_error(err); 1525 1526 return ret; 1527 } 1528 1529 /* 1530 * Reuse write_begin here for simplicity. 1531 */ 1532 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf) 1533 { 1534 struct vm_area_struct *vma = vmf->vma; 1535 struct inode *inode = file_inode(vma->vm_file); 1536 struct ceph_inode_info *ci = ceph_inode(inode); 1537 struct ceph_file_info *fi = vma->vm_file->private_data; 1538 struct ceph_cap_flush *prealloc_cf; 1539 struct page *page = vmf->page; 1540 loff_t off = page_offset(page); 1541 loff_t size = i_size_read(inode); 1542 size_t len; 1543 int want, got, err; 1544 sigset_t oldset; 1545 vm_fault_t ret = VM_FAULT_SIGBUS; 1546 1547 prealloc_cf = ceph_alloc_cap_flush(); 1548 if (!prealloc_cf) 1549 return VM_FAULT_OOM; 1550 1551 sb_start_pagefault(inode->i_sb); 1552 ceph_block_sigs(&oldset); 1553 1554 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1555 struct page *locked_page = NULL; 1556 if (off == 0) { 1557 lock_page(page); 1558 locked_page = page; 1559 } 1560 err = ceph_uninline_data(vma->vm_file, locked_page); 1561 if (locked_page) 1562 unlock_page(locked_page); 1563 if (err < 0) 1564 goto out_free; 1565 } 1566 1567 if (off + PAGE_SIZE <= size) 1568 len = PAGE_SIZE; 1569 else 1570 len = size & ~PAGE_MASK; 1571 1572 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", 1573 inode, ceph_vinop(inode), off, len, size); 1574 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1575 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 1576 else 1577 want = CEPH_CAP_FILE_BUFFER; 1578 1579 got = 0; 1580 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, 1581 &got, NULL); 1582 if (err < 0) 1583 goto out_free; 1584 1585 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", 1586 inode, off, len, ceph_cap_string(got)); 1587 1588 /* Update time before taking page lock */ 1589 file_update_time(vma->vm_file); 1590 inode_inc_iversion_raw(inode); 1591 1592 do { 1593 lock_page(page); 1594 1595 if ((off > size) || (page->mapping != inode->i_mapping)) { 1596 unlock_page(page); 1597 ret = VM_FAULT_NOPAGE; 1598 break; 1599 } 1600 1601 err = ceph_update_writeable_page(vma->vm_file, off, len, page); 1602 if (err >= 0) { 1603 /* success. we'll keep the page locked. */ 1604 set_page_dirty(page); 1605 ret = VM_FAULT_LOCKED; 1606 } 1607 } while (err == -EAGAIN); 1608 1609 if (ret == VM_FAULT_LOCKED || 1610 ci->i_inline_version != CEPH_INLINE_NONE) { 1611 int dirty; 1612 spin_lock(&ci->i_ceph_lock); 1613 ci->i_inline_version = CEPH_INLINE_NONE; 1614 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1615 &prealloc_cf); 1616 spin_unlock(&ci->i_ceph_lock); 1617 if (dirty) 1618 __mark_inode_dirty(inode, dirty); 1619 } 1620 1621 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n", 1622 inode, off, len, ceph_cap_string(got), ret); 1623 ceph_put_cap_refs(ci, got); 1624 out_free: 1625 ceph_restore_sigs(&oldset); 1626 sb_end_pagefault(inode->i_sb); 1627 ceph_free_cap_flush(prealloc_cf); 1628 if (err < 0) 1629 ret = vmf_error(err); 1630 return ret; 1631 } 1632 1633 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, 1634 char *data, size_t len) 1635 { 1636 struct address_space *mapping = inode->i_mapping; 1637 struct page *page; 1638 1639 if (locked_page) { 1640 page = locked_page; 1641 } else { 1642 if (i_size_read(inode) == 0) 1643 return; 1644 page = find_or_create_page(mapping, 0, 1645 mapping_gfp_constraint(mapping, 1646 ~__GFP_FS)); 1647 if (!page) 1648 return; 1649 if (PageUptodate(page)) { 1650 unlock_page(page); 1651 put_page(page); 1652 return; 1653 } 1654 } 1655 1656 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", 1657 inode, ceph_vinop(inode), len, locked_page); 1658 1659 if (len > 0) { 1660 void *kaddr = kmap_atomic(page); 1661 memcpy(kaddr, data, len); 1662 kunmap_atomic(kaddr); 