Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 #include <linux/ceph/ceph_debug.h>
0003 
0004 #include <linux/backing-dev.h>
0005 #include <linux/fs.h>
0006 #include <linux/mm.h>
0007 #include <linux/swap.h>
0008 #include <linux/pagemap.h>
0009 #include <linux/slab.h>
0010 #include <linux/pagevec.h>
0011 #include <linux/task_io_accounting_ops.h>
0012 #include <linux/signal.h>
0013 #include <linux/iversion.h>
0014 #include <linux/ktime.h>
0015 #include <linux/netfs.h>
0016 
0017 #include "super.h"
0018 #include "mds_client.h"
0019 #include "cache.h"
0020 #include "metric.h"
0021 #include <linux/ceph/osd_client.h>
0022 #include <linux/ceph/striper.h>
0023 
0024 /*
0025  * Ceph address space ops.
0026  *
0027  * There are a few funny things going on here.
0028  *
0029  * The page->private field is used to reference a struct
0030  * ceph_snap_context for _every_ dirty page.  This indicates which
0031  * snapshot the page was logically dirtied in, and thus which snap
0032  * context needs to be associated with the osd write during writeback.
0033  *
0034  * Similarly, struct ceph_inode_info maintains a set of counters to
0035  * count dirty pages on the inode.  In the absence of snapshots,
0036  * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
0037  *
0038  * When a snapshot is taken (that is, when the client receives
0039  * notification that a snapshot was taken), each inode with caps and
0040  * with dirty pages (dirty pages implies there is a cap) gets a new
0041  * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
0042  * order, new snaps go to the tail).  The i_wrbuffer_ref_head count is
0043  * moved to capsnap->dirty. (Unless a sync write is currently in
0044  * progress.  In that case, the capsnap is said to be "pending", new
0045  * writes cannot start, and the capsnap isn't "finalized" until the
0046  * write completes (or fails) and a final size/mtime for the inode for
0047  * that snap can be settled upon.)  i_wrbuffer_ref_head is reset to 0.
0048  *
0049  * On writeback, we must submit writes to the osd IN SNAP ORDER.  So,
0050  * we look for the first capsnap in i_cap_snaps and write out pages in
0051  * that snap context _only_.  Then we move on to the next capsnap,
0052  * eventually reaching the "live" or "head" context (i.e., pages that
0053  * are not yet snapped) and are writing the most recently dirtied
0054  * pages.
0055  *
0056  * Invalidate and so forth must take care to ensure the dirty page
0057  * accounting is preserved.
0058  */
0059 
0060 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
0061 #define CONGESTION_OFF_THRESH(congestion_kb)                \
0062     (CONGESTION_ON_THRESH(congestion_kb) -              \
0063      (CONGESTION_ON_THRESH(congestion_kb) >> 2))
0064 
0065 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
0066                     struct folio **foliop, void **_fsdata);
0067 
0068 static inline struct ceph_snap_context *page_snap_context(struct page *page)
0069 {
0070     if (PagePrivate(page))
0071         return (void *)page->private;
0072     return NULL;
0073 }
0074 
0075 /*
0076  * Dirty a page.  Optimistically adjust accounting, on the assumption
0077  * that we won't race with invalidate.  If we do, readjust.
0078  */
0079 static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
0080 {
0081     struct inode *inode;
0082     struct ceph_inode_info *ci;
0083     struct ceph_snap_context *snapc;
0084 
0085     if (folio_test_dirty(folio)) {
0086         dout("%p dirty_folio %p idx %lu -- already dirty\n",
0087              mapping->host, folio, folio->index);
0088         VM_BUG_ON_FOLIO(!folio_test_private(folio), folio);
0089         return false;
0090     }
0091 
0092     inode = mapping->host;
0093     ci = ceph_inode(inode);
0094 
0095     /* dirty the head */
0096     spin_lock(&ci->i_ceph_lock);
0097     BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
0098     if (__ceph_have_pending_cap_snap(ci)) {
0099         struct ceph_cap_snap *capsnap =
0100                 list_last_entry(&ci->i_cap_snaps,
0101                         struct ceph_cap_snap,
0102                         ci_item);
0103         snapc = ceph_get_snap_context(capsnap->context);
0104         capsnap->dirty_pages++;
0105     } else {
0106         BUG_ON(!ci->i_head_snapc);
0107         snapc = ceph_get_snap_context(ci->i_head_snapc);
0108         ++ci->i_wrbuffer_ref_head;
0109     }
0110     if (ci->i_wrbuffer_ref == 0)
0111         ihold(inode);
0112     ++ci->i_wrbuffer_ref;
0113     dout("%p dirty_folio %p idx %lu head %d/%d -> %d/%d "
0114          "snapc %p seq %lld (%d snaps)\n",
0115          mapping->host, folio, folio->index,
0116          ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
0117          ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
0118          snapc, snapc->seq, snapc->num_snaps);
0119     spin_unlock(&ci->i_ceph_lock);
0120 
0121     /*
0122      * Reference snap context in folio->private.  Also set
0123      * PagePrivate so that we get invalidate_folio callback.
0124      */
0125     VM_WARN_ON_FOLIO(folio->private, folio);
0126     folio_attach_private(folio, snapc);
0127 
0128     return ceph_fscache_dirty_folio(mapping, folio);
0129 }
0130 
0131 /*
0132  * If we are truncating the full folio (i.e. offset == 0), adjust the
0133  * dirty folio counters appropriately.  Only called if there is private
0134  * data on the folio.
0135  */
0136 static void ceph_invalidate_folio(struct folio *folio, size_t offset,
0137                 size_t length)
0138 {
0139     struct inode *inode;
0140     struct ceph_inode_info *ci;
0141     struct ceph_snap_context *snapc;
0142 
0143     inode = folio->mapping->host;
0144     ci = ceph_inode(inode);
0145 
0146     if (offset != 0 || length != folio_size(folio)) {
0147         dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n",
0148              inode, folio->index, offset, length);
0149         return;
0150     }
0151 
0152     WARN_ON(!folio_test_locked(folio));
0153     if (folio_test_private(folio)) {
0154         dout("%p invalidate_folio idx %lu full dirty page\n",
0155              inode, folio->index);
0156 
0157         snapc = folio_detach_private(folio);
0158         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
0159         ceph_put_snap_context(snapc);
0160     }
0161 
0162     folio_wait_fscache(folio);
0163 }
0164 
0165 static bool ceph_release_folio(struct folio *folio, gfp_t gfp)
0166 {
0167     struct inode *inode = folio->mapping->host;
0168 
0169     dout("%llx:%llx release_folio idx %lu (%sdirty)\n",
0170          ceph_vinop(inode),
0171          folio->index, folio_test_dirty(folio) ? "" : "not ");
0172 
0173     if (folio_test_private(folio))
0174         return false;
0175 
0176     if (folio_test_fscache(folio)) {
0177         if (current_is_kswapd() || !(gfp & __GFP_FS))
0178             return false;
0179         folio_wait_fscache(folio);
0180     }
0181     ceph_fscache_note_page_release(inode);
0182     return true;
0183 }
0184 
0185 static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq)
0186 {
0187     struct inode *inode = rreq->inode;
0188     struct ceph_inode_info *ci = ceph_inode(inode);
0189     struct ceph_file_layout *lo = &ci->i_layout;
0190     u32 blockoff;
0191     u64 blockno;
0192 
0193     /* Expand the start downward */
0194     blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
0195     rreq->start = blockno * lo->stripe_unit;
0196     rreq->len += blockoff;
0197 
0198     /* Now, round up the length to the next block */
0199     rreq->len = roundup(rreq->len, lo->stripe_unit);
0200 }
0201 
0202 static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq)
0203 {
0204     struct inode *inode = subreq->rreq->inode;
0205     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
0206     struct ceph_inode_info *ci = ceph_inode(inode);
0207     u64 objno, objoff;
0208     u32 xlen;
0209 
0210     /* Truncate the extent at the end of the current block */
0211     ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
0212                       &objno, &objoff, &xlen);
0213     subreq->len = min(xlen, fsc->mount_options->rsize);
0214     return true;
0215 }
0216 
0217 static void finish_netfs_read(struct ceph_osd_request *req)
0218 {
0219     struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
0220     struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
0221     struct netfs_io_subrequest *subreq = req->r_priv;
0222     int num_pages;
0223     int err = req->r_result;
0224 
0225     ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
0226                  req->r_end_latency, osd_data->length, err);
0227 
0228     dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
0229          subreq->len, i_size_read(req->r_inode));
0230 
0231     /* no object means success but no data */
0232     if (err == -ENOENT)
0233         err = 0;
0234     else if (err == -EBLOCKLISTED)
0235         fsc->blocklisted = true;
0236 
0237     if (err >= 0 && err < subreq->len)
0238         __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
0239 
0240     netfs_subreq_terminated(subreq, err, false);
0241 
0242     num_pages = calc_pages_for(osd_data->alignment, osd_data->length);
0243     ceph_put_page_vector(osd_data->pages, num_pages, false);
0244     iput(req->r_inode);
0245 }
0246 
0247 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
0248 {
0249     struct netfs_io_request *rreq = subreq->rreq;
0250     struct inode *inode = rreq->inode;
0251     struct ceph_mds_reply_info_parsed *rinfo;
0252     struct ceph_mds_reply_info_in *iinfo;
0253     struct ceph_mds_request *req;
0254     struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
0255     struct ceph_inode_info *ci = ceph_inode(inode);
