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/swap.h> 8 #include <linux/pagemap.h> 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 #include <linux/ktime.h> 15 #include <linux/netfs.h> 16 17 #include "super.h" 18 #include "mds_client.h" 19 #include "cache.h" 20 #include "metric.h" 21 #include <linux/ceph/osd_client.h> 22 #include <linux/ceph/striper.h> 23 24 /* 25 * Ceph address space ops. 26 * 27 * There are a few funny things going on here. 28 * 29 * The page->private field is used to reference a struct 30 * ceph_snap_context for _every_ dirty page. This indicates which 31 * snapshot the page was logically dirtied in, and thus which snap 32 * context needs to be associated with the osd write during writeback. 33 * 34 * Similarly, struct ceph_inode_info maintains a set of counters to 35 * count dirty pages on the inode. In the absence of snapshots, 36 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. 37 * 38 * When a snapshot is taken (that is, when the client receives 39 * notification that a snapshot was taken), each inode with caps and 40 * with dirty pages (dirty pages implies there is a cap) gets a new 41 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending 42 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is 43 * moved to capsnap->dirty. (Unless a sync write is currently in 44 * progress. In that case, the capsnap is said to be "pending", new 45 * writes cannot start, and the capsnap isn't "finalized" until the 46 * write completes (or fails) and a final size/mtime for the inode for 47 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. 48 * 49 * On writeback, we must submit writes to the osd IN SNAP ORDER. So, 50 * we look for the first capsnap in i_cap_snaps and write out pages in 51 * that snap context _only_. Then we move on to the next capsnap, 52 * eventually reaching the "live" or "head" context (i.e., pages that 53 * are not yet snapped) and are writing the most recently dirtied 54 * pages. 55 * 56 * Invalidate and so forth must take care to ensure the dirty page 57 * accounting is preserved. 58 */ 59 60 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) 61 #define CONGESTION_OFF_THRESH(congestion_kb) \ 62 (CONGESTION_ON_THRESH(congestion_kb) - \ 63 (CONGESTION_ON_THRESH(congestion_kb) >> 2)) 64 65 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, 66 struct folio *folio, void **_fsdata); 67 68 static inline struct ceph_snap_context *page_snap_context(struct page *page) 69 { 70 if (PagePrivate(page)) 71 return (void *)page->private; 72 return NULL; 73 } 74 75 /* 76 * Dirty a page. Optimistically adjust accounting, on the assumption 77 * that we won't race with invalidate. If we do, readjust. 78 */ 79 static int ceph_set_page_dirty(struct page *page) 80 { 81 struct address_space *mapping = page->mapping; 82 struct inode *inode; 83 struct ceph_inode_info *ci; 84 struct ceph_snap_context *snapc; 85 86 if (PageDirty(page)) { 87 dout("%p set_page_dirty %p idx %lu -- already dirty\n", 88 mapping->host, page, page->index); 89 BUG_ON(!PagePrivate(page)); 90 return 0; 91 } 92 93 inode = mapping->host; 94 ci = ceph_inode(inode); 95 96 /* dirty the head */ 97 spin_lock(&ci->i_ceph_lock); 98 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference 99 if (__ceph_have_pending_cap_snap(ci)) { 100 struct ceph_cap_snap *capsnap = 101 list_last_entry(&ci->i_cap_snaps, 102 struct ceph_cap_snap, 103 ci_item); 104 snapc = ceph_get_snap_context(capsnap->context); 105 capsnap->dirty_pages++; 106 } else { 107 BUG_ON(!ci->i_head_snapc); 108 snapc = ceph_get_snap_context(ci->i_head_snapc); 109 ++ci->i_wrbuffer_ref_head; 110 } 111 if (ci->i_wrbuffer_ref == 0) 112 ihold(inode); 113 ++ci->i_wrbuffer_ref; 114 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " 115 "snapc %p seq %lld (%d snaps)\n", 116 mapping->host, page, page->index, 117 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, 118 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, 119 snapc, snapc->seq, snapc->num_snaps); 120 spin_unlock(&ci->i_ceph_lock); 121 122 /* 123 * Reference snap context in page->private. Also set 124 * PagePrivate so that we get invalidatepage callback. 125 */ 126 BUG_ON(PagePrivate(page)); 127 attach_page_private(page, snapc); 128 129 return ceph_fscache_set_page_dirty(page); 130 } 131 132 /* 133 * If we are truncating the full page (i.e. offset == 0), adjust the 134 * dirty page counters appropriately. Only called if there is private 135 * data on the page. 136 */ 137 static void ceph_invalidatepage(struct page *page, unsigned int offset, 138 unsigned int length) 139 { 140 struct inode *inode; 141 struct ceph_inode_info *ci; 142 struct ceph_snap_context *snapc; 143 144 inode = page->mapping->host; 145 ci = ceph_inode(inode); 146 147 if (offset != 0 || length != thp_size(page)) { 148 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n", 149 inode, page, page->index, offset, length); 150 return; 151 } 152 153 WARN_ON(!PageLocked(page)); 154 if (PagePrivate(page)) { 155 dout("%p invalidatepage %p idx %lu full dirty page\n", 156 inode, page, page->index); 157 158 snapc = detach_page_private(page); 159 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 160 ceph_put_snap_context(snapc); 161 } 162 163 wait_on_page_fscache(page); 164 } 165 166 static int ceph_releasepage(struct page *page, gfp_t gfp) 167 { 168 struct inode *inode = page->mapping->host; 169 170 dout("%llx:%llx releasepage %p idx %lu (%sdirty)\n", 171 ceph_vinop(inode), page, 172 page->index, PageDirty(page) ? "" : "not "); 173 174 if (PagePrivate(page)) 175 return 0; 176 177 if (PageFsCache(page)) { 178 if (current_is_kswapd() || !(gfp & __GFP_FS)) 179 return 0; 180 wait_on_page_fscache(page); 181 } 182 ceph_fscache_note_page_release(inode); 183 return 1; 184 } 185 186 static void ceph_netfs_expand_readahead(struct netfs_read_request *rreq) 187 { 188 struct inode *inode = rreq->mapping->host; 189 struct ceph_inode_info *ci = ceph_inode(inode); 190 struct ceph_file_layout *lo = &ci->i_layout; 191 u32 blockoff; 192 u64 blockno; 193 194 /* Expand the start downward */ 195 blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff); 196 rreq->start = blockno * lo->stripe_unit; 197 rreq->len += blockoff; 198 199 /* Now, round up the length to the next block */ 200 rreq->len = roundup(rreq->len, lo->stripe_unit); 201 } 202 203 static bool ceph_netfs_clamp_length(struct netfs_read_subrequest *subreq) 204 { 205 struct inode *inode = subreq->rreq->mapping->host; 206 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 207 struct ceph_inode_info *ci = ceph_inode(inode); 208 u64 objno, objoff; 209 u32 xlen; 210 211 /* Truncate the extent at the end of the current block */ 212 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len, 213 &objno, &objoff, &xlen); 214 subreq->len = min(xlen, fsc->mount_options->rsize); 215 return true; 216 } 217 218 static void finish_netfs_read(struct ceph_osd_request *req) 219 { 220 struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode); 221 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); 222 struct netfs_read_subrequest *subreq = req->r_priv; 223 int num_pages; 224 int err = req->r_result; 225 226 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency, 227 req->r_end_latency, osd_data->length, err); 228 229 dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result, 230 subreq->len, i_size_read(req->r_inode)); 231 232 /* no object means success but no data */ 233 if (err == -ENOENT) 234 err = 0; 235 else if (err == -EBLOCKLISTED) 236 fsc->blocklisted = true; 237 238 if (err >= 0 && err < subreq->len) 239 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags); 240 241 netfs_subreq_terminated(subreq, err, true); 242 243 num_pages = calc_pages_for(osd_data->alignment, osd_data->length); 244 ceph_put_page_vector(osd_data->pages, num_pages, false); 245 iput(req->r_inode); 246 } 247 248 static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq) 249 { 250 struct netfs_read_request *rreq = subreq->rreq; 251 struct inode *inode = rreq->mapping->host; 252 struct ceph_inode_info *ci = ceph_inode(inode); 253 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 254 struct ceph_osd_request *req; 255 struct ceph_vino vino = ceph_vino(inode); 256 struct iov_iter iter; 257 struct page **pages; 258 size_t page_off; 259 int err = 0; 260 u64 len = subreq->len; 261 262 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len, 263 0, 1, CEPH_OSD_OP_READ, 264 CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica, 265 NULL, ci->i_truncate_seq, ci->i_truncate_size, false); 266 if (IS_ERR(req)) { 267 err = PTR_ERR(req); 268 req = NULL; 269 goto out; 270 } 271 272 dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len); 273 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len); 274 err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off); 275 if (err < 0) { 276 dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err); 277 goto out; 278 } 279 280 /* should always give us a page-aligned read */ 281 WARN_ON_ONCE(page_off); 282 len = err; 283 284 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false); 285 req->r_callback = finish_netfs_read; 286 req->r_priv = subreq; 287 req->r_inode = inode; 288 ihold(inode); 289 290 err = ceph_osdc_start_request(req->r_osdc, req, false); 291 if (err) 292 iput(inode); 293 out: 294 ceph_osdc_put_request(req); 295 if (err) 296 netfs_subreq_terminated(subreq, err, false); 297 dout("%s: result %d\n", __func__, err); 298 } 299 300 static void ceph_init_rreq(struct netfs_read_request *rreq, struct file *file) 301 { 302 } 303 304 static void ceph_readahead_cleanup(struct address_space *mapping, void *priv) 305 { 306 struct inode *inode = mapping->host; 307 struct ceph_inode_info *ci = ceph_inode(inode); 308 int got = (uintptr_t)priv; 309 310 if (got) 311 ceph_put_cap_refs(ci, got); 312 } 313 314 static const struct netfs_read_request_ops ceph_netfs_read_ops = { 315 .init_rreq = ceph_init_rreq, 316 .is_cache_enabled = ceph_is_cache_enabled, 317 .begin_cache_operation = ceph_begin_cache_operation, 318 .issue_op = ceph_netfs_issue_op, 319 .expand_readahead = ceph_netfs_expand_readahead, 320 .clamp_length = ceph_netfs_clamp_length, 321 .check_write_begin = ceph_netfs_check_write_begin, 322 .cleanup = ceph_readahead_cleanup, 323 }; 324 325 /* read a single page, without unlocking it. */ 326 static int ceph_readpage(struct file *file, struct page *subpage) 327 { 328 struct folio *folio = page_folio(subpage); 329 struct inode *inode = file_inode(file); 330 struct ceph_inode_info *ci = ceph_inode(inode); 331 struct ceph_vino vino = ceph_vino(inode); 332 size_t len = folio_size(folio); 333 u64 off = folio_file_pos(folio); 334 335 if (ci->i_inline_version != CEPH_INLINE_NONE) { 336 /* 337 * Uptodate inline data should have been added 338 * into page cache while getting Fcr caps. 339 */ 340 if (off == 0) { 341 folio_unlock(folio); 342 return -EINVAL; 343 } 344 zero_user_segment(&folio->page, 0, folio_size(folio)); 345 folio_mark_uptodate(folio); 346 folio_unlock(folio); 347 return 0; 348 } 349 350 dout("readpage ino %llx.%llx file %p off %llu len %zu folio %p index %lu\n", 351 vino.ino, vino.snap, file, off, len, folio, folio_index(folio)); 352 353 return netfs_readpage(file, folio, &ceph_netfs_read_ops, NULL); 354 } 355 356 static void ceph_readahead(struct readahead_control *ractl) 357 { 358 struct inode *inode = file_inode(ractl->file); 359 struct ceph_file_info *fi = ractl->file->private_data; 360 struct ceph_rw_context *rw_ctx; 361 int got = 0; 362 int ret = 0; 363 364 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE) 365 return; 366 367 rw_ctx = ceph_find_rw_context(fi); 368 if (!rw_ctx) { 369 /* 370 * readahead callers do not necessarily hold Fcb caps 371 * (e.g. fadvise, madvise). 372 */ 373 int want = CEPH_CAP_FILE_CACHE; 374 375 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got); 376 if (ret < 0) 377 dout("start_read %p, error getting cap\n", inode); 378 else if (!(got & want)) 379 dout("start_read %p, no cache cap\n", inode); 380 381 if (ret <= 0) 382 return; 383 } 384 netfs_readahead(ractl, &ceph_netfs_read_ops, (void *)(uintptr_t)got); 385 } 386 387 #ifdef CONFIG_CEPH_FSCACHE 388 static void ceph_set_page_fscache(struct page *page) 389 { 390 set_page_fscache(page); 391 } 392 393 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async) 394 { 395 struct inode *inode = priv; 396 397 if (IS_ERR_VALUE(error) && error != -ENOBUFS) 398 ceph_fscache_invalidate(inode, false); 399 } 400 401 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) 402 { 403 struct ceph_inode_info *ci = ceph_inode(inode); 404 struct fscache_cookie *cookie = ceph_fscache_cookie(ci); 405 406 fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode), 407 ceph_fscache_write_terminated, inode, caching); 408 } 409 #else 410 static inline void ceph_set_page_fscache(struct page *page) 411 { 412 } 413 414 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching) 415 { 416 } 417 #endif /* CONFIG_CEPH_FSCACHE */ 418 419 struct ceph_writeback_ctl 420 { 421 loff_t i_size; 422 u64 truncate_size; 423 u32 truncate_seq; 424 bool size_stable; 425 bool head_snapc; 426 }; 427 428 /* 429 * Get ref for the oldest snapc for an inode with dirty data... that is, the 430 * only snap context we are allowed to write back. 431 */ 432 static struct ceph_snap_context * 433 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl, 434 struct ceph_snap_context *page_snapc) 435 { 436 struct ceph_inode_info *ci = ceph_inode(inode); 437 struct ceph_snap_context *snapc = NULL; 438 struct ceph_cap_snap *capsnap = NULL; 439 440 spin_lock(&ci->i_ceph_lock); 441 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 442 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, 443 capsnap->context, capsnap->dirty_pages); 444 if (!capsnap->dirty_pages) 445 continue; 446 447 /* get i_size, truncate_{seq,size} for page_snapc? */ 448 if (snapc && capsnap->context != page_snapc) 449 continue; 450 451 if (ctl) { 452 if (capsnap->writing) { 453 ctl->i_size = i_size_read(inode); 454 ctl->size_stable = false; 455 } else { 456 ctl->i_size = capsnap->size; 457 ctl->size_stable = true; 458 } 459 ctl->truncate_size = capsnap->truncate_size; 460 ctl->truncate_seq = capsnap->truncate_seq; 461 ctl->head_snapc = false; 462 } 463 464 if (snapc) 465 break; 466 467 snapc = ceph_get_snap_context(capsnap->context); 468 if (!page_snapc || 469 page_snapc == snapc || 470 page_snapc->seq > snapc->seq) 471 break; 472 } 473 if (!snapc && ci->i_wrbuffer_ref_head) { 474 snapc = ceph_get_snap_context(ci->i_head_snapc); 475 dout(" head snapc %p has %d dirty pages\n", 476 snapc, ci->i_wrbuffer_ref_head); 477 if (ctl) { 478 ctl->i_size = i_size_read(inode); 479 ctl->truncate_size = ci->i_truncate_size; 480 ctl->truncate_seq = ci->i_truncate_seq; 481 ctl->size_stable = false; 482 ctl->head_snapc = true; 483 } 484 } 485 spin_unlock(&ci->i_ceph_lock); 486 return snapc; 487 } 488 489 static u64 get_writepages_data_length(struct inode *inode, 490 struct page *page, u64 start) 491 { 492 struct ceph_inode_info *ci = ceph_inode(inode); 493 struct ceph_snap_context *snapc = page_snap_context(page); 494 struct ceph_cap_snap *capsnap = NULL; 495 u64 end = i_size_read(inode); 496 497 if (snapc != ci->i_head_snapc) { 498 bool found = false; 499 spin_lock(&ci->i_ceph_lock); 500 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { 501 if (capsnap->context == snapc) { 502 if (!capsnap->writing) 503 end = capsnap->size; 504 found = true; 505 break; 506 } 507 } 508 spin_unlock(&ci->i_ceph_lock); 509 WARN_ON(!found); 510 } 511 if (end > page_offset(page) + thp_size(page)) 512 end = page_offset(page) + thp_size(page); 513 return end > start ? end - start : 0; 514 } 515 516 /* 517 * Write a single page, but leave the page locked. 