fs/ceph/addr.c

   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 "crypto.h"
  22 #include <linux/ceph/osd_client.h>
  23 #include <linux/ceph/striper.h>
  24
  25 /*
  26  * Ceph address space ops.
  27  *
  28  * There are a few funny things going on here.
  29  *
  30  * The page->private field is used to reference a struct
  31  * ceph_snap_context for _every_ dirty page.  This indicates which
  32  * snapshot the page was logically dirtied in, and thus which snap
  33  * context needs to be associated with the osd write during writeback.
  34  *
  35  * Similarly, struct ceph_inode_info maintains a set of counters to
  36  * count dirty pages on the inode.  In the absence of snapshots,
  37  * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
  38  *
  39  * When a snapshot is taken (that is, when the client receives
  40  * notification that a snapshot was taken), each inode with caps and
  41  * with dirty pages (dirty pages implies there is a cap) gets a new
  42  * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
  43  * order, new snaps go to the tail).  The i_wrbuffer_ref_head count is
  44  * moved to capsnap->dirty. (Unless a sync write is currently in
  45  * progress.  In that case, the capsnap is said to be "pending", new
  46  * writes cannot start, and the capsnap isn't "finalized" until the
  47  * write completes (or fails) and a final size/mtime for the inode for
  48  * that snap can be settled upon.)  i_wrbuffer_ref_head is reset to 0.
  49  *
  50  * On writeback, we must submit writes to the osd IN SNAP ORDER.  So,
  51  * we look for the first capsnap in i_cap_snaps and write out pages in
  52  * that snap context _only_.  Then we move on to the next capsnap,
  53  * eventually reaching the "live" or "head" context (i.e., pages that
  54  * are not yet snapped) and are writing the most recently dirtied
  55  * pages.
  56  *
  57  * Invalidate and so forth must take care to ensure the dirty page
  58  * accounting is preserved.
  59  */
  60
  61 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
  62 #define CONGESTION_OFF_THRESH(congestion_kb)                            \
  63         (CONGESTION_ON_THRESH(congestion_kb) -                          \
  64          (CONGESTION_ON_THRESH(congestion_kb) >> 2))
  65
  66 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
  67                                         struct folio **foliop, void **_fsdata);
  68
  69 static inline struct ceph_snap_context *page_snap_context(struct page *page)
  70 {
  71         if (PagePrivate(page))
  72                 return (void *)page->private;
  73         return NULL;
  74 }
  75
  76 /*
  77  * Dirty a page.  Optimistically adjust accounting, on the assumption
  78  * that we won't race with invalidate.  If we do, readjust.
  79  */
  80 static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
  81 {
  82         struct inode *inode;
  83         struct ceph_inode_info *ci;
  84         struct ceph_snap_context *snapc;
  85
  86         if (folio_test_dirty(folio)) {
  87                 dout("%p dirty_folio %p idx %lu -- already dirty\n",
  88                      mapping->host, folio, folio->index);
  89                 VM_BUG_ON_FOLIO(!folio_test_private(folio), folio);
  90                 return false;
  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 dirty_folio %p idx %lu head %d/%d -> %d/%d "
 115              "snapc %p seq %lld (%d snaps)\n",
 116              mapping->host, folio, folio->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 folio->private.  Also set
 124          * PagePrivate so that we get invalidate_folio callback.
 125          */
 126         VM_WARN_ON_FOLIO(folio->private, folio);
 127         folio_attach_private(folio, snapc);
 128
 129         return ceph_fscache_dirty_folio(mapping, folio);
 130 }
 131
 132 /*
 133  * If we are truncating the full folio (i.e. offset == 0), adjust the
 134  * dirty folio counters appropriately.  Only called if there is private
 135  * data on the folio.
 136  */
 137 static void ceph_invalidate_folio(struct folio *folio, size_t offset,
 138                                 size_t length)
 139 {
 140         struct inode *inode;
 141         struct ceph_inode_info *ci;
 142         struct ceph_snap_context *snapc;
 143
 144         inode = folio->mapping->host;
 145         ci = ceph_inode(inode);
 146
 147         if (offset != 0 || length != folio_size(folio)) {
 148                 dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n",
 149                      inode, folio->index, offset, length);
 150                 return;
 151         }
 152
 153         WARN_ON(!folio_test_locked(folio));
 154         if (folio_test_private(folio)) {
 155                 dout("%p invalidate_folio idx %lu full dirty page\n",
 156                      inode, folio->index);
 157
 158                 snapc = folio_detach_private(folio);
 159                 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
 160                 ceph_put_snap_context(snapc);
 161         }
 162
 163         folio_wait_fscache(folio);
 164 }
 165
 166 static bool ceph_release_folio(struct folio *folio, gfp_t gfp)
 167 {
 168         struct inode *inode = folio->mapping->host;
 169
 170         dout("%llx:%llx release_folio idx %lu (%sdirty)\n",
 171              ceph_vinop(inode),
 172              folio->index, folio_test_dirty(folio) ? "" : "not ");
 173
 174         if (folio_test_private(folio))
 175                 return false;
 176
 177         if (folio_test_fscache(folio)) {
 178                 if (current_is_kswapd() || !(gfp & __GFP_FS))
 179                         return false;
 180                 folio_wait_fscache(folio);
 181         }
 182         ceph_fscache_note_page_release(inode);
 183         return true;
 184 }
 185
 186 static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq)
 187 {
 188         struct inode *inode = rreq->inode;
 189         struct ceph_inode_info *ci = ceph_inode(inode);
 190         struct ceph_file_layout *lo = &ci->i_layout;
 191         unsigned long max_pages = inode->i_sb->s_bdi->ra_pages;
 192         loff_t end = rreq->start + rreq->len, new_end;
 193         struct ceph_netfs_request_data *priv = rreq->netfs_priv;
 194         unsigned long max_len;
 195         u32 blockoff;
 196
 197         if (priv) {
 198                 /* Readahead is disabled by posix_fadvise POSIX_FADV_RANDOM */
 199                 if (priv->file_ra_disabled)
 200                         max_pages = 0;
 201                 else
 202                         max_pages = priv->file_ra_pages;
 203
 204         }
 205
 206         /* Readahead is disabled */
 207         if (!max_pages)
 208                 return;
 209
 210         max_len = max_pages << PAGE_SHIFT;
 211
 212         /*
 213          * Try to expand the length forward by rounding up it to the next
 214          * block, but do not exceed the file size, unless the original
 215          * request already exceeds it.
 216          */
 217         new_end = min(round_up(end, lo->stripe_unit), rreq->i_size);
 218         if (new_end > end && new_end <= rreq->start + max_len)
 219                 rreq->len = new_end - rreq->start;
 220
 221         /* Try to expand the start downward */
 222         div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
 223         if (rreq->len + blockoff <= max_len) {
 224                 rreq->start -= blockoff;
 225                 rreq->len += blockoff;
 226         }
 227 }
 228
 229 static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq)
 230 {
 231         struct inode *inode = subreq->rreq->inode;
 232         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 233         struct ceph_inode_info *ci = ceph_inode(inode);
 234         u64 objno, objoff;
 235         u32 xlen;
 236
 237         /* Truncate the extent at the end of the current block */
 238         ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
 239                                       &objno, &objoff, &xlen);
 240         subreq->len = min(xlen, fsc->mount_options->rsize);
 241         return true;
 242 }
 243
 244 static void finish_netfs_read(struct ceph_osd_request *req)
 245 {
 246         struct inode *inode = req->r_inode;
 247         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 248         struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
 249         struct netfs_io_subrequest *subreq = req->r_priv;
 250         struct ceph_osd_req_op *op = &req->r_ops[0];
 251         int err = req->r_result;
 252         bool sparse = (op->op == CEPH_OSD_OP_SPARSE_READ);
 253
 254         ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
 255                                  req->r_end_latency, osd_data->length, err);
 256
 257         dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
 258              subreq->len, i_size_read(req->r_inode));
 259
 260         /* no object means success but no data */
 261         if (err == -ENOENT)
 262                 err = 0;
 263         else if (err == -EBLOCKLISTED)
 264                 fsc->blocklisted = true;
 265
 266         if (err >= 0) {
 267                 if (sparse && err > 0)
 268                         err = ceph_sparse_ext_map_end(op);
 269                 if (err < subreq->len)
 270                         __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
 271                 if (IS_ENCRYPTED(inode) && err > 0) {
 272                         err = ceph_fscrypt_decrypt_extents(inode,
 273                                         osd_data->pages, subreq->start,
 274                                         op->extent.sparse_ext,
 275                                         op->extent.sparse_ext_cnt);
 276                         if (err > subreq->len)
 277                                 err = subreq->len;
 278                 }
 279         }
 280
 281         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
 282                 ceph_put_page_vector(osd_data->pages,
 283                                      calc_pages_for(osd_data->alignment,
 284                                         osd_data->length), false);
 285         }
 286         netfs_subreq_terminated(subreq, err, false);
 287         iput(req->r_inode);
 288 }
 289
 290 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 291 {
 292         struct netfs_io_request *rreq = subreq->rreq;
 293         struct inode *inode = rreq->inode;
 294         struct ceph_mds_reply_info_parsed *rinfo;
 295         struct ceph_mds_reply_info_in *iinfo;
 296         struct ceph_mds_request *req;
 297         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 298         struct ceph_inode_info *ci = ceph_inode(inode);
 299         struct iov_iter iter;
 300         ssize_t err = 0;
 301         size_t len;
 302         int mode;
 303
 304         __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
 305         __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
 306
 307         if (subreq->start >= inode->i_size)
 308                 goto out;
 309
 310         /* We need to fetch the inline data. */
 311         mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA);
 312         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
 313         if (IS_ERR(req)) {
 314                 err = PTR_ERR(req);
 315                 goto out;
 316         }
 317         req->r_ino1 = ci->i_vino;
 318         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA);
 319         req->r_num_caps = 2;
 320
 321         err = ceph_mdsc_do_request(mdsc, NULL, req);
 322         if (err < 0)
 323                 goto out;
 324
 325         rinfo = &req->r_reply_info;
 326         iinfo = &rinfo->targeti;
 327         if (iinfo->inline_version == CEPH_INLINE_NONE) {
 328                 /* The data got uninlined */
 329                 ceph_mdsc_put_request(req);
 330                 return false;
 331         }
 332
 333         len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len);
 334         iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
 335         err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter);
 336         if (err == 0)
 337                 err = -EFAULT;
 338
 339         ceph_mdsc_put_request(req);
 340 out:
 341         netfs_subreq_terminated(subreq, err, false);
 342         return true;
 343 }
 344
 345 static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq)
 346 {
 347         struct netfs_io_request *rreq = subreq->rreq;
 348         struct inode *inode = rreq->inode;
 349         struct ceph_inode_info *ci = ceph_inode(inode);
 350         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 351         struct ceph_osd_request *req = NULL;
 352         struct ceph_vino vino = ceph_vino(inode);
 353         struct iov_iter iter;
 354         int err = 0;
 355         u64 len = subreq->len;
 356         bool sparse = IS_ENCRYPTED(inode) || ceph_test_mount_opt(fsc, SPARSEREAD);
 357         u64 off = subreq->start;
 358
 359         if (ceph_inode_is_shutdown(inode)) {
 360                 err = -EIO;
 361                 goto out;
 362         }
 363
 364         if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
 365                 return;
 366
 367         ceph_fscrypt_adjust_off_and_len(inode, &off, &len);
 368
 369         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino,
 370                         off, &len, 0, 1, sparse ? CEPH_OSD_OP_SPARSE_READ : CEPH_OSD_OP_READ,
 371                         CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
 372                         NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
 373         if (IS_ERR(req)) {
 374                 err = PTR_ERR(req);
 375                 req = NULL;
 376                 goto out;
 377         }
 378
 379         if (sparse) {
 380                 err = ceph_alloc_sparse_ext_map(&req->r_ops[0]);
 381                 if (err)
 382                         goto out;
 383         }
 384
 385         dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
 386
 387         iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
 388
 389         /*
 390          * FIXME: For now, use CEPH_OSD_DATA_TYPE_PAGES instead of _ITER for
 391          * encrypted inodes. We'd need infrastructure that handles an iov_iter
 392          * instead of page arrays, and we don't have that as of yet. Once the
 393          * dust settles on the write helpers and encrypt/decrypt routines for
 394          * netfs, we should be able to rework this.