0256     struct iov_iter iter;
0257     ssize_t err = 0;
0258     size_t len;
0259     int mode;
0260 
0261     __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
0262     __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
0263 
0264     if (subreq->start >= inode->i_size)
0265         goto out;
0266 
0267     /* We need to fetch the inline data. */
0268     mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA);
0269     req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
0270     if (IS_ERR(req)) {
0271         err = PTR_ERR(req);
0272         goto out;
0273     }
0274     req->r_ino1 = ci->i_vino;
0275     req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA);
0276     req->r_num_caps = 2;
0277 
0278     err = ceph_mdsc_do_request(mdsc, NULL, req);
0279     if (err < 0)
0280         goto out;
0281 
0282     rinfo = &req->r_reply_info;
0283     iinfo = &rinfo->targeti;
0284     if (iinfo->inline_version == CEPH_INLINE_NONE) {
0285         /* The data got uninlined */
0286         ceph_mdsc_put_request(req);
0287         return false;
0288     }
0289 
0290     len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len);
0291     iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
0292     err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter);
0293     if (err == 0)
0294         err = -EFAULT;
0295 
0296     ceph_mdsc_put_request(req);
0297 out:
0298     netfs_subreq_terminated(subreq, err, false);
0299     return true;
0300 }
0301 
0302 static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq)
0303 {
0304     struct netfs_io_request *rreq = subreq->rreq;
0305     struct inode *inode = rreq->inode;
0306     struct ceph_inode_info *ci = ceph_inode(inode);
0307     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
0308     struct ceph_osd_request *req;
0309     struct ceph_vino vino = ceph_vino(inode);
0310     struct iov_iter iter;
0311     struct page **pages;
0312     size_t page_off;
0313     int err = 0;
0314     u64 len = subreq->len;
0315 
0316     if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
0317         return;
0318 
0319     req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
0320             0, 1, CEPH_OSD_OP_READ,
0321             CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
0322             NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
0323     if (IS_ERR(req)) {
0324         err = PTR_ERR(req);
0325         req = NULL;
0326         goto out;
0327     }
0328 
0329     dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
0330     iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
0331     err = iov_iter_get_pages_alloc2(&iter, &pages, len, &page_off);
0332     if (err < 0) {
0333         dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err);
0334         goto out;
0335     }
0336 
0337     /* should always give us a page-aligned read */
0338     WARN_ON_ONCE(page_off);
0339     len = err;
0340     err = 0;
0341 
0342     osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
0343     req->r_callback = finish_netfs_read;
0344     req->r_priv = subreq;
0345     req->r_inode = inode;
0346     ihold(inode);
0347 
0348     ceph_osdc_start_request(req->r_osdc, req);
0349 out:
0350     ceph_osdc_put_request(req);
0351     if (err)
0352         netfs_subreq_terminated(subreq, err, false);
0353     dout("%s: result %d\n", __func__, err);
0354 }
0355 
0356 static int ceph_init_request(struct netfs_io_request *rreq, struct file *file)
0357 {
0358     struct inode *inode = rreq->inode;
0359     int got = 0, want = CEPH_CAP_FILE_CACHE;
0360     int ret = 0;
0361 
0362     if (rreq->origin != NETFS_READAHEAD)
0363         return 0;
0364 
0365     if (file) {
0366         struct ceph_rw_context *rw_ctx;
0367         struct ceph_file_info *fi = file->private_data;
0368 
0369         rw_ctx = ceph_find_rw_context(fi);
0370         if (rw_ctx)
0371             return 0;
0372     }
0373 
0374     /*
0375      * readahead callers do not necessarily hold Fcb caps
0376      * (e.g. fadvise, madvise).
0377      */
0378     ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
0379     if (ret < 0) {
0380         dout("start_read %p, error getting cap\n", inode);
0381         return ret;
0382     }
0383 
0384     if (!(got & want)) {
0385         dout("start_read %p, no cache cap\n", inode);
0386         return -EACCES;
0387     }
0388     if (ret == 0)
0389         return -EACCES;
0390 
0391     rreq->netfs_priv = (void *)(uintptr_t)got;
0392     return 0;
0393 }
0394 
0395 static void ceph_netfs_free_request(struct netfs_io_request *rreq)
0396 {
0397     struct ceph_inode_info *ci = ceph_inode(rreq->inode);
0398     int got = (uintptr_t)rreq->netfs_priv;
0399 
0400     if (got)
0401         ceph_put_cap_refs(ci, got);
0402 }
0403 
0404 const struct netfs_request_ops ceph_netfs_ops = {
0405     .init_request       = ceph_init_request,
0406     .free_request       = ceph_netfs_free_request,
0407     .begin_cache_operation  = ceph_begin_cache_operation,
0408     .issue_read     = ceph_netfs_issue_read,
0409     .expand_readahead   = ceph_netfs_expand_readahead,
0410     .clamp_length       = ceph_netfs_clamp_length,
0411     .check_write_begin  = ceph_netfs_check_write_begin,
0412 };
0413 
0414 #ifdef CONFIG_CEPH_FSCACHE
0415 static void ceph_set_page_fscache(struct page *page)
0416 {
0417     set_page_fscache(page);
0418 }
0419 
0420 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async)
0421 {
0422     struct inode *inode = priv;
0423 
0424     if (IS_ERR_VALUE(error) && error != -ENOBUFS)
0425         ceph_fscache_invalidate(inode, false);
0426 }
0427 
0428 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
0429 {
0430     struct ceph_inode_info *ci = ceph_inode(inode);
0431     struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
0432 
0433     fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
0434                    ceph_fscache_write_terminated, inode, caching);
0435 }
0436 #else
0437 static inline void ceph_set_page_fscache(struct page *page)
0438 {
0439 }
0440 
0441 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
0442 {
0443 }
0444 #endif /* CONFIG_CEPH_FSCACHE */
0445 
0446 struct ceph_writeback_ctl
0447 {
0448     loff_t i_size;
0449     u64 truncate_size;
0450     u32 truncate_seq;
0451     bool size_stable;
0452     bool head_snapc;
0453 };
0454 
0455 /*
0456  * Get ref for the oldest snapc for an inode with dirty data... that is, the
0457  * only snap context we are allowed to write back.
0458  */
0459 static struct ceph_snap_context *
0460 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
0461            struct ceph_snap_context *page_snapc)
0462 {
0463     struct ceph_inode_info *ci = ceph_inode(inode);
0464     struct ceph_snap_context *snapc = NULL;
0465     struct ceph_cap_snap *capsnap = NULL;
0466 
0467     spin_lock(&ci->i_ceph_lock);
0468     list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
0469         dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
0470              capsnap->context, capsnap->dirty_pages);
0471         if (!capsnap->dirty_pages)
0472             continue;
0473 
0474         /* get i_size, truncate_{seq,size} for page_snapc? */
0475         if (snapc && capsnap->context != page_snapc)
0476             continue;
0477 
0478         if (ctl) {
0479             if (capsnap->writing) {
0480                 ctl->i_size = i_size_read(inode);
0481                 ctl->size_stable = false;
0482             } else {
0483                 ctl->i_size = capsnap->size;
0484                 ctl->size_stable = true;
0485             }
0486             ctl->truncate_size = capsnap->truncate_size;
0487             ctl->truncate_seq = capsnap->truncate_seq;
0488             ctl->head_snapc = false;
0489         }
0490 
0491         if (snapc)
0492             break;
0493 
0494         snapc = ceph_get_snap_context(capsnap->context);
0495         if (!page_snapc ||
0496             page_snapc == snapc ||
0497             page_snapc->seq > snapc->seq)
0498             break;
0499     }
0500     if (!snapc && ci->i_wrbuffer_ref_head) {
0501         snapc = ceph_get_snap_context(ci->i_head_snapc);
0502         dout(" head snapc %p has %d dirty pages\n",
0503              snapc, ci->i_wrbuffer_ref_head);
0504         if (ctl) {
0505             ctl->i_size = i_size_read(inode);
0506             ctl->truncate_size = ci->i_truncate_size;
0507             ctl->truncate_seq = ci->i_truncate_seq;
0508             ctl->size_stable = false;
0509             ctl->head_snapc = true;
0510         }
0511     }
0512     spin_unlock(&ci->i_ceph_lock);
0513     return snapc;
0514 }
0515 
0516 static u64 get_writepages_data_length(struct inode *inode,
0517                       struct page *page, u64 start)
0518 {
0519     struct ceph_inode_info *ci = ceph_inode(inode);
0520     struct ceph_snap_context *snapc = page_snap_context(page);
0521     struct ceph_cap_snap *capsnap = NULL;
0522     u64 end = i_size_read(inode);
0523 
0524     if (snapc != ci->i_head_snapc) {
0525         bool found = false;
0526         spin_lock(&ci->i_ceph_lock);
0527         list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
0528             if (capsnap->context == snapc) {
0529                 if (!capsnap->writing)
0530                     end = capsnap->size;
0531                 found = true;
0532                 break;
0533             }
0534         }
0535         spin_unlock(&ci->i_ceph_lock);
0536         WARN_ON(!found);
0537     }
0538     if (end > page_offset(page) + thp_size(page))
0539         end = page_offset(page) + thp_size(page);
0540     return end > start ? end - start : 0;
0541 }
0542 
0543 /*
0544  * Write a single page, but leave the page locked.