518 * 519 * If we get a write error, mark the mapping for error, but still adjust the 520 * dirty page accounting (i.e., page is no longer dirty). 521 */ 522 static int writepage_nounlock(struct page *page, struct writeback_control *wbc) 523 { 524 struct inode *inode = page->mapping->host; 525 struct ceph_inode_info *ci = ceph_inode(inode); 526 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 527 struct ceph_snap_context *snapc, *oldest; 528 loff_t page_off = page_offset(page); 529 int err; 530 loff_t len = thp_size(page); 531 struct ceph_writeback_ctl ceph_wbc; 532 struct ceph_osd_client *osdc = &fsc->client->osdc; 533 struct ceph_osd_request *req; 534 bool caching = ceph_is_cache_enabled(inode); 535 536 dout("writepage %p idx %lu\n", page, page->index); 537 538 /* verify this is a writeable snap context */ 539 snapc = page_snap_context(page); 540 if (!snapc) { 541 dout("writepage %p page %p not dirty?\n", inode, page); 542 return 0; 543 } 544 oldest = get_oldest_context(inode, &ceph_wbc, snapc); 545 if (snapc->seq > oldest->seq) { 546 dout("writepage %p page %p snapc %p not writeable - noop\n", 547 inode, page, snapc); 548 /* we should only noop if called by kswapd */ 549 WARN_ON(!(current->flags & PF_MEMALLOC)); 550 ceph_put_snap_context(oldest); 551 redirty_page_for_writepage(wbc, page); 552 return 0; 553 } 554 ceph_put_snap_context(oldest); 555 556 /* is this a partial page at end of file? */ 557 if (page_off >= ceph_wbc.i_size) { 558 dout("%p page eof %llu\n", page, ceph_wbc.i_size); 559 page->mapping->a_ops->invalidatepage(page, 0, thp_size(page)); 560 return 0; 561 } 562 563 if (ceph_wbc.i_size < page_off + len) 564 len = ceph_wbc.i_size - page_off; 565 566 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n", 567 inode, page, page->index, page_off, len, snapc, snapc->seq); 568 569 if (atomic_long_inc_return(&fsc->writeback_count) > 570 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) 571 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 572 573 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1, 574 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc, 575 ceph_wbc.truncate_seq, ceph_wbc.truncate_size, 576 true); 577 if (IS_ERR(req)) 578 return PTR_ERR(req); 579 580 set_page_writeback(page); 581 if (caching) 582 ceph_set_page_fscache(page); 583 ceph_fscache_write_to_cache(inode, page_off, len, caching); 584 585 /* it may be a short write due to an object boundary */ 586 WARN_ON_ONCE(len > thp_size(page)); 587 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false); 588 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len); 589 590 req->r_mtime = inode->i_mtime; 591 err = ceph_osdc_start_request(osdc, req, true); 592 if (!err) 593 err = ceph_osdc_wait_request(osdc, req); 594 595 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 596 req->r_end_latency, len, err); 597 598 ceph_osdc_put_request(req); 599 if (err == 0) 600 err = len; 601 602 if (err < 0) { 603 struct writeback_control tmp_wbc; 604 if (!wbc) 605 wbc = &tmp_wbc; 606 if (err == -ERESTARTSYS) { 607 /* killed by SIGKILL */ 608 dout("writepage interrupted page %p\n", page); 609 redirty_page_for_writepage(wbc, page); 610 end_page_writeback(page); 611 return err; 612 } 613 if (err == -EBLOCKLISTED) 614 fsc->blocklisted = true; 615 dout("writepage setting page/mapping error %d %p\n", 616 err, page); 617 mapping_set_error(&inode->i_data, err); 618 wbc->pages_skipped++; 619 } else { 620 dout("writepage cleaned page %p\n", page); 621 err = 0; /* vfs expects us to return 0 */ 622 } 623 oldest = detach_page_private(page); 624 WARN_ON_ONCE(oldest != snapc); 625 end_page_writeback(page); 626 ceph_put_wrbuffer_cap_refs(ci, 1, snapc); 627 ceph_put_snap_context(snapc); /* page's reference */ 628 629 if (atomic_long_dec_return(&fsc->writeback_count) < 630 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) 631 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 632 633 return err; 634 } 635 636 static int ceph_writepage(struct page *page, struct writeback_control *wbc) 637 { 638 int err; 639 struct inode *inode = page->mapping->host; 640 BUG_ON(!inode); 641 ihold(inode); 642 643 wait_on_page_fscache(page); 644 645 err = writepage_nounlock(page, wbc); 646 if (err == -ERESTARTSYS) { 647 /* direct memory reclaimer was killed by SIGKILL. return 0 648 * to prevent caller from setting mapping/page error */ 649 err = 0; 650 } 651 unlock_page(page); 652 iput(inode); 653 return err; 654 } 655 656 /* 657 * async writeback completion handler. 658 * 659 * If we get an error, set the mapping error bit, but not the individual 660 * page error bits. 661 */ 662 static void writepages_finish(struct ceph_osd_request *req) 663 { 664 struct inode *inode = req->r_inode; 665 struct ceph_inode_info *ci = ceph_inode(inode); 666 struct ceph_osd_data *osd_data; 667 struct page *page; 668 int num_pages, total_pages = 0; 669 int i, j; 670 int rc = req->r_result; 671 struct ceph_snap_context *snapc = req->r_snapc; 672 struct address_space *mapping = inode->i_mapping; 673 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 674 unsigned int len = 0; 675 bool remove_page; 676 677 dout("writepages_finish %p rc %d\n", inode, rc); 678 if (rc < 0) { 679 mapping_set_error(mapping, rc); 680 ceph_set_error_write(ci); 681 if (rc == -EBLOCKLISTED) 682 fsc->blocklisted = true; 683 } else { 684 ceph_clear_error_write(ci); 685 } 686 687 /* 688 * We lost the cache cap, need to truncate the page before 689 * it is unlocked, otherwise we'd truncate it later in the 690 * page truncation thread, possibly losing some data that 691 * raced its way in 692 */ 693 remove_page = !(ceph_caps_issued(ci) & 694 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); 695 696 /* clean all pages */ 697 for (i = 0; i < req->r_num_ops; i++) { 698 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) 699 break; 700 701 osd_data = osd_req_op_extent_osd_data(req, i); 702 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); 703 len += osd_data->length; 704 num_pages = calc_pages_for((u64)osd_data->alignment, 705 (u64)osd_data->length); 706 total_pages += num_pages; 707 for (j = 0; j < num_pages; j++) { 708 page = osd_data->pages[j]; 709 BUG_ON(!page); 710 WARN_ON(!PageUptodate(page)); 711 712 if (atomic_long_dec_return(&fsc->writeback_count) < 713 CONGESTION_OFF_THRESH( 714 fsc->mount_options->congestion_kb)) 715 clear_bdi_congested(inode_to_bdi(inode), 716 BLK_RW_ASYNC); 717 718 ceph_put_snap_context(detach_page_private(page)); 719 end_page_writeback(page); 720 dout("unlocking %p\n", page); 721 722 if (remove_page) 723 generic_error_remove_page(inode->i_mapping, 724 page); 725 726 unlock_page(page); 727 } 728 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", 729 inode, osd_data->length, rc >= 0 ? num_pages : 0); 730 731 release_pages(osd_data->pages, num_pages); 732 } 733 734 