pstore: gracefully handle NULL pstore_info functions
PATCH 1/1 pstore: gracefully handle NULL pstore_info functions
If a pstore backend doesn't want to support various portions of the pstore interface, it can just leave those functions NULL instead of creating no-op stubs.
--- fs/pstore/inode.c | 3 ++- fs/pstore/platform.c | 11 ++++++----- 2 files changed, 8 insertions(+), 6 deletions(-)
fs/pstore/inode.c +2 -1
@@ -80,7 +80,8 @@ static int pstore_unlink(struct inode *dir, struct dentry *dentry) {struct pstore_private *p = dentry->d_inode->i_private; - p->psi->erase (p->type, p->id, p->psi); +if (p->psi->erase) + p->psi->erase (p->type, p->id, p->psi);return simple_unlink (dir, dentry); }
fs/pstore/platform.c +6 -5
@@ -178,7 +178,8 @@ int pstore_register(struct pstore_info *psi)if (pstore_is_mounted ())pstore_get_records (0 ); -kmsg_dump_register (&pstore_dumper); +if (psi->write && psi->buf && psi->bufsize) +kmsg_dump_register (&pstore_dumper); pstore_timer.expires = jiffies +PSTORE_INTERVAL ;add_timer (&pstore_timer);
@@ -203,12 +204,11 @@ void pstore_get_records(int quiet)struct timespec time;int failed =0 , rc; -if (!psi) +if (!psi || !psi->read)return ;mutex_lock (&psi->read_mutex); - rc = psi->open (psi); -if (rc) +if (psi->open && psi->open (psi)) goto out;while ((size = psi->read (&id, &type, &time, &buf, psi)) >0 ) {
@@ -219,7 +219,8 @@ void pstore_get_records(int quiet)if (rc && (rc != -EEXIST || !quiet)) failed++; } - psi->close (psi); +if (psi->close) + psi->close (psi); out:mutex_unlock (&psi->read_mutex);
Signature · 1 line
1.7.0.4
Quoted 3 lines · Kees Cook
-
kmsg_dump_register(&pstore_dumper);
+ if (psi->write&&psi->buf&&psi->bufsize)
+kmsg_dump_register(&pstore_dumper);
I can see that you might not need the other parts of the pstore interface if your back-end is very simple (e.g. only has space for a single record). But stub functions are cheap - so it isn't clear who should pay the overhead.
BUT - without psi->write!? What use is it!? I think in an early version I refused the registration with -EINVAL - but akpm convinced me it was a waste of electrons:
Quoted 7 lines · On Wed, 1 Dec 2010 16:51:39 Andrew Morton <[email protected]> wrote:
Review
Quoted 3 lines
(quoted diff) +3
-0+ if (!psi->reader || !psi->writer || !psi->eraser) { + spin_unlock(&pstore_lock); + return -EINVAL;It doesn't seem appropriate to check this here. It's a programming error! Just install the thing and let the kernel oops - the programmer will notice.
Hiding the lack of a ->write function by quietly accepting the registration, but not hooking into kmsg_dump doesn't sound useful in any scenario.
-Tony
Dang - pstore is coming up on a year old - time flies when you're having fun.
Quoted 9 lines · Tony Luck
Quoted 3 lines · Kees Cook
-
kmsg_dump_register(&pstore_dumper);
+ if (psi->write&&psi->buf&&psi->bufsize)
+kmsg_dump_register(&pstore_dumper);I can see that you might not need the other parts of the pstore interface if your back-end is very simple (e.g. only has space for a single record). But stub functions are cheap - so it isn't clear who should pay the overhead.
Okay, fine by me. :) I'll resend with just the open/read/close/erase checked.
-Kees
Signature · 2 lines
Kees Cook ChromeOS Security