0545  *
0546  * If we get a write error, mark the mapping for error, but still adjust the
0547  * dirty page accounting (i.e., page is no longer dirty).
0548  */
0549 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
0550 {
0551     struct folio *folio = page_folio(page);
0552     struct inode *inode = page->mapping->host;
0553     struct ceph_inode_info *ci = ceph_inode(inode);
0554     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
0555     struct ceph_snap_context *snapc, *oldest;
0556     loff_t page_off = page_offset(page);
0557     int err;
0558     loff_t len = thp_size(page);
0559     struct ceph_writeback_ctl ceph_wbc;
0560     struct ceph_osd_client *osdc = &fsc->client->osdc;
0561     struct ceph_osd_request *req;
0562     bool caching = ceph_is_cache_enabled(inode);
0563 
0564     dout("writepage %p idx %lu\n", page, page->index);
0565 
0566     /* verify this is a writeable snap context */
0567     snapc = page_snap_context(page);
0568     if (!snapc) {
0569         dout("writepage %p page %p not dirty?\n", inode, page);
0570         return 0;
0571     }
0572     oldest = get_oldest_context(inode, &ceph_wbc, snapc);
0573     if (snapc->seq > oldest->seq) {
0574         dout("writepage %p page %p snapc %p not writeable - noop\n",
0575              inode, page, snapc);
0576         /* we should only noop if called by kswapd */
0577         WARN_ON(!(current->flags & PF_MEMALLOC));
0578         ceph_put_snap_context(oldest);
0579         redirty_page_for_writepage(wbc, page);
0580         return 0;
0581     }
0582     ceph_put_snap_context(oldest);
0583 
0584     /* is this a partial page at end of file? */
0585     if (page_off >= ceph_wbc.i_size) {
0586         dout("folio at %lu beyond eof %llu\n", folio->index,
0587                 ceph_wbc.i_size);
0588         folio_invalidate(folio, 0, folio_size(folio));
0589         return 0;
0590     }
0591 
0592     if (ceph_wbc.i_size < page_off + len)
0593         len = ceph_wbc.i_size - page_off;
0594 
0595     dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
0596          inode, page, page->index, page_off, len, snapc, snapc->seq);
0597 
0598     if (atomic_long_inc_return(&fsc->writeback_count) >
0599         CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
0600         fsc->write_congested = true;
0601 
0602     req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
0603                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
0604                     ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
0605                     true);
0606     if (IS_ERR(req)) {
0607         redirty_page_for_writepage(wbc, page);
0608         return PTR_ERR(req);
0609     }
0610 
0611     set_page_writeback(page);
0612     if (caching)
0613         ceph_set_page_fscache(page);
0614     ceph_fscache_write_to_cache(inode, page_off, len, caching);
0615 
0616     /* it may be a short write due to an object boundary */
0617     WARN_ON_ONCE(len > thp_size(page));
0618     osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
0619     dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
0620 
0621     req->r_mtime = inode->i_mtime;
0622     ceph_osdc_start_request(osdc, req);
0623     err = ceph_osdc_wait_request(osdc, req);
0624 
0625     ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
0626                   req->r_end_latency, len, err);
0627 
0628     ceph_osdc_put_request(req);
0629     if (err == 0)
0630         err = len;
0631 
0632     if (err < 0) {
0633         struct writeback_control tmp_wbc;
0634         if (!wbc)
0635             wbc = &tmp_wbc;
0636         if (err == -ERESTARTSYS) {
0637             /* killed by SIGKILL */
0638             dout("writepage interrupted page %p\n", page);
0639             redirty_page_for_writepage(wbc, page);
0640             end_page_writeback(page);
0641             return err;
0642         }
0643         if (err == -EBLOCKLISTED)
0644             fsc->blocklisted = true;
0645         dout("writepage setting page/mapping error %d %p\n",
0646              err, page);
0647         mapping_set_error(&inode->i_data, err);
0648         wbc->pages_skipped++;
0649     } else {
0650         dout("writepage cleaned page %p\n", page);
0651         err = 0;  /* vfs expects us to return 0 */
0652     }
0653     oldest = detach_page_private(page);
0654     WARN_ON_ONCE(oldest != snapc);
0655     end_page_writeback(page);
0656     ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
0657     ceph_put_snap_context(snapc);  /* page's reference */
0658 
0659     if (atomic_long_dec_return(&fsc->writeback_count) <
0660         CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
0661         fsc->write_congested = false;
0662 
0663     return err;
0664 }
0665 
0666 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
0667 {
0668     int err;
0669     struct inode *inode = page->mapping->host;
0670     BUG_ON(!inode);
0671     ihold(inode);
0672 
0673     if (wbc->sync_mode == WB_SYNC_NONE &&
0674         ceph_inode_to_client(inode)->write_congested)
0675         return AOP_WRITEPAGE_ACTIVATE;
0676 
0677     wait_on_page_fscache(page);
0678 
0679     err = writepage_nounlock(page, wbc);
0680     if (err == -ERESTARTSYS) {
0681         /* direct memory reclaimer was killed by SIGKILL. return 0
0682          * to prevent caller from setting mapping/page error */
0683         err = 0;
0684     }
0685     unlock_page(page);
0686     iput(inode);
0687     return err;
0688 }
0689 
0690 /*
0691  * async writeback completion handler.
0692  *
0693  * If we get an error, set the mapping error bit, but not the individual
0694  * page error bits.