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 735 req->r_end_latency, len, rc); 736 737 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); 738 739 osd_data = osd_req_op_extent_osd_data(req, 0); 740 if (osd_data->pages_from_pool) 741 mempool_free(osd_data->pages, ceph_wb_pagevec_pool); 742 else 743 kfree(osd_data->pages); 744 ceph_osdc_put_request(req); 745 } 746 747 /* 748 * initiate async writeback 749 */ 750 static int ceph_writepages_start(struct address_space *mapping, 751 struct writeback_control *wbc) 752 { 753 struct inode *inode = mapping->host; 754 struct ceph_inode_info *ci = ceph_inode(inode); 755 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 756 struct ceph_vino vino = ceph_vino(inode); 757 pgoff_t index, start_index, end = -1; 758 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; 759 struct pagevec pvec; 760 int rc = 0; 761 unsigned int wsize = i_blocksize(inode); 762 struct ceph_osd_request *req = NULL; 763 struct ceph_writeback_ctl ceph_wbc; 764 bool should_loop, range_whole = false; 765 bool done = false; 766 bool caching = ceph_is_cache_enabled(inode); 767 768 dout("writepages_start %p (mode=%s)\n", inode, 769 wbc->sync_mode == WB_SYNC_NONE ? "NONE" : 770 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); 771 772 if (ceph_inode_is_shutdown(inode)) { 773 if (ci->i_wrbuffer_ref > 0) { 774 pr_warn_ratelimited( 775 "writepage_start %p %lld forced umount\n", 776 inode, ceph_ino(inode)); 777 } 778 mapping_set_error(mapping, -EIO); 779 return -EIO; /* we're in a forced umount, don't write! */ 780 } 781 if (fsc->mount_options->wsize < wsize) 782 wsize = fsc->mount_options->wsize; 783 784 pagevec_init(&pvec); 785 786 start_index = wbc->range_cyclic ? mapping->writeback_index : 0; 787 index = start_index; 788 789 retry: 790 /* find oldest snap context with dirty data */ 791 snapc = get_oldest_context(inode, &ceph_wbc, NULL); 792 if (!snapc) { 793 /* hmm, why does writepages get called when there 794 is no dirty data? */ 795 dout(" no snap context with dirty data?\n"); 796 goto out; 797 } 798 dout(" oldest snapc is %p seq %lld (%d snaps)\n", 799 snapc, snapc->seq, snapc->num_snaps); 800 801 should_loop = false; 802 if (ceph_wbc.head_snapc && snapc != last_snapc) { 803 /* where to start/end? */ 804 if (wbc->range_cyclic) { 805 index = start_index; 806 end = -1; 807 if (index > 0) 808 should_loop = true; 809 dout(" cyclic, start at %lu\n", index); 810 } else { 811 index = wbc->range_start >> PAGE_SHIFT; 812 end = wbc->range_end >> PAGE_SHIFT; 813 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 814 range_whole = true; 815 dout(" not cyclic, %lu to %lu\n", index, end); 816 } 817 } else if (!ceph_wbc.head_snapc) { 818 /* Do not respect wbc->range_{start,end}. Dirty pages 819 * in that range can be associated with newer snapc. 820 * They are not writeable until we write all dirty pages 821 * associated with 'snapc' get written */ 822 if (index > 0) 823 should_loop = true; 824 dout(" non-head snapc, range whole\n"); 825 } 826 827 ceph_put_snap_context(last_snapc); 828 last_snapc = snapc; 829 830 while (!done && index <= end) { 831 int num_ops = 0, op_idx; 832 unsigned i, pvec_pages, max_pages, locked_pages = 0; 833 struct page **pages = NULL, **data_pages; 834 struct page *page; 835 pgoff_t strip_unit_end = 0; 836 u64 offset = 0, len = 0; 837 bool from_pool = false; 838 839 max_pages = wsize >> PAGE_SHIFT; 840 841 get_more_pages: 842 pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, 843 end, PAGECACHE_TAG_DIRTY); 844 dout("pagevec_lookup_range_tag got %d\n", pvec_pages); 845 if (!pvec_pages && !locked_pages) 846 break; 847 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { 848 page = pvec.pages[i]; 849 dout("? %p idx %lu\n", page, page->index); 850 if (locked_pages == 0) 851 lock_page(page); /* first page */ 852 else if (!trylock_page(page)) 853 break; 854 855 /* only dirty pages, or our accounting breaks */ 856 if (unlikely(!PageDirty(page)) || 857 unlikely(page->mapping != mapping)) { 858 dout("!dirty or !mapping %p\n", page); 859 unlock_page(page); 860 continue; 861 } 862 /* only if matching snap context */ 863 pgsnapc = page_snap_context(page); 864 if (pgsnapc != snapc) { 865 dout("page snapc %p %lld != oldest %p %lld\n", 866 pgsnapc, pgsnapc->seq, snapc, snapc->seq); 867 if (!should_loop && 868 !ceph_wbc.head_snapc && 869 wbc->sync_mode != WB_SYNC_NONE) 870 should_loop = true; 871 unlock_page(page); 872 continue; 873 } 874 if (page_offset(page) >= ceph_wbc.i_size) { 875 dout("%p page eof %llu\n", 876 page, ceph_wbc.i_size); 877 if ((ceph_wbc.size_stable || 878 page_offset(page) >= i_size_read(inode)) && 879 clear_page_dirty_for_io(page)) 880 mapping->a_ops->invalidatepage(page, 881 0, thp_size(page)); 882 unlock_page(page); 883 continue; 884 } 885 if (strip_unit_end && (page->index > strip_unit_end)) { 886 dout("end of strip unit %p\n", page); 887 unlock_page(page); 888 break; 889 } 890 if (PageWriteback(page) || PageFsCache(page)) { 891 if (wbc->sync_mode == WB_SYNC_NONE) { 892 dout("%p under writeback\n", page); 893 unlock_page(page); 894 continue; 895 } 896 dout("waiting on writeback %p\n", page); 897 wait_on_page_writeback(page); 898 wait_on_page_fscache(page); 899 } 900 901 if (!clear_page_dirty_for_io(page)) { 902 dout("%p !clear_page_dirty_for_io\n", page); 903 unlock_page(page); 904 continue; 905 } 906 907 /* 908 * We have something to write. If this is 909 * the first locked page this time through, 910 * calculate max possinle write size and 911 * allocate a page array 912 */ 913 if (locked_pages == 0) { 914 u64 objnum; 915 u64 objoff; 916 u32 xlen; 917 918 /* prepare async write request */ 919 offset = (u64)page_offset(page); 920 ceph_calc_file_object_mapping(&ci->i_layout, 921 offset, wsize, 922 &objnum, &objoff, 923 &xlen); 924 len = xlen; 925 926 num_ops = 1; 927 strip_unit_end = page->index + 928 ((len - 1) >> PAGE_SHIFT); 929 930 BUG_ON(pages); 931 max_pages = calc_pages_for(0, (u64)len); 932 pages = kmalloc_array(max_pages, 933 sizeof(*pages), 934 GFP_NOFS); 935 if (!pages) { 936 from_pool = true; 937 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); 938 BUG_ON(!pages); 939 } 940 941 len = 0; 942 } else if (page->index != 943 (offset + len) >> PAGE_SHIFT) { 944 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS : 945 CEPH_OSD_MAX_OPS)) { 946 redirty_page_for_writepage(wbc, page); 947 unlock_page(page); 948 break; 949 } 950 951 num_ops++; 952 offset = (u64)page_offset(page); 953 len = 0; 954 } 955 956 /* note position of first page in pvec */ 957 dout("%p will write page %p idx %lu\n", 958 inode, page, page->index); 959 960 if (atomic_long_inc_return(&fsc->writeback_count) > 961 CONGESTION_ON_THRESH( 962 fsc->mount_options->congestion_kb)) { 963 set_bdi_congested(inode_to_bdi(inode), 964 BLK_RW_ASYNC); 965 } 966 967 968 pages[locked_pages++] = page; 969 pvec.pages[i] = NULL; 970 971 len += thp_size(page); 972 } 973 974 /* did we get anything? */ 975 if (!locked_pages) 976 goto release_pvec_pages; 977 if (i) { 978 unsigned j, n = 0; 979 /* shift unused page to beginning of pvec */ 980 for (j = 0; j < pvec_pages; j++) { 981 if (!pvec.pages[j]) 982 continue; 983 if (n < j) 984 pvec.pages[n] = pvec.pages[j]; 985 n++; 986 } 987 pvec.nr = n; 988 989 if (pvec_pages && i == pvec_pages && 990 locked_pages < max_pages) { 991 dout("reached end pvec, trying for more\n"); 992 pagevec_release(&pvec); 993 goto get_more_pages; 994 } 995 } 996 997 new_request: 998 offset = page_offset(pages[0]); 999 len = wsize; 1000 1001 req = ceph_osdc_new_request(&fsc->client->osdc, 