 395          */
 396         if (IS_ENCRYPTED(inode)) {
 397                 struct page **pages;
 398                 size_t page_off;
 399
 400                 err = iov_iter_get_pages_alloc2(&iter, &pages, len, &page_off);
 401                 if (err < 0) {
 402                         dout("%s: iov_ter_get_pages_alloc returned %d\n",
 403                              __func__, err);
 404                         goto out;
 405                 }
 406
 407                 /* should always give us a page-aligned read */
 408                 WARN_ON_ONCE(page_off);
 409                 len = err;
 410                 err = 0;
 411
 412                 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false,
 413                                                  false);
 414         } else {
 415                 osd_req_op_extent_osd_iter(req, 0, &iter);
 416         }
 417         req->r_callback = finish_netfs_read;
 418         req->r_priv = subreq;
 419         req->r_inode = inode;
 420         ihold(inode);
 421
 422         ceph_osdc_start_request(req->r_osdc, req);
 423 out:
 424         ceph_osdc_put_request(req);
 425         if (err)
 426                 netfs_subreq_terminated(subreq, err, false);
 427         dout("%s: result %d\n", __func__, err);
 428 }
 429
 430 static int ceph_init_request(struct netfs_io_request *rreq, struct file *file)
 431 {
 432         struct inode *inode = rreq->inode;
 433         int got = 0, want = CEPH_CAP_FILE_CACHE;
 434         struct ceph_netfs_request_data *priv;
 435         int ret = 0;
 436
 437         if (rreq->origin != NETFS_READAHEAD)
 438                 return 0;
 439
 440         priv = kzalloc(sizeof(*priv), GFP_NOFS);
 441         if (!priv)
 442                 return -ENOMEM;
 443
 444         if (file) {
 445                 struct ceph_rw_context *rw_ctx;
 446                 struct ceph_file_info *fi = file->private_data;
 447
 448                 priv->file_ra_pages = file->f_ra.ra_pages;
 449                 priv->file_ra_disabled = file->f_mode & FMODE_RANDOM;
 450
 451                 rw_ctx = ceph_find_rw_context(fi);
 452                 if (rw_ctx) {
 453                         rreq->netfs_priv = priv;
 454                         return 0;
 455                 }
 456         }
 457
 458         /*
 459          * readahead callers do not necessarily hold Fcb caps
 460          * (e.g. fadvise, madvise).
 461          */
 462         ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
 463         if (ret < 0) {
 464                 dout("start_read %p, error getting cap\n", inode);
 465                 goto out;
 466         }
 467
 468         if (!(got & want)) {
 469                 dout("start_read %p, no cache cap\n", inode);
 470                 ret = -EACCES;
 471                 goto out;
 472         }
 473         if (ret == 0) {
 474                 ret = -EACCES;
 475                 goto out;
 476         }
 477
 478         priv->caps = got;
 479         rreq->netfs_priv = priv;
 480
 481 out:
 482         if (ret < 0)
 483                 kfree(priv);
 484
 485         return ret;
 486 }
 487
 488 static void ceph_netfs_free_request(struct netfs_io_request *rreq)
 489 {
 490         struct ceph_netfs_request_data *priv = rreq->netfs_priv;
 491
 492         if (!priv)
 493                 return;
 494
 495         if (priv->caps)
 496                 ceph_put_cap_refs(ceph_inode(rreq->inode), priv->caps);
 497         kfree(priv);
 498         rreq->netfs_priv = NULL;
 499 }
 500
 501 const struct netfs_request_ops ceph_netfs_ops = {
 502         .init_request           = ceph_init_request,
 503         .free_request           = ceph_netfs_free_request,
 504         .begin_cache_operation  = ceph_begin_cache_operation,
 505         .issue_read             = ceph_netfs_issue_read,
 506         .expand_readahead       = ceph_netfs_expand_readahead,
 507         .clamp_length           = ceph_netfs_clamp_length,
 508         .check_write_begin      = ceph_netfs_check_write_begin,
 509 };
 510
 511 #ifdef CONFIG_CEPH_FSCACHE
 512 static void ceph_set_page_fscache(struct page *page)
 513 {
 514         set_page_fscache(page);
 515 }
 516
 517 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async)
 518 {
 519         struct inode *inode = priv;
 520
 521         if (IS_ERR_VALUE(error) && error != -ENOBUFS)
 522                 ceph_fscache_invalidate(inode, false);
 523 }
 524
 525 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
 526 {
 527         struct ceph_inode_info *ci = ceph_inode(inode);
 528         struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
 529
 530         fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
 531                                ceph_fscache_write_terminated, inode, caching);
 532 }
 533 #else
 534 static inline void ceph_set_page_fscache(struct page *page)
 535 {
 536 }
 537
 538 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
 539 {
 540 }
 541 #endif /* CONFIG_CEPH_FSCACHE */
 542
 543 struct ceph_writeback_ctl
 544 {
 545         loff_t i_size;
 546         u64 truncate_size;
 547         u32 truncate_seq;
 548         bool size_stable;
 549         bool head_snapc;
 550 };
 551
 552 /*
 553  * Get ref for the oldest snapc for an inode with dirty data... that is, the
 554  * only snap context we are allowed to write back.
 555  */
 556 static struct ceph_snap_context *
 557 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
 558                    struct ceph_snap_context *page_snapc)
 559 {
 560         struct ceph_inode_info *ci = ceph_inode(inode);
 561         struct ceph_snap_context *snapc = NULL;
 562         struct ceph_cap_snap *capsnap = NULL;
 563
 564         spin_lock(&ci->i_ceph_lock);
 565         list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
 566                 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
 567                      capsnap->context, capsnap->dirty_pages);
 568                 if (!capsnap->dirty_pages)
 569                         continue;
 570
 571                 /* get i_size, truncate_{seq,size} for page_snapc? */
 572                 if (snapc && capsnap->context != page_snapc)
 573                         continue;
 574
 575                 if (ctl) {
 576                         if (capsnap->writing) {
 577                                 ctl->i_size = i_size_read(inode);
 578                                 ctl->size_stable = false;
 579                         } else {
 580                                 ctl->i_size = capsnap->size;
 581                                 ctl->size_stable = true;
 582                         }
 583                         ctl->truncate_size = capsnap->truncate_size;
 584                         ctl->truncate_seq = capsnap->truncate_seq;
 585                         ctl->head_snapc = false;
 586                 }
 587
 588                 if (snapc)
 589                         break;
 590
 591                 snapc = ceph_get_snap_context(capsnap->context);
 592                 if (!page_snapc ||
 593                     page_snapc == snapc ||
 594                     page_snapc->seq > snapc->seq)
 595                         break;
 596         }
 597         if (!snapc && ci->i_wrbuffer_ref_head) {
 598                 snapc = ceph_get_snap_context(ci->i_head_snapc);
 599                 dout(" head snapc %p has %d dirty pages\n",
 600                      snapc, ci->i_wrbuffer_ref_head);
 601                 if (ctl) {
 602                         ctl->i_size = i_size_read(inode);
 603                         ctl->truncate_size = ci->i_truncate_size;
 604                         ctl->truncate_seq = ci->i_truncate_seq;
 605                         ctl->size_stable = false;
 606                         ctl->head_snapc = true;
 607                 }
 608         }
 609         spin_unlock(&ci->i_ceph_lock);
 610         return snapc;
 611 }
 612
 613 static u64 get_writepages_data_length(struct inode *inode,
 614                                       struct page *page, u64 start)
 615 {
 616         struct ceph_inode_info *ci = ceph_inode(inode);
 617         struct ceph_snap_context *snapc;
 618         struct ceph_cap_snap *capsnap = NULL;
 619         u64 end = i_size_read(inode);
 620         u64 ret;
 621
 622         snapc = page_snap_context(ceph_fscrypt_pagecache_page(page));
 623         if (snapc != ci->i_head_snapc) {
 624                 bool found = false;
 625                 spin_lock(&ci->i_ceph_lock);
 626                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
 627                         if (capsnap->context == snapc) {
 628                                 if (!capsnap->writing)
 629                                         end = capsnap->size;
 630                                 found = true;
 631                                 break;
 632                         }
 633                 }
 634                 spin_unlock(&ci->i_ceph_lock);
 635                 WARN_ON(!found);
 636         }
 637         if (end > ceph_fscrypt_page_offset(page) + thp_size(page))
 638                 end = ceph_fscrypt_page_offset(page) + thp_size(page);
 639         ret = end > start ? end - start : 0;
 640         if (ret && fscrypt_is_bounce_page(page))
 641                 ret = round_up(ret, CEPH_FSCRYPT_BLOCK_SIZE);
 642         return ret;
 643 }
 644
 645 /*
 646  * Write a single page, but leave the page locked.