0695  */
0696 static void writepages_finish(struct ceph_osd_request *req)
0697 {
0698     struct inode *inode = req->r_inode;
0699     struct ceph_inode_info *ci = ceph_inode(inode);
0700     struct ceph_osd_data *osd_data;
0701     struct page *page;
0702     int num_pages, total_pages = 0;
0703     int i, j;
0704     int rc = req->r_result;
0705     struct ceph_snap_context *snapc = req->r_snapc;
0706     struct address_space *mapping = inode->i_mapping;
0707     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
0708     unsigned int len = 0;
0709     bool remove_page;
0710 
0711     dout("writepages_finish %p rc %d\n", inode, rc);
0712     if (rc < 0) {
0713         mapping_set_error(mapping, rc);
0714         ceph_set_error_write(ci);
0715         if (rc == -EBLOCKLISTED)
0716             fsc->blocklisted = true;
0717     } else {
0718         ceph_clear_error_write(ci);
0719     }
0720 
0721     /*
0722      * We lost the cache cap, need to truncate the page before
0723      * it is unlocked, otherwise we'd truncate it later in the
0724      * page truncation thread, possibly losing some data that
0725      * raced its way in
0726      */
0727     remove_page = !(ceph_caps_issued(ci) &
0728             (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
0729 
0730     /* clean all pages */
0731     for (i = 0; i < req->r_num_ops; i++) {
0732         if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) {
0733             pr_warn("%s incorrect op %d req %p index %d tid %llu\n",
0734                 __func__, req->r_ops[i].op, req, i, req->r_tid);
0735             break;
0736         }
0737 
0738         osd_data = osd_req_op_extent_osd_data(req, i);
0739         BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
0740         len += osd_data->length;
0741         num_pages = calc_pages_for((u64)osd_data->alignment,
0742                        (u64)osd_data->length);
0743         total_pages += num_pages;
0744         for (j = 0; j < num_pages; j++) {
0745             page = osd_data->pages[j];
0746             BUG_ON(!page);
0747             WARN_ON(!PageUptodate(page));
0748 
0749             if (atomic_long_dec_return(&fsc->writeback_count) <
0750                  CONGESTION_OFF_THRESH(
0751                     fsc->mount_options->congestion_kb))
0752                 fsc->write_congested = false;
0753 
0754             ceph_put_snap_context(detach_page_private(page));
0755             end_page_writeback(page);
0756             dout("unlocking %p\n", page);
0757 
0758             if (remove_page)
0759                 generic_error_remove_page(inode->i_mapping,
0760                               page);
0761 
0762             unlock_page(page);
0763         }
0764         dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
0765              inode, osd_data->length, rc >= 0 ? num_pages : 0);
0766 
0767         release_pages(osd_data->pages, num_pages);
0768     }
0769 
0770     ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
0771                   req->r_end_latency, len, rc);
0772 
0773     ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
0774 
0775     osd_data = osd_req_op_extent_osd_data(req, 0);
0776     if (osd_data->pages_from_pool)
0777         mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
0778     else
0779         kfree(osd_data->pages);
0780     ceph_osdc_put_request(req);
0781 }
0782 
0783 /*
0784  * initiate async writeback
0785  */
0786 static int ceph_writepages_start(struct address_space *mapping,
0787                  struct writeback_control *wbc)
0788 {
0789     struct inode *inode = mapping->host;
0790     struct ceph_inode_info *ci = ceph_inode(inode);
0791     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
0792     struct ceph_vino vino = ceph_vino(inode);
0793     pgoff_t index, start_index, end = -1;
0794     struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
0795     struct pagevec pvec;
0796     int rc = 0;
0797     unsigned int wsize = i_blocksize(inode);
0798     struct ceph_osd_request *req = NULL;
0799     struct ceph_writeback_ctl ceph_wbc;
0800     bool should_loop, range_whole = false;
0801     bool done = false;
0802     bool caching = ceph_is_cache_enabled(inode);
0803 
0804     if (wbc->sync_mode == WB_SYNC_NONE &&
0805         fsc->write_congested)
0806         return 0;
0807 
0808     dout("writepages_start %p (mode=%s)\n", inode,
0809          wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
0810          (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
0811 
0812     if (ceph_inode_is_shutdown(inode)) {
0813         if (ci->i_wrbuffer_ref > 0) {
0814             pr_warn_ratelimited(
0815                 "writepage_start %p %lld forced umount\n",
0816                 inode, ceph_ino(inode));
0817         }
0818         mapping_set_error(mapping, -EIO);
0819         return -EIO; /* we're in a forced umount, don't write! */
0820     }
0821     if (fsc->mount_options->wsize < wsize)
0822         wsize = fsc->mount_options->wsize;
0823 
0824     pagevec_init(&pvec);
0825 
0826     start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
0827     index = start_index;
0828 
0829 retry:
0830     /* find oldest snap context with dirty data */
0831     snapc = get_oldest_context(inode, &ceph_wbc, NULL);
0832     if (!snapc) {
0833         /* hmm, why does writepages get called when there
0834            is no dirty data? */
0835         dout(" no snap context with dirty data?\n");
0836         goto out;
0837     }
0838     dout(" oldest snapc is %p seq %lld (%d snaps)\n",
0839          snapc, snapc->seq, snapc->num_snaps);
0840 
0841     should_loop = false;
0842     if (ceph_wbc.head_snapc && snapc != last_snapc) {
0843         /* where to start/end? */
0844         if (wbc->range_cyclic) {
0845             index = start_index;
0846             end = -1;
0847             if (index > 0)
0848                 should_loop = true;
0849             dout(" cyclic, start at %lu\n", index);
0850         } else {
0851             index = wbc->range_start >> PAGE_SHIFT;
0852             end = wbc->range_end >> PAGE_SHIFT;
0853             if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
0854                 range_whole = true;
0855             dout(" not cyclic, %lu to %lu\n", index, end);
0856         }
0857     } else if (!ceph_wbc.head_snapc) {
0858         /* Do not respect wbc->range_{start,end}. Dirty pages
0859          * in that range can be associated with newer snapc.
0860          * They are not writeable until we write all dirty pages
0861          * associated with 'snapc' get written */
0862         if (index > 0)
0863             should_loop = true;
0864         dout(" non-head snapc, range whole\n");
0865     }
0866 
0867     ceph_put_snap_context(last_snapc);
0868     last_snapc = snapc;
0869 
0870     while (!done && index <= end) {
0871         int num_ops = 0, op_idx;
0872         unsigned i, pvec_pages, max_pages, locked_pages = 0;
0873         struct page **pages = NULL, **data_pages;
0874         struct page *page;
0875         pgoff_t strip_unit_end = 0;
0876         u64 offset = 0, len = 0;
0877         bool from_pool = false;
0878 
0879         max_pages = wsize >> PAGE_SHIFT;
0880 
0881 get_more_pages:
0882         pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
0883                         end, PAGECACHE_TAG_DIRTY);
0884         dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
0885         if (!pvec_pages && !locked_pages)
0886             break;
0887         for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
0888             page = pvec.pages[i];
0889             dout("? %p idx %lu\n", page, page->index);
0890             if (locked_pages == 0)
0891                 lock_page(page);  /* first page */
0892             else if (!trylock_page(page))
0893                 break;
0894 
0895             /* only dirty pages, or our accounting breaks */
0896             if (unlikely(!PageDirty(page)) ||
0897                 unlikely(page->mapping != mapping)) {
0898                 dout("!dirty or !mapping %p\n", page);
0899                 unlock_page(page);
0900                 continue;
0901             }
0902             /* only if matching snap context */
0903             pgsnapc = page_snap_context(page);
0904             if (pgsnapc != snapc) {
0905                 dout("page snapc %p %lld != oldest %p %lld\n",
0906                      pgsnapc, pgsnapc->seq, snapc, snapc->seq);
0907                 if (!should_loop &&
0908                     !ceph_wbc.head_snapc &&
0909                     wbc->sync_mode != WB_SYNC_NONE)
0910                     should_loop = true;
0911                 unlock_page(page);
0912                 continue;
0913             }
0914             if (page_offset(page) >= ceph_wbc.i_size) {
0915                 struct folio *folio = page_folio(page);
0916 
0917                 dout("folio at %lu beyond eof %llu\n",
0918                      folio->index, ceph_wbc.i_size);
0919                 if ((ceph_wbc.size_stable ||
0920                     folio_pos(folio) >= i_size_read(inode)) &&
0921                     folio_clear_dirty_for_io(folio))
0922                     folio_invalidate(folio, 0,
0923                             folio_size(folio));
0924                 folio_unlock(folio);
0925                 continue;
0926             }
0927             if (strip_unit_end && (page->index > strip_unit_end)) {
0928                 dout("end of strip unit %p\n", page);
0929                 unlock_page(page);
0930                 break;
0931             }
0932             if (PageWriteback(page) || PageFsCache(page)) {
0933                 if (wbc->sync_mode == WB_SYNC_NONE) {
0934                     dout("%p under writeback\n", page);
0935                     unlock_page(page);
0936                     continue;
0937                 }
0938                 dout("waiting on writeback %p\n", page);
0939                 wait_on_page_writeback(page);
0940                 wait_on_page_fscache(page);
0941             }
0942 
0943             if (!clear_page_dirty_for_io(page)) {
0944                 dout("%p !clear_page_dirty_for_io\n", page);
0945                 unlock_page(page);
0946                 continue;
0947             }
0948 
0949             /*
0950              * We have something to write.  If this is
0951              * the first locked page this time through,
0952              * calculate max possinle write size and
0953              * allocate a page array
0954              */
0955             if (locked_pages == 0) {
0956                 u64 objnum;
0957                 u64 objoff;
0958                 u32 xlen;
0959 
0960                 /* prepare async write request */
0961                 offset = (u64)page_offset(page);
0962                 ceph_calc_file_object_mapping(&ci->i_layout,
0963                                   offset, wsize,