1002 &ci->i_layout, vino, 1003 offset, &len, 0, num_ops, 1004 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1005 snapc, ceph_wbc.truncate_seq, 1006 ceph_wbc.truncate_size, false); 1007 if (IS_ERR(req)) { 1008 req = ceph_osdc_new_request(&fsc->client->osdc, 1009 &ci->i_layout, vino, 1010 offset, &len, 0, 1011 min(num_ops, 1012 CEPH_OSD_SLAB_OPS), 1013 CEPH_OSD_OP_WRITE, 1014 CEPH_OSD_FLAG_WRITE, 1015 snapc, ceph_wbc.truncate_seq, 1016 ceph_wbc.truncate_size, true); 1017 BUG_ON(IS_ERR(req)); 1018 } 1019 BUG_ON(len < page_offset(pages[locked_pages - 1]) + 1020 thp_size(page) - offset); 1021 1022 req->r_callback = writepages_finish; 1023 req->r_inode = inode; 1024 1025 /* Format the osd request message and submit the write */ 1026 len = 0; 1027 data_pages = pages; 1028 op_idx = 0; 1029 for (i = 0; i < locked_pages; i++) { 1030 u64 cur_offset = page_offset(pages[i]); 1031 /* 1032 * Discontinuity in page range? Ceph can handle that by just passing 1033 * multiple extents in the write op. 1034 */ 1035 if (offset + len != cur_offset) { 1036 /* If it's full, stop here */ 1037 if (op_idx + 1 == req->r_num_ops) 1038 break; 1039 1040 /* Kick off an fscache write with what we have so far. */ 1041 ceph_fscache_write_to_cache(inode, offset, len, caching); 1042 1043 /* Start a new extent */ 1044 osd_req_op_extent_dup_last(req, op_idx, 1045 cur_offset - offset); 1046 dout("writepages got pages at %llu~%llu\n", 1047 offset, len); 1048 osd_req_op_extent_osd_data_pages(req, op_idx, 1049 data_pages, len, 0, 1050 from_pool, false); 1051 osd_req_op_extent_update(req, op_idx, len); 1052 1053 len = 0; 1054 offset = cur_offset; 1055 data_pages = pages + i; 1056 op_idx++; 1057 } 1058 1059 set_page_writeback(pages[i]); 1060 if (caching) 1061 ceph_set_page_fscache(pages[i]); 1062 len += thp_size(page); 1063 } 1064 ceph_fscache_write_to_cache(inode, offset, len, caching); 1065 1066 if (ceph_wbc.size_stable) { 1067 len = min(len, ceph_wbc.i_size - offset); 1068 } else if (i == locked_pages) { 1069 /* writepages_finish() clears writeback pages 1070 * according to the data length, so make sure 1071 * data length covers all locked pages */ 1072 u64 min_len = len + 1 - thp_size(page); 1073 len = get_writepages_data_length(inode, pages[i - 1], 1074 offset); 1075 len = max(len, min_len); 1076 } 1077 dout("writepages got pages at %llu~%llu\n", offset, len); 1078 1079 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len, 1080 0, from_pool, false); 1081 osd_req_op_extent_update(req, op_idx, len); 1082 1083 BUG_ON(op_idx + 1 != req->r_num_ops); 1084 1085 from_pool = false; 1086 if (i < locked_pages) { 1087 BUG_ON(num_ops <= req->r_num_ops); 1088 num_ops -= req->r_num_ops; 1089 locked_pages -= i; 1090 1091 /* allocate new pages array for next request */ 1092 data_pages = pages; 1093 pages = kmalloc_array(locked_pages, sizeof(*pages), 1094 GFP_NOFS); 1095 if (!pages) { 1096 from_pool = true; 1097 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS); 1098 BUG_ON(!pages); 1099 } 1100 memcpy(pages, data_pages + i, 1101 locked_pages * sizeof(*pages)); 1102 memset(data_pages + i, 0, 1103 locked_pages * sizeof(*pages)); 1104 } else { 1105 BUG_ON(num_ops != req->r_num_ops); 1106 index = pages[i - 1]->index + 1; 1107 /* request message now owns the pages array */ 1108 pages = NULL; 1109 } 1110 1111 req->r_mtime = inode->i_mtime; 1112 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); 1113 BUG_ON(rc); 1114 req = NULL; 1115 1116 wbc->nr_to_write -= i; 1117 if (pages) 1118 goto new_request; 1119 1120 /* 1121 * We stop writing back only if we are not doing 1122 * integrity sync. In case of integrity sync we have to 1123 * keep going until we have written all the pages 1124 * we tagged for writeback prior to entering this loop. 1125 */ 1126 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) 1127 done = true; 1128 1129 release_pvec_pages: 1130 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, 1131 pvec.nr ? pvec.pages[0] : NULL); 1132 pagevec_release(&pvec); 1133 } 1134 1135 if (should_loop && !done) { 1136 /* more to do; loop back to beginning of file */ 1137 dout("writepages looping back to beginning of file\n"); 1138 end = start_index - 1; /* OK even when start_index == 0 */ 1139 1140 /* to write dirty pages associated with next snapc, 1141 * we need to wait until current writes complete */ 1142 if (wbc->sync_mode != WB_SYNC_NONE && 1143 start_index == 0 && /* all dirty pages were checked */ 1144 !ceph_wbc.head_snapc) { 1145 struct page *page; 1146 unsigned i, nr; 1147 index = 0; 1148 while ((index <= end) && 1149 (nr = pagevec_lookup_tag(&pvec, mapping, &index, 1150 PAGECACHE_TAG_WRITEBACK))) { 1151 for (i = 0; i < nr; i++) { 1152 page = pvec.pages[i]; 1153 if (page_snap_context(page) != snapc) 1154 continue; 1155 wait_on_page_writeback(page); 1156 } 1157 pagevec_release(&pvec); 1158 cond_resched(); 1159 } 1160 } 1161 1162 start_index = 0; 1163 index = 0; 1164 goto retry; 1165 } 1166 1167 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 1168 mapping->writeback_index = index; 1169 1170 out: 1171 ceph_osdc_put_request(req); 1172 ceph_put_snap_context(last_snapc); 1173 dout("writepages dend - startone, rc = %d\n", rc); 1174 return rc; 1175 } 1176 1177 1178 1179 /* 1180 * See if a given @snapc is either writeable, or already written. 1181 */ 1182 static int context_is_writeable_or_written(struct inode *inode, 1183 struct ceph_snap_context *snapc) 1184 { 1185 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL); 1186 int ret = !oldest || snapc->seq <= oldest->seq; 1187 1188 ceph_put_snap_context(oldest); 1189 return ret; 1190 } 1191 1192 /** 1193 * ceph_find_incompatible - find an incompatible context and return it 1194 * @page: page being dirtied 1195 * 1196 * We are only allowed to write into/dirty a page if the page is 1197 * clean, or already dirty within the same snap context. Returns a 1198 * conflicting context if there is one, NULL if there isn't, or a 1199 * negative error code on other errors. 1200 * 1201 * Must be called with page lock held. 1202 */ 1203 static struct ceph_snap_context * 1204 ceph_find_incompatible(struct page *page) 1205 { 1206 struct inode *inode = page->mapping->host; 1207 struct ceph_inode_info *ci = ceph_inode(inode); 1208 1209 if (ceph_inode_is_shutdown(inode)) { 1210 dout(" page %p %llx:%llx is shutdown\n", page, 1211 ceph_vinop(inode)); 1212 return ERR_PTR(-ESTALE); 1213 } 1214 1215 for (;;) { 1216 struct ceph_snap_context *snapc, *oldest; 1217 1218 wait_on_page_writeback(page); 1219 1220 snapc = page_snap_context(page); 1221 if (!snapc || snapc == ci->i_head_snapc) 1222 break; 1223 1224 /* 1225 * this page is already dirty in another (older) snap 1226 * context! is it writeable now? 1227 */ 1228 oldest = get_oldest_context(inode, NULL, NULL); 1229 if (snapc->seq > oldest->seq) { 1230 /* not writeable -- return it for the caller to deal with */ 1231 ceph_put_snap_context(oldest); 1232 dout(" page %p snapc %p not current or oldest\n", page, snapc); 1233 return ceph_get_snap_context(snapc); 1234 } 1235 ceph_put_snap_context(oldest); 1236 1237 /* yay, writeable, do it now (without dropping page lock) */ 1238 dout(" page %p snapc %p not current, but oldest\n", page, snapc); 1239 if (clear_page_dirty_for_io(page)) { 1240 int r = writepage_nounlock(page, NULL); 1241 if (r < 0) 1242 return ERR_PTR(r); 1243 } 1244 } 1245 return NULL; 1246 } 1247 1248 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len, 1249 struct folio *folio, void **_fsdata) 1250 { 1251 struct inode *inode = file_inode(file); 1252 struct ceph_inode_info *ci = ceph_inode(inode); 1253 struct ceph_snap_context *snapc; 1254 1255 snapc = ceph_find_incompatible(folio_page(folio, 0)); 1256 if (snapc) { 1257 int r; 1258 1259 folio_unlock(folio); 1260 folio_put(folio); 1261 if (IS_ERR(snapc)) 1262 return PTR_ERR(snapc); 1263 1264 ceph_queue_writeback(inode); 1265 r = wait_event_killable(ci->i_cap_wq, 1266 context_is_writeable_or_written(inode, snapc)); 1267 ceph_put_snap_context(snapc); 1268 return r == 0 ? -EAGAIN : r; 1269 } 1270 return 0; 1271 } 1272 1273 /* 1274 * We are only allowed to write into/dirty the page if the page is 1275 * clean, or already dirty within the same snap context. 