1663 } 1664 1665 if (page != locked_page) { 1666 if (len < PAGE_SIZE) 1667 zero_user_segment(page, len, PAGE_SIZE); 1668 else 1669 flush_dcache_page(page); 1670 1671 SetPageUptodate(page); 1672 unlock_page(page); 1673 put_page(page); 1674 } 1675 } 1676 1677 int ceph_uninline_data(struct file *filp, struct page *locked_page) 1678 { 1679 struct inode *inode = file_inode(filp); 1680 struct ceph_inode_info *ci = ceph_inode(inode); 1681 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1682 struct ceph_osd_request *req; 1683 struct page *page = NULL; 1684 u64 len, inline_version; 1685 int err = 0; 1686 bool from_pagecache = false; 1687 1688 spin_lock(&ci->i_ceph_lock); 1689 inline_version = ci->i_inline_version; 1690 spin_unlock(&ci->i_ceph_lock); 1691 1692 dout("uninline_data %p %llx.%llx inline_version %llu\n", 1693 inode, ceph_vinop(inode), inline_version); 1694 1695 if (inline_version == 1 || /* initial version, no data */ 1696 inline_version == CEPH_INLINE_NONE) 1697 goto out; 1698 1699 if (locked_page) { 1700 page = locked_page; 1701 WARN_ON(!PageUptodate(page)); 1702 } else if (ceph_caps_issued(ci) & 1703 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { 1704 page = find_get_page(inode->i_mapping, 0); 1705 if (page) { 1706 if (PageUptodate(page)) { 1707 from_pagecache = true; 1708 lock_page(page); 1709 } else { 1710 put_page(page); 1711 page = NULL; 1712 } 1713 } 1714 } 1715 1716 if (page) { 1717 len = i_size_read(inode); 1718 if (len > PAGE_SIZE) 1719 len = PAGE_SIZE; 1720 } else { 1721 page = __page_cache_alloc(GFP_NOFS); 1722 if (!page) { 1723 err = -ENOMEM; 1724 goto out; 1725 } 1726 err = __ceph_do_getattr(inode, page, 1727 CEPH_STAT_CAP_INLINE_DATA, true); 1728 if (err < 0) { 1729 /* no inline data */ 1730 if (err == -ENODATA) 1731 err = 0; 1732 goto out; 1733 } 1734 len = err; 1735 } 1736 1737 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1738 ceph_vino(inode), 0, &len, 0, 1, 1739 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, 1740 NULL, 0, 0, false); 1741 if (IS_ERR(req)) { 1742 err = PTR_ERR(req); 1743 goto out; 1744 } 1745 1746 req->r_mtime = inode->i_mtime; 1747 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1748 if (!err) 1749 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1750 ceph_osdc_put_request(req); 1751 if (err < 0) 1752 goto out; 1753 1754 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1755 ceph_vino(inode), 0, &len, 1, 3, 1756 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1757 NULL, ci->i_truncate_seq, 1758 ci->i_truncate_size, false); 1759 if (IS_ERR(req)) { 1760 err = PTR_ERR(req); 1761 goto out; 1762 } 1763 1764 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); 1765 1766 { 1767 __le64 xattr_buf = cpu_to_le64(inline_version); 1768 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, 1769 "inline_version", &xattr_buf, 1770 sizeof(xattr_buf), 1771 CEPH_OSD_CMPXATTR_OP_GT, 1772 CEPH_OSD_CMPXATTR_MODE_U64); 1773 if (err) 1774 goto out_put; 1775 } 1776 1777 { 1778 char xattr_buf[32]; 1779 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), 1780 "%llu", inline_version); 1781 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, 1782 "inline_version", 1783 xattr_buf, xattr_len, 0, 0); 1784 if (err) 1785 goto out_put; 1786 } 1787 1788 req->r_mtime = inode->i_mtime; 1789 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1790 if (!err) 1791 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1792 out_put: 1793 ceph_osdc_put_request(req); 1794 if (err == -ECANCELED) 1795 err = 0; 1796 out: 1797 if (page && page != locked_page) { 1798 if (from_pagecache) { 1799 unlock_page(page); 1800 put_page(page); 1801 } else 1802 __free_pages(page, 0); 1803 } 1804 1805 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", 1806 