0964                                   &objnum, &objoff,
0965                                   &xlen);
0966                 len = xlen;
0967 
0968                 num_ops = 1;
0969                 strip_unit_end = page->index +
0970                     ((len - 1) >> PAGE_SHIFT);
0971 
0972                 BUG_ON(pages);
0973                 max_pages = calc_pages_for(0, (u64)len);
0974                 pages = kmalloc_array(max_pages,
0975                               sizeof(*pages),
0976                               GFP_NOFS);
0977                 if (!pages) {
0978                     from_pool = true;
0979                     pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
0980                     BUG_ON(!pages);
0981                 }
0982 
0983                 len = 0;
0984             } else if (page->index !=
0985                    (offset + len) >> PAGE_SHIFT) {
0986                 if (num_ops >= (from_pool ?  CEPH_OSD_SLAB_OPS :
0987                                  CEPH_OSD_MAX_OPS)) {
0988                     redirty_page_for_writepage(wbc, page);
0989                     unlock_page(page);
0990                     break;
0991                 }
0992 
0993                 num_ops++;
0994                 offset = (u64)page_offset(page);
0995                 len = 0;
0996             }
0997 
0998             /* note position of first page in pvec */
0999             dout("%p will write page %p idx %lu\n",
1000                  inode, page, page->index);
1001 
1002             if (atomic_long_inc_return(&fsc->writeback_count) >
1003                 CONGESTION_ON_THRESH(
1004                     fsc->mount_options->congestion_kb))
1005                 fsc->write_congested = true;
1006 
1007             pages[locked_pages++] = page;
1008             pvec.pages[i] = NULL;
1009 
1010             len += thp_size(page);
1011         }
1012 
1013         /* did we get anything? */
1014         if (!locked_pages)
1015             goto release_pvec_pages;
1016         if (i) {
1017             unsigned j, n = 0;
1018             /* shift unused page to beginning of pvec */
1019             for (j = 0; j < pvec_pages; j++) {
1020                 if (!pvec.pages[j])
1021                     continue;
1022                 if (n < j)
1023                     pvec.pages[n] = pvec.pages[j];
1024                 n++;
1025             }
1026             pvec.nr = n;
1027 
1028             if (pvec_pages && i == pvec_pages &&
1029                 locked_pages < max_pages) {
1030                 dout("reached end pvec, trying for more\n");
1031                 pagevec_release(&pvec);
1032                 goto get_more_pages;
1033             }
1034         }
1035 
1036 new_request:
1037         offset = page_offset(pages[0]);
1038         len = wsize;
1039 
1040         req = ceph_osdc_new_request(&fsc->client->osdc,
1041                     &ci->i_layout, vino,
1042                     offset, &len, 0, num_ops,
1043                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1044                     snapc, ceph_wbc.truncate_seq,
1045                     ceph_wbc.truncate_size, false);
1046         if (IS_ERR(req)) {
1047             req = ceph_osdc_new_request(&fsc->client->osdc,
1048                         &ci->i_layout, vino,
1049                         offset, &len, 0,
1050                         min(num_ops,
1051                             CEPH_OSD_SLAB_OPS),
1052                         CEPH_OSD_OP_WRITE,
1053                         CEPH_OSD_FLAG_WRITE,
1054                         snapc, ceph_wbc.truncate_seq,
1055                         ceph_wbc.truncate_size, true);
1056             BUG_ON(IS_ERR(req));
1057         }
1058         BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1059                  thp_size(page) - offset);
1060 
1061         req->r_callback = writepages_finish;
1062         req->r_inode = inode;
1063 
1064         /* Format the osd request message and submit the write */
1065         len = 0;
1066         data_pages = pages;
1067         op_idx = 0;
1068         for (i = 0; i < locked_pages; i++) {
1069             u64 cur_offset = page_offset(pages[i]);
1070             /*
1071              * Discontinuity in page range? Ceph can handle that by just passing
1072              * multiple extents in the write op.
1073              */
1074             if (offset + len != cur_offset) {
1075                 /* If it's full, stop here */
1076                 if (op_idx + 1 == req->r_num_ops)
1077                     break;
1078 
1079                 /* Kick off an fscache write with what we have so far. */
1080                 ceph_fscache_write_to_cache(inode, offset, len, caching);
1081 
1082                 /* Start a new extent */
1083                 osd_req_op_extent_dup_last(req, op_idx,
1084                                cur_offset - offset);
1085                 dout("writepages got pages at %llu~%llu\n",
1086                      offset, len);
1087                 osd_req_op_extent_osd_data_pages(req, op_idx,
1088                             data_pages, len, 0,
1089                             from_pool, false);
1090                 osd_req_op_extent_update(req, op_idx, len);
1091 
1092                 len = 0;
1093                 offset = cur_offset;
1094                 data_pages = pages + i;
1095                 op_idx++;
1096             }
1097 
1098             set_page_writeback(pages[i]);
1099             if (caching)
1100                 ceph_set_page_fscache(pages[i]);
1101             len += thp_size(page);
1102         }
1103         ceph_fscache_write_to_cache(inode, offset, len, caching);
1104 
1105         if (ceph_wbc.size_stable) {
1106             len = min(len, ceph_wbc.i_size - offset);
1107         } else if (i == locked_pages) {
1108             /* writepages_finish() clears writeback pages
1109              * according to the data length, so make sure
1110              * data length covers all locked pages */
1111             u64 min_len = len + 1 - thp_size(page);
1112             len = get_writepages_data_length(inode, pages[i - 1],
1113                              offset);
1114             len = max(len, min_len);
1115         }
1116         dout("writepages got pages at %llu~%llu\n", offset, len);
1117 
1118         osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1119                          0, from_pool, false);
1120         osd_req_op_extent_update(req, op_idx, len);
1121 
1122         BUG_ON(op_idx + 1 != req->r_num_ops);
1123 
1124         from_pool = false;
1125         if (i < locked_pages) {
1126             BUG_ON(num_ops <= req->r_num_ops);
1127             num_ops -= req->r_num_ops;
1128             locked_pages -= i;
1129 
1130             /* allocate new pages array for next request */
1131             data_pages = pages;
1132             pages = kmalloc_array(locked_pages, sizeof(*pages),
1133                           GFP_NOFS);
1134             if (!pages) {
1135                 from_pool = true;
1136                 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1137                 BUG_ON(!pages);
1138             }
1139             memcpy(pages, data_pages + i,
1140                    locked_pages * sizeof(*pages));
1141             memset(data_pages + i, 0,
1142                    locked_pages * sizeof(*pages));
1143         } else {
1144             BUG_ON(num_ops != req->r_num_ops);
1145             index = pages[i - 1]->index + 1;
1146             /* request message now owns the pages array */
1147             pages = NULL;
1148         }
1149 
1150         req->r_mtime = inode->i_mtime;
1151         ceph_osdc_start_request(&fsc->client->osdc, req);
1152         req = NULL;
1153 
1154         wbc->nr_to_write -= i;
1155         if (pages)
1156             goto new_request;
1157 
1158         /*
1159          * We stop writing back only if we are not doing
1160          * integrity sync. In case of integrity sync we have to
1161          * keep going until we have written all the pages
1162          * we tagged for writeback prior to entering this loop.
1163          */
1164         if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1165             done = true;
1166 
1167 release_pvec_pages:
1168         dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1169              pvec.nr ? pvec.pages[0] : NULL);
1170         pagevec_release(&pvec);
1171     }
1172 
1173     if (should_loop && !done) {
1174         /* more to do; loop back to beginning of file */
1175         dout("writepages looping back to beginning of file\n");
1176         end = start_index - 1; /* OK even when start_index == 0 */
1177 
1178         /* to write dirty pages associated with next snapc,
1179          * we need to wait until current writes complete */
1180         if (wbc->sync_mode != WB_SYNC_NONE &&
1181             start_index == 0 && /* all dirty pages were checked */
1182             !ceph_wbc.head_snapc) {
1183             struct page *page;
1184             unsigned i, nr;
1185             index = 0;
1186             while ((index <= end) &&
1187                    (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1188                         PAGECACHE_TAG_WRITEBACK))) {
1189                 for (i = 0; i < nr; i++) {
1190                     page = pvec.pages[i];
1191                     if (page_snap_context(page) != snapc)
1192                         continue;
1193                     wait_on_page_writeback(page);
1194                 }
1195                 pagevec_release(&pvec);
1196                 cond_resched();
1197             }
1198         }
1199 
1200         start_index = 0;
1201         index = 0;
1202         goto retry;
1203     }
1204 
1205     if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1206         mapping->writeback_index = index;
1207 
1208 out:
1209     ceph_osdc_put_request(req);
1210     ceph_put_snap_context(last_snapc);
1211     dout("writepages dend - startone, rc = %d\n", rc);
1212     return rc;
1213 }
1214 
1215 
1216 
1217 /*
1218  * See if a given @snapc is either writeable, or already written.
1219  */
1220 static int context_is_writeable_or_written(struct inode *inode,
1221                        struct ceph_snap_context *snapc)
1222 {
1223     struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1224     int ret = !oldest || snapc->seq <= oldest->seq;
1225 
1226     ceph_put_snap_context(oldest);
1227     return ret;
1228 }
1229 
1230 /**
1231  * ceph_find_incompatible - find an incompatible context and return it
1232  * @page: page being dirtied
1233  *
1234  * We are only allowed to write into/dirty a page if the page is
1235  * clean, or already dirty within the same snap context. Returns a
1236  * conflicting context if there is one, NULL if there isn't, or a
1237  * negative error code on other errors.
1238  *
1239  * Must be called with page lock held.