 647  *
 648  * If we get a write error, mark the mapping for error, but still adjust the
 649  * dirty page accounting (i.e., page is no longer dirty).
 650  */
 651 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
 652 {
 653         struct folio *folio = page_folio(page);
 654         struct inode *inode = page->mapping->host;
 655         struct ceph_inode_info *ci = ceph_inode(inode);
 656         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 657         struct ceph_snap_context *snapc, *oldest;
 658         loff_t page_off = page_offset(page);
 659         int err;
 660         loff_t len = thp_size(page);
 661         loff_t wlen;
 662         struct ceph_writeback_ctl ceph_wbc;
 663         struct ceph_osd_client *osdc = &fsc->client->osdc;
 664         struct ceph_osd_request *req;
 665         bool caching = ceph_is_cache_enabled(inode);
 666         struct page *bounce_page = NULL;
 667
 668         dout("writepage %p idx %lu\n", page, page->index);
 669
 670         if (ceph_inode_is_shutdown(inode))
 671                 return -EIO;
 672
 673         /* verify this is a writeable snap context */
 674         snapc = page_snap_context(page);
 675         if (!snapc) {
 676                 dout("writepage %p page %p not dirty?\n", inode, page);
 677                 return 0;
 678         }
 679         oldest = get_oldest_context(inode, &ceph_wbc, snapc);
 680         if (snapc->seq > oldest->seq) {
 681                 dout("writepage %p page %p snapc %p not writeable - noop\n",
 682                      inode, page, snapc);
 683                 /* we should only noop if called by kswapd */
 684                 WARN_ON(!(current->flags & PF_MEMALLOC));
 685                 ceph_put_snap_context(oldest);
 686                 redirty_page_for_writepage(wbc, page);
 687                 return 0;
 688         }
 689         ceph_put_snap_context(oldest);
 690
 691         /* is this a partial page at end of file? */
 692         if (page_off >= ceph_wbc.i_size) {
 693                 dout("folio at %lu beyond eof %llu\n", folio->index,
 694                                 ceph_wbc.i_size);
 695                 folio_invalidate(folio, 0, folio_size(folio));
 696                 return 0;
 697         }
 698
 699         if (ceph_wbc.i_size < page_off + len)
 700                 len = ceph_wbc.i_size - page_off;
 701
 702         wlen = IS_ENCRYPTED(inode) ? round_up(len, CEPH_FSCRYPT_BLOCK_SIZE) : len;
 703         dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
 704              inode, page, page->index, page_off, wlen, snapc, snapc->seq);
 705
 706         if (atomic_long_inc_return(&fsc->writeback_count) >
 707             CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
 708                 fsc->write_congested = true;
 709
 710         req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode),
 711                                     page_off, &wlen, 0, 1, CEPH_OSD_OP_WRITE,
 712                                     CEPH_OSD_FLAG_WRITE, snapc,
 713                                     ceph_wbc.truncate_seq,
 714                                     ceph_wbc.truncate_size, true);
 715         if (IS_ERR(req)) {
 716                 redirty_page_for_writepage(wbc, page);
 717                 return PTR_ERR(req);
 718         }
 719
 720         if (wlen < len)
 721                 len = wlen;
 722
 723         set_page_writeback(page);
 724         if (caching)
 725                 ceph_set_page_fscache(page);
 726         ceph_fscache_write_to_cache(inode, page_off, len, caching);
 727
 728         if (IS_ENCRYPTED(inode)) {
 729                 bounce_page = fscrypt_encrypt_pagecache_blocks(page,
 730                                                     CEPH_FSCRYPT_BLOCK_SIZE, 0,
 731                                                     GFP_NOFS);
 732                 if (IS_ERR(bounce_page)) {
 733                         redirty_page_for_writepage(wbc, page);
 734                         end_page_writeback(page);
 735                         ceph_osdc_put_request(req);
 736                         return PTR_ERR(bounce_page);
 737                 }
 738         }
 739
 740         /* it may be a short write due to an object boundary */
 741         WARN_ON_ONCE(len > thp_size(page));
 742         osd_req_op_extent_osd_data_pages(req, 0,
 743                         bounce_page ? &bounce_page : &page, wlen, 0,
 744                         false, false);
 745         dout("writepage %llu~%llu (%llu bytes, %sencrypted)\n",
 746              page_off, len, wlen, IS_ENCRYPTED(inode) ? "" : "not ");
 747
 748         req->r_mtime = inode->i_mtime;
 749         ceph_osdc_start_request(osdc, req);
 750         err = ceph_osdc_wait_request(osdc, req);
 751
 752         ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
 753                                   req->r_end_latency, len, err);
 754         fscrypt_free_bounce_page(bounce_page);
 755         ceph_osdc_put_request(req);
 756         if (err == 0)
 757                 err = len;
 758
 759         if (err < 0) {
 760                 struct writeback_control tmp_wbc;
 761                 if (!wbc)
 762                         wbc = &tmp_wbc;
 763                 if (err == -ERESTARTSYS) {
 764                         /* killed by SIGKILL */
 765                         dout("writepage interrupted page %p\n", page);
 766                         redirty_page_for_writepage(wbc, page);
 767                         end_page_writeback(page);
 768                         return err;
 769                 }
 770                 if (err == -EBLOCKLISTED)
 771                         fsc->blocklisted = true;
 772                 dout("writepage setting page/mapping error %d %p\n",
 773                      err, page);
 774                 mapping_set_error(&inode->i_data, err);
 775                 wbc->pages_skipped++;
 776         } else {
 777                 dout("writepage cleaned page %p\n", page);
 778                 err = 0;  /* vfs expects us to return 0 */
 779         }
 780         oldest = detach_page_private(page);
 781         WARN_ON_ONCE(oldest != snapc);
 782         end_page_writeback(page);
 783         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
 784         ceph_put_snap_context(snapc);  /* page's reference */
 785
 786         if (atomic_long_dec_return(&fsc->writeback_count) <
 787             CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
 788                 fsc->write_congested = false;
 789
 790         return err;
 791 }
 792
 793 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
 794 {
 795         int err;
 796         struct inode *inode = page->mapping->host;
 797         BUG_ON(!inode);
 798         ihold(inode);
 799
 800         if (wbc->sync_mode == WB_SYNC_NONE &&
 801             ceph_inode_to_client(inode)->write_congested)
 802                 return AOP_WRITEPAGE_ACTIVATE;
 803
 804         wait_on_page_fscache(page);
 805
 806         err = writepage_nounlock(page, wbc);
 807         if (err == -ERESTARTSYS) {
 808                 /* direct memory reclaimer was killed by SIGKILL. return 0
 809                  * to prevent caller from setting mapping/page error */
 810                 err = 0;
 811         }
 812         unlock_page(page);
 813         iput(inode);
 814         return err;
 815 }
 816
 817 /*
 818  * async writeback completion handler.
 819  *
 820  * If we get an error, set the mapping error bit, but not the individual
 821  * page error bits.
