On Wed, 2020-08-19 at 15:28 +0200, Ilya Dryomov wrote: > On Wed, Aug 19, 2020 at 1:25 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > On Wed, 2020-08-19 at 09:17 +0200, Ilya Dryomov wrote: > > > On Tue, Aug 18, 2020 at 9:49 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > > > Tuan and Ulrich mentioned that they were hitting a problem on s390x, > > > > which has a 32-bit ino_t value, even though it's a 64-bit arch (for > > > > historical reasons). > > > > > > > > I think the current handling of inode numbers in the ceph driver is > > > > wrong. It tries to use 32-bit inode numbers on 32-bit arches, but that's > > > > actually not a problem. 32-bit arches can deal with 64-bit inode numbers > > > > just fine when userland code is compiled with LFS support (the common > > > > case these days). > > > > > > > > What we really want to do is just use 64-bit numbers everywhere, unless > > > > someone has mounted with the ino32 mount option. In that case, we want > > > > to ensure that we hash the inode number down to something that will fit > > > > in 32 bits before presenting the value to userland. > > > > > > > > Add new helper functions that do this, and only do the conversion before > > > > presenting these values to userland in getattr, readdir, debugfs, and > > > > (some) debug log messages. > > > > > > > > The inode table hashvalue is changed to just cast the inode number to > > > > unsigned long, as low-order bits are the most likely to vary anyway. > > > > > > > > While it's not strictly required, we do want to put something in > > > > inode->i_ino. Instead of basing it on BITS_PER_LONG, however, base it on > > > > the size of the ino_t type. > > > > > > > > Reported-by: Tuan Hoang1 <Tuan.Hoang1@xxxxxxx> > > > > Reported-by: Ulrich Weigand <Ulrich.Weigand@xxxxxxxxxx> > > > > URL: https://tracker.ceph.com/issues/46828 > > > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx> > > > > --- > > > > fs/ceph/caps.c | 14 +++++----- > > > > fs/ceph/debugfs.c | 4 +-- > > > > fs/ceph/dir.c | 31 +++++++++------------- > > > > fs/ceph/file.c | 4 +-- > > > > fs/ceph/inode.c | 18 +++++++------ > > > > fs/ceph/mds_client.h | 2 +- > > > > fs/ceph/quota.c | 4 +-- > > > > fs/ceph/super.h | 63 +++++++++++++++++++------------------------- > > > > 8 files changed, 64 insertions(+), 76 deletions(-) > > > > > > > > v2: > > > > - fix dir_emit inode number for ".." > > > > - fix ino_t size test > > > > > > > > diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c > > > > index 55ccccf77cea..d5b1f781f398 100644 > > > > --- a/fs/ceph/caps.c > > > > +++ b/fs/ceph/caps.c > > > > @@ -887,8 +887,8 @@ int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) > > > > int have = ci->i_snap_caps; > > > > > > > > if ((have & mask) == mask) { > > > > - dout("__ceph_caps_issued_mask ino 0x%lx snap issued %s" > > > > - " (mask %s)\n", ci->vfs_inode.i_ino, > > > > + dout("__ceph_caps_issued_mask ino 0x%llx snap issued %s" > > > > + " (mask %s)\n", ceph_present_inode(&ci->vfs_inode), > > > > > > Hi Jeff, > > > > > > I agree with Zheng. douts should output the real 64-bit values. > > > If the presentation value is of interest in some places, they should > > > output it alongside the real value. > > > > > > > Ok. > > > > FWIW, the rationale for using ceph_present_inode in these is that the > > values would better match what stat() reports in st_ino. The debug > > messages are really inconsistent here though, so I'm fine with moving > > them to always print the 64-bit values. My guess is that hardly anyone > > uses -o ino32 in this day and age anyway... > > > > > > > > ceph_cap_string(have), > > > > ceph_cap_string(mask)); > > > > return 1; > > > > @@ -899,8 +899,8 @@ int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) > > > > if (!