1240  */
1241 static struct ceph_snap_context *
1242 ceph_find_incompatible(struct page *page)
1243 {
1244     struct inode *inode = page->mapping->host;
1245     struct ceph_inode_info *ci = ceph_inode(inode);
1246 
1247     if (ceph_inode_is_shutdown(inode)) {
1248         dout(" page %p %llx:%llx is shutdown\n", page,
1249              ceph_vinop(inode));
1250         return ERR_PTR(-ESTALE);
1251     }
1252 
1253     for (;;) {
1254         struct ceph_snap_context *snapc, *oldest;
1255 
1256         wait_on_page_writeback(page);
1257 
1258         snapc = page_snap_context(page);
1259         if (!snapc || snapc == ci->i_head_snapc)
1260             break;
1261 
1262         /*
1263          * this page is already dirty in another (older) snap
1264          * context!  is it writeable now?
1265          */
1266         oldest = get_oldest_context(inode, NULL, NULL);
1267         if (snapc->seq > oldest->seq) {
1268             /* not writeable -- return it for the caller to deal with */
1269             ceph_put_snap_context(oldest);
1270             dout(" page %p snapc %p not current or oldest\n", page, snapc);
1271             return ceph_get_snap_context(snapc);
1272         }
1273         ceph_put_snap_context(oldest);
1274 
1275         /* yay, writeable, do it now (without dropping page lock) */
1276         dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1277         if (clear_page_dirty_for_io(page)) {
1278             int r = writepage_nounlock(page, NULL);
1279             if (r < 0)
1280                 return ERR_PTR(r);
1281         }
1282     }
1283     return NULL;
1284 }
1285 
1286 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1287                     struct folio **foliop, void **_fsdata)
1288 {
1289     struct inode *inode = file_inode(file);
1290     struct ceph_inode_info *ci = ceph_inode(inode);
1291     struct ceph_snap_context *snapc;
1292 
1293     snapc = ceph_find_incompatible(folio_page(*foliop, 0));
1294     if (snapc) {
1295         int r;
1296 
1297         folio_unlock(*foliop);
1298         folio_put(*foliop);
1299         *foliop = NULL;
1300         if (IS_ERR(snapc))
1301             return PTR_ERR(snapc);
1302 
1303         ceph_queue_writeback(inode);
1304         r = wait_event_killable(ci->i_cap_wq,
1305                     context_is_writeable_or_written(inode, snapc));
1306         ceph_put_snap_context(snapc);
1307         return r == 0 ? -EAGAIN : r;
1308     }
1309     return 0;
1310 }
1311 
1312 /*
1313  * We are only allowed to write into/dirty the page if the page is
1314  * clean, or already dirty within the same snap context.
1315  */
1316 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1317                 loff_t pos, unsigned len,
1318                 struct page **pagep, void **fsdata)
1319 {
1320     struct inode *inode = file_inode(file);
1321     struct ceph_inode_info *ci = ceph_inode(inode);
1322     struct folio *folio = NULL;
1323     int r;
1324 
1325     r = netfs_write_begin(&ci->netfs, file, inode->i_mapping, pos, len, &folio, NULL);
1326     if (r < 0)
1327         return r;
1328 
1329     folio_wait_fscache(folio);
1330     WARN_ON_ONCE(!folio_test_locked(folio));
1331     *pagep = &folio->page;
1332     return 0;
1333 }
1334 
1335 /*
1336  * we don't do anything in here that simple_write_end doesn't do
1337  * except adjust dirty page accounting
1338  */
1339 static int ceph_write_end(struct file *file, struct address_space *mapping,
1340               loff_t pos, unsigned len, unsigned copied,
1341               struct page *subpage, void *fsdata)
1342 {
1343     struct folio *folio = page_folio(subpage);
1344     struct inode *inode = file_inode(file);
1345     bool check_cap = false;
1346 
1347     dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file,
1348          inode, folio, (int)pos, (int)copied, (int)len);
1349 
1350     if (!folio_test_uptodate(folio)) {
1351         /* just return that nothing was copied on a short copy */
1352         if (copied < len) {
1353             copied = 0;
1354             goto out;
1355         }
1356         folio_mark_uptodate(folio);
1357     }
1358 
1359     /* did file size increase? */
1360     if (pos+copied > i_size_read(inode))
1361         check_cap = ceph_inode_set_size(inode, pos+copied);
1362 
1363     folio_mark_dirty(folio);
1364 
1365 out:
1366     folio_unlock(folio);
1367     folio_put(folio);
1368 
1369     if (check_cap)
1370         ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1371 
1372     return copied;
1373 }
1374 
1375 const struct address_space_operations ceph_aops = {
1376     .read_folio = netfs_read_folio,
1377     .readahead = netfs_readahead,
1378     .writepage = ceph_writepage,
1379     .writepages = ceph_writepages_start,
1380     .write_begin = ceph_write_begin,
1381     .write_end = ceph_write_end,
1382     .dirty_folio = ceph_dirty_folio,
1383     .invalidate_folio = ceph_invalidate_folio,
1384     .release_folio = ceph_release_folio,
1385     .direct_IO = noop_direct_IO,
1386 };
1387 
1388 static void ceph_block_sigs(sigset_t *oldset)
1389 {
1390     sigset_t mask;
1391     siginitsetinv(&mask, sigmask(SIGKILL));
1392     sigprocmask(SIG_BLOCK, &mask, oldset);
1393 }
1394 
1395 static void ceph_restore_sigs(sigset_t *oldset)
1396 {
1397     sigprocmask(SIG_SETMASK, oldset, NULL);
1398 }
1399 
1400 /*
1401  * vm ops
1402  */
1403 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1404 {
1405     struct vm_area_struct *vma = vmf->vma;
1406     struct inode *inode = file_inode(vma->vm_file);
1407     struct ceph_inode_info *ci = ceph_inode(inode);
1408     struct ceph_file_info *fi = vma->vm_file->private_data;
1409     loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1410     int want, got, err;
1411     sigset_t oldset;
1412     vm_fault_t ret = VM_FAULT_SIGBUS;
1413 
1414     if (ceph_inode_is_shutdown(inode))
1415         return ret;
1416 
1417     ceph_block_sigs(&oldset);
1418 
1419     dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1420          inode, ceph_vinop(inode), off);
1421     if (fi->fmode & CEPH_FILE_MODE_LAZY)
1422         want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1423     else
1424         want = CEPH_CAP_FILE_CACHE;
1425 
1426     got = 0;
1427     err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1428     if (err < 0)
1429         goto out_restore;
1430 
1431     dout("filemap_fault %p %llu got cap refs on %s\n",
1432          inode, off, ceph_cap_string(got));
1433 
1434     if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1435         !ceph_has_inline_data(ci)) {
1436         CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1437         ceph_add_rw_context(fi, &rw_ctx);
1438         ret = filemap_fault(vmf);
1439         ceph_del_rw_context(fi, &rw_ctx);
1440         dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1441              inode, off, ceph_cap_string(got), ret);
1442     } else
1443         err = -EAGAIN;
1444 
1445     ceph_put_cap_refs(ci, got);
1446 
1447     if (err != -EAGAIN)
1448         goto out_restore;
1449 
1450     /* read inline data */
1451     if (off >= PAGE_SIZE) {
1452         /* does not support inline data > PAGE_SIZE */
1453         ret = VM_FAULT_SIGBUS;
1454     } else {
1455         struct address_space *mapping = inode->i_mapping;
1456         struct page *page;
1457 
1458         filemap_invalidate_lock_shared(mapping);
1459         page = find_or_create_page(mapping, 0,
1460                 mapping_gfp_constraint(mapping, ~__GFP_FS));
1461         if (!page) {
1462             ret = VM_FAULT_OOM;
1463             goto out_inline;
1464         }
1465         err = __ceph_do_getattr(inode, page,
1466                      CEPH_STAT_CAP_INLINE_DATA, true);
1467         if (err < 0 || off >= i_size_read(inode)) {
1468             unlock_page(page);
1469             put_page(page);
1470             ret = vmf_error(err);
1471             goto out_inline;
1472         }
1473         if (err < PAGE_SIZE)
1474             zero_user_segment(page, err, PAGE_SIZE);
1475         else
1476             flush_dcache_page(page);
1477         SetPageUptodate(page);
1478         vmf->page = page;
1479         ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1480 out_inline:
1481         filemap_invalidate_unlock_shared(mapping);
1482         dout("filemap_fault %p %llu read inline data ret %x\n",
1483              inode, off, ret);
1484     }
1485 out_restore:
1486     ceph_restore_sigs(&oldset);
1487     if (err < 0)
1488         ret = vmf_error(err);
1489 
1490     return ret;
1491 }
1492 
1493 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1494 {
1495     struct vm_area_struct *vma = vmf->vma;
1496     struct inode *inode = file_inode(vma->vm_file);
1497     struct ceph_inode_info *ci = ceph_inode(inode);
1498     struct ceph_file_info *fi = vma->vm_file->private_data;
1499     struct ceph_cap_flush *prealloc_cf;
1500     struct page *page = vmf->page;
1501     loff_t off = page_offset(page);