 822  */
 823 static void writepages_finish(struct ceph_osd_request *req)
 824 {
 825         struct inode *inode = req->r_inode;
 826         struct ceph_inode_info *ci = ceph_inode(inode);
 827         struct ceph_osd_data *osd_data;
 828         struct page *page;
 829         int num_pages, total_pages = 0;
 830         int i, j;
 831         int rc = req->r_result;
 832         struct ceph_snap_context *snapc = req->r_snapc;
 833         struct address_space *mapping = inode->i_mapping;
 834         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 835         unsigned int len = 0;
 836         bool remove_page;
 837
 838         dout("writepages_finish %p rc %d\n", inode, rc);
 839         if (rc < 0) {
 840                 mapping_set_error(mapping, rc);
 841                 ceph_set_error_write(ci);
 842                 if (rc == -EBLOCKLISTED)
 843                         fsc->blocklisted = true;
 844         } else {
 845                 ceph_clear_error_write(ci);
 846         }
 847
 848         /*
 849          * We lost the cache cap, need to truncate the page before
 850          * it is unlocked, otherwise we'd truncate it later in the
 851          * page truncation thread, possibly losing some data that
 852          * raced its way in
 853          */
 854         remove_page = !(ceph_caps_issued(ci) &
 855                         (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
 856
 857         /* clean all pages */
 858         for (i = 0; i < req->r_num_ops; i++) {
 859                 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) {
 860                         pr_warn("%s incorrect op %d req %p index %d tid %llu\n",
 861                                 __func__, req->r_ops[i].op, req, i, req->r_tid);
 862                         break;
 863                 }
 864
 865                 osd_data = osd_req_op_extent_osd_data(req, i);
 866                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
 867                 len += osd_data->length;
 868                 num_pages = calc_pages_for((u64)osd_data->alignment,
 869                                            (u64)osd_data->length);
 870                 total_pages += num_pages;
 871                 for (j = 0; j < num_pages; j++) {
 872                         page = osd_data->pages[j];
 873                         if (fscrypt_is_bounce_page(page)) {
 874                                 page = fscrypt_pagecache_page(page);
 875                                 fscrypt_free_bounce_page(osd_data->pages[j]);
 876                                 osd_data->pages[j] = page;
 877                         }
 878                         BUG_ON(!page);
 879                         WARN_ON(!PageUptodate(page));
 880
 881                         if (atomic_long_dec_return(&fsc->writeback_count) <
 882                              CONGESTION_OFF_THRESH(
 883                                         fsc->mount_options->congestion_kb))
 884                                 fsc->write_congested = false;
 885
 886                         ceph_put_snap_context(detach_page_private(page));
 887                         end_page_writeback(page);
 888                         dout("unlocking %p\n", page);
 889
 890                         if (remove_page)
 891                                 generic_error_remove_page(inode->i_mapping,
 892                                                           page);
 893
 894                         unlock_page(page);
 895                 }
 896                 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
 897                      inode, osd_data->length, rc >= 0 ? num_pages : 0);
 898
 899                 release_pages(osd_data->pages, num_pages);
 900         }
 901
 902         ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
 903                                   req->r_end_latency, len, rc);
 904
 905         ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
 906
 907         osd_data = osd_req_op_extent_osd_data(req, 0);
 908         if (osd_data->pages_from_pool)
 909                 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
 910         else
 911                 kfree(osd_data->pages);
 912         ceph_osdc_put_request(req);
 913 }
 914
 915 /*
 916  * initiate async writeback
 917  */
 918 static int ceph_writepages_start(struct address_space *mapping,
 919                                  struct writeback_control *wbc)
 920 {
 921         struct inode *inode = mapping->host;
 922         struct ceph_inode_info *ci = ceph_inode(inode);
 923         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 924         struct ceph_vino vino = ceph_vino(inode);
 925         pgoff_t index, start_index, end = -1;
 926         struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
 927         struct folio_batch fbatch;
 928         int rc = 0;
 929         unsigned int wsize = i_blocksize(inode);
 930         struct ceph_osd_request *req = NULL;
 931         struct ceph_writeback_ctl ceph_wbc;
 932         bool should_loop, range_whole = false;
 933         bool done = false;
 934         bool caching = ceph_is_cache_enabled(inode);
 935         xa_mark_t tag;
 936
 937         if (wbc->sync_mode == WB_SYNC_NONE &&
 938             fsc->write_congested)
 939                 return 0;
 940
 941         dout("writepages_start %p (mode=%s)\n", inode,
 942              wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
 943              (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
 944
 945         if (ceph_inode_is_shutdown(inode)) {
 946                 if (ci->i_wrbuffer_ref > 0) {
 947                         pr_warn_ratelimited(
 948                                 "writepage_start %p %lld forced umount\n",
 949                                 inode, ceph_ino(inode));
 950                 }
 951                 mapping_set_error(mapping, -EIO);
 952                 return -EIO; /* we're in a forced umount, don't write! */
 953         }
 954         if (fsc->mount_options->wsize < wsize)
 955                 wsize = fsc->mount_options->wsize;
 956
 957         folio_batch_init(&fbatch);
 958
 959         start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
 960         index = start_index;
 961
 962         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) {
 963                 tag = PAGECACHE_TAG_TOWRITE;
 964         } else {
 965                 tag = PAGECACHE_TAG_DIRTY;
 966         }
 967 retry:
 968         /* find oldest snap context with dirty data */
 969         snapc = get_oldest_context(inode, &ceph_wbc, NULL);
 970         if (!snapc) {
 971                 /* hmm, why does writepages get called when there
 972                    is no dirty data? */
 973                 dout(" no snap context with dirty data?\n");
 974                 goto out;
 975         }
 976         dout(" oldest snapc is %p seq %lld (%d snaps)\n",
 977              snapc, snapc->seq, snapc->num_snaps);
 978
 979         should_loop = false;
 980         if (ceph_wbc.head_snapc && snapc != last_snapc) {
 981                 /* where to start/end? */
 982                 if (wbc->range_cyclic) {
 983                         index = start_index;
 984                         end = -1;
 985                         if (index > 0)
 986                                 should_loop = true;
 987                         dout(" cyclic, start at %lu\n", index);
 988                 } else {
 989                         index = wbc->range_start >> PAGE_SHIFT;
 990                         end = wbc->range_end >> PAGE_SHIFT;
 991                         if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 992                                 range_whole = true;
 993                         dout(" not cyclic, %lu to %lu\n", index, end);
 994                 }
 995         } else if (!ceph_wbc.head_snapc) {
 996                 /* Do not respect wbc->range_{start,end}. Dirty pages
 997                  * in that range can be associated with newer snapc.
 998                  * They are not writeable until we write all dirty pages
 999                  * associated with 'snapc' get written */
1000                 if (index > 0)
1001                         should_loop = true;
1002                 dout(" non-head snapc, range whole\n");
1003         }
1004
1005         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1006                 tag_pages_for_writeback(mapping, index, end);
1007
1008         ceph_put_snap_context(last_snapc);
1009         last_snapc = snapc;
1010
1011         while (!done && index <= end) {
1012                 int num_ops = 0, op_idx;
1013                 unsigned i, nr_folios, max_pages, locked_pages = 0;
1014                 struct page **pages = NULL, **data_pages;
1015                 struct page *page;
1016                 pgoff_t strip_unit_end = 0;
1017                 u64 offset = 0, len = 0;
1018                 bool from_pool = false;
1019
1020                 max_pages = wsize >> PAGE_SHIFT;
1021
1022 get_more_pages:
1023                 nr_folios = filemap_get_folios_tag(mapping, &index,
1024                                                    end, tag, &fbatch);
1025                 dout("pagevec_lookup_range_tag got %d\n", nr_folios);
1026                 if (!nr_folios && !locked_pages)
1027                         break;
1028                 for (i = 0; i < nr_folios && locked_pages < max_pages; i++) {
1029                         page = &fbatch.folios[i]->page;
1030                         dout("? %p idx %lu\n", page, page->index);
1031                         if (locked_pages == 0)
1032                                 lock_page(page);  /* first page */
1033                         else if (!trylock_page(page))
1034                                 break;
1035
1036                         /* only dirty pages, or our accounting breaks */
1037                         if (unlikely(!PageDirty(page)) ||
1038                             unlikely(page->mapping != mapping)) {
1039                                 dout("!dirty or !mapping %p\n", page);
1040                                 unlock_page(page);
1041                                 continue;
1042                         }
1043                         /* only if matching snap context */
1044                         pgsnapc = page_snap_context(page);
1045                         if (pgsnapc != snapc) {
1046                                 dout("page snapc %p %lld != oldest %p %lld\n",
1047                                      pgsnapc, pgsnapc->seq, snapc, snapc->seq);
1048                                 if (!should_loop &&
1049                                     !ceph_wbc.head_snapc &&
1050                                     wbc->sync_mode != WB_SYNC_NONE)
1051                                         should_loop = true;
1052                                 unlock_page(page);
1053                                 continue;
1054                         }
1055                         if (page_offset(page) >= ceph_wbc.i_size) {
1056                                 struct folio *folio = page_folio(page);
1057
1058                                 dout("folio at %lu beyond eof %llu\n",
1059                                      folio->index, ceph_wbc.i_size);
1060                                 if ((ceph_wbc.size_stable ||
1061                                     folio_pos(folio) >= i_size_read(inode)) &&
1062                                     folio_clear_dirty_for_io(folio))
1063                                         folio_invalidate(folio, 0,
1064                                                         folio_size(folio));
1065                                 folio_unlock(folio);
1066                                 continue;
1067                         }
1068                         if (strip_unit_end && (page->index > strip_unit_end)) {
1069                                 dout("end of strip unit %p\n", page);
1070                                 unlock_page(page);
1071                                 break;
1072                         }
1073                         if (PageWriteback(page) || PageFsCache(page)) {
1074                                 if (wbc->sync_mode == WB_SYNC_NONE) {
1075                                         dout("%p under writeback\n", page);
1076                                         unlock_page(page);
1077                                         continue;
1078                                 }
1079                                 dout("waiting on writeback %p\n", page);
1080                                 wait_on_page_writeback(page);
1081                                 wait_on_page_fscache(page);
1082                         }
1083
1084                         if (!clear_page_dirty_for_io(page)) {
1085                                 dout("%p !clear_page_dirty_for_io\n", page);
1086                                 unlock_page(page);
1087                                 continue;
1088                         }
1089
1090                         /*
1091                          * We have something to write.  If this is