__cap_is_valid(cap)) > > > > continue; > > > > if ((cap->issued & mask) == mask) { > > > > - dout("__ceph_caps_issued_mask ino 0x%lx cap %p issued %s" > > > > - " (mask %s)\n", ci->vfs_inode.i_ino, cap, > > > > + dout("__ceph_caps_issued_mask ino 0x%llx cap %p issued %s" > > > > + " (mask %s)\n", ceph_present_inode(&ci->vfs_inode), cap, > > > > ceph_cap_string(cap->issued), > > > > ceph_cap_string(mask)); > > > > if (touch) > > > > @@ -911,8 +911,8 @@ int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) > > > > /* does a combination of caps satisfy mask? */ > > > > have |= cap->issued; > > > > if ((have & mask) == mask) { > > > > - dout("__ceph_caps_issued_mask ino 0x%lx combo issued %s" > > > > - " (mask %s)\n", ci->vfs_inode.i_ino, > > > > + dout("__ceph_caps_issued_mask ino 0x%llx combo issued %s" > > > > + " (mask %s)\n", ceph_present_inode(&ci->vfs_inode), > > > > ceph_cap_string(cap->issued), > > > > ceph_cap_string(mask)); > > > > if (touch) { > > > > @@ -2872,7 +2872,7 @@ int ceph_get_caps(struct file *filp, int need, int want, > > > > struct cap_wait cw; > > > > DEFINE_WAIT_FUNC(wait, woken_wake_function); > > > > > > > > - cw.ino = inode->i_ino; > > > > + cw.ino = ceph_present_inode(inode); > > > > cw.tgid = current->tgid; > > > > cw.need = need; > > > > cw.want = want; > > > > diff --git a/fs/ceph/debugfs.c b/fs/ceph/debugfs.c > > > > index 20043382d825..5eeca7987717 100644 > > > > --- a/fs/ceph/debugfs.c > > > > +++ b/fs/ceph/debugfs.c > > > > @@ -211,7 +211,7 @@ static int caps_show_cb(struct inode *inode, struct ceph_cap *cap, void *p) > > > > { > > > > struct seq_file *s = p; > > > > > > > > - seq_printf(s, "0x%-17lx%-17s%-17s\n", inode->i_ino, > > > > + seq_printf(s, "0x%-17llx%-17s%-17s\n", ceph_present_inode(inode), > > > > ceph_cap_string(cap->issued), > > > > ceph_cap_string(cap->implemented)); > > > > return 0; > > > > @@ -256,7 +256,7 @@ static int caps_show(struct seq_file *s, void *p) > > > > > > > > spin_lock(&mdsc->caps_list_lock); > > > > list_for_each_entry(cw, &mdsc->cap_wait_list, list) { > > > > - seq_printf(s, "%-13d0x%-17lx%-17s%-17s\n", cw->tgid, cw->ino, > > > > + seq_printf(s, "%-13d0x%-17llx%-17s%-17s\n", cw->tgid, cw->ino, > > > > ceph_cap_string(cw->need), > > > > ceph_cap_string(cw->want)); > > > > } > > > > diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c > > > > index 060bdcc5ce32..1a61a435a1cd 100644 > > > > --- a/fs/ceph/dir.c > > > > +++ b/fs/ceph/dir.c > > > > @@ -259,9 +259,7 @@ static int __dcache_readdir(struct file *file, struct dir_context *ctx, > > > > dentry, dentry, d_inode(dentry)); > > > > ctx->pos = di->offset; > > > > if (!dir_emit(ctx, dentry->d_name.name, > > > > - dentry->d_name.len, > > > > - ceph_translate_ino(dentry->d_sb, > > > > - d_inode(dentry)->i_ino), > > > > + dentry->d_name.len, ceph_present_inode(d_inode(dentry)), > > > > d_inode(dentry)->i_mode >> 12)) { > > > > dput(dentry); > > > > err = 0; > > > > @@ -324,18 +322,21 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) > > > > /* always start with . and .. */ > > > > if (ctx->pos == 0) { > > > > dout("readdir off 0 -> '.'\n"); > > > > - if (!dir_emit(ctx, ".", 1, > > > > - ceph_translate_ino(inode->i_sb, inode->i_ino), > > > > + if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode), > > > > inode->i_mode >> 12)) > > > > return 0; > > > > ctx->pos = 1; > > > > } > > > > if (ctx->pos == 1) { > > > > - ino_t ino = parent_ino(file->f_path.dentry); > > > > + u64 ino; > > > > + struct dentry *dentry = file->f_path.dentry; > > > > + > > > > + spin_lock(&dentry->d_lock); > > > > + ino = ceph_present_inode(dentry->d_parent->d_inode); > > > > + spin_unlock(&dentry->d_lock); > > > > + > > > > dout("readdir off 1 -> '..'