inode, ceph_vinop(inode), inline_version, err); 1807 return err; 1808 } 1809 1810 static const struct vm_operations_struct ceph_vmops = { 1811 .fault = ceph_filemap_fault, 1812 .page_mkwrite = ceph_page_mkwrite, 1813 }; 1814 1815 int ceph_mmap(struct file *file, struct vm_area_struct *vma) 1816 { 1817 struct address_space *mapping = file->f_mapping; 1818 1819 if (!mapping->a_ops->readpage) 1820 return -ENOEXEC; 1821 file_accessed(file); 1822 vma->vm_ops = &ceph_vmops; 1823 return 0; 1824 } 1825 1826 enum { 1827 POOL_READ = 1, 1828 POOL_WRITE = 2, 1829 }; 1830 1831 static int __ceph_pool_perm_get(struct ceph_inode_info *ci, 1832 s64 pool, struct ceph_string *pool_ns) 1833 { 1834 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 1835 struct ceph_mds_client *mdsc = fsc->mdsc; 1836 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; 1837 struct rb_node **p, *parent; 1838 struct ceph_pool_perm *perm; 1839 struct page **pages; 1840 size_t pool_ns_len; 1841 int err = 0, err2 = 0, have = 0; 1842 1843 down_read(&mdsc->pool_perm_rwsem); 1844 p = &mdsc->pool_perm_tree.rb_node; 1845 while (*p) { 1846 perm = rb_entry(*p, struct ceph_pool_perm, node); 1847 if (pool < perm->pool) 1848 p = &(*p)->rb_left; 1849 else if (pool > perm->pool) 1850 p = &(*p)->rb_right; 1851 else { 1852 int ret = ceph_compare_string(pool_ns, 1853 perm->pool_ns, 1854 perm->pool_ns_len); 1855 if (ret < 0) 1856 p = &(*p)->rb_left; 1857 else if (ret > 0) 1858 p = &(*p)->rb_right; 1859 else { 1860 have = perm->perm; 1861 break; 1862 } 1863 } 1864 } 1865 up_read(&mdsc->pool_perm_rwsem); 1866 if (*p) 1867 goto out; 1868 1869 if (pool_ns) 1870 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", 1871 pool, (int)pool_ns->len, pool_ns->str); 1872 else 1873 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); 1874 1875 down_write(&mdsc->pool_perm_rwsem); 1876 p = &mdsc->pool_perm_tree.rb_node; 1877 parent = NULL; 1878 while (*p) { 1879 parent = *p; 1880 perm = rb_entry(parent, struct ceph_pool_perm, node); 1881 if (pool < perm->pool) 1882 p = &(*p)->rb_left; 1883 else if (pool > perm->pool) 1884 p = &(*p)->rb_right; 1885 else { 1886 int ret = ceph_compare_string(pool_ns, 1887 perm->pool_ns, 1888 perm->pool_ns_len); 1889 if (ret < 0) 1890 p = &(*p)->rb_left; 1891 else if (ret > 0) 1892 p = &(*p)->rb_right; 1893 else { 1894 have = perm->perm; 1895 break; 1896 } 1897 } 1898 } 1899 if (*p) { 1900 up_write(&mdsc->pool_perm_rwsem); 1901 goto out; 1902 } 1903 1904 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1905 1, false, GFP_NOFS); 1906 if (!rd_req) { 1907 err = -ENOMEM; 1908 goto out_unlock; 1909 } 1910 1911 rd_req->r_flags = CEPH_OSD_FLAG_READ; 1912 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); 1913 rd_req->r_base_oloc.pool = pool; 1914 if (pool_ns) 1915 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); 1916 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); 1917 1918 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); 1919 if (err) 1920 goto out_unlock; 1921 1922 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1923 1, false, GFP_NOFS); 1924 if (!wr_req) { 1925 err = -ENOMEM; 1926 goto out_unlock; 1927 } 1928 1929 wr_req->r_flags = CEPH_OSD_FLAG_WRITE; 1930 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); 1931 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); 1932 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); 1933 1934 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); 1935 if (err) 1936 goto out_unlock; 1937 1938 /* one page should be large enough for STAT data */ 1939 pages = ceph_alloc_page_vector(1, GFP_KERNEL); 1940 if (IS_ERR(pages)) { 1941 err = PTR_ERR(pages); 1942 goto out_unlock; 1943 } 1944 1945 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, 1946 0, false, true); 