1092                          * the first locked page this time through,
1093                          * calculate max possinle write size and
1094                          * allocate a page array
1095                          */
1096                         if (locked_pages == 0) {
1097                                 u64 objnum;
1098                                 u64 objoff;
1099                                 u32 xlen;
1100
1101                                 /* prepare async write request */
1102                                 offset = (u64)page_offset(page);
1103                                 ceph_calc_file_object_mapping(&ci->i_layout,
1104                                                               offset, wsize,
1105                                                               &objnum, &objoff,
1106                                                               &xlen);
1107                                 len = xlen;
1108
1109                                 num_ops = 1;
1110                                 strip_unit_end = page->index +
1111                                         ((len - 1) >> PAGE_SHIFT);
1112
1113                                 BUG_ON(pages);
1114                                 max_pages = calc_pages_for(0, (u64)len);
1115                                 pages = kmalloc_array(max_pages,
1116                                                       sizeof(*pages),
1117                                                       GFP_NOFS);
1118                                 if (!pages) {
1119                                         from_pool = true;
1120                                         pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1121                                         BUG_ON(!pages);
1122                                 }
1123
1124                                 len = 0;
1125                         } else if (page->index !=
1126                                    (offset + len) >> PAGE_SHIFT) {
1127                                 if (num_ops >= (from_pool ?  CEPH_OSD_SLAB_OPS :
1128                                                              CEPH_OSD_MAX_OPS)) {
1129                                         redirty_page_for_writepage(wbc, page);
1130                                         unlock_page(page);
1131                                         break;
1132                                 }
1133
1134                                 num_ops++;
1135                                 offset = (u64)page_offset(page);
1136                                 len = 0;
1137                         }
1138
1139                         /* note position of first page in fbatch */
1140                         dout("%p will write page %p idx %lu\n",
1141                              inode, page, page->index);
1142
1143                         if (atomic_long_inc_return(&fsc->writeback_count) >
1144                             CONGESTION_ON_THRESH(
1145                                     fsc->mount_options->congestion_kb))
1146                                 fsc->write_congested = true;
1147
1148                         if (IS_ENCRYPTED(inode)) {
1149                                 pages[locked_pages] =
1150                                         fscrypt_encrypt_pagecache_blocks(page,
1151                                                 PAGE_SIZE, 0,
1152                                                 locked_pages ? GFP_NOWAIT : GFP_NOFS);
1153                                 if (IS_ERR(pages[locked_pages])) {
1154                                         if (PTR_ERR(pages[locked_pages]) == -EINVAL)
1155                                                 pr_err("%s: inode->i_blkbits=%hhu\n",
1156                                                         __func__, inode->i_blkbits);
1157                                         /* better not fail on first page! */
1158                                         BUG_ON(locked_pages == 0);
1159                                         pages[locked_pages] = NULL;
1160                                         redirty_page_for_writepage(wbc, page);
1161                                         unlock_page(page);
1162                                         break;
1163                                 }
1164                                 ++locked_pages;
1165                         } else {
1166                                 pages[locked_pages++] = page;
1167                         }
1168
1169                         fbatch.folios[i] = NULL;
1170                         len += thp_size(page);
1171                 }
1172
1173                 /* did we get anything? */
1174                 if (!locked_pages)
1175                         goto release_folios;
1176                 if (i) {
1177                         unsigned j, n = 0;
1178                         /* shift unused page to beginning of fbatch */
1179                         for (j = 0; j < nr_folios; j++) {
1180                                 if (!fbatch.folios[j])
1181                                         continue;
1182                                 if (n < j)
1183                                         fbatch.folios[n] = fbatch.folios[j];
1184                                 n++;
1185                         }
1186                         fbatch.nr = n;
1187
1188                         if (nr_folios && i == nr_folios &&
1189                             locked_pages < max_pages) {
1190                                 dout("reached end fbatch, trying for more\n");
1191                                 folio_batch_release(&fbatch);
1192                                 goto get_more_pages;
1193                         }
1194                 }
1195
1196 new_request:
1197                 offset = ceph_fscrypt_page_offset(pages[0]);
1198                 len = wsize;
1199
1200                 req = ceph_osdc_new_request(&fsc->client->osdc,
1201                                         &ci->i_layout, vino,
1202                                         offset, &len, 0, num_ops,
1203                                         CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1204                                         snapc, ceph_wbc.truncate_seq,
1205                                         ceph_wbc.truncate_size, false);
1206                 if (IS_ERR(req)) {
1207                         req = ceph_osdc_new_request(&fsc->client->osdc,
1208                                                 &ci->i_layout, vino,
1209                                                 offset, &len, 0,
1210                                                 min(num_ops,
1211                                                     CEPH_OSD_SLAB_OPS),
1212                                                 CEPH_OSD_OP_WRITE,
1213                                                 CEPH_OSD_FLAG_WRITE,
1214                                                 snapc, ceph_wbc.truncate_seq,
1215                                                 ceph_wbc.truncate_size, true);
1216                         BUG_ON(IS_ERR(req));
1217                 }
1218                 BUG_ON(len < ceph_fscrypt_page_offset(pages[locked_pages - 1]) +
1219                              thp_size(pages[locked_pages - 1]) - offset);
1220
1221                 req->r_callback = writepages_finish;
1222                 req->r_inode = inode;
1223
1224                 /* Format the osd request message and submit the write */
1225                 len = 0;
1226                 data_pages = pages;
1227                 op_idx = 0;
1228                 for (i = 0; i < locked_pages; i++) {
1229                         struct page *page = ceph_fscrypt_pagecache_page(pages[i]);
1230
1231                         u64 cur_offset = page_offset(page);
1232                         /*
1233                          * Discontinuity in page range? Ceph can handle that by just passing
1234                          * multiple extents in the write op.
1235                          */
1236                         if (offset + len != cur_offset) {
1237                                 /* If it's full, stop here */
1238                                 if (op_idx + 1 == req->r_num_ops)
1239                                         break;
1240
1241                                 /* Kick off an fscache write with what we have so far. */
1242                                 ceph_fscache_write_to_cache(inode, offset, len, caching);
1243
1244                                 /* Start a new extent */
1245                                 osd_req_op_extent_dup_last(req, op_idx,
1246                                                            cur_offset - offset);
1247                                 dout("writepages got pages at %llu~%llu\n",
1248                                      offset, len);
1249                                 osd_req_op_extent_osd_data_pages(req, op_idx,
1250                                                         data_pages, len, 0,
1251                                                         from_pool, false);
1252                                 osd_req_op_extent_update(req, op_idx, len);
1253
1254                                 len = 0;
1255                                 offset = cur_offset;
1256                                 data_pages = pages + i;
1257                                 op_idx++;
1258                         }
1259
1260                         set_page_writeback(page);
1261                         if (caching)
1262                                 ceph_set_page_fscache(page);
1263                         len += thp_size(page);
1264                 }
1265                 ceph_fscache_write_to_cache(inode, offset, len, caching);
1266
1267                 if (ceph_wbc.size_stable) {
1268                         len = min(len, ceph_wbc.i_size - offset);
1269                 } else if (i == locked_pages) {
1270                         /* writepages_finish() clears writeback pages
1271                          * according to the data length, so make sure
1272                          * data length covers all locked pages */
1273                         u64 min_len = len + 1 - thp_size(page);
1274                         len = get_writepages_data_length(inode, pages[i - 1],
1275                                                          offset);
1276                         len = max(len, min_len);
1277                 }
1278                 if (IS_ENCRYPTED(inode))
1279                         len = round_up(len, CEPH_FSCRYPT_BLOCK_SIZE);
1280
1281                 dout("writepages got pages at %llu~%llu\n", offset, len);
1282
1283                 if (IS_ENCRYPTED(inode) &&
1284                     ((offset | len) & ~CEPH_FSCRYPT_BLOCK_MASK))
1285                         pr_warn("%s: bad encrypted write offset=%lld len=%llu\n",
1286                                 __func__, offset, len);
1287
1288                 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1289                                                  0, from_pool, false);
1290                 osd_req_op_extent_update(req, op_idx, len);
1291
1292                 BUG_ON(op_idx + 1 != req->r_num_ops);
1293
1294                 from_pool = false;
1295                 if (i < locked_pages) {
1296                         BUG_ON(num_ops <= req->r_num_ops);
1297                         num_ops -= req->r_num_ops;
1298                         locked_pages -= i;
1299
1300                         /* allocate new pages array for next request */
1301                         data_pages = pages;
1302                         pages = kmalloc_array(locked_pages, sizeof(*pages),
1303                                               GFP_NOFS);
1304                         if (!pages) {
1305                                 from_pool = true;
1306                                 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1307                                 BUG_ON(!pages);
1308                         }
1309                         memcpy(pages, data_pages + i,
1310                                locked_pages * sizeof(*pages));
1311                         memset(data_pages + i, 0,
1312                                locked_pages * sizeof(*pages));
1313                 } else {
1314                         BUG_ON(num_ops != req->r_num_ops);
1315                         index = pages[i - 1]->index + 1;
1316                         /* request message now owns the pages array */
1317                         pages = NULL;
1318                 }
1319
1320                 req->r_mtime = inode->i_mtime;
1321                 ceph_osdc_start_request(&fsc->client->osdc, req);
1322                 req = NULL;
1323
1324                 wbc->nr_to_write -= i;
1325                 if (pages)
1326                         goto new_request;
1327
1328                 /*
1329                  * We stop writing back only if we are not doing
1330                  * integrity sync. In case of integrity sync we have to
1331                  * keep going until we have written all the pages
1332                  * we tagged for writeback prior to entering this loop.