1502     loff_t size = i_size_read(inode);
1503     size_t len;
1504     int want, got, err;
1505     sigset_t oldset;
1506     vm_fault_t ret = VM_FAULT_SIGBUS;
1507 
1508     if (ceph_inode_is_shutdown(inode))
1509         return ret;
1510 
1511     prealloc_cf = ceph_alloc_cap_flush();
1512     if (!prealloc_cf)
1513         return VM_FAULT_OOM;
1514 
1515     sb_start_pagefault(inode->i_sb);
1516     ceph_block_sigs(&oldset);
1517 
1518     if (off + thp_size(page) <= size)
1519         len = thp_size(page);
1520     else
1521         len = offset_in_thp(page, size);
1522 
1523     dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1524          inode, ceph_vinop(inode), off, len, size);
1525     if (fi->fmode & CEPH_FILE_MODE_LAZY)
1526         want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1527     else
1528         want = CEPH_CAP_FILE_BUFFER;
1529 
1530     got = 0;
1531     err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1532     if (err < 0)
1533         goto out_free;
1534 
1535     dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1536          inode, off, len, ceph_cap_string(got));
1537 
1538     /* Update time before taking page lock */
1539     file_update_time(vma->vm_file);
1540     inode_inc_iversion_raw(inode);
1541 
1542     do {
1543         struct ceph_snap_context *snapc;
1544 
1545         lock_page(page);
1546 
1547         if (page_mkwrite_check_truncate(page, inode) < 0) {
1548             unlock_page(page);
1549             ret = VM_FAULT_NOPAGE;
1550             break;
1551         }
1552 
1553         snapc = ceph_find_incompatible(page);
1554         if (!snapc) {
1555             /* success.  we'll keep the page locked. */
1556             set_page_dirty(page);
1557             ret = VM_FAULT_LOCKED;
1558             break;
1559         }
1560 
1561         unlock_page(page);
1562 
1563         if (IS_ERR(snapc)) {
1564             ret = VM_FAULT_SIGBUS;
1565             break;
1566         }
1567 
1568         ceph_queue_writeback(inode);
1569         err = wait_event_killable(ci->i_cap_wq,
1570                 context_is_writeable_or_written(inode, snapc));
1571         ceph_put_snap_context(snapc);
1572     } while (err == 0);
1573 
1574     if (ret == VM_FAULT_LOCKED) {
1575         int dirty;
1576         spin_lock(&ci->i_ceph_lock);
1577         dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1578                            &prealloc_cf);
1579         spin_unlock(&ci->i_ceph_lock);
1580         if (dirty)
1581             __mark_inode_dirty(inode, dirty);
1582     }
1583 
1584     dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1585          inode, off, len, ceph_cap_string(got), ret);
1586     ceph_put_cap_refs_async(ci, got);
1587 out_free:
1588     ceph_restore_sigs(&oldset);
1589     sb_end_pagefault(inode->i_sb);
1590     ceph_free_cap_flush(prealloc_cf);
1591     if (err < 0)
1592         ret = vmf_error(err);
1593     return ret;
1594 }
1595 
1596 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1597                char *data, size_t len)
1598 {
1599     struct address_space *mapping = inode->i_mapping;
1600     struct page *page;
1601 
1602     if (locked_page) {
1603         page = locked_page;
1604     } else {
1605         if (i_size_read(inode) == 0)
1606             return;
1607         page = find_or_create_page(mapping, 0,
1608                        mapping_gfp_constraint(mapping,
1609                        ~__GFP_FS));
1610         if (!page)
1611             return;
1612         if (PageUptodate(page)) {
1613             unlock_page(page);
1614             put_page(page);
1615             return;
1616         }
1617     }
1618 
1619     dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1620          inode, ceph_vinop(inode), len, locked_page);
1621 
1622     if (len > 0) {
1623         void *kaddr = kmap_atomic(page);
1624         memcpy(kaddr, data, len);
1625         kunmap_atomic(kaddr);
1626     }
1627 
1628     if (page != locked_page) {
1629         if (len < PAGE_SIZE)
1630             zero_user_segment(page, len, PAGE_SIZE);
1631         else
1632             flush_dcache_page(page);
1633 
1634         SetPageUptodate(page);
1635         unlock_page(page);
1636         put_page(page);
1637     }
1638 }
1639 
1640 int ceph_uninline_data(struct file *file)
1641 {
1642     struct inode *inode = file_inode(file);
1643     struct ceph_inode_info *ci = ceph_inode(inode);
1644     struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1645     struct ceph_osd_request *req = NULL;
1646     struct ceph_cap_flush *prealloc_cf;
1647     struct folio *folio = NULL;
1648     u64 inline_version = CEPH_INLINE_NONE;
1649     struct page *pages[1];
1650     int err = 0;
1651     u64 len;
1652 
1653     spin_lock(&ci->i_ceph_lock);
1654     inline_version = ci->i_inline_version;
1655     spin_unlock(&ci->i_ceph_lock);
1656 
1657     dout("uninline_data %p %llx.%llx inline_version %llu\n",
1658          inode, ceph_vinop(inode), inline_version);
1659 
1660     if (inline_version == CEPH_INLINE_NONE)
1661         return 0;
1662 
1663     prealloc_cf = ceph_alloc_cap_flush();
1664     if (!prealloc_cf)
1665         return -ENOMEM;
1666 
1667     if (inline_version == 1) /* initial version, no data */
1668         goto out_uninline;
1669 
1670     folio = read_mapping_folio(inode->i_mapping, 0, file);
1671     if (IS_ERR(folio)) {
1672         err = PTR_ERR(folio);
1673         goto out;
1674     }
1675 
1676     folio_lock(folio);
1677 
1678     len = i_size_read(inode);
1679     if (len > folio_size(folio))
1680         len = folio_size(folio);
1681 
1682     req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1683                     ceph_vino(inode), 0, &len, 0, 1,
1684                     CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1685                     NULL, 0, 0, false);
1686     if (IS_ERR(req)) {
1687         err = PTR_ERR(req);
1688         goto out_unlock;
1689     }
1690 
1691     req->r_mtime = inode->i_mtime;
1692     ceph_osdc_start_request(&fsc->client->osdc, req);
1693     err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1694     ceph_osdc_put_request(req);
1695     if (err < 0)
1696         goto out_unlock;
1697 
1698     req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1699                     ceph_vino(inode), 0, &len, 1, 3,
1700                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1701                     NULL, ci->i_truncate_seq,
1702                     ci->i_truncate_size, false);
1703     if (IS_ERR(req)) {
1704         err = PTR_ERR(req);
1705         goto out_unlock;
1706     }
1707 
1708     pages[0] = folio_page(folio, 0);
1709     osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false);
1710 
1711     {
1712         __le64 xattr_buf = cpu_to_le64(inline_version);
1713         err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1714                         "inline_version", &xattr_buf,
1715                         sizeof(xattr_buf),
1716                         CEPH_OSD_CMPXATTR_OP_GT,
1717                         CEPH_OSD_CMPXATTR_MODE_U64);
1718         if (err)
1719             goto out_put_req;
1720     }
1721 
1722     {
1723         char xattr_buf[32];
1724         int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1725                      "%llu", inline_version);
1726         err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1727                         "inline_version",
1728                         xattr_buf, xattr_len, 0, 0);
1729         if (err)
1730             goto out_put_req;
1731     }
1732 
1733     req->r_mtime = inode->i_mtime;
1734     ceph_osdc_start_request(&fsc->client->osdc, req);
1735     err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1736 
1737     ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1738                   req->r_end_latency, len, err);
1739 
1740 out_uninline:
1741     if (!err) {
1742         int dirty;
1743 
1744         /* Set to CAP_INLINE_NONE and dirty the caps */
1745         down_read(&fsc->mdsc->snap_rwsem);
1746         spin_lock(&ci->i_ceph_lock);
1747         ci->i_inline_version = CEPH_INLINE_NONE;
1748         dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf);
1749         spin_unlock(&ci->i_ceph_lock);
1750         up_read(&fsc->mdsc->snap_rwsem);
1751         if (dirty)
1752             __mark_inode_dirty(inode, dirty);
1753     }
1754 out_put_req:
1755     ceph_osdc_put_request(req);
1756     if (err == -ECANCELED)
1757         err = 0;
1758 out_unlock:
1759     if (folio) {
1760         folio_unlock(folio);
1761         folio_put(folio);
1762     }
1763 out:
1764     ceph_free_cap_flush(prealloc_cf);
1765     dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1766          inode, ceph_vinop(inode), inline_version, err);
1767     return err;
1768 }
1769 
1770 static const struct vm_operations_struct ceph_vmops = {
1771     .fault      = ceph_filemap_fault,
1772     .page_mkwrite   = ceph_page_mkwrite,
1773 };
1774 