\n"); > > > > - if (!dir_emit(ctx, "..", 2, > > > > - ceph_translate_ino(inode->i_sb, ino), > > > > - inode->i_mode >> 12)) > > > > + if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12)) > > > > return 0; > > > > ctx->pos = 2; > > > > } > > > > @@ -507,9 +508,6 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) > > > > } > > > > for (; i < rinfo->dir_nr; i++) { > > > > struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; > > > > - struct ceph_vino vino; > > > > - ino_t ino; > > > > - u32 ftype; > > > > > > > > BUG_ON(rde->offset < ctx->pos); > > > > > > > > @@ -519,13 +517,10 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) > > > > rde->name_len, rde->name, &rde->inode.in); > > > > > > > > BUG_ON(!rde->inode.in); > > > > - ftype = le32_to_cpu(rde->inode.in->mode) >> 12; > > > > - vino.ino = le64_to_cpu(rde->inode.in->ino); > > > > - vino.snap = le64_to_cpu(rde->inode.in->snapid); > > > > - ino = ceph_vino_to_ino(vino); > > > > > > > > if (!dir_emit(ctx, rde->name, rde->name_len, > > > > - ceph_translate_ino(inode->i_sb, ino), ftype)) { > > > > + ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)), > > > > + le32_to_cpu(rde->inode.in->mode) >> 12)) { > > > > dout("filldir stopping us...\n"); > > > > return 0; > > > > } > > > > @@ -1161,7 +1156,7 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry) > > > > > > > > if (try_async && op == CEPH_MDS_OP_UNLINK && > > > > (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) { > > > > - dout("async unlink on %lu/%.*s caps=%s", dir->i_ino, > > > > + dout("async unlink on %llu/%.*s caps=%s", ceph_present_inode(dir), > > > > dentry->d_name.len, dentry->d_name.name, > > > > ceph_cap_string(req->r_dir_caps)); > > > > set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags); > > > > diff --git a/fs/ceph/file.c b/fs/ceph/file.c > > > > index 5fa28a620932..20e0c6f1d8f2 100644 > > > > --- a/fs/ceph/file.c > > > > +++ b/fs/ceph/file.c > > > > @@ -642,8 +642,8 @@ static int ceph_finish_async_create(struct inode *dir, struct dentry *dentry, > > > > } else { > > > > struct dentry *dn; > > > > > > > > - dout("%s d_adding new inode 0x%llx to 0x%lx/%s\n", __func__, > > > > - vino.ino, dir->i_ino, dentry->d_name.name); > > > > + dout("%s d_adding new inode 0x%llx to 0x%llx/%s\n", __func__, > > > > + vino.ino, ceph_ino(dir), dentry->d_name.name); > > > > ceph_dir_clear_ordered(dir); > > > > ceph_init_inode_acls(inode, as_ctx); > > > > if (inode->i_state & I_NEW) { > > > > diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c > > > > index 357c937699d5..8dbb56d2edef 100644 > > > > --- a/fs/ceph/inode.c > > > > +++ b/fs/ceph/inode.c > > > > @@ -41,8 +41,10 @@ static void ceph_inode_work(struct work_struct *work); > > > > */ > > > > static int ceph_set_ino_cb(struct inode *inode, void *data) > > > > { > > > > - ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; > > > > - inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); > > > > + struct ceph_inode_info *ci = ceph_inode(inode); > > > > + > > > > + ci->i_vino = *(struct ceph_vino *)data; > > > > + inode->i_ino = ceph_vino_to_ino(ci->i_vino); > > > > inode_set_iversion_raw(inode, 0); > > > > return 0; > > > > } > > > > @@ -50,17 +52,17 @@ static int ceph_set_ino_cb(struct inode *inode, void *data) > > > > struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) > > > > { > > > > struct inode *inode; > > > > - ino_t t = ceph_vino_to_ino(vino); > > > > > > > > - inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); > > > > + inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare, > > > > + ceph_set_ino_cb, &vino); > > > > if (!inode) > > > > return ERR_PTR(-ENOMEM); > > > > if (inode->i_state & I_NEW) > > > > dout("get_inode created new inode %p %llx.%llx ino %llx\n", > > > > - inode, ceph_vinop(inode), (u64)inode->i_ino); > > > > + inode, ceph_vinop(inode), ceph_present_inode(inode)); > > > > > > > > - dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, > > > > - vino.snap, inode); > > > > + dout("get_inode on %llu=%llx.