int journal_get_undo_access(handle_t *handle, struct buffer_head *bh) { int err; struct journal_head *jh = journal_add_journal_head(bh); char *committed_data = NULL; JBUFFER_TRACE(jh, "entry"); /* * Do this first --- it can drop the journal lock, so we want to * make sure that obtaining the committed_data is done * atomically wrt. completion of any outstanding commits. */ err = do_get_write_access(handle, jh, 1); if (err) goto out; repeat: if (!jh->b_committed_data) { committed_data = jbd_alloc(jh2bh(jh)->b_size, GFP_NOFS); if (!committed_data) { printk(KERN_EMERG "%s: No memory for committed data\n", __func__); err = -ENOMEM; goto out; } } jbd_lock_bh_state(bh); if (!jh->b_committed_data) { /* Copy out the current buffer contents into the * preserved, committed copy. */ JBUFFER_TRACE(jh, "generate b_committed data"); if (!committed_data) { jbd_unlock_bh_state(bh); goto repeat; } jh->b_committed_data = committed_data; committed_data = NULL; memcpy(jh->b_committed_data, bh->b_data, bh->b_size); } jbd_unlock_bh_state(bh); out: journal_put_journal_head(jh); if (unlikely(committed_data)) jbd_free(committed_data, bh->b_size); return err; }
static int do_get_write_access(handle_t *handle, struct journal_head *jh, int force_copy) { struct buffer_head *bh; transaction_t *transaction; journal_t *journal; int error; char *frozen_buffer = NULL; int need_copy = 0; if (is_handle_aborted(handle)) return -EROFS; transaction = handle->h_transaction; journal = transaction->t_journal; jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy); JBUFFER_TRACE(jh, "entry"); repeat: bh = jh2bh(jh); /* @@@ Need to check for errors here at some point. */ lock_buffer(bh); jbd_lock_bh_state(bh); /* We now hold the buffer lock so it is safe to query the buffer * state. Is the buffer dirty? * * If so, there are two possibilities. The buffer may be * non-journaled, and undergoing a quite legitimate writeback. * Otherwise, it is journaled, and we don't expect dirty buffers * in that state (the buffers should be marked JBD_Dirty * instead.) So either the IO is being done under our own * control and this is a bug, or it's a third party IO such as * dump(8) (which may leave the buffer scheduled for read --- * ie. locked but not dirty) or tune2fs (which may actually have * the buffer dirtied, ugh.) */ if (buffer_dirty(bh)) { /* * First question: is this buffer already part of the current * transaction or the existing committing transaction? */ if (jh->b_transaction) { J_ASSERT_JH(jh, jh->b_transaction == transaction || jh->b_transaction == journal->j_committing_transaction); if (jh->b_next_transaction) J_ASSERT_JH(jh, jh->b_next_transaction == transaction); warn_dirty_buffer(bh); } /* * In any case we need to clean the dirty flag and we must * do it under the buffer lock to be sure we don't race * with running write-out. */ JBUFFER_TRACE(jh, "Journalling dirty buffer"); clear_buffer_dirty(bh); set_buffer_jbddirty(bh); } unlock_buffer(bh); error = -EROFS; if (is_handle_aborted(handle)) { jbd_unlock_bh_state(bh); goto out; } error = 0; /* * The buffer is already part of this transaction if b_transaction or * b_next_transaction points to it */ if (jh->b_transaction == transaction || jh->b_next_transaction == transaction) goto done; /* * this is the first time this transaction is touching this buffer, * reset the modified flag */ jh->b_modified = 0; /* * If there is already a copy-out version of this buffer, then we don't * need to make another one */ if (jh->b_frozen_data) { JBUFFER_TRACE(jh, "has frozen data"); J_ASSERT_JH(jh, jh->b_next_transaction == NULL); jh->b_next_transaction = transaction; goto done; } /* Is there data here we need to preserve? */ if (jh->b_transaction && jh->b_transaction != transaction) { JBUFFER_TRACE(jh, "owned by older transaction"); J_ASSERT_JH(jh, jh->b_next_transaction == NULL); J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction); /* There is one case we have to be very careful about. * If the committing transaction is currently writing * this buffer out to disk and has NOT made a copy-out, * then we cannot modify the buffer contents at all * right now. The essence of copy-out is that it is the * extra copy, not the primary copy, which gets * journaled. If the primary copy is already going to * disk then we cannot do copy-out here. */ if (jh->b_jlist == BJ_Shadow) { DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); wait_queue_head_t *wqh; wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); JBUFFER_TRACE(jh, "on shadow: sleep"); jbd_unlock_bh_state(bh); /* commit wakes up all shadow buffers after IO */ for ( ; ; ) { prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); if (jh->b_jlist != BJ_Shadow) break; schedule(); } finish_wait(wqh, &wait.wait); goto repeat; } /* Only do the copy if the currently-owning transaction * still needs it. If it is on the Forget list, the * committing transaction is past that stage. The * buffer had better remain locked during the kmalloc, * but that should be true --- we hold the journal lock * still and the buffer is