1333                  */
1334                 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1335                         done = true;
1336
1337 release_folios:
1338                 dout("folio_batch release on %d folios (%p)\n", (int)fbatch.nr,
1339                      fbatch.nr ? fbatch.folios[0] : NULL);
1340                 folio_batch_release(&fbatch);
1341         }
1342
1343         if (should_loop && !done) {
1344                 /* more to do; loop back to beginning of file */
1345                 dout("writepages looping back to beginning of file\n");
1346                 end = start_index - 1; /* OK even when start_index == 0 */
1347
1348                 /* to write dirty pages associated with next snapc,
1349                  * we need to wait until current writes complete */
1350                 if (wbc->sync_mode != WB_SYNC_NONE &&
1351                     start_index == 0 && /* all dirty pages were checked */
1352                     !ceph_wbc.head_snapc) {
1353                         struct page *page;
1354                         unsigned i, nr;
1355                         index = 0;
1356                         while ((index <= end) &&
1357                                (nr = filemap_get_folios_tag(mapping, &index,
1358                                                 (pgoff_t)-1,
1359                                                 PAGECACHE_TAG_WRITEBACK,
1360                                                 &fbatch))) {
1361                                 for (i = 0; i < nr; i++) {
1362                                         page = &fbatch.folios[i]->page;
1363                                         if (page_snap_context(page) != snapc)
1364                                                 continue;
1365                                         wait_on_page_writeback(page);
1366                                 }
1367                                 folio_batch_release(&fbatch);
1368                                 cond_resched();
1369                         }
1370                 }
1371
1372                 start_index = 0;
1373                 index = 0;
1374                 goto retry;
1375         }
1376
1377         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1378                 mapping->writeback_index = index;
1379
1380 out:
1381         ceph_osdc_put_request(req);
1382         ceph_put_snap_context(last_snapc);
1383         dout("writepages dend - startone, rc = %d\n", rc);
1384         return rc;
1385 }
1386
1387
1388
1389 /*
1390  * See if a given @snapc is either writeable, or already written.
1391  */
1392 static int context_is_writeable_or_written(struct inode *inode,
1393                                            struct ceph_snap_context *snapc)
1394 {
1395         struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1396         int ret = !oldest || snapc->seq <= oldest->seq;
1397
1398         ceph_put_snap_context(oldest);
1399         return ret;
1400 }
1401
1402 /**
1403  * ceph_find_incompatible - find an incompatible context and return it
1404  * @page: page being dirtied
1405  *
1406  * We are only allowed to write into/dirty a page if the page is
1407  * clean, or already dirty within the same snap context. Returns a
1408  * conflicting context if there is one, NULL if there isn't, or a
1409  * negative error code on other errors.
1410  *
1411  * Must be called with page lock held.
1412  */
1413 static struct ceph_snap_context *
1414 ceph_find_incompatible(struct page *page)
1415 {
1416         struct inode *inode = page->mapping->host;
1417         struct ceph_inode_info *ci = ceph_inode(inode);
1418
1419         if (ceph_inode_is_shutdown(inode)) {
1420                 dout(" page %p %llx:%llx is shutdown\n", page,
1421                      ceph_vinop(inode));
1422                 return ERR_PTR(-ESTALE);
1423         }
1424
1425         for (;;) {
1426                 struct ceph_snap_context *snapc, *oldest;
1427
1428                 wait_on_page_writeback(page);
1429
1430                 snapc = page_snap_context(page);
1431                 if (!snapc || snapc == ci->i_head_snapc)
1432                         break;
1433
1434                 /*
1435                  * this page is already dirty in another (older) snap
1436                  * context!  is it writeable now?
1437                  */
1438                 oldest = get_oldest_context(inode, NULL, NULL);
1439                 if (snapc->seq > oldest->seq) {
1440                         /* not writeable -- return it for the caller to deal with */
1441                         ceph_put_snap_context(oldest);
1442                         dout(" page %p snapc %p not current or oldest\n", page, snapc);
1443                         return ceph_get_snap_context(snapc);
1444                 }
1445                 ceph_put_snap_context(oldest);
1446
1447                 /* yay, writeable, do it now (without dropping page lock) */
1448                 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1449                 if (clear_page_dirty_for_io(page)) {
1450                         int r = writepage_nounlock(page, NULL);
1451                         if (r < 0)
1452                                 return ERR_PTR(r);
1453                 }
1454         }
1455         return NULL;
1456 }
1457
1458 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1459                                         struct folio **foliop, void **_fsdata)
1460 {
1461         struct inode *inode = file_inode(file);
1462         struct ceph_inode_info *ci = ceph_inode(inode);
1463         struct ceph_snap_context *snapc;
1464
1465         snapc = ceph_find_incompatible(folio_page(*foliop, 0));
1466         if (snapc) {
1467                 int r;
1468
1469                 folio_unlock(*foliop);
1470                 folio_put(*foliop);
1471                 *foliop = NULL;
1472                 if (IS_ERR(snapc))
1473                         return PTR_ERR(snapc);
1474
1475                 ceph_queue_writeback(inode);
1476                 r = wait_event_killable(ci->i_cap_wq,
1477                                         context_is_writeable_or_written(inode, snapc));
1478                 ceph_put_snap_context(snapc);
1479                 return r == 0 ? -EAGAIN : r;
1480         }
1481         return 0;
1482 }
1483
1484 /*
1485  * We are only allowed to write into/dirty the page if the page is
1486  * clean, or already dirty within the same snap context.
1487  */
1488 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1489                             loff_t pos, unsigned len,
1490                             struct page **pagep, void **fsdata)
1491 {
1492         struct inode *inode = file_inode(file);
1493         struct ceph_inode_info *ci = ceph_inode(inode);
1494         struct folio *folio = NULL;
1495         int r;
1496
1497         r = netfs_write_begin(&ci->netfs, file, inode->i_mapping, pos, len, &folio, NULL);
1498         if (r < 0)
1499                 return r;
1500
1501         folio_wait_fscache(folio);
1502         WARN_ON_ONCE(!folio_test_locked(folio));
1503         *pagep = &folio->page;
1504         return 0;
1505 }
1506
1507 /*
1508  * we don't do anything in here that simple_write_end doesn't do
1509  * except adjust dirty page accounting
1510  */
1511 static int ceph_write_end(struct file *file, struct address_space *mapping,
1512                           loff_t pos, unsigned len, unsigned copied,
1513                           struct page *subpage, void *fsdata)
1514 {
1515         struct folio *folio = page_folio(subpage);
1516         struct inode *inode = file_inode(file);
1517         bool check_cap = false;
1518
1519         dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file,
1520              inode, folio, (int)pos, (int)copied, (int)len);
1521
1522         if (!folio_test_uptodate(folio)) {
1523                 /* just return that nothing was copied on a short copy */
1524                 if (copied < len) {
1525                         copied = 0;
1526                         goto out;
1527                 }
1528                 folio_mark_uptodate(folio);
1529         }
1530
1531         /* did file size increase? */
1532         if (pos+copied > i_size_read(inode))
1533                 check_cap = ceph_inode_set_size(inode, pos+copied);
1534
1535         folio_mark_dirty(folio);
1536
1537 out:
1538         folio_unlock(folio);
1539         folio_put(folio);
1540
1541         if (check_cap)
1542                 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY);
1543
1544         return copied;
1545 }
1546
1547 const struct address_space_operations ceph_aops = {
1548         .read_folio = netfs_read_folio,
1549         .readahead = netfs_readahead,
1550         .writepage = ceph_writepage,
1551         .writepages = ceph_writepages_start,
1552         .write_begin = ceph_write_begin,
1553         .write_end = ceph_write_end,
1554         .dirty_folio = ceph_dirty_folio,
1555         .invalidate_folio = ceph_invalidate_folio,
1556         .release_folio = ceph_release_folio,
1557         .direct_IO = noop_direct_IO,
1558 };
1559
1560 static void ceph_block_sigs(sigset_t *oldset)
1561 {
1562         sigset_t mask;
1563         siginitsetinv(&mask, sigmask(SIGKILL));
1564         sigprocmask(SIG_BLOCK, &mask, oldset);
1565 }
1566
1567 static void ceph_restore_sigs(sigset_t *oldset)
1568 {
1569         sigprocmask(SIG_SETMASK, oldset, NULL);
1570 }
1571
1572 /*
1573  * vm ops
1574  */
1575 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1576 {
1577         struct vm_area_struct *vma = vmf->vma;
1578         struct inode *inode = file_inode(vma->vm_file);
1579         struct ceph_inode_info *ci = ceph_inode(inode);
1580         struct ceph_file_info *fi = vma->vm_file->private_data;
1581         loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1582         int want, got, err;
1583         sigset_t oldset;
1584         vm_fault_t ret = VM_FAULT_SIGBUS;
1585
1586         if (ceph_inode_is_shutdown(inode))
1587                 return ret;
1588
1589         ceph_block_sigs(&oldset);
1590
1591         dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1592              inode, ceph_vinop(inode), off);
1593         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1594                 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1595         else
1596                 want = CEPH_CAP_FILE_CACHE;
1597
1598         got = 0;
1599         err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1600         if (err < 0)
1601                 goto out_restore;
1602
1603         dout("filemap_fault %p %llu got cap refs on %s\n",
1604              inode, off, ceph_cap_string(got));
1605
1606         if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1607             !ceph_has_inline_data(ci)) {
1608                 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1609                 ceph_add_rw_context(fi, &rw_ctx);
1610                 ret = filemap_fault(vmf);
1611                 ceph_del_rw_context(fi, &rw_ctx);
1612                 dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1613                      inode, off, ceph_cap_string(got), ret);
1614         } else
1615                 err = -EAGAIN;
1616
1617         ceph_put_cap_refs(ci, got);
1618
1619         if (err != -EAGAIN)
1620                 goto out_restore;
1621
1622         /* read inline data */
1623         if (off >= PAGE_SIZE) {
1624                 /* does not support inline data > PAGE_SIZE */
1625                 ret = VM_FAULT_SIGBUS;
1626         } else {