%llx got %p\n", ceph_present_inode(inode), > > > > + vino.ino, vino.snap, inode); > > > > return inode; > > > > } > > > > > > > > @@ -2378,7 +2380,7 @@ int ceph_getattr(const struct path *path, struct kstat *stat, > > > > } > > > > > > > > generic_fillattr(inode, stat); > > > > - stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); > > > > + stat->ino = ceph_present_inode(inode); > > > > > > > > /* > > > > * btime on newly-allocated inodes is 0, so if this is still set to > > > > diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h > > > > index bc9e95937d7c..658800605bfb 100644 > > > > --- a/fs/ceph/mds_client.h > > > > +++ b/fs/ceph/mds_client.h > > > > @@ -372,7 +372,7 @@ struct ceph_quotarealm_inode { > > > > > > > > struct cap_wait { > > > > struct list_head list; > > > > - unsigned long ino; > > > > + u64 ino; > > > > pid_t tgid; > > > > int need; > > > > int want; > > > > diff --git a/fs/ceph/quota.c b/fs/ceph/quota.c > > > > index 198ddde5c1e6..cc2c4d40b022 100644 > > > > --- a/fs/ceph/quota.c > > > > +++ b/fs/ceph/quota.c > > > > @@ -23,12 +23,12 @@ static inline bool ceph_has_realms_with_quotas(struct inode *inode) > > > > { > > > > struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; > > > > struct super_block *sb = mdsc->fsc->sb; > > > > + struct inode *root = d_inode(sb->s_root); > > > > > > > > if (atomic64_read(&mdsc->quotarealms_count) > 0) > > > > return true; > > > > /* if root is the real CephFS root, we don't have quota realms */ > > > > - if (sb->s_root->d_inode && > > > > - (sb->s_root->d_inode->i_ino == CEPH_INO_ROOT)) > > > > + if (root && ceph_ino(root) == CEPH_INO_ROOT) > > > > return false; > > > > /* otherwise, we can't know for sure */ > > > > return true; > > > > diff --git a/fs/ceph/super.h b/fs/ceph/super.h > > > > index 4c3c964b1c54..0dd272a4a410 100644 > > > > --- a/fs/ceph/super.h > > > > +++ b/fs/ceph/super.h > > > > @@ -457,15 +457,7 @@ ceph_vino(const struct inode *inode) > > > > return ceph_inode(inode)->i_vino; > > > > } > > > > > > > > -/* > > > > - * ino_t is <64 bits on many architectures, blech. > > > > - * > > > > - * i_ino (kernel inode) st_ino (userspace) > > > > - * i386 32 32 > > > > - * x86_64+ino32 64 32 > > > > - * x86_64 64 64 > > > > - */ > > > > -static inline u32 ceph_ino_to_ino32(__u64 vino) > > > > +static inline u32 ceph_ino_to_ino32(u64 vino) > > > > { > > > > u32 ino = vino & 0xffffffff; > > > > ino ^= vino >> 32; > > > > @@ -474,36 +466,13 @@ static inline u32 ceph_ino_to_ino32(__u64 vino) > > > > return ino; > > > > } > > > > > > > > -/* > > > > - * kernel i_ino value > > > > - */ > > > > static inline ino_t ceph_vino_to_ino(struct ceph_vino vino) > > > > { > > > > -#if BITS_PER_LONG == 32 > > > > - return ceph_ino_to_ino32(vino.ino); > > > > -#else > > > > + if (sizeof(ino_t) == sizeof(u32)) > > > > + return ceph_ino_to_ino32(vino.ino); > > > > return (ino_t)vino.ino; > > > > -#endif > > > > } > > > > > > > > -/* > > > > - * user-visible ino (stat, filldir) > > > > - */ > > > > -#if BITS_PER_LONG == 32 > > > > -static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino) > > > > -{ > > > > - return ino; > > > > -} > > > > -#else > > > > -static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino) > > > > -{ > > > > - if (ceph_test_mount_opt(ceph_sb_to_client(sb), INO32)) > > > > - ino = ceph_ino_to_ino32(ino); > > > > - return ino; > > > > -} > > > > -#endif > > > > - > > > > - > > > > /* for printf-style formatting */ > > > > #define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap > > > > > > > > @@ -511,11 +480,34 @@ static inline u64 ceph_ino(struct inode *inode) > > > > { > > > > return ceph_inode(inode)->i_vino.ino; > > > > } > > > > + > > > > static inline u64 ceph_snap(struct inode *inode) > > > > { > > > > return ceph_inode(inode)->i_vino.snap; > > > > } > > > > > > > > +/** > > > > + * ceph_present_ino - format an inode number for presentation to userland > > > > + * @sb: superblock where the inode lives > > > > + * @ino: inode number to (possibly) convert > > > > + * > > > > + * If