already on the BUF_JOURNAL * list so won't be flushed. * * Subtle point, though: if this is a get_undo_access, * then we will be relying on the frozen_data to contain * the new value of the committed_data record after the * transaction, so we HAVE to force the frozen_data copy * in that case. */ if (jh->b_jlist != BJ_Forget || force_copy) { JBUFFER_TRACE(jh, "generate frozen data"); if (!frozen_buffer) { JBUFFER_TRACE(jh, "allocate memory for buffer"); jbd_unlock_bh_state(bh); frozen_buffer = jbd_alloc(jh2bh(jh)->b_size, GFP_NOFS); if (!frozen_buffer) { printk(KERN_EMERG "%s: OOM for frozen_buffer\n", __func__); JBUFFER_TRACE(jh, "oom!"); error = -ENOMEM; jbd_lock_bh_state(bh); goto done; } goto repeat; } jh->b_frozen_data = frozen_buffer; frozen_buffer = NULL; need_copy = 1; } jh->b_next_transaction = transaction; } /* * Finally, if the buffer is not journaled right now, we need to make * sure it doesn't get written to disk before the caller actually * commits the new data */ if (!jh->b_transaction) { JBUFFER_TRACE(jh, "no transaction"); J_ASSERT_JH(jh, !jh->b_next_transaction); jh->b_transaction = transaction; JBUFFER_TRACE(jh, "file as BJ_Reserved"); spin_lock(&journal->j_list_lock); __journal_file_buffer(jh, transaction, BJ_Reserved); spin_unlock(&journal->j_list_lock); } done: if (need_copy) { struct page *page; int offset; char *source; J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), "Possible IO failure.\n"); page = jh2bh(jh)->b_page; offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK; source = kmap_atomic(page, KM_USER0); memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); kunmap_atomic(source, KM_USER0); } jbd_unlock_bh_state(bh); /* * If we are about to journal a buffer, then any revoke pending on it is * no longer valid */ journal_cancel_revoke(handle, jh); out: if (unlikely(frozen_buffer)) /* It's usually NULL */ jbd_free(frozen_buffer, bh->b_size); JBUFFER_TRACE(jh, "exit"); return error; }
int journal_write_metadata_buffer(transaction_t *transaction, struct journal_head *jh_in, struct journal_head **jh_out, unsigned int blocknr) { int need_copy_out = 0; int done_copy_out = 0; int do_escape = 0; char *mapped_data; struct buffer_head *new_bh; struct journal_head *new_jh; struct page *new_page; unsigned int new_offset; struct buffer_head *bh_in = jh2bh(jh_in); journal_t *journal = transaction->t_journal; /* * The buffer really shouldn't be locked: only the current committing * transaction is allowed to write it, so nobody else is allowed * to do any IO. * * akpm: except if we're journalling data, and write() output is * also part of a shared mapping, and another thread has * decided to launch a writepage() against this buffer. */ J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); /* keep subsequent assertions sane */ new_bh->b_state = 0; init_buffer(new_bh, NULL, NULL); atomic_set(&new_bh->b_count, 1); new_jh = journal_add_journal_head(new_bh); /* This sleeps */ /* * If a new transaction has already done a buffer copy-out, then * we use that version of the data for the commit. */ jbd_lock_bh_state(bh_in); repeat: if (jh_in->b_frozen_data) { done_copy_out = 1; new_page = virt_to_page(jh_in->b_frozen_data); new_offset = offset_in_page(jh_in->b_frozen_data); } else { new_page = jh2bh(jh_in)->b_page; new_offset = offset_in_page(jh2bh(jh_in)->b_data); } mapped_data = kmap_atomic(new_page); /* * Check for escaping */ if (*((__be32 *)(mapped_data + new_offset)) == cpu_to_be32(JFS_MAGIC_NUMBER)) { need_copy_out = 1; do_escape = 1; } kunmap_atomic(mapped_data); /* * Do we need to do a data copy? */ if (need_copy_out && !done_copy_out) { char *tmp; jbd_unlock_bh_state(bh_in); tmp = jbd_alloc(bh_in->b_size, GFP_NOFS); jbd_lock_bh_state(bh_in); if (jh_in->b_frozen_data) { jbd_free(tmp, bh_in->b_size); goto repeat; } jh_in->b_frozen_data = tmp; mapped_data = kmap_atomic(new_page); memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size); kunmap_atomic(mapped_data); new_page = virt_to_page(tmp); new_offset = offset_in_page(tmp); done_copy_out = 1; } /* * Did we need to do an escaping? Now we've done all the * copying, we can finally do so. */ if (do_escape) { mapped_data = kmap_atomic(new_page); *((unsigned int *)(mapped_data + new_offset)) = 0; kunmap_atomic(mapped_data); } set_bh_page(new_bh, new_page, new_offset); new_jh->b_transaction = NULL; new_bh->b_size = jh2bh(jh_in)->b_size; new_bh->b_bdev = transaction->t_journal->j_dev; new_bh->b_blocknr = blocknr; set_buffer_mapped(new_bh); set_buffer_dirty(new_bh); *jh_out = new_jh; /* * The to-be-written buffer needs to get moved to the io queue, * and the original buffer whose contents we are shadowing or * copying is moved to the transaction's shadow queue. */ JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); spin_lock(&journal->j_list_lock); __journal_file_buffer(jh_in, transaction, BJ_Shadow); spin_unlock(&journal->j_list_lock); jbd_unlock_bh_state(bh_in); JBUFFER_TRACE(new_jh, "file as BJ_IO"); journal_file_buffer(new_jh, transaction, BJ_IO); return do_escape | (done_copy_out << 1); }