1627                 struct address_space *mapping = inode->i_mapping;
1628                 struct page *page;
1629
1630                 filemap_invalidate_lock_shared(mapping);
1631                 page = find_or_create_page(mapping, 0,
1632                                 mapping_gfp_constraint(mapping, ~__GFP_FS));
1633                 if (!page) {
1634                         ret = VM_FAULT_OOM;
1635                         goto out_inline;
1636                 }
1637                 err = __ceph_do_getattr(inode, page,
1638                                          CEPH_STAT_CAP_INLINE_DATA, true);
1639                 if (err < 0 || off >= i_size_read(inode)) {
1640                         unlock_page(page);
1641                         put_page(page);
1642                         ret = vmf_error(err);
1643                         goto out_inline;
1644                 }
1645                 if (err < PAGE_SIZE)
1646                         zero_user_segment(page, err, PAGE_SIZE);
1647                 else
1648                         flush_dcache_page(page);
1649                 SetPageUptodate(page);
1650                 vmf->page = page;
1651                 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1652 out_inline:
1653                 filemap_invalidate_unlock_shared(mapping);
1654                 dout("filemap_fault %p %llu read inline data ret %x\n",
1655                      inode, off, ret);
1656         }
1657 out_restore:
1658         ceph_restore_sigs(&oldset);
1659         if (err < 0)
1660                 ret = vmf_error(err);
1661
1662         return ret;
1663 }
1664
1665 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1666 {
1667         struct vm_area_struct *vma = vmf->vma;
1668         struct inode *inode = file_inode(vma->vm_file);
1669         struct ceph_inode_info *ci = ceph_inode(inode);
1670         struct ceph_file_info *fi = vma->vm_file->private_data;
1671         struct ceph_cap_flush *prealloc_cf;
1672         struct page *page = vmf->page;
1673         loff_t off = page_offset(page);
1674         loff_t size = i_size_read(inode);
1675         size_t len;
1676         int want, got, err;
1677         sigset_t oldset;
1678         vm_fault_t ret = VM_FAULT_SIGBUS;
1679
1680         if (ceph_inode_is_shutdown(inode))
1681                 return ret;
1682
1683         prealloc_cf = ceph_alloc_cap_flush();
1684         if (!prealloc_cf)
1685                 return VM_FAULT_OOM;
1686
1687         sb_start_pagefault(inode->i_sb);
1688         ceph_block_sigs(&oldset);
1689
1690         if (off + thp_size(page) <= size)
1691                 len = thp_size(page);
1692         else
1693                 len = offset_in_thp(page, size);
1694
1695         dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1696              inode, ceph_vinop(inode), off, len, size);
1697         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1698                 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1699         else
1700                 want = CEPH_CAP_FILE_BUFFER;
1701
1702         got = 0;
1703         err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1704         if (err < 0)
1705                 goto out_free;
1706
1707         dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1708              inode, off, len, ceph_cap_string(got));
1709
1710         /* Update time before taking page lock */
1711         file_update_time(vma->vm_file);
1712         inode_inc_iversion_raw(inode);
1713
1714         do {
1715                 struct ceph_snap_context *snapc;
1716
1717                 lock_page(page);
1718
1719                 if (page_mkwrite_check_truncate(page, inode) < 0) {
1720                         unlock_page(page);
1721                         ret = VM_FAULT_NOPAGE;
1722                         break;
1723                 }
1724
1725                 snapc = ceph_find_incompatible(page);
1726                 if (!snapc) {
1727                         /* success.  we'll keep the page locked. */
1728                         set_page_dirty(page);
1729                         ret = VM_FAULT_LOCKED;
1730                         break;
1731                 }
1732
1733                 unlock_page(page);
1734
1735                 if (IS_ERR(snapc)) {
1736                         ret = VM_FAULT_SIGBUS;
1737                         break;
1738                 }
1739
1740                 ceph_queue_writeback(inode);
1741                 err = wait_event_killable(ci->i_cap_wq,
1742                                 context_is_writeable_or_written(inode, snapc));
1743                 ceph_put_snap_context(snapc);
1744         } while (err == 0);
1745
1746         if (ret == VM_FAULT_LOCKED) {
1747                 int dirty;
1748                 spin_lock(&ci->i_ceph_lock);
1749                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1750                                                &prealloc_cf);
1751                 spin_unlock(&ci->i_ceph_lock);
1752                 if (dirty)
1753                         __mark_inode_dirty(inode, dirty);
1754         }
1755
1756         dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1757              inode, off, len, ceph_cap_string(got), ret);
1758         ceph_put_cap_refs_async(ci, got);
1759 out_free:
1760         ceph_restore_sigs(&oldset);
1761         sb_end_pagefault(inode->i_sb);
1762         ceph_free_cap_flush(prealloc_cf);
1763         if (err < 0)
1764                 ret = vmf_error(err);
1765         return ret;
1766 }
1767
1768 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1769                            char *data, size_t len)
1770 {
1771         struct address_space *mapping = inode->i_mapping;
1772         struct page *page;
1773
1774         if (locked_page) {
1775                 page = locked_page;
1776         } else {
1777                 if (i_size_read(inode) == 0)
1778                         return;
1779                 page = find_or_create_page(mapping, 0,
1780                                            mapping_gfp_constraint(mapping,
1781                                            ~__GFP_FS));
1782                 if (!page)
1783                         return;
1784                 if (PageUptodate(page)) {
1785                         unlock_page(page);
1786                         put_page(page);
1787                         return;
1788                 }
1789         }
1790
1791         dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1792              inode, ceph_vinop(inode), len, locked_page);
1793
1794         if (len > 0) {
1795                 void *kaddr = kmap_atomic(page);
1796                 memcpy(kaddr, data, len);
1797                 kunmap_atomic(kaddr);
1798         }
1799
1800         if (page != locked_page) {
1801                 if (len < PAGE_SIZE)
1802                         zero_user_segment(page, len, PAGE_SIZE);
1803                 else
1804                         flush_dcache_page(page);
1805
1806                 SetPageUptodate(page);
1807                 unlock_page(page);
1808                 put_page(page);
1809         }
1810 }
1811
1812 int ceph_uninline_data(struct file *file)
1813 {
1814         struct inode *inode = file_inode(file);
1815         struct ceph_inode_info *ci = ceph_inode(inode);
1816         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1817         struct ceph_osd_request *req = NULL;
1818         struct ceph_cap_flush *prealloc_cf = NULL;
1819         struct folio *folio = NULL;
1820         u64 inline_version = CEPH_INLINE_NONE;
1821         struct page *pages[1];
1822         int err = 0;
1823         u64 len;
1824
1825         spin_lock(&ci->i_ceph_lock);
1826         inline_version = ci->i_inline_version;
1827         spin_unlock(&ci->i_ceph_lock);
1828
1829         dout("uninline_data %p %llx.%llx inline_version %llu\n",
1830              inode, ceph_vinop(inode), inline_version);
1831
1832         if (ceph_inode_is_shutdown(inode)) {
1833                 err = -EIO;
1834                 goto out;
1835         }
1836
1837         if (inline_version == CEPH_INLINE_NONE)
1838                 return 0;
1839
1840         prealloc_cf = ceph_alloc_cap_flush();
1841         if (!prealloc_cf)
1842                 return -ENOMEM;
1843
1844         if (inline_version == 1) /* initial version, no data */
1845                 goto out_uninline;
1846
1847         folio = read_mapping_folio(inode->i_mapping, 0, file);
1848         if (IS_ERR(folio)) {
1849                 err = PTR_ERR(folio);
1850                 goto out;
1851         }
1852
1853         folio_lock(folio);
1854
1855         len = i_size_read(inode);
1856         if (len > folio_size(folio))
1857                 len = folio_size(folio);
1858
1859         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1860                                     ceph_vino(inode), 0, &len, 0, 1,
1861                                     CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1862                                     NULL, 0, 0, false);
1863         if (IS_ERR(req)) {
1864                 err = PTR_ERR(req);
1865                 goto out_unlock;
1866         }
1867
1868         req->r_mtime = inode->i_mtime;
1869         ceph_osdc_start_request(&fsc->client->osdc, req);
1870         err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1871         ceph_osdc_put_request(req);
1872         if (err < 0)
1873                 goto out_unlock;
1874
1875         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1876                                     ceph_vino(inode), 0, &len, 1, 3,
1877                                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1878                                     NULL, ci->i_truncate_seq,
1879                                     ci->i_truncate_size, false);
1880         if (IS_ERR(req)) {
1881                 err = PTR_ERR(req);
1882                 goto out_unlock;
1883         }
1884
1885         pages[0] = folio_page(folio, 0);
1886         osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false);
1887
1888         {
1889                 __le64 xattr_buf = cpu_to_le64(inline_version);
1890                 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1891                                             "inline_version", &xattr_buf,
1892                                             sizeof(xattr_buf),
1893                                             CEPH_OSD_CMPXATTR_OP_GT,
1894                                             CEPH_OSD_CMPXATTR_MODE_U64);
1895                 if (err)
1896                         goto out_put_req;
1897         }
1898
1899         {
1900                 char xattr_buf[32];
1901                 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1902                                          "%llu", inline_version);
1903                 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1904                                             "inline_version",
1905                                             xattr_buf, xattr_len, 0, 0);
1906                 if (err)
1907                         goto out_put_req;
1908         }
1909
1910         req->r_mtime = inode->i_mtime;
1911         ceph_osdc_start_request(&fsc->client->osdc, req);
1912         err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1913
1914         ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1915                                   req->r_end_latency, len, err);
1916
1917 out_uninline:
1918         if (!err) {
1919                 int dirty;
1920
1921                 /* Set to CAP_INLINE_NONE and dirty the caps */