1775 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1776 {
1777     struct address_space *mapping = file->f_mapping;
1778 
1779     if (!mapping->a_ops->read_folio)
1780         return -ENOEXEC;
1781     vma->vm_ops = &ceph_vmops;
1782     return 0;
1783 }
1784 
1785 enum {
1786     POOL_READ   = 1,
1787     POOL_WRITE  = 2,
1788 };
1789 
1790 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1791                 s64 pool, struct ceph_string *pool_ns)
1792 {
1793     struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->netfs.inode);
1794     struct ceph_mds_client *mdsc = fsc->mdsc;
1795     struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1796     struct rb_node **p, *parent;
1797     struct ceph_pool_perm *perm;
1798     struct page **pages;
1799     size_t pool_ns_len;
1800     int err = 0, err2 = 0, have = 0;
1801 
1802     down_read(&mdsc->pool_perm_rwsem);
1803     p = &mdsc->pool_perm_tree.rb_node;
1804     while (*p) {
1805         perm = rb_entry(*p, struct ceph_pool_perm, node);
1806         if (pool < perm->pool)
1807             p = &(*p)->rb_left;
1808         else if (pool > perm->pool)
1809             p = &(*p)->rb_right;
1810         else {
1811             int ret = ceph_compare_string(pool_ns,
1812                         perm->pool_ns,
1813                         perm->pool_ns_len);
1814             if (ret < 0)
1815                 p = &(*p)->rb_left;
1816             else if (ret > 0)
1817                 p = &(*p)->rb_right;
1818             else {
1819                 have = perm->perm;
1820                 break;
1821             }
1822         }
1823     }
1824     up_read(&mdsc->pool_perm_rwsem);
1825     if (*p)
1826         goto out;
1827 
1828     if (pool_ns)
1829         dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1830              pool, (int)pool_ns->len, pool_ns->str);
1831     else
1832         dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1833 
1834     down_write(&mdsc->pool_perm_rwsem);
1835     p = &mdsc->pool_perm_tree.rb_node;
1836     parent = NULL;
1837     while (*p) {
1838         parent = *p;
1839         perm = rb_entry(parent, struct ceph_pool_perm, node);
1840         if (pool < perm->pool)
1841             p = &(*p)->rb_left;
1842         else if (pool > perm->pool)
1843             p = &(*p)->rb_right;
1844         else {
1845             int ret = ceph_compare_string(pool_ns,
1846                         perm->pool_ns,
1847                         perm->pool_ns_len);
1848             if (ret < 0)
1849                 p = &(*p)->rb_left;
1850             else if (ret > 0)
1851                 p = &(*p)->rb_right;
1852             else {
1853                 have = perm->perm;
1854                 break;
1855             }
1856         }
1857     }
1858     if (*p) {
1859         up_write(&mdsc->pool_perm_rwsem);
1860         goto out;
1861     }
1862 
1863     rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1864                      1, false, GFP_NOFS);
1865     if (!rd_req) {
1866         err = -ENOMEM;
1867         goto out_unlock;
1868     }
1869 
1870     rd_req->r_flags = CEPH_OSD_FLAG_READ;
1871     osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1872     rd_req->r_base_oloc.pool = pool;
1873     if (pool_ns)
1874         rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1875     ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1876 
1877     err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1878     if (err)
1879         goto out_unlock;
1880 
1881     wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1882                      1, false, GFP_NOFS);
1883     if (!wr_req) {
1884         err = -ENOMEM;
1885         goto out_unlock;
1886     }
1887 
1888     wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1889     osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1890     ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1891     ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1892 
1893     err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1894     if (err)
1895         goto out_unlock;
1896 
1897     /* one page should be large enough for STAT data */
1898     pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1899     if (IS_ERR(pages)) {
1900         err = PTR_ERR(pages);
1901         goto out_unlock;
1902     }
1903 
1904     osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1905                      0, false, true);
1906     ceph_osdc_start_request(&fsc->client->osdc, rd_req);
1907 
1908     wr_req->r_mtime = ci->netfs.inode.i_mtime;
1909     ceph_osdc_start_request(&fsc->client->osdc, wr_req);
1910 
1911     err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1912     err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1913 
1914     if (err >= 0 || err == -ENOENT)
1915         have |= POOL_READ;
1916     else if (err != -EPERM) {
1917         if (err == -EBLOCKLISTED)
1918             fsc->blocklisted = true;
1919         goto out_unlock;
1920     }
1921 
1922     if (err2 == 0 || err2 == -EEXIST)
1923         have |= POOL_WRITE;
1924     else if (err2 != -EPERM) {
1925         if (err2 == -EBLOCKLISTED)
1926             fsc->blocklisted = true;
1927         err = err2;
1928         goto out_unlock;
1929     }
1930 
1931     pool_ns_len = pool_ns ? pool_ns->len : 0;
1932     perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1933     if (!perm) {
1934         err = -ENOMEM;
1935         goto out_unlock;
1936     }
1937 
1938     perm->pool = pool;
1939     perm->perm = have;
1940     perm->pool_ns_len = pool_ns_len;
1941     if (pool_ns_len > 0)
1942         memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1943     perm->pool_ns[pool_ns_len] = 0;
1944 
1945     rb_link_node(&perm->node, parent, p);
1946     rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1947     err = 0;
1948 out_unlock:
1949     up_write(&mdsc->pool_perm_rwsem);
1950 
1951     ceph_osdc_put_request(rd_req);
1952     ceph_osdc_put_request(wr_req);
1953 out:
1954     if (!err)
1955         err = have;
1956     if (pool_ns)
1957         dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1958              pool, (int)pool_ns->len, pool_ns->str, err);
1959     else
1960         dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1961     return err;
1962 }
1963 
1964 int ceph_pool_perm_check(struct inode *inode, int need)
1965 {
1966     struct ceph_inode_info *ci = ceph_inode(inode);
1967     struct ceph_string *pool_ns;
1968     s64 pool;
1969     int ret, flags;
1970 
1971     /* Only need to do this for regular files */
1972     if (!S_ISREG(inode->i_mode))
1973         return 0;
1974 
1975     if (ci->i_vino.snap != CEPH_NOSNAP) {
1976         /*
1977          * Pool permission check needs to write to the first object.
1978          * But for snapshot, head of the first object may have alread
1979          * been deleted. Skip check to avoid creating orphan object.
1980          */
1981         return 0;
1982     }
1983 
1984     if (ceph_test_mount_opt(ceph_inode_to_client(inode),
1985                 NOPOOLPERM))
1986         return 0;
1987 
1988     spin_lock(&ci->i_ceph_lock);
1989     flags = ci->i_ceph_flags;
1990     pool = ci->i_layout.pool_id;
1991     spin_unlock(&ci->i_ceph_lock);
1992 check:
1993     if (flags & CEPH_I_POOL_PERM) {
1994         if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1995             dout("ceph_pool_perm_check pool %lld no read perm\n",
1996                  pool);
1997             return -EPERM;
1998         }
1999         if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2000             dout("ceph_pool_perm_check pool %lld no write perm\n",
2001                  pool);
2002             return -EPERM;
2003         }
2004         return 0;
2005     }
2006 
2007     pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2008     ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2009     ceph_put_string(pool_ns);
2010     if (ret < 0)
2011         return ret;
2012 
2013     flags = CEPH_I_POOL_PERM;
2014     if (ret & POOL_READ)
2015         flags |= CEPH_I_POOL_RD;
2016     if (ret & POOL_WRITE)
2017         flags |= CEPH_I_POOL_WR;
2018 
2019     spin_lock(&ci->i_ceph_lock);
2020     if (pool == ci->i_layout.pool_id &&
2021         pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2022         ci->i_ceph_flags |= flags;
2023         } else {
2024         pool = ci->i_layout.pool_id;
2025         flags = ci->i_ceph_flags;
2026     }
2027     spin_unlock(&ci->i_ceph_lock);
2028     goto check;
2029 }
2030 
2031 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2032 {
2033     struct ceph_pool_perm *perm;
2034     struct rb_node *n;
2035 
2036     while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2037         n = rb_first(&mdsc->pool_perm_tree);
2038         perm = rb_entry(n, struct ceph_pool_perm, node);
2039         rb_erase(n, &mdsc->pool_perm_tree);
2040         kfree(perm);
2041     }
2042 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!