1276 */ 1277 static int ceph_write_begin(struct file *file, struct address_space *mapping, 1278 loff_t pos, unsigned len, unsigned aop_flags, 1279 struct page **pagep, void **fsdata) 1280 { 1281 struct inode *inode = file_inode(file); 1282 struct ceph_inode_info *ci = ceph_inode(inode); 1283 struct folio *folio = NULL; 1284 pgoff_t index = pos >> PAGE_SHIFT; 1285 int r; 1286 1287 /* 1288 * Uninlining should have already been done and everything updated, EXCEPT 1289 * for inline_version sent to the MDS. 1290 */ 1291 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1292 unsigned int fgp_flags = FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE; 1293 if (aop_flags & AOP_FLAG_NOFS) 1294 fgp_flags |= FGP_NOFS; 1295 folio = __filemap_get_folio(mapping, index, fgp_flags, 1296 mapping_gfp_mask(mapping)); 1297 if (!folio) 1298 return -ENOMEM; 1299 1300 /* 1301 * The inline_version on a new inode is set to 1. If that's the 1302 * case, then the folio is brand new and isn't yet Uptodate. 1303 */ 1304 r = 0; 1305 if (index == 0 && ci->i_inline_version != 1) { 1306 if (!folio_test_uptodate(folio)) { 1307 WARN_ONCE(1, "ceph: write_begin called on still-inlined inode (inline_version %llu)!\n", 1308 ci->i_inline_version); 1309 r = -EINVAL; 1310 } 1311 goto out; 1312 } 1313 zero_user_segment(&folio->page, 0, folio_size(folio)); 1314 folio_mark_uptodate(folio); 1315 goto out; 1316 } 1317 1318 r = netfs_write_begin(file, inode->i_mapping, pos, len, 0, &folio, NULL, 1319 &ceph_netfs_read_ops, NULL); 1320 out: 1321 if (r == 0) 1322 folio_wait_fscache(folio); 1323 if (r < 0) { 1324 if (folio) 1325 folio_put(folio); 1326 } else { 1327 WARN_ON_ONCE(!folio_test_locked(folio)); 1328 *pagep = &folio->page; 1329 } 1330 return r; 1331 } 1332 1333 /* 1334 * we don't do anything in here that simple_write_end doesn't do 1335 * except adjust dirty page accounting 1336 */ 1337 static int ceph_write_end(struct file *file, struct address_space *mapping, 1338 loff_t pos, unsigned len, unsigned copied, 1339 struct page *subpage, void *fsdata) 1340 { 1341 struct folio *folio = page_folio(subpage); 1342 struct inode *inode = file_inode(file); 1343 bool check_cap = false; 1344 1345 dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file, 1346 inode, folio, (int)pos, (int)copied, (int)len); 1347 1348 if (!folio_test_uptodate(folio)) { 1349 /* just return that nothing was copied on a short copy */ 1350 if (copied < len) { 1351 copied = 0; 1352 goto out; 1353 } 1354 folio_mark_uptodate(folio); 1355 } 1356 1357 /* did file size increase? */ 1358 if (pos+copied > i_size_read(inode)) 1359 check_cap = ceph_inode_set_size(inode, pos+copied); 1360 1361 folio_mark_dirty(folio); 1362 1363 out: 1364 folio_unlock(folio); 1365 folio_put(folio); 1366 1367 if (check_cap) 1368 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); 1369 1370 return copied; 1371 } 1372 1373 const struct address_space_operations ceph_aops = { 1374 .readpage = ceph_readpage, 1375 .readahead = ceph_readahead, 1376 .writepage = ceph_writepage, 1377 .writepages = ceph_writepages_start, 1378 .write_begin = ceph_write_begin, 1379 .write_end = ceph_write_end, 1380 .set_page_dirty = ceph_set_page_dirty, 1381 .invalidatepage = ceph_invalidatepage, 1382 .releasepage = ceph_releasepage, 1383 .direct_IO = noop_direct_IO, 1384 }; 1385 1386 static void ceph_block_sigs(sigset_t *oldset) 1387 { 1388 sigset_t mask; 1389 siginitsetinv(&mask, sigmask(SIGKILL)); 1390 sigprocmask(SIG_BLOCK, &mask, oldset); 1391 } 1392 1393 static void ceph_restore_sigs(sigset_t *oldset) 1394 { 1395 sigprocmask(SIG_SETMASK, oldset, NULL); 1396 } 1397 1398 /* 1399 * vm ops 1400 */ 1401 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf) 1402 { 1403 struct vm_area_struct *vma = vmf->vma; 1404 struct inode *inode = file_inode(vma->vm_file); 1405 struct ceph_inode_info *ci = ceph_inode(inode); 1406 struct ceph_file_info *fi = vma->vm_file->private_data; 1407 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT; 1408 int want, got, err; 1409 sigset_t oldset; 1410 vm_fault_t ret = VM_FAULT_SIGBUS; 1411 1412 if (ceph_inode_is_shutdown(inode)) 1413 return ret; 1414 1415 ceph_block_sigs(&oldset); 1416 1417 dout("filemap_fault %p %llx.%llx %llu trying to get caps\n", 1418 inode, ceph_vinop(inode), off); 1419 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1420 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1421 else 1422 want = CEPH_CAP_FILE_CACHE; 1423 1424 got = 0; 1425 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got); 1426 if (err < 0) 1427 goto out_restore; 1428 1429 dout("filemap_fault %p %llu got cap refs on %s\n", 1430 inode, off, ceph_cap_string(got)); 1431 1432 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || 1433 ci->i_inline_version == CEPH_INLINE_NONE) { 1434 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); 1435 ceph_add_rw_context(fi, &rw_ctx); 1436 ret = filemap_fault(vmf); 1437 ceph_del_rw_context(fi, &rw_ctx); 1438 dout("filemap_fault %p %llu drop cap refs %s ret %x\n", 1439 inode, off, ceph_cap_string(got), ret); 1440 } else 1441 err = -EAGAIN; 1442 1443 ceph_put_cap_refs(ci, got); 1444 1445 if (err != -EAGAIN) 1446 goto out_restore; 1447 1448 /* read inline data */ 1449 if (off >= PAGE_SIZE) { 1450 /* does not support inline data > PAGE_SIZE */ 1451 ret = VM_FAULT_SIGBUS; 1452 } else { 1453 struct address_space *mapping = inode->i_mapping; 1454 struct page *page; 1455 1456 filemap_invalidate_lock_shared(mapping); 1457 page = find_or_create_page(mapping, 0, 1458 mapping_gfp_constraint(mapping, ~__GFP_FS)); 1459 if (!page) { 1460 ret = VM_FAULT_OOM; 1461 goto out_inline; 1462 } 1463 err = __ceph_do_getattr(inode, page, 1464 CEPH_STAT_CAP_INLINE_DATA, true); 1465 if (err < 0 || off >= i_size_read(inode)) { 1466 unlock_page(page); 1467 put_page(page); 1468 ret = vmf_error(err); 1469 goto out_inline; 1470 } 1471 if (err < PAGE_SIZE) 1472 zero_user_segment(page, err, PAGE_SIZE); 1473 else 1474 flush_dcache_page(page); 1475 SetPageUptodate(page); 1476 vmf->page = page; 1477 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED; 1478 out_inline: 1479 filemap_invalidate_unlock_shared(mapping); 1480 dout("filemap_fault %p %llu read inline data ret %x\n", 1481 inode, off, ret); 1482 } 1483 out_restore: 1484 ceph_restore_sigs(&oldset); 1485 if (err < 0) 1486 ret = vmf_error(err); 1487 1488 return ret; 1489 } 1490 1491 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf) 1492 { 1493 struct vm_area_struct *vma = vmf->vma; 1494 struct inode *inode = file_inode(vma->vm_file); 1495 struct ceph_inode_info *ci = ceph_inode(inode); 1496 struct ceph_file_info *fi = vma->vm_file->private_data; 1497 struct ceph_cap_flush *prealloc_cf; 1498 struct page *page = vmf->page; 1499 loff_t off = page_offset(page); 1500 loff_t size = i_size_read(inode); 1501 size_t len; 1502 int want, got, err; 1503 sigset_t oldset; 1504 vm_fault_t ret = VM_FAULT_SIGBUS; 1505 1506 if (ceph_inode_is_shutdown(inode)) 1507 return ret; 1508 1509 prealloc_cf = ceph_alloc_cap_flush(); 1510 if (!prealloc_cf) 1511 return VM_FAULT_OOM; 1512 1513 sb_start_pagefault(inode->i_sb); 1514 ceph_block_sigs(&oldset); 1515 1516 if (ci->i_inline_version != CEPH_INLINE_NONE) { 1517 struct page *locked_page = NULL; 1518 if (off == 0) { 1519 lock_page(page); 1520 locked_page = page; 1521 } 1522 err = ceph_uninline_data(vma->vm_file, locked_page); 1523 if (locked_page) 1524 unlock_page(locked_page); 1525 if (err < 0) 1526 goto out_free; 1527 } 1528 1529 if (off + thp_size(page) <= size) 1530 len = thp_size(page); 1531 else 1532 len = offset_in_thp(page, size); 1533 1534 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n", 1535 inode, ceph_vinop(inode), off, len, size); 1536 if (fi->fmode & CEPH_FILE_MODE_LAZY) 1537 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; 1538 else 1539 want = CEPH_CAP_FILE_BUFFER; 1540 1541 got = 0; 1542 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got); 1543 if (err < 0) 1544 goto out_free; 1545 1546 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n", 1547 inode, off, len, ceph_cap_string(got)); 1548 1549 /* Update time before taking page lock */ 1550 file_update_time(vma->vm_file); 1551 inode_inc_iversion_raw(inode); 1552 1553 do { 1554 struct ceph_snap_context *snapc; 1555 1556 lock_page(page); 1557 1558 if (page_mkwrite_check_truncate(page, inode) < 0) { 1559 unlock_page(page); 1560 ret = VM_FAULT_NOPAGE; 1561 break; 1562 } 1563 1564 snapc = ceph_find_incompatible(page); 1565 if (!snapc) { 1566 /* success. we'll keep the page locked. */ 1567 set_page_dirty(page); 1568 ret = VM_FAULT_LOCKED; 1569 break; 1570 } 1571 1572 unlock_page(page); 1573 1574 if (IS_ERR(snapc)) { 1575 ret = VM_FAULT_SIGBUS; 1576 break; 1577 } 1578 1579 ceph_queue_writeback(inode); 1580 err = wait_event_killable(ci->i_cap_wq, 1581 context_is_writeable_or_written(inode, snapc)); 1582 ceph_put_snap_context(snapc); 1583 } while (err == 0); 1584 1585 if (ret == VM_FAULT_LOCKED || 1586 ci->i_inline_version != CEPH_INLINE_NONE) { 1587 int dirty; 1588 spin_lock(&ci->i_ceph_lock); 1589 ci->i_inline_version = CEPH_INLINE_NONE; 1590 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, 1591 &prealloc_cf); 1592 spin_unlock(&ci->i_ceph_lock); 1593 if (dirty) 1594 __mark_inode_dirty(inode, dirty); 1595 } 1596 1597 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n", 1598 inode, off, len, ceph_cap_string(got), ret); 1599 ceph_put_cap_refs_async(ci, got); 1600 out_free: 1601 ceph_restore_sigs(&oldset); 1602 sb_end_pagefault(inode->i_sb); 1603 ceph_free_cap_flush(prealloc_cf); 1604 if (err < 0) 1605 ret = vmf_error(err); 1606 return ret; 1607 } 1608 1609 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page, 1610 char *data, size_t len) 1611 { 1612 struct address_space *mapping = inode->i_mapping; 1613 struct page *page; 1614 1615 if (locked_page) { 1616 page = locked_page; 1617 } else { 1618 if (i_size_read(inode) == 0) 1619 return; 1620 page = find_or_create_page(mapping, 0, 1621 mapping_gfp_constraint(mapping, 1622 ~__GFP_FS)); 1623 if (!page) 1624 return; 1625 if (PageUptodate(page)) { 1626 unlock_page(page); 1627 put_page(page); 1628 return; 1629 } 1630 } 1631 1632 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n", 1633 inode, ceph_vinop(inode), len, locked_page); 1634 1635 if (len > 0) { 1636 void *kaddr = kmap_atomic(page); 1637 memcpy(kaddr, data, len); 1638 kunmap_atomic(kaddr); 1639 } 1640 1641 if (page != locked_page) { 1642 if (len < PAGE_SIZE) 1643 zero_user_segment(page, len, PAGE_SIZE); 1644 else 1645 flush_dcache_page(page); 1646 1647 SetPageUptodate(page); 1648 unlock_page(page); 1649 put_page(page); 1650 } 1651 } 1652 1653 int ceph_uninline_data(struct file *filp, struct page *locked_page) 1654 { 1655 struct inode *inode = file_inode(filp); 1656 struct ceph_inode_info *ci = ceph_inode(inode); 1657 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1658 struct ceph_osd_request *req; 1659 struct page *page = NULL; 1660 u64 len, inline_version; 1661 int err = 0; 1662 bool from_pagecache = false; 1663 1664 spin_lock(&ci->i_ceph_lock); 1665 inline_version = ci->i_inline_version; 1666 spin_unlock(&ci->i_ceph_lock); 1667 1668 dout("uninline_data %p %llx.%llx inline_version %llu\n", 1669 inode, ceph_vinop(inode), inline_version); 1670 1671 if (inline_version == 1 || /* initial version, no data */ 1672 inline_version == CEPH_INLINE_NONE) 1673 goto out; 1674 1675 if (locked_page) { 1676 page = locked_page; 1677 WARN_ON(!PageUptodate(page)); 1678 } else if (ceph_caps_issued(ci) & 1679 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) { 1680 page = find_get_page(inode->i_mapping, 0); 1681 if (page) { 1682 if (PageUptodate(page)) { 1683 from_pagecache = true; 1684 lock_page(page); 1685 } else { 1686 put_page(page); 1687 page = NULL; 1688 } 1689 } 1690 } 1691 1692 if (page) { 1693 len = i_size_read(inode); 1694 if (len > PAGE_SIZE) 1695 len = PAGE_SIZE; 1696 } else { 1697 page = __page_cache_alloc(GFP_NOFS); 1698 if (!page) { 1699 err = -ENOMEM; 1700 goto out; 1701 } 1702 err = __ceph_do_getattr(inode, page, 1703 CEPH_STAT_CAP_INLINE_DATA, true); 1704 if (err < 0) { 1705 /* no inline data */ 1706 if (err == -ENODATA) 1707 err = 0; 1708 goto out; 1709 } 1710 len = err; 1711 } 1712 1713 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1714 ceph_vino(inode), 0, &len, 0, 1, 1715 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE, 1716 NULL, 0, 0, false); 1717 if (IS_ERR(req)) { 1718 err = PTR_ERR(req); 1719 goto out; 1720 } 1721 1722 req->r_mtime = inode->i_mtime; 1723 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1724 if (!err) 1725 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1726 ceph_osdc_put_request(req); 1727 if (err < 0) 1728 goto out; 1729 1730 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, 1731 ceph_vino(inode), 0, &len, 1, 3, 1732 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, 1733 NULL, ci->i_truncate_seq, 1734 ci->i_truncate_size, false); 1735 if (IS_ERR(req)) { 1736 err = PTR_ERR(req); 1737 goto out; 1738 } 1739 1740 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false); 1741 1742 { 1743 __le64 xattr_buf = cpu_to_le64(inline_version); 1744 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR, 1745 "inline_version", &xattr_buf, 1746 sizeof(xattr_buf), 1747 CEPH_OSD_CMPXATTR_OP_GT, 1748 CEPH_OSD_CMPXATTR_MODE_U64); 1749 if (err) 1750 goto out_put; 1751 } 1752 1753 { 1754 char xattr_buf[32]; 1755 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf), 1756 "%llu", inline_version); 1757 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR, 1758 "inline_version", 1759 xattr_buf, xattr_len, 0, 0); 1760 if (err) 1761 goto out_put; 1762 } 1763 1764 req->r_mtime = inode->i_mtime; 1765 err = ceph_osdc_start_request(&fsc->client->osdc, req, false); 1766 if (!err) 1767 err = ceph_osdc_wait_request(&fsc->client->osdc, req); 1768 1769 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency, 1770 req->r_end_latency, len, err); 1771 1772 out_put: 1773 ceph_osdc_put_request(req); 1774 if (err == -ECANCELED) 1775 err = 0; 1776 out: 1777 if (page && page != locked_page) { 1778 if (from_pagecache) { 1779 unlock_page(page); 1780 put_page(page); 1781 } else 1782 __free_pages(page, 0); 1783 } 1784 1785 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n", 1786 inode, ceph_vinop(inode), inline_version, err); 1787 return err; 1788 } 1789 1790 static const struct vm_operations_struct ceph_vmops = { 1791 .fault = ceph_filemap_fault, 1792 .page_mkwrite = ceph_page_mkwrite, 1793 }; 1794 1795 int ceph_mmap(struct file *file, struct vm_area_struct *vma) 1796 { 1797 struct address_space *mapping = file->f_mapping; 1798 1799 if (!mapping->a_ops->readpage) 1800 return -ENOEXEC; 1801 file_accessed(file); 1802 vma->vm_ops = &ceph_vmops; 1803 return 0; 1804 } 1805 1806 enum { 1807 POOL_READ = 1, 1808 POOL_WRITE = 2, 1809 }; 1810 1811 static int __ceph_pool_perm_get(struct ceph_inode_info *ci, 1812 s64 pool, struct ceph_string *pool_ns) 1813 { 1814 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); 1815 struct ceph_mds_client *mdsc = fsc->mdsc; 1816 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL; 1817 struct rb_node **p, *parent; 1818 struct ceph_pool_perm *perm; 1819 struct page **pages; 1820 size_t pool_ns_len; 1821 int err = 0, err2 = 0, have = 0; 1822 1823 down_read(&mdsc->pool_perm_rwsem); 1824 p = &mdsc->pool_perm_tree.rb_node; 1825 while (*p) { 1826 perm = rb_entry(*p, struct ceph_pool_perm, node); 1827 if (pool < perm->pool) 1828 p = &(*p)->rb_left; 1829 else if (pool > perm->pool) 1830 p = &(*p)->rb_right; 1831 else { 1832 int ret = ceph_compare_string(pool_ns, 1833 perm->pool_ns, 1834 perm->pool_ns_len); 1835 if (ret < 0) 1836 p = &(*p)->rb_left; 1837 else if (ret > 0) 1838 p = &(*p)->rb_right; 1839 else { 1840 have = perm->perm; 1841 break; 1842 } 1843 } 1844 } 1845 up_read(&mdsc->pool_perm_rwsem); 1846 if (*p) 1847 goto out; 1848 1849 if (pool_ns) 1850 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n", 1851 pool, (int)pool_ns->len, pool_ns->str); 1852 else 1853 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool); 1854 1855 down_write(&mdsc->pool_perm_rwsem); 1856 p = &mdsc->pool_perm_tree.rb_node; 1857 parent = NULL; 1858 while (*p) { 1859 parent = *p; 1860 perm = rb_entry(parent, struct ceph_pool_perm, node); 1861 if (pool < perm->pool) 1862 p = &(*p)->rb_left; 1863 else if (pool > perm->pool) 1864 p = &(*p)->rb_right; 1865 else { 1866 int ret = ceph_compare_string(pool_ns, 1867 perm->pool_ns, 1868 perm->pool_ns_len); 1869 if (ret < 0) 1870 p = &(*p)->rb_left; 1871 else if (ret > 0) 1872 p = &(*p)->rb_right; 1873 else { 1874 have = perm->perm; 1875 break; 1876 } 1877 } 1878 } 1879 if (*p) { 1880 up_write(&mdsc->pool_perm_rwsem); 1881 goto out; 1882 } 1883 1884 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1885 1, false, GFP_NOFS); 1886 if (!rd_req) { 1887 err = -ENOMEM; 1888 goto out_unlock; 1889 } 1890 1891 rd_req->r_flags = CEPH_OSD_FLAG_READ; 1892 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0); 1893 rd_req->r_base_oloc.pool = pool; 1894 if (pool_ns) 1895 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns); 1896 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino); 1897 1898 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS); 1899 if (err) 1900 goto out_unlock; 1901 1902 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL, 1903 1, false, GFP_NOFS); 1904 if (!wr_req) { 1905 err = -ENOMEM; 1906 goto out_unlock; 1907 } 1908 1909 wr_req->r_flags = CEPH_OSD_FLAG_WRITE; 1910 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL); 1911 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc); 1912 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid); 1913 1914 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS); 1915 if (err) 1916 goto out_unlock; 1917 1918 /* one page should be large enough for STAT data */ 1919 pages = ceph_alloc_page_vector(1, GFP_KERNEL); 1920 if (IS_ERR(pages)) { 1921 err = PTR_ERR(pages); 1922 goto out_unlock; 1923 } 1924 1925 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE, 1926 0, false, true); 1927 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false); 1928 1929 wr_req->r_mtime = ci->vfs_inode.i_mtime; 1930 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false); 1931 1932 if (!err) 1933 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req); 1934 if (!err2) 1935 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req); 1936 1937 if (err >= 0 || err == -ENOENT) 1938 have |= POOL_READ; 1939 else if (err != -EPERM) { 1940 if (err == -EBLOCKLISTED) 1941 fsc->blocklisted = true; 1942 goto out_unlock; 1943 } 1944 1945 if (err2 == 0 || err2 == -EEXIST) 1946 have |= POOL_WRITE; 1947 else if (err2 != -EPERM) { 1948 if (err2 == -EBLOCKLISTED) 1949 fsc->blocklisted = true; 1950 err = err2; 1951 goto out_unlock; 1952 } 1953 1954 pool_ns_len = pool_ns ? pool_ns->len : 0; 1955 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS); 1956 if (!perm) { 1957 err = -ENOMEM; 1958 goto out_unlock; 1959 } 1960 1961 perm->pool = pool; 1962 perm->perm = have; 1963 perm->pool_ns_len = pool_ns_len; 1964 if (pool_ns_len > 0) 1965 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len); 1966 perm->pool_ns[pool_ns_len] = 0; 1967 1968 rb_link_node(&perm->node, parent, p); 1969 rb_insert_color(&perm->node, &mdsc->pool_perm_tree); 1970 err = 0; 1971 out_unlock: 1972 up_write(&mdsc->pool_perm_rwsem); 1973 1974 ceph_osdc_put_request(rd_req); 1975 ceph_osdc_put_request(wr_req); 1976 out: 1977 if (!err) 1978 err = have; 1979 if (pool_ns) 1980 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n", 1981 pool, (int)pool_ns->len, pool_ns->str, err); 1982 else 1983 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err); 1984 return err; 1985 } 1986 1987 int ceph_pool_perm_check(struct inode *inode, int need) 1988 { 1989 struct ceph_inode_info *ci = ceph_inode(inode); 1990 struct ceph_string *pool_ns; 1991 s64 pool; 1992 int ret, flags; 1993 1994 /* Only need to do this for regular files */ 1995 if (!S_ISREG(inode->i_mode)) 1996 return 0; 1997 1998 if (ci->i_vino.snap != CEPH_NOSNAP) { 1999 /* 2000 * Pool permission check needs to write to the first object. 2001 * But for snapshot, head of the first object may have alread 2002 * been deleted. Skip check to avoid creating orphan object. 2003 */ 2004 return 0; 2005 } 2006 2007 if (ceph_test_mount_opt(ceph_inode_to_client(inode), 2008 NOPOOLPERM)) 2009 return 0; 2010 2011 spin_lock(&ci->i_ceph_lock); 2012 flags = ci->i_ceph_flags; 2013 pool = ci->i_layout.pool_id; 2014 spin_unlock(&ci->i_ceph_lock); 2015 check: 2016 if (flags & CEPH_I_POOL_PERM) { 2017 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) { 2018 dout("ceph_pool_perm_check pool %lld no read perm\n", 2019 pool); 2020 return -EPERM; 2021 } 2022 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) { 2023 dout("ceph_pool_perm_check pool %lld no write perm\n", 2024 pool); 2025 return -EPERM; 2026 } 2027 return 0; 2028 } 2029 2030 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns); 2031 ret = __ceph_pool_perm_get(ci, pool, pool_ns); 2032 ceph_put_string(pool_ns); 2033 if (ret < 0) 2034 return ret; 2035 2036 flags = CEPH_I_POOL_PERM; 2037 if (ret & POOL_READ) 2038 flags |= CEPH_I_POOL_RD; 2039 if (ret & POOL_WRITE) 2040 flags |= CEPH_I_POOL_WR; 2041 2042 spin_lock(&ci->i_ceph_lock); 2043 if (pool == ci->i_layout.pool_id && 2044 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) { 2045 ci->i_ceph_flags |= flags; 2046 } else { 2047 pool = ci->i_layout.pool_id; 2048 flags = ci->i_ceph_flags; 2049 } 2050 spin_unlock(&ci->i_ceph_lock); 2051 goto check; 2052 } 2053 2054 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc) 2055 { 2056 struct ceph_pool_perm *perm; 2057 struct rb_node *n; 2058 2059 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) { 2060 n = rb_first(&mdsc->pool_perm_tree); 2061 perm = rb_entry(n, struct ceph_pool_perm, node); 2062 rb_erase(n, &mdsc->pool_perm_tree); 2063 kfree(perm); 2064 } 2065 } 2066