the user mounted with the ino32 option, then the 64-bit value needs > > > > + * to be converted to something that can fit inside 32 bits. Note that > > > > + * internal kernel code never uses this value, so this is entirely for > > > > + * userland consumption. > > > > + */ > > > > +static inline u64 ceph_present_ino(struct super_block *sb, u64 ino) > > > > +{ > > > > + if (unlikely(ceph_test_mount_opt(ceph_sb_to_client(sb), INO32))) > > > > + return ceph_ino_to_ino32(ino); > > > > + return ino; > > > > +} > > > > + > > > > +static inline u64 ceph_present_inode(struct inode *inode) > > > > +{ > > > > + return ceph_present_ino(inode->i_sb, ceph_ino(inode)); > > > > +} > > > > + > > > > static inline int ceph_ino_compare(struct inode *inode, void *data) > > > > { > > > > struct ceph_vino *pvino = (struct ceph_vino *)data; > > > > @@ -527,8 +519,7 @@ static inline int ceph_ino_compare(struct inode *inode, void *data) > > > > static inline struct inode *ceph_find_inode(struct super_block *sb, > > > > struct ceph_vino vino) > > > > { > > > > - ino_t t = ceph_vino_to_ino(vino); > > > > - return ilookup5(sb, t, ceph_ino_compare, &vino); > > > > + return ilookup5(sb, (unsigned long)vino.ino, ceph_ino_compare, &vino); > > > > > > What is the rationale for this bit? There is a paragraph in the > > > changelog that mentions that "low-order bits are the most likely > > > to vary anyway", but this way we seem to have three sets of values > > > instead of two: the real 64-bit values, the ino_t values and the > > > truncated values. > > > > > > > Yep. It's a bit confusing, but we have: > > > > 1) the real 64-bit values: these are what we generally want to use > > everywhere internally in ceph.ko > > > > 2) the presentation value: the values that get reported in getattr and > > readdir to userland. This is usually the same as the real 64-bit value > > unless you mount with -o ino32. > > > > 3) inode->i_ino: we'll no longer use this value for anything in ceph, > > but we want to put _something_ in there so things like tracepoints that > > display inode->i_ino don't show all 0's. > > > > The second arg to ilookup5 is just the value that the ilookup5/iget5 use > > to determine the hash bucket to use. This value won't ever be used for > > anything beyond that. Most existing filesystems just use the inode > > number for this. > > > > So the rationale is that while we could just use the same value as what > > ends up in inode->i_ino, that would mean that arches like s390x would > > end up passing down a 32-bit value here when they could pass in the full > > 64 bits. 32 bit arches might end up just sending the low-order bits, but > > this gets run through hash() anyway, so I doubt it will harm anything. > > I guess I still I don't see a good reason to special case this for > 64-bit arches with 32-bit ino_t (just s390x?) and I'd keep it as is. > It's just more consistent and less to keep in mind. > > This is just a feeling and not a strong opinion though, ultimately > it doesn't seem to matter too much. If you go ahead with truncating > to BITS_PER_LONG, I'd suggest turning your last paragraph into a > comment for ceph_find_inode(). > I guess it comes down to what makes a better hashval for the inode table on arches with a 32-bit long? Should we just use the lowest bits of the inode number or do a ceph_ino_to_ino32 conversion first? My rationale was that we're going to end up hashing this value through hash() in fs/inode.c anyway, so why bother hashing it down into 32 bits first? Most inodes (at least for a single MDS) end up clustered together so that the upper 32 bits are not generally significant. We might end up with collisions when 2 MDS's generate 2 inode numbers with the lower 32 bits all the same, but that's not a huge problem. The hash chain just ends up being a little longer than is optimal. -- Jeff Layton <jlayton@xxxxxxxxxx>