1947 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); 1948 1949 wr_req->r_mtime = ci->vfs_inode.i_mtime; 1950 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); 1951 1952 if (!err) 1953 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); 1954 if (!err2) 1955 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); 1956 1957 if (err >= 0 || err == -ENOENT) 1958 have |= POOL_READ; 1959 else if (err != -EPERM) { 1960 if (err == -EBLACKLISTED) 1961 fsc->blacklisted = true; 1962 goto out_unlock; 1963 } 1964 1965 if (err2 == 0 || err2 == -EEXIST) 1966 have |= POOL_WRITE; 1967 else if (err2 != -EPERM) { 1968 if (err2 == -EBLACKLISTED) 1969 fsc->blacklisted = true; 1970 err = err2; 1971 goto out_unlock; 1972 } 1973 1974 pool_ns_len = pool_ns ? pool_ns->len : 0; 1975 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); 1976 if (!perm) { 1977 err = -ENOMEM; 1978 goto out_unlock; 1979 } 1980 1981 perm->pool = pool; 1982 perm->perm = have; 1983 perm->pool_ns_len = pool_ns_len; 1984 if (pool_ns_len > 0) 1985 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); 1986 perm->pool_ns[pool_ns_len] = 0; 1987 1988 rb_link_node(&perm->node, parent, p); 1989 rb_insert_color(&perm->node, &mdsc->pool_perm_tree); 1990 err = 0; 1991 out_unlock: 1992 up_write(&mdsc->pool_perm_rwsem); 1993 1994 ceph_osdc_put_request(rd_req); 1995 ceph_osdc_put_request(wr_req); 1996 out: 1997 if (!err) 1998 err = have; 1999 if (pool_ns) 2000 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", 2001 pool, (int)pool_ns->len, pool_ns->str, err); 2002 else 2003 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); 2004 return err; 2005 } 2006 2007 int ceph_pool_perm_check(struct inode *inode, int need) 2008 { 2009 struct ceph_inode_info *ci = ceph_inode(inode); 2010 struct ceph_string *pool_ns; 2011 s64 pool; 2012 int ret, flags; 2013 2014 if (ci->i_vino.snap != CEPH_NOSNAP) { 2015 /* 2016 * Pool permission check needs to write to the first object. 2017 * But for snapshot, head of the first object may have alread 2018 * been deleted. Skip check to avoid creating orphan object. 2019 */ 2020 return 0; 2021 } 2022 2023 if (ceph_test_mount_opt(ceph_inode_to_client(inode), 2024 NOPOOLPERM)) 2025 return 0; 2026 2027 spin_lock(&ci->i_ceph_lock); 2028 flags = ci->i_ceph_flags; 2029 pool = ci->i_layout.pool_id; 2030 spin_unlock(&ci->i_ceph_lock); 2031 check: 2032 if (flags & CEPH_I_POOL_PERM) { 2033 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { 2034 dout("ceph_pool_perm_check pool %lld no read perm\n", 2035 pool); 2036 return -EPERM; 2037 } 2038 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { 2039 dout("ceph_pool_perm_check pool %lld no write perm\n", 2040 pool); 2041 return -EPERM; 2042 } 2043 return 0; 2044 } 2045 2046 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); 2047 ret = __ceph_pool_perm_get(ci, pool, pool_ns); 2048 ceph_put_string(pool_ns); 2049 if (ret < 0) 2050 return ret; 2051 2052 flags = CEPH_I_POOL_PERM; 2053 if (ret & POOL_READ) 2054 flags |= CEPH_I_POOL_RD; 2055 if (ret & POOL_WRITE) 2056 flags |= CEPH_I_POOL_WR; 2057 2058 spin_lock(&ci->i_ceph_lock); 2059 if (pool == ci->i_layout.pool_id && 2060 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { 2061 ci->i_ceph_flags |= flags; 2062 } else { 2063 pool = ci->i_layout.pool_id; 2064 flags = ci->i_ceph_flags; 2065 } 2066 spin_unlock(&ci->i_ceph_lock); 2067 goto check; 2068 } 2069 2070 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) 2071 { 2072 struct ceph_pool_perm *perm; 2073 struct rb_node *n; 2074 2075 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { 2076 n = rb_first(&mdsc->pool_perm_tree); 2077 perm = rb_entry(n, struct ceph_pool_perm, node); 2078 rb_erase(n, &mdsc->pool_perm_tree); 2079 kfree(perm); 2080 } 2081 } 2082