1922                 down_read(&fsc->mdsc->snap_rwsem);
1923                 spin_lock(&ci->i_ceph_lock);
1924                 ci->i_inline_version = CEPH_INLINE_NONE;
1925                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf);
1926                 spin_unlock(&ci->i_ceph_lock);
1927                 up_read(&fsc->mdsc->snap_rwsem);
1928                 if (dirty)
1929                         __mark_inode_dirty(inode, dirty);
1930         }
1931 out_put_req:
1932         ceph_osdc_put_request(req);
1933         if (err == -ECANCELED)
1934                 err = 0;
1935 out_unlock:
1936         if (folio) {
1937                 folio_unlock(folio);
1938                 folio_put(folio);
1939         }
1940 out:
1941         ceph_free_cap_flush(prealloc_cf);
1942         dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1943              inode, ceph_vinop(inode), inline_version, err);
1944         return err;
1945 }
1946
1947 static const struct vm_operations_struct ceph_vmops = {
1948         .fault          = ceph_filemap_fault,
1949         .page_mkwrite   = ceph_page_mkwrite,
1950 };
1951
1952 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1953 {
1954         struct address_space *mapping = file->f_mapping;
1955
1956         if (!mapping->a_ops->read_folio)
1957                 return -ENOEXEC;
1958         vma->vm_ops = &ceph_vmops;
1959         return 0;
1960 }
1961
1962 enum {
1963         POOL_READ       = 1,
1964         POOL_WRITE      = 2,
1965 };
1966
1967 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1968                                 s64 pool, struct ceph_string *pool_ns)
1969 {
1970         struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->netfs.inode);
1971         struct ceph_mds_client *mdsc = fsc->mdsc;
1972         struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1973         struct rb_node **p, *parent;
1974         struct ceph_pool_perm *perm;
1975         struct page **pages;
1976         size_t pool_ns_len;
1977         int err = 0, err2 = 0, have = 0;
1978
1979         down_read(&mdsc->pool_perm_rwsem);
1980         p = &mdsc->pool_perm_tree.rb_node;
1981         while (*p) {
1982                 perm = rb_entry(*p, struct ceph_pool_perm, node);
1983                 if (pool < perm->pool)
1984                         p = &(*p)->rb_left;
1985                 else if (pool > perm->pool)
1986                         p = &(*p)->rb_right;
1987                 else {
1988                         int ret = ceph_compare_string(pool_ns,
1989                                                 perm->pool_ns,
1990                                                 perm->pool_ns_len);
1991                         if (ret < 0)
1992                                 p = &(*p)->rb_left;
1993                         else if (ret > 0)
1994                                 p = &(*p)->rb_right;
1995                         else {
1996                                 have = perm->perm;
1997                                 break;
1998                         }
1999                 }
2000         }
2001         up_read(&mdsc->pool_perm_rwsem);
2002         if (*p)
2003                 goto out;
2004
2005         if (pool_ns)
2006                 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
2007                      pool, (int)pool_ns->len, pool_ns->str);
2008         else
2009                 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
2010
2011         down_write(&mdsc->pool_perm_rwsem);
2012         p = &mdsc->pool_perm_tree.rb_node;
2013         parent = NULL;
2014         while (*p) {
2015                 parent = *p;
2016                 perm = rb_entry(parent, struct ceph_pool_perm, node);
2017                 if (pool < perm->pool)
2018                         p = &(*p)->rb_left;
2019                 else if (pool > perm->pool)
2020                         p = &(*p)->rb_right;
2021                 else {
2022                         int ret = ceph_compare_string(pool_ns,
2023                                                 perm->pool_ns,
2024                                                 perm->pool_ns_len);
2025                         if (ret < 0)
2026                                 p = &(*p)->rb_left;
2027                         else if (ret > 0)
2028                                 p = &(*p)->rb_right;
2029                         else {
2030                                 have = perm->perm;
2031                                 break;
2032                         }
2033                 }
2034         }
2035         if (*p) {
2036                 up_write(&mdsc->pool_perm_rwsem);
2037                 goto out;
2038         }
2039
2040         rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
2041                                          1, false, GFP_NOFS);
2042         if (!rd_req) {
2043                 err = -ENOMEM;
2044                 goto out_unlock;
2045         }
2046
2047         rd_req->r_flags = CEPH_OSD_FLAG_READ;
2048         osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
2049         rd_req->r_base_oloc.pool = pool;
2050         if (pool_ns)
2051                 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
2052         ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
2053
2054         err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
2055         if (err)
2056                 goto out_unlock;
2057
2058         wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
2059                                          1, false, GFP_NOFS);
2060         if (!wr_req) {
2061                 err = -ENOMEM;
2062                 goto out_unlock;
2063         }
2064
2065         wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
2066         osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
2067         ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
2068         ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
2069
2070         err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
2071         if (err)
2072                 goto out_unlock;
2073
2074         /* one page should be large enough for STAT data */
2075         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
2076         if (IS_ERR(pages)) {
2077                 err = PTR_ERR(pages);
2078                 goto out_unlock;
2079         }
2080
2081         osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
2082                                      0, false, true);
2083         ceph_osdc_start_request(&fsc->client->osdc, rd_req);
2084
2085         wr_req->r_mtime = ci->netfs.inode.i_mtime;
2086         ceph_osdc_start_request(&fsc->client->osdc, wr_req);
2087
2088         err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
2089         err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
2090
2091         if (err >= 0 || err == -ENOENT)
2092                 have |= POOL_READ;
2093         else if (err != -EPERM) {
2094                 if (err == -EBLOCKLISTED)
2095                         fsc->blocklisted = true;
2096                 goto out_unlock;
2097         }
2098
2099         if (err2 == 0 || err2 == -EEXIST)
2100                 have |= POOL_WRITE;
2101         else if (err2 != -EPERM) {
2102                 if (err2 == -EBLOCKLISTED)
2103                         fsc->blocklisted = true;
2104                 err = err2;
2105                 goto out_unlock;
2106         }
2107
2108         pool_ns_len = pool_ns ? pool_ns->len : 0;
2109         perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
2110         if (!perm) {
2111                 err = -ENOMEM;
2112                 goto out_unlock;
2113         }
2114
2115         perm->pool = pool;
2116         perm->perm = have;
2117         perm->pool_ns_len = pool_ns_len;
2118         if (pool_ns_len > 0)
2119                 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
2120         perm->pool_ns[pool_ns_len] = 0;
2121
2122         rb_link_node(&perm->node, parent, p);
2123         rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
2124         err = 0;
2125 out_unlock:
2126         up_write(&mdsc->pool_perm_rwsem);
2127
2128         ceph_osdc_put_request(rd_req);
2129         ceph_osdc_put_request(wr_req);
2130 out:
2131         if (!err)
2132                 err = have;
2133         if (pool_ns)
2134                 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
2135                      pool, (int)pool_ns->len, pool_ns->str, err);
2136         else
2137                 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
2138         return err;
2139 }
2140
2141 int ceph_pool_perm_check(struct inode *inode, int need)
2142 {
2143         struct ceph_inode_info *ci = ceph_inode(inode);
2144         struct ceph_string *pool_ns;
2145         s64 pool;
2146         int ret, flags;
2147
2148         /* Only need to do this for regular files */
2149         if (!S_ISREG(inode->i_mode))
2150                 return 0;
2151
2152         if (ci->i_vino.snap != CEPH_NOSNAP) {
2153                 /*
2154                  * Pool permission check needs to write to the first object.
2155                  * But for snapshot, head of the first object may have alread
2156                  * been deleted. Skip check to avoid creating orphan object.
2157                  */
2158                 return 0;
2159         }
2160
2161         if (ceph_test_mount_opt(ceph_inode_to_client(inode),
2162                                 NOPOOLPERM))
2163                 return 0;
2164
2165         spin_lock(&ci->i_ceph_lock);
2166         flags = ci->i_ceph_flags;
2167         pool = ci->i_layout.pool_id;
2168         spin_unlock(&ci->i_ceph_lock);
2169 check:
2170         if (flags & CEPH_I_POOL_PERM) {
2171                 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2172                         dout("ceph_pool_perm_check pool %lld no read perm\n",
2173                              pool);
2174                         return -EPERM;
2175                 }
2176                 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2177                         dout("ceph_pool_perm_check pool %lld no write perm\n",
2178                              pool);
2179                         return -EPERM;
2180                 }
2181                 return 0;
2182         }
2183
2184         pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2185         ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2186         ceph_put_string(pool_ns);
2187         if (ret < 0)
2188                 return ret;
2189
2190         flags = CEPH_I_POOL_PERM;
2191         if (ret & POOL_READ)
2192                 flags |= CEPH_I_POOL_RD;
2193         if (ret & POOL_WRITE)
2194                 flags |= CEPH_I_POOL_WR;
2195
2196         spin_lock(&ci->i_ceph_lock);
2197         if (pool == ci->i_layout.pool_id &&
2198             pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2199                 ci->i_ceph_flags |= flags;
2200         } else {
2201                 pool = ci->i_layout.pool_id;
2202                 flags = ci->i_ceph_flags;
2203         }
2204         spin_unlock(&ci->i_ceph_lock);
2205         goto check;
2206 }
2207
2208 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2209 {
2210         struct ceph_pool_perm *perm;
2211         struct rb_node *n;
2212
2213         while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2214                 n = rb_first(&mdsc->pool_perm_tree);
2215                 perm = rb_entry(n, struct ceph_pool_perm, node);
2216                 rb_erase(n, &mdsc->pool_perm_tree);
2217                 kfree(perm);
2218         }
2219 }